* [RFC 00/11] QLogic RDMA Driver (qedr) RFC
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
This series introduces RoCE RDMA driver for the 579xx RDMA products by Qlogic.
The RDMA support is added as an additional loadable module (qedr) over the Ethernet qede driver.
The qedr module will support both RoCE and iWarp, although this series only adds RoCE support.
The qed and qede drivers are enhanced with functionality required for RDMA support.
Any review/comment is appreciated.
Based on net-next e19ac1578fd8eb3f7c93f2be2657deb2ccefffd7.
Rajesh Borundia (10):
Add RoCE driver framework
Add support for PD,PKEY and CQ verbs
Add support for memory registeration verbs
Add support for data path
Add LL2 RoCE interface
Add events support and register IB device
Add support for RoCE HW init
Add support for user context verbs
Add support for QP verbs
Add GSI support
Yuval Mintz (1):
qed: Add LL2
drivers/infiniband/Kconfig | 2 +
drivers/infiniband/hw/Makefile | 1 +
drivers/infiniband/hw/qedr/Kconfig | 7 +
drivers/infiniband/hw/qedr/Makefile | 3 +
drivers/infiniband/hw/qedr/main.c | 907 ++++++
drivers/infiniband/hw/qedr/qedr.h | 494 ++++
drivers/infiniband/hw/qedr/qedr_cm.c | 626 +++++
drivers/infiniband/hw/qedr/qedr_cm.h | 61 +
drivers/infiniband/hw/qedr/qedr_hsi.h | 56 +
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 748 +++++
drivers/infiniband/hw/qedr/qedr_user.h | 80 +
drivers/infiniband/hw/qedr/verbs.c | 3562 ++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 75 +
drivers/net/ethernet/qlogic/Kconfig | 8 +
drivers/net/ethernet/qlogic/qed/Makefile | 2 +
drivers/net/ethernet/qlogic/qed/qed.h | 37 +-
drivers/net/ethernet/qlogic/qed/qed_cxt.c | 8 +
drivers/net/ethernet/qlogic/qed/qed_cxt.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_dev.c | 276 +-
drivers/net/ethernet/qlogic/qed/qed_dev_api.h | 20 +
drivers/net/ethernet/qlogic/qed/qed_hsi.h | 3 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 1790 ++++++++++++
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 316 +++
drivers/net/ethernet/qlogic/qed/qed_main.c | 67 +-
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 29 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 2975 ++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 210 ++
drivers/net/ethernet/qlogic/qed/qed_sp.h | 5 +
drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
drivers/net/ethernet/qlogic/qede/Makefile | 1 +
drivers/net/ethernet/qlogic/qede/qede.h | 9 +
drivers/net/ethernet/qlogic/qede/qede_main.c | 35 +-
drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 ++
include/linux/qed/common_hsi.h | 1 +
include/linux/qed/qed_if.h | 9 +-
include/linux/qed/qed_ll2_if.h | 140 +
include/linux/qed/qed_roce_if.h | 604 ++++
include/linux/qed/qede_roce.h | 88 +
include/linux/qed/rdma_common.h | 1 +
include/uapi/linux/pci_regs.h | 3 +
40 files changed, 13564 insertions(+), 19 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/Kconfig
create mode 100644 drivers/infiniband/hw/qedr/Makefile
create mode 100644 drivers/infiniband/hw/qedr/main.c
create mode 100644 drivers/infiniband/hw/qedr/qedr.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.c
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_user.h
create mode 100644 drivers/infiniband/hw/qedr/verbs.c
create mode 100644 drivers/infiniband/hw/qedr/verbs.h
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.h
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
create mode 100644 include/linux/qed/qed_ll2_if.h
create mode 100644 include/linux/qed/qed_roce_if.h
create mode 100644 include/linux/qed/qede_roce.h
--
1.8.3.1
^ permalink raw reply [flat|nested] 78+ messages in thread
* [RFC 00/11] QLogic RDMA Driver (qedr) RFC
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
This series introduces RoCE RDMA driver for the 579xx RDMA products by Qlogic.
The RDMA support is added as an additional loadable module (qedr) over the Ethernet qede driver.
The qedr module will support both RoCE and iWarp, although this series only adds RoCE support.
The qed and qede drivers are enhanced with functionality required for RDMA support.
Any review/comment is appreciated.
Based on net-next e19ac1578fd8eb3f7c93f2be2657deb2ccefffd7.
Rajesh Borundia (10):
Add RoCE driver framework
Add support for PD,PKEY and CQ verbs
Add support for memory registeration verbs
Add support for data path
Add LL2 RoCE interface
Add events support and register IB device
Add support for RoCE HW init
Add support for user context verbs
Add support for QP verbs
Add GSI support
Yuval Mintz (1):
qed: Add LL2
drivers/infiniband/Kconfig | 2 +
drivers/infiniband/hw/Makefile | 1 +
drivers/infiniband/hw/qedr/Kconfig | 7 +
drivers/infiniband/hw/qedr/Makefile | 3 +
drivers/infiniband/hw/qedr/main.c | 907 ++++++
drivers/infiniband/hw/qedr/qedr.h | 494 ++++
drivers/infiniband/hw/qedr/qedr_cm.c | 626 +++++
drivers/infiniband/hw/qedr/qedr_cm.h | 61 +
drivers/infiniband/hw/qedr/qedr_hsi.h | 56 +
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 748 +++++
drivers/infiniband/hw/qedr/qedr_user.h | 80 +
drivers/infiniband/hw/qedr/verbs.c | 3562 ++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 75 +
drivers/net/ethernet/qlogic/Kconfig | 8 +
drivers/net/ethernet/qlogic/qed/Makefile | 2 +
drivers/net/ethernet/qlogic/qed/qed.h | 37 +-
drivers/net/ethernet/qlogic/qed/qed_cxt.c | 8 +
drivers/net/ethernet/qlogic/qed/qed_cxt.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_dev.c | 276 +-
drivers/net/ethernet/qlogic/qed/qed_dev_api.h | 20 +
drivers/net/ethernet/qlogic/qed/qed_hsi.h | 3 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 1790 ++++++++++++
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 316 +++
drivers/net/ethernet/qlogic/qed/qed_main.c | 67 +-
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 29 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 2975 ++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 210 ++
drivers/net/ethernet/qlogic/qed/qed_sp.h | 5 +
drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
drivers/net/ethernet/qlogic/qede/Makefile | 1 +
drivers/net/ethernet/qlogic/qede/qede.h | 9 +
drivers/net/ethernet/qlogic/qede/qede_main.c | 35 +-
drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 ++
include/linux/qed/common_hsi.h | 1 +
include/linux/qed/qed_if.h | 9 +-
include/linux/qed/qed_ll2_if.h | 140 +
include/linux/qed/qed_roce_if.h | 604 ++++
include/linux/qed/qede_roce.h | 88 +
include/linux/qed/rdma_common.h | 1 +
include/uapi/linux/pci_regs.h | 3 +
40 files changed, 13564 insertions(+), 19 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/Kconfig
create mode 100644 drivers/infiniband/hw/qedr/Makefile
create mode 100644 drivers/infiniband/hw/qedr/main.c
create mode 100644 drivers/infiniband/hw/qedr/qedr.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.c
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_user.h
create mode 100644 drivers/infiniband/hw/qedr/verbs.c
create mode 100644 drivers/infiniband/hw/qedr/verbs.h
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.h
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
create mode 100644 include/linux/qed/qed_ll2_if.h
create mode 100644 include/linux/qed/qed_roce_if.h
create mode 100644 include/linux/qed/qede_roce.h
--
1.8.3.1
^ permalink raw reply [flat|nested] 78+ messages in thread
* [RFC 01/11] qed: Add LL2
[not found] ` <1473696465-27986-1-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` Ram Amrani
` (4 subsequent siblings)
5 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
From: Yuval Mintz <Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
Add support for qed light L2 feature. It is used by the RDMA driver and for other purposes as well.
Signed-off-by: Yuval Mintz <Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
drivers/net/ethernet/qlogic/Kconfig | 8 +
drivers/net/ethernet/qlogic/qed/Makefile | 1 +
drivers/net/ethernet/qlogic/qed/qed.h | 8 +
drivers/net/ethernet/qlogic/qed/qed_cxt.c | 2 +
drivers/net/ethernet/qlogic/qed/qed_dev.c | 120 +-
drivers/net/ethernet/qlogic/qed/qed_dev_api.h | 20 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 1698 ++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 289 ++++
drivers/net/ethernet/qlogic/qed/qed_main.c | 23 +-
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 22 +
drivers/net/ethernet/qlogic/qed/qed_sp.h | 4 +
include/linux/qed/qed_if.h | 1 +
include/linux/qed/qed_ll2_if.h | 140 ++
13 files changed, 2333 insertions(+), 3 deletions(-)
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.h
create mode 100644 include/linux/qed/qed_ll2_if.h
diff --git a/drivers/net/ethernet/qlogic/Kconfig b/drivers/net/ethernet/qlogic/Kconfig
index 6ba4840..9eb3b19 100644
--- a/drivers/net/ethernet/qlogic/Kconfig
+++ b/drivers/net/ethernet/qlogic/Kconfig
@@ -88,6 +88,14 @@ config QED
---help---
This enables the support for ...
+config QED_LL2
+ bool "Qlogic QED Light L2 interface"
+ default n
+ depends on QED
+ ---help---
+ This enables support for Light L2 interface which is required
+ by all qed protocol drivers other than qede.
+
config QED_SRIOV
bool "QLogic QED 25/40/100Gb SR-IOV support"
depends on QED && PCI_IOV
diff --git a/drivers/net/ethernet/qlogic/qed/Makefile b/drivers/net/ethernet/qlogic/qed/Makefile
index d1f157e..02f4842 100644
--- a/drivers/net/ethernet/qlogic/qed/Makefile
+++ b/drivers/net/ethernet/qlogic/qed/Makefile
@@ -4,3 +4,4 @@ qed-y := qed_cxt.o qed_dev.o qed_hw.o qed_init_fw_funcs.o qed_init_ops.o \
qed_int.o qed_main.o qed_mcp.o qed_sp_commands.o qed_spq.o qed_l2.o \
qed_selftest.o qed_dcbx.o
qed-$(CONFIG_QED_SRIOV) += qed_sriov.o qed_vf.o
+qed-$(CONFIG_QED_LL2) += qed_ll2.o
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index 9ba7a7e..a42824e 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -71,6 +71,7 @@ struct qed_sb_info;
struct qed_sb_attn_info;
struct qed_cxt_mngr;
struct qed_sb_sp_info;
+struct qed_ll2_info;
struct qed_mcp_info;
struct qed_rt_data {
@@ -151,6 +152,7 @@ enum QED_RESOURCES {
QED_MAC,
QED_VLAN,
QED_ILT,
+ QED_LL2_QUEUE,
QED_MAX_RESC,
};
@@ -359,6 +361,8 @@ struct qed_hwfn {
struct qed_sb_attn_info *p_sb_attn;
/* Protocol related */
+ bool using_ll2;
+ struct qed_ll2_info *p_ll2_info;
struct qed_pf_params pf_params;
bool b_rdma_enabled_in_prs;
@@ -544,6 +548,10 @@ struct qed_dev {
} protocol_ops;
void *ops_cookie;
+#ifdef CONFIG_QED_LL2
+ struct qed_cb_ll2_info *ll2;
+ u8 ll2_mac_address[ETH_ALEN];
+#endif
const struct firmware *firmware;
};
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.c b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
index 5476927..34af48d 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
@@ -1849,6 +1849,8 @@ int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn)
/* Set the number of required CORE connections */
u32 core_cids = 1; /* SPQ */
+ if (p_hwfn->using_ll2)
+ core_cids += 4;
qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
switch (p_hwfn->hw_info.personality) {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 5ae27f2..4afcda5 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -29,6 +29,7 @@
#include "qed_hw.h"
#include "qed_init_ops.h"
#include "qed_int.h"
+#include "qed_ll2.h"
#include "qed_mcp.h"
#include "qed_reg_addr.h"
#include "qed_sp.h"
@@ -147,6 +148,9 @@ void qed_resc_free(struct qed_dev *cdev)
qed_eq_free(p_hwfn, p_hwfn->p_eq);
qed_consq_free(p_hwfn, p_hwfn->p_consq);
qed_int_free(p_hwfn);
+#ifdef CONFIG_QED_LL2
+ qed_ll2_free(p_hwfn, p_hwfn->p_ll2_info);
+#endif
qed_iov_free(p_hwfn);
qed_dmae_info_free(p_hwfn);
qed_dcbx_info_free(p_hwfn, p_hwfn->p_dcbx_info);
@@ -404,6 +408,9 @@ int qed_qm_reconf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
int qed_resc_alloc(struct qed_dev *cdev)
{
+#ifdef CONFIG_QED_LL2
+ struct qed_ll2_info *p_ll2_info;
+#endif
struct qed_consq *p_consq;
struct qed_eq *p_eq;
int i, rc = 0;
@@ -520,6 +527,15 @@ int qed_resc_alloc(struct qed_dev *cdev)
goto alloc_no_mem;
p_hwfn->p_consq = p_consq;
+#ifdef CONFIG_QED_LL2
+ if (p_hwfn->using_ll2) {
+ p_ll2_info = qed_ll2_alloc(p_hwfn);
+ if (!p_ll2_info)
+ goto alloc_no_mem;
+ p_hwfn->p_ll2_info = p_ll2_info;
+ }
+#endif
+
/* DMA info initialization */
rc = qed_dmae_info_alloc(p_hwfn);
if (rc) {
@@ -576,6 +592,10 @@ void qed_resc_setup(struct qed_dev *cdev)
qed_int_setup(p_hwfn, p_hwfn->p_main_ptt);
qed_iov_setup(p_hwfn, p_hwfn->p_main_ptt);
+#ifdef CONFIG_QED_LL2
+ if (p_hwfn->using_ll2)
+ qed_ll2_setup(p_hwfn, p_hwfn->p_ll2_info);
+#endif
}
}
@@ -1319,6 +1339,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
resc_num[QED_VLAN] = (ETH_NUM_VLAN_FILTERS - 1 /*For vlan0*/) /
num_funcs;
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
+ resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
@@ -1342,7 +1363,8 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
"RL = %d start = %d\n"
"MAC = %d start = %d\n"
"VLAN = %d start = %d\n"
- "ILT = %d start = %d\n",
+ "ILT = %d start = %d\n"
+ "LL2_QUEUE = %d start = %d\n",
p_hwfn->hw_info.resc_num[QED_SB],
p_hwfn->hw_info.resc_start[QED_SB],
p_hwfn->hw_info.resc_num[QED_L2_QUEUE],
@@ -1358,7 +1380,9 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
p_hwfn->hw_info.resc_num[QED_VLAN],
p_hwfn->hw_info.resc_start[QED_VLAN],
p_hwfn->hw_info.resc_num[QED_ILT],
- p_hwfn->hw_info.resc_start[QED_ILT]);
+ p_hwfn->hw_info.resc_start[QED_ILT],
+ RESC_NUM(p_hwfn, QED_LL2_QUEUE),
+ RESC_START(p_hwfn, QED_LL2_QUEUE));
return 0;
}
@@ -2163,6 +2187,98 @@ int qed_fw_rss_eng(struct qed_hwfn *p_hwfn, u8 src_id, u8 *dst_id)
return 0;
}
+static void qed_llh_mac_to_filter(u32 *p_high, u32 *p_low,
+ u8 *p_filter)
+{
+ *p_high = p_filter[1] | (p_filter[0] << 8);
+ *p_low = p_filter[5] | (p_filter[4] << 8) |
+ (p_filter[3] << 16) | (p_filter[2] << 24);
+}
+
+int qed_llh_add_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter)
+{
+ u32 high = 0, low = 0, en;
+ int i;
+
+ if (!(IS_MF_SI(p_hwfn) || IS_MF_DEFAULT(p_hwfn)))
+ return 0;
+
+ qed_llh_mac_to_filter(&high, &low, p_filter);
+
+ /* Find a free entry and utilize it */
+ for (i = 0; i < NIG_REG_LLH_FUNC_FILTER_EN_SIZE; i++) {
+ en = qed_rd(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32));
+ if (en)
+ continue;
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ 2 * i * sizeof(u32), low);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ (2 * i + 1) * sizeof(u32), high);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_MODE + i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE +
+ i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32), 1);
+ break;
+ }
+ if (i >= NIG_REG_LLH_FUNC_FILTER_EN_SIZE) {
+ DP_NOTICE(p_hwfn,
+ "Failed to find an empty LLH filter to utilize\n");
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
+ "MAC: %pM is added at %d\n",
+ p_filter, i);
+
+ return 0;
+}
+
+void qed_llh_remove_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter)
+{
+ u32 high = 0, low = 0;
+ int i;
+
+ if (!(IS_MF_SI(p_hwfn) || IS_MF_DEFAULT(p_hwfn)))
+ return;
+
+ qed_llh_mac_to_filter(&high, &low, p_filter);
+
+ /* Find the entry and clean it */
+ for (i = 0; i < NIG_REG_LLH_FUNC_FILTER_EN_SIZE; i++) {
+ if (qed_rd(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ 2 * i * sizeof(u32)) != low)
+ continue;
+ if (qed_rd(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ (2 * i + 1) * sizeof(u32)) != high)
+ continue;
+
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE + 2 * i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ (2 * i + 1) * sizeof(u32), 0);
+
+ DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
+ "MAC: %pM is removed from %d\n",
+ p_filter, i);
+ break;
+ }
+ if (i >= NIG_REG_LLH_FUNC_FILTER_EN_SIZE)
+ DP_NOTICE(p_hwfn, "Tried to remove a non-configured filter\n");
+}
+
static int qed_set_coalesce(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
u32 hw_addr, void *p_eth_qzone,
size_t eth_qzone_size, u8 timeset)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev_api.h b/drivers/net/ethernet/qlogic/qed/qed_dev_api.h
index 343bb03..2b4294b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev_api.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev_api.h
@@ -310,6 +310,26 @@ int qed_fw_rss_eng(struct qed_hwfn *p_hwfn,
u8 *dst_id);
/**
+ * @brief qed_llh_add_mac_filter - configures a MAC filter in llh
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param p_filter - MAC to add
+ */
+int qed_llh_add_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter);
+
+/**
+ * @brief qed_llh_remove_mac_filter - removes a MAC filtre from llh
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param p_filter - MAC to remove
+ */
+void qed_llh_remove_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter);
+
+/**
* *@brief Cleanup of previous driver remains prior to load
*
* @param p_hwfn
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
new file mode 100644
index 0000000..1a96a2d
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -0,0 +1,1698 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#include <linux/types.h>
+#include <asm/byteorder.h>
+#include <linux/dma-mapping.h>
+#include <linux/if_vlan.h>
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/stddef.h>
+#include <linux/version.h>
+#include <linux/workqueue.h>
+#include <net/ipv6.h>
+#include <linux/bitops.h>
+#include <linux/delay.h>
+#include <linux/errno.h>
+#include <linux/etherdevice.h>
+#include <linux/io.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/qed/qed_ll2_if.h>
+#include "qed.h"
+#include "qed_cxt.h"
+#include "qed_dev_api.h"
+#include "qed_hsi.h"
+#include "qed_hw.h"
+#include "qed_int.h"
+#include "qed_ll2.h"
+#include "qed_mcp.h"
+#include "qed_reg_addr.h"
+#include "qed_sp.h"
+
+#define QED_LL2_RX_REGISTERED(ll2) ((ll2)->rx_queue.b_cb_registred)
+#define QED_LL2_TX_REGISTERED(ll2) ((ll2)->tx_queue.b_cb_registred)
+
+#define QED_LL2_TX_SIZE (256)
+#define QED_LL2_RX_SIZE (4096)
+
+struct qed_cb_ll2_info {
+ int rx_cnt;
+ u32 rx_size;
+ u8 handle;
+ bool frags_mapped;
+
+ /* Lock protecting LL2 buffer lists in sleepless context */
+ spinlock_t lock;
+ struct list_head list;
+
+ const struct qed_ll2_cb_ops *cbs;
+ void *cb_cookie;
+};
+
+struct qed_ll2_buffer {
+ struct list_head list;
+ void *data;
+ dma_addr_t phys_addr;
+};
+
+static void qed_ll2b_complete_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment,
+ bool b_last_packet)
+{
+ struct qed_dev *cdev = p_hwfn->cdev;
+ struct sk_buff *skb = cookie;
+
+ /* All we need to do is release the mapping */
+ dma_unmap_single(&p_hwfn->cdev->pdev->dev, first_frag_addr,
+ skb_headlen(skb), DMA_TO_DEVICE);
+
+ if (cdev->ll2->cbs && cdev->ll2->cbs->tx_cb)
+ cdev->ll2->cbs->tx_cb(cdev->ll2->cb_cookie, skb,
+ b_last_fragment);
+
+ if (cdev->ll2->frags_mapped)
+ /* Case where mapped frags were received, need to
+ * free skb with nr_frags marked as 0
+ */
+ skb_shinfo(skb)->nr_frags = 0;
+
+ dev_kfree_skb_any(skb);
+}
+
+static int qed_ll2_alloc_buffer(struct qed_dev *cdev,
+ u8 **data, dma_addr_t *phys_addr)
+{
+ *data = kmalloc(cdev->ll2->rx_size, GFP_ATOMIC);
+ if (!(*data)) {
+ DP_INFO(cdev, "Failed to allocate LL2 buffer data\n");
+ return -ENOMEM;
+ }
+
+ *phys_addr = dma_map_single(&cdev->pdev->dev,
+ ((*data) + NET_SKB_PAD),
+ cdev->ll2->rx_size, DMA_FROM_DEVICE);
+ if (dma_mapping_error(&cdev->pdev->dev, *phys_addr)) {
+ DP_INFO(cdev, "Failed to map LL2 buffer data\n");
+ kfree((*data));
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+static int qed_ll2_dealloc_buffer(struct qed_dev *cdev,
+ struct qed_ll2_buffer *buffer)
+{
+ spin_lock_bh(&cdev->ll2->lock);
+
+ dma_unmap_single(&cdev->pdev->dev, buffer->phys_addr,
+ cdev->ll2->rx_size, DMA_FROM_DEVICE);
+ kfree(buffer->data);
+ list_del(&buffer->list);
+
+ cdev->ll2->rx_cnt--;
+ if (!cdev->ll2->rx_cnt)
+ DP_INFO(cdev, "All LL2 entries were removed\n");
+
+ spin_unlock_bh(&cdev->ll2->lock);
+
+ return 0;
+}
+
+static void qed_ll2_kill_buffers(struct qed_dev *cdev)
+{
+ struct qed_ll2_buffer *buffer, *tmp_buffer;
+
+ list_for_each_entry_safe(buffer, tmp_buffer, &cdev->ll2->list, list)
+ qed_ll2_dealloc_buffer(cdev, buffer);
+}
+
+void qed_ll2b_complete_rx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ struct qed_ll2_rx_packet *p_pkt,
+ struct core_rx_fast_path_cqe *p_cqe,
+ bool b_last_packet)
+{
+ u16 packet_length = le16_to_cpu(p_cqe->packet_length);
+ struct qed_ll2_buffer *buffer = p_pkt->cookie;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ u16 vlan = le16_to_cpu(p_cqe->vlan);
+ u32 opaque_data_0, opaque_data_1;
+ u8 pad = p_cqe->placement_offset;
+ dma_addr_t new_phys_addr;
+ struct sk_buff *skb;
+ bool reuse = false;
+ int rc = -EINVAL;
+ u8 *new_data;
+
+ opaque_data_0 = le32_to_cpu(p_cqe->opaque_data.data[0]);
+ opaque_data_1 = le32_to_cpu(p_cqe->opaque_data.data[1]);
+
+ DP_VERBOSE(p_hwfn,
+ (NETIF_MSG_RX_STATUS | QED_MSG_STORAGE | NETIF_MSG_PKTDATA),
+ "Got an LL2 Rx completion: [Buffer at phys 0x%llx, offset 0x%02x] Length 0x%04x Parse_flags 0x%04x vlan 0x%04x Opaque data [0x%08x:0x%08x]\n",
+ (u64)p_pkt->rx_buf_addr, pad, packet_length,
+ le16_to_cpu(p_cqe->parse_flags.flags), vlan,
+ opaque_data_0, opaque_data_1);
+
+ if ((cdev->dp_module & NETIF_MSG_PKTDATA) && buffer->data) {
+ print_hex_dump(KERN_INFO, "",
+ DUMP_PREFIX_OFFSET, 16, 1,
+ buffer->data, packet_length, false);
+ }
+
+ /* Determine if data is valid */
+ if (packet_length < ETH_HLEN)
+ reuse = true;
+
+ /* Allocate a replacement for buffer; Reuse upon failure */
+ if (!reuse)
+ rc = qed_ll2_alloc_buffer(p_hwfn->cdev, &new_data,
+ &new_phys_addr);
+
+ /* If need to reuse or there's no replacement buffer, repost this */
+ if (rc)
+ goto out_post;
+
+ skb = build_skb(buffer->data, 0);
+ if (!skb) {
+ rc = -ENOMEM;
+ goto out_post;
+ }
+
+ pad += NET_SKB_PAD;
+ skb_reserve(skb, pad);
+ skb_put(skb, packet_length);
+ skb_checksum_none_assert(skb);
+
+ /* Get parital ethernet information instead of eth_type_trans(),
+ * Since we don't have an associated net_device.
+ */
+ skb_reset_mac_header(skb);
+ skb->protocol = eth_hdr(skb)->h_proto;
+
+ /* Pass SKB onward */
+ if (cdev->ll2->cbs && cdev->ll2->cbs->rx_cb) {
+ if (vlan)
+ __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan);
+ cdev->ll2->cbs->rx_cb(cdev->ll2->cb_cookie, skb,
+ opaque_data_0, opaque_data_1);
+ }
+
+ /* Update Buffer information and update FW producer */
+ buffer->data = new_data;
+ buffer->phys_addr = new_phys_addr;
+
+out_post:
+ rc = qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev), cdev->ll2->handle,
+ buffer->phys_addr, 0, buffer, 1);
+
+ if (rc)
+ qed_ll2_dealloc_buffer(cdev, buffer);
+}
+
+static struct qed_ll2_info *__qed_ll2_handle_sanity(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ bool b_lock,
+ bool b_only_active)
+{
+ struct qed_ll2_info *p_ll2_conn, *p_ret = NULL;
+
+ if (connection_handle >= QED_MAX_NUM_OF_LL2_CONNECTIONS)
+ return NULL;
+
+ if (!p_hwfn->p_ll2_info)
+ return NULL;
+
+ p_ll2_conn = &p_hwfn->p_ll2_info[connection_handle];
+
+ if (b_only_active) {
+ if (b_lock)
+ mutex_lock(&p_ll2_conn->mutex);
+ if (p_ll2_conn->b_active)
+ p_ret = p_ll2_conn;
+ if (b_lock)
+ mutex_unlock(&p_ll2_conn->mutex);
+ } else {
+ p_ret = p_ll2_conn;
+ }
+
+ return p_ret;
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity(struct qed_hwfn *p_hwfn,
+ u8 connection_handle)
+{
+ return __qed_ll2_handle_sanity(p_hwfn, connection_handle, false, true);
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity_lock(struct qed_hwfn *p_hwfn,
+ u8 connection_handle)
+{
+ return __qed_ll2_handle_sanity(p_hwfn, connection_handle, true, true);
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity_inactive(struct qed_hwfn
+ *p_hwfn,
+ u8 connection_handle)
+{
+ return __qed_ll2_handle_sanity(p_hwfn, connection_handle, false, false);
+}
+
+static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ bool b_last_packet = false, b_last_frag = false;
+ struct qed_ll2_tx_packet *p_pkt = NULL;
+ struct qed_ll2_info *p_ll2_conn;
+ struct qed_ll2_tx_queue *p_tx;
+
+ p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return;
+
+ p_tx = &p_ll2_conn->tx_queue;
+
+ while (!list_empty(&p_tx->active_descq)) {
+ p_pkt = list_first_entry(&p_tx->active_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (!p_pkt)
+ break;
+
+ list_del(&p_pkt->list_entry);
+ b_last_packet = list_empty(&p_tx->active_descq);
+ list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+ p_tx->cur_completing_packet = *p_pkt;
+ p_tx->cur_completing_bd_idx = 1;
+ b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
+
+ qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
+ p_pkt->cookie,
+ p_pkt->bds_set[0].tx_frag,
+ b_last_frag, b_last_packet);
+ }
+}
+
+static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
+{
+ struct qed_ll2_info *p_ll2_conn = p_cookie;
+ struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+ u16 new_idx = 0, num_bds = 0, num_bds_in_packet = 0;
+ struct qed_ll2_tx_packet *p_pkt;
+ bool b_last_frag = false;
+ unsigned long flags;
+ int rc = -EINVAL;
+
+ spin_lock_irqsave(&p_tx->lock, flags);
+ if (p_tx->b_completing_packet) {
+ rc = -EBUSY;
+ goto out;
+ }
+
+ new_idx = le16_to_cpu(*p_tx->p_fw_cons);
+ num_bds = ((s16)new_idx - (s16)p_tx->bds_idx);
+ while (num_bds) {
+ if (list_empty(&p_tx->active_descq))
+ goto out;
+
+ p_pkt = list_first_entry(&p_tx->active_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (!p_pkt)
+ goto out;
+
+ p_tx->b_completing_packet = true;
+ p_tx->cur_completing_packet = *p_pkt;
+ num_bds_in_packet = p_pkt->bd_used;
+ list_del(&p_pkt->list_entry);
+
+ if (num_bds < num_bds_in_packet) {
+ DP_NOTICE(p_hwfn,
+ "Rest of BDs does not cover whole packet\n");
+ goto out;
+ }
+
+ num_bds -= num_bds_in_packet;
+ p_tx->bds_idx += num_bds_in_packet;
+ while (num_bds_in_packet--)
+ qed_chain_consume(&p_tx->txq_chain);
+
+ p_tx->cur_completing_bd_idx = 1;
+ b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
+ list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+
+ spin_unlock_irqrestore(&p_tx->lock, flags);
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ p_pkt->bds_set[0].tx_frag,
+ b_last_frag, !num_bds);
+ spin_lock_irqsave(&p_tx->lock, flags);
+ }
+
+ p_tx->b_completing_packet = false;
+ rc = 0;
+out:
+ spin_unlock_irqrestore(&p_tx->lock, flags);
+ return rc;
+}
+
+static int qed_ll2_rxq_completion_reg(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn,
+ union core_rx_cqe_union *p_cqe,
+ unsigned long lock_flags,
+ bool b_last_cqe)
+{
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+ struct qed_ll2_rx_packet *p_pkt = NULL;
+
+ if (!list_empty(&p_rx->active_descq))
+ p_pkt = list_first_entry(&p_rx->active_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (!p_pkt) {
+ DP_NOTICE(p_hwfn,
+ "LL2 Rx completion but active_descq is empty\n");
+ return -EIO;
+ }
+ list_del(&p_pkt->list_entry);
+
+ if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
+ DP_NOTICE(p_hwfn,
+ "Mismatch between active_descq and the LL2 Rx chain\n");
+ list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+ spin_unlock_irqrestore(&p_rx->lock, lock_flags);
+ qed_ll2b_complete_rx_packet(p_hwfn, p_ll2_conn->my_id,
+ p_pkt, &p_cqe->rx_cqe_fp, b_last_cqe);
+ spin_lock_irqsave(&p_rx->lock, lock_flags);
+
+ return 0;
+}
+
+static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
+{
+ struct qed_ll2_info *p_ll2_conn = cookie;
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+ union core_rx_cqe_union *cqe = NULL;
+ u16 cq_new_idx = 0, cq_old_idx = 0;
+ unsigned long flags = 0;
+ int rc = 0;
+
+ spin_lock_irqsave(&p_rx->lock, flags);
+ cq_new_idx = le16_to_cpu(*p_rx->p_fw_cons);
+ cq_old_idx = qed_chain_get_cons_idx(&p_rx->rcq_chain);
+
+ while (cq_new_idx != cq_old_idx) {
+ bool b_last_cqe = (cq_new_idx == cq_old_idx);
+
+ cqe = qed_chain_consume(&p_rx->rcq_chain);
+ cq_old_idx = qed_chain_get_cons_idx(&p_rx->rcq_chain);
+
+ DP_VERBOSE(p_hwfn,
+ QED_MSG_LL2,
+ "LL2 [sw. cons %04x, fw. at %04x] - Got Packet of type %02x\n",
+ cq_old_idx, cq_new_idx, cqe->rx_cqe_sp.type);
+
+ switch (cqe->rx_cqe_sp.type) {
+ case CORE_RX_CQE_TYPE_SLOW_PATH:
+ DP_NOTICE(p_hwfn, "LL2 - unexpected Rx CQE slowpath\n");
+ rc = -EINVAL;
+ break;
+ case CORE_RX_CQE_TYPE_REGULAR:
+ rc = qed_ll2_rxq_completion_reg(p_hwfn, p_ll2_conn,
+ cqe, flags, b_last_cqe);
+ break;
+ default:
+ rc = -EIO;
+ }
+ }
+
+ spin_unlock_irqrestore(&p_rx->lock, flags);
+ return rc;
+}
+
+void qed_ll2_rxq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ struct qed_ll2_rx_packet *p_pkt = NULL;
+ struct qed_ll2_rx_queue *p_rx;
+
+ p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return;
+
+ p_rx = &p_ll2_conn->rx_queue;
+
+ while (!list_empty(&p_rx->active_descq)) {
+ dma_addr_t rx_buf_addr;
+ void *cookie;
+ bool b_last;
+
+ p_pkt = list_first_entry(&p_rx->active_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (!p_pkt)
+ break;
+
+ list_del(&p_pkt->list_entry);
+ list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+ rx_buf_addr = p_pkt->rx_buf_addr;
+ cookie = p_pkt->cookie;
+
+ b_last = list_empty(&p_rx->active_descq);
+ }
+}
+
+static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn,
+ u8 action_on_error)
+{
+ enum qed_ll2_conn_type conn_type = p_ll2_conn->conn_type;
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+ struct core_rx_start_ramrod_data *p_ramrod = NULL;
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ u16 cqe_pbl_size;
+ int rc = 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_RX_QUEUE_START,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.core_rx_queue_start;
+
+ p_ramrod->sb_id = cpu_to_le16(qed_int_get_sp_sb_id(p_hwfn));
+ p_ramrod->sb_index = p_rx->rx_sb_index;
+ p_ramrod->complete_event_flg = 1;
+
+ p_ramrod->mtu = cpu_to_le16(p_ll2_conn->mtu);
+ DMA_REGPAIR_LE(p_ramrod->bd_base,
+ p_rx->rxq_chain.p_phys_addr);
+ cqe_pbl_size = (u16)qed_chain_get_page_cnt(&p_rx->rcq_chain);
+ p_ramrod->num_of_pbl_pages = cpu_to_le16(cqe_pbl_size);
+ DMA_REGPAIR_LE(p_ramrod->cqe_pbl_addr,
+ qed_chain_get_pbl_phys(&p_rx->rcq_chain));
+
+ p_ramrod->drop_ttl0_flg = p_ll2_conn->rx_drop_ttl0_flg;
+ p_ramrod->inner_vlan_removal_en = p_ll2_conn->rx_vlan_removal_en;
+ p_ramrod->queue_id = p_ll2_conn->queue_id;
+
+ if ((IS_MF_DEFAULT(p_hwfn) || IS_MF_SI(p_hwfn)) &&
+ p_ramrod->main_func_queue && (conn_type != QED_LL2_TYPE_ROCE)) {
+ p_ramrod->mf_si_bcast_accept_all = 1;
+ p_ramrod->mf_si_mcast_accept_all = 1;
+ } else {
+ p_ramrod->mf_si_bcast_accept_all = 0;
+ p_ramrod->mf_si_mcast_accept_all = 0;
+ }
+
+ p_ramrod->action_on_error.error_type = action_on_error;
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ enum qed_ll2_conn_type conn_type = p_ll2_conn->conn_type;
+ struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+ struct core_tx_start_ramrod_data *p_ramrod = NULL;
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ union qed_qm_pq_params pq_params;
+ u16 pq_id = 0, pbl_size;
+ int rc = -EINVAL;
+
+ if (!QED_LL2_TX_REGISTERED(p_ll2_conn))
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_TX_QUEUE_START,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.core_tx_queue_start;
+
+ p_ramrod->sb_id = cpu_to_le16(qed_int_get_sp_sb_id(p_hwfn));
+ p_ramrod->sb_index = p_tx->tx_sb_index;
+ p_ramrod->mtu = cpu_to_le16(p_ll2_conn->mtu);
+ p_ll2_conn->tx_stats_en = 1;
+ p_ramrod->stats_en = p_ll2_conn->tx_stats_en;
+ p_ramrod->stats_id = p_ll2_conn->tx_stats_id;
+
+ DMA_REGPAIR_LE(p_ramrod->pbl_base_addr,
+ qed_chain_get_pbl_phys(&p_tx->txq_chain));
+ pbl_size = qed_chain_get_page_cnt(&p_tx->txq_chain);
+ p_ramrod->pbl_size = cpu_to_le16(pbl_size);
+
+ memset(&pq_params, 0, sizeof(pq_params));
+ pq_params.core.tc = p_ll2_conn->tx_tc;
+ pq_id = qed_get_qm_pq(p_hwfn, PROTOCOLID_CORE, &pq_params);
+ p_ramrod->qm_pq_id = cpu_to_le16(pq_id);
+
+ switch (conn_type) {
+ case QED_LL2_TYPE_ISCSI:
+ case QED_LL2_TYPE_ISCSI_OOO:
+ p_ramrod->conn_type = PROTOCOLID_ISCSI;
+ break;
+ case QED_LL2_TYPE_ROCE:
+ p_ramrod->conn_type = PROTOCOLID_ROCE;
+ break;
+ default:
+ p_ramrod->conn_type = PROTOCOLID_ETH;
+ DP_NOTICE(p_hwfn, "Unknown connection type: %d\n", conn_type);
+ }
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_rx_queue_stop(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ struct core_rx_stop_ramrod_data *p_ramrod = NULL;
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ int rc = -EINVAL;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_RX_QUEUE_STOP,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.core_rx_queue_stop;
+
+ p_ramrod->complete_event_flg = 1;
+ p_ramrod->queue_id = p_ll2_conn->queue_id;
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_tx_queue_stop(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ int rc = -EINVAL;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_TX_QUEUE_STOP,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int
+qed_ll2_acquire_connection_rx(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info, u16 rx_num_desc)
+{
+ struct qed_ll2_rx_packet *p_descq;
+ u32 capacity;
+ int rc = 0;
+
+ if (!rx_num_desc)
+ goto out;
+
+ rc = qed_chain_alloc(p_hwfn->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_NEXT_PTR,
+ QED_CHAIN_CNT_TYPE_U16,
+ rx_num_desc,
+ sizeof(struct core_rx_bd),
+ &p_ll2_info->rx_queue.rxq_chain);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "Failed to allocate ll2 rxq chain\n");
+ goto out;
+ }
+
+ capacity = qed_chain_get_capacity(&p_ll2_info->rx_queue.rxq_chain);
+ p_descq = kcalloc(capacity, sizeof(struct qed_ll2_rx_packet),
+ GFP_KERNEL);
+ if (!p_descq) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn, "Failed to allocate ll2 Rx desc\n");
+ goto out;
+ }
+ p_ll2_info->rx_queue.descq_array = p_descq;
+
+ rc = qed_chain_alloc(p_hwfn->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U16,
+ rx_num_desc,
+ sizeof(struct core_rx_fast_path_cqe),
+ &p_ll2_info->rx_queue.rcq_chain);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "Failed to allocate ll2 rcq chain\n");
+ goto out;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_LL2,
+ "Allocated LL2 Rxq [Type %08x] with 0x%08x buffers\n",
+ p_ll2_info->conn_type, rx_num_desc);
+
+out:
+ return rc;
+}
+
+static int qed_ll2_acquire_connection_tx(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info,
+ u16 tx_num_desc)
+{
+ struct qed_ll2_tx_packet *p_descq;
+ u32 capacity;
+ int rc = 0;
+
+ if (!tx_num_desc)
+ goto out;
+
+ rc = qed_chain_alloc(p_hwfn->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U16,
+ tx_num_desc,
+ sizeof(struct core_tx_bd),
+ &p_ll2_info->tx_queue.txq_chain);
+ if (rc)
+ goto out;
+
+ capacity = qed_chain_get_capacity(&p_ll2_info->tx_queue.txq_chain);
+ p_descq = kcalloc(capacity, sizeof(struct qed_ll2_tx_packet),
+ GFP_KERNEL);
+ if (!p_descq) {
+ rc = -ENOMEM;
+ goto out;
+ }
+ p_ll2_info->tx_queue.descq_array = p_descq;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_LL2,
+ "Allocated LL2 Txq [Type %08x] with 0x%08x buffers\n",
+ p_ll2_info->conn_type, tx_num_desc);
+
+out:
+ if (rc)
+ DP_NOTICE(p_hwfn,
+ "Can't allocate memory for Tx LL2 with 0x%08x buffers\n",
+ tx_num_desc);
+ return rc;
+}
+
+int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_params,
+ u16 rx_num_desc,
+ u16 tx_num_desc,
+ u8 *p_connection_handle)
+{
+ qed_int_comp_cb_t comp_rx_cb, comp_tx_cb;
+ struct qed_ll2_info *p_ll2_info = NULL;
+ int rc;
+ u8 i;
+
+ if (!p_connection_handle || !p_hwfn->p_ll2_info)
+ return -EINVAL;
+
+ /* Find a free connection to be used */
+ for (i = 0; (i < QED_MAX_NUM_OF_LL2_CONNECTIONS); i++) {
+ mutex_lock(&p_hwfn->p_ll2_info[i].mutex);
+ if (p_hwfn->p_ll2_info[i].b_active) {
+ mutex_unlock(&p_hwfn->p_ll2_info[i].mutex);
+ continue;
+ }
+
+ p_hwfn->p_ll2_info[i].b_active = true;
+ p_ll2_info = &p_hwfn->p_ll2_info[i];
+ mutex_unlock(&p_hwfn->p_ll2_info[i].mutex);
+ break;
+ }
+ if (!p_ll2_info)
+ return -EBUSY;
+
+ p_ll2_info->conn_type = p_params->conn_type;
+ p_ll2_info->mtu = p_params->mtu;
+ p_ll2_info->rx_drop_ttl0_flg = p_params->rx_drop_ttl0_flg;
+ p_ll2_info->rx_vlan_removal_en = p_params->rx_vlan_removal_en;
+ p_ll2_info->tx_tc = p_params->tx_tc;
+ p_ll2_info->tx_dest = p_params->tx_dest;
+ p_ll2_info->ai_err_packet_too_big = p_params->ai_err_packet_too_big;
+ p_ll2_info->ai_err_no_buf = p_params->ai_err_no_buf;
+
+ rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info, rx_num_desc);
+ if (rc)
+ goto q_allocate_fail;
+
+ rc = qed_ll2_acquire_connection_tx(p_hwfn, p_ll2_info, tx_num_desc);
+ if (rc)
+ goto q_allocate_fail;
+
+ /* Register callbacks for the Rx/Tx queues */
+ comp_rx_cb = qed_ll2_rxq_completion;
+ comp_tx_cb = qed_ll2_txq_completion;
+
+ if (rx_num_desc) {
+ qed_int_register_cb(p_hwfn, comp_rx_cb,
+ &p_hwfn->p_ll2_info[i],
+ &p_ll2_info->rx_queue.rx_sb_index,
+ &p_ll2_info->rx_queue.p_fw_cons);
+ p_ll2_info->rx_queue.b_cb_registred = true;
+ }
+
+ if (tx_num_desc) {
+ qed_int_register_cb(p_hwfn,
+ comp_tx_cb,
+ &p_hwfn->p_ll2_info[i],
+ &p_ll2_info->tx_queue.tx_sb_index,
+ &p_ll2_info->tx_queue.p_fw_cons);
+ p_ll2_info->tx_queue.b_cb_registred = true;
+ }
+
+ *p_connection_handle = i;
+ return rc;
+
+q_allocate_fail:
+ qed_ll2_release_connection(p_hwfn, i);
+ return -ENOMEM;
+}
+
+static int qed_ll2_establish_connection_rx(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ u8 action_on_error = 0;
+
+ if (!QED_LL2_RX_REGISTERED(p_ll2_conn))
+ return 0;
+
+ DIRECT_REG_WR(p_ll2_conn->rx_queue.set_prod_addr, 0x0);
+
+ SET_FIELD(action_on_error,
+ CORE_RX_ACTION_ON_ERROR_PACKET_TOO_BIG,
+ p_ll2_conn->ai_err_packet_too_big);
+ SET_FIELD(action_on_error,
+ CORE_RX_ACTION_ON_ERROR_NO_BUFF, p_ll2_conn->ai_err_no_buf);
+
+ return qed_sp_ll2_rx_queue_start(p_hwfn, p_ll2_conn, action_on_error);
+}
+
+int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn;
+ struct qed_ll2_rx_queue *p_rx;
+ struct qed_ll2_tx_queue *p_tx;
+ int rc = -EINVAL;
+ u32 i, capacity;
+ u8 qid;
+
+ p_ll2_conn = qed_ll2_handle_sanity_lock(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+ p_rx = &p_ll2_conn->rx_queue;
+ p_tx = &p_ll2_conn->tx_queue;
+
+ qed_chain_reset(&p_rx->rxq_chain);
+ qed_chain_reset(&p_rx->rcq_chain);
+ INIT_LIST_HEAD(&p_rx->active_descq);
+ INIT_LIST_HEAD(&p_rx->free_descq);
+ INIT_LIST_HEAD(&p_rx->posting_descq);
+ spin_lock_init(&p_rx->lock);
+ capacity = qed_chain_get_capacity(&p_rx->rxq_chain);
+ for (i = 0; i < capacity; i++)
+ list_add_tail(&p_rx->descq_array[i].list_entry,
+ &p_rx->free_descq);
+ *p_rx->p_fw_cons = 0;
+
+ qed_chain_reset(&p_tx->txq_chain);
+ INIT_LIST_HEAD(&p_tx->active_descq);
+ INIT_LIST_HEAD(&p_tx->free_descq);
+ INIT_LIST_HEAD(&p_tx->sending_descq);
+ spin_lock_init(&p_tx->lock);
+ capacity = qed_chain_get_capacity(&p_tx->txq_chain);
+ for (i = 0; i < capacity; i++)
+ list_add_tail(&p_tx->descq_array[i].list_entry,
+ &p_tx->free_descq);
+ p_tx->cur_completing_bd_idx = 0;
+ p_tx->bds_idx = 0;
+ p_tx->b_completing_packet = false;
+ p_tx->cur_send_packet = NULL;
+ p_tx->cur_send_frag_num = 0;
+ p_tx->cur_completing_frag_num = 0;
+ *p_tx->p_fw_cons = 0;
+
+ qed_cxt_acquire_cid(p_hwfn, PROTOCOLID_CORE, &p_ll2_conn->cid);
+
+ qid = p_hwfn->hw_info.resc_start[QED_LL2_QUEUE] + connection_handle;
+ p_ll2_conn->queue_id = qid;
+ p_ll2_conn->tx_stats_id = qid;
+ p_rx->set_prod_addr = (u8 __iomem *)p_hwfn->regview +
+ GTT_BAR0_MAP_REG_TSDM_RAM +
+ TSTORM_LL2_RX_PRODS_OFFSET(qid);
+ p_tx->doorbell_addr = (u8 __iomem *)p_hwfn->doorbells +
+ qed_db_addr(p_ll2_conn->cid,
+ DQ_DEMS_LEGACY);
+
+ rc = qed_ll2_establish_connection_rx(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+
+ rc = qed_sp_ll2_tx_queue_start(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+
+ if (p_hwfn->hw_info.personality != QED_PCI_ETH_ROCE)
+ qed_wr(p_hwfn, p_hwfn->p_main_ptt, PRS_REG_USE_LIGHT_L2, 1);
+
+ return rc;
+}
+
+static void qed_ll2_post_rx_buffer_notify_fw(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_rx_queue *p_rx,
+ struct qed_ll2_rx_packet *p_curp)
+{
+ struct qed_ll2_rx_packet *p_posting_packet = NULL;
+ struct core_ll2_rx_prod rx_prod = { 0, 0, 0 };
+ bool b_notify_fw = false;
+ u16 bd_prod, cq_prod;
+
+ /* This handles the flushing of already posted buffers */
+ while (!list_empty(&p_rx->posting_descq)) {
+ p_posting_packet = list_first_entry(&p_rx->posting_descq,
+ struct qed_ll2_rx_packet,
+ list_entry);
+ list_del(&p_posting_packet->list_entry);
+ list_add_tail(&p_posting_packet->list_entry,
+ &p_rx->active_descq);
+ b_notify_fw = true;
+ }
+
+ /* This handles the supplied packet [if there is one] */
+ if (p_curp) {
+ list_add_tail(&p_curp->list_entry, &p_rx->active_descq);
+ b_notify_fw = true;
+ }
+
+ if (!b_notify_fw)
+ return;
+
+ bd_prod = qed_chain_get_prod_idx(&p_rx->rxq_chain);
+ cq_prod = qed_chain_get_prod_idx(&p_rx->rcq_chain);
+ rx_prod.bd_prod = cpu_to_le16(bd_prod);
+ rx_prod.cqe_prod = cpu_to_le16(cq_prod);
+ DIRECT_REG_WR(p_rx->set_prod_addr, *((u32 *)&rx_prod));
+}
+
+int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr,
+ u16 buf_len, void *cookie, u8 notify_fw)
+{
+ struct core_rx_bd_with_buff_len *p_curb = NULL;
+ struct qed_ll2_rx_packet *p_curp = NULL;
+ struct qed_ll2_info *p_ll2_conn;
+ struct qed_ll2_rx_queue *p_rx;
+ unsigned long flags;
+ void *p_data;
+ int rc = 0;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+ p_rx = &p_ll2_conn->rx_queue;
+
+ spin_lock_irqsave(&p_rx->lock, flags);
+ if (!list_empty(&p_rx->free_descq))
+ p_curp = list_first_entry(&p_rx->free_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (p_curp) {
+ if (qed_chain_get_elem_left(&p_rx->rxq_chain) &&
+ qed_chain_get_elem_left(&p_rx->rcq_chain)) {
+ p_data = qed_chain_produce(&p_rx->rxq_chain);
+ p_curb = (struct core_rx_bd_with_buff_len *)p_data;
+ qed_chain_produce(&p_rx->rcq_chain);
+ }
+ }
+
+ /* If we're lacking entires, let's try to flush buffers to FW */
+ if (!p_curp || !p_curb) {
+ rc = -EBUSY;
+ p_curp = NULL;
+ goto out_notify;
+ }
+
+ /* We have an Rx packet we can fill */
+ DMA_REGPAIR_LE(p_curb->addr, addr);
+ p_curb->buff_length = cpu_to_le16(buf_len);
+ p_curp->rx_buf_addr = addr;
+ p_curp->cookie = cookie;
+ p_curp->rxq_bd = p_curb;
+ p_curp->buf_length = buf_len;
+ list_del(&p_curp->list_entry);
+
+ /* Check if we only want to enqueue this packet without informing FW */
+ if (!notify_fw) {
+ list_add_tail(&p_curp->list_entry, &p_rx->posting_descq);
+ goto out;
+ }
+
+out_notify:
+ qed_ll2_post_rx_buffer_notify_fw(p_hwfn, p_rx, p_curp);
+out:
+ spin_unlock_irqrestore(&p_rx->lock, flags);
+ return rc;
+}
+
+static void qed_ll2_prepare_tx_packet_set(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_tx_queue *p_tx,
+ struct qed_ll2_tx_packet *p_curp,
+ u8 num_of_bds,
+ dma_addr_t first_frag,
+ u16 first_frag_len, void *p_cookie,
+ u8 notify_fw)
+{
+ list_del(&p_curp->list_entry);
+ p_curp->cookie = p_cookie;
+ p_curp->bd_used = num_of_bds;
+ p_curp->notify_fw = notify_fw;
+ p_tx->cur_send_packet = p_curp;
+ p_tx->cur_send_frag_num = 0;
+
+ p_curp->bds_set[p_tx->cur_send_frag_num].tx_frag = first_frag;
+ p_curp->bds_set[p_tx->cur_send_frag_num].frag_len = first_frag_len;
+ p_tx->cur_send_frag_num++;
+}
+
+static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2,
+ struct qed_ll2_tx_packet *p_curp,
+ u8 num_of_bds,
+ enum core_tx_dest tx_dest,
+ u16 vlan,
+ u8 bd_flags,
+ u16 l4_hdr_offset_w,
+ dma_addr_t first_frag,
+ u16 first_frag_len)
+{
+ struct qed_chain *p_tx_chain = &p_ll2->tx_queue.txq_chain;
+ u16 prod_idx = qed_chain_get_prod_idx(p_tx_chain);
+ struct core_tx_bd *start_bd = NULL;
+ u16 frag_idx;
+
+ start_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
+ start_bd->nw_vlan_or_lb_echo = cpu_to_le16(vlan);
+ SET_FIELD(start_bd->bitfield1, CORE_TX_BD_L4_HDR_OFFSET_W,
+ cpu_to_le16(l4_hdr_offset_w));
+ SET_FIELD(start_bd->bitfield1, CORE_TX_BD_TX_DST, tx_dest);
+ start_bd->bd_flags.as_bitfield = bd_flags;
+ start_bd->bd_flags.as_bitfield |= CORE_TX_BD_FLAGS_START_BD_MASK <<
+ CORE_TX_BD_FLAGS_START_BD_SHIFT;
+ SET_FIELD(start_bd->bitfield0, CORE_TX_BD_NBDS, num_of_bds);
+ DMA_REGPAIR_LE(start_bd->addr, first_frag);
+ start_bd->nbytes = cpu_to_le16(first_frag_len);
+
+ DP_VERBOSE(p_hwfn,
+ (NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
+ "LL2 [q 0x%02x cid 0x%08x type 0x%08x] Tx Producer at [0x%04x] - set with a %04x bytes %02x BDs buffer at %08x:%08x\n",
+ p_ll2->queue_id,
+ p_ll2->cid,
+ p_ll2->conn_type,
+ prod_idx,
+ first_frag_len,
+ num_of_bds,
+ le32_to_cpu(start_bd->addr.hi),
+ le32_to_cpu(start_bd->addr.lo));
+
+ if (p_ll2->tx_queue.cur_send_frag_num == num_of_bds)
+ return;
+
+ /* Need to provide the packet with additional BDs for frags */
+ for (frag_idx = p_ll2->tx_queue.cur_send_frag_num;
+ frag_idx < num_of_bds; frag_idx++) {
+ struct core_tx_bd **p_bd = &p_curp->bds_set[frag_idx].txq_bd;
+
+ *p_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
+ (*p_bd)->bd_flags.as_bitfield = 0;
+ (*p_bd)->bitfield1 = 0;
+ (*p_bd)->bitfield0 = 0;
+ p_curp->bds_set[frag_idx].tx_frag = 0;
+ p_curp->bds_set[frag_idx].frag_len = 0;
+ }
+}
+
+/* This should be called while the Txq spinlock is being held */
+static void qed_ll2_tx_packet_notify(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ bool b_notify = p_ll2_conn->tx_queue.cur_send_packet->notify_fw;
+ struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+ struct qed_ll2_tx_packet *p_pkt = NULL;
+ struct core_db_data db_msg = { 0, 0, 0 };
+ u16 bd_prod;
+
+ /* If there are missing BDs, don't do anything now */
+ if (p_ll2_conn->tx_queue.cur_send_frag_num !=
+ p_ll2_conn->tx_queue.cur_send_packet->bd_used)
+ return;
+
+ /* Push the current packet to the list and clean after it */
+ list_add_tail(&p_ll2_conn->tx_queue.cur_send_packet->list_entry,
+ &p_ll2_conn->tx_queue.sending_descq);
+ p_ll2_conn->tx_queue.cur_send_packet = NULL;
+ p_ll2_conn->tx_queue.cur_send_frag_num = 0;
+
+ /* Notify FW of packet only if requested to */
+ if (!b_notify)
+ return;
+
+ bd_prod = qed_chain_get_prod_idx(&p_ll2_conn->tx_queue.txq_chain);
+
+ while (!list_empty(&p_tx->sending_descq)) {
+ p_pkt = list_first_entry(&p_tx->sending_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (!p_pkt)
+ break;
+
+ list_del(&p_pkt->list_entry);
+ list_add_tail(&p_pkt->list_entry, &p_tx->active_descq);
+ }
+
+ SET_FIELD(db_msg.params, CORE_DB_DATA_DEST, DB_DEST_XCM);
+ SET_FIELD(db_msg.params, CORE_DB_DATA_AGG_CMD, DB_AGG_CMD_SET);
+ SET_FIELD(db_msg.params, CORE_DB_DATA_AGG_VAL_SEL,
+ DQ_XCM_CORE_TX_BD_PROD_CMD);
+ db_msg.agg_flags = DQ_XCM_CORE_DQ_CF_CMD;
+ db_msg.spq_prod = cpu_to_le16(bd_prod);
+
+ /* Make sure the BDs data is updated before ringing the doorbell */
+ wmb();
+
+ DIRECT_REG_WR(p_tx->doorbell_addr, *((u32 *)&db_msg));
+
+ DP_VERBOSE(p_hwfn,
+ (NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
+ "LL2 [q 0x%02x cid 0x%08x type 0x%08x] Doorbelled [producer 0x%04x]\n",
+ p_ll2_conn->queue_id,
+ p_ll2_conn->cid, p_ll2_conn->conn_type, db_msg.spq_prod);
+}
+
+int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ u8 num_of_bds,
+ u16 vlan,
+ u8 bd_flags,
+ u16 l4_hdr_offset_w,
+ dma_addr_t first_frag,
+ u16 first_frag_len, void *cookie, u8 notify_fw)
+{
+ struct qed_ll2_tx_packet *p_curp = NULL;
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ struct qed_ll2_tx_queue *p_tx;
+ struct qed_chain *p_tx_chain;
+ unsigned long flags;
+ int rc = 0;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+ p_tx = &p_ll2_conn->tx_queue;
+ p_tx_chain = &p_tx->txq_chain;
+
+ if (num_of_bds > CORE_LL2_TX_MAX_BDS_PER_PACKET)
+ return -EIO;
+
+ spin_lock_irqsave(&p_tx->lock, flags);
+ if (p_tx->cur_send_packet) {
+ rc = -EEXIST;
+ goto out;
+ }
+
+ /* Get entry, but only if we have tx elements for it */
+ if (!list_empty(&p_tx->free_descq))
+ p_curp = list_first_entry(&p_tx->free_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (p_curp && qed_chain_get_elem_left(p_tx_chain) < num_of_bds)
+ p_curp = NULL;
+
+ if (!p_curp) {
+ rc = -EBUSY;
+ goto out;
+ }
+
+ /* Prepare packet and BD, and perhaps send a doorbell to FW */
+ qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp,
+ num_of_bds, first_frag,
+ first_frag_len, cookie, notify_fw);
+ qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp,
+ num_of_bds, CORE_TX_DEST_NW,
+ vlan, bd_flags, l4_hdr_offset_w,
+ first_frag, first_frag_len);
+
+ qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
+
+out:
+ spin_unlock_irqrestore(&p_tx->lock, flags);
+ return rc;
+}
+
+int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr, u16 nbytes)
+{
+ struct qed_ll2_tx_packet *p_cur_send_packet = NULL;
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ u16 cur_send_frag_num = 0;
+ struct core_tx_bd *p_bd;
+ unsigned long flags;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+
+ if (!p_ll2_conn->tx_queue.cur_send_packet)
+ return -EINVAL;
+
+ p_cur_send_packet = p_ll2_conn->tx_queue.cur_send_packet;
+ cur_send_frag_num = p_ll2_conn->tx_queue.cur_send_frag_num;
+
+ if (cur_send_frag_num >= p_cur_send_packet->bd_used)
+ return -EINVAL;
+
+ /* Fill the BD information, and possibly notify FW */
+ p_bd = p_cur_send_packet->bds_set[cur_send_frag_num].txq_bd;
+ DMA_REGPAIR_LE(p_bd->addr, addr);
+ p_bd->nbytes = cpu_to_le16(nbytes);
+ p_cur_send_packet->bds_set[cur_send_frag_num].tx_frag = addr;
+ p_cur_send_packet->bds_set[cur_send_frag_num].frag_len = nbytes;
+
+ p_ll2_conn->tx_queue.cur_send_frag_num++;
+
+ spin_lock_irqsave(&p_ll2_conn->tx_queue.lock, flags);
+ qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
+ spin_unlock_irqrestore(&p_ll2_conn->tx_queue.lock, flags);
+
+ return 0;
+}
+
+int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ int rc = -EINVAL;
+
+ p_ll2_conn = qed_ll2_handle_sanity_lock(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+
+ /* Stop Tx & Rx of connection, if needed */
+ if (QED_LL2_TX_REGISTERED(p_ll2_conn)) {
+ rc = qed_sp_ll2_tx_queue_stop(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+ qed_ll2_txq_flush(p_hwfn, connection_handle);
+ }
+
+ if (QED_LL2_RX_REGISTERED(p_ll2_conn)) {
+ rc = qed_sp_ll2_rx_queue_stop(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+ qed_ll2_rxq_flush(p_hwfn, connection_handle);
+ }
+
+ return rc;
+}
+
+void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return;
+
+ if (QED_LL2_RX_REGISTERED(p_ll2_conn)) {
+ p_ll2_conn->rx_queue.b_cb_registred = false;
+ qed_int_unregister_cb(p_hwfn, p_ll2_conn->rx_queue.rx_sb_index);
+ }
+
+ if (QED_LL2_TX_REGISTERED(p_ll2_conn)) {
+ p_ll2_conn->tx_queue.b_cb_registred = false;
+ qed_int_unregister_cb(p_hwfn, p_ll2_conn->tx_queue.tx_sb_index);
+ }
+
+ kfree(p_ll2_conn->tx_queue.descq_array);
+ qed_chain_free(p_hwfn->cdev, &p_ll2_conn->tx_queue.txq_chain);
+
+ kfree(p_ll2_conn->rx_queue.descq_array);
+ qed_chain_free(p_hwfn->cdev, &p_ll2_conn->rx_queue.rxq_chain);
+ qed_chain_free(p_hwfn->cdev, &p_ll2_conn->rx_queue.rcq_chain);
+
+ qed_cxt_release_cid(p_hwfn, p_ll2_conn->cid);
+
+ mutex_lock(&p_ll2_conn->mutex);
+ p_ll2_conn->b_active = false;
+ mutex_unlock(&p_ll2_conn->mutex);
+}
+
+struct qed_ll2_info *qed_ll2_alloc(struct qed_hwfn *p_hwfn)
+{
+ struct qed_ll2_info *p_ll2_connections;
+ u8 i;
+
+ /* Allocate LL2's set struct */
+ p_ll2_connections = kcalloc(QED_MAX_NUM_OF_LL2_CONNECTIONS,
+ sizeof(struct qed_ll2_info), GFP_KERNEL);
+ if (!p_ll2_connections) {
+ DP_NOTICE(p_hwfn, "Failed to allocate `struct qed_ll2'\n");
+ return NULL;
+ }
+
+ for (i = 0; i < QED_MAX_NUM_OF_LL2_CONNECTIONS; i++)
+ p_ll2_connections[i].my_id = i;
+
+ return p_ll2_connections;
+}
+
+void qed_ll2_setup(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections)
+{
+ int i;
+
+ for (i = 0; i < QED_MAX_NUM_OF_LL2_CONNECTIONS; i++)
+ mutex_init(&p_ll2_connections[i].mutex);
+}
+
+void qed_ll2_free(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections)
+{
+ kfree(p_ll2_connections);
+}
+
+static void _qed_ll2_get_tstats(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_ll2_info *p_ll2_conn,
+ struct qed_ll2_stats *p_stats)
+{
+ struct core_ll2_tstorm_per_queue_stat tstats;
+ u8 qid = p_ll2_conn->queue_id;
+ u32 tstats_addr;
+
+ memset(&tstats, 0, sizeof(tstats));
+ tstats_addr = BAR0_MAP_REG_TSDM_RAM +
+ CORE_LL2_TSTORM_PER_QUEUE_STAT_OFFSET(qid);
+ qed_memcpy_from(p_hwfn, p_ptt, &tstats, tstats_addr, sizeof(tstats));
+
+ p_stats->packet_too_big_discard =
+ HILO_64_REGPAIR(tstats.packet_too_big_discard);
+ p_stats->no_buff_discard = HILO_64_REGPAIR(tstats.no_buff_discard);
+}
+
+static void _qed_ll2_get_ustats(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_ll2_info *p_ll2_conn,
+ struct qed_ll2_stats *p_stats)
+{
+ struct core_ll2_ustorm_per_queue_stat ustats;
+ u8 qid = p_ll2_conn->queue_id;
+ u32 ustats_addr;
+
+ memset(&ustats, 0, sizeof(ustats));
+ ustats_addr = BAR0_MAP_REG_USDM_RAM +
+ CORE_LL2_USTORM_PER_QUEUE_STAT_OFFSET(qid);
+ qed_memcpy_from(p_hwfn, p_ptt, &ustats, ustats_addr, sizeof(ustats));
+
+ p_stats->rcv_ucast_bytes = HILO_64_REGPAIR(ustats.rcv_ucast_bytes);
+ p_stats->rcv_mcast_bytes = HILO_64_REGPAIR(ustats.rcv_mcast_bytes);
+ p_stats->rcv_bcast_bytes = HILO_64_REGPAIR(ustats.rcv_bcast_bytes);
+ p_stats->rcv_ucast_pkts = HILO_64_REGPAIR(ustats.rcv_ucast_pkts);
+ p_stats->rcv_mcast_pkts = HILO_64_REGPAIR(ustats.rcv_mcast_pkts);
+ p_stats->rcv_bcast_pkts = HILO_64_REGPAIR(ustats.rcv_bcast_pkts);
+}
+
+static void _qed_ll2_get_pstats(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_ll2_info *p_ll2_conn,
+ struct qed_ll2_stats *p_stats)
+{
+ struct core_ll2_pstorm_per_queue_stat pstats;
+ u8 stats_id = p_ll2_conn->tx_stats_id;
+ u32 pstats_addr;
+
+ memset(&pstats, 0, sizeof(pstats));
+ pstats_addr = BAR0_MAP_REG_PSDM_RAM +
+ CORE_LL2_PSTORM_PER_QUEUE_STAT_OFFSET(stats_id);
+ qed_memcpy_from(p_hwfn, p_ptt, &pstats, pstats_addr, sizeof(pstats));
+
+ p_stats->sent_ucast_bytes = HILO_64_REGPAIR(pstats.sent_ucast_bytes);
+ p_stats->sent_mcast_bytes = HILO_64_REGPAIR(pstats.sent_mcast_bytes);
+ p_stats->sent_bcast_bytes = HILO_64_REGPAIR(pstats.sent_bcast_bytes);
+ p_stats->sent_ucast_pkts = HILO_64_REGPAIR(pstats.sent_ucast_pkts);
+ p_stats->sent_mcast_pkts = HILO_64_REGPAIR(pstats.sent_mcast_pkts);
+ p_stats->sent_bcast_pkts = HILO_64_REGPAIR(pstats.sent_bcast_pkts);
+}
+
+int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+ u8 connection_handle, struct qed_ll2_stats *p_stats)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ struct qed_ptt *p_ptt;
+
+ memset(p_stats, 0, sizeof(*p_stats));
+
+ if ((connection_handle >= QED_MAX_NUM_OF_LL2_CONNECTIONS) ||
+ !p_hwfn->p_ll2_info)
+ return -EINVAL;
+
+ p_ll2_conn = &p_hwfn->p_ll2_info[connection_handle];
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ DP_ERR(p_hwfn, "Failed to acquire ptt\n");
+ return -EINVAL;
+ }
+
+ _qed_ll2_get_tstats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+ _qed_ll2_get_ustats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+ if (p_ll2_conn->tx_stats_en)
+ _qed_ll2_get_pstats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+
+ qed_ptt_release(p_hwfn, p_ptt);
+ return 0;
+}
+
+static void qed_ll2_register_cb_ops(struct qed_dev *cdev,
+ const struct qed_ll2_cb_ops *ops,
+ void *cookie)
+{
+ cdev->ll2->cbs = ops;
+ cdev->ll2->cb_cookie = cookie;
+}
+
+static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
+{
+ struct qed_ll2_info ll2_info;
+ struct qed_ll2_buffer *buffer;
+ enum qed_ll2_conn_type conn_type;
+ struct qed_ptt *p_ptt;
+ int rc, i;
+
+ /* Initialize LL2 locks & lists */
+ INIT_LIST_HEAD(&cdev->ll2->list);
+ spin_lock_init(&cdev->ll2->lock);
+ cdev->ll2->rx_size = NET_SKB_PAD + ETH_HLEN +
+ L1_CACHE_BYTES + params->mtu;
+ cdev->ll2->frags_mapped = params->frags_mapped;
+
+ /*Allocate memory for LL2 */
+ DP_INFO(cdev, "Allocating LL2 buffers of size %08x bytes\n",
+ cdev->ll2->rx_size);
+ for (i = 0; i < QED_LL2_RX_SIZE; i++) {
+ buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
+ if (!buffer) {
+ DP_INFO(cdev, "Failed to allocate LL2 buffers\n");
+ goto fail;
+ }
+
+ rc = qed_ll2_alloc_buffer(cdev, (u8 **)&buffer->data,
+ &buffer->phys_addr);
+ if (rc) {
+ kfree(buffer);
+ goto fail;
+ }
+
+ list_add_tail(&buffer->list, &cdev->ll2->list);
+ }
+
+ switch (QED_LEADING_HWFN(cdev)->hw_info.personality) {
+ case QED_PCI_ISCSI:
+ conn_type = QED_LL2_TYPE_ISCSI;
+ break;
+ case QED_PCI_ETH_ROCE:
+ conn_type = QED_LL2_TYPE_ROCE;
+ break;
+ default:
+ conn_type = QED_LL2_TYPE_TEST;
+ }
+
+ /* Prepare the temporary ll2 information */
+ memset(&ll2_info, 0, sizeof(ll2_info));
+ ll2_info.conn_type = conn_type;
+ ll2_info.mtu = params->mtu;
+ ll2_info.rx_drop_ttl0_flg = params->drop_ttl0_packets;
+ ll2_info.rx_vlan_removal_en = params->rx_vlan_stripping;
+ ll2_info.tx_tc = 0;
+ ll2_info.tx_dest = CORE_TX_DEST_NW;
+
+ rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_info,
+ QED_LL2_RX_SIZE, QED_LL2_TX_SIZE,
+ &cdev->ll2->handle);
+ if (rc) {
+ DP_INFO(cdev, "Failed to acquire LL2 connection\n");
+ goto fail;
+ }
+
+ rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle);
+ if (rc) {
+ DP_INFO(cdev, "Failed to establish LL2 connection\n");
+ goto release_fail;
+ }
+
+ /* Post all Rx buffers to FW */
+ spin_lock_bh(&cdev->ll2->lock);
+ list_for_each_entry(buffer, &cdev->ll2->list, list) {
+ rc = qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle,
+ buffer->phys_addr, 0, buffer, 1);
+ if (rc) {
+ DP_INFO(cdev,
+ "Failed to post an Rx buffer; Deleting it\n");
+ dma_unmap_single(&cdev->pdev->dev, buffer->phys_addr,
+ cdev->ll2->rx_size, DMA_FROM_DEVICE);
+ kfree(buffer->data);
+ list_del(&buffer->list);
+ kfree(buffer);
+ } else {
+ cdev->ll2->rx_cnt++;
+ }
+ }
+ spin_unlock_bh(&cdev->ll2->lock);
+
+ if (!cdev->ll2->rx_cnt) {
+ DP_INFO(cdev, "Failed passing even a single Rx buffer\n");
+ goto release_terminate;
+ }
+
+ if (!is_valid_ether_addr(params->ll2_mac_address)) {
+ DP_INFO(cdev, "Invalid Ethernet address\n");
+ goto release_terminate;
+ }
+
+ p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ if (!p_ptt) {
+ DP_INFO(cdev, "Failed to acquire PTT\n");
+ goto release_terminate;
+ }
+
+ rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ params->ll2_mac_address);
+ qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+ if (rc) {
+ DP_ERR(cdev, "Failed to allocate LLH filter\n");
+ goto release_terminate_all;
+ }
+
+ ether_addr_copy(cdev->ll2_mac_address, params->ll2_mac_address);
+
+ return 0;
+
+release_terminate_all:
+
+release_terminate:
+ qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+release_fail:
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+fail:
+ qed_ll2_kill_buffers(cdev);
+ cdev->ll2->handle = QED_LL2_UNUSED_HANDLE;
+ return -EINVAL;
+}
+
+static int qed_ll2_stop(struct qed_dev *cdev)
+{
+ struct qed_ptt *p_ptt;
+ int rc;
+
+ if (cdev->ll2->handle == QED_LL2_UNUSED_HANDLE)
+ return 0;
+
+ p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ if (!p_ptt) {
+ DP_INFO(cdev, "Failed to acquire PTT\n");
+ goto fail;
+ }
+
+ qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ cdev->ll2_mac_address);
+ qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+ eth_zero_addr(cdev->ll2_mac_address);
+
+ rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle);
+ if (rc)
+ DP_INFO(cdev, "Failed to terminate LL2 connection\n");
+
+ qed_ll2_kill_buffers(cdev);
+
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+ cdev->ll2->handle = QED_LL2_UNUSED_HANDLE;
+
+ return rc;
+fail:
+ return -EINVAL;
+}
+
+static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
+{
+ const skb_frag_t *frag;
+ int rc = -EINVAL, i;
+ dma_addr_t mapping;
+ u16 vlan = 0;
+ u8 flags = 0;
+
+ if (unlikely(skb->ip_summed != CHECKSUM_NONE)) {
+ DP_INFO(cdev, "Cannot transmit a checksumed packet\n");
+ return -EINVAL;
+ }
+
+ if (1 + skb_shinfo(skb)->nr_frags > CORE_LL2_TX_MAX_BDS_PER_PACKET) {
+ DP_ERR(cdev, "Cannot transmit a packet with %d fragments\n",
+ 1 + skb_shinfo(skb)->nr_frags);
+ return -EINVAL;
+ }
+
+ mapping = dma_map_single(&cdev->pdev->dev, skb->data,
+ skb->len, DMA_TO_DEVICE);
+ if (unlikely(dma_mapping_error(&cdev->pdev->dev, mapping))) {
+ DP_NOTICE(cdev, "SKB mapping failed\n");
+ return -EINVAL;
+ }
+
+ /* Request HW to calculate IP csum */
+ if (!((vlan_get_protocol(skb) == htons(ETH_P_IPV6)) &&
+ ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
+ flags |= BIT(CORE_TX_BD_FLAGS_IP_CSUM_SHIFT);
+
+ if (skb_vlan_tag_present(skb)) {
+ vlan = skb_vlan_tag_get(skb);
+ flags |= BIT(CORE_TX_BD_FLAGS_VLAN_INSERTION_SHIFT);
+ }
+
+ rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle,
+ 1 + skb_shinfo(skb)->nr_frags,
+ vlan, flags, 0, mapping,
+ skb->len, skb, 1);
+ if (rc)
+ goto err;
+
+ for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+ frag = &skb_shinfo(skb)->frags[i];
+ if (!cdev->ll2->frags_mapped) {
+ mapping = skb_frag_dma_map(&cdev->pdev->dev, frag, 0,
+ skb_frag_size(frag),
+ DMA_TO_DEVICE);
+
+ if (unlikely(dma_mapping_error(&cdev->pdev->dev,
+ mapping))) {
+ DP_NOTICE(cdev,
+ "Unable to map frag - dropping packet\n");
+ goto err;
+ }
+ } else {
+ mapping = page_to_phys(skb_frag_page(frag)) |
+ frag->page_offset;
+ }
+
+ rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle,
+ mapping,
+ skb_frag_size(frag));
+
+ /* if failed not much to do here, partial packet has been posted
+ * we can't free memory, will need to wait for completion.
+ */
+ if (rc)
+ goto err2;
+ }
+
+ return 0;
+
+err:
+ dma_unmap_single(&cdev->pdev->dev, mapping, skb->len, DMA_TO_DEVICE);
+
+err2:
+ return rc;
+}
+
+static int qed_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
+{
+ if (!cdev->ll2)
+ return -EINVAL;
+
+ return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle, stats);
+}
+
+const struct qed_ll2_ops qed_ll2_ops_pass = {
+ .start = &qed_ll2_start,
+ .stop = &qed_ll2_stop,
+ .start_xmit = &qed_ll2_start_xmit,
+ .register_cb_ops = &qed_ll2_register_cb_ops,
+ .get_stats = &qed_ll2_stats,
+};
+
+int qed_ll2_alloc_if(struct qed_dev *cdev)
+{
+ cdev->ll2 = kzalloc(sizeof(*cdev->ll2), GFP_KERNEL);
+ return cdev->ll2 ? 0 : -ENOMEM;
+}
+
+void qed_ll2_dealloc_if(struct qed_dev *cdev)
+{
+ kfree(cdev->ll2);
+ cdev->ll2 = NULL;
+}
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
new file mode 100644
index 0000000..a037c48
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -0,0 +1,289 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#ifndef _QED_LL2_H
+#define _QED_LL2_H
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_ll2_if.h>
+#include "qed.h"
+#include "qed_hsi.h"
+#include "qed_sp.h"
+
+#define QED_MAX_NUM_OF_LL2_CONNECTIONS (4)
+
+enum qed_ll2_conn_type {
+ QED_LL2_TYPE_RESERVED,
+ QED_LL2_TYPE_ISCSI,
+ QED_LL2_TYPE_TEST,
+ QED_LL2_TYPE_ISCSI_OOO,
+ QED_LL2_TYPE_RESERVED2,
+ QED_LL2_TYPE_ROCE,
+ QED_LL2_TYPE_RESERVED3,
+ MAX_QED_LL2_RX_CONN_TYPE
+};
+
+struct qed_ll2_rx_packet {
+ struct list_head list_entry;
+ struct core_rx_bd_with_buff_len *rxq_bd;
+ dma_addr_t rx_buf_addr;
+ u16 buf_length;
+ void *cookie;
+ u8 placement_offset;
+ u16 parse_flags;
+ u16 packet_length;
+ u16 vlan;
+ u32 opaque_data[2];
+};
+
+struct qed_ll2_tx_packet {
+ struct list_head list_entry;
+ u16 bd_used;
+ u16 vlan;
+ u16 l4_hdr_offset_w;
+ u8 bd_flags;
+ bool notify_fw;
+ void *cookie;
+
+ struct {
+ struct core_tx_bd *txq_bd;
+ dma_addr_t tx_frag;
+ u16 frag_len;
+ } bds_set[ETH_TX_MAX_BDS_PER_NON_LSO_PACKET];
+};
+
+struct qed_ll2_rx_queue {
+ /* Lock protecting the Rx queue manipulation */
+ spinlock_t lock;
+ struct qed_chain rxq_chain;
+ struct qed_chain rcq_chain;
+ u8 rx_sb_index;
+ bool b_cb_registred;
+ __le16 *p_fw_cons;
+ struct list_head active_descq;
+ struct list_head free_descq;
+ struct list_head posting_descq;
+ struct qed_ll2_rx_packet *descq_array;
+ void __iomem *set_prod_addr;
+};
+
+struct qed_ll2_tx_queue {
+ /* Lock protecting the Tx queue manipulation */
+ spinlock_t lock;
+ struct qed_chain txq_chain;
+ u8 tx_sb_index;
+ bool b_cb_registred;
+ __le16 *p_fw_cons;
+ struct list_head active_descq;
+ struct list_head free_descq;
+ struct list_head sending_descq;
+ struct qed_ll2_tx_packet *descq_array;
+ struct qed_ll2_tx_packet *cur_send_packet;
+ struct qed_ll2_tx_packet cur_completing_packet;
+ u16 cur_completing_bd_idx;
+ void __iomem *doorbell_addr;
+ u16 bds_idx;
+ u16 cur_send_frag_num;
+ u16 cur_completing_frag_num;
+ bool b_completing_packet;
+};
+
+struct qed_ll2_info {
+ /* Lock protecting the state of LL2 */
+ struct mutex mutex;
+ enum qed_ll2_conn_type conn_type;
+ u32 cid;
+ u8 my_id;
+ u8 queue_id;
+ u8 tx_stats_id;
+ bool b_active;
+ u16 mtu;
+ u8 rx_drop_ttl0_flg;
+ u8 rx_vlan_removal_en;
+ u8 tx_tc;
+ enum core_tx_dest tx_dest;
+ enum core_error_handle ai_err_packet_too_big;
+ enum core_error_handle ai_err_no_buf;
+ u8 tx_stats_en;
+ struct qed_ll2_rx_queue rx_queue;
+ struct qed_ll2_tx_queue tx_queue;
+};
+
+/**
+ * @brief qed_ll2_acquire_connection - allocate resources,
+ * starts rx & tx (if relevant) queues pair. Provides
+ * connecion handler as output parameter.
+ *
+ * @param p_hwfn
+ * @param p_params Contain various configuration properties
+ * @param rx_num_desc
+ * @param tx_num_desc
+ *
+ * @param p_connection_handle Output container for LL2 connection's handle
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_params,
+ u16 rx_num_desc,
+ u16 tx_num_desc,
+ u8 *p_connection_handle);
+
+/**
+ * @brief qed_ll2_establish_connection - start previously
+ * allocated LL2 queues pair
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_post_rx_buffers - submit buffers to LL2 Rx queue.
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ * @param addr rx (physical address) buffers to submit
+ * @param cookie
+ * @param notify_fw produce corresponding Rx BD immediately
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr,
+ u16 buf_len, void *cookie, u8 notify_fw);
+
+/**
+ * @brief qed_ll2_prepare_tx_packet - request for start Tx BD
+ * to prepare Tx packet submission to FW.
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ * @param num_of_bds a number of requested BD equals a number of
+ * fragments in Tx packet
+ * @param vlan VLAN to insert to packet (if insertion set)
+ * @param bd_flags
+ * @param l4_hdr_offset_w L4 Header Offset from start of packet
+ * (in words). This is needed if both l4_csum
+ * and ipv6_ext are set
+ * @param first_frag
+ * @param first_frag_len
+ * @param cookie
+ *
+ * @param notify_fw
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ u8 num_of_bds,
+ u16 vlan,
+ u8 bd_flags,
+ u16 l4_hdr_offset_w,
+ dma_addr_t first_frag,
+ u16 first_frag_len, void *cookie, u8 notify_fw);
+
+/**
+ * @brief qed_ll2_release_connection - releases resources
+ * allocated for LL2 connection
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ */
+void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_set_fragment_of_tx_packet - provides fragments to fill
+ * Tx BD of BDs requested by
+ * qed_ll2_prepare_tx_packet
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle
+ * obtained from
+ * qed_ll2_require_connection
+ * @param addr
+ * @param nbytes
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr, u16 nbytes);
+
+/**
+ * @brief qed_ll2_terminate_connection - stops Tx/Rx queues
+ *
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle
+ * obtained from
+ * qed_ll2_require_connection
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_get_stats - get LL2 queue's statistics
+ *
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ * @param p_stats
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+ u8 connection_handle, struct qed_ll2_stats *p_stats);
+
+/**
+ * @brief qed_ll2_alloc - Allocates LL2 connections set
+ *
+ * @param p_hwfn
+ *
+ * @return pointer to alocated qed_ll2_info or NULL
+ */
+struct qed_ll2_info *qed_ll2_alloc(struct qed_hwfn *p_hwfn);
+
+/**
+ * @brief qed_ll2_setup - Inits LL2 connections set
+ *
+ * @param p_hwfn
+ * @param p_ll2_connections
+ *
+ */
+void qed_ll2_setup(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections);
+
+/**
+ * @brief qed_ll2_free - Releases LL2 connections set
+ *
+ * @param p_hwfn
+ * @param p_ll2_connections
+ *
+ */
+void qed_ll2_free(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections);
+
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_main.c b/drivers/net/ethernet/qlogic/qed/qed_main.c
index 32f71ee..0225382 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_main.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_main.c
@@ -22,11 +22,13 @@
#include <linux/etherdevice.h>
#include <linux/vmalloc.h>
#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_ll2_if.h>
#include "qed.h"
#include "qed_sriov.h"
#include "qed_sp.h"
#include "qed_dev_api.h"
+#include "qed_ll2.h"
#include "qed_mcp.h"
#include "qed_hw.h"
#include "qed_selftest.h"
@@ -606,7 +608,16 @@ static int qed_nic_reset(struct qed_dev *cdev)
static int qed_nic_setup(struct qed_dev *cdev)
{
- int rc;
+ int rc, i;
+
+ /* Determine if interface is going to require LL2 */
+ if (QED_LEADING_HWFN(cdev)->hw_info.personality != QED_PCI_ETH) {
+ for (i = 0; i < cdev->num_hwfns; i++) {
+ struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
+
+ p_hwfn->using_ll2 = true;
+ }
+ }
rc = qed_resc_alloc(cdev);
if (rc)
@@ -871,6 +882,12 @@ static int qed_slowpath_start(struct qed_dev *cdev,
DP_INFO(cdev,
"HW initialization and function start completed successfully\n");
+ /* Allocate LL2 interface if needed */
+ if (QED_LEADING_HWFN(cdev)->using_ll2) {
+ rc = qed_ll2_alloc_if(cdev);
+ if (rc)
+ goto err3;
+ }
if (IS_PF(cdev)) {
hwfn = QED_LEADING_HWFN(cdev);
drv_version.version = (params->drv_major << 24) |
@@ -891,6 +908,8 @@ static int qed_slowpath_start(struct qed_dev *cdev,
return 0;
+err3:
+ qed_hw_stop(cdev);
err2:
qed_hw_timers_stop_all(cdev);
if (IS_PF(cdev))
@@ -913,6 +932,8 @@ static int qed_slowpath_stop(struct qed_dev *cdev)
if (!cdev)
return -ENODEV;
+ qed_ll2_dealloc_if(cdev);
+
if (IS_PF(cdev)) {
qed_free_stream_mem(cdev);
if (IS_QED_ETH_IF(cdev))
diff --git a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
index b49d47f..103a32c 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
@@ -208,6 +208,26 @@
0x50196cUL
#define NIG_REG_LLH_CLS_TYPE_DUALMODE \
0x501964UL
+#define NIG_REG_LLH_FUNC_FILTER_VALUE \
+ 0x501a00UL
+#define NIG_REG_LLH_FUNC_FILTER_VALUE_SIZE \
+ 32
+#define NIG_REG_LLH_FUNC_FILTER_EN \
+ 0x501a80UL
+#define NIG_REG_LLH_FUNC_FILTER_EN_SIZE \
+ 16
+#define NIG_REG_LLH_FUNC_FILTER_MODE \
+ 0x501ac0UL
+#define NIG_REG_LLH_FUNC_FILTER_MODE_SIZE \
+ 16
+#define NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE \
+ 0x501b00UL
+#define NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE_SIZE \
+ 16
+#define NIG_REG_LLH_FUNC_FILTER_HDR_SEL \
+ 0x501b40UL
+#define NIG_REG_LLH_FUNC_FILTER_HDR_SEL_SIZE \
+ 16
#define NCSI_REG_CONFIG \
0x040200UL
#define PBF_REG_INIT \
@@ -264,6 +284,8 @@
0x1f0a1cUL
#define PRS_REG_ROCE_DEST_QP_MAX_PF \
0x1f0430UL
+#define PRS_REG_USE_LIGHT_L2 \
+ 0x1f096cUL
#define PSDM_REG_ENABLE_IN1 \
0xfa0004UL
#define PSEM_REG_ENABLE_IN \
diff --git a/drivers/net/ethernet/qlogic/qed/qed_sp.h b/drivers/net/ethernet/qlogic/qed/qed_sp.h
index a548504..a3c539f 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_sp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_sp.h
@@ -61,6 +61,10 @@ union ramrod_data {
struct vport_start_ramrod_data vport_start;
struct vport_stop_ramrod_data vport_stop;
struct vport_update_ramrod_data vport_update;
+ struct core_rx_start_ramrod_data core_rx_queue_start;
+ struct core_rx_stop_ramrod_data core_rx_queue_stop;
+ struct core_tx_start_ramrod_data core_tx_queue_start;
+ struct core_tx_stop_ramrod_data core_tx_queue_stop;
struct vport_filter_update_ramrod_data vport_filter_update;
struct rdma_init_func_ramrod_data rdma_init_func;
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index d8dc5c2..6611441 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -623,6 +623,7 @@ enum DP_MODULE {
QED_MSG_SP = 0x100000,
QED_MSG_STORAGE = 0x200000,
QED_MSG_CXT = 0x800000,
+ QED_MSG_LL2 = 0x1000000,
QED_MSG_ILT = 0x2000000,
QED_MSG_ROCE = 0x4000000,
QED_MSG_DEBUG = 0x8000000,
diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h
new file mode 100644
index 0000000..0164cca
--- /dev/null
+++ b/include/linux/qed/qed_ll2_if.h
@@ -0,0 +1,140 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#ifndef _QED_LL2_IF_H
+#define _QED_LL2_IF_H
+
+#include <linux/types.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/pci.h>
+#include <linux/skbuff.h>
+#include <linux/types.h>
+#include <linux/version.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/qed/qed_if.h>
+
+struct qed_ll2_stats {
+ u64 gsi_invalid_hdr;
+ u64 gsi_invalid_pkt_length;
+ u64 gsi_unsupported_pkt_typ;
+ u64 gsi_crcchksm_error;
+
+ u64 packet_too_big_discard;
+ u64 no_buff_discard;
+
+ u64 rcv_ucast_bytes;
+ u64 rcv_mcast_bytes;
+ u64 rcv_bcast_bytes;
+ u64 rcv_ucast_pkts;
+ u64 rcv_mcast_pkts;
+ u64 rcv_bcast_pkts;
+
+ u64 sent_ucast_bytes;
+ u64 sent_mcast_bytes;
+ u64 sent_bcast_bytes;
+ u64 sent_ucast_pkts;
+ u64 sent_mcast_pkts;
+ u64 sent_bcast_pkts;
+};
+
+#define QED_LL2_UNUSED_HANDLE (0xff)
+
+struct qed_ll2_cb_ops {
+ int (*rx_cb)(void *, struct sk_buff *, u32, u32);
+ int (*tx_cb)(void *, struct sk_buff *, bool);
+};
+
+struct qed_ll2_params {
+ u16 mtu;
+ bool drop_ttl0_packets;
+ bool rx_vlan_stripping;
+ u8 tx_tc;
+ bool frags_mapped;
+ u8 ll2_mac_address[ETH_ALEN];
+};
+
+struct qed_ll2_ops {
+/**
+ * @brief start - initializes ll2
+ *
+ * @param cdev
+ * @param params - protocol driver configuration for the ll2.
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ int (*start)(struct qed_dev *cdev, struct qed_ll2_params *params);
+
+/**
+ * @brief stop - stops the ll2
+ *
+ * @param cdev
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ int (*stop)(struct qed_dev *cdev);
+
+/**
+ * @brief start_xmit - transmits an skb over the ll2 interface
+ *
+ * @param cdev
+ * @param skb
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ int (*start_xmit)(struct qed_dev *cdev, struct sk_buff *skb);
+
+/**
+ * @brief register_cb_ops - protocol driver register the callback for Rx/Tx
+ * packets. Should be called before `start'.
+ *
+ * @param cdev
+ * @param cookie - to be passed to the callback functions.
+ * @param ops - the callback functions to register for Rx / Tx.
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ void (*register_cb_ops)(struct qed_dev *cdev,
+ const struct qed_ll2_cb_ops *ops,
+ void *cookie);
+
+/**
+ * @brief get LL2 related statistics
+ *
+ * @param cdev
+ * @param stats - pointer to struct that would be filled with stats
+ *
+ * @return 0 on success, error otherwise.
+ */
+ int (*get_stats)(struct qed_dev *cdev, struct qed_ll2_stats *stats);
+};
+
+#ifdef CONFIG_QED_LL2
+int qed_ll2_alloc_if(struct qed_dev *);
+void qed_ll2_dealloc_if(struct qed_dev *);
+#else
+static const struct qed_ll2_ops qed_ll2_ops_pass = {
+ .start = NULL,
+ .stop = NULL,
+ .start_xmit = NULL,
+ .register_cb_ops = NULL,
+ .get_stats = NULL,
+};
+
+static inline int qed_ll2_alloc_if(struct qed_dev *cdev)
+{
+ return 0;
+}
+
+static inline void qed_ll2_dealloc_if(struct qed_dev *cdev)
+{
+}
+#endif
+#endif
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 01/11] qed: Add LL2
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
From: Yuval Mintz <Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
Add support for qed light L2 feature. It is used by the RDMA driver and for other purposes as well.
Signed-off-by: Yuval Mintz <Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
drivers/net/ethernet/qlogic/Kconfig | 8 +
drivers/net/ethernet/qlogic/qed/Makefile | 1 +
drivers/net/ethernet/qlogic/qed/qed.h | 8 +
drivers/net/ethernet/qlogic/qed/qed_cxt.c | 2 +
drivers/net/ethernet/qlogic/qed/qed_dev.c | 120 +-
drivers/net/ethernet/qlogic/qed/qed_dev_api.h | 20 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 1698 ++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 289 ++++
drivers/net/ethernet/qlogic/qed/qed_main.c | 23 +-
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 22 +
drivers/net/ethernet/qlogic/qed/qed_sp.h | 4 +
include/linux/qed/qed_if.h | 1 +
include/linux/qed/qed_ll2_if.h | 140 ++
13 files changed, 2333 insertions(+), 3 deletions(-)
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_ll2.h
create mode 100644 include/linux/qed/qed_ll2_if.h
diff --git a/drivers/net/ethernet/qlogic/Kconfig b/drivers/net/ethernet/qlogic/Kconfig
index 6ba4840..9eb3b19 100644
--- a/drivers/net/ethernet/qlogic/Kconfig
+++ b/drivers/net/ethernet/qlogic/Kconfig
@@ -88,6 +88,14 @@ config QED
---help---
This enables the support for ...
+config QED_LL2
+ bool "Qlogic QED Light L2 interface"
+ default n
+ depends on QED
+ ---help---
+ This enables support for Light L2 interface which is required
+ by all qed protocol drivers other than qede.
+
config QED_SRIOV
bool "QLogic QED 25/40/100Gb SR-IOV support"
depends on QED && PCI_IOV
diff --git a/drivers/net/ethernet/qlogic/qed/Makefile b/drivers/net/ethernet/qlogic/qed/Makefile
index d1f157e..02f4842 100644
--- a/drivers/net/ethernet/qlogic/qed/Makefile
+++ b/drivers/net/ethernet/qlogic/qed/Makefile
@@ -4,3 +4,4 @@ qed-y := qed_cxt.o qed_dev.o qed_hw.o qed_init_fw_funcs.o qed_init_ops.o \
qed_int.o qed_main.o qed_mcp.o qed_sp_commands.o qed_spq.o qed_l2.o \
qed_selftest.o qed_dcbx.o
qed-$(CONFIG_QED_SRIOV) += qed_sriov.o qed_vf.o
+qed-$(CONFIG_QED_LL2) += qed_ll2.o
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index 9ba7a7e..a42824e 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -71,6 +71,7 @@ struct qed_sb_info;
struct qed_sb_attn_info;
struct qed_cxt_mngr;
struct qed_sb_sp_info;
+struct qed_ll2_info;
struct qed_mcp_info;
struct qed_rt_data {
@@ -151,6 +152,7 @@ enum QED_RESOURCES {
QED_MAC,
QED_VLAN,
QED_ILT,
+ QED_LL2_QUEUE,
QED_MAX_RESC,
};
@@ -359,6 +361,8 @@ struct qed_hwfn {
struct qed_sb_attn_info *p_sb_attn;
/* Protocol related */
+ bool using_ll2;
+ struct qed_ll2_info *p_ll2_info;
struct qed_pf_params pf_params;
bool b_rdma_enabled_in_prs;
@@ -544,6 +548,10 @@ struct qed_dev {
} protocol_ops;
void *ops_cookie;
+#ifdef CONFIG_QED_LL2
+ struct qed_cb_ll2_info *ll2;
+ u8 ll2_mac_address[ETH_ALEN];
+#endif
const struct firmware *firmware;
};
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.c b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
index 5476927..34af48d 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
@@ -1849,6 +1849,8 @@ int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn)
/* Set the number of required CORE connections */
u32 core_cids = 1; /* SPQ */
+ if (p_hwfn->using_ll2)
+ core_cids += 4;
qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
switch (p_hwfn->hw_info.personality) {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 5ae27f2..4afcda5 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -29,6 +29,7 @@
#include "qed_hw.h"
#include "qed_init_ops.h"
#include "qed_int.h"
+#include "qed_ll2.h"
#include "qed_mcp.h"
#include "qed_reg_addr.h"
#include "qed_sp.h"
@@ -147,6 +148,9 @@ void qed_resc_free(struct qed_dev *cdev)
qed_eq_free(p_hwfn, p_hwfn->p_eq);
qed_consq_free(p_hwfn, p_hwfn->p_consq);
qed_int_free(p_hwfn);
+#ifdef CONFIG_QED_LL2
+ qed_ll2_free(p_hwfn, p_hwfn->p_ll2_info);
+#endif
qed_iov_free(p_hwfn);
qed_dmae_info_free(p_hwfn);
qed_dcbx_info_free(p_hwfn, p_hwfn->p_dcbx_info);
@@ -404,6 +408,9 @@ int qed_qm_reconf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
int qed_resc_alloc(struct qed_dev *cdev)
{
+#ifdef CONFIG_QED_LL2
+ struct qed_ll2_info *p_ll2_info;
+#endif
struct qed_consq *p_consq;
struct qed_eq *p_eq;
int i, rc = 0;
@@ -520,6 +527,15 @@ int qed_resc_alloc(struct qed_dev *cdev)
goto alloc_no_mem;
p_hwfn->p_consq = p_consq;
+#ifdef CONFIG_QED_LL2
+ if (p_hwfn->using_ll2) {
+ p_ll2_info = qed_ll2_alloc(p_hwfn);
+ if (!p_ll2_info)
+ goto alloc_no_mem;
+ p_hwfn->p_ll2_info = p_ll2_info;
+ }
+#endif
+
/* DMA info initialization */
rc = qed_dmae_info_alloc(p_hwfn);
if (rc) {
@@ -576,6 +592,10 @@ void qed_resc_setup(struct qed_dev *cdev)
qed_int_setup(p_hwfn, p_hwfn->p_main_ptt);
qed_iov_setup(p_hwfn, p_hwfn->p_main_ptt);
+#ifdef CONFIG_QED_LL2
+ if (p_hwfn->using_ll2)
+ qed_ll2_setup(p_hwfn, p_hwfn->p_ll2_info);
+#endif
}
}
@@ -1319,6 +1339,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
resc_num[QED_VLAN] = (ETH_NUM_VLAN_FILTERS - 1 /*For vlan0*/) /
num_funcs;
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
+ resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
@@ -1342,7 +1363,8 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
"RL = %d start = %d\n"
"MAC = %d start = %d\n"
"VLAN = %d start = %d\n"
- "ILT = %d start = %d\n",
+ "ILT = %d start = %d\n"
+ "LL2_QUEUE = %d start = %d\n",
p_hwfn->hw_info.resc_num[QED_SB],
p_hwfn->hw_info.resc_start[QED_SB],
p_hwfn->hw_info.resc_num[QED_L2_QUEUE],
@@ -1358,7 +1380,9 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
p_hwfn->hw_info.resc_num[QED_VLAN],
p_hwfn->hw_info.resc_start[QED_VLAN],
p_hwfn->hw_info.resc_num[QED_ILT],
- p_hwfn->hw_info.resc_start[QED_ILT]);
+ p_hwfn->hw_info.resc_start[QED_ILT],
+ RESC_NUM(p_hwfn, QED_LL2_QUEUE),
+ RESC_START(p_hwfn, QED_LL2_QUEUE));
return 0;
}
@@ -2163,6 +2187,98 @@ int qed_fw_rss_eng(struct qed_hwfn *p_hwfn, u8 src_id, u8 *dst_id)
return 0;
}
+static void qed_llh_mac_to_filter(u32 *p_high, u32 *p_low,
+ u8 *p_filter)
+{
+ *p_high = p_filter[1] | (p_filter[0] << 8);
+ *p_low = p_filter[5] | (p_filter[4] << 8) |
+ (p_filter[3] << 16) | (p_filter[2] << 24);
+}
+
+int qed_llh_add_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter)
+{
+ u32 high = 0, low = 0, en;
+ int i;
+
+ if (!(IS_MF_SI(p_hwfn) || IS_MF_DEFAULT(p_hwfn)))
+ return 0;
+
+ qed_llh_mac_to_filter(&high, &low, p_filter);
+
+ /* Find a free entry and utilize it */
+ for (i = 0; i < NIG_REG_LLH_FUNC_FILTER_EN_SIZE; i++) {
+ en = qed_rd(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32));
+ if (en)
+ continue;
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ 2 * i * sizeof(u32), low);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ (2 * i + 1) * sizeof(u32), high);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_MODE + i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE +
+ i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32), 1);
+ break;
+ }
+ if (i >= NIG_REG_LLH_FUNC_FILTER_EN_SIZE) {
+ DP_NOTICE(p_hwfn,
+ "Failed to find an empty LLH filter to utilize\n");
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
+ "MAC: %pM is added at %d\n",
+ p_filter, i);
+
+ return 0;
+}
+
+void qed_llh_remove_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter)
+{
+ u32 high = 0, low = 0;
+ int i;
+
+ if (!(IS_MF_SI(p_hwfn) || IS_MF_DEFAULT(p_hwfn)))
+ return;
+
+ qed_llh_mac_to_filter(&high, &low, p_filter);
+
+ /* Find the entry and clean it */
+ for (i = 0; i < NIG_REG_LLH_FUNC_FILTER_EN_SIZE; i++) {
+ if (qed_rd(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ 2 * i * sizeof(u32)) != low)
+ continue;
+ if (qed_rd(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ (2 * i + 1) * sizeof(u32)) != high)
+ continue;
+
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE + 2 * i * sizeof(u32), 0);
+ qed_wr(p_hwfn, p_ptt,
+ NIG_REG_LLH_FUNC_FILTER_VALUE +
+ (2 * i + 1) * sizeof(u32), 0);
+
+ DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
+ "MAC: %pM is removed from %d\n",
+ p_filter, i);
+ break;
+ }
+ if (i >= NIG_REG_LLH_FUNC_FILTER_EN_SIZE)
+ DP_NOTICE(p_hwfn, "Tried to remove a non-configured filter\n");
+}
+
static int qed_set_coalesce(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
u32 hw_addr, void *p_eth_qzone,
size_t eth_qzone_size, u8 timeset)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev_api.h b/drivers/net/ethernet/qlogic/qed/qed_dev_api.h
index 343bb03..2b4294b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev_api.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev_api.h
@@ -310,6 +310,26 @@ int qed_fw_rss_eng(struct qed_hwfn *p_hwfn,
u8 *dst_id);
/**
+ * @brief qed_llh_add_mac_filter - configures a MAC filter in llh
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param p_filter - MAC to add
+ */
+int qed_llh_add_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter);
+
+/**
+ * @brief qed_llh_remove_mac_filter - removes a MAC filtre from llh
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param p_filter - MAC to remove
+ */
+void qed_llh_remove_mac_filter(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u8 *p_filter);
+
+/**
* *@brief Cleanup of previous driver remains prior to load
*
* @param p_hwfn
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
new file mode 100644
index 0000000..1a96a2d
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -0,0 +1,1698 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#include <linux/types.h>
+#include <asm/byteorder.h>
+#include <linux/dma-mapping.h>
+#include <linux/if_vlan.h>
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/stddef.h>
+#include <linux/version.h>
+#include <linux/workqueue.h>
+#include <net/ipv6.h>
+#include <linux/bitops.h>
+#include <linux/delay.h>
+#include <linux/errno.h>
+#include <linux/etherdevice.h>
+#include <linux/io.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/qed/qed_ll2_if.h>
+#include "qed.h"
+#include "qed_cxt.h"
+#include "qed_dev_api.h"
+#include "qed_hsi.h"
+#include "qed_hw.h"
+#include "qed_int.h"
+#include "qed_ll2.h"
+#include "qed_mcp.h"
+#include "qed_reg_addr.h"
+#include "qed_sp.h"
+
+#define QED_LL2_RX_REGISTERED(ll2) ((ll2)->rx_queue.b_cb_registred)
+#define QED_LL2_TX_REGISTERED(ll2) ((ll2)->tx_queue.b_cb_registred)
+
+#define QED_LL2_TX_SIZE (256)
+#define QED_LL2_RX_SIZE (4096)
+
+struct qed_cb_ll2_info {
+ int rx_cnt;
+ u32 rx_size;
+ u8 handle;
+ bool frags_mapped;
+
+ /* Lock protecting LL2 buffer lists in sleepless context */
+ spinlock_t lock;
+ struct list_head list;
+
+ const struct qed_ll2_cb_ops *cbs;
+ void *cb_cookie;
+};
+
+struct qed_ll2_buffer {
+ struct list_head list;
+ void *data;
+ dma_addr_t phys_addr;
+};
+
+static void qed_ll2b_complete_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment,
+ bool b_last_packet)
+{
+ struct qed_dev *cdev = p_hwfn->cdev;
+ struct sk_buff *skb = cookie;
+
+ /* All we need to do is release the mapping */
+ dma_unmap_single(&p_hwfn->cdev->pdev->dev, first_frag_addr,
+ skb_headlen(skb), DMA_TO_DEVICE);
+
+ if (cdev->ll2->cbs && cdev->ll2->cbs->tx_cb)
+ cdev->ll2->cbs->tx_cb(cdev->ll2->cb_cookie, skb,
+ b_last_fragment);
+
+ if (cdev->ll2->frags_mapped)
+ /* Case where mapped frags were received, need to
+ * free skb with nr_frags marked as 0
+ */
+ skb_shinfo(skb)->nr_frags = 0;
+
+ dev_kfree_skb_any(skb);
+}
+
+static int qed_ll2_alloc_buffer(struct qed_dev *cdev,
+ u8 **data, dma_addr_t *phys_addr)
+{
+ *data = kmalloc(cdev->ll2->rx_size, GFP_ATOMIC);
+ if (!(*data)) {
+ DP_INFO(cdev, "Failed to allocate LL2 buffer data\n");
+ return -ENOMEM;
+ }
+
+ *phys_addr = dma_map_single(&cdev->pdev->dev,
+ ((*data) + NET_SKB_PAD),
+ cdev->ll2->rx_size, DMA_FROM_DEVICE);
+ if (dma_mapping_error(&cdev->pdev->dev, *phys_addr)) {
+ DP_INFO(cdev, "Failed to map LL2 buffer data\n");
+ kfree((*data));
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+static int qed_ll2_dealloc_buffer(struct qed_dev *cdev,
+ struct qed_ll2_buffer *buffer)
+{
+ spin_lock_bh(&cdev->ll2->lock);
+
+ dma_unmap_single(&cdev->pdev->dev, buffer->phys_addr,
+ cdev->ll2->rx_size, DMA_FROM_DEVICE);
+ kfree(buffer->data);
+ list_del(&buffer->list);
+
+ cdev->ll2->rx_cnt--;
+ if (!cdev->ll2->rx_cnt)
+ DP_INFO(cdev, "All LL2 entries were removed\n");
+
+ spin_unlock_bh(&cdev->ll2->lock);
+
+ return 0;
+}
+
+static void qed_ll2_kill_buffers(struct qed_dev *cdev)
+{
+ struct qed_ll2_buffer *buffer, *tmp_buffer;
+
+ list_for_each_entry_safe(buffer, tmp_buffer, &cdev->ll2->list, list)
+ qed_ll2_dealloc_buffer(cdev, buffer);
+}
+
+void qed_ll2b_complete_rx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ struct qed_ll2_rx_packet *p_pkt,
+ struct core_rx_fast_path_cqe *p_cqe,
+ bool b_last_packet)
+{
+ u16 packet_length = le16_to_cpu(p_cqe->packet_length);
+ struct qed_ll2_buffer *buffer = p_pkt->cookie;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ u16 vlan = le16_to_cpu(p_cqe->vlan);
+ u32 opaque_data_0, opaque_data_1;
+ u8 pad = p_cqe->placement_offset;
+ dma_addr_t new_phys_addr;
+ struct sk_buff *skb;
+ bool reuse = false;
+ int rc = -EINVAL;
+ u8 *new_data;
+
+ opaque_data_0 = le32_to_cpu(p_cqe->opaque_data.data[0]);
+ opaque_data_1 = le32_to_cpu(p_cqe->opaque_data.data[1]);
+
+ DP_VERBOSE(p_hwfn,
+ (NETIF_MSG_RX_STATUS | QED_MSG_STORAGE | NETIF_MSG_PKTDATA),
+ "Got an LL2 Rx completion: [Buffer at phys 0x%llx, offset 0x%02x] Length 0x%04x Parse_flags 0x%04x vlan 0x%04x Opaque data [0x%08x:0x%08x]\n",
+ (u64)p_pkt->rx_buf_addr, pad, packet_length,
+ le16_to_cpu(p_cqe->parse_flags.flags), vlan,
+ opaque_data_0, opaque_data_1);
+
+ if ((cdev->dp_module & NETIF_MSG_PKTDATA) && buffer->data) {
+ print_hex_dump(KERN_INFO, "",
+ DUMP_PREFIX_OFFSET, 16, 1,
+ buffer->data, packet_length, false);
+ }
+
+ /* Determine if data is valid */
+ if (packet_length < ETH_HLEN)
+ reuse = true;
+
+ /* Allocate a replacement for buffer; Reuse upon failure */
+ if (!reuse)
+ rc = qed_ll2_alloc_buffer(p_hwfn->cdev, &new_data,
+ &new_phys_addr);
+
+ /* If need to reuse or there's no replacement buffer, repost this */
+ if (rc)
+ goto out_post;
+
+ skb = build_skb(buffer->data, 0);
+ if (!skb) {
+ rc = -ENOMEM;
+ goto out_post;
+ }
+
+ pad += NET_SKB_PAD;
+ skb_reserve(skb, pad);
+ skb_put(skb, packet_length);
+ skb_checksum_none_assert(skb);
+
+ /* Get parital ethernet information instead of eth_type_trans(),
+ * Since we don't have an associated net_device.
+ */
+ skb_reset_mac_header(skb);
+ skb->protocol = eth_hdr(skb)->h_proto;
+
+ /* Pass SKB onward */
+ if (cdev->ll2->cbs && cdev->ll2->cbs->rx_cb) {
+ if (vlan)
+ __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan);
+ cdev->ll2->cbs->rx_cb(cdev->ll2->cb_cookie, skb,
+ opaque_data_0, opaque_data_1);
+ }
+
+ /* Update Buffer information and update FW producer */
+ buffer->data = new_data;
+ buffer->phys_addr = new_phys_addr;
+
+out_post:
+ rc = qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev), cdev->ll2->handle,
+ buffer->phys_addr, 0, buffer, 1);
+
+ if (rc)
+ qed_ll2_dealloc_buffer(cdev, buffer);
+}
+
+static struct qed_ll2_info *__qed_ll2_handle_sanity(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ bool b_lock,
+ bool b_only_active)
+{
+ struct qed_ll2_info *p_ll2_conn, *p_ret = NULL;
+
+ if (connection_handle >= QED_MAX_NUM_OF_LL2_CONNECTIONS)
+ return NULL;
+
+ if (!p_hwfn->p_ll2_info)
+ return NULL;
+
+ p_ll2_conn = &p_hwfn->p_ll2_info[connection_handle];
+
+ if (b_only_active) {
+ if (b_lock)
+ mutex_lock(&p_ll2_conn->mutex);
+ if (p_ll2_conn->b_active)
+ p_ret = p_ll2_conn;
+ if (b_lock)
+ mutex_unlock(&p_ll2_conn->mutex);
+ } else {
+ p_ret = p_ll2_conn;
+ }
+
+ return p_ret;
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity(struct qed_hwfn *p_hwfn,
+ u8 connection_handle)
+{
+ return __qed_ll2_handle_sanity(p_hwfn, connection_handle, false, true);
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity_lock(struct qed_hwfn *p_hwfn,
+ u8 connection_handle)
+{
+ return __qed_ll2_handle_sanity(p_hwfn, connection_handle, true, true);
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity_inactive(struct qed_hwfn
+ *p_hwfn,
+ u8 connection_handle)
+{
+ return __qed_ll2_handle_sanity(p_hwfn, connection_handle, false, false);
+}
+
+static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ bool b_last_packet = false, b_last_frag = false;
+ struct qed_ll2_tx_packet *p_pkt = NULL;
+ struct qed_ll2_info *p_ll2_conn;
+ struct qed_ll2_tx_queue *p_tx;
+
+ p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return;
+
+ p_tx = &p_ll2_conn->tx_queue;
+
+ while (!list_empty(&p_tx->active_descq)) {
+ p_pkt = list_first_entry(&p_tx->active_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (!p_pkt)
+ break;
+
+ list_del(&p_pkt->list_entry);
+ b_last_packet = list_empty(&p_tx->active_descq);
+ list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+ p_tx->cur_completing_packet = *p_pkt;
+ p_tx->cur_completing_bd_idx = 1;
+ b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
+
+ qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
+ p_pkt->cookie,
+ p_pkt->bds_set[0].tx_frag,
+ b_last_frag, b_last_packet);
+ }
+}
+
+static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
+{
+ struct qed_ll2_info *p_ll2_conn = p_cookie;
+ struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+ u16 new_idx = 0, num_bds = 0, num_bds_in_packet = 0;
+ struct qed_ll2_tx_packet *p_pkt;
+ bool b_last_frag = false;
+ unsigned long flags;
+ int rc = -EINVAL;
+
+ spin_lock_irqsave(&p_tx->lock, flags);
+ if (p_tx->b_completing_packet) {
+ rc = -EBUSY;
+ goto out;
+ }
+
+ new_idx = le16_to_cpu(*p_tx->p_fw_cons);
+ num_bds = ((s16)new_idx - (s16)p_tx->bds_idx);
+ while (num_bds) {
+ if (list_empty(&p_tx->active_descq))
+ goto out;
+
+ p_pkt = list_first_entry(&p_tx->active_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (!p_pkt)
+ goto out;
+
+ p_tx->b_completing_packet = true;
+ p_tx->cur_completing_packet = *p_pkt;
+ num_bds_in_packet = p_pkt->bd_used;
+ list_del(&p_pkt->list_entry);
+
+ if (num_bds < num_bds_in_packet) {
+ DP_NOTICE(p_hwfn,
+ "Rest of BDs does not cover whole packet\n");
+ goto out;
+ }
+
+ num_bds -= num_bds_in_packet;
+ p_tx->bds_idx += num_bds_in_packet;
+ while (num_bds_in_packet--)
+ qed_chain_consume(&p_tx->txq_chain);
+
+ p_tx->cur_completing_bd_idx = 1;
+ b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
+ list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+
+ spin_unlock_irqrestore(&p_tx->lock, flags);
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ p_pkt->bds_set[0].tx_frag,
+ b_last_frag, !num_bds);
+ spin_lock_irqsave(&p_tx->lock, flags);
+ }
+
+ p_tx->b_completing_packet = false;
+ rc = 0;
+out:
+ spin_unlock_irqrestore(&p_tx->lock, flags);
+ return rc;
+}
+
+static int qed_ll2_rxq_completion_reg(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn,
+ union core_rx_cqe_union *p_cqe,
+ unsigned long lock_flags,
+ bool b_last_cqe)
+{
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+ struct qed_ll2_rx_packet *p_pkt = NULL;
+
+ if (!list_empty(&p_rx->active_descq))
+ p_pkt = list_first_entry(&p_rx->active_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (!p_pkt) {
+ DP_NOTICE(p_hwfn,
+ "LL2 Rx completion but active_descq is empty\n");
+ return -EIO;
+ }
+ list_del(&p_pkt->list_entry);
+
+ if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
+ DP_NOTICE(p_hwfn,
+ "Mismatch between active_descq and the LL2 Rx chain\n");
+ list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+ spin_unlock_irqrestore(&p_rx->lock, lock_flags);
+ qed_ll2b_complete_rx_packet(p_hwfn, p_ll2_conn->my_id,
+ p_pkt, &p_cqe->rx_cqe_fp, b_last_cqe);
+ spin_lock_irqsave(&p_rx->lock, lock_flags);
+
+ return 0;
+}
+
+static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
+{
+ struct qed_ll2_info *p_ll2_conn = cookie;
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+ union core_rx_cqe_union *cqe = NULL;
+ u16 cq_new_idx = 0, cq_old_idx = 0;
+ unsigned long flags = 0;
+ int rc = 0;
+
+ spin_lock_irqsave(&p_rx->lock, flags);
+ cq_new_idx = le16_to_cpu(*p_rx->p_fw_cons);
+ cq_old_idx = qed_chain_get_cons_idx(&p_rx->rcq_chain);
+
+ while (cq_new_idx != cq_old_idx) {
+ bool b_last_cqe = (cq_new_idx == cq_old_idx);
+
+ cqe = qed_chain_consume(&p_rx->rcq_chain);
+ cq_old_idx = qed_chain_get_cons_idx(&p_rx->rcq_chain);
+
+ DP_VERBOSE(p_hwfn,
+ QED_MSG_LL2,
+ "LL2 [sw. cons %04x, fw. at %04x] - Got Packet of type %02x\n",
+ cq_old_idx, cq_new_idx, cqe->rx_cqe_sp.type);
+
+ switch (cqe->rx_cqe_sp.type) {
+ case CORE_RX_CQE_TYPE_SLOW_PATH:
+ DP_NOTICE(p_hwfn, "LL2 - unexpected Rx CQE slowpath\n");
+ rc = -EINVAL;
+ break;
+ case CORE_RX_CQE_TYPE_REGULAR:
+ rc = qed_ll2_rxq_completion_reg(p_hwfn, p_ll2_conn,
+ cqe, flags, b_last_cqe);
+ break;
+ default:
+ rc = -EIO;
+ }
+ }
+
+ spin_unlock_irqrestore(&p_rx->lock, flags);
+ return rc;
+}
+
+void qed_ll2_rxq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ struct qed_ll2_rx_packet *p_pkt = NULL;
+ struct qed_ll2_rx_queue *p_rx;
+
+ p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return;
+
+ p_rx = &p_ll2_conn->rx_queue;
+
+ while (!list_empty(&p_rx->active_descq)) {
+ dma_addr_t rx_buf_addr;
+ void *cookie;
+ bool b_last;
+
+ p_pkt = list_first_entry(&p_rx->active_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (!p_pkt)
+ break;
+
+ list_del(&p_pkt->list_entry);
+ list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+ rx_buf_addr = p_pkt->rx_buf_addr;
+ cookie = p_pkt->cookie;
+
+ b_last = list_empty(&p_rx->active_descq);
+ }
+}
+
+static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn,
+ u8 action_on_error)
+{
+ enum qed_ll2_conn_type conn_type = p_ll2_conn->conn_type;
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+ struct core_rx_start_ramrod_data *p_ramrod = NULL;
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ u16 cqe_pbl_size;
+ int rc = 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_RX_QUEUE_START,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.core_rx_queue_start;
+
+ p_ramrod->sb_id = cpu_to_le16(qed_int_get_sp_sb_id(p_hwfn));
+ p_ramrod->sb_index = p_rx->rx_sb_index;
+ p_ramrod->complete_event_flg = 1;
+
+ p_ramrod->mtu = cpu_to_le16(p_ll2_conn->mtu);
+ DMA_REGPAIR_LE(p_ramrod->bd_base,
+ p_rx->rxq_chain.p_phys_addr);
+ cqe_pbl_size = (u16)qed_chain_get_page_cnt(&p_rx->rcq_chain);
+ p_ramrod->num_of_pbl_pages = cpu_to_le16(cqe_pbl_size);
+ DMA_REGPAIR_LE(p_ramrod->cqe_pbl_addr,
+ qed_chain_get_pbl_phys(&p_rx->rcq_chain));
+
+ p_ramrod->drop_ttl0_flg = p_ll2_conn->rx_drop_ttl0_flg;
+ p_ramrod->inner_vlan_removal_en = p_ll2_conn->rx_vlan_removal_en;
+ p_ramrod->queue_id = p_ll2_conn->queue_id;
+
+ if ((IS_MF_DEFAULT(p_hwfn) || IS_MF_SI(p_hwfn)) &&
+ p_ramrod->main_func_queue && (conn_type != QED_LL2_TYPE_ROCE)) {
+ p_ramrod->mf_si_bcast_accept_all = 1;
+ p_ramrod->mf_si_mcast_accept_all = 1;
+ } else {
+ p_ramrod->mf_si_bcast_accept_all = 0;
+ p_ramrod->mf_si_mcast_accept_all = 0;
+ }
+
+ p_ramrod->action_on_error.error_type = action_on_error;
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ enum qed_ll2_conn_type conn_type = p_ll2_conn->conn_type;
+ struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+ struct core_tx_start_ramrod_data *p_ramrod = NULL;
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ union qed_qm_pq_params pq_params;
+ u16 pq_id = 0, pbl_size;
+ int rc = -EINVAL;
+
+ if (!QED_LL2_TX_REGISTERED(p_ll2_conn))
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_TX_QUEUE_START,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.core_tx_queue_start;
+
+ p_ramrod->sb_id = cpu_to_le16(qed_int_get_sp_sb_id(p_hwfn));
+ p_ramrod->sb_index = p_tx->tx_sb_index;
+ p_ramrod->mtu = cpu_to_le16(p_ll2_conn->mtu);
+ p_ll2_conn->tx_stats_en = 1;
+ p_ramrod->stats_en = p_ll2_conn->tx_stats_en;
+ p_ramrod->stats_id = p_ll2_conn->tx_stats_id;
+
+ DMA_REGPAIR_LE(p_ramrod->pbl_base_addr,
+ qed_chain_get_pbl_phys(&p_tx->txq_chain));
+ pbl_size = qed_chain_get_page_cnt(&p_tx->txq_chain);
+ p_ramrod->pbl_size = cpu_to_le16(pbl_size);
+
+ memset(&pq_params, 0, sizeof(pq_params));
+ pq_params.core.tc = p_ll2_conn->tx_tc;
+ pq_id = qed_get_qm_pq(p_hwfn, PROTOCOLID_CORE, &pq_params);
+ p_ramrod->qm_pq_id = cpu_to_le16(pq_id);
+
+ switch (conn_type) {
+ case QED_LL2_TYPE_ISCSI:
+ case QED_LL2_TYPE_ISCSI_OOO:
+ p_ramrod->conn_type = PROTOCOLID_ISCSI;
+ break;
+ case QED_LL2_TYPE_ROCE:
+ p_ramrod->conn_type = PROTOCOLID_ROCE;
+ break;
+ default:
+ p_ramrod->conn_type = PROTOCOLID_ETH;
+ DP_NOTICE(p_hwfn, "Unknown connection type: %d\n", conn_type);
+ }
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_rx_queue_stop(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ struct core_rx_stop_ramrod_data *p_ramrod = NULL;
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ int rc = -EINVAL;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_RX_QUEUE_STOP,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.core_rx_queue_stop;
+
+ p_ramrod->complete_event_flg = 1;
+ p_ramrod->queue_id = p_ll2_conn->queue_id;
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_tx_queue_stop(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ struct qed_spq_entry *p_ent = NULL;
+ struct qed_sp_init_data init_data;
+ int rc = -EINVAL;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = p_ll2_conn->cid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ CORE_RAMROD_TX_QUEUE_STOP,
+ PROTOCOLID_CORE, &init_data);
+ if (rc)
+ return rc;
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int
+qed_ll2_acquire_connection_rx(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info, u16 rx_num_desc)
+{
+ struct qed_ll2_rx_packet *p_descq;
+ u32 capacity;
+ int rc = 0;
+
+ if (!rx_num_desc)
+ goto out;
+
+ rc = qed_chain_alloc(p_hwfn->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_NEXT_PTR,
+ QED_CHAIN_CNT_TYPE_U16,
+ rx_num_desc,
+ sizeof(struct core_rx_bd),
+ &p_ll2_info->rx_queue.rxq_chain);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "Failed to allocate ll2 rxq chain\n");
+ goto out;
+ }
+
+ capacity = qed_chain_get_capacity(&p_ll2_info->rx_queue.rxq_chain);
+ p_descq = kcalloc(capacity, sizeof(struct qed_ll2_rx_packet),
+ GFP_KERNEL);
+ if (!p_descq) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn, "Failed to allocate ll2 Rx desc\n");
+ goto out;
+ }
+ p_ll2_info->rx_queue.descq_array = p_descq;
+
+ rc = qed_chain_alloc(p_hwfn->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U16,
+ rx_num_desc,
+ sizeof(struct core_rx_fast_path_cqe),
+ &p_ll2_info->rx_queue.rcq_chain);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "Failed to allocate ll2 rcq chain\n");
+ goto out;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_LL2,
+ "Allocated LL2 Rxq [Type %08x] with 0x%08x buffers\n",
+ p_ll2_info->conn_type, rx_num_desc);
+
+out:
+ return rc;
+}
+
+static int qed_ll2_acquire_connection_tx(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info,
+ u16 tx_num_desc)
+{
+ struct qed_ll2_tx_packet *p_descq;
+ u32 capacity;
+ int rc = 0;
+
+ if (!tx_num_desc)
+ goto out;
+
+ rc = qed_chain_alloc(p_hwfn->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U16,
+ tx_num_desc,
+ sizeof(struct core_tx_bd),
+ &p_ll2_info->tx_queue.txq_chain);
+ if (rc)
+ goto out;
+
+ capacity = qed_chain_get_capacity(&p_ll2_info->tx_queue.txq_chain);
+ p_descq = kcalloc(capacity, sizeof(struct qed_ll2_tx_packet),
+ GFP_KERNEL);
+ if (!p_descq) {
+ rc = -ENOMEM;
+ goto out;
+ }
+ p_ll2_info->tx_queue.descq_array = p_descq;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_LL2,
+ "Allocated LL2 Txq [Type %08x] with 0x%08x buffers\n",
+ p_ll2_info->conn_type, tx_num_desc);
+
+out:
+ if (rc)
+ DP_NOTICE(p_hwfn,
+ "Can't allocate memory for Tx LL2 with 0x%08x buffers\n",
+ tx_num_desc);
+ return rc;
+}
+
+int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_params,
+ u16 rx_num_desc,
+ u16 tx_num_desc,
+ u8 *p_connection_handle)
+{
+ qed_int_comp_cb_t comp_rx_cb, comp_tx_cb;
+ struct qed_ll2_info *p_ll2_info = NULL;
+ int rc;
+ u8 i;
+
+ if (!p_connection_handle || !p_hwfn->p_ll2_info)
+ return -EINVAL;
+
+ /* Find a free connection to be used */
+ for (i = 0; (i < QED_MAX_NUM_OF_LL2_CONNECTIONS); i++) {
+ mutex_lock(&p_hwfn->p_ll2_info[i].mutex);
+ if (p_hwfn->p_ll2_info[i].b_active) {
+ mutex_unlock(&p_hwfn->p_ll2_info[i].mutex);
+ continue;
+ }
+
+ p_hwfn->p_ll2_info[i].b_active = true;
+ p_ll2_info = &p_hwfn->p_ll2_info[i];
+ mutex_unlock(&p_hwfn->p_ll2_info[i].mutex);
+ break;
+ }
+ if (!p_ll2_info)
+ return -EBUSY;
+
+ p_ll2_info->conn_type = p_params->conn_type;
+ p_ll2_info->mtu = p_params->mtu;
+ p_ll2_info->rx_drop_ttl0_flg = p_params->rx_drop_ttl0_flg;
+ p_ll2_info->rx_vlan_removal_en = p_params->rx_vlan_removal_en;
+ p_ll2_info->tx_tc = p_params->tx_tc;
+ p_ll2_info->tx_dest = p_params->tx_dest;
+ p_ll2_info->ai_err_packet_too_big = p_params->ai_err_packet_too_big;
+ p_ll2_info->ai_err_no_buf = p_params->ai_err_no_buf;
+
+ rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info, rx_num_desc);
+ if (rc)
+ goto q_allocate_fail;
+
+ rc = qed_ll2_acquire_connection_tx(p_hwfn, p_ll2_info, tx_num_desc);
+ if (rc)
+ goto q_allocate_fail;
+
+ /* Register callbacks for the Rx/Tx queues */
+ comp_rx_cb = qed_ll2_rxq_completion;
+ comp_tx_cb = qed_ll2_txq_completion;
+
+ if (rx_num_desc) {
+ qed_int_register_cb(p_hwfn, comp_rx_cb,
+ &p_hwfn->p_ll2_info[i],
+ &p_ll2_info->rx_queue.rx_sb_index,
+ &p_ll2_info->rx_queue.p_fw_cons);
+ p_ll2_info->rx_queue.b_cb_registred = true;
+ }
+
+ if (tx_num_desc) {
+ qed_int_register_cb(p_hwfn,
+ comp_tx_cb,
+ &p_hwfn->p_ll2_info[i],
+ &p_ll2_info->tx_queue.tx_sb_index,
+ &p_ll2_info->tx_queue.p_fw_cons);
+ p_ll2_info->tx_queue.b_cb_registred = true;
+ }
+
+ *p_connection_handle = i;
+ return rc;
+
+q_allocate_fail:
+ qed_ll2_release_connection(p_hwfn, i);
+ return -ENOMEM;
+}
+
+static int qed_ll2_establish_connection_rx(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ u8 action_on_error = 0;
+
+ if (!QED_LL2_RX_REGISTERED(p_ll2_conn))
+ return 0;
+
+ DIRECT_REG_WR(p_ll2_conn->rx_queue.set_prod_addr, 0x0);
+
+ SET_FIELD(action_on_error,
+ CORE_RX_ACTION_ON_ERROR_PACKET_TOO_BIG,
+ p_ll2_conn->ai_err_packet_too_big);
+ SET_FIELD(action_on_error,
+ CORE_RX_ACTION_ON_ERROR_NO_BUFF, p_ll2_conn->ai_err_no_buf);
+
+ return qed_sp_ll2_rx_queue_start(p_hwfn, p_ll2_conn, action_on_error);
+}
+
+int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn;
+ struct qed_ll2_rx_queue *p_rx;
+ struct qed_ll2_tx_queue *p_tx;
+ int rc = -EINVAL;
+ u32 i, capacity;
+ u8 qid;
+
+ p_ll2_conn = qed_ll2_handle_sanity_lock(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+ p_rx = &p_ll2_conn->rx_queue;
+ p_tx = &p_ll2_conn->tx_queue;
+
+ qed_chain_reset(&p_rx->rxq_chain);
+ qed_chain_reset(&p_rx->rcq_chain);
+ INIT_LIST_HEAD(&p_rx->active_descq);
+ INIT_LIST_HEAD(&p_rx->free_descq);
+ INIT_LIST_HEAD(&p_rx->posting_descq);
+ spin_lock_init(&p_rx->lock);
+ capacity = qed_chain_get_capacity(&p_rx->rxq_chain);
+ for (i = 0; i < capacity; i++)
+ list_add_tail(&p_rx->descq_array[i].list_entry,
+ &p_rx->free_descq);
+ *p_rx->p_fw_cons = 0;
+
+ qed_chain_reset(&p_tx->txq_chain);
+ INIT_LIST_HEAD(&p_tx->active_descq);
+ INIT_LIST_HEAD(&p_tx->free_descq);
+ INIT_LIST_HEAD(&p_tx->sending_descq);
+ spin_lock_init(&p_tx->lock);
+ capacity = qed_chain_get_capacity(&p_tx->txq_chain);
+ for (i = 0; i < capacity; i++)
+ list_add_tail(&p_tx->descq_array[i].list_entry,
+ &p_tx->free_descq);
+ p_tx->cur_completing_bd_idx = 0;
+ p_tx->bds_idx = 0;
+ p_tx->b_completing_packet = false;
+ p_tx->cur_send_packet = NULL;
+ p_tx->cur_send_frag_num = 0;
+ p_tx->cur_completing_frag_num = 0;
+ *p_tx->p_fw_cons = 0;
+
+ qed_cxt_acquire_cid(p_hwfn, PROTOCOLID_CORE, &p_ll2_conn->cid);
+
+ qid = p_hwfn->hw_info.resc_start[QED_LL2_QUEUE] + connection_handle;
+ p_ll2_conn->queue_id = qid;
+ p_ll2_conn->tx_stats_id = qid;
+ p_rx->set_prod_addr = (u8 __iomem *)p_hwfn->regview +
+ GTT_BAR0_MAP_REG_TSDM_RAM +
+ TSTORM_LL2_RX_PRODS_OFFSET(qid);
+ p_tx->doorbell_addr = (u8 __iomem *)p_hwfn->doorbells +
+ qed_db_addr(p_ll2_conn->cid,
+ DQ_DEMS_LEGACY);
+
+ rc = qed_ll2_establish_connection_rx(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+
+ rc = qed_sp_ll2_tx_queue_start(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+
+ if (p_hwfn->hw_info.personality != QED_PCI_ETH_ROCE)
+ qed_wr(p_hwfn, p_hwfn->p_main_ptt, PRS_REG_USE_LIGHT_L2, 1);
+
+ return rc;
+}
+
+static void qed_ll2_post_rx_buffer_notify_fw(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_rx_queue *p_rx,
+ struct qed_ll2_rx_packet *p_curp)
+{
+ struct qed_ll2_rx_packet *p_posting_packet = NULL;
+ struct core_ll2_rx_prod rx_prod = { 0, 0, 0 };
+ bool b_notify_fw = false;
+ u16 bd_prod, cq_prod;
+
+ /* This handles the flushing of already posted buffers */
+ while (!list_empty(&p_rx->posting_descq)) {
+ p_posting_packet = list_first_entry(&p_rx->posting_descq,
+ struct qed_ll2_rx_packet,
+ list_entry);
+ list_del(&p_posting_packet->list_entry);
+ list_add_tail(&p_posting_packet->list_entry,
+ &p_rx->active_descq);
+ b_notify_fw = true;
+ }
+
+ /* This handles the supplied packet [if there is one] */
+ if (p_curp) {
+ list_add_tail(&p_curp->list_entry, &p_rx->active_descq);
+ b_notify_fw = true;
+ }
+
+ if (!b_notify_fw)
+ return;
+
+ bd_prod = qed_chain_get_prod_idx(&p_rx->rxq_chain);
+ cq_prod = qed_chain_get_prod_idx(&p_rx->rcq_chain);
+ rx_prod.bd_prod = cpu_to_le16(bd_prod);
+ rx_prod.cqe_prod = cpu_to_le16(cq_prod);
+ DIRECT_REG_WR(p_rx->set_prod_addr, *((u32 *)&rx_prod));
+}
+
+int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr,
+ u16 buf_len, void *cookie, u8 notify_fw)
+{
+ struct core_rx_bd_with_buff_len *p_curb = NULL;
+ struct qed_ll2_rx_packet *p_curp = NULL;
+ struct qed_ll2_info *p_ll2_conn;
+ struct qed_ll2_rx_queue *p_rx;
+ unsigned long flags;
+ void *p_data;
+ int rc = 0;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+ p_rx = &p_ll2_conn->rx_queue;
+
+ spin_lock_irqsave(&p_rx->lock, flags);
+ if (!list_empty(&p_rx->free_descq))
+ p_curp = list_first_entry(&p_rx->free_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (p_curp) {
+ if (qed_chain_get_elem_left(&p_rx->rxq_chain) &&
+ qed_chain_get_elem_left(&p_rx->rcq_chain)) {
+ p_data = qed_chain_produce(&p_rx->rxq_chain);
+ p_curb = (struct core_rx_bd_with_buff_len *)p_data;
+ qed_chain_produce(&p_rx->rcq_chain);
+ }
+ }
+
+ /* If we're lacking entires, let's try to flush buffers to FW */
+ if (!p_curp || !p_curb) {
+ rc = -EBUSY;
+ p_curp = NULL;
+ goto out_notify;
+ }
+
+ /* We have an Rx packet we can fill */
+ DMA_REGPAIR_LE(p_curb->addr, addr);
+ p_curb->buff_length = cpu_to_le16(buf_len);
+ p_curp->rx_buf_addr = addr;
+ p_curp->cookie = cookie;
+ p_curp->rxq_bd = p_curb;
+ p_curp->buf_length = buf_len;
+ list_del(&p_curp->list_entry);
+
+ /* Check if we only want to enqueue this packet without informing FW */
+ if (!notify_fw) {
+ list_add_tail(&p_curp->list_entry, &p_rx->posting_descq);
+ goto out;
+ }
+
+out_notify:
+ qed_ll2_post_rx_buffer_notify_fw(p_hwfn, p_rx, p_curp);
+out:
+ spin_unlock_irqrestore(&p_rx->lock, flags);
+ return rc;
+}
+
+static void qed_ll2_prepare_tx_packet_set(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_tx_queue *p_tx,
+ struct qed_ll2_tx_packet *p_curp,
+ u8 num_of_bds,
+ dma_addr_t first_frag,
+ u16 first_frag_len, void *p_cookie,
+ u8 notify_fw)
+{
+ list_del(&p_curp->list_entry);
+ p_curp->cookie = p_cookie;
+ p_curp->bd_used = num_of_bds;
+ p_curp->notify_fw = notify_fw;
+ p_tx->cur_send_packet = p_curp;
+ p_tx->cur_send_frag_num = 0;
+
+ p_curp->bds_set[p_tx->cur_send_frag_num].tx_frag = first_frag;
+ p_curp->bds_set[p_tx->cur_send_frag_num].frag_len = first_frag_len;
+ p_tx->cur_send_frag_num++;
+}
+
+static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2,
+ struct qed_ll2_tx_packet *p_curp,
+ u8 num_of_bds,
+ enum core_tx_dest tx_dest,
+ u16 vlan,
+ u8 bd_flags,
+ u16 l4_hdr_offset_w,
+ dma_addr_t first_frag,
+ u16 first_frag_len)
+{
+ struct qed_chain *p_tx_chain = &p_ll2->tx_queue.txq_chain;
+ u16 prod_idx = qed_chain_get_prod_idx(p_tx_chain);
+ struct core_tx_bd *start_bd = NULL;
+ u16 frag_idx;
+
+ start_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
+ start_bd->nw_vlan_or_lb_echo = cpu_to_le16(vlan);
+ SET_FIELD(start_bd->bitfield1, CORE_TX_BD_L4_HDR_OFFSET_W,
+ cpu_to_le16(l4_hdr_offset_w));
+ SET_FIELD(start_bd->bitfield1, CORE_TX_BD_TX_DST, tx_dest);
+ start_bd->bd_flags.as_bitfield = bd_flags;
+ start_bd->bd_flags.as_bitfield |= CORE_TX_BD_FLAGS_START_BD_MASK <<
+ CORE_TX_BD_FLAGS_START_BD_SHIFT;
+ SET_FIELD(start_bd->bitfield0, CORE_TX_BD_NBDS, num_of_bds);
+ DMA_REGPAIR_LE(start_bd->addr, first_frag);
+ start_bd->nbytes = cpu_to_le16(first_frag_len);
+
+ DP_VERBOSE(p_hwfn,
+ (NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
+ "LL2 [q 0x%02x cid 0x%08x type 0x%08x] Tx Producer at [0x%04x] - set with a %04x bytes %02x BDs buffer at %08x:%08x\n",
+ p_ll2->queue_id,
+ p_ll2->cid,
+ p_ll2->conn_type,
+ prod_idx,
+ first_frag_len,
+ num_of_bds,
+ le32_to_cpu(start_bd->addr.hi),
+ le32_to_cpu(start_bd->addr.lo));
+
+ if (p_ll2->tx_queue.cur_send_frag_num == num_of_bds)
+ return;
+
+ /* Need to provide the packet with additional BDs for frags */
+ for (frag_idx = p_ll2->tx_queue.cur_send_frag_num;
+ frag_idx < num_of_bds; frag_idx++) {
+ struct core_tx_bd **p_bd = &p_curp->bds_set[frag_idx].txq_bd;
+
+ *p_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
+ (*p_bd)->bd_flags.as_bitfield = 0;
+ (*p_bd)->bitfield1 = 0;
+ (*p_bd)->bitfield0 = 0;
+ p_curp->bds_set[frag_idx].tx_frag = 0;
+ p_curp->bds_set[frag_idx].frag_len = 0;
+ }
+}
+
+/* This should be called while the Txq spinlock is being held */
+static void qed_ll2_tx_packet_notify(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_conn)
+{
+ bool b_notify = p_ll2_conn->tx_queue.cur_send_packet->notify_fw;
+ struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+ struct qed_ll2_tx_packet *p_pkt = NULL;
+ struct core_db_data db_msg = { 0, 0, 0 };
+ u16 bd_prod;
+
+ /* If there are missing BDs, don't do anything now */
+ if (p_ll2_conn->tx_queue.cur_send_frag_num !=
+ p_ll2_conn->tx_queue.cur_send_packet->bd_used)
+ return;
+
+ /* Push the current packet to the list and clean after it */
+ list_add_tail(&p_ll2_conn->tx_queue.cur_send_packet->list_entry,
+ &p_ll2_conn->tx_queue.sending_descq);
+ p_ll2_conn->tx_queue.cur_send_packet = NULL;
+ p_ll2_conn->tx_queue.cur_send_frag_num = 0;
+
+ /* Notify FW of packet only if requested to */
+ if (!b_notify)
+ return;
+
+ bd_prod = qed_chain_get_prod_idx(&p_ll2_conn->tx_queue.txq_chain);
+
+ while (!list_empty(&p_tx->sending_descq)) {
+ p_pkt = list_first_entry(&p_tx->sending_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (!p_pkt)
+ break;
+
+ list_del(&p_pkt->list_entry);
+ list_add_tail(&p_pkt->list_entry, &p_tx->active_descq);
+ }
+
+ SET_FIELD(db_msg.params, CORE_DB_DATA_DEST, DB_DEST_XCM);
+ SET_FIELD(db_msg.params, CORE_DB_DATA_AGG_CMD, DB_AGG_CMD_SET);
+ SET_FIELD(db_msg.params, CORE_DB_DATA_AGG_VAL_SEL,
+ DQ_XCM_CORE_TX_BD_PROD_CMD);
+ db_msg.agg_flags = DQ_XCM_CORE_DQ_CF_CMD;
+ db_msg.spq_prod = cpu_to_le16(bd_prod);
+
+ /* Make sure the BDs data is updated before ringing the doorbell */
+ wmb();
+
+ DIRECT_REG_WR(p_tx->doorbell_addr, *((u32 *)&db_msg));
+
+ DP_VERBOSE(p_hwfn,
+ (NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
+ "LL2 [q 0x%02x cid 0x%08x type 0x%08x] Doorbelled [producer 0x%04x]\n",
+ p_ll2_conn->queue_id,
+ p_ll2_conn->cid, p_ll2_conn->conn_type, db_msg.spq_prod);
+}
+
+int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ u8 num_of_bds,
+ u16 vlan,
+ u8 bd_flags,
+ u16 l4_hdr_offset_w,
+ dma_addr_t first_frag,
+ u16 first_frag_len, void *cookie, u8 notify_fw)
+{
+ struct qed_ll2_tx_packet *p_curp = NULL;
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ struct qed_ll2_tx_queue *p_tx;
+ struct qed_chain *p_tx_chain;
+ unsigned long flags;
+ int rc = 0;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+ p_tx = &p_ll2_conn->tx_queue;
+ p_tx_chain = &p_tx->txq_chain;
+
+ if (num_of_bds > CORE_LL2_TX_MAX_BDS_PER_PACKET)
+ return -EIO;
+
+ spin_lock_irqsave(&p_tx->lock, flags);
+ if (p_tx->cur_send_packet) {
+ rc = -EEXIST;
+ goto out;
+ }
+
+ /* Get entry, but only if we have tx elements for it */
+ if (!list_empty(&p_tx->free_descq))
+ p_curp = list_first_entry(&p_tx->free_descq,
+ struct qed_ll2_tx_packet, list_entry);
+ if (p_curp && qed_chain_get_elem_left(p_tx_chain) < num_of_bds)
+ p_curp = NULL;
+
+ if (!p_curp) {
+ rc = -EBUSY;
+ goto out;
+ }
+
+ /* Prepare packet and BD, and perhaps send a doorbell to FW */
+ qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp,
+ num_of_bds, first_frag,
+ first_frag_len, cookie, notify_fw);
+ qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp,
+ num_of_bds, CORE_TX_DEST_NW,
+ vlan, bd_flags, l4_hdr_offset_w,
+ first_frag, first_frag_len);
+
+ qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
+
+out:
+ spin_unlock_irqrestore(&p_tx->lock, flags);
+ return rc;
+}
+
+int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr, u16 nbytes)
+{
+ struct qed_ll2_tx_packet *p_cur_send_packet = NULL;
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ u16 cur_send_frag_num = 0;
+ struct core_tx_bd *p_bd;
+ unsigned long flags;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+
+ if (!p_ll2_conn->tx_queue.cur_send_packet)
+ return -EINVAL;
+
+ p_cur_send_packet = p_ll2_conn->tx_queue.cur_send_packet;
+ cur_send_frag_num = p_ll2_conn->tx_queue.cur_send_frag_num;
+
+ if (cur_send_frag_num >= p_cur_send_packet->bd_used)
+ return -EINVAL;
+
+ /* Fill the BD information, and possibly notify FW */
+ p_bd = p_cur_send_packet->bds_set[cur_send_frag_num].txq_bd;
+ DMA_REGPAIR_LE(p_bd->addr, addr);
+ p_bd->nbytes = cpu_to_le16(nbytes);
+ p_cur_send_packet->bds_set[cur_send_frag_num].tx_frag = addr;
+ p_cur_send_packet->bds_set[cur_send_frag_num].frag_len = nbytes;
+
+ p_ll2_conn->tx_queue.cur_send_frag_num++;
+
+ spin_lock_irqsave(&p_ll2_conn->tx_queue.lock, flags);
+ qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
+ spin_unlock_irqrestore(&p_ll2_conn->tx_queue.lock, flags);
+
+ return 0;
+}
+
+int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ int rc = -EINVAL;
+
+ p_ll2_conn = qed_ll2_handle_sanity_lock(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return -EINVAL;
+
+ /* Stop Tx & Rx of connection, if needed */
+ if (QED_LL2_TX_REGISTERED(p_ll2_conn)) {
+ rc = qed_sp_ll2_tx_queue_stop(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+ qed_ll2_txq_flush(p_hwfn, connection_handle);
+ }
+
+ if (QED_LL2_RX_REGISTERED(p_ll2_conn)) {
+ rc = qed_sp_ll2_rx_queue_stop(p_hwfn, p_ll2_conn);
+ if (rc)
+ return rc;
+ qed_ll2_rxq_flush(p_hwfn, connection_handle);
+ }
+
+ return rc;
+}
+
+void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+
+ p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+ if (!p_ll2_conn)
+ return;
+
+ if (QED_LL2_RX_REGISTERED(p_ll2_conn)) {
+ p_ll2_conn->rx_queue.b_cb_registred = false;
+ qed_int_unregister_cb(p_hwfn, p_ll2_conn->rx_queue.rx_sb_index);
+ }
+
+ if (QED_LL2_TX_REGISTERED(p_ll2_conn)) {
+ p_ll2_conn->tx_queue.b_cb_registred = false;
+ qed_int_unregister_cb(p_hwfn, p_ll2_conn->tx_queue.tx_sb_index);
+ }
+
+ kfree(p_ll2_conn->tx_queue.descq_array);
+ qed_chain_free(p_hwfn->cdev, &p_ll2_conn->tx_queue.txq_chain);
+
+ kfree(p_ll2_conn->rx_queue.descq_array);
+ qed_chain_free(p_hwfn->cdev, &p_ll2_conn->rx_queue.rxq_chain);
+ qed_chain_free(p_hwfn->cdev, &p_ll2_conn->rx_queue.rcq_chain);
+
+ qed_cxt_release_cid(p_hwfn, p_ll2_conn->cid);
+
+ mutex_lock(&p_ll2_conn->mutex);
+ p_ll2_conn->b_active = false;
+ mutex_unlock(&p_ll2_conn->mutex);
+}
+
+struct qed_ll2_info *qed_ll2_alloc(struct qed_hwfn *p_hwfn)
+{
+ struct qed_ll2_info *p_ll2_connections;
+ u8 i;
+
+ /* Allocate LL2's set struct */
+ p_ll2_connections = kcalloc(QED_MAX_NUM_OF_LL2_CONNECTIONS,
+ sizeof(struct qed_ll2_info), GFP_KERNEL);
+ if (!p_ll2_connections) {
+ DP_NOTICE(p_hwfn, "Failed to allocate `struct qed_ll2'\n");
+ return NULL;
+ }
+
+ for (i = 0; i < QED_MAX_NUM_OF_LL2_CONNECTIONS; i++)
+ p_ll2_connections[i].my_id = i;
+
+ return p_ll2_connections;
+}
+
+void qed_ll2_setup(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections)
+{
+ int i;
+
+ for (i = 0; i < QED_MAX_NUM_OF_LL2_CONNECTIONS; i++)
+ mutex_init(&p_ll2_connections[i].mutex);
+}
+
+void qed_ll2_free(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections)
+{
+ kfree(p_ll2_connections);
+}
+
+static void _qed_ll2_get_tstats(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_ll2_info *p_ll2_conn,
+ struct qed_ll2_stats *p_stats)
+{
+ struct core_ll2_tstorm_per_queue_stat tstats;
+ u8 qid = p_ll2_conn->queue_id;
+ u32 tstats_addr;
+
+ memset(&tstats, 0, sizeof(tstats));
+ tstats_addr = BAR0_MAP_REG_TSDM_RAM +
+ CORE_LL2_TSTORM_PER_QUEUE_STAT_OFFSET(qid);
+ qed_memcpy_from(p_hwfn, p_ptt, &tstats, tstats_addr, sizeof(tstats));
+
+ p_stats->packet_too_big_discard =
+ HILO_64_REGPAIR(tstats.packet_too_big_discard);
+ p_stats->no_buff_discard = HILO_64_REGPAIR(tstats.no_buff_discard);
+}
+
+static void _qed_ll2_get_ustats(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_ll2_info *p_ll2_conn,
+ struct qed_ll2_stats *p_stats)
+{
+ struct core_ll2_ustorm_per_queue_stat ustats;
+ u8 qid = p_ll2_conn->queue_id;
+ u32 ustats_addr;
+
+ memset(&ustats, 0, sizeof(ustats));
+ ustats_addr = BAR0_MAP_REG_USDM_RAM +
+ CORE_LL2_USTORM_PER_QUEUE_STAT_OFFSET(qid);
+ qed_memcpy_from(p_hwfn, p_ptt, &ustats, ustats_addr, sizeof(ustats));
+
+ p_stats->rcv_ucast_bytes = HILO_64_REGPAIR(ustats.rcv_ucast_bytes);
+ p_stats->rcv_mcast_bytes = HILO_64_REGPAIR(ustats.rcv_mcast_bytes);
+ p_stats->rcv_bcast_bytes = HILO_64_REGPAIR(ustats.rcv_bcast_bytes);
+ p_stats->rcv_ucast_pkts = HILO_64_REGPAIR(ustats.rcv_ucast_pkts);
+ p_stats->rcv_mcast_pkts = HILO_64_REGPAIR(ustats.rcv_mcast_pkts);
+ p_stats->rcv_bcast_pkts = HILO_64_REGPAIR(ustats.rcv_bcast_pkts);
+}
+
+static void _qed_ll2_get_pstats(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_ll2_info *p_ll2_conn,
+ struct qed_ll2_stats *p_stats)
+{
+ struct core_ll2_pstorm_per_queue_stat pstats;
+ u8 stats_id = p_ll2_conn->tx_stats_id;
+ u32 pstats_addr;
+
+ memset(&pstats, 0, sizeof(pstats));
+ pstats_addr = BAR0_MAP_REG_PSDM_RAM +
+ CORE_LL2_PSTORM_PER_QUEUE_STAT_OFFSET(stats_id);
+ qed_memcpy_from(p_hwfn, p_ptt, &pstats, pstats_addr, sizeof(pstats));
+
+ p_stats->sent_ucast_bytes = HILO_64_REGPAIR(pstats.sent_ucast_bytes);
+ p_stats->sent_mcast_bytes = HILO_64_REGPAIR(pstats.sent_mcast_bytes);
+ p_stats->sent_bcast_bytes = HILO_64_REGPAIR(pstats.sent_bcast_bytes);
+ p_stats->sent_ucast_pkts = HILO_64_REGPAIR(pstats.sent_ucast_pkts);
+ p_stats->sent_mcast_pkts = HILO_64_REGPAIR(pstats.sent_mcast_pkts);
+ p_stats->sent_bcast_pkts = HILO_64_REGPAIR(pstats.sent_bcast_pkts);
+}
+
+int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+ u8 connection_handle, struct qed_ll2_stats *p_stats)
+{
+ struct qed_ll2_info *p_ll2_conn = NULL;
+ struct qed_ptt *p_ptt;
+
+ memset(p_stats, 0, sizeof(*p_stats));
+
+ if ((connection_handle >= QED_MAX_NUM_OF_LL2_CONNECTIONS) ||
+ !p_hwfn->p_ll2_info)
+ return -EINVAL;
+
+ p_ll2_conn = &p_hwfn->p_ll2_info[connection_handle];
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ DP_ERR(p_hwfn, "Failed to acquire ptt\n");
+ return -EINVAL;
+ }
+
+ _qed_ll2_get_tstats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+ _qed_ll2_get_ustats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+ if (p_ll2_conn->tx_stats_en)
+ _qed_ll2_get_pstats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+
+ qed_ptt_release(p_hwfn, p_ptt);
+ return 0;
+}
+
+static void qed_ll2_register_cb_ops(struct qed_dev *cdev,
+ const struct qed_ll2_cb_ops *ops,
+ void *cookie)
+{
+ cdev->ll2->cbs = ops;
+ cdev->ll2->cb_cookie = cookie;
+}
+
+static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
+{
+ struct qed_ll2_info ll2_info;
+ struct qed_ll2_buffer *buffer;
+ enum qed_ll2_conn_type conn_type;
+ struct qed_ptt *p_ptt;
+ int rc, i;
+
+ /* Initialize LL2 locks & lists */
+ INIT_LIST_HEAD(&cdev->ll2->list);
+ spin_lock_init(&cdev->ll2->lock);
+ cdev->ll2->rx_size = NET_SKB_PAD + ETH_HLEN +
+ L1_CACHE_BYTES + params->mtu;
+ cdev->ll2->frags_mapped = params->frags_mapped;
+
+ /*Allocate memory for LL2 */
+ DP_INFO(cdev, "Allocating LL2 buffers of size %08x bytes\n",
+ cdev->ll2->rx_size);
+ for (i = 0; i < QED_LL2_RX_SIZE; i++) {
+ buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
+ if (!buffer) {
+ DP_INFO(cdev, "Failed to allocate LL2 buffers\n");
+ goto fail;
+ }
+
+ rc = qed_ll2_alloc_buffer(cdev, (u8 **)&buffer->data,
+ &buffer->phys_addr);
+ if (rc) {
+ kfree(buffer);
+ goto fail;
+ }
+
+ list_add_tail(&buffer->list, &cdev->ll2->list);
+ }
+
+ switch (QED_LEADING_HWFN(cdev)->hw_info.personality) {
+ case QED_PCI_ISCSI:
+ conn_type = QED_LL2_TYPE_ISCSI;
+ break;
+ case QED_PCI_ETH_ROCE:
+ conn_type = QED_LL2_TYPE_ROCE;
+ break;
+ default:
+ conn_type = QED_LL2_TYPE_TEST;
+ }
+
+ /* Prepare the temporary ll2 information */
+ memset(&ll2_info, 0, sizeof(ll2_info));
+ ll2_info.conn_type = conn_type;
+ ll2_info.mtu = params->mtu;
+ ll2_info.rx_drop_ttl0_flg = params->drop_ttl0_packets;
+ ll2_info.rx_vlan_removal_en = params->rx_vlan_stripping;
+ ll2_info.tx_tc = 0;
+ ll2_info.tx_dest = CORE_TX_DEST_NW;
+
+ rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_info,
+ QED_LL2_RX_SIZE, QED_LL2_TX_SIZE,
+ &cdev->ll2->handle);
+ if (rc) {
+ DP_INFO(cdev, "Failed to acquire LL2 connection\n");
+ goto fail;
+ }
+
+ rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle);
+ if (rc) {
+ DP_INFO(cdev, "Failed to establish LL2 connection\n");
+ goto release_fail;
+ }
+
+ /* Post all Rx buffers to FW */
+ spin_lock_bh(&cdev->ll2->lock);
+ list_for_each_entry(buffer, &cdev->ll2->list, list) {
+ rc = qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle,
+ buffer->phys_addr, 0, buffer, 1);
+ if (rc) {
+ DP_INFO(cdev,
+ "Failed to post an Rx buffer; Deleting it\n");
+ dma_unmap_single(&cdev->pdev->dev, buffer->phys_addr,
+ cdev->ll2->rx_size, DMA_FROM_DEVICE);
+ kfree(buffer->data);
+ list_del(&buffer->list);
+ kfree(buffer);
+ } else {
+ cdev->ll2->rx_cnt++;
+ }
+ }
+ spin_unlock_bh(&cdev->ll2->lock);
+
+ if (!cdev->ll2->rx_cnt) {
+ DP_INFO(cdev, "Failed passing even a single Rx buffer\n");
+ goto release_terminate;
+ }
+
+ if (!is_valid_ether_addr(params->ll2_mac_address)) {
+ DP_INFO(cdev, "Invalid Ethernet address\n");
+ goto release_terminate;
+ }
+
+ p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ if (!p_ptt) {
+ DP_INFO(cdev, "Failed to acquire PTT\n");
+ goto release_terminate;
+ }
+
+ rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ params->ll2_mac_address);
+ qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+ if (rc) {
+ DP_ERR(cdev, "Failed to allocate LLH filter\n");
+ goto release_terminate_all;
+ }
+
+ ether_addr_copy(cdev->ll2_mac_address, params->ll2_mac_address);
+
+ return 0;
+
+release_terminate_all:
+
+release_terminate:
+ qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+release_fail:
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+fail:
+ qed_ll2_kill_buffers(cdev);
+ cdev->ll2->handle = QED_LL2_UNUSED_HANDLE;
+ return -EINVAL;
+}
+
+static int qed_ll2_stop(struct qed_dev *cdev)
+{
+ struct qed_ptt *p_ptt;
+ int rc;
+
+ if (cdev->ll2->handle == QED_LL2_UNUSED_HANDLE)
+ return 0;
+
+ p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ if (!p_ptt) {
+ DP_INFO(cdev, "Failed to acquire PTT\n");
+ goto fail;
+ }
+
+ qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ cdev->ll2_mac_address);
+ qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+ eth_zero_addr(cdev->ll2_mac_address);
+
+ rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle);
+ if (rc)
+ DP_INFO(cdev, "Failed to terminate LL2 connection\n");
+
+ qed_ll2_kill_buffers(cdev);
+
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+ cdev->ll2->handle = QED_LL2_UNUSED_HANDLE;
+
+ return rc;
+fail:
+ return -EINVAL;
+}
+
+static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
+{
+ const skb_frag_t *frag;
+ int rc = -EINVAL, i;
+ dma_addr_t mapping;
+ u16 vlan = 0;
+ u8 flags = 0;
+
+ if (unlikely(skb->ip_summed != CHECKSUM_NONE)) {
+ DP_INFO(cdev, "Cannot transmit a checksumed packet\n");
+ return -EINVAL;
+ }
+
+ if (1 + skb_shinfo(skb)->nr_frags > CORE_LL2_TX_MAX_BDS_PER_PACKET) {
+ DP_ERR(cdev, "Cannot transmit a packet with %d fragments\n",
+ 1 + skb_shinfo(skb)->nr_frags);
+ return -EINVAL;
+ }
+
+ mapping = dma_map_single(&cdev->pdev->dev, skb->data,
+ skb->len, DMA_TO_DEVICE);
+ if (unlikely(dma_mapping_error(&cdev->pdev->dev, mapping))) {
+ DP_NOTICE(cdev, "SKB mapping failed\n");
+ return -EINVAL;
+ }
+
+ /* Request HW to calculate IP csum */
+ if (!((vlan_get_protocol(skb) == htons(ETH_P_IPV6)) &&
+ ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
+ flags |= BIT(CORE_TX_BD_FLAGS_IP_CSUM_SHIFT);
+
+ if (skb_vlan_tag_present(skb)) {
+ vlan = skb_vlan_tag_get(skb);
+ flags |= BIT(CORE_TX_BD_FLAGS_VLAN_INSERTION_SHIFT);
+ }
+
+ rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle,
+ 1 + skb_shinfo(skb)->nr_frags,
+ vlan, flags, 0, mapping,
+ skb->len, skb, 1);
+ if (rc)
+ goto err;
+
+ for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+ frag = &skb_shinfo(skb)->frags[i];
+ if (!cdev->ll2->frags_mapped) {
+ mapping = skb_frag_dma_map(&cdev->pdev->dev, frag, 0,
+ skb_frag_size(frag),
+ DMA_TO_DEVICE);
+
+ if (unlikely(dma_mapping_error(&cdev->pdev->dev,
+ mapping))) {
+ DP_NOTICE(cdev,
+ "Unable to map frag - dropping packet\n");
+ goto err;
+ }
+ } else {
+ mapping = page_to_phys(skb_frag_page(frag)) |
+ frag->page_offset;
+ }
+
+ rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle,
+ mapping,
+ skb_frag_size(frag));
+
+ /* if failed not much to do here, partial packet has been posted
+ * we can't free memory, will need to wait for completion.
+ */
+ if (rc)
+ goto err2;
+ }
+
+ return 0;
+
+err:
+ dma_unmap_single(&cdev->pdev->dev, mapping, skb->len, DMA_TO_DEVICE);
+
+err2:
+ return rc;
+}
+
+static int qed_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
+{
+ if (!cdev->ll2)
+ return -EINVAL;
+
+ return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
+ cdev->ll2->handle, stats);
+}
+
+const struct qed_ll2_ops qed_ll2_ops_pass = {
+ .start = &qed_ll2_start,
+ .stop = &qed_ll2_stop,
+ .start_xmit = &qed_ll2_start_xmit,
+ .register_cb_ops = &qed_ll2_register_cb_ops,
+ .get_stats = &qed_ll2_stats,
+};
+
+int qed_ll2_alloc_if(struct qed_dev *cdev)
+{
+ cdev->ll2 = kzalloc(sizeof(*cdev->ll2), GFP_KERNEL);
+ return cdev->ll2 ? 0 : -ENOMEM;
+}
+
+void qed_ll2_dealloc_if(struct qed_dev *cdev)
+{
+ kfree(cdev->ll2);
+ cdev->ll2 = NULL;
+}
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
new file mode 100644
index 0000000..a037c48
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -0,0 +1,289 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#ifndef _QED_LL2_H
+#define _QED_LL2_H
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_ll2_if.h>
+#include "qed.h"
+#include "qed_hsi.h"
+#include "qed_sp.h"
+
+#define QED_MAX_NUM_OF_LL2_CONNECTIONS (4)
+
+enum qed_ll2_conn_type {
+ QED_LL2_TYPE_RESERVED,
+ QED_LL2_TYPE_ISCSI,
+ QED_LL2_TYPE_TEST,
+ QED_LL2_TYPE_ISCSI_OOO,
+ QED_LL2_TYPE_RESERVED2,
+ QED_LL2_TYPE_ROCE,
+ QED_LL2_TYPE_RESERVED3,
+ MAX_QED_LL2_RX_CONN_TYPE
+};
+
+struct qed_ll2_rx_packet {
+ struct list_head list_entry;
+ struct core_rx_bd_with_buff_len *rxq_bd;
+ dma_addr_t rx_buf_addr;
+ u16 buf_length;
+ void *cookie;
+ u8 placement_offset;
+ u16 parse_flags;
+ u16 packet_length;
+ u16 vlan;
+ u32 opaque_data[2];
+};
+
+struct qed_ll2_tx_packet {
+ struct list_head list_entry;
+ u16 bd_used;
+ u16 vlan;
+ u16 l4_hdr_offset_w;
+ u8 bd_flags;
+ bool notify_fw;
+ void *cookie;
+
+ struct {
+ struct core_tx_bd *txq_bd;
+ dma_addr_t tx_frag;
+ u16 frag_len;
+ } bds_set[ETH_TX_MAX_BDS_PER_NON_LSO_PACKET];
+};
+
+struct qed_ll2_rx_queue {
+ /* Lock protecting the Rx queue manipulation */
+ spinlock_t lock;
+ struct qed_chain rxq_chain;
+ struct qed_chain rcq_chain;
+ u8 rx_sb_index;
+ bool b_cb_registred;
+ __le16 *p_fw_cons;
+ struct list_head active_descq;
+ struct list_head free_descq;
+ struct list_head posting_descq;
+ struct qed_ll2_rx_packet *descq_array;
+ void __iomem *set_prod_addr;
+};
+
+struct qed_ll2_tx_queue {
+ /* Lock protecting the Tx queue manipulation */
+ spinlock_t lock;
+ struct qed_chain txq_chain;
+ u8 tx_sb_index;
+ bool b_cb_registred;
+ __le16 *p_fw_cons;
+ struct list_head active_descq;
+ struct list_head free_descq;
+ struct list_head sending_descq;
+ struct qed_ll2_tx_packet *descq_array;
+ struct qed_ll2_tx_packet *cur_send_packet;
+ struct qed_ll2_tx_packet cur_completing_packet;
+ u16 cur_completing_bd_idx;
+ void __iomem *doorbell_addr;
+ u16 bds_idx;
+ u16 cur_send_frag_num;
+ u16 cur_completing_frag_num;
+ bool b_completing_packet;
+};
+
+struct qed_ll2_info {
+ /* Lock protecting the state of LL2 */
+ struct mutex mutex;
+ enum qed_ll2_conn_type conn_type;
+ u32 cid;
+ u8 my_id;
+ u8 queue_id;
+ u8 tx_stats_id;
+ bool b_active;
+ u16 mtu;
+ u8 rx_drop_ttl0_flg;
+ u8 rx_vlan_removal_en;
+ u8 tx_tc;
+ enum core_tx_dest tx_dest;
+ enum core_error_handle ai_err_packet_too_big;
+ enum core_error_handle ai_err_no_buf;
+ u8 tx_stats_en;
+ struct qed_ll2_rx_queue rx_queue;
+ struct qed_ll2_tx_queue tx_queue;
+};
+
+/**
+ * @brief qed_ll2_acquire_connection - allocate resources,
+ * starts rx & tx (if relevant) queues pair. Provides
+ * connecion handler as output parameter.
+ *
+ * @param p_hwfn
+ * @param p_params Contain various configuration properties
+ * @param rx_num_desc
+ * @param tx_num_desc
+ *
+ * @param p_connection_handle Output container for LL2 connection's handle
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_params,
+ u16 rx_num_desc,
+ u16 tx_num_desc,
+ u8 *p_connection_handle);
+
+/**
+ * @brief qed_ll2_establish_connection - start previously
+ * allocated LL2 queues pair
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_post_rx_buffers - submit buffers to LL2 Rx queue.
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ * @param addr rx (physical address) buffers to submit
+ * @param cookie
+ * @param notify_fw produce corresponding Rx BD immediately
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr,
+ u16 buf_len, void *cookie, u8 notify_fw);
+
+/**
+ * @brief qed_ll2_prepare_tx_packet - request for start Tx BD
+ * to prepare Tx packet submission to FW.
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ * @param num_of_bds a number of requested BD equals a number of
+ * fragments in Tx packet
+ * @param vlan VLAN to insert to packet (if insertion set)
+ * @param bd_flags
+ * @param l4_hdr_offset_w L4 Header Offset from start of packet
+ * (in words). This is needed if both l4_csum
+ * and ipv6_ext are set
+ * @param first_frag
+ * @param first_frag_len
+ * @param cookie
+ *
+ * @param notify_fw
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ u8 num_of_bds,
+ u16 vlan,
+ u8 bd_flags,
+ u16 l4_hdr_offset_w,
+ dma_addr_t first_frag,
+ u16 first_frag_len, void *cookie, u8 notify_fw);
+
+/**
+ * @brief qed_ll2_release_connection - releases resources
+ * allocated for LL2 connection
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ */
+void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_set_fragment_of_tx_packet - provides fragments to fill
+ * Tx BD of BDs requested by
+ * qed_ll2_prepare_tx_packet
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle
+ * obtained from
+ * qed_ll2_require_connection
+ * @param addr
+ * @param nbytes
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ dma_addr_t addr, u16 nbytes);
+
+/**
+ * @brief qed_ll2_terminate_connection - stops Tx/Rx queues
+ *
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle
+ * obtained from
+ * qed_ll2_require_connection
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_get_stats - get LL2 queue's statistics
+ *
+ *
+ * @param p_hwfn
+ * @param connection_handle LL2 connection's handle obtained from
+ * qed_ll2_require_connection
+ * @param p_stats
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+ u8 connection_handle, struct qed_ll2_stats *p_stats);
+
+/**
+ * @brief qed_ll2_alloc - Allocates LL2 connections set
+ *
+ * @param p_hwfn
+ *
+ * @return pointer to alocated qed_ll2_info or NULL
+ */
+struct qed_ll2_info *qed_ll2_alloc(struct qed_hwfn *p_hwfn);
+
+/**
+ * @brief qed_ll2_setup - Inits LL2 connections set
+ *
+ * @param p_hwfn
+ * @param p_ll2_connections
+ *
+ */
+void qed_ll2_setup(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections);
+
+/**
+ * @brief qed_ll2_free - Releases LL2 connections set
+ *
+ * @param p_hwfn
+ * @param p_ll2_connections
+ *
+ */
+void qed_ll2_free(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_connections);
+
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_main.c b/drivers/net/ethernet/qlogic/qed/qed_main.c
index 32f71ee..0225382 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_main.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_main.c
@@ -22,11 +22,13 @@
#include <linux/etherdevice.h>
#include <linux/vmalloc.h>
#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_ll2_if.h>
#include "qed.h"
#include "qed_sriov.h"
#include "qed_sp.h"
#include "qed_dev_api.h"
+#include "qed_ll2.h"
#include "qed_mcp.h"
#include "qed_hw.h"
#include "qed_selftest.h"
@@ -606,7 +608,16 @@ static int qed_nic_reset(struct qed_dev *cdev)
static int qed_nic_setup(struct qed_dev *cdev)
{
- int rc;
+ int rc, i;
+
+ /* Determine if interface is going to require LL2 */
+ if (QED_LEADING_HWFN(cdev)->hw_info.personality != QED_PCI_ETH) {
+ for (i = 0; i < cdev->num_hwfns; i++) {
+ struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
+
+ p_hwfn->using_ll2 = true;
+ }
+ }
rc = qed_resc_alloc(cdev);
if (rc)
@@ -871,6 +882,12 @@ static int qed_slowpath_start(struct qed_dev *cdev,
DP_INFO(cdev,
"HW initialization and function start completed successfully\n");
+ /* Allocate LL2 interface if needed */
+ if (QED_LEADING_HWFN(cdev)->using_ll2) {
+ rc = qed_ll2_alloc_if(cdev);
+ if (rc)
+ goto err3;
+ }
if (IS_PF(cdev)) {
hwfn = QED_LEADING_HWFN(cdev);
drv_version.version = (params->drv_major << 24) |
@@ -891,6 +908,8 @@ static int qed_slowpath_start(struct qed_dev *cdev,
return 0;
+err3:
+ qed_hw_stop(cdev);
err2:
qed_hw_timers_stop_all(cdev);
if (IS_PF(cdev))
@@ -913,6 +932,8 @@ static int qed_slowpath_stop(struct qed_dev *cdev)
if (!cdev)
return -ENODEV;
+ qed_ll2_dealloc_if(cdev);
+
if (IS_PF(cdev)) {
qed_free_stream_mem(cdev);
if (IS_QED_ETH_IF(cdev))
diff --git a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
index b49d47f..103a32c 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
@@ -208,6 +208,26 @@
0x50196cUL
#define NIG_REG_LLH_CLS_TYPE_DUALMODE \
0x501964UL
+#define NIG_REG_LLH_FUNC_FILTER_VALUE \
+ 0x501a00UL
+#define NIG_REG_LLH_FUNC_FILTER_VALUE_SIZE \
+ 32
+#define NIG_REG_LLH_FUNC_FILTER_EN \
+ 0x501a80UL
+#define NIG_REG_LLH_FUNC_FILTER_EN_SIZE \
+ 16
+#define NIG_REG_LLH_FUNC_FILTER_MODE \
+ 0x501ac0UL
+#define NIG_REG_LLH_FUNC_FILTER_MODE_SIZE \
+ 16
+#define NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE \
+ 0x501b00UL
+#define NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE_SIZE \
+ 16
+#define NIG_REG_LLH_FUNC_FILTER_HDR_SEL \
+ 0x501b40UL
+#define NIG_REG_LLH_FUNC_FILTER_HDR_SEL_SIZE \
+ 16
#define NCSI_REG_CONFIG \
0x040200UL
#define PBF_REG_INIT \
@@ -264,6 +284,8 @@
0x1f0a1cUL
#define PRS_REG_ROCE_DEST_QP_MAX_PF \
0x1f0430UL
+#define PRS_REG_USE_LIGHT_L2 \
+ 0x1f096cUL
#define PSDM_REG_ENABLE_IN1 \
0xfa0004UL
#define PSEM_REG_ENABLE_IN \
diff --git a/drivers/net/ethernet/qlogic/qed/qed_sp.h b/drivers/net/ethernet/qlogic/qed/qed_sp.h
index a548504..a3c539f 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_sp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_sp.h
@@ -61,6 +61,10 @@ union ramrod_data {
struct vport_start_ramrod_data vport_start;
struct vport_stop_ramrod_data vport_stop;
struct vport_update_ramrod_data vport_update;
+ struct core_rx_start_ramrod_data core_rx_queue_start;
+ struct core_rx_stop_ramrod_data core_rx_queue_stop;
+ struct core_tx_start_ramrod_data core_tx_queue_start;
+ struct core_tx_stop_ramrod_data core_tx_queue_stop;
struct vport_filter_update_ramrod_data vport_filter_update;
struct rdma_init_func_ramrod_data rdma_init_func;
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index d8dc5c2..6611441 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -623,6 +623,7 @@ enum DP_MODULE {
QED_MSG_SP = 0x100000,
QED_MSG_STORAGE = 0x200000,
QED_MSG_CXT = 0x800000,
+ QED_MSG_LL2 = 0x1000000,
QED_MSG_ILT = 0x2000000,
QED_MSG_ROCE = 0x4000000,
QED_MSG_DEBUG = 0x8000000,
diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h
new file mode 100644
index 0000000..0164cca
--- /dev/null
+++ b/include/linux/qed/qed_ll2_if.h
@@ -0,0 +1,140 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#ifndef _QED_LL2_IF_H
+#define _QED_LL2_IF_H
+
+#include <linux/types.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/pci.h>
+#include <linux/skbuff.h>
+#include <linux/types.h>
+#include <linux/version.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/qed/qed_if.h>
+
+struct qed_ll2_stats {
+ u64 gsi_invalid_hdr;
+ u64 gsi_invalid_pkt_length;
+ u64 gsi_unsupported_pkt_typ;
+ u64 gsi_crcchksm_error;
+
+ u64 packet_too_big_discard;
+ u64 no_buff_discard;
+
+ u64 rcv_ucast_bytes;
+ u64 rcv_mcast_bytes;
+ u64 rcv_bcast_bytes;
+ u64 rcv_ucast_pkts;
+ u64 rcv_mcast_pkts;
+ u64 rcv_bcast_pkts;
+
+ u64 sent_ucast_bytes;
+ u64 sent_mcast_bytes;
+ u64 sent_bcast_bytes;
+ u64 sent_ucast_pkts;
+ u64 sent_mcast_pkts;
+ u64 sent_bcast_pkts;
+};
+
+#define QED_LL2_UNUSED_HANDLE (0xff)
+
+struct qed_ll2_cb_ops {
+ int (*rx_cb)(void *, struct sk_buff *, u32, u32);
+ int (*tx_cb)(void *, struct sk_buff *, bool);
+};
+
+struct qed_ll2_params {
+ u16 mtu;
+ bool drop_ttl0_packets;
+ bool rx_vlan_stripping;
+ u8 tx_tc;
+ bool frags_mapped;
+ u8 ll2_mac_address[ETH_ALEN];
+};
+
+struct qed_ll2_ops {
+/**
+ * @brief start - initializes ll2
+ *
+ * @param cdev
+ * @param params - protocol driver configuration for the ll2.
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ int (*start)(struct qed_dev *cdev, struct qed_ll2_params *params);
+
+/**
+ * @brief stop - stops the ll2
+ *
+ * @param cdev
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ int (*stop)(struct qed_dev *cdev);
+
+/**
+ * @brief start_xmit - transmits an skb over the ll2 interface
+ *
+ * @param cdev
+ * @param skb
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ int (*start_xmit)(struct qed_dev *cdev, struct sk_buff *skb);
+
+/**
+ * @brief register_cb_ops - protocol driver register the callback for Rx/Tx
+ * packets. Should be called before `start'.
+ *
+ * @param cdev
+ * @param cookie - to be passed to the callback functions.
+ * @param ops - the callback functions to register for Rx / Tx.
+ *
+ * @return 0 on success, otherwise error value.
+ */
+ void (*register_cb_ops)(struct qed_dev *cdev,
+ const struct qed_ll2_cb_ops *ops,
+ void *cookie);
+
+/**
+ * @brief get LL2 related statistics
+ *
+ * @param cdev
+ * @param stats - pointer to struct that would be filled with stats
+ *
+ * @return 0 on success, error otherwise.
+ */
+ int (*get_stats)(struct qed_dev *cdev, struct qed_ll2_stats *stats);
+};
+
+#ifdef CONFIG_QED_LL2
+int qed_ll2_alloc_if(struct qed_dev *);
+void qed_ll2_dealloc_if(struct qed_dev *);
+#else
+static const struct qed_ll2_ops qed_ll2_ops_pass = {
+ .start = NULL,
+ .stop = NULL,
+ .start_xmit = NULL,
+ .register_cb_ops = NULL,
+ .get_stats = NULL,
+};
+
+static inline int qed_ll2_alloc_if(struct qed_dev *cdev)
+{
+ return 0;
+}
+
+static inline void qed_ll2_dealloc_if(struct qed_dev *cdev)
+{
+}
+#endif
+#endif
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 02/11] Add RoCE driver framework
[not found] ` <1473696465-27986-1-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` Ram Amrani
` (4 subsequent siblings)
5 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
Ram Amrani
Adds a skeletal implementation of the qed* RoCE driver -
basically the ability to communicate with the qede driver and
receive notifications from it regarding various init/exit events.
Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/Kconfig | 2 +
drivers/infiniband/hw/Makefile | 1 +
drivers/infiniband/hw/qedr/Kconfig | 7 +
drivers/infiniband/hw/qedr/Makefile | 3 +
drivers/infiniband/hw/qedr/main.c | 293 +++++++++++++++++++++++++
drivers/infiniband/hw/qedr/qedr.h | 60 ++++++
drivers/net/ethernet/qlogic/qede/Makefile | 1 +
drivers/net/ethernet/qlogic/qede/qede.h | 9 +
drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 +++++++++++++++++++++++++++
include/linux/qed/qed_if.h | 3 +-
include/linux/qed/qede_roce.h | 88 ++++++++
include/uapi/linux/pci_regs.h | 3 +
13 files changed, 803 insertions(+), 11 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/Kconfig
create mode 100644 drivers/infiniband/hw/qedr/Makefile
create mode 100644 drivers/infiniband/hw/qedr/main.c
create mode 100644 drivers/infiniband/hw/qedr/qedr.h
create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
create mode 100644 include/linux/qed/qede_roce.h
diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index e9b7dc0..77ab0f3 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -88,4 +88,6 @@ source "drivers/infiniband/sw/rxe/Kconfig"
source "drivers/infiniband/hw/hfi1/Kconfig"
+source "drivers/infiniband/hw/qedr/Kconfig"
+
endif # INFINIBAND
diff --git a/drivers/infiniband/hw/Makefile b/drivers/infiniband/hw/Makefile
index c0c7cf8..c51e464 100644
--- a/drivers/infiniband/hw/Makefile
+++ b/drivers/infiniband/hw/Makefile
@@ -9,3 +9,4 @@ obj-$(CONFIG_INFINIBAND_NES) += nes/
obj-$(CONFIG_INFINIBAND_OCRDMA) += ocrdma/
obj-$(CONFIG_INFINIBAND_USNIC) += usnic/
obj-$(CONFIG_INFINIBAND_HFI1) += hfi1/
+obj-$(CONFIG_INFINIBAND_QEDR) += qedr/
diff --git a/drivers/infiniband/hw/qedr/Kconfig b/drivers/infiniband/hw/qedr/Kconfig
new file mode 100644
index 0000000..58b584f
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/Kconfig
@@ -0,0 +1,7 @@
+config INFINIBAND_QEDR
+ tristate "QLogic RoCE driver"
+ depends on NETDEVICES && ETHERNET && PCI && 64BIT && QEDE
+ select QED_LL2
+ ---help---
+ This driver provides low-level InfiniBand over Ethernet
+ support for QLogic QED host channel adapters (HCAs).
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
new file mode 100644
index 0000000..3a5b7a2
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -0,0 +1,3 @@
+obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
+
+qedr-y := main.o
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
new file mode 100644
index 0000000..3fe58a3
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -0,0 +1,293 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/module.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_addr.h>
+#include <linux/netdevice.h>
+#include <linux/iommu.h>
+#include <net/addrconf.h>
+#include <linux/qed/qede_roce.h>
+#include "qedr.h"
+
+MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
+MODULE_AUTHOR("QLogic Corporation");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(QEDR_MODULE_VERSION);
+
+uint debug;
+module_param(debug, uint, 0);
+MODULE_PARM_DESC(debug, "Default debug msglevel");
+
+static LIST_HEAD(qedr_dev_list);
+static DEFINE_SPINLOCK(qedr_devlist_lock);
+
+void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
+ enum ib_event_type type)
+{
+ struct ib_event ibev;
+
+ ibev.device = &dev->ibdev;
+ ibev.element.port_num = port_num;
+ ibev.event = type;
+
+ ib_dispatch_event(&ibev);
+}
+
+static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
+ u8 port_num)
+{
+ return IB_LINK_LAYER_ETHERNET;
+}
+
+static int qedr_register_device(struct qedr_dev *dev)
+{
+ strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
+
+ memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
+ dev->ibdev.owner = THIS_MODULE;
+
+ dev->ibdev.get_link_layer = qedr_link_layer;
+
+ return 0;
+}
+
+/* QEDR sysfs interface */
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+ char *buf)
+{
+ struct qedr_dev *dev = dev_get_drvdata(device);
+
+ return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
+}
+
+static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
+ char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", "FW_VER_TO_ADD");
+}
+
+static ssize_t show_hca_type(struct device *device,
+ struct device_attribute *attr, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
+}
+
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
+
+static struct device_attribute *qedr_attributes[] = {
+ &dev_attr_hw_rev,
+ &dev_attr_fw_ver,
+ &dev_attr_hca_type
+};
+
+static void qedr_remove_sysfiles(struct qedr_dev *dev)
+{
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
+ device_remove_file(&dev->ibdev.dev, qedr_attributes[i]);
+}
+
+void qedr_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
+{
+ *p_dp_level = 0;
+ *p_dp_module = 0;
+
+ if (debug & QED_LOG_VERBOSE_MASK) {
+ *p_dp_level = QED_LEVEL_VERBOSE;
+ *p_dp_module = (debug & 0x3FFFFFFF);
+ } else if (debug & QED_LOG_INFO_MASK) {
+ *p_dp_level = QED_LEVEL_INFO;
+ } else if (debug & QED_LOG_NOTICE_MASK) {
+ *p_dp_level = QED_LEVEL_NOTICE;
+ }
+}
+
+static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
+{
+ struct pci_dev *bridge;
+ u32 val;
+
+ dev->atomic_cap = IB_ATOMIC_NONE;
+
+ bridge = pdev->bus->self;
+ if (!bridge)
+ return;
+
+ /* Check whether we are connected directly or via a switch */
+ while (bridge && bridge->bus->parent) {
+ DP_NOTICE(dev,
+ "Device is not connected directly to root. bridge->bus->number=%d primary=%d\n",
+ bridge->bus->number, bridge->bus->primary);
+ /* Need to check Atomic Op Routing Supported all the way to
+ * root complex.
+ */
+ pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
+ if (!(val & PCI_EXP_DEVCAP2_ATOMIC_ROUTE)) {
+ pcie_capability_clear_word(pdev,
+ PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ return;
+ }
+ bridge = bridge->bus->parent->self;
+ }
+ bridge = pdev->bus->self;
+
+ /* according to bridge capability */
+ pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
+ if (val & PCI_EXP_DEVCAP2_ATOMIC_COMP64) {
+ pcie_capability_set_word(pdev, PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ dev->atomic_cap = IB_ATOMIC_GLOB;
+ } else {
+ pcie_capability_clear_word(pdev, PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ }
+}
+
+static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
+ struct net_device *ndev)
+{
+ struct qedr_dev *dev;
+ int rc = 0, i;
+
+ dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
+ if (!dev) {
+ pr_err("Unable to allocate ib device\n");
+ return NULL;
+ }
+
+ qedr_config_debug(debug, &dev->dp_module, &dev->dp_level);
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr add device called\n");
+
+ dev->pdev = pdev;
+ dev->ndev = ndev;
+ dev->cdev = cdev;
+
+ qedr_pci_set_atomic(dev, pdev);
+
+ rc = qedr_register_device(dev);
+ if (rc) {
+ DP_ERR(dev, "Unable to allocate register device\n");
+ goto init_err;
+ }
+
+ for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
+ if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
+ goto init_err;
+
+ spin_lock(&qedr_devlist_lock);
+ list_add_tail_rcu(&dev->entry, &qedr_dev_list);
+ spin_unlock(&qedr_devlist_lock);
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
+ return dev;
+
+init_err:
+ ib_dealloc_device(&dev->ibdev);
+ DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
+
+ return NULL;
+}
+
+static void qedr_remove(struct qedr_dev *dev)
+{
+ /* First unregister with stack to stop all the active traffic
+ * of the registered clients.
+ */
+ qedr_remove_sysfiles(dev);
+
+ spin_lock(&qedr_devlist_lock);
+ list_del_rcu(&dev->entry);
+ spin_unlock(&qedr_devlist_lock);
+
+ ib_dealloc_device(&dev->ibdev);
+}
+
+static int qedr_close(struct qedr_dev *dev)
+{
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
+
+ return 0;
+}
+
+static void qedr_shutdown(struct qedr_dev *dev)
+{
+ qedr_close(dev);
+ qedr_remove(dev);
+}
+
+/* event handling via NIC driver ensures that all the NIC specific
+ * initialization done before RoCE driver notifies
+ * event to stack.
+ */
+static void qedr_notify(struct qedr_dev *dev, enum qede_roce_event event)
+{
+ switch (event) {
+ case QEDE_UP:
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
+ break;
+ case QEDE_DOWN:
+ qedr_close(dev);
+ break;
+ case QEDE_CLOSE:
+ qedr_shutdown(dev);
+ break;
+ case QEDE_CHANGE_ADDR:
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+ break;
+ default:
+ pr_err("Event not supported\n");
+ }
+}
+
+static struct qedr_driver qedr_drv = {
+ .name = "qedr_driver",
+ .add = qedr_add,
+ .remove = qedr_remove,
+ .notify = qedr_notify,
+};
+
+static int __init qedr_init_module(void)
+{
+ return qede_roce_register_driver(&qedr_drv);
+}
+
+static void __exit qedr_exit_module(void)
+{
+ qede_roce_unregister_driver(&qedr_drv);
+}
+
+module_init(qedr_init_module);
+module_exit(qedr_exit_module);
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
new file mode 100644
index 0000000..c3f93b5
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -0,0 +1,60 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QEDR_H__
+#define __QEDR_H__
+
+#include <linux/pci.h>
+#include <rdma/ib_addr.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qede_roce.h>
+
+#define QEDR_MODULE_VERSION "8.10.10.0"
+#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
+#define DP_NAME(dev) ((dev)->ibdev.name)
+
+enum DP_QEDR_MODULE {
+ QEDR_MSG_INIT = 0x10000,
+};
+
+struct qedr_dev {
+ struct ib_device ibdev;
+ struct qed_dev *cdev;
+ struct pci_dev *pdev;
+ struct net_device *ndev;
+ struct list_head entry;
+
+ enum ib_atomic_cap atomic_cap;
+
+ u32 dp_module;
+ u8 dp_level;
+};
+#endif
diff --git a/drivers/net/ethernet/qlogic/qede/Makefile b/drivers/net/ethernet/qlogic/qede/Makefile
index 74a4985..28dc589 100644
--- a/drivers/net/ethernet/qlogic/qede/Makefile
+++ b/drivers/net/ethernet/qlogic/qede/Makefile
@@ -2,3 +2,4 @@ obj-$(CONFIG_QEDE) := qede.o
qede-y := qede_main.o qede_ethtool.o
qede-$(CONFIG_DCB) += qede_dcbnl.o
+qede-$(CONFIG_INFINIBAND_QEDR) += qede_roce.o
diff --git a/drivers/net/ethernet/qlogic/qede/qede.h b/drivers/net/ethernet/qlogic/qede/qede.h
index e01adce..28c0e9f 100644
--- a/drivers/net/ethernet/qlogic/qede/qede.h
+++ b/drivers/net/ethernet/qlogic/qede/qede.h
@@ -106,6 +106,13 @@ struct qede_vlan {
bool configured;
};
+struct qede_rdma_dev {
+ struct qedr_dev *qedr_dev;
+ struct list_head entry;
+ struct list_head roce_event_list;
+ struct workqueue_struct *roce_wq;
+};
+
struct qede_dev {
struct qed_dev *cdev;
struct net_device *ndev;
@@ -185,6 +192,8 @@ struct qede_dev {
unsigned long sp_flags;
u16 vxlan_dst_port;
u16 geneve_dst_port;
+
+ struct qede_rdma_dev rdma_info;
};
enum QEDE_STATE {
diff --git a/drivers/net/ethernet/qlogic/qede/qede_main.c b/drivers/net/ethernet/qlogic/qede/qede_main.c
index 4056219..4f744ee 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_main.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_main.c
@@ -36,7 +36,7 @@
#include <linux/random.h>
#include <net/ip6_checksum.h>
#include <linux/bitops.h>
-
+#include <linux/qed/qede_roce.h>
#include "qede.h"
static char version[] =
@@ -189,8 +189,7 @@ static int qede_netdev_event(struct notifier_block *this, unsigned long event,
struct ethtool_drvinfo drvinfo;
struct qede_dev *edev;
- /* Currently only support name change */
- if (event != NETDEV_CHANGENAME)
+ if ((event != NETDEV_CHANGENAME) && (event != NETDEV_CHANGEADDR))
goto done;
/* Check whether this is a qede device */
@@ -203,11 +202,18 @@ static int qede_netdev_event(struct notifier_block *this, unsigned long event,
goto done;
edev = netdev_priv(ndev);
- /* Notify qed of the name change */
- if (!edev->ops || !edev->ops->common)
- goto done;
- edev->ops->common->set_id(edev->cdev, edev->ndev->name,
- "qede");
+ switch (event) {
+ case NETDEV_CHANGENAME:
+ /* Notify qed of the name change */
+ if (!edev->ops || !edev->ops->common)
+ goto done;
+ edev->ops->common->set_id(edev->cdev, edev->ndev->name, "qede");
+ break;
+ case NETDEV_CHANGEADDR:
+ edev = netdev_priv(ndev);
+ qede_roce_event_changeaddr(edev);
+ break;
+ }
done:
return NOTIFY_DONE;
@@ -2538,10 +2544,14 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
qede_init_ndev(edev);
+ rc = qede_roce_dev_add(edev);
+ if (rc)
+ goto err3;
+
rc = register_netdev(edev->ndev);
if (rc) {
DP_NOTICE(edev, "Cannot register net-device\n");
- goto err3;
+ goto err4;
}
edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
@@ -2560,6 +2570,8 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
return 0;
+err4:
+ qede_roce_dev_remove(edev);
err3:
free_netdev(edev->ndev);
err2:
@@ -2606,8 +2618,11 @@ static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
DP_INFO(edev, "Starting qede_remove\n");
cancel_delayed_work_sync(&edev->sp_task);
+
unregister_netdev(ndev);
+ qede_roce_dev_remove(edev);
+
edev->ops->common->set_power_state(cdev, PCI_D0);
pci_set_drvdata(pdev, NULL);
@@ -3504,6 +3519,7 @@ static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
DP_INFO(edev, "Starting qede unload\n");
+ qede_roce_dev_event_close(edev);
mutex_lock(&edev->qede_lock);
edev->state = QEDE_STATE_CLOSED;
@@ -3604,6 +3620,7 @@ static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
/* Query whether link is already-up */
memset(&link_output, 0, sizeof(link_output));
edev->ops->common->get_link(edev->cdev, &link_output);
+ qede_roce_dev_event_open(edev);
qede_link_update(edev, &link_output);
DP_INFO(edev, "Ending successfully qede load\n");
diff --git a/drivers/net/ethernet/qlogic/qede/qede_roce.c b/drivers/net/ethernet/qlogic/qede/qede_roce.c
new file mode 100644
index 0000000..b7e0ed9
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qede/qede_roce.c
@@ -0,0 +1,309 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/pci.h>
+#include <linux/netdevice.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/qede/qede_roce.h>
+#include "qede.h"
+
+static struct qedr_driver *qedr_drv;
+static LIST_HEAD(qedr_dev_list);
+static DEFINE_MUTEX(qedr_dev_list_lock);
+
+bool qede_roce_supported(struct qede_dev *dev)
+{
+ return dev->dev_info.common.rdma_supported;
+}
+
+static void _qede_roce_dev_add(struct qede_dev *edev)
+{
+ if (!qedr_drv)
+ return;
+
+ edev->rdma_info.qedr_dev = qedr_drv->add(edev->cdev, edev->pdev,
+ edev->ndev);
+}
+
+static int qede_roce_create_wq(struct qede_dev *edev)
+{
+ INIT_LIST_HEAD(&edev->rdma_info.roce_event_list);
+ edev->rdma_info.roce_wq = create_singlethread_workqueue("roce_wq");
+ if (!edev->rdma_info.roce_wq) {
+ pr_err("qedr: Could not create workqueue\n");
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+static void qede_roce_cleanup_event(struct qede_dev *edev)
+{
+ struct list_head *head = &edev->rdma_info.roce_event_list;
+ struct qede_roce_event_work *event_node;
+
+ flush_workqueue(edev->rdma_info.roce_wq);
+ while (!list_empty(head)) {
+ event_node = list_entry(head->next, struct qede_roce_event_work,
+ list);
+ cancel_work_sync(&event_node->work);
+ list_del(&event_node->list);
+ kfree(event_node);
+ }
+}
+
+static void qede_roce_destroy_wq(struct qede_dev *edev)
+{
+ qede_roce_cleanup_event(edev);
+ destroy_workqueue(edev->rdma_info.roce_wq);
+}
+
+int qede_roce_dev_add(struct qede_dev *edev)
+{
+ int rc = 0;
+
+ if (qede_roce_supported(edev)) {
+ rc = qede_roce_create_wq(edev);
+ if (rc)
+ return rc;
+
+ INIT_LIST_HEAD(&edev->rdma_info.entry);
+ mutex_lock(&qedr_dev_list_lock);
+ list_add_tail(&edev->rdma_info.entry, &qedr_dev_list);
+ _qede_roce_dev_add(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+ }
+
+ return rc;
+}
+
+static void _qede_roce_dev_remove(struct qede_dev *edev)
+{
+ if (qedr_drv && qedr_drv->remove && edev->rdma_info.qedr_dev)
+ qedr_drv->remove(edev->rdma_info.qedr_dev);
+ edev->rdma_info.qedr_dev = NULL;
+}
+
+void qede_roce_dev_remove(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ qede_roce_destroy_wq(edev);
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_remove(edev);
+ list_del(&edev->rdma_info.entry);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+static void _qede_roce_dev_open(struct qede_dev *edev)
+{
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_UP);
+}
+
+void qede_roce_dev_open(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_open(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+static void _qede_roce_dev_close(struct qede_dev *edev)
+{
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_DOWN);
+}
+
+void qede_roce_dev_close(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_close(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+void qede_roce_dev_shutdown(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_CLOSE);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+int qede_roce_register_driver(struct qedr_driver *drv)
+{
+ struct qede_dev *edev;
+ u8 qedr_counter = 0;
+
+ mutex_lock(&qedr_dev_list_lock);
+ if (qedr_drv) {
+ mutex_unlock(&qedr_dev_list_lock);
+ return -EINVAL;
+ }
+ qedr_drv = drv;
+
+ list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) {
+ struct net_device *netdev;
+
+ qedr_counter++;
+ _qede_roce_dev_add(edev);
+ netdev = edev->ndev;
+ if (netif_running(netdev) && netif_oper_up(netdev))
+ _qede_roce_dev_open(edev);
+ }
+ mutex_unlock(&qedr_dev_list_lock);
+
+ pr_err("qedr: discovered and registered %d RoCE funcs\n", qedr_counter);
+
+ return 0;
+}
+EXPORT_SYMBOL(qede_roce_register_driver);
+
+void qede_roce_unregister_driver(struct qedr_driver *drv)
+{
+ struct qede_dev *edev;
+
+ mutex_lock(&qedr_dev_list_lock);
+ list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) {
+ if (edev->rdma_info.qedr_dev)
+ _qede_roce_dev_remove(edev);
+ }
+ qedr_drv = NULL;
+ mutex_unlock(&qedr_dev_list_lock);
+}
+EXPORT_SYMBOL(qede_roce_unregister_driver);
+
+void qede_roce_changeaddr(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_CHANGE_ADDR);
+}
+
+struct qede_roce_event_work *qede_roce_get_free_event_node(struct qede_dev
+ *edev)
+{
+ struct qede_roce_event_work *event_node = NULL;
+ struct list_head *list_node = NULL;
+ bool found = false;
+
+ list_for_each(list_node, &edev->rdma_info.roce_event_list) {
+ event_node = list_entry(list_node, struct qede_roce_event_work,
+ list);
+ if (!work_pending(&event_node->work)) {
+ found = true;
+ break;
+ }
+ }
+
+ if (!found) {
+ event_node = kzalloc(sizeof(*event_node), GFP_KERNEL);
+ if (!event_node) {
+ pr_err("qedr: Could not allocate memory for roce work\n");
+ return NULL;
+ }
+ list_add_tail(&event_node->list,
+ &edev->rdma_info.roce_event_list);
+ }
+
+ return event_node;
+}
+
+static void qede_roce_handle_event(struct work_struct *work)
+{
+ struct qede_roce_event_work *event_node;
+ enum qede_roce_event event;
+ struct qede_dev *edev;
+
+ event_node = container_of(work, struct qede_roce_event_work, work);
+ event = event_node->event;
+ edev = event_node->ptr;
+
+ switch (event) {
+ case QEDE_UP:
+ qede_roce_dev_open(edev);
+ break;
+ case QEDE_DOWN:
+ qede_roce_dev_close(edev);
+ break;
+ case QEDE_CLOSE:
+ qede_roce_dev_shutdown(edev);
+ break;
+ case QEDE_CHANGE_ADDR:
+ qede_roce_changeaddr(edev);
+ break;
+ default:
+ pr_err("qede: Invalid roce event %d", event);
+ }
+}
+
+static void qede_roce_add_event(struct qede_dev *edev,
+ enum qede_roce_event event)
+{
+ struct qede_roce_event_work *event_node;
+
+ event_node = qede_roce_get_free_event_node(edev);
+ if (!event_node)
+ return;
+
+ event_node->event = event;
+ event_node->ptr = edev;
+
+ INIT_WORK(&event_node->work, qede_roce_handle_event);
+ queue_work(edev->rdma_info.roce_wq, &event_node->work);
+}
+
+void qede_roce_dev_event_open(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_UP);
+}
+
+void qede_roce_dev_event_close(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_DOWN);
+}
+
+void qede_roce_event_changeaddr(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_CHANGE_ADDR);
+}
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index 6611441..cb899b2 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -260,11 +260,10 @@ struct qed_dev_info {
/* MFW version */
u32 mfw_rev;
- bool rdma_supported;
-
u32 flash_size;
u8 mf_mode;
bool tx_switching;
+ bool rdma_supported;
};
enum qed_sb_type {
diff --git a/include/linux/qed/qede_roce.h b/include/linux/qed/qede_roce.h
new file mode 100644
index 0000000..fd7acc1
--- /dev/null
+++ b/include/linux/qed/qede_roce.h
@@ -0,0 +1,88 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef QEDE_ROCE_H
+#define QEDE_ROCE_H
+
+struct qedr_dev;
+struct qed_dev;
+struct qede_dev;
+
+enum qede_roce_event {
+ QEDE_UP,
+ QEDE_DOWN,
+ QEDE_CHANGE_ADDR,
+ QEDE_CLOSE
+};
+
+struct qede_roce_event_work {
+ struct list_head list;
+ struct work_struct work;
+ void *ptr;
+ enum qede_roce_event event;
+};
+
+struct qedr_driver {
+ unsigned char name[32];
+
+ struct qedr_dev* (*add)(struct qed_dev *, struct pci_dev *,
+ struct net_device *);
+
+ void (*remove)(struct qedr_dev *);
+ void (*notify)(struct qedr_dev *, enum qede_roce_event);
+};
+
+/* APIs for RoCE driver to register callback handlers,
+ * which will be invoked when device is added, removed, ifup, ifdown
+ */
+int qede_roce_register_driver(struct qedr_driver *drv);
+void qede_roce_unregister_driver(struct qedr_driver *drv);
+
+bool qede_roce_supported(struct qede_dev *dev);
+
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+int qede_roce_dev_add(struct qede_dev *dev);
+void qede_roce_dev_event_open(struct qede_dev *dev);
+void qede_roce_dev_event_close(struct qede_dev *dev);
+void qede_roce_dev_remove(struct qede_dev *dev);
+void qede_roce_event_changeaddr(struct qede_dev *qedr);
+#else
+static int qede_roce_dev_add(struct qede_dev *dev)
+{
+ return 0;
+}
+
+static void qede_roce_dev_event_open(struct qede_dev *dev) {}
+static void qede_roce_dev_event_close(struct qede_dev *dev) {}
+static void qede_roce_dev_remove(struct qede_dev *dev) {}
+static void qede_roce_event_changeaddr(struct qede_dev *qedr) {}
+#endif
+#endif
diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
index 4040951..3973000 100644
--- a/include/uapi/linux/pci_regs.h
+++ b/include/uapi/linux/pci_regs.h
@@ -612,6 +612,8 @@
*/
#define PCI_EXP_DEVCAP2 36 /* Device Capabilities 2 */
#define PCI_EXP_DEVCAP2_ARI 0x00000020 /* Alternative Routing-ID */
+#define PCI_EXP_DEVCAP2_ATOMIC_ROUTE 0x00000040 /* Atomic Op routing */
+#define PCI_EXP_DEVCAP2_ATOMIC_COMP64 0x00000100 /* Atomic 64-bit comparison */
#define PCI_EXP_DEVCAP2_LTR 0x00000800 /* Latency tolerance reporting */
#define PCI_EXP_DEVCAP2_OBFF_MASK 0x000c0000 /* OBFF support mechanism */
#define PCI_EXP_DEVCAP2_OBFF_MSG 0x00040000 /* New message signaling */
@@ -619,6 +621,7 @@
#define PCI_EXP_DEVCTL2 40 /* Device Control 2 */
#define PCI_EXP_DEVCTL2_COMP_TIMEOUT 0x000f /* Completion Timeout Value */
#define PCI_EXP_DEVCTL2_ARI 0x0020 /* Alternative Routing-ID */
+#define PCI_EXP_DEVCTL2_ATOMIC_REQ 0x0040 /* Set Atomic requests */
#define PCI_EXP_DEVCTL2_IDO_REQ_EN 0x0100 /* Allow IDO for requests */
#define PCI_EXP_DEVCTL2_IDO_CMP_EN 0x0200 /* Allow IDO for completions */
#define PCI_EXP_DEVCTL2_LTR_EN 0x0400 /* Enable LTR mechanism */
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 02/11] Add RoCE driver framework
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
Ram Amrani
Adds a skeletal implementation of the qed* RoCE driver -
basically the ability to communicate with the qede driver and
receive notifications from it regarding various init/exit events.
Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/Kconfig | 2 +
drivers/infiniband/hw/Makefile | 1 +
drivers/infiniband/hw/qedr/Kconfig | 7 +
drivers/infiniband/hw/qedr/Makefile | 3 +
drivers/infiniband/hw/qedr/main.c | 293 +++++++++++++++++++++++++
drivers/infiniband/hw/qedr/qedr.h | 60 ++++++
drivers/net/ethernet/qlogic/qede/Makefile | 1 +
drivers/net/ethernet/qlogic/qede/qede.h | 9 +
drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 +++++++++++++++++++++++++++
include/linux/qed/qed_if.h | 3 +-
include/linux/qed/qede_roce.h | 88 ++++++++
include/uapi/linux/pci_regs.h | 3 +
13 files changed, 803 insertions(+), 11 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/Kconfig
create mode 100644 drivers/infiniband/hw/qedr/Makefile
create mode 100644 drivers/infiniband/hw/qedr/main.c
create mode 100644 drivers/infiniband/hw/qedr/qedr.h
create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
create mode 100644 include/linux/qed/qede_roce.h
diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index e9b7dc0..77ab0f3 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -88,4 +88,6 @@ source "drivers/infiniband/sw/rxe/Kconfig"
source "drivers/infiniband/hw/hfi1/Kconfig"
+source "drivers/infiniband/hw/qedr/Kconfig"
+
endif # INFINIBAND
diff --git a/drivers/infiniband/hw/Makefile b/drivers/infiniband/hw/Makefile
index c0c7cf8..c51e464 100644
--- a/drivers/infiniband/hw/Makefile
+++ b/drivers/infiniband/hw/Makefile
@@ -9,3 +9,4 @@ obj-$(CONFIG_INFINIBAND_NES) += nes/
obj-$(CONFIG_INFINIBAND_OCRDMA) += ocrdma/
obj-$(CONFIG_INFINIBAND_USNIC) += usnic/
obj-$(CONFIG_INFINIBAND_HFI1) += hfi1/
+obj-$(CONFIG_INFINIBAND_QEDR) += qedr/
diff --git a/drivers/infiniband/hw/qedr/Kconfig b/drivers/infiniband/hw/qedr/Kconfig
new file mode 100644
index 0000000..58b584f
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/Kconfig
@@ -0,0 +1,7 @@
+config INFINIBAND_QEDR
+ tristate "QLogic RoCE driver"
+ depends on NETDEVICES && ETHERNET && PCI && 64BIT && QEDE
+ select QED_LL2
+ ---help---
+ This driver provides low-level InfiniBand over Ethernet
+ support for QLogic QED host channel adapters (HCAs).
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
new file mode 100644
index 0000000..3a5b7a2
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -0,0 +1,3 @@
+obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
+
+qedr-y := main.o
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
new file mode 100644
index 0000000..3fe58a3
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -0,0 +1,293 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/module.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_addr.h>
+#include <linux/netdevice.h>
+#include <linux/iommu.h>
+#include <net/addrconf.h>
+#include <linux/qed/qede_roce.h>
+#include "qedr.h"
+
+MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
+MODULE_AUTHOR("QLogic Corporation");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(QEDR_MODULE_VERSION);
+
+uint debug;
+module_param(debug, uint, 0);
+MODULE_PARM_DESC(debug, "Default debug msglevel");
+
+static LIST_HEAD(qedr_dev_list);
+static DEFINE_SPINLOCK(qedr_devlist_lock);
+
+void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
+ enum ib_event_type type)
+{
+ struct ib_event ibev;
+
+ ibev.device = &dev->ibdev;
+ ibev.element.port_num = port_num;
+ ibev.event = type;
+
+ ib_dispatch_event(&ibev);
+}
+
+static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
+ u8 port_num)
+{
+ return IB_LINK_LAYER_ETHERNET;
+}
+
+static int qedr_register_device(struct qedr_dev *dev)
+{
+ strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
+
+ memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
+ dev->ibdev.owner = THIS_MODULE;
+
+ dev->ibdev.get_link_layer = qedr_link_layer;
+
+ return 0;
+}
+
+/* QEDR sysfs interface */
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+ char *buf)
+{
+ struct qedr_dev *dev = dev_get_drvdata(device);
+
+ return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
+}
+
+static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
+ char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", "FW_VER_TO_ADD");
+}
+
+static ssize_t show_hca_type(struct device *device,
+ struct device_attribute *attr, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
+}
+
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
+
+static struct device_attribute *qedr_attributes[] = {
+ &dev_attr_hw_rev,
+ &dev_attr_fw_ver,
+ &dev_attr_hca_type
+};
+
+static void qedr_remove_sysfiles(struct qedr_dev *dev)
+{
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
+ device_remove_file(&dev->ibdev.dev, qedr_attributes[i]);
+}
+
+void qedr_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
+{
+ *p_dp_level = 0;
+ *p_dp_module = 0;
+
+ if (debug & QED_LOG_VERBOSE_MASK) {
+ *p_dp_level = QED_LEVEL_VERBOSE;
+ *p_dp_module = (debug & 0x3FFFFFFF);
+ } else if (debug & QED_LOG_INFO_MASK) {
+ *p_dp_level = QED_LEVEL_INFO;
+ } else if (debug & QED_LOG_NOTICE_MASK) {
+ *p_dp_level = QED_LEVEL_NOTICE;
+ }
+}
+
+static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
+{
+ struct pci_dev *bridge;
+ u32 val;
+
+ dev->atomic_cap = IB_ATOMIC_NONE;
+
+ bridge = pdev->bus->self;
+ if (!bridge)
+ return;
+
+ /* Check whether we are connected directly or via a switch */
+ while (bridge && bridge->bus->parent) {
+ DP_NOTICE(dev,
+ "Device is not connected directly to root. bridge->bus->number=%d primary=%d\n",
+ bridge->bus->number, bridge->bus->primary);
+ /* Need to check Atomic Op Routing Supported all the way to
+ * root complex.
+ */
+ pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
+ if (!(val & PCI_EXP_DEVCAP2_ATOMIC_ROUTE)) {
+ pcie_capability_clear_word(pdev,
+ PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ return;
+ }
+ bridge = bridge->bus->parent->self;
+ }
+ bridge = pdev->bus->self;
+
+ /* according to bridge capability */
+ pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
+ if (val & PCI_EXP_DEVCAP2_ATOMIC_COMP64) {
+ pcie_capability_set_word(pdev, PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ dev->atomic_cap = IB_ATOMIC_GLOB;
+ } else {
+ pcie_capability_clear_word(pdev, PCI_EXP_DEVCTL2,
+ PCI_EXP_DEVCTL2_ATOMIC_REQ);
+ }
+}
+
+static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
+ struct net_device *ndev)
+{
+ struct qedr_dev *dev;
+ int rc = 0, i;
+
+ dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
+ if (!dev) {
+ pr_err("Unable to allocate ib device\n");
+ return NULL;
+ }
+
+ qedr_config_debug(debug, &dev->dp_module, &dev->dp_level);
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr add device called\n");
+
+ dev->pdev = pdev;
+ dev->ndev = ndev;
+ dev->cdev = cdev;
+
+ qedr_pci_set_atomic(dev, pdev);
+
+ rc = qedr_register_device(dev);
+ if (rc) {
+ DP_ERR(dev, "Unable to allocate register device\n");
+ goto init_err;
+ }
+
+ for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
+ if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
+ goto init_err;
+
+ spin_lock(&qedr_devlist_lock);
+ list_add_tail_rcu(&dev->entry, &qedr_dev_list);
+ spin_unlock(&qedr_devlist_lock);
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
+ return dev;
+
+init_err:
+ ib_dealloc_device(&dev->ibdev);
+ DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
+
+ return NULL;
+}
+
+static void qedr_remove(struct qedr_dev *dev)
+{
+ /* First unregister with stack to stop all the active traffic
+ * of the registered clients.
+ */
+ qedr_remove_sysfiles(dev);
+
+ spin_lock(&qedr_devlist_lock);
+ list_del_rcu(&dev->entry);
+ spin_unlock(&qedr_devlist_lock);
+
+ ib_dealloc_device(&dev->ibdev);
+}
+
+static int qedr_close(struct qedr_dev *dev)
+{
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
+
+ return 0;
+}
+
+static void qedr_shutdown(struct qedr_dev *dev)
+{
+ qedr_close(dev);
+ qedr_remove(dev);
+}
+
+/* event handling via NIC driver ensures that all the NIC specific
+ * initialization done before RoCE driver notifies
+ * event to stack.
+ */
+static void qedr_notify(struct qedr_dev *dev, enum qede_roce_event event)
+{
+ switch (event) {
+ case QEDE_UP:
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
+ break;
+ case QEDE_DOWN:
+ qedr_close(dev);
+ break;
+ case QEDE_CLOSE:
+ qedr_shutdown(dev);
+ break;
+ case QEDE_CHANGE_ADDR:
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+ break;
+ default:
+ pr_err("Event not supported\n");
+ }
+}
+
+static struct qedr_driver qedr_drv = {
+ .name = "qedr_driver",
+ .add = qedr_add,
+ .remove = qedr_remove,
+ .notify = qedr_notify,
+};
+
+static int __init qedr_init_module(void)
+{
+ return qede_roce_register_driver(&qedr_drv);
+}
+
+static void __exit qedr_exit_module(void)
+{
+ qede_roce_unregister_driver(&qedr_drv);
+}
+
+module_init(qedr_init_module);
+module_exit(qedr_exit_module);
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
new file mode 100644
index 0000000..c3f93b5
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -0,0 +1,60 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QEDR_H__
+#define __QEDR_H__
+
+#include <linux/pci.h>
+#include <rdma/ib_addr.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qede_roce.h>
+
+#define QEDR_MODULE_VERSION "8.10.10.0"
+#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
+#define DP_NAME(dev) ((dev)->ibdev.name)
+
+enum DP_QEDR_MODULE {
+ QEDR_MSG_INIT = 0x10000,
+};
+
+struct qedr_dev {
+ struct ib_device ibdev;
+ struct qed_dev *cdev;
+ struct pci_dev *pdev;
+ struct net_device *ndev;
+ struct list_head entry;
+
+ enum ib_atomic_cap atomic_cap;
+
+ u32 dp_module;
+ u8 dp_level;
+};
+#endif
diff --git a/drivers/net/ethernet/qlogic/qede/Makefile b/drivers/net/ethernet/qlogic/qede/Makefile
index 74a4985..28dc589 100644
--- a/drivers/net/ethernet/qlogic/qede/Makefile
+++ b/drivers/net/ethernet/qlogic/qede/Makefile
@@ -2,3 +2,4 @@ obj-$(CONFIG_QEDE) := qede.o
qede-y := qede_main.o qede_ethtool.o
qede-$(CONFIG_DCB) += qede_dcbnl.o
+qede-$(CONFIG_INFINIBAND_QEDR) += qede_roce.o
diff --git a/drivers/net/ethernet/qlogic/qede/qede.h b/drivers/net/ethernet/qlogic/qede/qede.h
index e01adce..28c0e9f 100644
--- a/drivers/net/ethernet/qlogic/qede/qede.h
+++ b/drivers/net/ethernet/qlogic/qede/qede.h
@@ -106,6 +106,13 @@ struct qede_vlan {
bool configured;
};
+struct qede_rdma_dev {
+ struct qedr_dev *qedr_dev;
+ struct list_head entry;
+ struct list_head roce_event_list;
+ struct workqueue_struct *roce_wq;
+};
+
struct qede_dev {
struct qed_dev *cdev;
struct net_device *ndev;
@@ -185,6 +192,8 @@ struct qede_dev {
unsigned long sp_flags;
u16 vxlan_dst_port;
u16 geneve_dst_port;
+
+ struct qede_rdma_dev rdma_info;
};
enum QEDE_STATE {
diff --git a/drivers/net/ethernet/qlogic/qede/qede_main.c b/drivers/net/ethernet/qlogic/qede/qede_main.c
index 4056219..4f744ee 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_main.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_main.c
@@ -36,7 +36,7 @@
#include <linux/random.h>
#include <net/ip6_checksum.h>
#include <linux/bitops.h>
-
+#include <linux/qed/qede_roce.h>
#include "qede.h"
static char version[] =
@@ -189,8 +189,7 @@ static int qede_netdev_event(struct notifier_block *this, unsigned long event,
struct ethtool_drvinfo drvinfo;
struct qede_dev *edev;
- /* Currently only support name change */
- if (event != NETDEV_CHANGENAME)
+ if ((event != NETDEV_CHANGENAME) && (event != NETDEV_CHANGEADDR))
goto done;
/* Check whether this is a qede device */
@@ -203,11 +202,18 @@ static int qede_netdev_event(struct notifier_block *this, unsigned long event,
goto done;
edev = netdev_priv(ndev);
- /* Notify qed of the name change */
- if (!edev->ops || !edev->ops->common)
- goto done;
- edev->ops->common->set_id(edev->cdev, edev->ndev->name,
- "qede");
+ switch (event) {
+ case NETDEV_CHANGENAME:
+ /* Notify qed of the name change */
+ if (!edev->ops || !edev->ops->common)
+ goto done;
+ edev->ops->common->set_id(edev->cdev, edev->ndev->name, "qede");
+ break;
+ case NETDEV_CHANGEADDR:
+ edev = netdev_priv(ndev);
+ qede_roce_event_changeaddr(edev);
+ break;
+ }
done:
return NOTIFY_DONE;
@@ -2538,10 +2544,14 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
qede_init_ndev(edev);
+ rc = qede_roce_dev_add(edev);
+ if (rc)
+ goto err3;
+
rc = register_netdev(edev->ndev);
if (rc) {
DP_NOTICE(edev, "Cannot register net-device\n");
- goto err3;
+ goto err4;
}
edev->ops->common->set_id(cdev, edev->ndev->name, DRV_MODULE_VERSION);
@@ -2560,6 +2570,8 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level,
return 0;
+err4:
+ qede_roce_dev_remove(edev);
err3:
free_netdev(edev->ndev);
err2:
@@ -2606,8 +2618,11 @@ static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode)
DP_INFO(edev, "Starting qede_remove\n");
cancel_delayed_work_sync(&edev->sp_task);
+
unregister_netdev(ndev);
+ qede_roce_dev_remove(edev);
+
edev->ops->common->set_power_state(cdev, PCI_D0);
pci_set_drvdata(pdev, NULL);
@@ -3504,6 +3519,7 @@ static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode)
DP_INFO(edev, "Starting qede unload\n");
+ qede_roce_dev_event_close(edev);
mutex_lock(&edev->qede_lock);
edev->state = QEDE_STATE_CLOSED;
@@ -3604,6 +3620,7 @@ static int qede_load(struct qede_dev *edev, enum qede_load_mode mode)
/* Query whether link is already-up */
memset(&link_output, 0, sizeof(link_output));
edev->ops->common->get_link(edev->cdev, &link_output);
+ qede_roce_dev_event_open(edev);
qede_link_update(edev, &link_output);
DP_INFO(edev, "Ending successfully qede load\n");
diff --git a/drivers/net/ethernet/qlogic/qede/qede_roce.c b/drivers/net/ethernet/qlogic/qede/qede_roce.c
new file mode 100644
index 0000000..b7e0ed9
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qede/qede_roce.c
@@ -0,0 +1,309 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/pci.h>
+#include <linux/netdevice.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/qede/qede_roce.h>
+#include "qede.h"
+
+static struct qedr_driver *qedr_drv;
+static LIST_HEAD(qedr_dev_list);
+static DEFINE_MUTEX(qedr_dev_list_lock);
+
+bool qede_roce_supported(struct qede_dev *dev)
+{
+ return dev->dev_info.common.rdma_supported;
+}
+
+static void _qede_roce_dev_add(struct qede_dev *edev)
+{
+ if (!qedr_drv)
+ return;
+
+ edev->rdma_info.qedr_dev = qedr_drv->add(edev->cdev, edev->pdev,
+ edev->ndev);
+}
+
+static int qede_roce_create_wq(struct qede_dev *edev)
+{
+ INIT_LIST_HEAD(&edev->rdma_info.roce_event_list);
+ edev->rdma_info.roce_wq = create_singlethread_workqueue("roce_wq");
+ if (!edev->rdma_info.roce_wq) {
+ pr_err("qedr: Could not create workqueue\n");
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+static void qede_roce_cleanup_event(struct qede_dev *edev)
+{
+ struct list_head *head = &edev->rdma_info.roce_event_list;
+ struct qede_roce_event_work *event_node;
+
+ flush_workqueue(edev->rdma_info.roce_wq);
+ while (!list_empty(head)) {
+ event_node = list_entry(head->next, struct qede_roce_event_work,
+ list);
+ cancel_work_sync(&event_node->work);
+ list_del(&event_node->list);
+ kfree(event_node);
+ }
+}
+
+static void qede_roce_destroy_wq(struct qede_dev *edev)
+{
+ qede_roce_cleanup_event(edev);
+ destroy_workqueue(edev->rdma_info.roce_wq);
+}
+
+int qede_roce_dev_add(struct qede_dev *edev)
+{
+ int rc = 0;
+
+ if (qede_roce_supported(edev)) {
+ rc = qede_roce_create_wq(edev);
+ if (rc)
+ return rc;
+
+ INIT_LIST_HEAD(&edev->rdma_info.entry);
+ mutex_lock(&qedr_dev_list_lock);
+ list_add_tail(&edev->rdma_info.entry, &qedr_dev_list);
+ _qede_roce_dev_add(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+ }
+
+ return rc;
+}
+
+static void _qede_roce_dev_remove(struct qede_dev *edev)
+{
+ if (qedr_drv && qedr_drv->remove && edev->rdma_info.qedr_dev)
+ qedr_drv->remove(edev->rdma_info.qedr_dev);
+ edev->rdma_info.qedr_dev = NULL;
+}
+
+void qede_roce_dev_remove(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ qede_roce_destroy_wq(edev);
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_remove(edev);
+ list_del(&edev->rdma_info.entry);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+static void _qede_roce_dev_open(struct qede_dev *edev)
+{
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_UP);
+}
+
+void qede_roce_dev_open(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_open(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+static void _qede_roce_dev_close(struct qede_dev *edev)
+{
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_DOWN);
+}
+
+void qede_roce_dev_close(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ _qede_roce_dev_close(edev);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+void qede_roce_dev_shutdown(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ mutex_lock(&qedr_dev_list_lock);
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_CLOSE);
+ mutex_unlock(&qedr_dev_list_lock);
+}
+
+int qede_roce_register_driver(struct qedr_driver *drv)
+{
+ struct qede_dev *edev;
+ u8 qedr_counter = 0;
+
+ mutex_lock(&qedr_dev_list_lock);
+ if (qedr_drv) {
+ mutex_unlock(&qedr_dev_list_lock);
+ return -EINVAL;
+ }
+ qedr_drv = drv;
+
+ list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) {
+ struct net_device *netdev;
+
+ qedr_counter++;
+ _qede_roce_dev_add(edev);
+ netdev = edev->ndev;
+ if (netif_running(netdev) && netif_oper_up(netdev))
+ _qede_roce_dev_open(edev);
+ }
+ mutex_unlock(&qedr_dev_list_lock);
+
+ pr_err("qedr: discovered and registered %d RoCE funcs\n", qedr_counter);
+
+ return 0;
+}
+EXPORT_SYMBOL(qede_roce_register_driver);
+
+void qede_roce_unregister_driver(struct qedr_driver *drv)
+{
+ struct qede_dev *edev;
+
+ mutex_lock(&qedr_dev_list_lock);
+ list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) {
+ if (edev->rdma_info.qedr_dev)
+ _qede_roce_dev_remove(edev);
+ }
+ qedr_drv = NULL;
+ mutex_unlock(&qedr_dev_list_lock);
+}
+EXPORT_SYMBOL(qede_roce_unregister_driver);
+
+void qede_roce_changeaddr(struct qede_dev *edev)
+{
+ if (!qede_roce_supported(edev))
+ return;
+
+ if (qedr_drv && edev->rdma_info.qedr_dev && qedr_drv->notify)
+ qedr_drv->notify(edev->rdma_info.qedr_dev, QEDE_CHANGE_ADDR);
+}
+
+struct qede_roce_event_work *qede_roce_get_free_event_node(struct qede_dev
+ *edev)
+{
+ struct qede_roce_event_work *event_node = NULL;
+ struct list_head *list_node = NULL;
+ bool found = false;
+
+ list_for_each(list_node, &edev->rdma_info.roce_event_list) {
+ event_node = list_entry(list_node, struct qede_roce_event_work,
+ list);
+ if (!work_pending(&event_node->work)) {
+ found = true;
+ break;
+ }
+ }
+
+ if (!found) {
+ event_node = kzalloc(sizeof(*event_node), GFP_KERNEL);
+ if (!event_node) {
+ pr_err("qedr: Could not allocate memory for roce work\n");
+ return NULL;
+ }
+ list_add_tail(&event_node->list,
+ &edev->rdma_info.roce_event_list);
+ }
+
+ return event_node;
+}
+
+static void qede_roce_handle_event(struct work_struct *work)
+{
+ struct qede_roce_event_work *event_node;
+ enum qede_roce_event event;
+ struct qede_dev *edev;
+
+ event_node = container_of(work, struct qede_roce_event_work, work);
+ event = event_node->event;
+ edev = event_node->ptr;
+
+ switch (event) {
+ case QEDE_UP:
+ qede_roce_dev_open(edev);
+ break;
+ case QEDE_DOWN:
+ qede_roce_dev_close(edev);
+ break;
+ case QEDE_CLOSE:
+ qede_roce_dev_shutdown(edev);
+ break;
+ case QEDE_CHANGE_ADDR:
+ qede_roce_changeaddr(edev);
+ break;
+ default:
+ pr_err("qede: Invalid roce event %d", event);
+ }
+}
+
+static void qede_roce_add_event(struct qede_dev *edev,
+ enum qede_roce_event event)
+{
+ struct qede_roce_event_work *event_node;
+
+ event_node = qede_roce_get_free_event_node(edev);
+ if (!event_node)
+ return;
+
+ event_node->event = event;
+ event_node->ptr = edev;
+
+ INIT_WORK(&event_node->work, qede_roce_handle_event);
+ queue_work(edev->rdma_info.roce_wq, &event_node->work);
+}
+
+void qede_roce_dev_event_open(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_UP);
+}
+
+void qede_roce_dev_event_close(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_DOWN);
+}
+
+void qede_roce_event_changeaddr(struct qede_dev *edev)
+{
+ qede_roce_add_event(edev, QEDE_CHANGE_ADDR);
+}
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index 6611441..cb899b2 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -260,11 +260,10 @@ struct qed_dev_info {
/* MFW version */
u32 mfw_rev;
- bool rdma_supported;
-
u32 flash_size;
u8 mf_mode;
bool tx_switching;
+ bool rdma_supported;
};
enum qed_sb_type {
diff --git a/include/linux/qed/qede_roce.h b/include/linux/qed/qede_roce.h
new file mode 100644
index 0000000..fd7acc1
--- /dev/null
+++ b/include/linux/qed/qede_roce.h
@@ -0,0 +1,88 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef QEDE_ROCE_H
+#define QEDE_ROCE_H
+
+struct qedr_dev;
+struct qed_dev;
+struct qede_dev;
+
+enum qede_roce_event {
+ QEDE_UP,
+ QEDE_DOWN,
+ QEDE_CHANGE_ADDR,
+ QEDE_CLOSE
+};
+
+struct qede_roce_event_work {
+ struct list_head list;
+ struct work_struct work;
+ void *ptr;
+ enum qede_roce_event event;
+};
+
+struct qedr_driver {
+ unsigned char name[32];
+
+ struct qedr_dev* (*add)(struct qed_dev *, struct pci_dev *,
+ struct net_device *);
+
+ void (*remove)(struct qedr_dev *);
+ void (*notify)(struct qedr_dev *, enum qede_roce_event);
+};
+
+/* APIs for RoCE driver to register callback handlers,
+ * which will be invoked when device is added, removed, ifup, ifdown
+ */
+int qede_roce_register_driver(struct qedr_driver *drv);
+void qede_roce_unregister_driver(struct qedr_driver *drv);
+
+bool qede_roce_supported(struct qede_dev *dev);
+
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+int qede_roce_dev_add(struct qede_dev *dev);
+void qede_roce_dev_event_open(struct qede_dev *dev);
+void qede_roce_dev_event_close(struct qede_dev *dev);
+void qede_roce_dev_remove(struct qede_dev *dev);
+void qede_roce_event_changeaddr(struct qede_dev *qedr);
+#else
+static int qede_roce_dev_add(struct qede_dev *dev)
+{
+ return 0;
+}
+
+static void qede_roce_dev_event_open(struct qede_dev *dev) {}
+static void qede_roce_dev_event_close(struct qede_dev *dev) {}
+static void qede_roce_dev_remove(struct qede_dev *dev) {}
+static void qede_roce_event_changeaddr(struct qede_dev *qedr) {}
+#endif
+#endif
diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
index 4040951..3973000 100644
--- a/include/uapi/linux/pci_regs.h
+++ b/include/uapi/linux/pci_regs.h
@@ -612,6 +612,8 @@
*/
#define PCI_EXP_DEVCAP2 36 /* Device Capabilities 2 */
#define PCI_EXP_DEVCAP2_ARI 0x00000020 /* Alternative Routing-ID */
+#define PCI_EXP_DEVCAP2_ATOMIC_ROUTE 0x00000040 /* Atomic Op routing */
+#define PCI_EXP_DEVCAP2_ATOMIC_COMP64 0x00000100 /* Atomic 64-bit comparison */
#define PCI_EXP_DEVCAP2_LTR 0x00000800 /* Latency tolerance reporting */
#define PCI_EXP_DEVCAP2_OBFF_MASK 0x000c0000 /* OBFF support mechanism */
#define PCI_EXP_DEVCAP2_OBFF_MSG 0x00040000 /* New message signaling */
@@ -619,6 +621,7 @@
#define PCI_EXP_DEVCTL2 40 /* Device Control 2 */
#define PCI_EXP_DEVCTL2_COMP_TIMEOUT 0x000f /* Completion Timeout Value */
#define PCI_EXP_DEVCTL2_ARI 0x0020 /* Alternative Routing-ID */
+#define PCI_EXP_DEVCTL2_ATOMIC_REQ 0x0040 /* Set Atomic requests */
#define PCI_EXP_DEVCTL2_IDO_REQ_EN 0x0100 /* Allow IDO for requests */
#define PCI_EXP_DEVCTL2_IDO_CMP_EN 0x0200 /* Allow IDO for completions */
#define PCI_EXP_DEVCTL2_LTR_EN 0x0400 /* Enable LTR mechanism */
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 03/11] Add support for RoCE HW init
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Allocate and setup RoCE resources, interrupts and completion queues.
Adds device attributes.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 408 +++++++++++-
drivers/infiniband/hw/qedr/qedr.h | 118 ++++
drivers/infiniband/hw/qedr/qedr_hsi.h | 56 ++
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 96 +++
drivers/net/ethernet/qlogic/qed/Makefile | 1 +
drivers/net/ethernet/qlogic/qed/qed.h | 26 +-
drivers/net/ethernet/qlogic/qed/qed_cxt.c | 6 +
drivers/net/ethernet/qlogic/qed/qed_cxt.h | 6 +
drivers/net/ethernet/qlogic/qed/qed_dev.c | 155 +++++
drivers/net/ethernet/qlogic/qed/qed_main.c | 44 +-
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 887 +++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 117 ++++
drivers/net/ethernet/qlogic/qed/qed_sp.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
drivers/net/ethernet/qlogic/qede/qede_roce.c | 2 +-
include/linux/qed/qed_if.h | 5 +-
include/linux/qed/qed_roce_if.h | 345 ++++++++++
18 files changed, 2281 insertions(+), 7 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
create mode 100644 include/linux/qed/qed_roce_if.h
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 3fe58a3..0b5274a 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -36,6 +36,8 @@
#include <linux/iommu.h>
#include <net/addrconf.h>
#include <linux/qed/qede_roce.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_if.h>
#include "qedr.h"
MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
@@ -80,6 +82,139 @@ static int qedr_register_device(struct qedr_dev *dev)
return 0;
}
+/* This function allocates fast-path status block memory */
+static int qedr_alloc_mem_sb(struct qedr_dev *dev,
+ struct qed_sb_info *sb_info, u16 sb_id)
+{
+ struct status_block *sb_virt;
+ dma_addr_t sb_phys;
+ int rc;
+
+ sb_virt = dma_alloc_coherent(&dev->pdev->dev,
+ sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
+ if (!sb_virt) {
+ pr_err("Status block allocation failed\n");
+ return -ENOMEM;
+ }
+
+ rc = dev->ops->common->sb_init(dev->cdev, sb_info,
+ sb_virt, sb_phys, sb_id,
+ QED_SB_TYPE_CNQ);
+ if (rc) {
+ pr_err("Status block initialization failed\n");
+ dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
+ sb_virt, sb_phys);
+ return rc;
+ }
+
+ return 0;
+}
+
+static void qedr_free_mem_sb(struct qedr_dev *dev,
+ struct qed_sb_info *sb_info, int sb_id)
+{
+ if (sb_info->sb_virt) {
+ dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
+ dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
+ (void *)sb_info->sb_virt, sb_info->sb_phys);
+ }
+}
+
+static void qedr_free_resources(struct qedr_dev *dev)
+{
+ int i;
+
+ for (i = 0; i < dev->num_cnq; i++) {
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
+ }
+
+ kfree(dev->cnq_array);
+ kfree(dev->sb_array);
+ kfree(dev->sgid_tbl);
+}
+
+static int qedr_alloc_resources(struct qedr_dev *dev)
+{
+ struct qedr_cnq *cnq;
+ __le16 *cons_pi;
+ u16 n_entries;
+ int i, rc;
+
+ dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
+ QEDR_MAX_SGID, GFP_KERNEL);
+ if (!dev->sgid_tbl)
+ return -ENOMEM;
+
+ spin_lock_init(&dev->sgid_lock);
+
+ /* Allocate Status blocks for CNQ */
+ dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
+ GFP_KERNEL);
+ if (!dev->sb_array) {
+ rc = -ENOMEM;
+ goto err1;
+ }
+
+ dev->cnq_array = kcalloc(dev->num_cnq,
+ sizeof(*dev->cnq_array), GFP_KERNEL);
+ if (!dev->cnq_array) {
+ rc = -ENOMEM;
+ goto err2;
+ }
+
+ dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
+
+ /* Allocate CNQ PBLs */
+ n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
+ for (i = 0; i < dev->num_cnq; i++) {
+ cnq = &dev->cnq_array[i];
+
+ rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
+ dev->sb_start + i);
+ if (rc)
+ goto err3;
+
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_CONSUME,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U16,
+ n_entries,
+ sizeof(struct regpair *),
+ &cnq->pbl);
+ if (rc)
+ goto err4;
+
+ cnq->dev = dev;
+ cnq->sb = &dev->sb_array[i];
+ cons_pi = dev->sb_array[i].sb_virt->pi_array;
+ cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
+ cnq->index = i;
+ sprintf(cnq->name, "qedr%d@pci:%s",
+ i, pci_name(dev->pdev));
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
+ i, qed_chain_get_cons_idx(&cnq->pbl));
+ }
+
+ return 0;
+err4:
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ for (--i; i >= 0; i--) {
+ dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ }
+err3:
+ for (--i; i >= 0; i--)
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ kfree(dev->cnq_array);
+err2:
+ kfree(dev->sb_array);
+err1:
+ kfree(dev->sgid_tbl);
+ return rc;
+}
+
/* QEDR sysfs interface */
static ssize_t show_rev(struct device *device, struct device_attribute *attr,
char *buf)
@@ -176,9 +311,237 @@ static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
}
}
+static const struct qed_rdma_ops *qed_ops;
+
+#define HILO_U64(hi, lo) ((((u64)(hi)) << 32) + (lo))
+
+static irqreturn_t qedr_irq_handler(int irq, void *handle)
+{
+ u16 hw_comp_cons, sw_comp_cons;
+ struct qedr_cnq *cnq = handle;
+
+ qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
+
+ qed_sb_update_sb_idx(cnq->sb);
+
+ hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
+ sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
+
+ /* Align protocol-index and chain reads */
+ rmb();
+
+ while (sw_comp_cons != hw_comp_cons) {
+ sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
+ cnq->n_comp++;
+ }
+
+ qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
+ sw_comp_cons);
+
+ qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
+
+ return IRQ_HANDLED;
+}
+
+static void qedr_sync_free_irqs(struct qedr_dev *dev)
+{
+ u32 vector;
+ int i;
+
+ for (i = 0; i < dev->int_info.used_cnt; i++) {
+ if (dev->int_info.msix_cnt) {
+ vector = dev->int_info.msix[i * dev->num_hwfns].vector;
+ synchronize_irq(vector);
+ free_irq(vector, &dev->cnq_array[i]);
+ }
+ }
+
+ dev->int_info.used_cnt = 0;
+}
+
+static int qedr_req_msix_irqs(struct qedr_dev *dev)
+{
+ int i, rc = 0;
+
+ if (dev->num_cnq > dev->int_info.msix_cnt) {
+ DP_ERR(dev,
+ "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
+ dev->num_cnq, dev->int_info.msix_cnt);
+ return -EINVAL;
+ }
+
+ for (i = 0; i < dev->num_cnq; i++) {
+ rc = request_irq(dev->int_info.msix[i * dev->num_hwfns].vector,
+ qedr_irq_handler, 0, dev->cnq_array[i].name,
+ &dev->cnq_array[i]);
+ if (rc) {
+ DP_ERR(dev, "Request cnq %d irq failed\n", i);
+ qedr_sync_free_irqs(dev);
+ } else {
+ DP_VERBOSE(dev, QEDR_MSG_INIT,
+ "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
+ dev->cnq_array[i].name, i,
+ &dev->cnq_array[i]);
+ dev->int_info.used_cnt++;
+ }
+ }
+
+ return rc;
+}
+
+static int qedr_setup_irqs(struct qedr_dev *dev)
+{
+ int rc;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
+
+ /* Learn Interrupt configuration */
+ rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
+ if (rc < 0)
+ return rc;
+ rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
+ if (rc) {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
+ return rc;
+ }
+
+ if (dev->int_info.msix_cnt) {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
+ dev->int_info.msix_cnt);
+ rc = qedr_req_msix_irqs(dev);
+ if (rc)
+ return rc;
+ }
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
+ return 0;
+}
+
+static int qedr_set_device_attr(struct qedr_dev *dev)
+{
+ struct qed_rdma_device *qed_attr;
+ struct qedr_device_attr *attr;
+ u32 page_size;
+
+ /* Part 1 - query core capabilities */
+ qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
+
+ /* Part 2 - check capabilities */
+ page_size = ~dev->attr.page_size_caps + 1;
+ if (page_size > PAGE_SIZE) {
+ DP_ERR(dev,
+ "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
+ PAGE_SIZE, page_size);
+ return -ENODEV;
+ }
+
+ /* Part 3 - copy and update capabilities */
+ attr = &dev->attr;
+ attr->vendor_id = qed_attr->vendor_id;
+ attr->vendor_part_id = qed_attr->vendor_part_id;
+ attr->hw_ver = qed_attr->hw_ver;
+ attr->fw_ver = qed_attr->fw_ver;
+ attr->node_guid = qed_attr->node_guid;
+ attr->sys_image_guid = qed_attr->sys_image_guid;
+ attr->max_cnq = qed_attr->max_cnq;
+ attr->max_sge = qed_attr->max_sge;
+ attr->max_inline = qed_attr->max_inline;
+ attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
+ attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
+ attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
+ attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
+ attr->max_dev_resp_rd_atomic_resc =
+ qed_attr->max_dev_resp_rd_atomic_resc;
+ attr->max_cq = qed_attr->max_cq;
+ attr->max_qp = qed_attr->max_qp;
+ attr->max_mr = qed_attr->max_mr;
+ attr->max_mr_size = qed_attr->max_mr_size;
+ attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
+ attr->max_mw = qed_attr->max_mw;
+ attr->max_fmr = qed_attr->max_fmr;
+ attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
+ attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
+ attr->max_pd = qed_attr->max_pd;
+ attr->max_ah = qed_attr->max_ah;
+ attr->max_pkey = qed_attr->max_pkey;
+ attr->max_srq = qed_attr->max_srq;
+ attr->max_srq_wr = qed_attr->max_srq_wr;
+ attr->dev_caps = qed_attr->dev_caps;
+ attr->page_size_caps = qed_attr->page_size_caps;
+ attr->dev_ack_delay = qed_attr->dev_ack_delay;
+ attr->reserved_lkey = qed_attr->reserved_lkey;
+ attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
+ attr->max_stats_queues = qed_attr->max_stats_queues;
+
+ return 0;
+}
+
+static int qedr_init_hw(struct qedr_dev *dev)
+{
+ struct qed_rdma_add_user_out_params out_params;
+ struct qed_rdma_start_in_params *in_params;
+ struct qed_rdma_cnq_params *cur_pbl;
+ struct qed_rdma_events events;
+ dma_addr_t p_phys_table;
+ u32 page_cnt;
+ int rc = 0;
+ int i;
+
+ in_params = kzalloc(sizeof(*in_params), GFP_KERNEL);
+ if (!in_params) {
+ rc = -ENOMEM;
+ goto out;
+ }
+
+ in_params->desired_cnq = dev->num_cnq;
+ for (i = 0; i < dev->num_cnq; i++) {
+ cur_pbl = &in_params->cnq_pbl_list[i];
+
+ page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
+ cur_pbl->num_pbl_pages = page_cnt;
+
+ p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
+ cur_pbl->pbl_ptr = (u64)p_phys_table;
+ }
+
+ events.context = dev;
+
+ in_params->events = &events;
+ in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
+ in_params->max_mtu = dev->ndev->mtu;
+ ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
+
+ rc = dev->ops->rdma_init(dev->cdev, in_params);
+ if (rc)
+ goto out;
+
+ rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
+ if (rc)
+ goto out;
+
+ dev->db_addr = (void *)(uintptr_t)out_params.dpi_addr;
+ dev->db_phys_addr = out_params.dpi_phys_addr;
+ dev->db_size = out_params.dpi_size;
+ dev->dpi = out_params.dpi;
+
+ rc = qedr_set_device_attr(dev);
+out:
+ kfree(in_params);
+ if (rc)
+ DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
+
+ return rc;
+}
+
+void qedr_stop_hw(struct qedr_dev *dev)
+{
+ dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
+ dev->ops->rdma_stop(dev->rdma_ctx);
+}
+
static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
struct net_device *ndev)
{
+ struct qed_dev_rdma_info dev_info;
struct qedr_dev *dev;
int rc = 0, i;
@@ -195,17 +558,49 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
dev->ndev = ndev;
dev->cdev = cdev;
+ qed_ops = qed_get_rdma_ops();
+ if (!qed_ops) {
+ DP_ERR(dev, "Failed to get qed roce operations\n");
+ goto init_err;
+ }
+
+ dev->ops = qed_ops;
+ rc = qed_ops->fill_dev_info(cdev, &dev_info);
+ if (rc)
+ goto init_err;
+
+ dev->num_hwfns = dev_info.common.num_hwfns;
+ dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
+
+ dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
+ if (!dev->num_cnq) {
+ DP_ERR(dev, "not enough CNQ resources.\n");
+ goto init_err;
+ }
+
qedr_pci_set_atomic(dev, pdev);
+ rc = qedr_alloc_resources(dev);
+ if (rc)
+ goto init_err;
+
+ rc = qedr_init_hw(dev);
+ if (rc)
+ goto alloc_err;
+
+ rc = qedr_setup_irqs(dev);
+ if (rc)
+ goto irq_err;
+
rc = qedr_register_device(dev);
if (rc) {
DP_ERR(dev, "Unable to allocate register device\n");
- goto init_err;
+ goto reg_err;
}
for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
- goto init_err;
+ goto reg_err;
spin_lock(&qedr_devlist_lock);
list_add_tail_rcu(&dev->entry, &qedr_dev_list);
@@ -214,6 +609,12 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
return dev;
+reg_err:
+ qedr_sync_free_irqs(dev);
+irq_err:
+ qedr_stop_hw(dev);
+alloc_err:
+ qedr_free_resources(dev);
init_err:
ib_dealloc_device(&dev->ibdev);
DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
@@ -232,6 +633,9 @@ static void qedr_remove(struct qedr_dev *dev)
list_del_rcu(&dev->entry);
spin_unlock(&qedr_devlist_lock);
+ qedr_stop_hw(dev);
+ qedr_sync_free_irqs(dev);
+ qedr_free_resources(dev);
ib_dealloc_device(&dev->ibdev);
}
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index c3f93b5..0663a4c9 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -35,7 +35,10 @@
#include <linux/pci.h>
#include <rdma/ib_addr.h>
#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_roce_if.h>
#include <linux/qed/qede_roce.h>
+#include "qedr_hsi.h"
#define QEDR_MODULE_VERSION "8.10.10.0"
#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
@@ -45,6 +48,60 @@ enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
};
+struct qedr_dev;
+
+struct qedr_cnq {
+ struct qedr_dev *dev;
+ struct qed_chain pbl;
+ struct qed_sb_info *sb;
+ char name[32];
+ u64 n_comp;
+ __le16 *hw_cons_ptr;
+ u8 index;
+};
+
+#define QEDR_MAX_SGID 128
+
+struct qedr_device_attr {
+ u32 vendor_id;
+ u32 vendor_part_id;
+ u32 hw_ver;
+ u64 fw_ver;
+ u64 node_guid;
+ u64 sys_image_guid;
+ u8 max_cnq;
+ u8 max_sge;
+ u16 max_inline;
+ u32 max_sqe;
+ u32 max_rqe;
+ u8 max_qp_resp_rd_atomic_resc;
+ u8 max_qp_req_rd_atomic_resc;
+ u64 max_dev_resp_rd_atomic_resc;
+ u32 max_cq;
+ u32 max_qp;
+ u32 max_mr;
+ u64 max_mr_size;
+ u32 max_cqe;
+ u32 max_mw;
+ u32 max_fmr;
+ u32 max_mr_mw_fmr_pbl;
+ u64 max_mr_mw_fmr_size;
+ u32 max_pd;
+ u32 max_ah;
+ u8 max_pkey;
+ u32 max_srq;
+ u32 max_srq_wr;
+ u8 max_srq_sge;
+ u8 max_stats_queues;
+ u32 dev_caps;
+
+ u64 page_size_caps;
+ u8 dev_ack_delay;
+ u32 reserved_lkey;
+ u32 bad_pkey_counter;
+ struct qed_rdma_events events;
+};
+
struct qedr_dev {
struct ib_device ibdev;
struct qed_dev *cdev;
@@ -54,7 +111,68 @@ struct qedr_dev {
enum ib_atomic_cap atomic_cap;
+ void *rdma_ctx;
+ struct qedr_device_attr attr;
+
+ const struct qed_rdma_ops *ops;
+ struct qed_int_info int_info;
+
+ struct qed_sb_info *sb_array;
+ struct qedr_cnq *cnq_array;
+ int num_cnq;
+ int sb_start;
+
+ void __iomem *db_addr;
+ u64 db_phys_addr;
+ u32 db_size;
+ u16 dpi;
+
+ union ib_gid *sgid_tbl;
+
+ /* Lock for sgid table */
+ spinlock_t sgid_lock;
+
+ u64 guid;
+
u32 dp_module;
u8 dp_level;
+ u8 num_hwfns;
};
+
+#define QEDR_MAX_SQ_PBL (0x8000)
+#define QEDR_MAX_SQ_PBL_ENTRIES (0x10000 / sizeof(void *))
+#define QEDR_SQE_ELEMENT_SIZE (sizeof(struct rdma_sq_sge))
+#define QEDR_MAX_SQE_ELEMENTS_PER_SQE (ROCE_REQ_MAX_SINGLE_SQ_WQE_SIZE / \
+ QEDR_SQE_ELEMENT_SIZE)
+#define QEDR_MAX_SQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \
+ QEDR_SQE_ELEMENT_SIZE)
+#define QEDR_MAX_SQE ((QEDR_MAX_SQ_PBL_ENTRIES) *\
+ (RDMA_RING_PAGE_SIZE) / \
+ (QEDR_SQE_ELEMENT_SIZE) /\
+ (QEDR_MAX_SQE_ELEMENTS_PER_SQE))
+/* RQ */
+#define QEDR_MAX_RQ_PBL (0x2000)
+#define QEDR_MAX_RQ_PBL_ENTRIES (0x10000 / sizeof(void *))
+#define QEDR_RQE_ELEMENT_SIZE (sizeof(struct rdma_rq_sge))
+#define QEDR_MAX_RQE_ELEMENTS_PER_RQE (RDMA_MAX_SGE_PER_RQ_WQE)
+#define QEDR_MAX_RQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \
+ QEDR_RQE_ELEMENT_SIZE)
+#define QEDR_MAX_RQE ((QEDR_MAX_RQ_PBL_ENTRIES) *\
+ (RDMA_RING_PAGE_SIZE) / \
+ (QEDR_RQE_ELEMENT_SIZE) /\
+ (QEDR_MAX_RQE_ELEMENTS_PER_RQE))
+
+#define QEDR_CQE_SIZE (sizeof(union rdma_cqe))
+#define QEDR_MAX_CQE_PBL_SIZE (512 * 1024)
+#define QEDR_MAX_CQE_PBL_ENTRIES (((QEDR_MAX_CQE_PBL_SIZE) / \
+ sizeof(u64)) - 1)
+#define QEDR_MAX_CQES ((u32)((QEDR_MAX_CQE_PBL_ENTRIES) * \
+ (QED_CHAIN_PAGE_SIZE) / QEDR_CQE_SIZE))
+
+#define QEDR_ROCE_MAX_CNQ_SIZE (0x4000)
+
+#define QEDR_MAX_PORT (1)
+
+#define QEDR_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
+
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi.h b/drivers/infiniband/hw/qedr/qedr_hsi.h
new file mode 100644
index 0000000..66d2752
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_hsi.h
@@ -0,0 +1,56 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QED_HSI_ROCE__
+#define __QED_HSI_ROCE__
+
+#include <linux/qed/common_hsi.h>
+#include <linux/qed/roce_common.h>
+#include "qedr_hsi_rdma.h"
+
+/* Affiliated asynchronous events / errors enumeration */
+enum roce_async_events_type {
+ ROCE_ASYNC_EVENT_NONE = 0,
+ ROCE_ASYNC_EVENT_COMM_EST = 1,
+ ROCE_ASYNC_EVENT_SQ_DRAINED,
+ ROCE_ASYNC_EVENT_SRQ_LIMIT,
+ ROCE_ASYNC_EVENT_LAST_WQE_REACHED,
+ ROCE_ASYNC_EVENT_CQ_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_CATASTROPHIC_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR,
+ ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR,
+ ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR,
+ ROCE_ASYNC_EVENT_SRQ_EMPTY,
+ MAX_ROCE_ASYNC_EVENTS_TYPE
+};
+
+#endif /* __QED_HSI_ROCE__ */
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
new file mode 100644
index 0000000..3e508fb
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -0,0 +1,96 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QED_HSI_RDMA__
+#define __QED_HSI_RDMA__
+
+#include <linux/qed/rdma_common.h>
+
+/* rdma completion notification queue element */
+struct rdma_cnqe {
+ struct regpair cq_handle;
+};
+
+struct rdma_cqe_responder {
+ struct regpair srq_wr_id;
+ struct regpair qp_handle;
+ __le32 imm_data_or_inv_r_Key;
+ __le32 length;
+ __le32 imm_data_hi;
+ __le16 rq_cons;
+ u8 flags;
+};
+
+struct rdma_cqe_requester {
+ __le16 sq_cons;
+ __le16 reserved0;
+ __le32 reserved1;
+ struct regpair qp_handle;
+ struct regpair reserved2;
+ __le32 reserved3;
+ __le16 reserved4;
+ u8 flags;
+ u8 status;
+};
+
+struct rdma_cqe_common {
+ struct regpair reserved0;
+ struct regpair qp_handle;
+ __le16 reserved1[7];
+ u8 flags;
+ u8 status;
+};
+
+/* rdma completion queue element */
+union rdma_cqe {
+ struct rdma_cqe_responder resp;
+ struct rdma_cqe_requester req;
+ struct rdma_cqe_common cmn;
+};
+
+struct rdma_sq_sge {
+ __le32 length;
+ struct regpair addr;
+ __le32 l_key;
+};
+
+struct rdma_rq_sge {
+ struct regpair addr;
+ __le32 length;
+ __le32 flags;
+};
+
+struct rdma_srq_sge {
+ struct regpair addr;
+ __le32 length;
+ __le32 l_key;
+};
+#endif /* __QED_HSI_RDMA__ */
diff --git a/drivers/net/ethernet/qlogic/qed/Makefile b/drivers/net/ethernet/qlogic/qed/Makefile
index 02f4842..2fb352f 100644
--- a/drivers/net/ethernet/qlogic/qed/Makefile
+++ b/drivers/net/ethernet/qlogic/qed/Makefile
@@ -5,3 +5,4 @@ qed-y := qed_cxt.o qed_dev.o qed_hw.o qed_init_fw_funcs.o qed_init_ops.o \
qed_selftest.o qed_dcbx.o
qed-$(CONFIG_QED_SRIOV) += qed_sriov.o qed_vf.o
qed-$(CONFIG_QED_LL2) += qed_ll2.o
+qed-$(CONFIG_INFINIBAND_QEDR) += qed_roce.o
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index a42824e..f5ff97b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -34,6 +34,9 @@ extern const struct qed_common_ops qed_common_ops_pass;
#define QED_WFQ_UNIT 100
+#define QED_WID_SIZE (1024)
+#define QED_PF_DEMS_SIZE (4)
+
/* cau states */
enum qed_coalescing_mode {
QED_COAL_MODE_DISABLE,
@@ -151,14 +154,17 @@ enum QED_RESOURCES {
QED_RL,
QED_MAC,
QED_VLAN,
+ QED_RDMA_CNQ_RAM,
QED_ILT,
QED_LL2_QUEUE,
+ QED_RDMA_STATS_QUEUE,
QED_MAX_RESC,
};
enum QED_FEATURE {
QED_PF_L2_QUE,
QED_VF,
+ QED_RDMA_CNQ,
QED_MAX_FEATURES,
};
@@ -363,6 +369,7 @@ struct qed_hwfn {
/* Protocol related */
bool using_ll2;
struct qed_ll2_info *p_ll2_info;
+ struct qed_rdma_info *p_rdma_info;
struct qed_pf_params pf_params;
bool b_rdma_enabled_in_prs;
@@ -399,6 +406,17 @@ struct qed_hwfn {
/* Buffer for unzipping firmware data */
void *unzip_buf;
+ /* PWM region specific data */
+ u32 dpi_size;
+ u32 dpi_count;
+
+ /* This is used to calculate the doorbell address */
+ u32 dpi_start_offset;
+
+ /* If one of the following is set then EDPM shouldn't be used */
+ u8 dcbx_no_edpm;
+ u8 db_bar_no_edpm;
+
struct qed_simd_fp_handler simd_proto_handler[64];
#ifdef CONFIG_QED_SRIOV
@@ -432,6 +450,8 @@ struct qed_int_params {
bool fp_initialized;
u8 fp_msix_base;
u8 fp_msix_cnt;
+ u8 rdma_msix_base;
+ u8 rdma_msix_cnt;
};
struct qed_dev {
@@ -523,7 +543,6 @@ struct qed_dev {
bool b_is_vf;
u32 drv_type;
-
struct qed_eth_stats *reset_stats;
struct qed_fw_data *fw_data;
@@ -553,6 +572,10 @@ struct qed_dev {
u8 ll2_mac_address[ETH_ALEN];
#endif
const struct firmware *firmware;
+
+ u32 rdma_max_sge;
+ u32 rdma_max_inline;
+ u32 rdma_max_srq_sge;
};
#define NUM_OF_VFS(dev) MAX_NUM_VFS_BB
@@ -611,5 +634,6 @@ void qed_get_protocol_stats(struct qed_dev *cdev,
enum qed_mcp_protocol_type type,
union qed_mcp_protocol_stats *stats);
int qed_slowpath_irq_req(struct qed_hwfn *hwfn);
+void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
#endif /* _QED_H */
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.c b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
index 34af48d..4a4e7a9 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
@@ -48,7 +48,13 @@
#define TM_ELEM_SIZE 4
/* ILT constants */
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+/* For RoCE we configure to 64K to cover for RoCE max tasks 256K purpose. */
+#define ILT_DEFAULT_HW_P_SIZE 4
+#else
#define ILT_DEFAULT_HW_P_SIZE 3
+#endif
+
#define ILT_PAGE_IN_BYTES(hw_p_size) (1U << ((hw_p_size) + 12))
#define ILT_CFG_REG(cli, reg) PSWRQ2_REG_ ## cli ## _ ## reg ## _RT_OFFSET
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.h b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
index c6f6f2e..d00ad05 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
@@ -170,6 +170,12 @@ int qed_qm_reconf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
*/
void qed_cxt_release_cid(struct qed_hwfn *p_hwfn,
u32 cid);
+int qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
+ enum qed_cxt_elem_type elem_type, u32 iid);
+u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
+ enum protocol_type type);
+u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
+ enum protocol_type type);
#define QED_CTX_WORKING_MEM 0
#define QED_CTX_FL_MEM 1
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 4afcda5..219b3b7 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -38,6 +38,9 @@
static DEFINE_SPINLOCK(qm_lock);
+#define QED_MIN_DPIS (4)
+#define QED_MIN_PWM_REGION (QED_WID_SIZE * QED_MIN_DPIS)
+
/* API common to all protocols */
enum BAR_ID {
BAR_ID_0, /* used for GRC */
@@ -802,6 +805,140 @@ static int qed_hw_init_common(struct qed_hwfn *p_hwfn,
return rc;
}
+static int
+qed_hw_init_dpi_size(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u32 pwm_region_size, u32 n_cpus)
+{
+ u32 dpi_page_size_1, dpi_page_size_2, dpi_page_size;
+ u32 dpi_bit_shift, dpi_count;
+ u32 min_dpis;
+
+ /* Calculate DPI size */
+ dpi_page_size_1 = QED_WID_SIZE * n_cpus;
+ dpi_page_size_2 = max_t(u32, QED_WID_SIZE, PAGE_SIZE);
+ dpi_page_size = max_t(u32, dpi_page_size_1, dpi_page_size_2);
+ dpi_page_size = roundup_pow_of_two(dpi_page_size);
+ dpi_bit_shift = ilog2(dpi_page_size / 4096);
+
+ dpi_count = pwm_region_size / dpi_page_size;
+
+ min_dpis = p_hwfn->pf_params.rdma_pf_params.min_dpis;
+ min_dpis = max_t(u32, QED_MIN_DPIS, min_dpis);
+
+ /* Update hwfn */
+ p_hwfn->dpi_size = dpi_page_size;
+ p_hwfn->dpi_count = dpi_count;
+
+ /* Update registers */
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_DPI_BIT_SHIFT, dpi_bit_shift);
+
+ if (dpi_count < min_dpis)
+ return -EINVAL;
+
+ return 0;
+}
+
+enum QED_ROCE_EDPM_MODE {
+ QED_ROCE_EDPM_MODE_ENABLE = 0,
+ QED_ROCE_EDPM_MODE_FORCE_ON = 1,
+ QED_ROCE_EDPM_MODE_DISABLE = 2,
+};
+
+static int
+qed_hw_init_pf_doorbell_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 pwm_regsize, norm_regsize;
+ u32 non_pwm_conn, min_addr_reg1;
+ u32 db_bar_size, n_cpus;
+ u32 roce_edpm_mode;
+ u32 pf_dems_shift;
+ int rc = 0;
+ u8 cond;
+
+ db_bar_size = qed_hw_bar_size(p_hwfn, BAR_ID_1);
+ if (p_hwfn->cdev->num_hwfns > 1)
+ db_bar_size /= 2;
+
+ /* Calculate doorbell regions */
+ non_pwm_conn = qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_CORE) +
+ qed_cxt_get_proto_cid_count(p_hwfn, PROTOCOLID_CORE,
+ NULL) +
+ qed_cxt_get_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
+ NULL);
+ norm_regsize = roundup(QED_PF_DEMS_SIZE * non_pwm_conn, 4096);
+ min_addr_reg1 = norm_regsize / 4096;
+ pwm_regsize = db_bar_size - norm_regsize;
+
+ /* Check that the normal and PWM sizes are valid */
+ if (db_bar_size < norm_regsize) {
+ DP_ERR(p_hwfn->cdev,
+ "Doorbell BAR size 0x%x is too small (normal region is 0x%0x )\n",
+ db_bar_size, norm_regsize);
+ return -EINVAL;
+ }
+
+ if (pwm_regsize < QED_MIN_PWM_REGION) {
+ DP_ERR(p_hwfn->cdev,
+ "PWM region size 0x%0x is too small. Should be at least 0x%0x (Doorbell BAR size is 0x%x and normal region size is 0x%0x)\n",
+ pwm_regsize,
+ QED_MIN_PWM_REGION, db_bar_size, norm_regsize);
+ return -EINVAL;
+ }
+
+ /* Calculate number of DPIs */
+ roce_edpm_mode = p_hwfn->pf_params.rdma_pf_params.roce_edpm_mode;
+ if ((roce_edpm_mode == QED_ROCE_EDPM_MODE_ENABLE) ||
+ ((roce_edpm_mode == QED_ROCE_EDPM_MODE_FORCE_ON))) {
+ /* Either EDPM is mandatory, or we are attempting to allocate a
+ * WID per CPU.
+ */
+ n_cpus = num_active_cpus();
+ rc = qed_hw_init_dpi_size(p_hwfn, p_ptt, pwm_regsize, n_cpus);
+ }
+
+ cond = ((rc) && (roce_edpm_mode == QED_ROCE_EDPM_MODE_ENABLE)) ||
+ (roce_edpm_mode == QED_ROCE_EDPM_MODE_DISABLE);
+ if (cond || p_hwfn->dcbx_no_edpm) {
+ /* Either EDPM is disabled from user configuration, or it is
+ * disabled via DCBx, or it is not mandatory and we failed to
+ * allocated a WID per CPU.
+ */
+ n_cpus = 1;
+ rc = qed_hw_init_dpi_size(p_hwfn, p_ptt, pwm_regsize, n_cpus);
+
+ if (cond)
+ qed_rdma_dpm_bar(p_hwfn, p_ptt);
+ }
+
+ DP_INFO(p_hwfn,
+ "doorbell bar: normal_region_size=%d, pwm_region_size=%d, dpi_size=%d, dpi_count=%d, roce_edpm=%s\n",
+ norm_regsize,
+ pwm_regsize,
+ p_hwfn->dpi_size,
+ p_hwfn->dpi_count,
+ ((p_hwfn->dcbx_no_edpm) || (p_hwfn->db_bar_no_edpm)) ?
+ "disabled" : "enabled");
+
+ if (rc) {
+ DP_ERR(p_hwfn,
+ "Failed to allocate enough DPIs. Allocated %d but the current minimum is %d.\n",
+ p_hwfn->dpi_count,
+ p_hwfn->pf_params.rdma_pf_params.min_dpis);
+ return -EINVAL;
+ }
+
+ /* Update hwfn */
+ p_hwfn->dpi_start_offset = norm_regsize;
+
+ /* Update registers */
+ /* DEMS size is configured log2 of DWORDs, hence the division by 4 */
+ pf_dems_shift = ilog2(QED_PF_DEMS_SIZE / 4);
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_ICID_BIT_SHIFT_NORM, pf_dems_shift);
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_MIN_ADDR_REG1, min_addr_reg1);
+
+ return 0;
+}
+
static int qed_hw_init_port(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, int hw_mode)
{
@@ -875,6 +1012,10 @@ static int qed_hw_init_pf(struct qed_hwfn *p_hwfn,
/* Pure runtime initializations - directly to the HW */
qed_int_igu_init_pure_rt(p_hwfn, p_ptt, true, true);
+ rc = qed_hw_init_pf_doorbell_bar(p_hwfn, p_ptt);
+ if (rc)
+ return rc;
+
if (b_hw_start) {
/* enable interrupts */
qed_int_igu_enable(p_hwfn, p_ptt, int_mode);
@@ -1299,6 +1440,19 @@ static void qed_hw_set_feat(struct qed_hwfn *p_hwfn)
u32 *feat_num = p_hwfn->hw_info.feat_num;
int num_features = 1;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ /* Roce CNQ require each: 1 status block. 1 CNQ, we divide the
+ * status blocks equally between L2 / RoCE but with consideration as
+ * to how many l2 queues / cnqs we have
+ */
+ if (p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE) {
+ num_features++;
+
+ feat_num[QED_RDMA_CNQ] =
+ min_t(u32, RESC_NUM(p_hwfn, QED_SB) / num_features,
+ RESC_NUM(p_hwfn, QED_RDMA_CNQ_RAM));
+ }
+#endif
feat_num[QED_PF_L2_QUE] = min_t(u32, RESC_NUM(p_hwfn, QED_SB) /
num_features,
RESC_NUM(p_hwfn, QED_L2_QUEUE));
@@ -1340,6 +1494,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
num_funcs;
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
+ resc_num[QED_RDMA_CNQ_RAM] = NUM_OF_CMDQS_CQS / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_main.c b/drivers/net/ethernet/qlogic/qed/qed_main.c
index 0225382..9dab843 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_main.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_main.c
@@ -33,6 +33,11 @@
#include "qed_hw.h"
#include "qed_selftest.h"
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+#define QED_ROCE_QPS (8192)
+#define QED_ROCE_DPIS (8)
+#endif
+
static char version[] =
"QLogic FastLinQ 4xxxx Core Module qed " DRV_MODULE_VERSION "\n";
@@ -206,8 +211,8 @@ int qed_fill_dev_info(struct qed_dev *cdev,
dev_info->pci_mem_start = cdev->pci_params.mem_start;
dev_info->pci_mem_end = cdev->pci_params.mem_end;
dev_info->pci_irq = cdev->pci_params.irq;
- dev_info->rdma_supported =
- (cdev->hwfns[0].hw_info.personality == QED_PCI_ETH_ROCE);
+ dev_info->rdma_supported = (cdev->hwfns[0].hw_info.personality ==
+ QED_PCI_ETH_ROCE);
dev_info->is_mf_default = IS_MF_DEFAULT(&cdev->hwfns[0]);
ether_addr_copy(dev_info->hw_mac, cdev->hwfns[0].hw_info.hw_mac_addr);
@@ -675,6 +680,9 @@ static int qed_slowpath_setup_int(struct qed_dev *cdev,
enum qed_int_mode int_mode)
{
struct qed_sb_cnt_info sb_cnt_info;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ int num_l2_queues;
+#endif
int rc;
int i;
@@ -705,6 +713,31 @@ static int qed_slowpath_setup_int(struct qed_dev *cdev,
cdev->int_params.fp_msix_cnt = cdev->int_params.out.num_vectors -
cdev->num_hwfns;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ num_l2_queues = 0;
+ for_each_hwfn(cdev, i)
+ num_l2_queues += FEAT_NUM(&cdev->hwfns[i], QED_PF_L2_QUE);
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA,
+ "cdev->int_params.fp_msix_cnt=%d num_l2_queues=%d\n",
+ cdev->int_params.fp_msix_cnt, num_l2_queues);
+
+ if (cdev->int_params.fp_msix_cnt > num_l2_queues) {
+ cdev->int_params.rdma_msix_cnt =
+ (cdev->int_params.fp_msix_cnt - num_l2_queues)
+ / cdev->num_hwfns;
+ cdev->int_params.rdma_msix_base =
+ cdev->int_params.fp_msix_base + num_l2_queues;
+ cdev->int_params.fp_msix_cnt = num_l2_queues;
+ } else {
+ cdev->int_params.rdma_msix_cnt = 0;
+ }
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA, "roce_msix_cnt=%d roce_msix_base=%d\n",
+ cdev->int_params.rdma_msix_cnt,
+ cdev->int_params.rdma_msix_base);
+#endif
+
return 0;
}
@@ -808,6 +841,13 @@ static void qed_update_pf_params(struct qed_dev *cdev,
{
int i;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ params->rdma_pf_params.num_qps = QED_ROCE_QPS;
+ params->rdma_pf_params.min_dpis = QED_ROCE_DPIS;
+ /* divide by 3 the MRs to avoid MF ILT overflow */
+ params->rdma_pf_params.num_mrs = RDMA_MAX_TIDS;
+ params->rdma_pf_params.gl_pi = QED_ROCE_PROTOCOL_INDEX;
+#endif
for (i = 0; i < cdev->num_hwfns; i++) {
struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
diff --git a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
index 103a32c..57c4f1f 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
@@ -549,4 +549,11 @@
#define QM_REG_WFQPFWEIGHT 0x2f4e80UL
#define QM_REG_WFQVPWEIGHT 0x2fa000UL
+#define NIG_REG_ROCE_DUPLICATE_TO_HOST 0x5088f0UL
+#define PRS_REG_LIGHT_L2_ETHERTYPE_EN 0x1f0968UL
+#define NIG_REG_LLH_ENG_CLS_ENG_ID_TBL 0x501b90UL
+#define DORQ_REG_PF_DPM_ENABLE 0x100510UL
+#define DORQ_REG_PF_ICID_BIT_SHIFT_NORM 0x100448UL
+#define DORQ_REG_PF_MIN_ADDR_REG1 0x100400UL
+#define DORQ_REG_PF_DPI_BIT_SHIFT 0x100450UL
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
new file mode 100644
index 0000000..e65a130
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -0,0 +1,887 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/types.h>
+#include <asm/byteorder.h>
+#include <linux/bitops.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/errno.h>
+#include <linux/etherdevice.h>
+#include <linux/if_ether.h>
+#include <linux/if_vlan.h>
+#include <linux/io.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/tcp.h>
+#include <linux/bitops.h>
+#include <linux/qed/qed_roce_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qed.h"
+#include "qed_cxt.h"
+#include "qed_hsi.h"
+#include "qed_hw.h"
+#include "qed_init_ops.h"
+#include "qed_int.h"
+#include "qed_ll2.h"
+#include "qed_mcp.h"
+#include "qed_reg_addr.h"
+#include "qed_sp.h"
+#include "qed_roce.h"
+
+void qed_async_roce_event(struct qed_hwfn *p_hwfn,
+ struct event_ring_entry *p_eqe)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+
+ p_rdma_info->events.affiliated_event(p_rdma_info->events.context,
+ p_eqe->opcode, &p_eqe->data);
+}
+
+static int qed_rdma_bmap_alloc(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 max_count)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "max_count = %08x\n", max_count);
+
+ bmap->max_count = max_count;
+
+ bmap->bitmap = kzalloc(BITS_TO_LONGS(max_count) * sizeof(long),
+ GFP_KERNEL);
+ if (!bmap->bitmap) {
+ DP_NOTICE(p_hwfn,
+ "qed bmap alloc failed: cannot allocate memory (bitmap)\n");
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocated bitmap %p\n",
+ bmap->bitmap);
+ return 0;
+}
+
+static int qed_rdma_bmap_alloc_id(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 *id_num)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "bmap = %p\n", bmap);
+
+ *id_num = find_first_zero_bit(bmap->bitmap, bmap->max_count);
+
+ if (*id_num >= bmap->max_count) {
+ DP_NOTICE(p_hwfn, "no id available max_count=%d\n",
+ bmap->max_count);
+ return -EINVAL;
+ }
+
+ __set_bit(*id_num, bmap->bitmap);
+
+ return 0;
+}
+
+static void qed_bmap_release_id(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 id_num)
+{
+ bool b_acquired;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "id_num = %08x", id_num);
+ if (id_num >= bmap->max_count)
+ return;
+
+ b_acquired = test_and_clear_bit(id_num, bmap->bitmap);
+ if (!b_acquired) {
+ DP_NOTICE(p_hwfn, "ID %d already released\n", id_num);
+ return;
+ }
+}
+
+u32 qed_rdma_get_sb_id(void *p_hwfn, u32 rel_sb_id)
+{
+ /* First sb id for RoCE is after all the l2 sb */
+ return FEAT_NUM((struct qed_hwfn *)p_hwfn, QED_PF_L2_QUE) + rel_sb_id;
+}
+
+u32 qed_rdma_query_cau_timer_res(void *rdma_cxt)
+{
+ return QED_CAU_DEF_RX_TIMER_RES;
+}
+
+static int qed_rdma_alloc(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_info *p_rdma_info;
+ u32 num_cons, num_tasks;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocating RDMA\n");
+
+ /* Allocate a struct with current pf rdma info */
+ p_rdma_info = kzalloc(sizeof(*p_rdma_info), GFP_KERNEL);
+ if (!p_rdma_info) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ p_rdma_info = p_rdma_info;
+ p_rdma_info->proto = PROTOCOLID_ROCE;
+
+ num_cons = qed_cxt_get_proto_cid_count(p_hwfn, p_rdma_info->proto, 0);
+
+ p_rdma_info->num_qps = num_cons / 2;
+
+ num_tasks = qed_cxt_get_proto_tid_count(p_hwfn, PROTOCOLID_ROCE);
+
+ /* Each MR uses a single task */
+ p_rdma_info->num_mrs = num_tasks;
+
+ /* Queue zone lines are shared between RoCE and L2 in such a way that
+ * they can be used by each without obstructing the other.
+ */
+ p_rdma_info->queue_zone_base = (u16)FEAT_NUM(p_hwfn, QED_L2_QUEUE);
+
+ /* Allocate a struct with device params and fill it */
+ p_rdma_info->dev = kzalloc(sizeof(*p_rdma_info->dev), GFP_KERNEL);
+ if (!p_rdma_info->dev) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info dev). rc = %d\n",
+ rc);
+ goto free_rdma_info;
+ }
+
+ /* Allocate a struct with port params and fill it */
+ p_rdma_info->port = kzalloc(sizeof(*p_rdma_info->port), GFP_KERNEL);
+ if (!p_rdma_info->port) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info port). rc = %d\n",
+ rc);
+ goto free_rdma_dev;
+ }
+
+ /* Allocate bit map for pd's */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->pd_map, RDMA_MAX_PDS);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate pd_map, rc = %d\n",
+ rc);
+ goto free_rdma_port;
+ }
+
+ /* Allocate DPI bitmap */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->dpi_map,
+ p_hwfn->dpi_count);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate DPI bitmap, rc = %d\n", rc);
+ goto free_pd_map;
+ }
+
+ /* Allocate bitmap for cq's. The maximum number of CQs is bounded to
+ * twice the number of QPs.
+ */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cq_map,
+ p_rdma_info->num_qps * 2);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate cq bitmap, rc = %d\n", rc);
+ goto free_dpi_map;
+ }
+
+ /* Allocate bitmap for toggle bit for cq icids
+ * We toggle the bit every time we create or resize cq for a given icid.
+ * The maximum number of CQs is bounded to twice the number of QPs.
+ */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->toggle_bits,
+ p_rdma_info->num_qps * 2);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate toogle bits, rc = %d\n", rc);
+ goto free_cq_map;
+ }
+
+ /* Allocate bitmap for itids */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->tid_map,
+ p_rdma_info->num_mrs);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate itids bitmaps, rc = %d\n", rc);
+ goto free_toggle_map;
+ }
+
+ /* Allocate bitmap for cids used for qps. */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cid_map, num_cons);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate cid bitmap, rc = %d\n", rc);
+ goto free_tid_map;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocation successful\n");
+ return 0;
+
+free_tid_map:
+ kfree(p_rdma_info->tid_map.bitmap);
+free_toggle_map:
+ kfree(p_rdma_info->toggle_bits.bitmap);
+free_cq_map:
+ kfree(p_rdma_info->cq_map.bitmap);
+free_dpi_map:
+ kfree(p_rdma_info->dpi_map.bitmap);
+free_pd_map:
+ kfree(p_rdma_info->pd_map.bitmap);
+free_rdma_port:
+ kfree(p_rdma_info->port);
+free_rdma_dev:
+ kfree(p_rdma_info->dev);
+free_rdma_info:
+ kfree(p_rdma_info);
+
+ return rc;
+}
+
+void qed_rdma_resc_free(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+
+ kfree(p_rdma_info->cid_map.bitmap);
+ kfree(p_rdma_info->tid_map.bitmap);
+ kfree(p_rdma_info->toggle_bits.bitmap);
+ kfree(p_rdma_info->cq_map.bitmap);
+ kfree(p_rdma_info->dpi_map.bitmap);
+ kfree(p_rdma_info->pd_map.bitmap);
+
+ kfree(p_rdma_info->port);
+ kfree(p_rdma_info->dev);
+
+ kfree(p_rdma_info);
+}
+
+static void qed_rdma_free(struct qed_hwfn *p_hwfn)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Freeing RDMA\n");
+
+ qed_rdma_resc_free(p_hwfn);
+}
+
+static void qed_rdma_get_guid(struct qed_hwfn *p_hwfn, u8 *guid)
+{
+ guid[0] = p_hwfn->hw_info.hw_mac_addr[0] ^ 2;
+ guid[1] = p_hwfn->hw_info.hw_mac_addr[1];
+ guid[2] = p_hwfn->hw_info.hw_mac_addr[2];
+ guid[3] = 0xff;
+ guid[4] = 0xfe;
+ guid[5] = p_hwfn->hw_info.hw_mac_addr[3];
+ guid[6] = p_hwfn->hw_info.hw_mac_addr[4];
+ guid[7] = p_hwfn->hw_info.hw_mac_addr[5];
+}
+
+static void qed_rdma_init_events(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_events *events;
+
+ events = &p_hwfn->p_rdma_info->events;
+
+ events->unaffiliated_event = params->events->unaffiliated_event;
+ events->affiliated_event = params->events->affiliated_event;
+ events->context = params->events->context;
+}
+
+static void qed_rdma_init_devinfo(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ u32 pci_status_control;
+ u32 num_qps;
+
+ /* Vendor specific information */
+ dev->vendor_id = cdev->vendor_id;
+ dev->vendor_part_id = cdev->device_id;
+ dev->hw_ver = 0;
+ dev->fw_ver = (FW_MAJOR_VERSION << 24) | (FW_MINOR_VERSION << 16) |
+ (FW_REVISION_VERSION << 8) | (FW_ENGINEERING_VERSION);
+
+ qed_rdma_get_guid(p_hwfn, (u8 *)&dev->sys_image_guid);
+ dev->node_guid = dev->sys_image_guid;
+
+ dev->max_sge = min_t(u32, RDMA_MAX_SGE_PER_SQ_WQE,
+ RDMA_MAX_SGE_PER_RQ_WQE);
+
+ if (cdev->rdma_max_sge)
+ dev->max_sge = min_t(u32, cdev->rdma_max_sge, dev->max_sge);
+
+ dev->max_inline = ROCE_REQ_MAX_INLINE_DATA_SIZE;
+
+ dev->max_inline = (cdev->rdma_max_inline) ?
+ min_t(u32, cdev->rdma_max_inline, dev->max_inline) :
+ dev->max_inline;
+
+ dev->max_wqe = QED_RDMA_MAX_WQE;
+ dev->max_cnq = (u8)FEAT_NUM(p_hwfn, QED_RDMA_CNQ);
+
+ /* The number of QPs may be higher than QED_ROCE_MAX_QPS. because
+ * it is up-aligned to 16 and then to ILT page size within qed cxt.
+ * This is OK in terms of ILT but we don't want to configure the FW
+ * above its abilities
+ */
+ num_qps = ROCE_MAX_QPS;
+ num_qps = min_t(u64, num_qps, p_hwfn->p_rdma_info->num_qps);
+ dev->max_qp = num_qps;
+
+ /* CQs uses the same icids that QPs use hence they are limited by the
+ * number of icids. There are two icids per QP.
+ */
+ dev->max_cq = num_qps * 2;
+
+ /* The number of mrs is smaller by 1 since the first is reserved */
+ dev->max_mr = p_hwfn->p_rdma_info->num_mrs - 1;
+ dev->max_mr_size = QED_RDMA_MAX_MR_SIZE;
+
+ /* The maximum CQE capacity per CQ supported.
+ * max number of cqes will be in two layer pbl,
+ * 8 is the pointer size in bytes
+ * 32 is the size of cq element in bytes
+ */
+ if (params->cq_mode == QED_RDMA_CQ_MODE_32_BITS)
+ dev->max_cqe = QED_RDMA_MAX_CQE_32_BIT;
+ else
+ dev->max_cqe = QED_RDMA_MAX_CQE_16_BIT;
+
+ dev->max_mw = 0;
+ dev->max_fmr = QED_RDMA_MAX_FMR;
+ dev->max_mr_mw_fmr_pbl = (PAGE_SIZE / 8) * (PAGE_SIZE / 8);
+ dev->max_mr_mw_fmr_size = dev->max_mr_mw_fmr_pbl * PAGE_SIZE;
+ dev->max_pkey = QED_RDMA_MAX_P_KEY;
+
+ dev->max_qp_resp_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
+ (RDMA_RESP_RD_ATOMIC_ELM_SIZE * 2);
+ dev->max_qp_req_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
+ RDMA_REQ_RD_ATOMIC_ELM_SIZE;
+ dev->max_dev_resp_rd_atomic_resc = dev->max_qp_resp_rd_atomic_resc *
+ p_hwfn->p_rdma_info->num_qps;
+ dev->page_size_caps = QED_RDMA_PAGE_SIZE_CAPS;
+ dev->dev_ack_delay = QED_RDMA_ACK_DELAY;
+ dev->max_pd = RDMA_MAX_PDS;
+ dev->max_ah = p_hwfn->p_rdma_info->num_qps;
+ dev->max_stats_queues = (u8)RESC_NUM(p_hwfn, QED_RDMA_STATS_QUEUE);
+
+ /* Set capablities */
+ dev->dev_caps = 0;
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RNR_NAK, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RESIZE_CQ, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_MEMORY_EXT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_QUEUE_EXT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ZBVA, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_LOCAL_INV_FENCE, 1);
+
+ /* Check atomic operations support in PCI configuration space. */
+ pci_read_config_dword(cdev->pdev,
+ cdev->pdev->pcie_cap + PCI_EXP_DEVCTL2,
+ &pci_status_control);
+
+ if (pci_status_control & PCI_EXP_DEVCTL2_LTR_EN)
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ATOMIC_OP, 1);
+}
+
+static void qed_rdma_init_port(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_port *port = p_hwfn->p_rdma_info->port;
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+
+ port->port_state = p_hwfn->mcp_info->link_output.link_up ?
+ QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
+
+ port->max_msg_size = min_t(u64,
+ (dev->max_mr_mw_fmr_size *
+ p_hwfn->cdev->rdma_max_sge),
+ BIT(31));
+
+ port->pkey_bad_counter = 0;
+}
+
+static int qed_rdma_init_hw(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 ll2_ethertype_en;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW\n");
+ p_hwfn->b_rdma_enabled_in_prs = false;
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
+
+ p_hwfn->rdma_prs_search_reg = PRS_REG_SEARCH_ROCE;
+
+ /* We delay writing to this reg until first cid is allocated. See
+ * qed_cxt_dynamic_ilt_alloc function for more details
+ */
+ ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
+ qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
+ (ll2_ethertype_en | 0x01));
+
+ if (qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_ROCE) % 2) {
+ DP_NOTICE(p_hwfn, "The first RoCE's cid should be even\n");
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW - Done\n");
+ return 0;
+}
+
+static int qed_rdma_start_fw(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params,
+ struct qed_ptt *p_ptt)
+{
+ struct rdma_init_func_ramrod_data *p_ramrod;
+ struct qed_rdma_cnq_params *p_cnq_pbl_list;
+ struct rdma_init_func_hdr *p_params_header;
+ struct rdma_cnq_params *p_cnq_params;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ u32 cnq_id, sb_id;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Starting FW\n");
+
+ /* Save the number of cnqs for the function close ramrod */
+ p_hwfn->p_rdma_info->num_cnqs = params->desired_cnq;
+
+ /* Start Roce */
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_INIT,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.roce_init_func.rdma;
+
+ p_params_header = &p_ramrod->params_header;
+ p_params_header->cnq_start_offset = (u8)RESC_START(p_hwfn,
+ QED_RDMA_CNQ_RAM);
+ p_params_header->num_cnqs = params->desired_cnq;
+
+ if (params->cq_mode == QED_RDMA_CQ_MODE_16_BITS)
+ p_params_header->cq_ring_mode = 1;
+ else
+ p_params_header->cq_ring_mode = 0;
+
+ for (cnq_id = 0; cnq_id < params->desired_cnq; cnq_id++) {
+ sb_id = qed_rdma_get_sb_id(p_hwfn, cnq_id);
+ p_cnq_params = &p_ramrod->cnq_params[cnq_id];
+ p_cnq_pbl_list = ¶ms->cnq_pbl_list[cnq_id];
+ p_cnq_params->sb_num =
+ cpu_to_le16(p_hwfn->sbs_info[sb_id]->igu_sb_id);
+
+ p_cnq_params->sb_index = p_hwfn->pf_params.rdma_pf_params.gl_pi;
+ p_cnq_params->num_pbl_pages = p_cnq_pbl_list->num_pbl_pages;
+
+ DMA_REGPAIR_LE(p_cnq_params->pbl_base_addr,
+ p_cnq_pbl_list->pbl_ptr);
+
+ /* we assume here that cnq_id and qz_offset are the same */
+ p_cnq_params->queue_zone_num =
+ cpu_to_le16(p_hwfn->p_rdma_info->queue_zone_base +
+ cnq_id);
+ }
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_rdma_reserve_lkey(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+
+ /* The first DPI is reserved for the Kernel */
+ __set_bit(0, p_hwfn->p_rdma_info->dpi_map.bitmap);
+
+ /* Tid 0 will be used as the key for "reserved MR".
+ * The driver should allocate memory for it so it can be loaded but no
+ * ramrod should be passed on it.
+ */
+ qed_rdma_alloc_tid(p_hwfn, &dev->reserved_lkey);
+ if (dev->reserved_lkey != RDMA_RESERVED_LKEY) {
+ DP_NOTICE(p_hwfn,
+ "Reserved lkey should be equal to RDMA_RESERVED_LKEY\n");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int qed_rdma_setup(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_rdma_start_in_params *params)
+{
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA setup\n");
+
+ spin_lock_init(&p_hwfn->p_rdma_info->lock);
+
+ qed_rdma_init_devinfo(p_hwfn, params);
+ qed_rdma_init_port(p_hwfn);
+ qed_rdma_init_events(p_hwfn, params);
+
+ rc = qed_rdma_reserve_lkey(p_hwfn);
+ if (rc)
+ return rc;
+
+ rc = qed_rdma_init_hw(p_hwfn, p_ptt);
+ if (rc)
+ return rc;
+
+ return qed_rdma_start_fw(p_hwfn, params, p_ptt);
+}
+
+int qed_rdma_stop(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_close_func_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ struct qed_ptt *p_ptt;
+ u32 ll2_ethertype_en;
+ int rc = -EBUSY;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop\n");
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Failed to acquire PTT\n");
+ return rc;
+ }
+
+ /* Disable RoCE search */
+ qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 0);
+ p_hwfn->b_rdma_enabled_in_prs = false;
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
+
+ ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
+ (ll2_ethertype_en & 0xFFFE));
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ /* Stop RoCE */
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_CLOSE,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ goto out;
+
+ p_ramrod = &p_ent->ramrod.rdma_close_func;
+
+ p_ramrod->num_cnqs = p_hwfn->p_rdma_info->num_cnqs;
+ p_ramrod->cnq_start_offset = (u8)RESC_START(p_hwfn, QED_RDMA_CNQ_RAM);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+out:
+ qed_rdma_free(p_hwfn);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop done, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_add_user(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ u32 dpi_start_offset;
+ u32 returned_id = 0;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding User\n");
+
+ /* Allocate DPI */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map,
+ &returned_id);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ out_params->dpi = (u16)returned_id;
+
+ /* Calculate the corresponding DPI address */
+ dpi_start_offset = p_hwfn->dpi_start_offset;
+
+ out_params->dpi_addr = (u64)((u8 __iomem *)p_hwfn->doorbells +
+ dpi_start_offset +
+ ((out_params->dpi) * p_hwfn->dpi_size));
+
+ out_params->dpi_phys_addr = p_hwfn->cdev->db_phys_addr +
+ dpi_start_offset +
+ ((out_params->dpi) * p_hwfn->dpi_size);
+
+ out_params->dpi_size = p_hwfn->dpi_size;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding user - done, rc = %d\n", rc);
+ return rc;
+}
+
+struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Query device\n");
+
+ /* Return struct with device parameters */
+ return p_hwfn->p_rdma_info->dev;
+}
+
+int qed_rdma_alloc_tid(void *rdma_cxt, u32 *itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID\n");
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn,
+ &p_hwfn->p_rdma_info->tid_map, itid);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+ if (rc)
+ goto out;
+
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_TASK, *itid);
+out:
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID - done, rc = %d\n", rc);
+ return rc;
+}
+
+void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 qz_offset, u16 prod)
+{
+ struct qed_hwfn *p_hwfn;
+ u16 qz_num;
+ u32 addr;
+
+ p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ qz_num = p_hwfn->p_rdma_info->queue_zone_base + qz_offset;
+ addr = GTT_BAR0_MAP_REG_USDM_RAM +
+ USTORM_COMMON_QUEUE_CONS_OFFSET(qz_num);
+
+ REG_WR16(p_hwfn, addr, prod);
+
+ /* keep prod updates ordered */
+ wmb();
+}
+
+static int qed_fill_rdma_dev_info(struct qed_dev *cdev,
+ struct qed_dev_rdma_info *info)
+{
+ memset(info, 0, sizeof(*info));
+
+ info->rdma_type = QED_RDMA_TYPE_ROCE;
+
+ qed_fill_dev_info(cdev, &info->common);
+
+ return 0;
+}
+
+static int qed_rdma_get_sb_start(struct qed_dev *cdev)
+{
+ int feat_num;
+
+ if (cdev->num_hwfns > 1)
+ feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE);
+ else
+ feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE) *
+ cdev->num_hwfns;
+
+ return feat_num;
+}
+
+static int qed_rdma_get_min_cnq_msix(struct qed_dev *cdev)
+{
+ int n_cnq = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_RDMA_CNQ);
+ int n_msix = cdev->int_params.rdma_msix_cnt;
+
+ return min_t(int, n_cnq, n_msix);
+}
+
+static int qed_rdma_set_int(struct qed_dev *cdev, u16 cnt)
+{
+ int limit = 0;
+
+ /* Mark the fastpath as free/used */
+ cdev->int_params.fp_initialized = cnt ? true : false;
+
+ if (cdev->int_params.out.int_mode != QED_INT_MODE_MSIX) {
+ DP_ERR(cdev,
+ "qed roce supports only MSI-X interrupts (detected %d).\n",
+ cdev->int_params.out.int_mode);
+ return -EINVAL;
+ } else if (cdev->int_params.fp_msix_cnt) {
+ limit = cdev->int_params.rdma_msix_cnt;
+ }
+
+ if (!limit)
+ return -ENOMEM;
+
+ return min_t(int, cnt, limit);
+}
+
+static int qed_rdma_get_int(struct qed_dev *cdev, struct qed_int_info *info)
+{
+ memset(info, 0, sizeof(*info));
+
+ if (!cdev->int_params.fp_initialized) {
+ DP_INFO(cdev,
+ "Protocol driver requested interrupt information, but its support is not yet configured\n");
+ return -EINVAL;
+ }
+
+ if (cdev->int_params.out.int_mode == QED_INT_MODE_MSIX) {
+ int msix_base = cdev->int_params.rdma_msix_base;
+
+ info->msix_cnt = cdev->int_params.rdma_msix_cnt;
+ info->msix = &cdev->int_params.msix_table[msix_base];
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA, "msix_cnt = %d msix_base=%d\n",
+ info->msix_cnt, msix_base);
+ }
+
+ return 0;
+}
+
+static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
+{
+ return QED_LEADING_HWFN(cdev);
+}
+
+static void qed_rdma_dpm_conf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 val;
+
+ val = ((p_hwfn->dcbx_no_edpm) || (p_hwfn->db_bar_no_edpm)) ? 0 : 1;
+
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_DPM_ENABLE, val);
+ DP_VERBOSE(p_hwfn, (QED_MSG_DCB | QED_MSG_RDMA),
+ "Changing DPM_EN state to %d (DCBX=%d, DB_BAR=%d)\n",
+ val, p_hwfn->dcbx_no_edpm, p_hwfn->db_bar_no_edpm);
+}
+
+void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ p_hwfn->db_bar_no_edpm = true;
+
+ qed_rdma_dpm_conf(p_hwfn, p_ptt);
+}
+
+int qed_rdma_start(void *rdma_cxt, struct qed_rdma_start_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_ptt *p_ptt;
+ int rc = -EBUSY;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "desired_cnq = %08x\n", params->desired_cnq);
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt)
+ goto err;
+
+ rc = qed_rdma_alloc(p_hwfn, p_ptt, params);
+ if (rc)
+ goto err1;
+
+ rc = qed_rdma_setup(p_hwfn, p_ptt, params);
+ if (rc)
+ goto err2;
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ return rc;
+
+err2:
+ qed_rdma_free(p_hwfn);
+err1:
+ qed_ptt_release(p_hwfn, p_ptt);
+err:
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA start - error, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_rdma_init(struct qed_dev *cdev,
+ struct qed_rdma_start_in_params *params)
+{
+ return qed_rdma_start(QED_LEADING_HWFN(cdev), params);
+}
+
+void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "dpi = %08x\n", dpi);
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map, dpi);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
+static const struct qed_rdma_ops qed_rdma_ops_pass = {
+ .common = &qed_common_ops_pass,
+ .fill_dev_info = &qed_fill_rdma_dev_info,
+ .rdma_get_rdma_ctx = &qed_rdma_get_rdma_ctx,
+ .rdma_init = &qed_rdma_init,
+ .rdma_add_user = &qed_rdma_add_user,
+ .rdma_remove_user = &qed_rdma_remove_user,
+ .rdma_stop = &qed_rdma_stop,
+ .rdma_query_device = &qed_rdma_query_device,
+ .rdma_get_start_sb = &qed_rdma_get_sb_start,
+ .rdma_get_rdma_int = &qed_rdma_get_int,
+ .rdma_set_rdma_int = &qed_rdma_set_int,
+ .rdma_get_min_cnq_msix = &qed_rdma_get_min_cnq_msix,
+ .rdma_cnq_prod_update = &qed_rdma_cnq_prod_update,
+};
+
+const struct qed_rdma_ops *qed_get_rdma_ops()
+{
+ return &qed_rdma_ops_pass;
+}
+EXPORT_SYMBOL(qed_get_rdma_ops);
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
new file mode 100644
index 0000000..46f14b2
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -0,0 +1,117 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _QED_ROCE_H
+#define _QED_ROCE_H
+#include <linux/types.h>
+#include <linux/bitops.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qed.h"
+#include "qed_dev_api.h"
+#include "qed_hsi.h"
+
+#define QED_RDMA_MAX_FMR (RDMA_MAX_TIDS)
+#define QED_RDMA_MAX_P_KEY (1)
+#define QED_RDMA_MAX_WQE (0x7FFF)
+#define QED_RDMA_MAX_SRQ_WQE_ELEM (0x7FFF)
+#define QED_RDMA_PAGE_SIZE_CAPS (0xFFFFF000)
+#define QED_RDMA_ACK_DELAY (15)
+#define QED_RDMA_MAX_MR_SIZE (0x10000000000ULL)
+#define QED_RDMA_MAX_CQS (RDMA_MAX_CQS)
+#define QED_RDMA_MAX_MRS (RDMA_MAX_TIDS)
+/* Add 1 for header element */
+#define QED_RDMA_MAX_SRQ_ELEM_PER_WQE (RDMA_MAX_SGE_PER_RQ_WQE + 1)
+#define QED_RDMA_MAX_SGE_PER_SRQ_WQE (RDMA_MAX_SGE_PER_RQ_WQE)
+#define QED_RDMA_SRQ_WQE_ELEM_SIZE (16)
+#define QED_RDMA_MAX_SRQS (32 * 1024)
+
+#define QED_RDMA_MAX_CQE_32_BIT (0x7FFFFFFF - 1)
+#define QED_RDMA_MAX_CQE_16_BIT (0x7FFF - 1)
+
+enum qed_rdma_toggle_bit {
+ QED_RDMA_TOGGLE_BIT_CLEAR = 0,
+ QED_RDMA_TOGGLE_BIT_SET = 1
+};
+
+struct qed_bmap {
+ unsigned long *bitmap;
+ u32 max_count;
+};
+
+struct qed_rdma_info {
+ /* spin lock to protect bitmaps */
+ spinlock_t lock;
+
+ struct qed_bmap cq_map;
+ struct qed_bmap pd_map;
+ struct qed_bmap tid_map;
+ struct qed_bmap qp_map;
+ struct qed_bmap srq_map;
+ struct qed_bmap cid_map;
+ struct qed_bmap dpi_map;
+ struct qed_bmap toggle_bits;
+ struct qed_rdma_events events;
+ struct qed_rdma_device *dev;
+ struct qed_rdma_port *port;
+ u32 last_tid;
+ u8 num_cnqs;
+ u32 num_qps;
+ u32 num_mrs;
+ u16 queue_zone_base;
+ enum protocol_type proto;
+};
+
+int
+qed_rdma_add_user(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *out_params);
+int qed_rdma_alloc_pd(void *rdma_cxt, u16 *pd);
+int qed_rdma_alloc_tid(void *rdma_cxt, u32 *tid);
+int qed_rdma_deregister_tid(void *rdma_cxt, u32 tid);
+void qed_rdma_free_tid(void *rdma_cxt, u32 tid);
+struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt);
+int
+qed_rdma_register_tid(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *params);
+void qed_rdma_remove_user(void *rdma_cxt, u16 dpi);
+int qed_rdma_start(void *p_hwfn, struct qed_rdma_start_in_params *params);
+int qed_rdma_stop(void *rdma_cxt);
+u32 qed_rdma_get_sb_id(void *p_hwfn, u32 rel_sb_id);
+u32 qed_rdma_query_cau_timer_res(void *p_hwfn);
+void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 cnq_index, u16 prod);
+void qed_rdma_resc_free(struct qed_hwfn *p_hwfn);
+void qed_async_roce_event(struct qed_hwfn *p_hwfn,
+ struct event_ring_entry *p_eqe);
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_sp.h b/drivers/net/ethernet/qlogic/qed/qed_sp.h
index a3c539f..652c908 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_sp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_sp.h
@@ -85,6 +85,7 @@ union ramrod_data {
struct rdma_srq_create_ramrod_data rdma_create_srq;
struct rdma_srq_destroy_ramrod_data rdma_destroy_srq;
struct rdma_srq_modify_ramrod_data rdma_modify_srq;
+ struct roce_init_func_ramrod_data roce_init_func;
struct iscsi_slow_path_hdr iscsi_empty;
struct iscsi_init_ramrod_params iscsi_init;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_spq.c b/drivers/net/ethernet/qlogic/qed/qed_spq.c
index 0265a32..4e7dd90 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_spq.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_spq.c
@@ -28,6 +28,9 @@
#include "qed_reg_addr.h"
#include "qed_sp.h"
#include "qed_sriov.h"
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+#include "qed_roce.h"
+#endif
/***************************************************************************
* Structures & Definitions
@@ -237,6 +240,11 @@ qed_async_event_completion(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe)
{
switch (p_eqe->protocol_id) {
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ case PROTOCOLID_ROCE:
+ qed_async_roce_event(p_hwfn, p_eqe);
+ return 0;
+#endif
case PROTOCOLID_COMMON:
return qed_sriov_eqe_event(p_hwfn,
p_eqe->opcode,
diff --git a/drivers/net/ethernet/qlogic/qede/qede_roce.c b/drivers/net/ethernet/qlogic/qede/qede_roce.c
index b7e0ed9..3d12a44 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_roce.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_roce.c
@@ -33,7 +33,7 @@
#include <linux/netdevice.h>
#include <linux/list.h>
#include <linux/mutex.h>
-#include <linux/qede/qede_roce.h>
+#include <linux/qed/qede_roce.h>
#include "qede.h"
static struct qedr_driver *qedr_drv;
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index cb899b2..1c6e5e0 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -34,6 +34,8 @@ enum dcbx_protocol_type {
DCBX_MAX_PROTOCOL_TYPE
};
+#define QED_ROCE_PROTOCOL_INDEX (3)
+
#ifdef CONFIG_DCB
#define QED_LLDP_CHASSIS_ID_STAT_LEN 4
#define QED_LLDP_PORT_ID_STAT_LEN 4
@@ -268,6 +270,7 @@ struct qed_dev_info {
enum qed_sb_type {
QED_SB_TYPE_L2_QUEUE,
+ QED_SB_TYPE_CNQ,
};
enum qed_protocol {
@@ -624,7 +627,7 @@ enum DP_MODULE {
QED_MSG_CXT = 0x800000,
QED_MSG_LL2 = 0x1000000,
QED_MSG_ILT = 0x2000000,
- QED_MSG_ROCE = 0x4000000,
+ QED_MSG_RDMA = 0x4000000,
QED_MSG_DEBUG = 0x8000000,
/* to be added...up to 0x8000000 */
};
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
new file mode 100644
index 0000000..0f7d527
--- /dev/null
+++ b/include/linux/qed/qed_roce_if.h
@@ -0,0 +1,345 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _QED_ROCE_IF_H
+#define _QED_ROCE_IF_H
+#include <linux/types.h>
+#include <linux/delay.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_ll2_if.h>
+
+#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
+
+/* rdma interface */
+enum qed_rdma_tid_type {
+ QED_RDMA_TID_REGISTERED_MR,
+ QED_RDMA_TID_FMR,
+ QED_RDMA_TID_MW_TYPE1,
+ QED_RDMA_TID_MW_TYPE2A
+};
+
+struct qed_rdma_events {
+ void *context;
+ void (*affiliated_event)(void *context, u8 fw_event_code,
+ void *fw_handle);
+ void (*unaffiliated_event)(void *context, u8 event_code);
+};
+
+struct qed_rdma_device {
+ u32 vendor_id;
+ u32 vendor_part_id;
+ u32 hw_ver;
+ u64 fw_ver;
+
+ u64 node_guid;
+ u64 sys_image_guid;
+
+ u8 max_cnq;
+ u8 max_sge;
+ u8 max_srq_sge;
+ u16 max_inline;
+ u32 max_wqe;
+ u32 max_srq_wqe;
+ u8 max_qp_resp_rd_atomic_resc;
+ u8 max_qp_req_rd_atomic_resc;
+ u64 max_dev_resp_rd_atomic_resc;
+ u32 max_cq;
+ u32 max_qp;
+ u32 max_srq;
+ u32 max_mr;
+ u64 max_mr_size;
+ u32 max_cqe;
+ u32 max_mw;
+ u32 max_fmr;
+ u32 max_mr_mw_fmr_pbl;
+ u64 max_mr_mw_fmr_size;
+ u32 max_pd;
+ u32 max_ah;
+ u8 max_pkey;
+ u16 max_srq_wr;
+ u8 max_stats_queues;
+ u32 dev_caps;
+
+ /* Abilty to support RNR-NAK generation */
+
+#define QED_RDMA_DEV_CAP_RNR_NAK_MASK 0x1
+#define QED_RDMA_DEV_CAP_RNR_NAK_SHIFT 0
+ /* Abilty to support shutdown port */
+#define QED_RDMA_DEV_CAP_SHUTDOWN_PORT_MASK 0x1
+#define QED_RDMA_DEV_CAP_SHUTDOWN_PORT_SHIFT 1
+ /* Abilty to support port active event */
+#define QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT_MASK 0x1
+#define QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT_SHIFT 2
+ /* Abilty to support port change event */
+#define QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT_MASK 0x1
+#define QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT_SHIFT 3
+ /* Abilty to support system image GUID */
+#define QED_RDMA_DEV_CAP_SYS_IMAGE_MASK 0x1
+#define QED_RDMA_DEV_CAP_SYS_IMAGE_SHIFT 4
+ /* Abilty to support bad P_Key counter support */
+#define QED_RDMA_DEV_CAP_BAD_PKEY_CNT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BAD_PKEY_CNT_SHIFT 5
+ /* Abilty to support atomic operations */
+#define QED_RDMA_DEV_CAP_ATOMIC_OP_MASK 0x1
+#define QED_RDMA_DEV_CAP_ATOMIC_OP_SHIFT 6
+#define QED_RDMA_DEV_CAP_RESIZE_CQ_MASK 0x1
+#define QED_RDMA_DEV_CAP_RESIZE_CQ_SHIFT 7
+ /* Abilty to support modifying the maximum number of
+ * outstanding work requests per QP
+ */
+#define QED_RDMA_DEV_CAP_RESIZE_MAX_WR_MASK 0x1
+#define QED_RDMA_DEV_CAP_RESIZE_MAX_WR_SHIFT 8
+ /* Abilty to support automatic path migration */
+#define QED_RDMA_DEV_CAP_AUTO_PATH_MIG_MASK 0x1
+#define QED_RDMA_DEV_CAP_AUTO_PATH_MIG_SHIFT 9
+ /* Abilty to support the base memory management extensions */
+#define QED_RDMA_DEV_CAP_BASE_MEMORY_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BASE_MEMORY_EXT_SHIFT 10
+#define QED_RDMA_DEV_CAP_BASE_QUEUE_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BASE_QUEUE_EXT_SHIFT 11
+ /* Abilty to support multipile page sizes per memory region */
+#define QED_RDMA_DEV_CAP_MULTI_PAGE_PER_MR_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_MULTI_PAGE_PER_MR_EXT_SHIFT 12
+ /* Abilty to support block list physical buffer list */
+#define QED_RDMA_DEV_CAP_BLOCK_MODE_MASK 0x1
+#define QED_RDMA_DEV_CAP_BLOCK_MODE_SHIFT 13
+ /* Abilty to support zero based virtual addresses */
+#define QED_RDMA_DEV_CAP_ZBVA_MASK 0x1
+#define QED_RDMA_DEV_CAP_ZBVA_SHIFT 14
+ /* Abilty to support local invalidate fencing */
+#define QED_RDMA_DEV_CAP_LOCAL_INV_FENCE_MASK 0x1
+#define QED_RDMA_DEV_CAP_LOCAL_INV_FENCE_SHIFT 15
+ /* Abilty to support Loopback on QP */
+#define QED_RDMA_DEV_CAP_LB_INDICATOR_MASK 0x1
+#define QED_RDMA_DEV_CAP_LB_INDICATOR_SHIFT 16
+ u64 page_size_caps;
+ u8 dev_ack_delay;
+ u32 reserved_lkey;
+ u32 bad_pkey_counter;
+ struct qed_rdma_events events;
+};
+
+enum qed_port_state {
+ QED_RDMA_PORT_UP,
+ QED_RDMA_PORT_DOWN,
+};
+
+enum qed_roce_capability {
+ QED_ROCE_V1 = 1 << 0,
+ QED_ROCE_V2 = 1 << 1,
+};
+
+struct qed_rdma_port {
+ enum qed_port_state port_state;
+ int link_speed;
+ u64 max_msg_size;
+ u8 source_gid_table_len;
+ void *source_gid_table_ptr;
+ u8 pkey_table_len;
+ void *pkey_table_ptr;
+ u32 pkey_bad_counter;
+ enum qed_roce_capability capability;
+};
+
+struct qed_rdma_cnq_params {
+ u8 num_pbl_pages;
+ u64 pbl_ptr;
+};
+
+/* The CQ Mode affects the CQ doorbell transaction size.
+ * 64/32 bit machines should configure to 32/16 bits respectively.
+ */
+enum qed_rdma_cq_mode {
+ QED_RDMA_CQ_MODE_16_BITS,
+ QED_RDMA_CQ_MODE_32_BITS,
+};
+
+struct qed_roce_dcqcn_params {
+ u8 notification_point;
+ u8 reaction_point;
+
+ /* fields for notification point */
+ u32 cnp_send_timeout;
+
+ /* fields for reaction point */
+ u32 rl_bc_rate;
+ u16 rl_max_rate;
+ u16 rl_r_ai;
+ u16 rl_r_hai;
+ u16 dcqcn_g;
+ u32 dcqcn_k_us;
+ u32 dcqcn_timeout_us;
+};
+
+struct qed_rdma_start_in_params {
+ struct qed_rdma_events *events;
+ struct qed_rdma_cnq_params cnq_pbl_list[128];
+ u8 desired_cnq;
+ enum qed_rdma_cq_mode cq_mode;
+ struct qed_roce_dcqcn_params dcqcn_params;
+ u16 max_mtu;
+ u8 mac_addr[ETH_ALEN];
+ u8 iwarp_flags;
+};
+
+struct qed_rdma_add_user_out_params {
+ u16 dpi;
+ u64 dpi_addr;
+ u64 dpi_phys_addr;
+ u32 dpi_size;
+};
+
+enum roce_mode {
+ ROCE_V1,
+ ROCE_V2_IPV4,
+ ROCE_V2_IPV6,
+ MAX_ROCE_MODE
+};
+
+union qed_gid {
+ u8 bytes[16];
+ u16 words[8];
+ u32 dwords[4];
+ u64 qwords[2];
+ u32 ipv4_addr;
+};
+
+struct qed_rdma_register_tid_in_params {
+ u32 itid;
+ enum qed_rdma_tid_type tid_type;
+ u8 key;
+ u16 pd;
+ bool local_read;
+ bool local_write;
+ bool remote_read;
+ bool remote_write;
+ bool remote_atomic;
+ bool mw_bind;
+ u64 pbl_ptr;
+ bool pbl_two_level;
+ u8 pbl_page_size_log;
+ u8 page_size_log;
+ u32 fbo;
+ u64 length;
+ u64 vaddr;
+ bool zbva;
+ bool phy_mr;
+ bool dma_mr;
+
+ bool dif_enabled;
+ u64 dif_error_addr;
+ u64 dif_runt_addr;
+};
+
+struct qed_rdma_create_srq_in_params {
+ u64 pbl_base_addr;
+ u64 prod_pair_addr;
+ u16 num_pages;
+ u16 pd_id;
+ u16 page_size;
+};
+
+struct qed_rdma_create_srq_out_params {
+ u16 srq_id;
+};
+
+struct qed_rdma_destroy_srq_in_params {
+ u16 srq_id;
+};
+
+struct qed_rdma_modify_srq_in_params {
+ u32 wqe_limit;
+ u16 srq_id;
+};
+
+struct qed_rdma_stats_out_params {
+ u64 sent_bytes;
+ u64 sent_pkts;
+ u64 rcv_bytes;
+ u64 rcv_pkts;
+};
+
+struct qed_rdma_counters_out_params {
+ u64 pd_count;
+ u64 max_pd;
+ u64 dpi_count;
+ u64 max_dpi;
+ u64 cq_count;
+ u64 max_cq;
+ u64 qp_count;
+ u64 max_qp;
+ u64 tid_count;
+ u64 max_tid;
+};
+
+#define QED_ROCE_TX_HEAD_FAILURE (1)
+#define QED_ROCE_TX_FRAG_FAILURE (2)
+
+enum qed_rdma_type {
+ QED_RDMA_TYPE_ROCE,
+};
+
+struct qed_dev_rdma_info {
+ struct qed_dev_info common;
+ enum qed_rdma_type rdma_type;
+};
+
+struct qed_rdma_ops {
+ const struct qed_common_ops *common;
+
+ int (*fill_dev_info)(struct qed_dev *cdev,
+ struct qed_dev_rdma_info *info);
+ void *(*rdma_get_rdma_ctx)(struct qed_dev *cdev);
+
+ int (*rdma_init)(struct qed_dev *dev,
+ struct qed_rdma_start_in_params *iparams);
+
+ int (*rdma_add_user)(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *oparams);
+
+ void (*rdma_remove_user)(void *rdma_cxt, u16 dpi);
+ int (*rdma_stop)(void *rdma_cxt);
+ struct qed_rdma_device* (*rdma_query_device)(void *rdma_cxt);
+ int (*rdma_get_start_sb)(struct qed_dev *cdev);
+ int (*rdma_get_min_cnq_msix)(struct qed_dev *cdev);
+ void (*rdma_cnq_prod_update)(void *rdma_cxt, u8 cnq_index, u16 prod);
+ int (*rdma_get_rdma_int)(struct qed_dev *cdev,
+ struct qed_int_info *info);
+ int (*rdma_set_rdma_int)(struct qed_dev *cdev, u16 cnt);
+};
+
+const struct qed_rdma_ops *qed_get_rdma_ops(void);
+
+#endif
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 03/11] Add support for RoCE HW init
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Allocate and setup RoCE resources, interrupts and completion queues.
Adds device attributes.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 408 +++++++++++-
drivers/infiniband/hw/qedr/qedr.h | 118 ++++
drivers/infiniband/hw/qedr/qedr_hsi.h | 56 ++
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 96 +++
drivers/net/ethernet/qlogic/qed/Makefile | 1 +
drivers/net/ethernet/qlogic/qed/qed.h | 26 +-
drivers/net/ethernet/qlogic/qed/qed_cxt.c | 6 +
drivers/net/ethernet/qlogic/qed/qed_cxt.h | 6 +
drivers/net/ethernet/qlogic/qed/qed_dev.c | 155 +++++
drivers/net/ethernet/qlogic/qed/qed_main.c | 44 +-
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 887 +++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 117 ++++
drivers/net/ethernet/qlogic/qed/qed_sp.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
drivers/net/ethernet/qlogic/qede/qede_roce.c | 2 +-
include/linux/qed/qed_if.h | 5 +-
include/linux/qed/qed_roce_if.h | 345 ++++++++++
18 files changed, 2281 insertions(+), 7 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
create mode 100644 include/linux/qed/qed_roce_if.h
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 3fe58a3..0b5274a 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -36,6 +36,8 @@
#include <linux/iommu.h>
#include <net/addrconf.h>
#include <linux/qed/qede_roce.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_if.h>
#include "qedr.h"
MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
@@ -80,6 +82,139 @@ static int qedr_register_device(struct qedr_dev *dev)
return 0;
}
+/* This function allocates fast-path status block memory */
+static int qedr_alloc_mem_sb(struct qedr_dev *dev,
+ struct qed_sb_info *sb_info, u16 sb_id)
+{
+ struct status_block *sb_virt;
+ dma_addr_t sb_phys;
+ int rc;
+
+ sb_virt = dma_alloc_coherent(&dev->pdev->dev,
+ sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
+ if (!sb_virt) {
+ pr_err("Status block allocation failed\n");
+ return -ENOMEM;
+ }
+
+ rc = dev->ops->common->sb_init(dev->cdev, sb_info,
+ sb_virt, sb_phys, sb_id,
+ QED_SB_TYPE_CNQ);
+ if (rc) {
+ pr_err("Status block initialization failed\n");
+ dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
+ sb_virt, sb_phys);
+ return rc;
+ }
+
+ return 0;
+}
+
+static void qedr_free_mem_sb(struct qedr_dev *dev,
+ struct qed_sb_info *sb_info, int sb_id)
+{
+ if (sb_info->sb_virt) {
+ dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
+ dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
+ (void *)sb_info->sb_virt, sb_info->sb_phys);
+ }
+}
+
+static void qedr_free_resources(struct qedr_dev *dev)
+{
+ int i;
+
+ for (i = 0; i < dev->num_cnq; i++) {
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
+ }
+
+ kfree(dev->cnq_array);
+ kfree(dev->sb_array);
+ kfree(dev->sgid_tbl);
+}
+
+static int qedr_alloc_resources(struct qedr_dev *dev)
+{
+ struct qedr_cnq *cnq;
+ __le16 *cons_pi;
+ u16 n_entries;
+ int i, rc;
+
+ dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
+ QEDR_MAX_SGID, GFP_KERNEL);
+ if (!dev->sgid_tbl)
+ return -ENOMEM;
+
+ spin_lock_init(&dev->sgid_lock);
+
+ /* Allocate Status blocks for CNQ */
+ dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
+ GFP_KERNEL);
+ if (!dev->sb_array) {
+ rc = -ENOMEM;
+ goto err1;
+ }
+
+ dev->cnq_array = kcalloc(dev->num_cnq,
+ sizeof(*dev->cnq_array), GFP_KERNEL);
+ if (!dev->cnq_array) {
+ rc = -ENOMEM;
+ goto err2;
+ }
+
+ dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
+
+ /* Allocate CNQ PBLs */
+ n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
+ for (i = 0; i < dev->num_cnq; i++) {
+ cnq = &dev->cnq_array[i];
+
+ rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
+ dev->sb_start + i);
+ if (rc)
+ goto err3;
+
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_CONSUME,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U16,
+ n_entries,
+ sizeof(struct regpair *),
+ &cnq->pbl);
+ if (rc)
+ goto err4;
+
+ cnq->dev = dev;
+ cnq->sb = &dev->sb_array[i];
+ cons_pi = dev->sb_array[i].sb_virt->pi_array;
+ cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
+ cnq->index = i;
+ sprintf(cnq->name, "qedr%d@pci:%s",
+ i, pci_name(dev->pdev));
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
+ i, qed_chain_get_cons_idx(&cnq->pbl));
+ }
+
+ return 0;
+err4:
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ for (--i; i >= 0; i--) {
+ dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ }
+err3:
+ for (--i; i >= 0; i--)
+ qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
+ kfree(dev->cnq_array);
+err2:
+ kfree(dev->sb_array);
+err1:
+ kfree(dev->sgid_tbl);
+ return rc;
+}
+
/* QEDR sysfs interface */
static ssize_t show_rev(struct device *device, struct device_attribute *attr,
char *buf)
@@ -176,9 +311,237 @@ static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
}
}
+static const struct qed_rdma_ops *qed_ops;
+
+#define HILO_U64(hi, lo) ((((u64)(hi)) << 32) + (lo))
+
+static irqreturn_t qedr_irq_handler(int irq, void *handle)
+{
+ u16 hw_comp_cons, sw_comp_cons;
+ struct qedr_cnq *cnq = handle;
+
+ qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
+
+ qed_sb_update_sb_idx(cnq->sb);
+
+ hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
+ sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
+
+ /* Align protocol-index and chain reads */
+ rmb();
+
+ while (sw_comp_cons != hw_comp_cons) {
+ sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
+ cnq->n_comp++;
+ }
+
+ qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
+ sw_comp_cons);
+
+ qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
+
+ return IRQ_HANDLED;
+}
+
+static void qedr_sync_free_irqs(struct qedr_dev *dev)
+{
+ u32 vector;
+ int i;
+
+ for (i = 0; i < dev->int_info.used_cnt; i++) {
+ if (dev->int_info.msix_cnt) {
+ vector = dev->int_info.msix[i * dev->num_hwfns].vector;
+ synchronize_irq(vector);
+ free_irq(vector, &dev->cnq_array[i]);
+ }
+ }
+
+ dev->int_info.used_cnt = 0;
+}
+
+static int qedr_req_msix_irqs(struct qedr_dev *dev)
+{
+ int i, rc = 0;
+
+ if (dev->num_cnq > dev->int_info.msix_cnt) {
+ DP_ERR(dev,
+ "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
+ dev->num_cnq, dev->int_info.msix_cnt);
+ return -EINVAL;
+ }
+
+ for (i = 0; i < dev->num_cnq; i++) {
+ rc = request_irq(dev->int_info.msix[i * dev->num_hwfns].vector,
+ qedr_irq_handler, 0, dev->cnq_array[i].name,
+ &dev->cnq_array[i]);
+ if (rc) {
+ DP_ERR(dev, "Request cnq %d irq failed\n", i);
+ qedr_sync_free_irqs(dev);
+ } else {
+ DP_VERBOSE(dev, QEDR_MSG_INIT,
+ "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
+ dev->cnq_array[i].name, i,
+ &dev->cnq_array[i]);
+ dev->int_info.used_cnt++;
+ }
+ }
+
+ return rc;
+}
+
+static int qedr_setup_irqs(struct qedr_dev *dev)
+{
+ int rc;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
+
+ /* Learn Interrupt configuration */
+ rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
+ if (rc < 0)
+ return rc;
+ rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
+ if (rc) {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
+ return rc;
+ }
+
+ if (dev->int_info.msix_cnt) {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
+ dev->int_info.msix_cnt);
+ rc = qedr_req_msix_irqs(dev);
+ if (rc)
+ return rc;
+ }
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
+ return 0;
+}
+
+static int qedr_set_device_attr(struct qedr_dev *dev)
+{
+ struct qed_rdma_device *qed_attr;
+ struct qedr_device_attr *attr;
+ u32 page_size;
+
+ /* Part 1 - query core capabilities */
+ qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
+
+ /* Part 2 - check capabilities */
+ page_size = ~dev->attr.page_size_caps + 1;
+ if (page_size > PAGE_SIZE) {
+ DP_ERR(dev,
+ "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
+ PAGE_SIZE, page_size);
+ return -ENODEV;
+ }
+
+ /* Part 3 - copy and update capabilities */
+ attr = &dev->attr;
+ attr->vendor_id = qed_attr->vendor_id;
+ attr->vendor_part_id = qed_attr->vendor_part_id;
+ attr->hw_ver = qed_attr->hw_ver;
+ attr->fw_ver = qed_attr->fw_ver;
+ attr->node_guid = qed_attr->node_guid;
+ attr->sys_image_guid = qed_attr->sys_image_guid;
+ attr->max_cnq = qed_attr->max_cnq;
+ attr->max_sge = qed_attr->max_sge;
+ attr->max_inline = qed_attr->max_inline;
+ attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
+ attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
+ attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
+ attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
+ attr->max_dev_resp_rd_atomic_resc =
+ qed_attr->max_dev_resp_rd_atomic_resc;
+ attr->max_cq = qed_attr->max_cq;
+ attr->max_qp = qed_attr->max_qp;
+ attr->max_mr = qed_attr->max_mr;
+ attr->max_mr_size = qed_attr->max_mr_size;
+ attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
+ attr->max_mw = qed_attr->max_mw;
+ attr->max_fmr = qed_attr->max_fmr;
+ attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
+ attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
+ attr->max_pd = qed_attr->max_pd;
+ attr->max_ah = qed_attr->max_ah;
+ attr->max_pkey = qed_attr->max_pkey;
+ attr->max_srq = qed_attr->max_srq;
+ attr->max_srq_wr = qed_attr->max_srq_wr;
+ attr->dev_caps = qed_attr->dev_caps;
+ attr->page_size_caps = qed_attr->page_size_caps;
+ attr->dev_ack_delay = qed_attr->dev_ack_delay;
+ attr->reserved_lkey = qed_attr->reserved_lkey;
+ attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
+ attr->max_stats_queues = qed_attr->max_stats_queues;
+
+ return 0;
+}
+
+static int qedr_init_hw(struct qedr_dev *dev)
+{
+ struct qed_rdma_add_user_out_params out_params;
+ struct qed_rdma_start_in_params *in_params;
+ struct qed_rdma_cnq_params *cur_pbl;
+ struct qed_rdma_events events;
+ dma_addr_t p_phys_table;
+ u32 page_cnt;
+ int rc = 0;
+ int i;
+
+ in_params = kzalloc(sizeof(*in_params), GFP_KERNEL);
+ if (!in_params) {
+ rc = -ENOMEM;
+ goto out;
+ }
+
+ in_params->desired_cnq = dev->num_cnq;
+ for (i = 0; i < dev->num_cnq; i++) {
+ cur_pbl = &in_params->cnq_pbl_list[i];
+
+ page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
+ cur_pbl->num_pbl_pages = page_cnt;
+
+ p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
+ cur_pbl->pbl_ptr = (u64)p_phys_table;
+ }
+
+ events.context = dev;
+
+ in_params->events = &events;
+ in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
+ in_params->max_mtu = dev->ndev->mtu;
+ ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
+
+ rc = dev->ops->rdma_init(dev->cdev, in_params);
+ if (rc)
+ goto out;
+
+ rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
+ if (rc)
+ goto out;
+
+ dev->db_addr = (void *)(uintptr_t)out_params.dpi_addr;
+ dev->db_phys_addr = out_params.dpi_phys_addr;
+ dev->db_size = out_params.dpi_size;
+ dev->dpi = out_params.dpi;
+
+ rc = qedr_set_device_attr(dev);
+out:
+ kfree(in_params);
+ if (rc)
+ DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
+
+ return rc;
+}
+
+void qedr_stop_hw(struct qedr_dev *dev)
+{
+ dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
+ dev->ops->rdma_stop(dev->rdma_ctx);
+}
+
static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
struct net_device *ndev)
{
+ struct qed_dev_rdma_info dev_info;
struct qedr_dev *dev;
int rc = 0, i;
@@ -195,17 +558,49 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
dev->ndev = ndev;
dev->cdev = cdev;
+ qed_ops = qed_get_rdma_ops();
+ if (!qed_ops) {
+ DP_ERR(dev, "Failed to get qed roce operations\n");
+ goto init_err;
+ }
+
+ dev->ops = qed_ops;
+ rc = qed_ops->fill_dev_info(cdev, &dev_info);
+ if (rc)
+ goto init_err;
+
+ dev->num_hwfns = dev_info.common.num_hwfns;
+ dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
+
+ dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
+ if (!dev->num_cnq) {
+ DP_ERR(dev, "not enough CNQ resources.\n");
+ goto init_err;
+ }
+
qedr_pci_set_atomic(dev, pdev);
+ rc = qedr_alloc_resources(dev);
+ if (rc)
+ goto init_err;
+
+ rc = qedr_init_hw(dev);
+ if (rc)
+ goto alloc_err;
+
+ rc = qedr_setup_irqs(dev);
+ if (rc)
+ goto irq_err;
+
rc = qedr_register_device(dev);
if (rc) {
DP_ERR(dev, "Unable to allocate register device\n");
- goto init_err;
+ goto reg_err;
}
for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
- goto init_err;
+ goto reg_err;
spin_lock(&qedr_devlist_lock);
list_add_tail_rcu(&dev->entry, &qedr_dev_list);
@@ -214,6 +609,12 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
return dev;
+reg_err:
+ qedr_sync_free_irqs(dev);
+irq_err:
+ qedr_stop_hw(dev);
+alloc_err:
+ qedr_free_resources(dev);
init_err:
ib_dealloc_device(&dev->ibdev);
DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
@@ -232,6 +633,9 @@ static void qedr_remove(struct qedr_dev *dev)
list_del_rcu(&dev->entry);
spin_unlock(&qedr_devlist_lock);
+ qedr_stop_hw(dev);
+ qedr_sync_free_irqs(dev);
+ qedr_free_resources(dev);
ib_dealloc_device(&dev->ibdev);
}
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index c3f93b5..0663a4c9 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -35,7 +35,10 @@
#include <linux/pci.h>
#include <rdma/ib_addr.h>
#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_roce_if.h>
#include <linux/qed/qede_roce.h>
+#include "qedr_hsi.h"
#define QEDR_MODULE_VERSION "8.10.10.0"
#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
@@ -45,6 +48,60 @@ enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
};
+struct qedr_dev;
+
+struct qedr_cnq {
+ struct qedr_dev *dev;
+ struct qed_chain pbl;
+ struct qed_sb_info *sb;
+ char name[32];
+ u64 n_comp;
+ __le16 *hw_cons_ptr;
+ u8 index;
+};
+
+#define QEDR_MAX_SGID 128
+
+struct qedr_device_attr {
+ u32 vendor_id;
+ u32 vendor_part_id;
+ u32 hw_ver;
+ u64 fw_ver;
+ u64 node_guid;
+ u64 sys_image_guid;
+ u8 max_cnq;
+ u8 max_sge;
+ u16 max_inline;
+ u32 max_sqe;
+ u32 max_rqe;
+ u8 max_qp_resp_rd_atomic_resc;
+ u8 max_qp_req_rd_atomic_resc;
+ u64 max_dev_resp_rd_atomic_resc;
+ u32 max_cq;
+ u32 max_qp;
+ u32 max_mr;
+ u64 max_mr_size;
+ u32 max_cqe;
+ u32 max_mw;
+ u32 max_fmr;
+ u32 max_mr_mw_fmr_pbl;
+ u64 max_mr_mw_fmr_size;
+ u32 max_pd;
+ u32 max_ah;
+ u8 max_pkey;
+ u32 max_srq;
+ u32 max_srq_wr;
+ u8 max_srq_sge;
+ u8 max_stats_queues;
+ u32 dev_caps;
+
+ u64 page_size_caps;
+ u8 dev_ack_delay;
+ u32 reserved_lkey;
+ u32 bad_pkey_counter;
+ struct qed_rdma_events events;
+};
+
struct qedr_dev {
struct ib_device ibdev;
struct qed_dev *cdev;
@@ -54,7 +111,68 @@ struct qedr_dev {
enum ib_atomic_cap atomic_cap;
+ void *rdma_ctx;
+ struct qedr_device_attr attr;
+
+ const struct qed_rdma_ops *ops;
+ struct qed_int_info int_info;
+
+ struct qed_sb_info *sb_array;
+ struct qedr_cnq *cnq_array;
+ int num_cnq;
+ int sb_start;
+
+ void __iomem *db_addr;
+ u64 db_phys_addr;
+ u32 db_size;
+ u16 dpi;
+
+ union ib_gid *sgid_tbl;
+
+ /* Lock for sgid table */
+ spinlock_t sgid_lock;
+
+ u64 guid;
+
u32 dp_module;
u8 dp_level;
+ u8 num_hwfns;
};
+
+#define QEDR_MAX_SQ_PBL (0x8000)
+#define QEDR_MAX_SQ_PBL_ENTRIES (0x10000 / sizeof(void *))
+#define QEDR_SQE_ELEMENT_SIZE (sizeof(struct rdma_sq_sge))
+#define QEDR_MAX_SQE_ELEMENTS_PER_SQE (ROCE_REQ_MAX_SINGLE_SQ_WQE_SIZE / \
+ QEDR_SQE_ELEMENT_SIZE)
+#define QEDR_MAX_SQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \
+ QEDR_SQE_ELEMENT_SIZE)
+#define QEDR_MAX_SQE ((QEDR_MAX_SQ_PBL_ENTRIES) *\
+ (RDMA_RING_PAGE_SIZE) / \
+ (QEDR_SQE_ELEMENT_SIZE) /\
+ (QEDR_MAX_SQE_ELEMENTS_PER_SQE))
+/* RQ */
+#define QEDR_MAX_RQ_PBL (0x2000)
+#define QEDR_MAX_RQ_PBL_ENTRIES (0x10000 / sizeof(void *))
+#define QEDR_RQE_ELEMENT_SIZE (sizeof(struct rdma_rq_sge))
+#define QEDR_MAX_RQE_ELEMENTS_PER_RQE (RDMA_MAX_SGE_PER_RQ_WQE)
+#define QEDR_MAX_RQE_ELEMENTS_PER_PAGE ((RDMA_RING_PAGE_SIZE) / \
+ QEDR_RQE_ELEMENT_SIZE)
+#define QEDR_MAX_RQE ((QEDR_MAX_RQ_PBL_ENTRIES) *\
+ (RDMA_RING_PAGE_SIZE) / \
+ (QEDR_RQE_ELEMENT_SIZE) /\
+ (QEDR_MAX_RQE_ELEMENTS_PER_RQE))
+
+#define QEDR_CQE_SIZE (sizeof(union rdma_cqe))
+#define QEDR_MAX_CQE_PBL_SIZE (512 * 1024)
+#define QEDR_MAX_CQE_PBL_ENTRIES (((QEDR_MAX_CQE_PBL_SIZE) / \
+ sizeof(u64)) - 1)
+#define QEDR_MAX_CQES ((u32)((QEDR_MAX_CQE_PBL_ENTRIES) * \
+ (QED_CHAIN_PAGE_SIZE) / QEDR_CQE_SIZE))
+
+#define QEDR_ROCE_MAX_CNQ_SIZE (0x4000)
+
+#define QEDR_MAX_PORT (1)
+
+#define QEDR_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
+
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi.h b/drivers/infiniband/hw/qedr/qedr_hsi.h
new file mode 100644
index 0000000..66d2752
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_hsi.h
@@ -0,0 +1,56 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QED_HSI_ROCE__
+#define __QED_HSI_ROCE__
+
+#include <linux/qed/common_hsi.h>
+#include <linux/qed/roce_common.h>
+#include "qedr_hsi_rdma.h"
+
+/* Affiliated asynchronous events / errors enumeration */
+enum roce_async_events_type {
+ ROCE_ASYNC_EVENT_NONE = 0,
+ ROCE_ASYNC_EVENT_COMM_EST = 1,
+ ROCE_ASYNC_EVENT_SQ_DRAINED,
+ ROCE_ASYNC_EVENT_SRQ_LIMIT,
+ ROCE_ASYNC_EVENT_LAST_WQE_REACHED,
+ ROCE_ASYNC_EVENT_CQ_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_CATASTROPHIC_ERR,
+ ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR,
+ ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR,
+ ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR,
+ ROCE_ASYNC_EVENT_SRQ_EMPTY,
+ MAX_ROCE_ASYNC_EVENTS_TYPE
+};
+
+#endif /* __QED_HSI_ROCE__ */
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
new file mode 100644
index 0000000..3e508fb
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -0,0 +1,96 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __QED_HSI_RDMA__
+#define __QED_HSI_RDMA__
+
+#include <linux/qed/rdma_common.h>
+
+/* rdma completion notification queue element */
+struct rdma_cnqe {
+ struct regpair cq_handle;
+};
+
+struct rdma_cqe_responder {
+ struct regpair srq_wr_id;
+ struct regpair qp_handle;
+ __le32 imm_data_or_inv_r_Key;
+ __le32 length;
+ __le32 imm_data_hi;
+ __le16 rq_cons;
+ u8 flags;
+};
+
+struct rdma_cqe_requester {
+ __le16 sq_cons;
+ __le16 reserved0;
+ __le32 reserved1;
+ struct regpair qp_handle;
+ struct regpair reserved2;
+ __le32 reserved3;
+ __le16 reserved4;
+ u8 flags;
+ u8 status;
+};
+
+struct rdma_cqe_common {
+ struct regpair reserved0;
+ struct regpair qp_handle;
+ __le16 reserved1[7];
+ u8 flags;
+ u8 status;
+};
+
+/* rdma completion queue element */
+union rdma_cqe {
+ struct rdma_cqe_responder resp;
+ struct rdma_cqe_requester req;
+ struct rdma_cqe_common cmn;
+};
+
+struct rdma_sq_sge {
+ __le32 length;
+ struct regpair addr;
+ __le32 l_key;
+};
+
+struct rdma_rq_sge {
+ struct regpair addr;
+ __le32 length;
+ __le32 flags;
+};
+
+struct rdma_srq_sge {
+ struct regpair addr;
+ __le32 length;
+ __le32 l_key;
+};
+#endif /* __QED_HSI_RDMA__ */
diff --git a/drivers/net/ethernet/qlogic/qed/Makefile b/drivers/net/ethernet/qlogic/qed/Makefile
index 02f4842..2fb352f 100644
--- a/drivers/net/ethernet/qlogic/qed/Makefile
+++ b/drivers/net/ethernet/qlogic/qed/Makefile
@@ -5,3 +5,4 @@ qed-y := qed_cxt.o qed_dev.o qed_hw.o qed_init_fw_funcs.o qed_init_ops.o \
qed_selftest.o qed_dcbx.o
qed-$(CONFIG_QED_SRIOV) += qed_sriov.o qed_vf.o
qed-$(CONFIG_QED_LL2) += qed_ll2.o
+qed-$(CONFIG_INFINIBAND_QEDR) += qed_roce.o
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index a42824e..f5ff97b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -34,6 +34,9 @@ extern const struct qed_common_ops qed_common_ops_pass;
#define QED_WFQ_UNIT 100
+#define QED_WID_SIZE (1024)
+#define QED_PF_DEMS_SIZE (4)
+
/* cau states */
enum qed_coalescing_mode {
QED_COAL_MODE_DISABLE,
@@ -151,14 +154,17 @@ enum QED_RESOURCES {
QED_RL,
QED_MAC,
QED_VLAN,
+ QED_RDMA_CNQ_RAM,
QED_ILT,
QED_LL2_QUEUE,
+ QED_RDMA_STATS_QUEUE,
QED_MAX_RESC,
};
enum QED_FEATURE {
QED_PF_L2_QUE,
QED_VF,
+ QED_RDMA_CNQ,
QED_MAX_FEATURES,
};
@@ -363,6 +369,7 @@ struct qed_hwfn {
/* Protocol related */
bool using_ll2;
struct qed_ll2_info *p_ll2_info;
+ struct qed_rdma_info *p_rdma_info;
struct qed_pf_params pf_params;
bool b_rdma_enabled_in_prs;
@@ -399,6 +406,17 @@ struct qed_hwfn {
/* Buffer for unzipping firmware data */
void *unzip_buf;
+ /* PWM region specific data */
+ u32 dpi_size;
+ u32 dpi_count;
+
+ /* This is used to calculate the doorbell address */
+ u32 dpi_start_offset;
+
+ /* If one of the following is set then EDPM shouldn't be used */
+ u8 dcbx_no_edpm;
+ u8 db_bar_no_edpm;
+
struct qed_simd_fp_handler simd_proto_handler[64];
#ifdef CONFIG_QED_SRIOV
@@ -432,6 +450,8 @@ struct qed_int_params {
bool fp_initialized;
u8 fp_msix_base;
u8 fp_msix_cnt;
+ u8 rdma_msix_base;
+ u8 rdma_msix_cnt;
};
struct qed_dev {
@@ -523,7 +543,6 @@ struct qed_dev {
bool b_is_vf;
u32 drv_type;
-
struct qed_eth_stats *reset_stats;
struct qed_fw_data *fw_data;
@@ -553,6 +572,10 @@ struct qed_dev {
u8 ll2_mac_address[ETH_ALEN];
#endif
const struct firmware *firmware;
+
+ u32 rdma_max_sge;
+ u32 rdma_max_inline;
+ u32 rdma_max_srq_sge;
};
#define NUM_OF_VFS(dev) MAX_NUM_VFS_BB
@@ -611,5 +634,6 @@ void qed_get_protocol_stats(struct qed_dev *cdev,
enum qed_mcp_protocol_type type,
union qed_mcp_protocol_stats *stats);
int qed_slowpath_irq_req(struct qed_hwfn *hwfn);
+void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
#endif /* _QED_H */
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.c b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
index 34af48d..4a4e7a9 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.c
@@ -48,7 +48,13 @@
#define TM_ELEM_SIZE 4
/* ILT constants */
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+/* For RoCE we configure to 64K to cover for RoCE max tasks 256K purpose. */
+#define ILT_DEFAULT_HW_P_SIZE 4
+#else
#define ILT_DEFAULT_HW_P_SIZE 3
+#endif
+
#define ILT_PAGE_IN_BYTES(hw_p_size) (1U << ((hw_p_size) + 12))
#define ILT_CFG_REG(cli, reg) PSWRQ2_REG_ ## cli ## _ ## reg ## _RT_OFFSET
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.h b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
index c6f6f2e..d00ad05 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
@@ -170,6 +170,12 @@ int qed_qm_reconf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
*/
void qed_cxt_release_cid(struct qed_hwfn *p_hwfn,
u32 cid);
+int qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
+ enum qed_cxt_elem_type elem_type, u32 iid);
+u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
+ enum protocol_type type);
+u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
+ enum protocol_type type);
#define QED_CTX_WORKING_MEM 0
#define QED_CTX_FL_MEM 1
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 4afcda5..219b3b7 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -38,6 +38,9 @@
static DEFINE_SPINLOCK(qm_lock);
+#define QED_MIN_DPIS (4)
+#define QED_MIN_PWM_REGION (QED_WID_SIZE * QED_MIN_DPIS)
+
/* API common to all protocols */
enum BAR_ID {
BAR_ID_0, /* used for GRC */
@@ -802,6 +805,140 @@ static int qed_hw_init_common(struct qed_hwfn *p_hwfn,
return rc;
}
+static int
+qed_hw_init_dpi_size(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt, u32 pwm_region_size, u32 n_cpus)
+{
+ u32 dpi_page_size_1, dpi_page_size_2, dpi_page_size;
+ u32 dpi_bit_shift, dpi_count;
+ u32 min_dpis;
+
+ /* Calculate DPI size */
+ dpi_page_size_1 = QED_WID_SIZE * n_cpus;
+ dpi_page_size_2 = max_t(u32, QED_WID_SIZE, PAGE_SIZE);
+ dpi_page_size = max_t(u32, dpi_page_size_1, dpi_page_size_2);
+ dpi_page_size = roundup_pow_of_two(dpi_page_size);
+ dpi_bit_shift = ilog2(dpi_page_size / 4096);
+
+ dpi_count = pwm_region_size / dpi_page_size;
+
+ min_dpis = p_hwfn->pf_params.rdma_pf_params.min_dpis;
+ min_dpis = max_t(u32, QED_MIN_DPIS, min_dpis);
+
+ /* Update hwfn */
+ p_hwfn->dpi_size = dpi_page_size;
+ p_hwfn->dpi_count = dpi_count;
+
+ /* Update registers */
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_DPI_BIT_SHIFT, dpi_bit_shift);
+
+ if (dpi_count < min_dpis)
+ return -EINVAL;
+
+ return 0;
+}
+
+enum QED_ROCE_EDPM_MODE {
+ QED_ROCE_EDPM_MODE_ENABLE = 0,
+ QED_ROCE_EDPM_MODE_FORCE_ON = 1,
+ QED_ROCE_EDPM_MODE_DISABLE = 2,
+};
+
+static int
+qed_hw_init_pf_doorbell_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 pwm_regsize, norm_regsize;
+ u32 non_pwm_conn, min_addr_reg1;
+ u32 db_bar_size, n_cpus;
+ u32 roce_edpm_mode;
+ u32 pf_dems_shift;
+ int rc = 0;
+ u8 cond;
+
+ db_bar_size = qed_hw_bar_size(p_hwfn, BAR_ID_1);
+ if (p_hwfn->cdev->num_hwfns > 1)
+ db_bar_size /= 2;
+
+ /* Calculate doorbell regions */
+ non_pwm_conn = qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_CORE) +
+ qed_cxt_get_proto_cid_count(p_hwfn, PROTOCOLID_CORE,
+ NULL) +
+ qed_cxt_get_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
+ NULL);
+ norm_regsize = roundup(QED_PF_DEMS_SIZE * non_pwm_conn, 4096);
+ min_addr_reg1 = norm_regsize / 4096;
+ pwm_regsize = db_bar_size - norm_regsize;
+
+ /* Check that the normal and PWM sizes are valid */
+ if (db_bar_size < norm_regsize) {
+ DP_ERR(p_hwfn->cdev,
+ "Doorbell BAR size 0x%x is too small (normal region is 0x%0x )\n",
+ db_bar_size, norm_regsize);
+ return -EINVAL;
+ }
+
+ if (pwm_regsize < QED_MIN_PWM_REGION) {
+ DP_ERR(p_hwfn->cdev,
+ "PWM region size 0x%0x is too small. Should be at least 0x%0x (Doorbell BAR size is 0x%x and normal region size is 0x%0x)\n",
+ pwm_regsize,
+ QED_MIN_PWM_REGION, db_bar_size, norm_regsize);
+ return -EINVAL;
+ }
+
+ /* Calculate number of DPIs */
+ roce_edpm_mode = p_hwfn->pf_params.rdma_pf_params.roce_edpm_mode;
+ if ((roce_edpm_mode == QED_ROCE_EDPM_MODE_ENABLE) ||
+ ((roce_edpm_mode == QED_ROCE_EDPM_MODE_FORCE_ON))) {
+ /* Either EDPM is mandatory, or we are attempting to allocate a
+ * WID per CPU.
+ */
+ n_cpus = num_active_cpus();
+ rc = qed_hw_init_dpi_size(p_hwfn, p_ptt, pwm_regsize, n_cpus);
+ }
+
+ cond = ((rc) && (roce_edpm_mode == QED_ROCE_EDPM_MODE_ENABLE)) ||
+ (roce_edpm_mode == QED_ROCE_EDPM_MODE_DISABLE);
+ if (cond || p_hwfn->dcbx_no_edpm) {
+ /* Either EDPM is disabled from user configuration, or it is
+ * disabled via DCBx, or it is not mandatory and we failed to
+ * allocated a WID per CPU.
+ */
+ n_cpus = 1;
+ rc = qed_hw_init_dpi_size(p_hwfn, p_ptt, pwm_regsize, n_cpus);
+
+ if (cond)
+ qed_rdma_dpm_bar(p_hwfn, p_ptt);
+ }
+
+ DP_INFO(p_hwfn,
+ "doorbell bar: normal_region_size=%d, pwm_region_size=%d, dpi_size=%d, dpi_count=%d, roce_edpm=%s\n",
+ norm_regsize,
+ pwm_regsize,
+ p_hwfn->dpi_size,
+ p_hwfn->dpi_count,
+ ((p_hwfn->dcbx_no_edpm) || (p_hwfn->db_bar_no_edpm)) ?
+ "disabled" : "enabled");
+
+ if (rc) {
+ DP_ERR(p_hwfn,
+ "Failed to allocate enough DPIs. Allocated %d but the current minimum is %d.\n",
+ p_hwfn->dpi_count,
+ p_hwfn->pf_params.rdma_pf_params.min_dpis);
+ return -EINVAL;
+ }
+
+ /* Update hwfn */
+ p_hwfn->dpi_start_offset = norm_regsize;
+
+ /* Update registers */
+ /* DEMS size is configured log2 of DWORDs, hence the division by 4 */
+ pf_dems_shift = ilog2(QED_PF_DEMS_SIZE / 4);
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_ICID_BIT_SHIFT_NORM, pf_dems_shift);
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_MIN_ADDR_REG1, min_addr_reg1);
+
+ return 0;
+}
+
static int qed_hw_init_port(struct qed_hwfn *p_hwfn,
struct qed_ptt *p_ptt, int hw_mode)
{
@@ -875,6 +1012,10 @@ static int qed_hw_init_pf(struct qed_hwfn *p_hwfn,
/* Pure runtime initializations - directly to the HW */
qed_int_igu_init_pure_rt(p_hwfn, p_ptt, true, true);
+ rc = qed_hw_init_pf_doorbell_bar(p_hwfn, p_ptt);
+ if (rc)
+ return rc;
+
if (b_hw_start) {
/* enable interrupts */
qed_int_igu_enable(p_hwfn, p_ptt, int_mode);
@@ -1299,6 +1440,19 @@ static void qed_hw_set_feat(struct qed_hwfn *p_hwfn)
u32 *feat_num = p_hwfn->hw_info.feat_num;
int num_features = 1;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ /* Roce CNQ require each: 1 status block. 1 CNQ, we divide the
+ * status blocks equally between L2 / RoCE but with consideration as
+ * to how many l2 queues / cnqs we have
+ */
+ if (p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE) {
+ num_features++;
+
+ feat_num[QED_RDMA_CNQ] =
+ min_t(u32, RESC_NUM(p_hwfn, QED_SB) / num_features,
+ RESC_NUM(p_hwfn, QED_RDMA_CNQ_RAM));
+ }
+#endif
feat_num[QED_PF_L2_QUE] = min_t(u32, RESC_NUM(p_hwfn, QED_SB) /
num_features,
RESC_NUM(p_hwfn, QED_L2_QUEUE));
@@ -1340,6 +1494,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
num_funcs;
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
+ resc_num[QED_RDMA_CNQ_RAM] = NUM_OF_CMDQS_CQS / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_main.c b/drivers/net/ethernet/qlogic/qed/qed_main.c
index 0225382..9dab843 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_main.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_main.c
@@ -33,6 +33,11 @@
#include "qed_hw.h"
#include "qed_selftest.h"
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+#define QED_ROCE_QPS (8192)
+#define QED_ROCE_DPIS (8)
+#endif
+
static char version[] =
"QLogic FastLinQ 4xxxx Core Module qed " DRV_MODULE_VERSION "\n";
@@ -206,8 +211,8 @@ int qed_fill_dev_info(struct qed_dev *cdev,
dev_info->pci_mem_start = cdev->pci_params.mem_start;
dev_info->pci_mem_end = cdev->pci_params.mem_end;
dev_info->pci_irq = cdev->pci_params.irq;
- dev_info->rdma_supported =
- (cdev->hwfns[0].hw_info.personality == QED_PCI_ETH_ROCE);
+ dev_info->rdma_supported = (cdev->hwfns[0].hw_info.personality ==
+ QED_PCI_ETH_ROCE);
dev_info->is_mf_default = IS_MF_DEFAULT(&cdev->hwfns[0]);
ether_addr_copy(dev_info->hw_mac, cdev->hwfns[0].hw_info.hw_mac_addr);
@@ -675,6 +680,9 @@ static int qed_slowpath_setup_int(struct qed_dev *cdev,
enum qed_int_mode int_mode)
{
struct qed_sb_cnt_info sb_cnt_info;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ int num_l2_queues;
+#endif
int rc;
int i;
@@ -705,6 +713,31 @@ static int qed_slowpath_setup_int(struct qed_dev *cdev,
cdev->int_params.fp_msix_cnt = cdev->int_params.out.num_vectors -
cdev->num_hwfns;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ num_l2_queues = 0;
+ for_each_hwfn(cdev, i)
+ num_l2_queues += FEAT_NUM(&cdev->hwfns[i], QED_PF_L2_QUE);
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA,
+ "cdev->int_params.fp_msix_cnt=%d num_l2_queues=%d\n",
+ cdev->int_params.fp_msix_cnt, num_l2_queues);
+
+ if (cdev->int_params.fp_msix_cnt > num_l2_queues) {
+ cdev->int_params.rdma_msix_cnt =
+ (cdev->int_params.fp_msix_cnt - num_l2_queues)
+ / cdev->num_hwfns;
+ cdev->int_params.rdma_msix_base =
+ cdev->int_params.fp_msix_base + num_l2_queues;
+ cdev->int_params.fp_msix_cnt = num_l2_queues;
+ } else {
+ cdev->int_params.rdma_msix_cnt = 0;
+ }
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA, "roce_msix_cnt=%d roce_msix_base=%d\n",
+ cdev->int_params.rdma_msix_cnt,
+ cdev->int_params.rdma_msix_base);
+#endif
+
return 0;
}
@@ -808,6 +841,13 @@ static void qed_update_pf_params(struct qed_dev *cdev,
{
int i;
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ params->rdma_pf_params.num_qps = QED_ROCE_QPS;
+ params->rdma_pf_params.min_dpis = QED_ROCE_DPIS;
+ /* divide by 3 the MRs to avoid MF ILT overflow */
+ params->rdma_pf_params.num_mrs = RDMA_MAX_TIDS;
+ params->rdma_pf_params.gl_pi = QED_ROCE_PROTOCOL_INDEX;
+#endif
for (i = 0; i < cdev->num_hwfns; i++) {
struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
diff --git a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
index 103a32c..57c4f1f 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
@@ -549,4 +549,11 @@
#define QM_REG_WFQPFWEIGHT 0x2f4e80UL
#define QM_REG_WFQVPWEIGHT 0x2fa000UL
+#define NIG_REG_ROCE_DUPLICATE_TO_HOST 0x5088f0UL
+#define PRS_REG_LIGHT_L2_ETHERTYPE_EN 0x1f0968UL
+#define NIG_REG_LLH_ENG_CLS_ENG_ID_TBL 0x501b90UL
+#define DORQ_REG_PF_DPM_ENABLE 0x100510UL
+#define DORQ_REG_PF_ICID_BIT_SHIFT_NORM 0x100448UL
+#define DORQ_REG_PF_MIN_ADDR_REG1 0x100400UL
+#define DORQ_REG_PF_DPI_BIT_SHIFT 0x100450UL
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
new file mode 100644
index 0000000..e65a130
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -0,0 +1,887 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/types.h>
+#include <asm/byteorder.h>
+#include <linux/bitops.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/errno.h>
+#include <linux/etherdevice.h>
+#include <linux/if_ether.h>
+#include <linux/if_vlan.h>
+#include <linux/io.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/tcp.h>
+#include <linux/bitops.h>
+#include <linux/qed/qed_roce_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qed.h"
+#include "qed_cxt.h"
+#include "qed_hsi.h"
+#include "qed_hw.h"
+#include "qed_init_ops.h"
+#include "qed_int.h"
+#include "qed_ll2.h"
+#include "qed_mcp.h"
+#include "qed_reg_addr.h"
+#include "qed_sp.h"
+#include "qed_roce.h"
+
+void qed_async_roce_event(struct qed_hwfn *p_hwfn,
+ struct event_ring_entry *p_eqe)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+
+ p_rdma_info->events.affiliated_event(p_rdma_info->events.context,
+ p_eqe->opcode, &p_eqe->data);
+}
+
+static int qed_rdma_bmap_alloc(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 max_count)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "max_count = %08x\n", max_count);
+
+ bmap->max_count = max_count;
+
+ bmap->bitmap = kzalloc(BITS_TO_LONGS(max_count) * sizeof(long),
+ GFP_KERNEL);
+ if (!bmap->bitmap) {
+ DP_NOTICE(p_hwfn,
+ "qed bmap alloc failed: cannot allocate memory (bitmap)\n");
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocated bitmap %p\n",
+ bmap->bitmap);
+ return 0;
+}
+
+static int qed_rdma_bmap_alloc_id(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 *id_num)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "bmap = %p\n", bmap);
+
+ *id_num = find_first_zero_bit(bmap->bitmap, bmap->max_count);
+
+ if (*id_num >= bmap->max_count) {
+ DP_NOTICE(p_hwfn, "no id available max_count=%d\n",
+ bmap->max_count);
+ return -EINVAL;
+ }
+
+ __set_bit(*id_num, bmap->bitmap);
+
+ return 0;
+}
+
+static void qed_bmap_release_id(struct qed_hwfn *p_hwfn,
+ struct qed_bmap *bmap, u32 id_num)
+{
+ bool b_acquired;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "id_num = %08x", id_num);
+ if (id_num >= bmap->max_count)
+ return;
+
+ b_acquired = test_and_clear_bit(id_num, bmap->bitmap);
+ if (!b_acquired) {
+ DP_NOTICE(p_hwfn, "ID %d already released\n", id_num);
+ return;
+ }
+}
+
+u32 qed_rdma_get_sb_id(void *p_hwfn, u32 rel_sb_id)
+{
+ /* First sb id for RoCE is after all the l2 sb */
+ return FEAT_NUM((struct qed_hwfn *)p_hwfn, QED_PF_L2_QUE) + rel_sb_id;
+}
+
+u32 qed_rdma_query_cau_timer_res(void *rdma_cxt)
+{
+ return QED_CAU_DEF_RX_TIMER_RES;
+}
+
+static int qed_rdma_alloc(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_info *p_rdma_info;
+ u32 num_cons, num_tasks;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocating RDMA\n");
+
+ /* Allocate a struct with current pf rdma info */
+ p_rdma_info = kzalloc(sizeof(*p_rdma_info), GFP_KERNEL);
+ if (!p_rdma_info) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ p_rdma_info = p_rdma_info;
+ p_rdma_info->proto = PROTOCOLID_ROCE;
+
+ num_cons = qed_cxt_get_proto_cid_count(p_hwfn, p_rdma_info->proto, 0);
+
+ p_rdma_info->num_qps = num_cons / 2;
+
+ num_tasks = qed_cxt_get_proto_tid_count(p_hwfn, PROTOCOLID_ROCE);
+
+ /* Each MR uses a single task */
+ p_rdma_info->num_mrs = num_tasks;
+
+ /* Queue zone lines are shared between RoCE and L2 in such a way that
+ * they can be used by each without obstructing the other.
+ */
+ p_rdma_info->queue_zone_base = (u16)FEAT_NUM(p_hwfn, QED_L2_QUEUE);
+
+ /* Allocate a struct with device params and fill it */
+ p_rdma_info->dev = kzalloc(sizeof(*p_rdma_info->dev), GFP_KERNEL);
+ if (!p_rdma_info->dev) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info dev). rc = %d\n",
+ rc);
+ goto free_rdma_info;
+ }
+
+ /* Allocate a struct with port params and fill it */
+ p_rdma_info->port = kzalloc(sizeof(*p_rdma_info->port), GFP_KERNEL);
+ if (!p_rdma_info->port) {
+ DP_NOTICE(p_hwfn,
+ "qed rdma alloc failed: cannot allocate memory (rdma info port). rc = %d\n",
+ rc);
+ goto free_rdma_dev;
+ }
+
+ /* Allocate bit map for pd's */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->pd_map, RDMA_MAX_PDS);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate pd_map, rc = %d\n",
+ rc);
+ goto free_rdma_port;
+ }
+
+ /* Allocate DPI bitmap */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->dpi_map,
+ p_hwfn->dpi_count);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate DPI bitmap, rc = %d\n", rc);
+ goto free_pd_map;
+ }
+
+ /* Allocate bitmap for cq's. The maximum number of CQs is bounded to
+ * twice the number of QPs.
+ */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cq_map,
+ p_rdma_info->num_qps * 2);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate cq bitmap, rc = %d\n", rc);
+ goto free_dpi_map;
+ }
+
+ /* Allocate bitmap for toggle bit for cq icids
+ * We toggle the bit every time we create or resize cq for a given icid.
+ * The maximum number of CQs is bounded to twice the number of QPs.
+ */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->toggle_bits,
+ p_rdma_info->num_qps * 2);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate toogle bits, rc = %d\n", rc);
+ goto free_cq_map;
+ }
+
+ /* Allocate bitmap for itids */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->tid_map,
+ p_rdma_info->num_mrs);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate itids bitmaps, rc = %d\n", rc);
+ goto free_toggle_map;
+ }
+
+ /* Allocate bitmap for cids used for qps. */
+ rc = qed_rdma_bmap_alloc(p_hwfn, &p_rdma_info->cid_map, num_cons);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to allocate cid bitmap, rc = %d\n", rc);
+ goto free_tid_map;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocation successful\n");
+ return 0;
+
+free_tid_map:
+ kfree(p_rdma_info->tid_map.bitmap);
+free_toggle_map:
+ kfree(p_rdma_info->toggle_bits.bitmap);
+free_cq_map:
+ kfree(p_rdma_info->cq_map.bitmap);
+free_dpi_map:
+ kfree(p_rdma_info->dpi_map.bitmap);
+free_pd_map:
+ kfree(p_rdma_info->pd_map.bitmap);
+free_rdma_port:
+ kfree(p_rdma_info->port);
+free_rdma_dev:
+ kfree(p_rdma_info->dev);
+free_rdma_info:
+ kfree(p_rdma_info);
+
+ return rc;
+}
+
+void qed_rdma_resc_free(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+
+ kfree(p_rdma_info->cid_map.bitmap);
+ kfree(p_rdma_info->tid_map.bitmap);
+ kfree(p_rdma_info->toggle_bits.bitmap);
+ kfree(p_rdma_info->cq_map.bitmap);
+ kfree(p_rdma_info->dpi_map.bitmap);
+ kfree(p_rdma_info->pd_map.bitmap);
+
+ kfree(p_rdma_info->port);
+ kfree(p_rdma_info->dev);
+
+ kfree(p_rdma_info);
+}
+
+static void qed_rdma_free(struct qed_hwfn *p_hwfn)
+{
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Freeing RDMA\n");
+
+ qed_rdma_resc_free(p_hwfn);
+}
+
+static void qed_rdma_get_guid(struct qed_hwfn *p_hwfn, u8 *guid)
+{
+ guid[0] = p_hwfn->hw_info.hw_mac_addr[0] ^ 2;
+ guid[1] = p_hwfn->hw_info.hw_mac_addr[1];
+ guid[2] = p_hwfn->hw_info.hw_mac_addr[2];
+ guid[3] = 0xff;
+ guid[4] = 0xfe;
+ guid[5] = p_hwfn->hw_info.hw_mac_addr[3];
+ guid[6] = p_hwfn->hw_info.hw_mac_addr[4];
+ guid[7] = p_hwfn->hw_info.hw_mac_addr[5];
+}
+
+static void qed_rdma_init_events(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_events *events;
+
+ events = &p_hwfn->p_rdma_info->events;
+
+ events->unaffiliated_event = params->events->unaffiliated_event;
+ events->affiliated_event = params->events->affiliated_event;
+ events->context = params->events->context;
+}
+
+static void qed_rdma_init_devinfo(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params)
+{
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ u32 pci_status_control;
+ u32 num_qps;
+
+ /* Vendor specific information */
+ dev->vendor_id = cdev->vendor_id;
+ dev->vendor_part_id = cdev->device_id;
+ dev->hw_ver = 0;
+ dev->fw_ver = (FW_MAJOR_VERSION << 24) | (FW_MINOR_VERSION << 16) |
+ (FW_REVISION_VERSION << 8) | (FW_ENGINEERING_VERSION);
+
+ qed_rdma_get_guid(p_hwfn, (u8 *)&dev->sys_image_guid);
+ dev->node_guid = dev->sys_image_guid;
+
+ dev->max_sge = min_t(u32, RDMA_MAX_SGE_PER_SQ_WQE,
+ RDMA_MAX_SGE_PER_RQ_WQE);
+
+ if (cdev->rdma_max_sge)
+ dev->max_sge = min_t(u32, cdev->rdma_max_sge, dev->max_sge);
+
+ dev->max_inline = ROCE_REQ_MAX_INLINE_DATA_SIZE;
+
+ dev->max_inline = (cdev->rdma_max_inline) ?
+ min_t(u32, cdev->rdma_max_inline, dev->max_inline) :
+ dev->max_inline;
+
+ dev->max_wqe = QED_RDMA_MAX_WQE;
+ dev->max_cnq = (u8)FEAT_NUM(p_hwfn, QED_RDMA_CNQ);
+
+ /* The number of QPs may be higher than QED_ROCE_MAX_QPS. because
+ * it is up-aligned to 16 and then to ILT page size within qed cxt.
+ * This is OK in terms of ILT but we don't want to configure the FW
+ * above its abilities
+ */
+ num_qps = ROCE_MAX_QPS;
+ num_qps = min_t(u64, num_qps, p_hwfn->p_rdma_info->num_qps);
+ dev->max_qp = num_qps;
+
+ /* CQs uses the same icids that QPs use hence they are limited by the
+ * number of icids. There are two icids per QP.
+ */
+ dev->max_cq = num_qps * 2;
+
+ /* The number of mrs is smaller by 1 since the first is reserved */
+ dev->max_mr = p_hwfn->p_rdma_info->num_mrs - 1;
+ dev->max_mr_size = QED_RDMA_MAX_MR_SIZE;
+
+ /* The maximum CQE capacity per CQ supported.
+ * max number of cqes will be in two layer pbl,
+ * 8 is the pointer size in bytes
+ * 32 is the size of cq element in bytes
+ */
+ if (params->cq_mode == QED_RDMA_CQ_MODE_32_BITS)
+ dev->max_cqe = QED_RDMA_MAX_CQE_32_BIT;
+ else
+ dev->max_cqe = QED_RDMA_MAX_CQE_16_BIT;
+
+ dev->max_mw = 0;
+ dev->max_fmr = QED_RDMA_MAX_FMR;
+ dev->max_mr_mw_fmr_pbl = (PAGE_SIZE / 8) * (PAGE_SIZE / 8);
+ dev->max_mr_mw_fmr_size = dev->max_mr_mw_fmr_pbl * PAGE_SIZE;
+ dev->max_pkey = QED_RDMA_MAX_P_KEY;
+
+ dev->max_qp_resp_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
+ (RDMA_RESP_RD_ATOMIC_ELM_SIZE * 2);
+ dev->max_qp_req_rd_atomic_resc = RDMA_RING_PAGE_SIZE /
+ RDMA_REQ_RD_ATOMIC_ELM_SIZE;
+ dev->max_dev_resp_rd_atomic_resc = dev->max_qp_resp_rd_atomic_resc *
+ p_hwfn->p_rdma_info->num_qps;
+ dev->page_size_caps = QED_RDMA_PAGE_SIZE_CAPS;
+ dev->dev_ack_delay = QED_RDMA_ACK_DELAY;
+ dev->max_pd = RDMA_MAX_PDS;
+ dev->max_ah = p_hwfn->p_rdma_info->num_qps;
+ dev->max_stats_queues = (u8)RESC_NUM(p_hwfn, QED_RDMA_STATS_QUEUE);
+
+ /* Set capablities */
+ dev->dev_caps = 0;
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RNR_NAK, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_RESIZE_CQ, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_MEMORY_EXT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_BASE_QUEUE_EXT, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ZBVA, 1);
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_LOCAL_INV_FENCE, 1);
+
+ /* Check atomic operations support in PCI configuration space. */
+ pci_read_config_dword(cdev->pdev,
+ cdev->pdev->pcie_cap + PCI_EXP_DEVCTL2,
+ &pci_status_control);
+
+ if (pci_status_control & PCI_EXP_DEVCTL2_LTR_EN)
+ SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ATOMIC_OP, 1);
+}
+
+static void qed_rdma_init_port(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_port *port = p_hwfn->p_rdma_info->port;
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+
+ port->port_state = p_hwfn->mcp_info->link_output.link_up ?
+ QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
+
+ port->max_msg_size = min_t(u64,
+ (dev->max_mr_mw_fmr_size *
+ p_hwfn->cdev->rdma_max_sge),
+ BIT(31));
+
+ port->pkey_bad_counter = 0;
+}
+
+static int qed_rdma_init_hw(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 ll2_ethertype_en;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW\n");
+ p_hwfn->b_rdma_enabled_in_prs = false;
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
+
+ p_hwfn->rdma_prs_search_reg = PRS_REG_SEARCH_ROCE;
+
+ /* We delay writing to this reg until first cid is allocated. See
+ * qed_cxt_dynamic_ilt_alloc function for more details
+ */
+ ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
+ qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
+ (ll2_ethertype_en | 0x01));
+
+ if (qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_ROCE) % 2) {
+ DP_NOTICE(p_hwfn, "The first RoCE's cid should be even\n");
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Initializing HW - Done\n");
+ return 0;
+}
+
+static int qed_rdma_start_fw(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_start_in_params *params,
+ struct qed_ptt *p_ptt)
+{
+ struct rdma_init_func_ramrod_data *p_ramrod;
+ struct qed_rdma_cnq_params *p_cnq_pbl_list;
+ struct rdma_init_func_hdr *p_params_header;
+ struct rdma_cnq_params *p_cnq_params;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ u32 cnq_id, sb_id;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Starting FW\n");
+
+ /* Save the number of cnqs for the function close ramrod */
+ p_hwfn->p_rdma_info->num_cnqs = params->desired_cnq;
+
+ /* Start Roce */
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_INIT,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.roce_init_func.rdma;
+
+ p_params_header = &p_ramrod->params_header;
+ p_params_header->cnq_start_offset = (u8)RESC_START(p_hwfn,
+ QED_RDMA_CNQ_RAM);
+ p_params_header->num_cnqs = params->desired_cnq;
+
+ if (params->cq_mode == QED_RDMA_CQ_MODE_16_BITS)
+ p_params_header->cq_ring_mode = 1;
+ else
+ p_params_header->cq_ring_mode = 0;
+
+ for (cnq_id = 0; cnq_id < params->desired_cnq; cnq_id++) {
+ sb_id = qed_rdma_get_sb_id(p_hwfn, cnq_id);
+ p_cnq_params = &p_ramrod->cnq_params[cnq_id];
+ p_cnq_pbl_list = ¶ms->cnq_pbl_list[cnq_id];
+ p_cnq_params->sb_num =
+ cpu_to_le16(p_hwfn->sbs_info[sb_id]->igu_sb_id);
+
+ p_cnq_params->sb_index = p_hwfn->pf_params.rdma_pf_params.gl_pi;
+ p_cnq_params->num_pbl_pages = p_cnq_pbl_list->num_pbl_pages;
+
+ DMA_REGPAIR_LE(p_cnq_params->pbl_base_addr,
+ p_cnq_pbl_list->pbl_ptr);
+
+ /* we assume here that cnq_id and qz_offset are the same */
+ p_cnq_params->queue_zone_num =
+ cpu_to_le16(p_hwfn->p_rdma_info->queue_zone_base +
+ cnq_id);
+ }
+
+ return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_rdma_reserve_lkey(struct qed_hwfn *p_hwfn)
+{
+ struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
+
+ /* The first DPI is reserved for the Kernel */
+ __set_bit(0, p_hwfn->p_rdma_info->dpi_map.bitmap);
+
+ /* Tid 0 will be used as the key for "reserved MR".
+ * The driver should allocate memory for it so it can be loaded but no
+ * ramrod should be passed on it.
+ */
+ qed_rdma_alloc_tid(p_hwfn, &dev->reserved_lkey);
+ if (dev->reserved_lkey != RDMA_RESERVED_LKEY) {
+ DP_NOTICE(p_hwfn,
+ "Reserved lkey should be equal to RDMA_RESERVED_LKEY\n");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int qed_rdma_setup(struct qed_hwfn *p_hwfn,
+ struct qed_ptt *p_ptt,
+ struct qed_rdma_start_in_params *params)
+{
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA setup\n");
+
+ spin_lock_init(&p_hwfn->p_rdma_info->lock);
+
+ qed_rdma_init_devinfo(p_hwfn, params);
+ qed_rdma_init_port(p_hwfn);
+ qed_rdma_init_events(p_hwfn, params);
+
+ rc = qed_rdma_reserve_lkey(p_hwfn);
+ if (rc)
+ return rc;
+
+ rc = qed_rdma_init_hw(p_hwfn, p_ptt);
+ if (rc)
+ return rc;
+
+ return qed_rdma_start_fw(p_hwfn, params, p_ptt);
+}
+
+int qed_rdma_stop(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_close_func_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ struct qed_ptt *p_ptt;
+ u32 ll2_ethertype_en;
+ int rc = -EBUSY;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop\n");
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Failed to acquire PTT\n");
+ return rc;
+ }
+
+ /* Disable RoCE search */
+ qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 0);
+ p_hwfn->b_rdma_enabled_in_prs = false;
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, 0);
+
+ ll2_ethertype_en = qed_rd(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN);
+
+ qed_wr(p_hwfn, p_ptt, PRS_REG_LIGHT_L2_ETHERTYPE_EN,
+ (ll2_ethertype_en & 0xFFFE));
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ /* Stop RoCE */
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_FUNC_CLOSE,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ goto out;
+
+ p_ramrod = &p_ent->ramrod.rdma_close_func;
+
+ p_ramrod->num_cnqs = p_hwfn->p_rdma_info->num_cnqs;
+ p_ramrod->cnq_start_offset = (u8)RESC_START(p_hwfn, QED_RDMA_CNQ_RAM);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+out:
+ qed_rdma_free(p_hwfn);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA stop done, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_add_user(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ u32 dpi_start_offset;
+ u32 returned_id = 0;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding User\n");
+
+ /* Allocate DPI */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map,
+ &returned_id);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ out_params->dpi = (u16)returned_id;
+
+ /* Calculate the corresponding DPI address */
+ dpi_start_offset = p_hwfn->dpi_start_offset;
+
+ out_params->dpi_addr = (u64)((u8 __iomem *)p_hwfn->doorbells +
+ dpi_start_offset +
+ ((out_params->dpi) * p_hwfn->dpi_size));
+
+ out_params->dpi_phys_addr = p_hwfn->cdev->db_phys_addr +
+ dpi_start_offset +
+ ((out_params->dpi) * p_hwfn->dpi_size);
+
+ out_params->dpi_size = p_hwfn->dpi_size;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Adding user - done, rc = %d\n", rc);
+ return rc;
+}
+
+struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Query device\n");
+
+ /* Return struct with device parameters */
+ return p_hwfn->p_rdma_info->dev;
+}
+
+int qed_rdma_alloc_tid(void *rdma_cxt, u32 *itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID\n");
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn,
+ &p_hwfn->p_rdma_info->tid_map, itid);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+ if (rc)
+ goto out;
+
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_TASK, *itid);
+out:
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Allocate TID - done, rc = %d\n", rc);
+ return rc;
+}
+
+void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 qz_offset, u16 prod)
+{
+ struct qed_hwfn *p_hwfn;
+ u16 qz_num;
+ u32 addr;
+
+ p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ qz_num = p_hwfn->p_rdma_info->queue_zone_base + qz_offset;
+ addr = GTT_BAR0_MAP_REG_USDM_RAM +
+ USTORM_COMMON_QUEUE_CONS_OFFSET(qz_num);
+
+ REG_WR16(p_hwfn, addr, prod);
+
+ /* keep prod updates ordered */
+ wmb();
+}
+
+static int qed_fill_rdma_dev_info(struct qed_dev *cdev,
+ struct qed_dev_rdma_info *info)
+{
+ memset(info, 0, sizeof(*info));
+
+ info->rdma_type = QED_RDMA_TYPE_ROCE;
+
+ qed_fill_dev_info(cdev, &info->common);
+
+ return 0;
+}
+
+static int qed_rdma_get_sb_start(struct qed_dev *cdev)
+{
+ int feat_num;
+
+ if (cdev->num_hwfns > 1)
+ feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE);
+ else
+ feat_num = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_PF_L2_QUE) *
+ cdev->num_hwfns;
+
+ return feat_num;
+}
+
+static int qed_rdma_get_min_cnq_msix(struct qed_dev *cdev)
+{
+ int n_cnq = FEAT_NUM(QED_LEADING_HWFN(cdev), QED_RDMA_CNQ);
+ int n_msix = cdev->int_params.rdma_msix_cnt;
+
+ return min_t(int, n_cnq, n_msix);
+}
+
+static int qed_rdma_set_int(struct qed_dev *cdev, u16 cnt)
+{
+ int limit = 0;
+
+ /* Mark the fastpath as free/used */
+ cdev->int_params.fp_initialized = cnt ? true : false;
+
+ if (cdev->int_params.out.int_mode != QED_INT_MODE_MSIX) {
+ DP_ERR(cdev,
+ "qed roce supports only MSI-X interrupts (detected %d).\n",
+ cdev->int_params.out.int_mode);
+ return -EINVAL;
+ } else if (cdev->int_params.fp_msix_cnt) {
+ limit = cdev->int_params.rdma_msix_cnt;
+ }
+
+ if (!limit)
+ return -ENOMEM;
+
+ return min_t(int, cnt, limit);
+}
+
+static int qed_rdma_get_int(struct qed_dev *cdev, struct qed_int_info *info)
+{
+ memset(info, 0, sizeof(*info));
+
+ if (!cdev->int_params.fp_initialized) {
+ DP_INFO(cdev,
+ "Protocol driver requested interrupt information, but its support is not yet configured\n");
+ return -EINVAL;
+ }
+
+ if (cdev->int_params.out.int_mode == QED_INT_MODE_MSIX) {
+ int msix_base = cdev->int_params.rdma_msix_base;
+
+ info->msix_cnt = cdev->int_params.rdma_msix_cnt;
+ info->msix = &cdev->int_params.msix_table[msix_base];
+
+ DP_VERBOSE(cdev, QED_MSG_RDMA, "msix_cnt = %d msix_base=%d\n",
+ info->msix_cnt, msix_base);
+ }
+
+ return 0;
+}
+
+static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
+{
+ return QED_LEADING_HWFN(cdev);
+}
+
+static void qed_rdma_dpm_conf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ u32 val;
+
+ val = ((p_hwfn->dcbx_no_edpm) || (p_hwfn->db_bar_no_edpm)) ? 0 : 1;
+
+ qed_wr(p_hwfn, p_ptt, DORQ_REG_PF_DPM_ENABLE, val);
+ DP_VERBOSE(p_hwfn, (QED_MSG_DCB | QED_MSG_RDMA),
+ "Changing DPM_EN state to %d (DCBX=%d, DB_BAR=%d)\n",
+ val, p_hwfn->dcbx_no_edpm, p_hwfn->db_bar_no_edpm);
+}
+
+void qed_rdma_dpm_bar(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
+{
+ p_hwfn->db_bar_no_edpm = true;
+
+ qed_rdma_dpm_conf(p_hwfn, p_ptt);
+}
+
+int qed_rdma_start(void *rdma_cxt, struct qed_rdma_start_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_ptt *p_ptt;
+ int rc = -EBUSY;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "desired_cnq = %08x\n", params->desired_cnq);
+
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt)
+ goto err;
+
+ rc = qed_rdma_alloc(p_hwfn, p_ptt, params);
+ if (rc)
+ goto err1;
+
+ rc = qed_rdma_setup(p_hwfn, p_ptt, params);
+ if (rc)
+ goto err2;
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ return rc;
+
+err2:
+ qed_rdma_free(p_hwfn);
+err1:
+ qed_ptt_release(p_hwfn, p_ptt);
+err:
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA start - error, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_rdma_init(struct qed_dev *cdev,
+ struct qed_rdma_start_in_params *params)
+{
+ return qed_rdma_start(QED_LEADING_HWFN(cdev), params);
+}
+
+void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "dpi = %08x\n", dpi);
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->dpi_map, dpi);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
+static const struct qed_rdma_ops qed_rdma_ops_pass = {
+ .common = &qed_common_ops_pass,
+ .fill_dev_info = &qed_fill_rdma_dev_info,
+ .rdma_get_rdma_ctx = &qed_rdma_get_rdma_ctx,
+ .rdma_init = &qed_rdma_init,
+ .rdma_add_user = &qed_rdma_add_user,
+ .rdma_remove_user = &qed_rdma_remove_user,
+ .rdma_stop = &qed_rdma_stop,
+ .rdma_query_device = &qed_rdma_query_device,
+ .rdma_get_start_sb = &qed_rdma_get_sb_start,
+ .rdma_get_rdma_int = &qed_rdma_get_int,
+ .rdma_set_rdma_int = &qed_rdma_set_int,
+ .rdma_get_min_cnq_msix = &qed_rdma_get_min_cnq_msix,
+ .rdma_cnq_prod_update = &qed_rdma_cnq_prod_update,
+};
+
+const struct qed_rdma_ops *qed_get_rdma_ops()
+{
+ return &qed_rdma_ops_pass;
+}
+EXPORT_SYMBOL(qed_get_rdma_ops);
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
new file mode 100644
index 0000000..46f14b2
--- /dev/null
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -0,0 +1,117 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _QED_ROCE_H
+#define _QED_ROCE_H
+#include <linux/types.h>
+#include <linux/bitops.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qed.h"
+#include "qed_dev_api.h"
+#include "qed_hsi.h"
+
+#define QED_RDMA_MAX_FMR (RDMA_MAX_TIDS)
+#define QED_RDMA_MAX_P_KEY (1)
+#define QED_RDMA_MAX_WQE (0x7FFF)
+#define QED_RDMA_MAX_SRQ_WQE_ELEM (0x7FFF)
+#define QED_RDMA_PAGE_SIZE_CAPS (0xFFFFF000)
+#define QED_RDMA_ACK_DELAY (15)
+#define QED_RDMA_MAX_MR_SIZE (0x10000000000ULL)
+#define QED_RDMA_MAX_CQS (RDMA_MAX_CQS)
+#define QED_RDMA_MAX_MRS (RDMA_MAX_TIDS)
+/* Add 1 for header element */
+#define QED_RDMA_MAX_SRQ_ELEM_PER_WQE (RDMA_MAX_SGE_PER_RQ_WQE + 1)
+#define QED_RDMA_MAX_SGE_PER_SRQ_WQE (RDMA_MAX_SGE_PER_RQ_WQE)
+#define QED_RDMA_SRQ_WQE_ELEM_SIZE (16)
+#define QED_RDMA_MAX_SRQS (32 * 1024)
+
+#define QED_RDMA_MAX_CQE_32_BIT (0x7FFFFFFF - 1)
+#define QED_RDMA_MAX_CQE_16_BIT (0x7FFF - 1)
+
+enum qed_rdma_toggle_bit {
+ QED_RDMA_TOGGLE_BIT_CLEAR = 0,
+ QED_RDMA_TOGGLE_BIT_SET = 1
+};
+
+struct qed_bmap {
+ unsigned long *bitmap;
+ u32 max_count;
+};
+
+struct qed_rdma_info {
+ /* spin lock to protect bitmaps */
+ spinlock_t lock;
+
+ struct qed_bmap cq_map;
+ struct qed_bmap pd_map;
+ struct qed_bmap tid_map;
+ struct qed_bmap qp_map;
+ struct qed_bmap srq_map;
+ struct qed_bmap cid_map;
+ struct qed_bmap dpi_map;
+ struct qed_bmap toggle_bits;
+ struct qed_rdma_events events;
+ struct qed_rdma_device *dev;
+ struct qed_rdma_port *port;
+ u32 last_tid;
+ u8 num_cnqs;
+ u32 num_qps;
+ u32 num_mrs;
+ u16 queue_zone_base;
+ enum protocol_type proto;
+};
+
+int
+qed_rdma_add_user(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *out_params);
+int qed_rdma_alloc_pd(void *rdma_cxt, u16 *pd);
+int qed_rdma_alloc_tid(void *rdma_cxt, u32 *tid);
+int qed_rdma_deregister_tid(void *rdma_cxt, u32 tid);
+void qed_rdma_free_tid(void *rdma_cxt, u32 tid);
+struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt);
+int
+qed_rdma_register_tid(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *params);
+void qed_rdma_remove_user(void *rdma_cxt, u16 dpi);
+int qed_rdma_start(void *p_hwfn, struct qed_rdma_start_in_params *params);
+int qed_rdma_stop(void *rdma_cxt);
+u32 qed_rdma_get_sb_id(void *p_hwfn, u32 rel_sb_id);
+u32 qed_rdma_query_cau_timer_res(void *p_hwfn);
+void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 cnq_index, u16 prod);
+void qed_rdma_resc_free(struct qed_hwfn *p_hwfn);
+void qed_async_roce_event(struct qed_hwfn *p_hwfn,
+ struct event_ring_entry *p_eqe);
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_sp.h b/drivers/net/ethernet/qlogic/qed/qed_sp.h
index a3c539f..652c908 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_sp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_sp.h
@@ -85,6 +85,7 @@ union ramrod_data {
struct rdma_srq_create_ramrod_data rdma_create_srq;
struct rdma_srq_destroy_ramrod_data rdma_destroy_srq;
struct rdma_srq_modify_ramrod_data rdma_modify_srq;
+ struct roce_init_func_ramrod_data roce_init_func;
struct iscsi_slow_path_hdr iscsi_empty;
struct iscsi_init_ramrod_params iscsi_init;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_spq.c b/drivers/net/ethernet/qlogic/qed/qed_spq.c
index 0265a32..4e7dd90 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_spq.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_spq.c
@@ -28,6 +28,9 @@
#include "qed_reg_addr.h"
#include "qed_sp.h"
#include "qed_sriov.h"
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+#include "qed_roce.h"
+#endif
/***************************************************************************
* Structures & Definitions
@@ -237,6 +240,11 @@ qed_async_event_completion(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe)
{
switch (p_eqe->protocol_id) {
+#if IS_ENABLED(CONFIG_INFINIBAND_QEDR)
+ case PROTOCOLID_ROCE:
+ qed_async_roce_event(p_hwfn, p_eqe);
+ return 0;
+#endif
case PROTOCOLID_COMMON:
return qed_sriov_eqe_event(p_hwfn,
p_eqe->opcode,
diff --git a/drivers/net/ethernet/qlogic/qede/qede_roce.c b/drivers/net/ethernet/qlogic/qede/qede_roce.c
index b7e0ed9..3d12a44 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_roce.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_roce.c
@@ -33,7 +33,7 @@
#include <linux/netdevice.h>
#include <linux/list.h>
#include <linux/mutex.h>
-#include <linux/qede/qede_roce.h>
+#include <linux/qed/qede_roce.h>
#include "qede.h"
static struct qedr_driver *qedr_drv;
diff --git a/include/linux/qed/qed_if.h b/include/linux/qed/qed_if.h
index cb899b2..1c6e5e0 100644
--- a/include/linux/qed/qed_if.h
+++ b/include/linux/qed/qed_if.h
@@ -34,6 +34,8 @@ enum dcbx_protocol_type {
DCBX_MAX_PROTOCOL_TYPE
};
+#define QED_ROCE_PROTOCOL_INDEX (3)
+
#ifdef CONFIG_DCB
#define QED_LLDP_CHASSIS_ID_STAT_LEN 4
#define QED_LLDP_PORT_ID_STAT_LEN 4
@@ -268,6 +270,7 @@ struct qed_dev_info {
enum qed_sb_type {
QED_SB_TYPE_L2_QUEUE,
+ QED_SB_TYPE_CNQ,
};
enum qed_protocol {
@@ -624,7 +627,7 @@ enum DP_MODULE {
QED_MSG_CXT = 0x800000,
QED_MSG_LL2 = 0x1000000,
QED_MSG_ILT = 0x2000000,
- QED_MSG_ROCE = 0x4000000,
+ QED_MSG_RDMA = 0x4000000,
QED_MSG_DEBUG = 0x8000000,
/* to be added...up to 0x8000000 */
};
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
new file mode 100644
index 0000000..0f7d527
--- /dev/null
+++ b/include/linux/qed/qed_roce_if.h
@@ -0,0 +1,345 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _QED_ROCE_IF_H
+#define _QED_ROCE_IF_H
+#include <linux/types.h>
+#include <linux/delay.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_ll2_if.h>
+
+#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
+
+/* rdma interface */
+enum qed_rdma_tid_type {
+ QED_RDMA_TID_REGISTERED_MR,
+ QED_RDMA_TID_FMR,
+ QED_RDMA_TID_MW_TYPE1,
+ QED_RDMA_TID_MW_TYPE2A
+};
+
+struct qed_rdma_events {
+ void *context;
+ void (*affiliated_event)(void *context, u8 fw_event_code,
+ void *fw_handle);
+ void (*unaffiliated_event)(void *context, u8 event_code);
+};
+
+struct qed_rdma_device {
+ u32 vendor_id;
+ u32 vendor_part_id;
+ u32 hw_ver;
+ u64 fw_ver;
+
+ u64 node_guid;
+ u64 sys_image_guid;
+
+ u8 max_cnq;
+ u8 max_sge;
+ u8 max_srq_sge;
+ u16 max_inline;
+ u32 max_wqe;
+ u32 max_srq_wqe;
+ u8 max_qp_resp_rd_atomic_resc;
+ u8 max_qp_req_rd_atomic_resc;
+ u64 max_dev_resp_rd_atomic_resc;
+ u32 max_cq;
+ u32 max_qp;
+ u32 max_srq;
+ u32 max_mr;
+ u64 max_mr_size;
+ u32 max_cqe;
+ u32 max_mw;
+ u32 max_fmr;
+ u32 max_mr_mw_fmr_pbl;
+ u64 max_mr_mw_fmr_size;
+ u32 max_pd;
+ u32 max_ah;
+ u8 max_pkey;
+ u16 max_srq_wr;
+ u8 max_stats_queues;
+ u32 dev_caps;
+
+ /* Abilty to support RNR-NAK generation */
+
+#define QED_RDMA_DEV_CAP_RNR_NAK_MASK 0x1
+#define QED_RDMA_DEV_CAP_RNR_NAK_SHIFT 0
+ /* Abilty to support shutdown port */
+#define QED_RDMA_DEV_CAP_SHUTDOWN_PORT_MASK 0x1
+#define QED_RDMA_DEV_CAP_SHUTDOWN_PORT_SHIFT 1
+ /* Abilty to support port active event */
+#define QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT_MASK 0x1
+#define QED_RDMA_DEV_CAP_PORT_ACTIVE_EVENT_SHIFT 2
+ /* Abilty to support port change event */
+#define QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT_MASK 0x1
+#define QED_RDMA_DEV_CAP_PORT_CHANGE_EVENT_SHIFT 3
+ /* Abilty to support system image GUID */
+#define QED_RDMA_DEV_CAP_SYS_IMAGE_MASK 0x1
+#define QED_RDMA_DEV_CAP_SYS_IMAGE_SHIFT 4
+ /* Abilty to support bad P_Key counter support */
+#define QED_RDMA_DEV_CAP_BAD_PKEY_CNT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BAD_PKEY_CNT_SHIFT 5
+ /* Abilty to support atomic operations */
+#define QED_RDMA_DEV_CAP_ATOMIC_OP_MASK 0x1
+#define QED_RDMA_DEV_CAP_ATOMIC_OP_SHIFT 6
+#define QED_RDMA_DEV_CAP_RESIZE_CQ_MASK 0x1
+#define QED_RDMA_DEV_CAP_RESIZE_CQ_SHIFT 7
+ /* Abilty to support modifying the maximum number of
+ * outstanding work requests per QP
+ */
+#define QED_RDMA_DEV_CAP_RESIZE_MAX_WR_MASK 0x1
+#define QED_RDMA_DEV_CAP_RESIZE_MAX_WR_SHIFT 8
+ /* Abilty to support automatic path migration */
+#define QED_RDMA_DEV_CAP_AUTO_PATH_MIG_MASK 0x1
+#define QED_RDMA_DEV_CAP_AUTO_PATH_MIG_SHIFT 9
+ /* Abilty to support the base memory management extensions */
+#define QED_RDMA_DEV_CAP_BASE_MEMORY_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BASE_MEMORY_EXT_SHIFT 10
+#define QED_RDMA_DEV_CAP_BASE_QUEUE_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_BASE_QUEUE_EXT_SHIFT 11
+ /* Abilty to support multipile page sizes per memory region */
+#define QED_RDMA_DEV_CAP_MULTI_PAGE_PER_MR_EXT_MASK 0x1
+#define QED_RDMA_DEV_CAP_MULTI_PAGE_PER_MR_EXT_SHIFT 12
+ /* Abilty to support block list physical buffer list */
+#define QED_RDMA_DEV_CAP_BLOCK_MODE_MASK 0x1
+#define QED_RDMA_DEV_CAP_BLOCK_MODE_SHIFT 13
+ /* Abilty to support zero based virtual addresses */
+#define QED_RDMA_DEV_CAP_ZBVA_MASK 0x1
+#define QED_RDMA_DEV_CAP_ZBVA_SHIFT 14
+ /* Abilty to support local invalidate fencing */
+#define QED_RDMA_DEV_CAP_LOCAL_INV_FENCE_MASK 0x1
+#define QED_RDMA_DEV_CAP_LOCAL_INV_FENCE_SHIFT 15
+ /* Abilty to support Loopback on QP */
+#define QED_RDMA_DEV_CAP_LB_INDICATOR_MASK 0x1
+#define QED_RDMA_DEV_CAP_LB_INDICATOR_SHIFT 16
+ u64 page_size_caps;
+ u8 dev_ack_delay;
+ u32 reserved_lkey;
+ u32 bad_pkey_counter;
+ struct qed_rdma_events events;
+};
+
+enum qed_port_state {
+ QED_RDMA_PORT_UP,
+ QED_RDMA_PORT_DOWN,
+};
+
+enum qed_roce_capability {
+ QED_ROCE_V1 = 1 << 0,
+ QED_ROCE_V2 = 1 << 1,
+};
+
+struct qed_rdma_port {
+ enum qed_port_state port_state;
+ int link_speed;
+ u64 max_msg_size;
+ u8 source_gid_table_len;
+ void *source_gid_table_ptr;
+ u8 pkey_table_len;
+ void *pkey_table_ptr;
+ u32 pkey_bad_counter;
+ enum qed_roce_capability capability;
+};
+
+struct qed_rdma_cnq_params {
+ u8 num_pbl_pages;
+ u64 pbl_ptr;
+};
+
+/* The CQ Mode affects the CQ doorbell transaction size.
+ * 64/32 bit machines should configure to 32/16 bits respectively.
+ */
+enum qed_rdma_cq_mode {
+ QED_RDMA_CQ_MODE_16_BITS,
+ QED_RDMA_CQ_MODE_32_BITS,
+};
+
+struct qed_roce_dcqcn_params {
+ u8 notification_point;
+ u8 reaction_point;
+
+ /* fields for notification point */
+ u32 cnp_send_timeout;
+
+ /* fields for reaction point */
+ u32 rl_bc_rate;
+ u16 rl_max_rate;
+ u16 rl_r_ai;
+ u16 rl_r_hai;
+ u16 dcqcn_g;
+ u32 dcqcn_k_us;
+ u32 dcqcn_timeout_us;
+};
+
+struct qed_rdma_start_in_params {
+ struct qed_rdma_events *events;
+ struct qed_rdma_cnq_params cnq_pbl_list[128];
+ u8 desired_cnq;
+ enum qed_rdma_cq_mode cq_mode;
+ struct qed_roce_dcqcn_params dcqcn_params;
+ u16 max_mtu;
+ u8 mac_addr[ETH_ALEN];
+ u8 iwarp_flags;
+};
+
+struct qed_rdma_add_user_out_params {
+ u16 dpi;
+ u64 dpi_addr;
+ u64 dpi_phys_addr;
+ u32 dpi_size;
+};
+
+enum roce_mode {
+ ROCE_V1,
+ ROCE_V2_IPV4,
+ ROCE_V2_IPV6,
+ MAX_ROCE_MODE
+};
+
+union qed_gid {
+ u8 bytes[16];
+ u16 words[8];
+ u32 dwords[4];
+ u64 qwords[2];
+ u32 ipv4_addr;
+};
+
+struct qed_rdma_register_tid_in_params {
+ u32 itid;
+ enum qed_rdma_tid_type tid_type;
+ u8 key;
+ u16 pd;
+ bool local_read;
+ bool local_write;
+ bool remote_read;
+ bool remote_write;
+ bool remote_atomic;
+ bool mw_bind;
+ u64 pbl_ptr;
+ bool pbl_two_level;
+ u8 pbl_page_size_log;
+ u8 page_size_log;
+ u32 fbo;
+ u64 length;
+ u64 vaddr;
+ bool zbva;
+ bool phy_mr;
+ bool dma_mr;
+
+ bool dif_enabled;
+ u64 dif_error_addr;
+ u64 dif_runt_addr;
+};
+
+struct qed_rdma_create_srq_in_params {
+ u64 pbl_base_addr;
+ u64 prod_pair_addr;
+ u16 num_pages;
+ u16 pd_id;
+ u16 page_size;
+};
+
+struct qed_rdma_create_srq_out_params {
+ u16 srq_id;
+};
+
+struct qed_rdma_destroy_srq_in_params {
+ u16 srq_id;
+};
+
+struct qed_rdma_modify_srq_in_params {
+ u32 wqe_limit;
+ u16 srq_id;
+};
+
+struct qed_rdma_stats_out_params {
+ u64 sent_bytes;
+ u64 sent_pkts;
+ u64 rcv_bytes;
+ u64 rcv_pkts;
+};
+
+struct qed_rdma_counters_out_params {
+ u64 pd_count;
+ u64 max_pd;
+ u64 dpi_count;
+ u64 max_dpi;
+ u64 cq_count;
+ u64 max_cq;
+ u64 qp_count;
+ u64 max_qp;
+ u64 tid_count;
+ u64 max_tid;
+};
+
+#define QED_ROCE_TX_HEAD_FAILURE (1)
+#define QED_ROCE_TX_FRAG_FAILURE (2)
+
+enum qed_rdma_type {
+ QED_RDMA_TYPE_ROCE,
+};
+
+struct qed_dev_rdma_info {
+ struct qed_dev_info common;
+ enum qed_rdma_type rdma_type;
+};
+
+struct qed_rdma_ops {
+ const struct qed_common_ops *common;
+
+ int (*fill_dev_info)(struct qed_dev *cdev,
+ struct qed_dev_rdma_info *info);
+ void *(*rdma_get_rdma_ctx)(struct qed_dev *cdev);
+
+ int (*rdma_init)(struct qed_dev *dev,
+ struct qed_rdma_start_in_params *iparams);
+
+ int (*rdma_add_user)(void *rdma_cxt,
+ struct qed_rdma_add_user_out_params *oparams);
+
+ void (*rdma_remove_user)(void *rdma_cxt, u16 dpi);
+ int (*rdma_stop)(void *rdma_cxt);
+ struct qed_rdma_device* (*rdma_query_device)(void *rdma_cxt);
+ int (*rdma_get_start_sb)(struct qed_dev *cdev);
+ int (*rdma_get_min_cnq_msix)(struct qed_dev *cdev);
+ void (*rdma_cnq_prod_update)(void *rdma_cxt, u8 cnq_index, u16 prod);
+ int (*rdma_get_rdma_int)(struct qed_dev *cdev,
+ struct qed_int_info *info);
+ int (*rdma_set_rdma_int)(struct qed_dev *cdev, u16 cnt);
+};
+
+const struct qed_rdma_ops *qed_get_rdma_ops(void);
+
+#endif
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 04/11] Add support for user context verbs
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for ucontext, query port, add and del gid verbs.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/Makefile | 2 +-
drivers/infiniband/hw/qedr/main.c | 44 ++-
drivers/infiniband/hw/qedr/qedr.h | 38 +++
drivers/infiniband/hw/qedr/qedr_user.h | 27 ++
drivers/infiniband/hw/qedr/verbs.c | 417 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 26 ++
drivers/net/ethernet/qlogic/qed/qed.h | 2 +-
drivers/net/ethernet/qlogic/qed/qed_dev.c | 1 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 19 +-
drivers/net/ethernet/qlogic/qed/qed_roce.h | 1 +
include/linux/qed/qed_roce_if.h | 1 +
include/linux/qed/rdma_common.h | 1 +
12 files changed, 568 insertions(+), 11 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_user.h
create mode 100644 drivers/infiniband/hw/qedr/verbs.c
create mode 100644 drivers/infiniband/hw/qedr/verbs.h
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
index 3a5b7a2..b10f2b1 100644
--- a/drivers/infiniband/hw/qedr/Makefile
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
-qedr-y := main.o
+qedr-y := main.o verbs.o
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 0b5274a..a01983c 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -32,6 +32,7 @@
#include <linux/module.h>
#include <rdma/ib_verbs.h>
#include <rdma/ib_addr.h>
+#include <rdma/ib_user_verbs.h>
#include <linux/netdevice.h>
#include <linux/iommu.h>
#include <net/addrconf.h>
@@ -39,6 +40,8 @@
#include <linux/qed/qed_chain.h>
#include <linux/qed/qed_if.h>
#include "qedr.h"
+#include "verbs.h"
+#include "qedr_user.h"
MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
MODULE_AUTHOR("QLogic Corporation");
@@ -70,14 +73,47 @@ static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
return IB_LINK_LAYER_ETHERNET;
}
+static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str,
+ size_t str_len)
+{
+ struct qedr_dev *qedr = get_qedr_dev(ibdev);
+ u32 fw_ver = (u32)qedr->attr.fw_ver;
+
+ snprintf(str, str_len, "%d. %d. %d. %d",
+ (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF,
+ (fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
+}
+
static int qedr_register_device(struct qedr_dev *dev)
{
strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
dev->ibdev.owner = THIS_MODULE;
+ dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
+
+ dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
+ QEDR_UVERBS(QUERY_DEVICE) |
+ QEDR_UVERBS(QUERY_PORT);
+
+ dev->ibdev.phys_port_cnt = 1;
+ dev->ibdev.num_comp_vectors = dev->num_cnq;
+ dev->ibdev.node_type = RDMA_NODE_IB_CA;
+ dev->ibdev.query_device = qedr_query_device;
+ dev->ibdev.query_port = qedr_query_port;
+
+ dev->ibdev.query_gid = qedr_query_gid;
+ dev->ibdev.add_gid = qedr_add_gid;
+ dev->ibdev.del_gid = qedr_del_gid;
+
+ dev->ibdev.alloc_ucontext = qedr_alloc_ucontext;
+ dev->ibdev.dealloc_ucontext = qedr_dealloc_ucontext;
+ dev->ibdev.mmap = qedr_mmap;
+
+ dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
+ dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
return 0;
}
@@ -224,12 +260,6 @@ static ssize_t show_rev(struct device *device, struct device_attribute *attr,
return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
}
-static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
- char *buf)
-{
- return scnprintf(buf, PAGE_SIZE, "%s\n", "FW_VER_TO_ADD");
-}
-
static ssize_t show_hca_type(struct device *device,
struct device_attribute *attr, char *buf)
{
@@ -237,12 +267,10 @@ static ssize_t show_hca_type(struct device *device,
}
static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
-static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
static struct device_attribute *qedr_attributes[] = {
&dev_attr_hw_rev,
- &dev_attr_fw_ver,
&dev_attr_hca_type
};
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 0663a4c9..ed3ef33 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -46,6 +46,7 @@
enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
+ QEDR_MSG_MISC = 0x400000,
};
struct qedr_dev;
@@ -175,4 +176,41 @@ struct qedr_dev {
#define QEDR_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
+#define QEDR_ROCE_PKEY_MAX 1
+#define QEDR_ROCE_PKEY_TABLE_LEN 1
+#define QEDR_ROCE_PKEY_DEFAULT 0xffff
+
+struct qedr_ucontext {
+ struct ib_ucontext ibucontext;
+ struct qedr_dev *dev;
+ struct qedr_pd *pd;
+ u64 dpi_addr;
+ u64 dpi_phys_addr;
+ u32 dpi_size;
+ u16 dpi;
+
+ struct list_head mm_head;
+
+ /* Lock to protect mm list */
+ struct mutex mm_list_lock;
+};
+
+struct qedr_mm {
+ struct {
+ u64 phy_addr;
+ unsigned long len;
+ } key;
+ struct list_head entry;
+};
+
+static inline struct qedr_dev *get_qedr_dev(struct ib_device *ibdev)
+{
+ return container_of(ibdev, struct qedr_dev, ibdev);
+}
+
+static inline
+struct qedr_ucontext *get_qedr_ucontext(struct ib_ucontext *ibucontext)
+{
+ return container_of(ibucontext, struct qedr_ucontext, ibucontext);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_user.h b/drivers/infiniband/hw/qedr/qedr_user.h
new file mode 100644
index 0000000..45288dc
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_user.h
@@ -0,0 +1,27 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#ifndef __QEDR_USER_H__
+#define __QEDR_USER_H__
+
+#define QEDR_ABI_VERSION (6)
+
+/* user kernel communication data structures. */
+
+struct qedr_alloc_ucontext_resp {
+ u64 db_pa;
+ u32 db_size;
+
+ u32 max_send_wr;
+ u32 max_recv_wr;
+ u32 max_srq_wr;
+ u32 sges_per_send_wr;
+ u32 sges_per_recv_wr;
+ u32 sges_per_srq_wr;
+ int max_cqes;
+};
+#endif /* __QEDR_USER_H__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
new file mode 100644
index 0000000..e765b50
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -0,0 +1,417 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#include <linux/dma-mapping.h>
+#include <linux/crc32.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/udp.h>
+#include <linux/iommu.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_cache.h>
+
+#include "qedr_hsi.h"
+#include <linux/qed/qed_if.h>
+#include "qedr.h"
+#include "verbs.h"
+#include "qedr_user.h"
+
+int qedr_query_gid(struct ib_device *ibdev, u8 port, int index,
+ union ib_gid *sgid)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ int rc = 0;
+
+ if (!rdma_cap_roce_gid_table(ibdev, port))
+ return -ENODEV;
+
+ rc = ib_get_cached_gid(ibdev, port, index, sgid, NULL);
+ if (rc == -EAGAIN) {
+ memcpy(sgid, &zgid, sizeof(*sgid));
+ return 0;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "query gid: index=%d %llx:%llx\n", index,
+ sgid->global.interface_id, sgid->global.subnet_prefix);
+
+ return rc;
+}
+
+int qedr_add_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, const union ib_gid *gid,
+ const struct ib_gid_attr *attr, void **context)
+{
+ if (!rdma_cap_roce_gid_table(device, port_num))
+ return -EINVAL;
+
+ if (port_num > QEDR_MAX_PORT)
+ return -EINVAL;
+
+ if (!context)
+ return -EINVAL;
+
+ return 0;
+}
+
+int qedr_del_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, void **context)
+{
+ if (!rdma_cap_roce_gid_table(device, port_num))
+ return -EINVAL;
+
+ if (port_num > QEDR_MAX_PORT)
+ return -EINVAL;
+
+ if (!context)
+ return -EINVAL;
+
+ return 0;
+}
+
+int qedr_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *attr, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qedr_device_attr *qattr = &dev->attr;
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev,
+ "qedr_query_device called with invalid params rdma_ctx=%p\n",
+ dev->rdma_ctx);
+ return -EINVAL;
+ }
+
+ memset(attr, 0, sizeof(*attr));
+
+ attr->fw_ver = qattr->fw_ver;
+ attr->sys_image_guid = qattr->sys_image_guid;
+ attr->max_mr_size = qattr->max_mr_size;
+ attr->page_size_cap = qattr->page_size_caps;
+ attr->vendor_id = qattr->vendor_id;
+ attr->vendor_part_id = qattr->vendor_part_id;
+ attr->hw_ver = qattr->hw_ver;
+ attr->max_qp = qattr->max_qp;
+ attr->max_qp_wr = max_t(u32, qattr->max_sqe, qattr->max_rqe);
+ attr->device_cap_flags = IB_DEVICE_CURR_QP_STATE_MOD |
+ IB_DEVICE_RC_RNR_NAK_GEN |
+ IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_MGT_EXTENSIONS;
+
+ attr->max_sge = qattr->max_sge;
+ attr->max_sge_rd = qattr->max_sge;
+ attr->max_cq = qattr->max_cq;
+ attr->max_cqe = qattr->max_cqe;
+ attr->max_mr = qattr->max_mr;
+ attr->max_mw = qattr->max_mw;
+ attr->max_pd = qattr->max_pd;
+ attr->atomic_cap = dev->atomic_cap;
+ attr->max_fmr = qattr->max_fmr;
+ attr->max_map_per_fmr = 16;
+ attr->max_qp_init_rd_atom =
+ 1 << (fls(qattr->max_qp_req_rd_atomic_resc) - 1);
+ attr->max_qp_rd_atom =
+ min(1 << (fls(qattr->max_qp_resp_rd_atomic_resc) - 1),
+ attr->max_qp_init_rd_atom);
+
+ attr->max_srq = qattr->max_srq;
+ attr->max_srq_sge = qattr->max_srq_sge;
+ attr->max_srq_wr = qattr->max_srq_wr;
+
+ attr->local_ca_ack_delay = qattr->dev_ack_delay;
+ attr->max_fast_reg_page_list_len = qattr->max_mr / 8;
+ attr->max_pkeys = QEDR_ROCE_PKEY_MAX;
+ attr->max_ah = qattr->max_ah;
+
+ return 0;
+}
+
+#define QEDR_SPEED_SDR (1)
+#define QEDR_SPEED_DDR (2)
+#define QEDR_SPEED_QDR (4)
+#define QEDR_SPEED_FDR10 (8)
+#define QEDR_SPEED_FDR (16)
+#define QEDR_SPEED_EDR (32)
+
+static inline void get_link_speed_and_width(int speed, u8 *ib_speed,
+ u8 *ib_width)
+{
+ switch (speed) {
+ case 1000:
+ *ib_speed = QEDR_SPEED_SDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+ case 10000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+
+ case 20000:
+ *ib_speed = QEDR_SPEED_DDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 25000:
+ *ib_speed = QEDR_SPEED_EDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+
+ case 40000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 50000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 100000:
+ *ib_speed = QEDR_SPEED_EDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ default:
+ /* Unsupported */
+ *ib_speed = QEDR_SPEED_SDR;
+ *ib_width = IB_WIDTH_1X;
+ }
+}
+
+int qedr_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr *attr)
+{
+ struct qedr_dev *dev;
+ struct qed_rdma_port *rdma_port;
+
+ dev = get_qedr_dev(ibdev);
+ if (port > 1) {
+ DP_ERR(dev, "invalid_port=0x%x\n", port);
+ return -EINVAL;
+ }
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev, "rdma_ctx is NULL\n");
+ return -EINVAL;
+ }
+
+ rdma_port = dev->ops->rdma_query_port(dev->rdma_ctx);
+ memset(attr, 0, sizeof(*attr));
+
+ if (rdma_port->port_state == QED_RDMA_PORT_UP) {
+ attr->state = IB_PORT_ACTIVE;
+ attr->phys_state = 5;
+ } else {
+ attr->state = IB_PORT_DOWN;
+ attr->phys_state = 3;
+ }
+ attr->max_mtu = IB_MTU_4096;
+ attr->active_mtu = iboe_get_mtu(dev->ndev->mtu);
+ attr->lid = 0;
+ attr->lmc = 0;
+ attr->sm_lid = 0;
+ attr->sm_sl = 0;
+ attr->port_cap_flags = IB_PORT_IP_BASED_GIDS;
+ attr->gid_tbl_len = QEDR_MAX_SGID;
+ attr->pkey_tbl_len = QEDR_ROCE_PKEY_TABLE_LEN;
+ attr->bad_pkey_cntr = rdma_port->pkey_bad_counter;
+ attr->qkey_viol_cntr = 0;
+ get_link_speed_and_width(rdma_port->link_speed,
+ &attr->active_speed, &attr->active_width);
+ attr->max_msg_sz = rdma_port->max_msg_size;
+ attr->max_vl_num = 4;
+
+ return 0;
+}
+
+static int qedr_add_mmap(struct qedr_ucontext *uctx, u64 phy_addr,
+ unsigned long len)
+{
+ struct qedr_mm *mm;
+
+ mm = kzalloc(sizeof(*mm), GFP_KERNEL);
+ if (!mm)
+ return -ENOMEM;
+ mm->key.phy_addr = phy_addr;
+ /* This function might be called with a length which is not a multiple
+ * of PAGE_SIZE, while the mapping is PAGE_SIZE grained and the kernel
+ * forces this granularity by increasing the requested size if needed.
+ * When qedr_mmap is called, it will search the list with the updated
+ * length as a key. To prevent search failures, the length is rounded up
+ * in advance to PAGE_SIZE.
+ */
+ mm->key.len = roundup(len, PAGE_SIZE);
+ INIT_LIST_HEAD(&mm->entry);
+
+ mutex_lock(&uctx->mm_list_lock);
+ list_add(&mm->entry, &uctx->mm_head);
+ mutex_unlock(&uctx->mm_list_lock);
+
+ DP_VERBOSE(uctx->dev, QEDR_MSG_MISC,
+ "added (addr=0x%llx,len=0x%lx) for ctx=%p\n",
+ (unsigned long long)mm->key.phy_addr,
+ (unsigned long)mm->key.len, uctx);
+
+ return 0;
+}
+
+static bool qedr_search_mmap(struct qedr_ucontext *uctx, u64 phy_addr,
+ unsigned long len)
+{
+ bool found = false;
+ struct qedr_mm *mm;
+
+ mutex_lock(&uctx->mm_list_lock);
+ list_for_each_entry(mm, &uctx->mm_head, entry) {
+ if (len != mm->key.len || phy_addr != mm->key.phy_addr)
+ continue;
+
+ found = true;
+ break;
+ }
+ mutex_unlock(&uctx->mm_list_lock);
+ DP_VERBOSE(uctx->dev, QEDR_MSG_MISC,
+ "searched for (addr=0x%llx,len=0x%lx) for ctx=%p, result=%d\n",
+ mm->key.phy_addr, mm->key.len, uctx, found);
+
+ return found;
+}
+
+struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *ibdev,
+ struct ib_udata *udata)
+{
+ int rc;
+ struct qedr_ucontext *ctx;
+ struct qedr_alloc_ucontext_resp uresp;
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qed_rdma_add_user_out_params oparams;
+
+ if (!udata)
+ return ERR_PTR(-EFAULT);
+
+ ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+ if (!ctx)
+ return ERR_PTR(-ENOMEM);
+
+ rc = dev->ops->rdma_add_user(dev->rdma_ctx, &oparams);
+ if (rc) {
+ DP_ERR(dev,
+ "failed to allocate a DPI for a new RoCE application, rc=%d. To overcome this consider to increase the number of DPIs, increase the doorbell BAR size or just close unnecessary RoCE applications. In order to increase the number of DPIs consult the QEDR readme\n",
+ rc);
+ goto err;
+ }
+
+ ctx->dpi = oparams.dpi;
+ ctx->dpi_addr = oparams.dpi_addr;
+ ctx->dpi_phys_addr = oparams.dpi_phys_addr;
+ ctx->dpi_size = oparams.dpi_size;
+ INIT_LIST_HEAD(&ctx->mm_head);
+ mutex_init(&ctx->mm_list_lock);
+
+ memset(&uresp, 0, sizeof(uresp));
+
+ uresp.db_pa = ctx->dpi_phys_addr;
+ uresp.db_size = ctx->dpi_size;
+ uresp.max_send_wr = dev->attr.max_sqe;
+ uresp.max_recv_wr = dev->attr.max_rqe;
+ uresp.max_srq_wr = dev->attr.max_srq_wr;
+ uresp.sges_per_send_wr = QEDR_MAX_SQE_ELEMENTS_PER_SQE;
+ uresp.sges_per_recv_wr = QEDR_MAX_RQE_ELEMENTS_PER_RQE;
+ uresp.sges_per_srq_wr = dev->attr.max_srq_sge;
+ uresp.max_cqes = QEDR_MAX_CQES;
+
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ goto err;
+
+ ctx->dev = dev;
+
+ rc = qedr_add_mmap(ctx, ctx->dpi_phys_addr, ctx->dpi_size);
+ if (rc)
+ goto err;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Allocating user context %p\n",
+ &ctx->ibucontext);
+ return &ctx->ibucontext;
+
+err:
+ kfree(ctx);
+ return ERR_PTR(rc);
+}
+
+int qedr_dealloc_ucontext(struct ib_ucontext *ibctx)
+{
+ struct qedr_ucontext *uctx = get_qedr_ucontext(ibctx);
+ struct qedr_mm *mm, *tmp;
+ int status = 0;
+
+ DP_VERBOSE(uctx->dev, QEDR_MSG_INIT, "Deallocating user context %p\n",
+ uctx);
+ uctx->dev->ops->rdma_remove_user(uctx->dev->rdma_ctx, uctx->dpi);
+
+ list_for_each_entry_safe(mm, tmp, &uctx->mm_head, entry) {
+ DP_VERBOSE(uctx->dev, QEDR_MSG_MISC,
+ "deleted (addr=0x%llx,len=0x%lx) for ctx=%p\n",
+ mm->key.phy_addr, mm->key.len, uctx);
+ list_del(&mm->entry);
+ kfree(mm);
+ }
+
+ kfree(uctx);
+ return status;
+}
+
+int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+ struct qedr_ucontext *ucontext = get_qedr_ucontext(context);
+ struct qedr_dev *dev = get_qedr_dev(context->device);
+ unsigned long vm_page = vma->vm_pgoff << PAGE_SHIFT;
+ u64 unmapped_db = dev->db_phys_addr;
+ unsigned long len = (vma->vm_end - vma->vm_start);
+ int rc = 0;
+ bool found;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT,
+ "qedr_mmap called vm_page=0x%lx vm_pgoff=0x%lx unmapped_db=0x%llx db_size=%x, len=%lx\n",
+ vm_page, vma->vm_pgoff, unmapped_db, dev->db_size, len);
+ if (vma->vm_start & (PAGE_SIZE - 1)) {
+ DP_ERR(dev, "Vma_start not page aligned = %ld\n",
+ vma->vm_start);
+ return -EINVAL;
+ }
+
+ found = qedr_search_mmap(ucontext, vm_page, len);
+ if (!found) {
+ DP_ERR(dev, "Vma_pgoff not found in mapped array = %ld\n",
+ vma->vm_pgoff);
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n");
+
+ if ((vm_page >= unmapped_db) && (vm_page <= (unmapped_db +
+ dev->db_size))) {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n");
+ if (vma->vm_flags & VM_READ) {
+ DP_ERR(dev, "Trying to map doorbell bar for read\n");
+ return -EPERM;
+ }
+
+ vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+
+ rc = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
+ PAGE_SIZE, vma->vm_page_prot);
+ } else {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Mapping chains\n");
+ rc = remap_pfn_range(vma, vma->vm_start,
+ vma->vm_pgoff, len, vma->vm_page_prot);
+ }
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_mmap return code: %d\n", rc);
+ return rc;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
new file mode 100644
index 0000000..f731063
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -0,0 +1,26 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#ifndef __QEDR_VERBS_H__
+#define __QEDR_VERBS_H__
+
+int qedr_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *attr, struct ib_udata *udata);
+int qedr_query_port(struct ib_device *, u8 port, struct ib_port_attr *props);
+
+int qedr_query_gid(struct ib_device *, u8 port, int index, union ib_gid *gid);
+
+struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *, struct ib_udata *);
+int qedr_dealloc_ucontext(struct ib_ucontext *);
+
+int qedr_mmap(struct ib_ucontext *, struct vm_area_struct *vma);
+int qedr_del_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, void **context);
+int qedr_add_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, const union ib_gid *gid,
+ const struct ib_gid_attr *attr, void **context);
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index f5ff97b..9640475 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -52,7 +52,7 @@ enum qed_mcp_protocol_type;
static inline u32 qed_db_addr(u32 cid, u32 DEMS)
{
u32 db_addr = FIELD_VALUE(DB_LEGACY_ADDR_DEMS, DEMS) |
- FIELD_VALUE(DB_LEGACY_ADDR_ICID, cid);
+ (cid * QED_PF_DEMS_SIZE);
return db_addr;
}
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 219b3b7..4ae8ed0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -1495,6 +1495,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
resc_num[QED_RDMA_CNQ_RAM] = NUM_OF_CMDQS_CQS / num_funcs;
+ resc_num[QED_RDMA_STATS_QUEUE] = RDMA_NUM_STATISTIC_COUNTERS_BB / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index e65a130..9cdb472 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -158,7 +158,7 @@ static int qed_rdma_alloc(struct qed_hwfn *p_hwfn,
return rc;
}
- p_rdma_info = p_rdma_info;
+ p_hwfn->p_rdma_info = p_rdma_info;
p_rdma_info->proto = PROTOCOLID_ROCE;
num_cons = qed_cxt_get_proto_cid_count(p_hwfn, p_rdma_info->proto, 0);
@@ -664,6 +664,22 @@ int qed_rdma_add_user(void *rdma_cxt,
return rc;
}
+struct qed_rdma_port *qed_rdma_query_port(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_rdma_port *p_port = p_hwfn->p_rdma_info->port;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA Query port\n");
+
+ /* Link may have changed */
+ p_port->port_state = p_hwfn->mcp_info->link_output.link_up ?
+ QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
+
+ p_port->link_speed = p_hwfn->mcp_info->link_output.speed;
+
+ return p_port;
+}
+
struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
{
struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
@@ -872,6 +888,7 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_add_user = &qed_rdma_add_user,
.rdma_remove_user = &qed_rdma_remove_user,
.rdma_stop = &qed_rdma_stop,
+ .rdma_query_port = &qed_rdma_query_port,
.rdma_query_device = &qed_rdma_query_device,
.rdma_get_start_sb = &qed_rdma_get_sb_start,
.rdma_get_rdma_int = &qed_rdma_get_int,
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 46f14b2..c94c39e 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -102,6 +102,7 @@ int qed_rdma_alloc_tid(void *rdma_cxt, u32 *tid);
int qed_rdma_deregister_tid(void *rdma_cxt, u32 tid);
void qed_rdma_free_tid(void *rdma_cxt, u32 tid);
struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt);
+struct qed_rdma_port *qed_rdma_query_port(void *rdma_cxt);
int
qed_rdma_register_tid(void *rdma_cxt,
struct qed_rdma_register_tid_in_params *params);
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 0f7d527..e99470c 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -332,6 +332,7 @@ struct qed_rdma_ops {
void (*rdma_remove_user)(void *rdma_cxt, u16 dpi);
int (*rdma_stop)(void *rdma_cxt);
struct qed_rdma_device* (*rdma_query_device)(void *rdma_cxt);
+ struct qed_rdma_port* (*rdma_query_port)(void *rdma_cxt);
int (*rdma_get_start_sb)(struct qed_dev *cdev);
int (*rdma_get_min_cnq_msix)(struct qed_dev *cdev);
void (*rdma_cnq_prod_update)(void *rdma_cxt, u8 cnq_index, u16 prod);
diff --git a/include/linux/qed/rdma_common.h b/include/linux/qed/rdma_common.h
index 187991c..7663725 100644
--- a/include/linux/qed/rdma_common.h
+++ b/include/linux/qed/rdma_common.h
@@ -28,6 +28,7 @@
#define RDMA_MAX_PDS (64 * 1024)
#define RDMA_NUM_STATISTIC_COUNTERS MAX_NUM_VPORTS
+#define RDMA_NUM_STATISTIC_COUNTERS_BB MAX_NUM_VPORTS_BB
#define RDMA_TASK_TYPE (PROTOCOLID_ROCE)
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 04/11] Add support for user context verbs
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for ucontext, query port, add and del gid verbs.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/Makefile | 2 +-
drivers/infiniband/hw/qedr/main.c | 44 ++-
drivers/infiniband/hw/qedr/qedr.h | 38 +++
drivers/infiniband/hw/qedr/qedr_user.h | 27 ++
drivers/infiniband/hw/qedr/verbs.c | 417 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 26 ++
drivers/net/ethernet/qlogic/qed/qed.h | 2 +-
drivers/net/ethernet/qlogic/qed/qed_dev.c | 1 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 19 +-
drivers/net/ethernet/qlogic/qed/qed_roce.h | 1 +
include/linux/qed/qed_roce_if.h | 1 +
include/linux/qed/rdma_common.h | 1 +
12 files changed, 568 insertions(+), 11 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_user.h
create mode 100644 drivers/infiniband/hw/qedr/verbs.c
create mode 100644 drivers/infiniband/hw/qedr/verbs.h
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
index 3a5b7a2..b10f2b1 100644
--- a/drivers/infiniband/hw/qedr/Makefile
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
-qedr-y := main.o
+qedr-y := main.o verbs.o
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 0b5274a..a01983c 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -32,6 +32,7 @@
#include <linux/module.h>
#include <rdma/ib_verbs.h>
#include <rdma/ib_addr.h>
+#include <rdma/ib_user_verbs.h>
#include <linux/netdevice.h>
#include <linux/iommu.h>
#include <net/addrconf.h>
@@ -39,6 +40,8 @@
#include <linux/qed/qed_chain.h>
#include <linux/qed/qed_if.h>
#include "qedr.h"
+#include "verbs.h"
+#include "qedr_user.h"
MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
MODULE_AUTHOR("QLogic Corporation");
@@ -70,14 +73,47 @@ static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
return IB_LINK_LAYER_ETHERNET;
}
+static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str,
+ size_t str_len)
+{
+ struct qedr_dev *qedr = get_qedr_dev(ibdev);
+ u32 fw_ver = (u32)qedr->attr.fw_ver;
+
+ snprintf(str, str_len, "%d. %d. %d. %d",
+ (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF,
+ (fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
+}
+
static int qedr_register_device(struct qedr_dev *dev)
{
strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
dev->ibdev.owner = THIS_MODULE;
+ dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
+
+ dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
+ QEDR_UVERBS(QUERY_DEVICE) |
+ QEDR_UVERBS(QUERY_PORT);
+
+ dev->ibdev.phys_port_cnt = 1;
+ dev->ibdev.num_comp_vectors = dev->num_cnq;
+ dev->ibdev.node_type = RDMA_NODE_IB_CA;
+ dev->ibdev.query_device = qedr_query_device;
+ dev->ibdev.query_port = qedr_query_port;
+
+ dev->ibdev.query_gid = qedr_query_gid;
+ dev->ibdev.add_gid = qedr_add_gid;
+ dev->ibdev.del_gid = qedr_del_gid;
+
+ dev->ibdev.alloc_ucontext = qedr_alloc_ucontext;
+ dev->ibdev.dealloc_ucontext = qedr_dealloc_ucontext;
+ dev->ibdev.mmap = qedr_mmap;
+
+ dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
+ dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
return 0;
}
@@ -224,12 +260,6 @@ static ssize_t show_rev(struct device *device, struct device_attribute *attr,
return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
}
-static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
- char *buf)
-{
- return scnprintf(buf, PAGE_SIZE, "%s\n", "FW_VER_TO_ADD");
-}
-
static ssize_t show_hca_type(struct device *device,
struct device_attribute *attr, char *buf)
{
@@ -237,12 +267,10 @@ static ssize_t show_hca_type(struct device *device,
}
static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
-static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
static struct device_attribute *qedr_attributes[] = {
&dev_attr_hw_rev,
- &dev_attr_fw_ver,
&dev_attr_hca_type
};
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 0663a4c9..ed3ef33 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -46,6 +46,7 @@
enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
+ QEDR_MSG_MISC = 0x400000,
};
struct qedr_dev;
@@ -175,4 +176,41 @@ struct qedr_dev {
#define QEDR_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
+#define QEDR_ROCE_PKEY_MAX 1
+#define QEDR_ROCE_PKEY_TABLE_LEN 1
+#define QEDR_ROCE_PKEY_DEFAULT 0xffff
+
+struct qedr_ucontext {
+ struct ib_ucontext ibucontext;
+ struct qedr_dev *dev;
+ struct qedr_pd *pd;
+ u64 dpi_addr;
+ u64 dpi_phys_addr;
+ u32 dpi_size;
+ u16 dpi;
+
+ struct list_head mm_head;
+
+ /* Lock to protect mm list */
+ struct mutex mm_list_lock;
+};
+
+struct qedr_mm {
+ struct {
+ u64 phy_addr;
+ unsigned long len;
+ } key;
+ struct list_head entry;
+};
+
+static inline struct qedr_dev *get_qedr_dev(struct ib_device *ibdev)
+{
+ return container_of(ibdev, struct qedr_dev, ibdev);
+}
+
+static inline
+struct qedr_ucontext *get_qedr_ucontext(struct ib_ucontext *ibucontext)
+{
+ return container_of(ibucontext, struct qedr_ucontext, ibucontext);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_user.h b/drivers/infiniband/hw/qedr/qedr_user.h
new file mode 100644
index 0000000..45288dc
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_user.h
@@ -0,0 +1,27 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#ifndef __QEDR_USER_H__
+#define __QEDR_USER_H__
+
+#define QEDR_ABI_VERSION (6)
+
+/* user kernel communication data structures. */
+
+struct qedr_alloc_ucontext_resp {
+ u64 db_pa;
+ u32 db_size;
+
+ u32 max_send_wr;
+ u32 max_recv_wr;
+ u32 max_srq_wr;
+ u32 sges_per_send_wr;
+ u32 sges_per_recv_wr;
+ u32 sges_per_srq_wr;
+ int max_cqes;
+};
+#endif /* __QEDR_USER_H__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
new file mode 100644
index 0000000..e765b50
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -0,0 +1,417 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#include <linux/dma-mapping.h>
+#include <linux/crc32.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/udp.h>
+#include <linux/iommu.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_cache.h>
+
+#include "qedr_hsi.h"
+#include <linux/qed/qed_if.h>
+#include "qedr.h"
+#include "verbs.h"
+#include "qedr_user.h"
+
+int qedr_query_gid(struct ib_device *ibdev, u8 port, int index,
+ union ib_gid *sgid)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ int rc = 0;
+
+ if (!rdma_cap_roce_gid_table(ibdev, port))
+ return -ENODEV;
+
+ rc = ib_get_cached_gid(ibdev, port, index, sgid, NULL);
+ if (rc == -EAGAIN) {
+ memcpy(sgid, &zgid, sizeof(*sgid));
+ return 0;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "query gid: index=%d %llx:%llx\n", index,
+ sgid->global.interface_id, sgid->global.subnet_prefix);
+
+ return rc;
+}
+
+int qedr_add_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, const union ib_gid *gid,
+ const struct ib_gid_attr *attr, void **context)
+{
+ if (!rdma_cap_roce_gid_table(device, port_num))
+ return -EINVAL;
+
+ if (port_num > QEDR_MAX_PORT)
+ return -EINVAL;
+
+ if (!context)
+ return -EINVAL;
+
+ return 0;
+}
+
+int qedr_del_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, void **context)
+{
+ if (!rdma_cap_roce_gid_table(device, port_num))
+ return -EINVAL;
+
+ if (port_num > QEDR_MAX_PORT)
+ return -EINVAL;
+
+ if (!context)
+ return -EINVAL;
+
+ return 0;
+}
+
+int qedr_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *attr, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qedr_device_attr *qattr = &dev->attr;
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev,
+ "qedr_query_device called with invalid params rdma_ctx=%p\n",
+ dev->rdma_ctx);
+ return -EINVAL;
+ }
+
+ memset(attr, 0, sizeof(*attr));
+
+ attr->fw_ver = qattr->fw_ver;
+ attr->sys_image_guid = qattr->sys_image_guid;
+ attr->max_mr_size = qattr->max_mr_size;
+ attr->page_size_cap = qattr->page_size_caps;
+ attr->vendor_id = qattr->vendor_id;
+ attr->vendor_part_id = qattr->vendor_part_id;
+ attr->hw_ver = qattr->hw_ver;
+ attr->max_qp = qattr->max_qp;
+ attr->max_qp_wr = max_t(u32, qattr->max_sqe, qattr->max_rqe);
+ attr->device_cap_flags = IB_DEVICE_CURR_QP_STATE_MOD |
+ IB_DEVICE_RC_RNR_NAK_GEN |
+ IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_MGT_EXTENSIONS;
+
+ attr->max_sge = qattr->max_sge;
+ attr->max_sge_rd = qattr->max_sge;
+ attr->max_cq = qattr->max_cq;
+ attr->max_cqe = qattr->max_cqe;
+ attr->max_mr = qattr->max_mr;
+ attr->max_mw = qattr->max_mw;
+ attr->max_pd = qattr->max_pd;
+ attr->atomic_cap = dev->atomic_cap;
+ attr->max_fmr = qattr->max_fmr;
+ attr->max_map_per_fmr = 16;
+ attr->max_qp_init_rd_atom =
+ 1 << (fls(qattr->max_qp_req_rd_atomic_resc) - 1);
+ attr->max_qp_rd_atom =
+ min(1 << (fls(qattr->max_qp_resp_rd_atomic_resc) - 1),
+ attr->max_qp_init_rd_atom);
+
+ attr->max_srq = qattr->max_srq;
+ attr->max_srq_sge = qattr->max_srq_sge;
+ attr->max_srq_wr = qattr->max_srq_wr;
+
+ attr->local_ca_ack_delay = qattr->dev_ack_delay;
+ attr->max_fast_reg_page_list_len = qattr->max_mr / 8;
+ attr->max_pkeys = QEDR_ROCE_PKEY_MAX;
+ attr->max_ah = qattr->max_ah;
+
+ return 0;
+}
+
+#define QEDR_SPEED_SDR (1)
+#define QEDR_SPEED_DDR (2)
+#define QEDR_SPEED_QDR (4)
+#define QEDR_SPEED_FDR10 (8)
+#define QEDR_SPEED_FDR (16)
+#define QEDR_SPEED_EDR (32)
+
+static inline void get_link_speed_and_width(int speed, u8 *ib_speed,
+ u8 *ib_width)
+{
+ switch (speed) {
+ case 1000:
+ *ib_speed = QEDR_SPEED_SDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+ case 10000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+
+ case 20000:
+ *ib_speed = QEDR_SPEED_DDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 25000:
+ *ib_speed = QEDR_SPEED_EDR;
+ *ib_width = IB_WIDTH_1X;
+ break;
+
+ case 40000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 50000:
+ *ib_speed = QEDR_SPEED_QDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ case 100000:
+ *ib_speed = QEDR_SPEED_EDR;
+ *ib_width = IB_WIDTH_4X;
+ break;
+
+ default:
+ /* Unsupported */
+ *ib_speed = QEDR_SPEED_SDR;
+ *ib_width = IB_WIDTH_1X;
+ }
+}
+
+int qedr_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr *attr)
+{
+ struct qedr_dev *dev;
+ struct qed_rdma_port *rdma_port;
+
+ dev = get_qedr_dev(ibdev);
+ if (port > 1) {
+ DP_ERR(dev, "invalid_port=0x%x\n", port);
+ return -EINVAL;
+ }
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev, "rdma_ctx is NULL\n");
+ return -EINVAL;
+ }
+
+ rdma_port = dev->ops->rdma_query_port(dev->rdma_ctx);
+ memset(attr, 0, sizeof(*attr));
+
+ if (rdma_port->port_state == QED_RDMA_PORT_UP) {
+ attr->state = IB_PORT_ACTIVE;
+ attr->phys_state = 5;
+ } else {
+ attr->state = IB_PORT_DOWN;
+ attr->phys_state = 3;
+ }
+ attr->max_mtu = IB_MTU_4096;
+ attr->active_mtu = iboe_get_mtu(dev->ndev->mtu);
+ attr->lid = 0;
+ attr->lmc = 0;
+ attr->sm_lid = 0;
+ attr->sm_sl = 0;
+ attr->port_cap_flags = IB_PORT_IP_BASED_GIDS;
+ attr->gid_tbl_len = QEDR_MAX_SGID;
+ attr->pkey_tbl_len = QEDR_ROCE_PKEY_TABLE_LEN;
+ attr->bad_pkey_cntr = rdma_port->pkey_bad_counter;
+ attr->qkey_viol_cntr = 0;
+ get_link_speed_and_width(rdma_port->link_speed,
+ &attr->active_speed, &attr->active_width);
+ attr->max_msg_sz = rdma_port->max_msg_size;
+ attr->max_vl_num = 4;
+
+ return 0;
+}
+
+static int qedr_add_mmap(struct qedr_ucontext *uctx, u64 phy_addr,
+ unsigned long len)
+{
+ struct qedr_mm *mm;
+
+ mm = kzalloc(sizeof(*mm), GFP_KERNEL);
+ if (!mm)
+ return -ENOMEM;
+ mm->key.phy_addr = phy_addr;
+ /* This function might be called with a length which is not a multiple
+ * of PAGE_SIZE, while the mapping is PAGE_SIZE grained and the kernel
+ * forces this granularity by increasing the requested size if needed.
+ * When qedr_mmap is called, it will search the list with the updated
+ * length as a key. To prevent search failures, the length is rounded up
+ * in advance to PAGE_SIZE.
+ */
+ mm->key.len = roundup(len, PAGE_SIZE);
+ INIT_LIST_HEAD(&mm->entry);
+
+ mutex_lock(&uctx->mm_list_lock);
+ list_add(&mm->entry, &uctx->mm_head);
+ mutex_unlock(&uctx->mm_list_lock);
+
+ DP_VERBOSE(uctx->dev, QEDR_MSG_MISC,
+ "added (addr=0x%llx,len=0x%lx) for ctx=%p\n",
+ (unsigned long long)mm->key.phy_addr,
+ (unsigned long)mm->key.len, uctx);
+
+ return 0;
+}
+
+static bool qedr_search_mmap(struct qedr_ucontext *uctx, u64 phy_addr,
+ unsigned long len)
+{
+ bool found = false;
+ struct qedr_mm *mm;
+
+ mutex_lock(&uctx->mm_list_lock);
+ list_for_each_entry(mm, &uctx->mm_head, entry) {
+ if (len != mm->key.len || phy_addr != mm->key.phy_addr)
+ continue;
+
+ found = true;
+ break;
+ }
+ mutex_unlock(&uctx->mm_list_lock);
+ DP_VERBOSE(uctx->dev, QEDR_MSG_MISC,
+ "searched for (addr=0x%llx,len=0x%lx) for ctx=%p, result=%d\n",
+ mm->key.phy_addr, mm->key.len, uctx, found);
+
+ return found;
+}
+
+struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *ibdev,
+ struct ib_udata *udata)
+{
+ int rc;
+ struct qedr_ucontext *ctx;
+ struct qedr_alloc_ucontext_resp uresp;
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qed_rdma_add_user_out_params oparams;
+
+ if (!udata)
+ return ERR_PTR(-EFAULT);
+
+ ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+ if (!ctx)
+ return ERR_PTR(-ENOMEM);
+
+ rc = dev->ops->rdma_add_user(dev->rdma_ctx, &oparams);
+ if (rc) {
+ DP_ERR(dev,
+ "failed to allocate a DPI for a new RoCE application, rc=%d. To overcome this consider to increase the number of DPIs, increase the doorbell BAR size or just close unnecessary RoCE applications. In order to increase the number of DPIs consult the QEDR readme\n",
+ rc);
+ goto err;
+ }
+
+ ctx->dpi = oparams.dpi;
+ ctx->dpi_addr = oparams.dpi_addr;
+ ctx->dpi_phys_addr = oparams.dpi_phys_addr;
+ ctx->dpi_size = oparams.dpi_size;
+ INIT_LIST_HEAD(&ctx->mm_head);
+ mutex_init(&ctx->mm_list_lock);
+
+ memset(&uresp, 0, sizeof(uresp));
+
+ uresp.db_pa = ctx->dpi_phys_addr;
+ uresp.db_size = ctx->dpi_size;
+ uresp.max_send_wr = dev->attr.max_sqe;
+ uresp.max_recv_wr = dev->attr.max_rqe;
+ uresp.max_srq_wr = dev->attr.max_srq_wr;
+ uresp.sges_per_send_wr = QEDR_MAX_SQE_ELEMENTS_PER_SQE;
+ uresp.sges_per_recv_wr = QEDR_MAX_RQE_ELEMENTS_PER_RQE;
+ uresp.sges_per_srq_wr = dev->attr.max_srq_sge;
+ uresp.max_cqes = QEDR_MAX_CQES;
+
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ goto err;
+
+ ctx->dev = dev;
+
+ rc = qedr_add_mmap(ctx, ctx->dpi_phys_addr, ctx->dpi_size);
+ if (rc)
+ goto err;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Allocating user context %p\n",
+ &ctx->ibucontext);
+ return &ctx->ibucontext;
+
+err:
+ kfree(ctx);
+ return ERR_PTR(rc);
+}
+
+int qedr_dealloc_ucontext(struct ib_ucontext *ibctx)
+{
+ struct qedr_ucontext *uctx = get_qedr_ucontext(ibctx);
+ struct qedr_mm *mm, *tmp;
+ int status = 0;
+
+ DP_VERBOSE(uctx->dev, QEDR_MSG_INIT, "Deallocating user context %p\n",
+ uctx);
+ uctx->dev->ops->rdma_remove_user(uctx->dev->rdma_ctx, uctx->dpi);
+
+ list_for_each_entry_safe(mm, tmp, &uctx->mm_head, entry) {
+ DP_VERBOSE(uctx->dev, QEDR_MSG_MISC,
+ "deleted (addr=0x%llx,len=0x%lx) for ctx=%p\n",
+ mm->key.phy_addr, mm->key.len, uctx);
+ list_del(&mm->entry);
+ kfree(mm);
+ }
+
+ kfree(uctx);
+ return status;
+}
+
+int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+ struct qedr_ucontext *ucontext = get_qedr_ucontext(context);
+ struct qedr_dev *dev = get_qedr_dev(context->device);
+ unsigned long vm_page = vma->vm_pgoff << PAGE_SHIFT;
+ u64 unmapped_db = dev->db_phys_addr;
+ unsigned long len = (vma->vm_end - vma->vm_start);
+ int rc = 0;
+ bool found;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT,
+ "qedr_mmap called vm_page=0x%lx vm_pgoff=0x%lx unmapped_db=0x%llx db_size=%x, len=%lx\n",
+ vm_page, vma->vm_pgoff, unmapped_db, dev->db_size, len);
+ if (vma->vm_start & (PAGE_SIZE - 1)) {
+ DP_ERR(dev, "Vma_start not page aligned = %ld\n",
+ vma->vm_start);
+ return -EINVAL;
+ }
+
+ found = qedr_search_mmap(ucontext, vm_page, len);
+ if (!found) {
+ DP_ERR(dev, "Vma_pgoff not found in mapped array = %ld\n",
+ vma->vm_pgoff);
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n");
+
+ if ((vm_page >= unmapped_db) && (vm_page <= (unmapped_db +
+ dev->db_size))) {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Mapping doorbell bar\n");
+ if (vma->vm_flags & VM_READ) {
+ DP_ERR(dev, "Trying to map doorbell bar for read\n");
+ return -EPERM;
+ }
+
+ vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+
+ rc = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
+ PAGE_SIZE, vma->vm_page_prot);
+ } else {
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Mapping chains\n");
+ rc = remap_pfn_range(vma, vma->vm_start,
+ vma->vm_pgoff, len, vma->vm_page_prot);
+ }
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_mmap return code: %d\n", rc);
+ return rc;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
new file mode 100644
index 0000000..f731063
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -0,0 +1,26 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#ifndef __QEDR_VERBS_H__
+#define __QEDR_VERBS_H__
+
+int qedr_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *attr, struct ib_udata *udata);
+int qedr_query_port(struct ib_device *, u8 port, struct ib_port_attr *props);
+
+int qedr_query_gid(struct ib_device *, u8 port, int index, union ib_gid *gid);
+
+struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *, struct ib_udata *);
+int qedr_dealloc_ucontext(struct ib_ucontext *);
+
+int qedr_mmap(struct ib_ucontext *, struct vm_area_struct *vma);
+int qedr_del_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, void **context);
+int qedr_add_gid(struct ib_device *device, u8 port_num,
+ unsigned int index, const union ib_gid *gid,
+ const struct ib_gid_attr *attr, void **context);
+#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index f5ff97b..9640475 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -52,7 +52,7 @@ enum qed_mcp_protocol_type;
static inline u32 qed_db_addr(u32 cid, u32 DEMS)
{
u32 db_addr = FIELD_VALUE(DB_LEGACY_ADDR_DEMS, DEMS) |
- FIELD_VALUE(DB_LEGACY_ADDR_ICID, cid);
+ (cid * QED_PF_DEMS_SIZE);
return db_addr;
}
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 219b3b7..4ae8ed0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -1495,6 +1495,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
resc_num[QED_RDMA_CNQ_RAM] = NUM_OF_CMDQS_CQS / num_funcs;
+ resc_num[QED_RDMA_STATS_QUEUE] = RDMA_NUM_STATISTIC_COUNTERS_BB / num_funcs;
for (i = 0; i < QED_MAX_RESC; i++)
resc_start[i] = resc_num[i] * enabled_func_idx;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index e65a130..9cdb472 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -158,7 +158,7 @@ static int qed_rdma_alloc(struct qed_hwfn *p_hwfn,
return rc;
}
- p_rdma_info = p_rdma_info;
+ p_hwfn->p_rdma_info = p_rdma_info;
p_rdma_info->proto = PROTOCOLID_ROCE;
num_cons = qed_cxt_get_proto_cid_count(p_hwfn, p_rdma_info->proto, 0);
@@ -664,6 +664,22 @@ int qed_rdma_add_user(void *rdma_cxt,
return rc;
}
+struct qed_rdma_port *qed_rdma_query_port(void *rdma_cxt)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_rdma_port *p_port = p_hwfn->p_rdma_info->port;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "RDMA Query port\n");
+
+ /* Link may have changed */
+ p_port->port_state = p_hwfn->mcp_info->link_output.link_up ?
+ QED_RDMA_PORT_UP : QED_RDMA_PORT_DOWN;
+
+ p_port->link_speed = p_hwfn->mcp_info->link_output.speed;
+
+ return p_port;
+}
+
struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
{
struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
@@ -872,6 +888,7 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_add_user = &qed_rdma_add_user,
.rdma_remove_user = &qed_rdma_remove_user,
.rdma_stop = &qed_rdma_stop,
+ .rdma_query_port = &qed_rdma_query_port,
.rdma_query_device = &qed_rdma_query_device,
.rdma_get_start_sb = &qed_rdma_get_sb_start,
.rdma_get_rdma_int = &qed_rdma_get_int,
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 46f14b2..c94c39e 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -102,6 +102,7 @@ int qed_rdma_alloc_tid(void *rdma_cxt, u32 *tid);
int qed_rdma_deregister_tid(void *rdma_cxt, u32 tid);
void qed_rdma_free_tid(void *rdma_cxt, u32 tid);
struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt);
+struct qed_rdma_port *qed_rdma_query_port(void *rdma_cxt);
int
qed_rdma_register_tid(void *rdma_cxt,
struct qed_rdma_register_tid_in_params *params);
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 0f7d527..e99470c 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -332,6 +332,7 @@ struct qed_rdma_ops {
void (*rdma_remove_user)(void *rdma_cxt, u16 dpi);
int (*rdma_stop)(void *rdma_cxt);
struct qed_rdma_device* (*rdma_query_device)(void *rdma_cxt);
+ struct qed_rdma_port* (*rdma_query_port)(void *rdma_cxt);
int (*rdma_get_start_sb)(struct qed_dev *cdev);
int (*rdma_get_min_cnq_msix)(struct qed_dev *cdev);
void (*rdma_cnq_prod_update)(void *rdma_cxt, u8 cnq_index, u16 prod);
diff --git a/include/linux/qed/rdma_common.h b/include/linux/qed/rdma_common.h
index 187991c..7663725 100644
--- a/include/linux/qed/rdma_common.h
+++ b/include/linux/qed/rdma_common.h
@@ -28,6 +28,7 @@
#define RDMA_MAX_PDS (64 * 1024)
#define RDMA_NUM_STATISTIC_COUNTERS MAX_NUM_VPORTS
+#define RDMA_NUM_STATISTIC_COUNTERS_BB MAX_NUM_VPORTS_BB
#define RDMA_TASK_TYPE (PROTOCOLID_ROCE)
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 05/11] Add support for PD,PKEY and CQ verbs
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for protection domain and completion queue verbs.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 18 +-
drivers/infiniband/hw/qedr/qedr.h | 78 +++++
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 79 +++++
drivers/infiniband/hw/qedr/qedr_user.h | 19 +
drivers/infiniband/hw/qedr/verbs.c | 543 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 14 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 313 +++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 20 ++
include/linux/qed/qed_roce_if.h | 29 ++
9 files changed, 1112 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index a01983c..02f0dbb 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -94,7 +94,13 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
QEDR_UVERBS(QUERY_DEVICE) |
- QEDR_UVERBS(QUERY_PORT);
+ QEDR_UVERBS(QUERY_PORT) |
+ QEDR_UVERBS(ALLOC_PD) |
+ QEDR_UVERBS(DEALLOC_PD) |
+ QEDR_UVERBS(CREATE_CQ) |
+ QEDR_UVERBS(RESIZE_CQ) |
+ QEDR_UVERBS(DESTROY_CQ) |
+ QEDR_UVERBS(REQ_NOTIFY_CQ);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -111,6 +117,16 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.dealloc_ucontext = qedr_dealloc_ucontext;
dev->ibdev.mmap = qedr_mmap;
+ dev->ibdev.alloc_pd = qedr_alloc_pd;
+ dev->ibdev.dealloc_pd = qedr_dealloc_pd;
+
+ dev->ibdev.create_cq = qedr_create_cq;
+ dev->ibdev.destroy_cq = qedr_destroy_cq;
+ dev->ibdev.resize_cq = qedr_resize_cq;
+ dev->ibdev.req_notify_cq = qedr_arm_cq;
+
+ dev->ibdev.query_pkey = qedr_query_pkey;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index ed3ef33..b2aa8fd 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -44,8 +44,12 @@
#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
#define DP_NAME(dev) ((dev)->ibdev.name)
+#define QEDR_CQ_MAGIC_NUMBER (0x11223344)
+
enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
+ QEDR_MSG_CQ = 0x20000,
+ QEDR_MSG_MR = 0x100000,
QEDR_MSG_MISC = 0x400000,
};
@@ -180,6 +184,12 @@ struct qedr_dev {
#define QEDR_ROCE_PKEY_TABLE_LEN 1
#define QEDR_ROCE_PKEY_DEFAULT 0xffff
+struct qedr_pbl {
+ struct list_head list_entry;
+ void *va;
+ dma_addr_t pa;
+};
+
struct qedr_ucontext {
struct ib_ucontext ibucontext;
struct qedr_dev *dev;
@@ -195,6 +205,64 @@ struct qedr_ucontext {
struct mutex mm_list_lock;
};
+union db_prod64 {
+ struct rdma_pwm_val32_data data;
+ u64 raw;
+};
+
+enum qedr_cq_type {
+ QEDR_CQ_TYPE_GSI,
+ QEDR_CQ_TYPE_KERNEL,
+ QEDR_CQ_TYPE_USER
+};
+
+struct qedr_pbl_info {
+ u32 num_pbls;
+ u32 num_pbes;
+ u32 pbl_size;
+ u32 pbe_size;
+ bool two_layered;
+};
+
+struct qedr_userq {
+ struct ib_umem *umem;
+ struct qedr_pbl_info pbl_info;
+ struct qedr_pbl *pbl_tbl;
+ u64 buf_addr;
+ size_t buf_len;
+};
+
+struct qedr_cq {
+ struct ib_cq ibcq;
+
+ enum qedr_cq_type cq_type;
+ u32 sig;
+
+ u16 icid;
+
+ /* Lock to protect multiplem CQ's */
+ spinlock_t cq_lock;
+ u8 arm_flags;
+ struct qed_chain pbl;
+
+ void __iomem *db_addr;
+ union db_prod64 db;
+
+ u8 pbl_toggle;
+ union rdma_cqe *latest_cqe;
+ union rdma_cqe *toggle_cqe;
+
+ u32 cq_cons;
+
+ struct qedr_userq q;
+};
+
+struct qedr_pd {
+ struct ib_pd ibpd;
+ u32 pd_id;
+ struct qedr_ucontext *uctx;
+};
+
struct qedr_mm {
struct {
u64 phy_addr;
@@ -213,4 +281,14 @@ struct qedr_ucontext *get_qedr_ucontext(struct ib_ucontext *ibucontext)
{
return container_of(ibucontext, struct qedr_ucontext, ibucontext);
}
+
+static inline struct qedr_pd *get_qedr_pd(struct ib_pd *ibpd)
+{
+ return container_of(ibpd, struct qedr_pd, ibpd);
+}
+
+static inline struct qedr_cq *get_qedr_cq(struct ib_cq *ibcq)
+{
+ return container_of(ibcq, struct qedr_cq, ibcq);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
index 3e508fb..84f6520 100644
--- a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -47,6 +47,19 @@ struct rdma_cqe_responder {
__le32 imm_data_hi;
__le16 rq_cons;
u8 flags;
+#define RDMA_CQE_RESPONDER_TOGGLE_BIT_MASK 0x1
+#define RDMA_CQE_RESPONDER_TOGGLE_BIT_SHIFT 0
+#define RDMA_CQE_RESPONDER_TYPE_MASK 0x3
+#define RDMA_CQE_RESPONDER_TYPE_SHIFT 1
+#define RDMA_CQE_RESPONDER_INV_FLG_MASK 0x1
+#define RDMA_CQE_RESPONDER_INV_FLG_SHIFT 3
+#define RDMA_CQE_RESPONDER_IMM_FLG_MASK 0x1
+#define RDMA_CQE_RESPONDER_IMM_FLG_SHIFT 4
+#define RDMA_CQE_RESPONDER_RDMA_FLG_MASK 0x1
+#define RDMA_CQE_RESPONDER_RDMA_FLG_SHIFT 5
+#define RDMA_CQE_RESPONDER_RESERVED2_MASK 0x3
+#define RDMA_CQE_RESPONDER_RESERVED2_SHIFT 6
+ u8 status;
};
struct rdma_cqe_requester {
@@ -58,6 +71,12 @@ struct rdma_cqe_requester {
__le32 reserved3;
__le16 reserved4;
u8 flags;
+#define RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK 0x1
+#define RDMA_CQE_REQUESTER_TOGGLE_BIT_SHIFT 0
+#define RDMA_CQE_REQUESTER_TYPE_MASK 0x3
+#define RDMA_CQE_REQUESTER_TYPE_SHIFT 1
+#define RDMA_CQE_REQUESTER_RESERVED5_MASK 0x1F
+#define RDMA_CQE_REQUESTER_RESERVED5_SHIFT 3
u8 status;
};
@@ -66,6 +85,12 @@ struct rdma_cqe_common {
struct regpair qp_handle;
__le16 reserved1[7];
u8 flags;
+#define RDMA_CQE_COMMON_TOGGLE_BIT_MASK 0x1
+#define RDMA_CQE_COMMON_TOGGLE_BIT_SHIFT 0
+#define RDMA_CQE_COMMON_TYPE_MASK 0x3
+#define RDMA_CQE_COMMON_TYPE_SHIFT 1
+#define RDMA_CQE_COMMON_RESERVED2_MASK 0x1F
+#define RDMA_CQE_COMMON_RESERVED2_SHIFT 3
u8 status;
};
@@ -76,6 +101,45 @@ union rdma_cqe {
struct rdma_cqe_common cmn;
};
+/* * CQE requester status enumeration */
+enum rdma_cqe_requester_status_enum {
+ RDMA_CQE_REQ_STS_OK,
+ RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR,
+ RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR,
+ RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR,
+ RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR,
+ RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR,
+ RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR,
+ RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR,
+ RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR,
+ RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR,
+ RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR,
+ RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR,
+ MAX_RDMA_CQE_REQUESTER_STATUS_ENUM
+};
+
+/* CQE responder status enumeration */
+enum rdma_cqe_responder_status_enum {
+ RDMA_CQE_RESP_STS_OK,
+ RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR,
+ RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR,
+ RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR,
+ RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR,
+ RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR,
+ RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR,
+ RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR,
+ MAX_RDMA_CQE_RESPONDER_STATUS_ENUM
+};
+
+/* CQE type enumeration */
+enum rdma_cqe_type {
+ RDMA_CQE_TYPE_REQUESTER,
+ RDMA_CQE_TYPE_RESPONDER_RQ,
+ RDMA_CQE_TYPE_RESPONDER_SRQ,
+ RDMA_CQE_TYPE_INVALID,
+ MAX_RDMA_CQE_TYPE
+};
+
struct rdma_sq_sge {
__le32 length;
struct regpair addr;
@@ -93,4 +157,19 @@ struct rdma_srq_sge {
__le32 length;
__le32 l_key;
};
+
+/* Rdma doorbell data for CQ */
+struct rdma_pwm_val32_data {
+ __le16 icid;
+ u8 agg_flags;
+ u8 params;
+#define RDMA_PWM_VAL32_DATA_AGG_CMD_MASK 0x3
+#define RDMA_PWM_VAL32_DATA_AGG_CMD_SHIFT 0
+#define RDMA_PWM_VAL32_DATA_BYPASS_EN_MASK 0x1
+#define RDMA_PWM_VAL32_DATA_BYPASS_EN_SHIFT 2
+#define RDMA_PWM_VAL32_DATA_RESERVED_MASK 0x1F
+#define RDMA_PWM_VAL32_DATA_RESERVED_SHIFT 3
+ __le32 value;
+};
+
#endif /* __QED_HSI_RDMA__ */
diff --git a/drivers/infiniband/hw/qedr/qedr_user.h b/drivers/infiniband/hw/qedr/qedr_user.h
index 45288dc..5832c30 100644
--- a/drivers/infiniband/hw/qedr/qedr_user.h
+++ b/drivers/infiniband/hw/qedr/qedr_user.h
@@ -24,4 +24,23 @@ struct qedr_alloc_ucontext_resp {
u32 sges_per_srq_wr;
int max_cqes;
};
+
+struct qedr_alloc_pd_ureq {
+ u64 rsvd1;
+};
+
+struct qedr_alloc_pd_uresp {
+ u32 pd_id;
+};
+
+struct qedr_create_cq_ureq {
+ uint64_t addr;
+ size_t len;
+};
+
+struct qedr_create_cq_uresp {
+ u32 db_offset;
+ u16 icid;
+};
+
#endif /* __QEDR_USER_H__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index e765b50..2c26957 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -25,6 +25,17 @@
#include "verbs.h"
#include "qedr_user.h"
+#define DB_ADDR_SHIFT(addr) ((addr) << DB_PWM_ADDR_OFFSET_SHIFT)
+
+int qedr_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
+{
+ if (index > QEDR_ROCE_PKEY_TABLE_LEN)
+ return -EINVAL;
+
+ *pkey = QEDR_ROCE_PKEY_DEFAULT;
+ return 0;
+}
+
int qedr_query_gid(struct ib_device *ibdev, u8 port, int index,
union ib_gid *sgid)
{
@@ -415,3 +426,535 @@ int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_mmap return code: %d\n", rc);
return rc;
}
+
+struct ib_pd *qedr_alloc_pd(struct ib_device *ibdev,
+ struct ib_ucontext *context, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qedr_ucontext *uctx = NULL;
+ struct qedr_alloc_pd_uresp uresp;
+ struct qedr_pd *pd;
+ u16 pd_id;
+ int rc;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Function called from: %s\n",
+ (udata && context) ? "User Lib" : "Kernel");
+ if (udata && context)
+ uctx = get_qedr_ucontext(context);
+
+ pd = kzalloc(sizeof(*pd), GFP_KERNEL);
+ if (!pd) {
+ DP_ERR(dev, "failed to alloce PD\n");
+ return ERR_PTR(-ENOMEM);
+ }
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev, "invlaid RDMA context\n");
+ return ERR_PTR(-EINVAL);
+ }
+
+ dev->ops->rdma_alloc_pd(dev->rdma_ctx, &pd_id);
+
+ uresp.pd_id = pd_id;
+ pd->pd_id = pd_id;
+
+ if (uctx) {
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ DP_ERR(dev, "copy error pd_id=0x%x.\n", pd_id);
+ uctx->pd = pd;
+ pd->uctx = uctx;
+ }
+
+ return &pd->ibpd;
+}
+
+int qedr_dealloc_pd(struct ib_pd *ibpd)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+
+ if (!pd)
+ pr_err("Invalid PD received in dealloc_pd\n");
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Deallocating PD %d\n", pd->pd_id);
+ dev->ops->rdma_dealloc_pd(dev->rdma_ctx, pd->pd_id);
+
+ kfree(pd);
+
+ return 0;
+}
+
+static void qedr_free_pbl(struct qedr_dev *dev,
+ struct qedr_pbl_info *pbl_info, struct qedr_pbl *pbl)
+{
+ struct pci_dev *pdev = dev->pdev;
+ int i;
+
+ for (i = 0; i < pbl_info->num_pbls; i++) {
+ if (!pbl[i].va)
+ continue;
+ dma_free_coherent(&pdev->dev, pbl_info->pbl_size,
+ pbl[i].va, pbl[i].pa);
+ }
+
+ kfree(pbl);
+}
+
+#define MIN_FW_PBL_PAGE_SIZE (4 * 1024)
+#define MAX_FW_PBL_PAGE_SIZE (64 * 1024)
+
+#define NUM_PBES_ON_PAGE(_page_size) (_page_size / sizeof(u64))
+#define MAX_PBES_ON_PAGE NUM_PBES_ON_PAGE(MAX_FW_PBL_PAGE_SIZE)
+#define MAX_PBES_TWO_LAYER (MAX_PBES_ON_PAGE * MAX_PBES_ON_PAGE)
+
+static struct qedr_pbl *qedr_alloc_pbl_tbl(struct qedr_dev *dev,
+ struct qedr_pbl_info *pbl_info,
+ gfp_t flags)
+{
+ struct pci_dev *pdev = dev->pdev;
+ struct qedr_pbl *pbl_table;
+ dma_addr_t *pbl_main_tbl;
+ dma_addr_t pa;
+ int rc = 0;
+ void *va;
+ int i;
+
+ pbl_table = kcalloc(pbl_info->num_pbls, sizeof(*pbl_table), flags);
+
+ if (!pbl_table) {
+ DP_ERR(dev, "pbl table is NULL\n");
+ return NULL;
+ }
+
+ for (i = 0; i < pbl_info->num_pbls; i++) {
+ va = dma_alloc_coherent(&pdev->dev, pbl_info->pbl_size,
+ &pa, flags);
+ if (!va) {
+ DP_ERR(dev, "Failed to allocate pbl#%d\n", i);
+ rc = -ENOMEM;
+ goto err;
+ }
+ memset(va, 0, pbl_info->pbl_size);
+ pbl_table[i].va = va;
+ pbl_table[i].pa = pa;
+ }
+
+ /* Two-Layer PBLs, if we have more than one pbl we need to initialize
+ * the first one with physical pointers to all of the rest
+ */
+ pbl_main_tbl = (dma_addr_t *)pbl_table[0].va;
+ for (i = 0; i < pbl_info->num_pbls - 1; i++)
+ pbl_main_tbl[i] = pbl_table[i + 1].pa;
+
+ return pbl_table;
+
+err:
+ qedr_free_pbl(dev, pbl_info, pbl_table);
+ return NULL;
+}
+
+static int qedr_prepare_pbl_tbl(struct qedr_dev *dev,
+ struct qedr_pbl_info *pbl_info,
+ u32 num_pbes, int two_layer_capable)
+{
+ u32 pbl_capacity;
+ u32 pbl_size;
+ u32 num_pbls;
+
+ if ((num_pbes > MAX_PBES_ON_PAGE) && two_layer_capable) {
+ if (num_pbes > MAX_PBES_TWO_LAYER) {
+ DP_ERR(dev, "prepare pbl table: too many pages %d\n",
+ num_pbes);
+ return -EINVAL;
+ }
+
+ /* calculate required pbl page size */
+ pbl_size = MIN_FW_PBL_PAGE_SIZE;
+ pbl_capacity = NUM_PBES_ON_PAGE(pbl_size) *
+ NUM_PBES_ON_PAGE(pbl_size);
+
+ while (pbl_capacity < num_pbes) {
+ pbl_size *= 2;
+ pbl_capacity = pbl_size / sizeof(u64);
+ pbl_capacity = pbl_capacity * pbl_capacity;
+ }
+
+ num_pbls = DIV_ROUND_UP(num_pbes, NUM_PBES_ON_PAGE(pbl_size));
+ num_pbls++; /* One for the layer0 ( points to the pbls) */
+ pbl_info->two_layered = true;
+ } else {
+ /* One layered PBL */
+ num_pbls = 1;
+ pbl_size = max_t(u32, MIN_FW_PBL_PAGE_SIZE,
+ roundup_pow_of_two((num_pbes * sizeof(u64))));
+ pbl_info->two_layered = false;
+ }
+
+ pbl_info->num_pbls = num_pbls;
+ pbl_info->pbl_size = pbl_size;
+ pbl_info->num_pbes = num_pbes;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "prepare pbl table: num_pbes=%d, num_pbls=%d, pbl_size=%d\n",
+ pbl_info->num_pbes, pbl_info->num_pbls, pbl_info->pbl_size);
+
+ return 0;
+}
+
+static void qedr_populate_pbls(struct qedr_dev *dev, struct ib_umem *umem,
+ struct qedr_pbl *pbl,
+ struct qedr_pbl_info *pbl_info)
+{
+ int shift, pg_cnt, pages, pbe_cnt, total_num_pbes = 0;
+ struct qedr_pbl *pbl_tbl;
+ struct scatterlist *sg;
+ struct regpair *pbe;
+ int entry;
+ u32 addr;
+
+ if (!pbl_info->num_pbes)
+ return;
+
+ /* If we have a two layered pbl, the first pbl points to the rest
+ * of the pbls and the first entry lays on the second pbl in the table
+ */
+ if (pbl_info->two_layered)
+ pbl_tbl = &pbl[1];
+ else
+ pbl_tbl = pbl;
+
+ pbe = (struct regpair *)pbl_tbl->va;
+ if (!pbe) {
+ DP_ERR(dev, "pbe is NULL\n");
+ return;
+ }
+
+ pbe_cnt = 0;
+
+ shift = ilog2(umem->page_size);
+
+ for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+ pages = sg_dma_len(sg) >> shift;
+ for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) {
+ /* store the page address in pbe */
+ pbe->lo = cpu_to_le32(sg_dma_address(sg) +
+ umem->page_size * pg_cnt);
+ addr = upper_32_bits(sg_dma_address(sg) +
+ umem->page_size * pg_cnt);
+ pbe->hi = cpu_to_le32(addr);
+ pbe_cnt++;
+ total_num_pbes++;
+ pbe++;
+
+ if (total_num_pbes == pbl_info->num_pbes)
+ return;
+
+ /* If the given pbl is full storing the pbes,
+ * move to next pbl.
+ */
+ if (pbe_cnt == (pbl_info->pbl_size / sizeof(u64))) {
+ pbl_tbl++;
+ pbe = (struct regpair *)pbl_tbl->va;
+ pbe_cnt = 0;
+ }
+ }
+ }
+}
+
+static int qedr_copy_cq_uresp(struct qedr_dev *dev,
+ struct qedr_cq *cq, struct ib_udata *udata)
+{
+ struct qedr_create_cq_uresp uresp;
+ int rc;
+
+ memset(&uresp, 0, sizeof(uresp));
+
+ uresp.db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT);
+ uresp.icid = cq->icid;
+
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ DP_ERR(dev, "copy error cqid=0x%x.\n", cq->icid);
+
+ return rc;
+}
+
+static void consume_cqe(struct qedr_cq *cq)
+{
+ if (cq->latest_cqe == cq->toggle_cqe)
+ cq->pbl_toggle ^= RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK;
+
+ cq->latest_cqe = qed_chain_consume(&cq->pbl);
+}
+
+static inline int qedr_align_cq_entries(int entries)
+{
+ u64 size, aligned_size;
+
+ /* We allocate an extra entry that we don't report to the FW. */
+ size = (entries + 1) * QEDR_CQE_SIZE;
+ aligned_size = ALIGN(size, PAGE_SIZE);
+
+ return aligned_size / QEDR_CQE_SIZE;
+}
+
+static inline int qedr_init_user_queue(struct ib_ucontext *ib_ctx,
+ struct qedr_dev *dev,
+ struct qedr_userq *q,
+ u64 buf_addr, size_t buf_len,
+ int access, int dmasync)
+{
+ int page_cnt;
+ int rc;
+
+ q->buf_addr = buf_addr;
+ q->buf_len = buf_len;
+ q->umem = ib_umem_get(ib_ctx, q->buf_addr, q->buf_len, access, dmasync);
+ if (IS_ERR(q->umem)) {
+ DP_ERR(dev, "create user queue: failed ib_umem_get, got %ld\n",
+ PTR_ERR(q->umem));
+ return PTR_ERR(q->umem);
+ }
+
+ page_cnt = ib_umem_page_count(q->umem);
+ rc = qedr_prepare_pbl_tbl(dev, &q->pbl_info, page_cnt, 0);
+ if (rc)
+ goto err0;
+
+ q->pbl_tbl = qedr_alloc_pbl_tbl(dev, &q->pbl_info, GFP_KERNEL);
+ if (!q->pbl_tbl)
+ goto err0;
+
+ qedr_populate_pbls(dev, q->umem, q->pbl_tbl, &q->pbl_info);
+
+ return 0;
+
+err0:
+ ib_umem_release(q->umem);
+
+ return rc;
+}
+
+static inline void qedr_init_cq_params(struct qedr_cq *cq,
+ struct qedr_ucontext *ctx,
+ struct qedr_dev *dev, int vector,
+ int chain_entries, int page_cnt,
+ u64 pbl_ptr,
+ struct qed_rdma_create_cq_in_params
+ *params)
+{
+ memset(params, 0, sizeof(*params));
+ params->cq_handle_hi = upper_32_bits((uintptr_t)cq);
+ params->cq_handle_lo = lower_32_bits((uintptr_t)cq);
+ params->cnq_id = vector;
+ params->cq_size = chain_entries - 1;
+ params->dpi = (ctx) ? ctx->dpi : dev->dpi;
+ params->pbl_num_pages = page_cnt;
+ params->pbl_ptr = pbl_ptr;
+ params->pbl_two_level = 0;
+}
+
+static void doorbell_cq(struct qedr_cq *cq, u32 cons, u8 flags)
+{
+ /* Flush data before signalling doorbell */
+ wmb();
+ cq->db.data.agg_flags = flags;
+ cq->db.data.value = cpu_to_le32(cons);
+ writeq(cq->db.raw, cq->db_addr);
+
+ /* Make sure write would stick */
+ mmiowb();
+}
+
+int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
+{
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+ unsigned long sflags;
+
+ if (cq->cq_type == QEDR_CQ_TYPE_GSI)
+ return 0;
+
+ spin_lock_irqsave(&cq->cq_lock, sflags);
+
+ cq->arm_flags = 0;
+
+ if (flags & IB_CQ_SOLICITED)
+ cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_SE_CF_CMD;
+
+ if (flags & IB_CQ_NEXT_COMP)
+ cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_CF_CMD;
+
+ doorbell_cq(cq, cq->cq_cons - 1, cq->arm_flags);
+
+ spin_unlock_irqrestore(&cq->cq_lock, sflags);
+
+ return 0;
+}
+
+struct ib_cq *qedr_create_cq(struct ib_device *ibdev,
+ const struct ib_cq_init_attr *attr,
+ struct ib_ucontext *ib_ctx, struct ib_udata *udata)
+{
+ struct qedr_ucontext *ctx = get_qedr_ucontext(ib_ctx);
+ struct qed_rdma_destroy_cq_out_params destroy_oparams;
+ struct qed_rdma_destroy_cq_in_params destroy_iparams;
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qed_rdma_create_cq_in_params params;
+ struct qedr_create_cq_ureq ureq;
+ int vector = attr->comp_vector;
+ int entries = attr->cqe;
+ struct qedr_cq *cq;
+ int chain_entries;
+ int page_cnt;
+ u64 pbl_ptr;
+ u16 icid;
+ int rc;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT,
+ "create_cq: called from %s. entries=%d, vector=%d\n",
+ udata ? "User Lib" : "Kernel", entries, vector);
+
+ if (entries > QEDR_MAX_CQES) {
+ DP_ERR(dev,
+ "create cq: the number of entries %d is too high. Must be equal or below %d.\n",
+ entries, QEDR_MAX_CQES);
+ return ERR_PTR(-EINVAL);
+ }
+
+ chain_entries = qedr_align_cq_entries(entries);
+ chain_entries = min_t(int, chain_entries, QEDR_MAX_CQES);
+
+ cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+ if (!cq)
+ return ERR_PTR(-ENOMEM);
+
+ if (udata) {
+ memset(&ureq, 0, sizeof(ureq));
+ if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) {
+ DP_ERR(dev,
+ "create cq: problem copying data from user space\n");
+ goto err0;
+ }
+
+ if (!ureq.len) {
+ DP_ERR(dev,
+ "create cq: cannot create a cq with 0 entries\n");
+ goto err0;
+ }
+
+ cq->cq_type = QEDR_CQ_TYPE_USER;
+
+ rc = qedr_init_user_queue(ib_ctx, dev, &cq->q, ureq.addr, ureq.len,
+ IB_ACCESS_LOCAL_WRITE, 1);
+ if (rc)
+ goto err0;
+
+ pbl_ptr = cq->q.pbl_tbl->pa;
+ page_cnt = cq->q.pbl_info.num_pbes;
+ } else {
+ cq->cq_type = QEDR_CQ_TYPE_KERNEL;
+
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_CONSUME,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U32,
+ chain_entries,
+ sizeof(union rdma_cqe),
+ &cq->pbl);
+ if (rc)
+ goto err1;
+
+ page_cnt = qed_chain_get_page_cnt(&cq->pbl);
+ pbl_ptr = qed_chain_get_pbl_phys(&cq->pbl);
+ }
+
+ qedr_init_cq_params(cq, ctx, dev, vector, chain_entries, page_cnt,
+ pbl_ptr, ¶ms);
+
+ rc = dev->ops->rdma_create_cq(dev->rdma_ctx, ¶ms, &icid);
+ if (rc)
+ goto err2;
+
+ cq->icid = icid;
+ cq->sig = QEDR_CQ_MAGIC_NUMBER;
+ spin_lock_init(&cq->cq_lock);
+
+ if (ib_ctx) {
+ rc = qedr_copy_cq_uresp(dev, cq, udata);
+ if (rc)
+ goto err3;
+ } else {
+ /* Generate doorbell address. */
+ cq->db_addr = dev->db_addr +
+ DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT);
+ cq->db.data.icid = cq->icid;
+ cq->db.data.params = DB_AGG_CMD_SET <<
+ RDMA_PWM_VAL32_DATA_AGG_CMD_SHIFT;
+
+ /* point to the very last element, passing it we will toggle */
+ cq->toggle_cqe = qed_chain_get_last_elem(&cq->pbl);
+ cq->pbl_toggle = RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK;
+ cq->latest_cqe = NULL;
+ consume_cqe(cq);
+ cq->cq_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_CQ,
+ "create cq: icid=0x%0x, addr=%p, size(entries)=0x%0x\n",
+ cq->icid, cq, params.cq_size);
+
+ return &cq->ibcq;
+
+err3:
+ destroy_iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &destroy_iparams,
+ &destroy_oparams);
+err2:
+ if (udata)
+ qedr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl);
+ else
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+err1:
+ if (udata)
+ ib_umem_release(cq->q.umem);
+err0:
+ kfree(cq);
+ return ERR_PTR(-EINVAL);
+}
+
+int qedr_resize_cq(struct ib_cq *ibcq, int new_cnt, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+
+ DP_ERR(dev, "cq %p RESIZE NOT SUPPORTED\n", cq);
+
+ return 0;
+}
+
+int qedr_destroy_cq(struct ib_cq *ibcq)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qed_rdma_destroy_cq_out_params oparams;
+ struct qed_rdma_destroy_cq_in_params iparams;
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+
+ DP_VERBOSE(dev, QEDR_MSG_CQ, "destroy cq: cq_id %d", cq->icid);
+
+ /* GSIs CQs are handled by driver, so they don't exist in the FW */
+ if (cq->cq_type != QEDR_CQ_TYPE_GSI) {
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+ }
+
+ if (ibcq->uobject && ibcq->uobject->context) {
+ qedr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl);
+ ib_umem_release(cq->q.umem);
+ }
+
+ kfree(cq);
+
+ return 0;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index f731063..d5e1017 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -14,6 +14,8 @@ int qedr_query_port(struct ib_device *, u8 port, struct ib_port_attr *props);
int qedr_query_gid(struct ib_device *, u8 port, int index, union ib_gid *gid);
+int qedr_query_pkey(struct ib_device *, u8 port, u16 index, u16 *pkey);
+
struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *, struct ib_udata *);
int qedr_dealloc_ucontext(struct ib_ucontext *);
@@ -23,4 +25,16 @@ int qedr_del_gid(struct ib_device *device, u8 port_num,
int qedr_add_gid(struct ib_device *device, u8 port_num,
unsigned int index, const union ib_gid *gid,
const struct ib_gid_attr *attr, void **context);
+struct ib_pd *qedr_alloc_pd(struct ib_device *,
+ struct ib_ucontext *, struct ib_udata *);
+int qedr_dealloc_pd(struct ib_pd *pd);
+
+struct ib_cq *qedr_create_cq(struct ib_device *ibdev,
+ const struct ib_cq_init_attr *attr,
+ struct ib_ucontext *ib_ctx,
+ struct ib_udata *udata);
+int qedr_resize_cq(struct ib_cq *, int cqe, struct ib_udata *);
+int qedr_destroy_cq(struct ib_cq *);
+int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
+
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 9cdb472..7b489f0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -805,6 +805,315 @@ static int qed_rdma_get_int(struct qed_dev *cdev, struct qed_int_info *info)
return 0;
}
+int qed_rdma_alloc_pd(void *rdma_cxt, u16 *pd)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ u32 returned_id;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Alloc PD\n");
+
+ /* Allocates an unused protection domain */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn,
+ &p_hwfn->p_rdma_info->pd_map, &returned_id);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ *pd = (u16)returned_id;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Alloc PD - done, rc = %d\n", rc);
+ return rc;
+}
+
+void qed_rdma_free_pd(void *rdma_cxt, u16 pd)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "pd = %08x\n", pd);
+
+ /* Returns a previously allocated protection domain for reuse */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->pd_map, pd);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
+static enum qed_rdma_toggle_bit
+qed_rdma_toggle_bit_create_resize_cq(struct qed_hwfn *p_hwfn, u16 icid)
+{
+ struct qed_rdma_info *p_info = p_hwfn->p_rdma_info;
+ enum qed_rdma_toggle_bit toggle_bit;
+ u32 bmap_id;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", icid);
+
+ /* the function toggle the bit that is related to a given icid
+ * and returns the new toggle bit's value
+ */
+ bmap_id = icid - qed_cxt_get_proto_cid_start(p_hwfn, p_info->proto);
+
+ spin_lock_bh(&p_info->lock);
+ toggle_bit = !test_and_change_bit(bmap_id,
+ p_info->toggle_bits.bitmap);
+ spin_unlock_bh(&p_info->lock);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QED_RDMA_TOGGLE_BIT_= %d\n",
+ toggle_bit);
+
+ return toggle_bit;
+}
+
+int qed_rdma_create_cq(void *rdma_cxt,
+ struct qed_rdma_create_cq_in_params *params, u16 *icid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_rdma_info *p_info = p_hwfn->p_rdma_info;
+ struct rdma_create_cq_ramrod_data *p_ramrod;
+ enum qed_rdma_toggle_bit toggle_bit;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ u32 returned_id, start_cid;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "cq_handle = %08x%08x\n",
+ params->cq_handle_hi, params->cq_handle_lo);
+
+ /* Allocate icid */
+ spin_lock_bh(&p_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn,
+ &p_info->cq_map, &returned_id);
+ spin_unlock_bh(&p_info->lock);
+
+ if (rc) {
+ DP_NOTICE(p_hwfn, "Can't create CQ, rc = %d\n", rc);
+ return rc;
+ }
+
+ start_cid = qed_cxt_get_proto_cid_start(p_hwfn,
+ p_info->proto);
+ *icid = returned_id + start_cid;
+
+ /* Check if icid requires a page allocation */
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, *icid);
+ if (rc)
+ goto err;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = *icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ /* Send create CQ ramrod */
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_CREATE_CQ,
+ p_info->proto, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.rdma_create_cq;
+
+ p_ramrod->cq_handle.hi = cpu_to_le32(params->cq_handle_hi);
+ p_ramrod->cq_handle.lo = cpu_to_le32(params->cq_handle_lo);
+ p_ramrod->dpi = cpu_to_le16(params->dpi);
+ p_ramrod->is_two_level_pbl = params->pbl_two_level;
+ p_ramrod->max_cqes = cpu_to_le32(params->cq_size);
+ p_ramrod->pbl_addr.hi = DMA_HI_LE(params->pbl_ptr);
+ p_ramrod->pbl_addr.lo = DMA_LO_LE(params->pbl_ptr);
+ p_ramrod->pbl_num_pages = cpu_to_le16(params->pbl_num_pages);
+ p_ramrod->cnq_id = (u8)RESC_START(p_hwfn, QED_RDMA_CNQ_RAM) +
+ params->cnq_id;
+ p_ramrod->int_timeout = params->int_timeout;
+
+ /* toggle the bit for every resize or create cq for a given icid */
+ toggle_bit = qed_rdma_toggle_bit_create_resize_cq(p_hwfn, *icid);
+
+ p_ramrod->toggle_bit = toggle_bit;
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc) {
+ /* restore toggle bit */
+ qed_rdma_toggle_bit_create_resize_cq(p_hwfn, *icid);
+ goto err;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Created CQ, rc = %d\n", rc);
+ return rc;
+
+err:
+ /* release allocated icid */
+ qed_bmap_release_id(p_hwfn, &p_info->cq_map, returned_id);
+ DP_NOTICE(p_hwfn, "Create CQ failed, rc = %d\n", rc);
+
+ return rc;
+}
+
+int qed_rdma_resize_cq(void *rdma_cxt,
+ struct qed_rdma_resize_cq_in_params *in_params,
+ struct qed_rdma_resize_cq_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_resize_cq_output_params *p_ramrod_res;
+ struct rdma_resize_cq_ramrod_data *p_ramrod;
+ enum qed_rdma_toggle_bit toggle_bit;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ u8 fw_return_code;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", in_params->icid);
+
+ p_ramrod_res =
+ (struct rdma_resize_cq_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_resize_cq_output_params),
+ &ramrod_res_phys, GFP_KERNEL);
+ if (!p_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed resize cq failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = in_params->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_RESIZE_CQ,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.rdma_resize_cq;
+
+ p_ramrod->flags = 0;
+
+ /* toggle the bit for every resize or create cq for a given icid */
+ toggle_bit = qed_rdma_toggle_bit_create_resize_cq(p_hwfn,
+ in_params->icid);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_RESIZE_CQ_RAMROD_DATA_TOGGLE_BIT, toggle_bit);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_RESIZE_CQ_RAMROD_DATA_IS_TWO_LEVEL_PBL,
+ in_params->pbl_two_level);
+
+ p_ramrod->pbl_log_page_size = in_params->pbl_page_size_log - 12;
+ p_ramrod->pbl_num_pages = cpu_to_le16(in_params->pbl_num_pages);
+ p_ramrod->max_cqes = cpu_to_le32(in_params->cq_size);
+ p_ramrod->pbl_addr.hi = DMA_HI_LE(in_params->pbl_ptr);
+ p_ramrod->pbl_addr.lo = DMA_LO_LE(in_params->pbl_ptr);
+
+ p_ramrod->output_params_addr.hi = DMA_HI_LE(ramrod_res_phys);
+ p_ramrod->output_params_addr.lo = DMA_LO_LE(ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc)
+ goto err;
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ out_params->prod = le32_to_cpu(p_ramrod_res->old_cq_prod);
+ out_params->cons = le32_to_cpu(p_ramrod_res->old_cq_cons);
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_resize_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Resized CQ, rc = %d\n", rc);
+
+ return rc;
+
+err: dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_resize_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+ DP_NOTICE(p_hwfn, "Resized CQ, Failed - rc = %d\n", rc);
+
+ return rc;
+}
+
+int qed_rdma_destroy_cq(void *rdma_cxt,
+ struct qed_rdma_destroy_cq_in_params *in_params,
+ struct qed_rdma_destroy_cq_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_destroy_cq_output_params *p_ramrod_res;
+ struct rdma_destroy_cq_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", in_params->icid);
+
+ p_ramrod_res =
+ (struct rdma_destroy_cq_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_destroy_cq_output_params),
+ &ramrod_res_phys, GFP_KERNEL);
+ if (!p_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed destroy cq failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = in_params->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ /* Send destroy CQ ramrod */
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_DESTROY_CQ,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.rdma_destroy_cq;
+ p_ramrod->output_params_addr.hi = DMA_HI_LE(ramrod_res_phys);
+ p_ramrod->output_params_addr.lo = DMA_LO_LE(ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err;
+
+ out_params->num_cq_notif = le16_to_cpu(p_ramrod_res->cnq_num);
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_destroy_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ /* Free icid */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+
+ qed_bmap_release_id(p_hwfn,
+ &p_hwfn->p_rdma_info->cq_map,
+ (in_params->icid -
+ qed_cxt_get_proto_cid_start(p_hwfn,
+ p_hwfn->
+ p_rdma_info->proto)));
+
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroyed CQ, rc = %d\n", rc);
+ return rc;
+
+err: dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_destroy_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ return rc;
+}
+
static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
{
return QED_LEADING_HWFN(cdev);
@@ -895,6 +1204,10 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_set_rdma_int = &qed_rdma_set_int,
.rdma_get_min_cnq_msix = &qed_rdma_get_min_cnq_msix,
.rdma_cnq_prod_update = &qed_rdma_cnq_prod_update,
+ .rdma_alloc_pd = &qed_rdma_alloc_pd,
+ .rdma_dealloc_pd = &qed_rdma_free_pd,
+ .rdma_create_cq = &qed_rdma_create_cq,
+ .rdma_destroy_cq = &qed_rdma_destroy_cq,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index c94c39e..6802cd0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -94,6 +94,26 @@ struct qed_rdma_info {
enum protocol_type proto;
};
+struct qed_rdma_resize_cq_in_params {
+ u16 icid;
+ u32 cq_size;
+ bool pbl_two_level;
+ u64 pbl_ptr;
+ u16 pbl_num_pages;
+ u8 pbl_page_size_log;
+};
+
+struct qed_rdma_resize_cq_out_params {
+ u32 prod;
+ u32 cons;
+};
+
+struct qed_rdma_resize_cnq_in_params {
+ u32 cnq_id;
+ u32 pbl_page_size_log;
+ u64 pbl_ptr;
+};
+
int
qed_rdma_add_user(void *rdma_cxt,
struct qed_rdma_add_user_out_params *out_params);
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index e99470c..b559b1c 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -263,6 +263,19 @@ struct qed_rdma_register_tid_in_params {
u64 dif_runt_addr;
};
+struct qed_rdma_create_cq_in_params {
+ u32 cq_handle_lo;
+ u32 cq_handle_hi;
+ u32 cq_size;
+ u16 dpi;
+ bool pbl_two_level;
+ u64 pbl_ptr;
+ u16 pbl_num_pages;
+ u8 pbl_page_size_log;
+ u8 cnq_id;
+ u16 int_timeout;
+};
+
struct qed_rdma_create_srq_in_params {
u64 pbl_base_addr;
u64 prod_pair_addr;
@@ -271,6 +284,14 @@ struct qed_rdma_create_srq_in_params {
u16 page_size;
};
+struct qed_rdma_destroy_cq_in_params {
+ u16 icid;
+};
+
+struct qed_rdma_destroy_cq_out_params {
+ u16 num_cq_notif;
+};
+
struct qed_rdma_create_srq_out_params {
u16 srq_id;
};
@@ -339,6 +360,14 @@ struct qed_rdma_ops {
int (*rdma_get_rdma_int)(struct qed_dev *cdev,
struct qed_int_info *info);
int (*rdma_set_rdma_int)(struct qed_dev *cdev, u16 cnt);
+ int (*rdma_alloc_pd)(void *rdma_cxt, u16 *pd);
+ void (*rdma_dealloc_pd)(void *rdma_cxt, u16 pd);
+ int (*rdma_create_cq)(void *rdma_cxt,
+ struct qed_rdma_create_cq_in_params *params,
+ u16 *icid);
+ int (*rdma_destroy_cq)(void *rdma_cxt,
+ struct qed_rdma_destroy_cq_in_params *iparams,
+ struct qed_rdma_destroy_cq_out_params *oparams);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 05/11] Add support for PD,PKEY and CQ verbs
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for protection domain and completion queue verbs.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 18 +-
drivers/infiniband/hw/qedr/qedr.h | 78 +++++
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 79 +++++
drivers/infiniband/hw/qedr/qedr_user.h | 19 +
drivers/infiniband/hw/qedr/verbs.c | 543 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 14 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 313 +++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 20 ++
include/linux/qed/qed_roce_if.h | 29 ++
9 files changed, 1112 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index a01983c..02f0dbb 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -94,7 +94,13 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.uverbs_cmd_mask = QEDR_UVERBS(GET_CONTEXT) |
QEDR_UVERBS(QUERY_DEVICE) |
- QEDR_UVERBS(QUERY_PORT);
+ QEDR_UVERBS(QUERY_PORT) |
+ QEDR_UVERBS(ALLOC_PD) |
+ QEDR_UVERBS(DEALLOC_PD) |
+ QEDR_UVERBS(CREATE_CQ) |
+ QEDR_UVERBS(RESIZE_CQ) |
+ QEDR_UVERBS(DESTROY_CQ) |
+ QEDR_UVERBS(REQ_NOTIFY_CQ);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -111,6 +117,16 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.dealloc_ucontext = qedr_dealloc_ucontext;
dev->ibdev.mmap = qedr_mmap;
+ dev->ibdev.alloc_pd = qedr_alloc_pd;
+ dev->ibdev.dealloc_pd = qedr_dealloc_pd;
+
+ dev->ibdev.create_cq = qedr_create_cq;
+ dev->ibdev.destroy_cq = qedr_destroy_cq;
+ dev->ibdev.resize_cq = qedr_resize_cq;
+ dev->ibdev.req_notify_cq = qedr_arm_cq;
+
+ dev->ibdev.query_pkey = qedr_query_pkey;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index ed3ef33..b2aa8fd 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -44,8 +44,12 @@
#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
#define DP_NAME(dev) ((dev)->ibdev.name)
+#define QEDR_CQ_MAGIC_NUMBER (0x11223344)
+
enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
+ QEDR_MSG_CQ = 0x20000,
+ QEDR_MSG_MR = 0x100000,
QEDR_MSG_MISC = 0x400000,
};
@@ -180,6 +184,12 @@ struct qedr_dev {
#define QEDR_ROCE_PKEY_TABLE_LEN 1
#define QEDR_ROCE_PKEY_DEFAULT 0xffff
+struct qedr_pbl {
+ struct list_head list_entry;
+ void *va;
+ dma_addr_t pa;
+};
+
struct qedr_ucontext {
struct ib_ucontext ibucontext;
struct qedr_dev *dev;
@@ -195,6 +205,64 @@ struct qedr_ucontext {
struct mutex mm_list_lock;
};
+union db_prod64 {
+ struct rdma_pwm_val32_data data;
+ u64 raw;
+};
+
+enum qedr_cq_type {
+ QEDR_CQ_TYPE_GSI,
+ QEDR_CQ_TYPE_KERNEL,
+ QEDR_CQ_TYPE_USER
+};
+
+struct qedr_pbl_info {
+ u32 num_pbls;
+ u32 num_pbes;
+ u32 pbl_size;
+ u32 pbe_size;
+ bool two_layered;
+};
+
+struct qedr_userq {
+ struct ib_umem *umem;
+ struct qedr_pbl_info pbl_info;
+ struct qedr_pbl *pbl_tbl;
+ u64 buf_addr;
+ size_t buf_len;
+};
+
+struct qedr_cq {
+ struct ib_cq ibcq;
+
+ enum qedr_cq_type cq_type;
+ u32 sig;
+
+ u16 icid;
+
+ /* Lock to protect multiplem CQ's */
+ spinlock_t cq_lock;
+ u8 arm_flags;
+ struct qed_chain pbl;
+
+ void __iomem *db_addr;
+ union db_prod64 db;
+
+ u8 pbl_toggle;
+ union rdma_cqe *latest_cqe;
+ union rdma_cqe *toggle_cqe;
+
+ u32 cq_cons;
+
+ struct qedr_userq q;
+};
+
+struct qedr_pd {
+ struct ib_pd ibpd;
+ u32 pd_id;
+ struct qedr_ucontext *uctx;
+};
+
struct qedr_mm {
struct {
u64 phy_addr;
@@ -213,4 +281,14 @@ struct qedr_ucontext *get_qedr_ucontext(struct ib_ucontext *ibucontext)
{
return container_of(ibucontext, struct qedr_ucontext, ibucontext);
}
+
+static inline struct qedr_pd *get_qedr_pd(struct ib_pd *ibpd)
+{
+ return container_of(ibpd, struct qedr_pd, ibpd);
+}
+
+static inline struct qedr_cq *get_qedr_cq(struct ib_cq *ibcq)
+{
+ return container_of(ibcq, struct qedr_cq, ibcq);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
index 3e508fb..84f6520 100644
--- a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -47,6 +47,19 @@ struct rdma_cqe_responder {
__le32 imm_data_hi;
__le16 rq_cons;
u8 flags;
+#define RDMA_CQE_RESPONDER_TOGGLE_BIT_MASK 0x1
+#define RDMA_CQE_RESPONDER_TOGGLE_BIT_SHIFT 0
+#define RDMA_CQE_RESPONDER_TYPE_MASK 0x3
+#define RDMA_CQE_RESPONDER_TYPE_SHIFT 1
+#define RDMA_CQE_RESPONDER_INV_FLG_MASK 0x1
+#define RDMA_CQE_RESPONDER_INV_FLG_SHIFT 3
+#define RDMA_CQE_RESPONDER_IMM_FLG_MASK 0x1
+#define RDMA_CQE_RESPONDER_IMM_FLG_SHIFT 4
+#define RDMA_CQE_RESPONDER_RDMA_FLG_MASK 0x1
+#define RDMA_CQE_RESPONDER_RDMA_FLG_SHIFT 5
+#define RDMA_CQE_RESPONDER_RESERVED2_MASK 0x3
+#define RDMA_CQE_RESPONDER_RESERVED2_SHIFT 6
+ u8 status;
};
struct rdma_cqe_requester {
@@ -58,6 +71,12 @@ struct rdma_cqe_requester {
__le32 reserved3;
__le16 reserved4;
u8 flags;
+#define RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK 0x1
+#define RDMA_CQE_REQUESTER_TOGGLE_BIT_SHIFT 0
+#define RDMA_CQE_REQUESTER_TYPE_MASK 0x3
+#define RDMA_CQE_REQUESTER_TYPE_SHIFT 1
+#define RDMA_CQE_REQUESTER_RESERVED5_MASK 0x1F
+#define RDMA_CQE_REQUESTER_RESERVED5_SHIFT 3
u8 status;
};
@@ -66,6 +85,12 @@ struct rdma_cqe_common {
struct regpair qp_handle;
__le16 reserved1[7];
u8 flags;
+#define RDMA_CQE_COMMON_TOGGLE_BIT_MASK 0x1
+#define RDMA_CQE_COMMON_TOGGLE_BIT_SHIFT 0
+#define RDMA_CQE_COMMON_TYPE_MASK 0x3
+#define RDMA_CQE_COMMON_TYPE_SHIFT 1
+#define RDMA_CQE_COMMON_RESERVED2_MASK 0x1F
+#define RDMA_CQE_COMMON_RESERVED2_SHIFT 3
u8 status;
};
@@ -76,6 +101,45 @@ union rdma_cqe {
struct rdma_cqe_common cmn;
};
+/* * CQE requester status enumeration */
+enum rdma_cqe_requester_status_enum {
+ RDMA_CQE_REQ_STS_OK,
+ RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR,
+ RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR,
+ RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR,
+ RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR,
+ RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR,
+ RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR,
+ RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR,
+ RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR,
+ RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR,
+ RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR,
+ RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR,
+ MAX_RDMA_CQE_REQUESTER_STATUS_ENUM
+};
+
+/* CQE responder status enumeration */
+enum rdma_cqe_responder_status_enum {
+ RDMA_CQE_RESP_STS_OK,
+ RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR,
+ RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR,
+ RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR,
+ RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR,
+ RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR,
+ RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR,
+ RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR,
+ MAX_RDMA_CQE_RESPONDER_STATUS_ENUM
+};
+
+/* CQE type enumeration */
+enum rdma_cqe_type {
+ RDMA_CQE_TYPE_REQUESTER,
+ RDMA_CQE_TYPE_RESPONDER_RQ,
+ RDMA_CQE_TYPE_RESPONDER_SRQ,
+ RDMA_CQE_TYPE_INVALID,
+ MAX_RDMA_CQE_TYPE
+};
+
struct rdma_sq_sge {
__le32 length;
struct regpair addr;
@@ -93,4 +157,19 @@ struct rdma_srq_sge {
__le32 length;
__le32 l_key;
};
+
+/* Rdma doorbell data for CQ */
+struct rdma_pwm_val32_data {
+ __le16 icid;
+ u8 agg_flags;
+ u8 params;
+#define RDMA_PWM_VAL32_DATA_AGG_CMD_MASK 0x3
+#define RDMA_PWM_VAL32_DATA_AGG_CMD_SHIFT 0
+#define RDMA_PWM_VAL32_DATA_BYPASS_EN_MASK 0x1
+#define RDMA_PWM_VAL32_DATA_BYPASS_EN_SHIFT 2
+#define RDMA_PWM_VAL32_DATA_RESERVED_MASK 0x1F
+#define RDMA_PWM_VAL32_DATA_RESERVED_SHIFT 3
+ __le32 value;
+};
+
#endif /* __QED_HSI_RDMA__ */
diff --git a/drivers/infiniband/hw/qedr/qedr_user.h b/drivers/infiniband/hw/qedr/qedr_user.h
index 45288dc..5832c30 100644
--- a/drivers/infiniband/hw/qedr/qedr_user.h
+++ b/drivers/infiniband/hw/qedr/qedr_user.h
@@ -24,4 +24,23 @@ struct qedr_alloc_ucontext_resp {
u32 sges_per_srq_wr;
int max_cqes;
};
+
+struct qedr_alloc_pd_ureq {
+ u64 rsvd1;
+};
+
+struct qedr_alloc_pd_uresp {
+ u32 pd_id;
+};
+
+struct qedr_create_cq_ureq {
+ uint64_t addr;
+ size_t len;
+};
+
+struct qedr_create_cq_uresp {
+ u32 db_offset;
+ u16 icid;
+};
+
#endif /* __QEDR_USER_H__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index e765b50..2c26957 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -25,6 +25,17 @@
#include "verbs.h"
#include "qedr_user.h"
+#define DB_ADDR_SHIFT(addr) ((addr) << DB_PWM_ADDR_OFFSET_SHIFT)
+
+int qedr_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
+{
+ if (index > QEDR_ROCE_PKEY_TABLE_LEN)
+ return -EINVAL;
+
+ *pkey = QEDR_ROCE_PKEY_DEFAULT;
+ return 0;
+}
+
int qedr_query_gid(struct ib_device *ibdev, u8 port, int index,
union ib_gid *sgid)
{
@@ -415,3 +426,535 @@ int qedr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr_mmap return code: %d\n", rc);
return rc;
}
+
+struct ib_pd *qedr_alloc_pd(struct ib_device *ibdev,
+ struct ib_ucontext *context, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qedr_ucontext *uctx = NULL;
+ struct qedr_alloc_pd_uresp uresp;
+ struct qedr_pd *pd;
+ u16 pd_id;
+ int rc;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Function called from: %s\n",
+ (udata && context) ? "User Lib" : "Kernel");
+ if (udata && context)
+ uctx = get_qedr_ucontext(context);
+
+ pd = kzalloc(sizeof(*pd), GFP_KERNEL);
+ if (!pd) {
+ DP_ERR(dev, "failed to alloce PD\n");
+ return ERR_PTR(-ENOMEM);
+ }
+
+ if (!dev->rdma_ctx) {
+ DP_ERR(dev, "invlaid RDMA context\n");
+ return ERR_PTR(-EINVAL);
+ }
+
+ dev->ops->rdma_alloc_pd(dev->rdma_ctx, &pd_id);
+
+ uresp.pd_id = pd_id;
+ pd->pd_id = pd_id;
+
+ if (uctx) {
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ DP_ERR(dev, "copy error pd_id=0x%x.\n", pd_id);
+ uctx->pd = pd;
+ pd->uctx = uctx;
+ }
+
+ return &pd->ibpd;
+}
+
+int qedr_dealloc_pd(struct ib_pd *ibpd)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+
+ if (!pd)
+ pr_err("Invalid PD received in dealloc_pd\n");
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT, "Deallocating PD %d\n", pd->pd_id);
+ dev->ops->rdma_dealloc_pd(dev->rdma_ctx, pd->pd_id);
+
+ kfree(pd);
+
+ return 0;
+}
+
+static void qedr_free_pbl(struct qedr_dev *dev,
+ struct qedr_pbl_info *pbl_info, struct qedr_pbl *pbl)
+{
+ struct pci_dev *pdev = dev->pdev;
+ int i;
+
+ for (i = 0; i < pbl_info->num_pbls; i++) {
+ if (!pbl[i].va)
+ continue;
+ dma_free_coherent(&pdev->dev, pbl_info->pbl_size,
+ pbl[i].va, pbl[i].pa);
+ }
+
+ kfree(pbl);
+}
+
+#define MIN_FW_PBL_PAGE_SIZE (4 * 1024)
+#define MAX_FW_PBL_PAGE_SIZE (64 * 1024)
+
+#define NUM_PBES_ON_PAGE(_page_size) (_page_size / sizeof(u64))
+#define MAX_PBES_ON_PAGE NUM_PBES_ON_PAGE(MAX_FW_PBL_PAGE_SIZE)
+#define MAX_PBES_TWO_LAYER (MAX_PBES_ON_PAGE * MAX_PBES_ON_PAGE)
+
+static struct qedr_pbl *qedr_alloc_pbl_tbl(struct qedr_dev *dev,
+ struct qedr_pbl_info *pbl_info,
+ gfp_t flags)
+{
+ struct pci_dev *pdev = dev->pdev;
+ struct qedr_pbl *pbl_table;
+ dma_addr_t *pbl_main_tbl;
+ dma_addr_t pa;
+ int rc = 0;
+ void *va;
+ int i;
+
+ pbl_table = kcalloc(pbl_info->num_pbls, sizeof(*pbl_table), flags);
+
+ if (!pbl_table) {
+ DP_ERR(dev, "pbl table is NULL\n");
+ return NULL;
+ }
+
+ for (i = 0; i < pbl_info->num_pbls; i++) {
+ va = dma_alloc_coherent(&pdev->dev, pbl_info->pbl_size,
+ &pa, flags);
+ if (!va) {
+ DP_ERR(dev, "Failed to allocate pbl#%d\n", i);
+ rc = -ENOMEM;
+ goto err;
+ }
+ memset(va, 0, pbl_info->pbl_size);
+ pbl_table[i].va = va;
+ pbl_table[i].pa = pa;
+ }
+
+ /* Two-Layer PBLs, if we have more than one pbl we need to initialize
+ * the first one with physical pointers to all of the rest
+ */
+ pbl_main_tbl = (dma_addr_t *)pbl_table[0].va;
+ for (i = 0; i < pbl_info->num_pbls - 1; i++)
+ pbl_main_tbl[i] = pbl_table[i + 1].pa;
+
+ return pbl_table;
+
+err:
+ qedr_free_pbl(dev, pbl_info, pbl_table);
+ return NULL;
+}
+
+static int qedr_prepare_pbl_tbl(struct qedr_dev *dev,
+ struct qedr_pbl_info *pbl_info,
+ u32 num_pbes, int two_layer_capable)
+{
+ u32 pbl_capacity;
+ u32 pbl_size;
+ u32 num_pbls;
+
+ if ((num_pbes > MAX_PBES_ON_PAGE) && two_layer_capable) {
+ if (num_pbes > MAX_PBES_TWO_LAYER) {
+ DP_ERR(dev, "prepare pbl table: too many pages %d\n",
+ num_pbes);
+ return -EINVAL;
+ }
+
+ /* calculate required pbl page size */
+ pbl_size = MIN_FW_PBL_PAGE_SIZE;
+ pbl_capacity = NUM_PBES_ON_PAGE(pbl_size) *
+ NUM_PBES_ON_PAGE(pbl_size);
+
+ while (pbl_capacity < num_pbes) {
+ pbl_size *= 2;
+ pbl_capacity = pbl_size / sizeof(u64);
+ pbl_capacity = pbl_capacity * pbl_capacity;
+ }
+
+ num_pbls = DIV_ROUND_UP(num_pbes, NUM_PBES_ON_PAGE(pbl_size));
+ num_pbls++; /* One for the layer0 ( points to the pbls) */
+ pbl_info->two_layered = true;
+ } else {
+ /* One layered PBL */
+ num_pbls = 1;
+ pbl_size = max_t(u32, MIN_FW_PBL_PAGE_SIZE,
+ roundup_pow_of_two((num_pbes * sizeof(u64))));
+ pbl_info->two_layered = false;
+ }
+
+ pbl_info->num_pbls = num_pbls;
+ pbl_info->pbl_size = pbl_size;
+ pbl_info->num_pbes = num_pbes;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "prepare pbl table: num_pbes=%d, num_pbls=%d, pbl_size=%d\n",
+ pbl_info->num_pbes, pbl_info->num_pbls, pbl_info->pbl_size);
+
+ return 0;
+}
+
+static void qedr_populate_pbls(struct qedr_dev *dev, struct ib_umem *umem,
+ struct qedr_pbl *pbl,
+ struct qedr_pbl_info *pbl_info)
+{
+ int shift, pg_cnt, pages, pbe_cnt, total_num_pbes = 0;
+ struct qedr_pbl *pbl_tbl;
+ struct scatterlist *sg;
+ struct regpair *pbe;
+ int entry;
+ u32 addr;
+
+ if (!pbl_info->num_pbes)
+ return;
+
+ /* If we have a two layered pbl, the first pbl points to the rest
+ * of the pbls and the first entry lays on the second pbl in the table
+ */
+ if (pbl_info->two_layered)
+ pbl_tbl = &pbl[1];
+ else
+ pbl_tbl = pbl;
+
+ pbe = (struct regpair *)pbl_tbl->va;
+ if (!pbe) {
+ DP_ERR(dev, "pbe is NULL\n");
+ return;
+ }
+
+ pbe_cnt = 0;
+
+ shift = ilog2(umem->page_size);
+
+ for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+ pages = sg_dma_len(sg) >> shift;
+ for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) {
+ /* store the page address in pbe */
+ pbe->lo = cpu_to_le32(sg_dma_address(sg) +
+ umem->page_size * pg_cnt);
+ addr = upper_32_bits(sg_dma_address(sg) +
+ umem->page_size * pg_cnt);
+ pbe->hi = cpu_to_le32(addr);
+ pbe_cnt++;
+ total_num_pbes++;
+ pbe++;
+
+ if (total_num_pbes == pbl_info->num_pbes)
+ return;
+
+ /* If the given pbl is full storing the pbes,
+ * move to next pbl.
+ */
+ if (pbe_cnt == (pbl_info->pbl_size / sizeof(u64))) {
+ pbl_tbl++;
+ pbe = (struct regpair *)pbl_tbl->va;
+ pbe_cnt = 0;
+ }
+ }
+ }
+}
+
+static int qedr_copy_cq_uresp(struct qedr_dev *dev,
+ struct qedr_cq *cq, struct ib_udata *udata)
+{
+ struct qedr_create_cq_uresp uresp;
+ int rc;
+
+ memset(&uresp, 0, sizeof(uresp));
+
+ uresp.db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT);
+ uresp.icid = cq->icid;
+
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ DP_ERR(dev, "copy error cqid=0x%x.\n", cq->icid);
+
+ return rc;
+}
+
+static void consume_cqe(struct qedr_cq *cq)
+{
+ if (cq->latest_cqe == cq->toggle_cqe)
+ cq->pbl_toggle ^= RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK;
+
+ cq->latest_cqe = qed_chain_consume(&cq->pbl);
+}
+
+static inline int qedr_align_cq_entries(int entries)
+{
+ u64 size, aligned_size;
+
+ /* We allocate an extra entry that we don't report to the FW. */
+ size = (entries + 1) * QEDR_CQE_SIZE;
+ aligned_size = ALIGN(size, PAGE_SIZE);
+
+ return aligned_size / QEDR_CQE_SIZE;
+}
+
+static inline int qedr_init_user_queue(struct ib_ucontext *ib_ctx,
+ struct qedr_dev *dev,
+ struct qedr_userq *q,
+ u64 buf_addr, size_t buf_len,
+ int access, int dmasync)
+{
+ int page_cnt;
+ int rc;
+
+ q->buf_addr = buf_addr;
+ q->buf_len = buf_len;
+ q->umem = ib_umem_get(ib_ctx, q->buf_addr, q->buf_len, access, dmasync);
+ if (IS_ERR(q->umem)) {
+ DP_ERR(dev, "create user queue: failed ib_umem_get, got %ld\n",
+ PTR_ERR(q->umem));
+ return PTR_ERR(q->umem);
+ }
+
+ page_cnt = ib_umem_page_count(q->umem);
+ rc = qedr_prepare_pbl_tbl(dev, &q->pbl_info, page_cnt, 0);
+ if (rc)
+ goto err0;
+
+ q->pbl_tbl = qedr_alloc_pbl_tbl(dev, &q->pbl_info, GFP_KERNEL);
+ if (!q->pbl_tbl)
+ goto err0;
+
+ qedr_populate_pbls(dev, q->umem, q->pbl_tbl, &q->pbl_info);
+
+ return 0;
+
+err0:
+ ib_umem_release(q->umem);
+
+ return rc;
+}
+
+static inline void qedr_init_cq_params(struct qedr_cq *cq,
+ struct qedr_ucontext *ctx,
+ struct qedr_dev *dev, int vector,
+ int chain_entries, int page_cnt,
+ u64 pbl_ptr,
+ struct qed_rdma_create_cq_in_params
+ *params)
+{
+ memset(params, 0, sizeof(*params));
+ params->cq_handle_hi = upper_32_bits((uintptr_t)cq);
+ params->cq_handle_lo = lower_32_bits((uintptr_t)cq);
+ params->cnq_id = vector;
+ params->cq_size = chain_entries - 1;
+ params->dpi = (ctx) ? ctx->dpi : dev->dpi;
+ params->pbl_num_pages = page_cnt;
+ params->pbl_ptr = pbl_ptr;
+ params->pbl_two_level = 0;
+}
+
+static void doorbell_cq(struct qedr_cq *cq, u32 cons, u8 flags)
+{
+ /* Flush data before signalling doorbell */
+ wmb();
+ cq->db.data.agg_flags = flags;
+ cq->db.data.value = cpu_to_le32(cons);
+ writeq(cq->db.raw, cq->db_addr);
+
+ /* Make sure write would stick */
+ mmiowb();
+}
+
+int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
+{
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+ unsigned long sflags;
+
+ if (cq->cq_type == QEDR_CQ_TYPE_GSI)
+ return 0;
+
+ spin_lock_irqsave(&cq->cq_lock, sflags);
+
+ cq->arm_flags = 0;
+
+ if (flags & IB_CQ_SOLICITED)
+ cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_SE_CF_CMD;
+
+ if (flags & IB_CQ_NEXT_COMP)
+ cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_CF_CMD;
+
+ doorbell_cq(cq, cq->cq_cons - 1, cq->arm_flags);
+
+ spin_unlock_irqrestore(&cq->cq_lock, sflags);
+
+ return 0;
+}
+
+struct ib_cq *qedr_create_cq(struct ib_device *ibdev,
+ const struct ib_cq_init_attr *attr,
+ struct ib_ucontext *ib_ctx, struct ib_udata *udata)
+{
+ struct qedr_ucontext *ctx = get_qedr_ucontext(ib_ctx);
+ struct qed_rdma_destroy_cq_out_params destroy_oparams;
+ struct qed_rdma_destroy_cq_in_params destroy_iparams;
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+ struct qed_rdma_create_cq_in_params params;
+ struct qedr_create_cq_ureq ureq;
+ int vector = attr->comp_vector;
+ int entries = attr->cqe;
+ struct qedr_cq *cq;
+ int chain_entries;
+ int page_cnt;
+ u64 pbl_ptr;
+ u16 icid;
+ int rc;
+
+ DP_VERBOSE(dev, QEDR_MSG_INIT,
+ "create_cq: called from %s. entries=%d, vector=%d\n",
+ udata ? "User Lib" : "Kernel", entries, vector);
+
+ if (entries > QEDR_MAX_CQES) {
+ DP_ERR(dev,
+ "create cq: the number of entries %d is too high. Must be equal or below %d.\n",
+ entries, QEDR_MAX_CQES);
+ return ERR_PTR(-EINVAL);
+ }
+
+ chain_entries = qedr_align_cq_entries(entries);
+ chain_entries = min_t(int, chain_entries, QEDR_MAX_CQES);
+
+ cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+ if (!cq)
+ return ERR_PTR(-ENOMEM);
+
+ if (udata) {
+ memset(&ureq, 0, sizeof(ureq));
+ if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) {
+ DP_ERR(dev,
+ "create cq: problem copying data from user space\n");
+ goto err0;
+ }
+
+ if (!ureq.len) {
+ DP_ERR(dev,
+ "create cq: cannot create a cq with 0 entries\n");
+ goto err0;
+ }
+
+ cq->cq_type = QEDR_CQ_TYPE_USER;
+
+ rc = qedr_init_user_queue(ib_ctx, dev, &cq->q, ureq.addr, ureq.len,
+ IB_ACCESS_LOCAL_WRITE, 1);
+ if (rc)
+ goto err0;
+
+ pbl_ptr = cq->q.pbl_tbl->pa;
+ page_cnt = cq->q.pbl_info.num_pbes;
+ } else {
+ cq->cq_type = QEDR_CQ_TYPE_KERNEL;
+
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_CONSUME,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U32,
+ chain_entries,
+ sizeof(union rdma_cqe),
+ &cq->pbl);
+ if (rc)
+ goto err1;
+
+ page_cnt = qed_chain_get_page_cnt(&cq->pbl);
+ pbl_ptr = qed_chain_get_pbl_phys(&cq->pbl);
+ }
+
+ qedr_init_cq_params(cq, ctx, dev, vector, chain_entries, page_cnt,
+ pbl_ptr, ¶ms);
+
+ rc = dev->ops->rdma_create_cq(dev->rdma_ctx, ¶ms, &icid);
+ if (rc)
+ goto err2;
+
+ cq->icid = icid;
+ cq->sig = QEDR_CQ_MAGIC_NUMBER;
+ spin_lock_init(&cq->cq_lock);
+
+ if (ib_ctx) {
+ rc = qedr_copy_cq_uresp(dev, cq, udata);
+ if (rc)
+ goto err3;
+ } else {
+ /* Generate doorbell address. */
+ cq->db_addr = dev->db_addr +
+ DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT);
+ cq->db.data.icid = cq->icid;
+ cq->db.data.params = DB_AGG_CMD_SET <<
+ RDMA_PWM_VAL32_DATA_AGG_CMD_SHIFT;
+
+ /* point to the very last element, passing it we will toggle */
+ cq->toggle_cqe = qed_chain_get_last_elem(&cq->pbl);
+ cq->pbl_toggle = RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK;
+ cq->latest_cqe = NULL;
+ consume_cqe(cq);
+ cq->cq_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_CQ,
+ "create cq: icid=0x%0x, addr=%p, size(entries)=0x%0x\n",
+ cq->icid, cq, params.cq_size);
+
+ return &cq->ibcq;
+
+err3:
+ destroy_iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &destroy_iparams,
+ &destroy_oparams);
+err2:
+ if (udata)
+ qedr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl);
+ else
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+err1:
+ if (udata)
+ ib_umem_release(cq->q.umem);
+err0:
+ kfree(cq);
+ return ERR_PTR(-EINVAL);
+}
+
+int qedr_resize_cq(struct ib_cq *ibcq, int new_cnt, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+
+ DP_ERR(dev, "cq %p RESIZE NOT SUPPORTED\n", cq);
+
+ return 0;
+}
+
+int qedr_destroy_cq(struct ib_cq *ibcq)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qed_rdma_destroy_cq_out_params oparams;
+ struct qed_rdma_destroy_cq_in_params iparams;
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+
+ DP_VERBOSE(dev, QEDR_MSG_CQ, "destroy cq: cq_id %d", cq->icid);
+
+ /* GSIs CQs are handled by driver, so they don't exist in the FW */
+ if (cq->cq_type != QEDR_CQ_TYPE_GSI) {
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+ }
+
+ if (ibcq->uobject && ibcq->uobject->context) {
+ qedr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl);
+ ib_umem_release(cq->q.umem);
+ }
+
+ kfree(cq);
+
+ return 0;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index f731063..d5e1017 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -14,6 +14,8 @@ int qedr_query_port(struct ib_device *, u8 port, struct ib_port_attr *props);
int qedr_query_gid(struct ib_device *, u8 port, int index, union ib_gid *gid);
+int qedr_query_pkey(struct ib_device *, u8 port, u16 index, u16 *pkey);
+
struct ib_ucontext *qedr_alloc_ucontext(struct ib_device *, struct ib_udata *);
int qedr_dealloc_ucontext(struct ib_ucontext *);
@@ -23,4 +25,16 @@ int qedr_del_gid(struct ib_device *device, u8 port_num,
int qedr_add_gid(struct ib_device *device, u8 port_num,
unsigned int index, const union ib_gid *gid,
const struct ib_gid_attr *attr, void **context);
+struct ib_pd *qedr_alloc_pd(struct ib_device *,
+ struct ib_ucontext *, struct ib_udata *);
+int qedr_dealloc_pd(struct ib_pd *pd);
+
+struct ib_cq *qedr_create_cq(struct ib_device *ibdev,
+ const struct ib_cq_init_attr *attr,
+ struct ib_ucontext *ib_ctx,
+ struct ib_udata *udata);
+int qedr_resize_cq(struct ib_cq *, int cqe, struct ib_udata *);
+int qedr_destroy_cq(struct ib_cq *);
+int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
+
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 9cdb472..7b489f0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -805,6 +805,315 @@ static int qed_rdma_get_int(struct qed_dev *cdev, struct qed_int_info *info)
return 0;
}
+int qed_rdma_alloc_pd(void *rdma_cxt, u16 *pd)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ u32 returned_id;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Alloc PD\n");
+
+ /* Allocates an unused protection domain */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn,
+ &p_hwfn->p_rdma_info->pd_map, &returned_id);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ *pd = (u16)returned_id;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Alloc PD - done, rc = %d\n", rc);
+ return rc;
+}
+
+void qed_rdma_free_pd(void *rdma_cxt, u16 pd)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "pd = %08x\n", pd);
+
+ /* Returns a previously allocated protection domain for reuse */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->pd_map, pd);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
+static enum qed_rdma_toggle_bit
+qed_rdma_toggle_bit_create_resize_cq(struct qed_hwfn *p_hwfn, u16 icid)
+{
+ struct qed_rdma_info *p_info = p_hwfn->p_rdma_info;
+ enum qed_rdma_toggle_bit toggle_bit;
+ u32 bmap_id;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", icid);
+
+ /* the function toggle the bit that is related to a given icid
+ * and returns the new toggle bit's value
+ */
+ bmap_id = icid - qed_cxt_get_proto_cid_start(p_hwfn, p_info->proto);
+
+ spin_lock_bh(&p_info->lock);
+ toggle_bit = !test_and_change_bit(bmap_id,
+ p_info->toggle_bits.bitmap);
+ spin_unlock_bh(&p_info->lock);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QED_RDMA_TOGGLE_BIT_= %d\n",
+ toggle_bit);
+
+ return toggle_bit;
+}
+
+int qed_rdma_create_cq(void *rdma_cxt,
+ struct qed_rdma_create_cq_in_params *params, u16 *icid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_rdma_info *p_info = p_hwfn->p_rdma_info;
+ struct rdma_create_cq_ramrod_data *p_ramrod;
+ enum qed_rdma_toggle_bit toggle_bit;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ u32 returned_id, start_cid;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "cq_handle = %08x%08x\n",
+ params->cq_handle_hi, params->cq_handle_lo);
+
+ /* Allocate icid */
+ spin_lock_bh(&p_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn,
+ &p_info->cq_map, &returned_id);
+ spin_unlock_bh(&p_info->lock);
+
+ if (rc) {
+ DP_NOTICE(p_hwfn, "Can't create CQ, rc = %d\n", rc);
+ return rc;
+ }
+
+ start_cid = qed_cxt_get_proto_cid_start(p_hwfn,
+ p_info->proto);
+ *icid = returned_id + start_cid;
+
+ /* Check if icid requires a page allocation */
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, *icid);
+ if (rc)
+ goto err;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = *icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ /* Send create CQ ramrod */
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_CREATE_CQ,
+ p_info->proto, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.rdma_create_cq;
+
+ p_ramrod->cq_handle.hi = cpu_to_le32(params->cq_handle_hi);
+ p_ramrod->cq_handle.lo = cpu_to_le32(params->cq_handle_lo);
+ p_ramrod->dpi = cpu_to_le16(params->dpi);
+ p_ramrod->is_two_level_pbl = params->pbl_two_level;
+ p_ramrod->max_cqes = cpu_to_le32(params->cq_size);
+ p_ramrod->pbl_addr.hi = DMA_HI_LE(params->pbl_ptr);
+ p_ramrod->pbl_addr.lo = DMA_LO_LE(params->pbl_ptr);
+ p_ramrod->pbl_num_pages = cpu_to_le16(params->pbl_num_pages);
+ p_ramrod->cnq_id = (u8)RESC_START(p_hwfn, QED_RDMA_CNQ_RAM) +
+ params->cnq_id;
+ p_ramrod->int_timeout = params->int_timeout;
+
+ /* toggle the bit for every resize or create cq for a given icid */
+ toggle_bit = qed_rdma_toggle_bit_create_resize_cq(p_hwfn, *icid);
+
+ p_ramrod->toggle_bit = toggle_bit;
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc) {
+ /* restore toggle bit */
+ qed_rdma_toggle_bit_create_resize_cq(p_hwfn, *icid);
+ goto err;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Created CQ, rc = %d\n", rc);
+ return rc;
+
+err:
+ /* release allocated icid */
+ qed_bmap_release_id(p_hwfn, &p_info->cq_map, returned_id);
+ DP_NOTICE(p_hwfn, "Create CQ failed, rc = %d\n", rc);
+
+ return rc;
+}
+
+int qed_rdma_resize_cq(void *rdma_cxt,
+ struct qed_rdma_resize_cq_in_params *in_params,
+ struct qed_rdma_resize_cq_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_resize_cq_output_params *p_ramrod_res;
+ struct rdma_resize_cq_ramrod_data *p_ramrod;
+ enum qed_rdma_toggle_bit toggle_bit;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ u8 fw_return_code;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", in_params->icid);
+
+ p_ramrod_res =
+ (struct rdma_resize_cq_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_resize_cq_output_params),
+ &ramrod_res_phys, GFP_KERNEL);
+ if (!p_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed resize cq failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = in_params->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_RESIZE_CQ,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.rdma_resize_cq;
+
+ p_ramrod->flags = 0;
+
+ /* toggle the bit for every resize or create cq for a given icid */
+ toggle_bit = qed_rdma_toggle_bit_create_resize_cq(p_hwfn,
+ in_params->icid);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_RESIZE_CQ_RAMROD_DATA_TOGGLE_BIT, toggle_bit);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_RESIZE_CQ_RAMROD_DATA_IS_TWO_LEVEL_PBL,
+ in_params->pbl_two_level);
+
+ p_ramrod->pbl_log_page_size = in_params->pbl_page_size_log - 12;
+ p_ramrod->pbl_num_pages = cpu_to_le16(in_params->pbl_num_pages);
+ p_ramrod->max_cqes = cpu_to_le32(in_params->cq_size);
+ p_ramrod->pbl_addr.hi = DMA_HI_LE(in_params->pbl_ptr);
+ p_ramrod->pbl_addr.lo = DMA_LO_LE(in_params->pbl_ptr);
+
+ p_ramrod->output_params_addr.hi = DMA_HI_LE(ramrod_res_phys);
+ p_ramrod->output_params_addr.lo = DMA_LO_LE(ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc)
+ goto err;
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ out_params->prod = le32_to_cpu(p_ramrod_res->old_cq_prod);
+ out_params->cons = le32_to_cpu(p_ramrod_res->old_cq_cons);
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_resize_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Resized CQ, rc = %d\n", rc);
+
+ return rc;
+
+err: dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_resize_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+ DP_NOTICE(p_hwfn, "Resized CQ, Failed - rc = %d\n", rc);
+
+ return rc;
+}
+
+int qed_rdma_destroy_cq(void *rdma_cxt,
+ struct qed_rdma_destroy_cq_in_params *in_params,
+ struct qed_rdma_destroy_cq_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_destroy_cq_output_params *p_ramrod_res;
+ struct rdma_destroy_cq_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", in_params->icid);
+
+ p_ramrod_res =
+ (struct rdma_destroy_cq_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_destroy_cq_output_params),
+ &ramrod_res_phys, GFP_KERNEL);
+ if (!p_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed destroy cq failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = in_params->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ /* Send destroy CQ ramrod */
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_DESTROY_CQ,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.rdma_destroy_cq;
+ p_ramrod->output_params_addr.hi = DMA_HI_LE(ramrod_res_phys);
+ p_ramrod->output_params_addr.lo = DMA_LO_LE(ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err;
+
+ out_params->num_cq_notif = le16_to_cpu(p_ramrod_res->cnq_num);
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_destroy_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ /* Free icid */
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+
+ qed_bmap_release_id(p_hwfn,
+ &p_hwfn->p_rdma_info->cq_map,
+ (in_params->icid -
+ qed_cxt_get_proto_cid_start(p_hwfn,
+ p_hwfn->
+ p_rdma_info->proto)));
+
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroyed CQ, rc = %d\n", rc);
+ return rc;
+
+err: dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct rdma_destroy_cq_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ return rc;
+}
+
static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
{
return QED_LEADING_HWFN(cdev);
@@ -895,6 +1204,10 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_set_rdma_int = &qed_rdma_set_int,
.rdma_get_min_cnq_msix = &qed_rdma_get_min_cnq_msix,
.rdma_cnq_prod_update = &qed_rdma_cnq_prod_update,
+ .rdma_alloc_pd = &qed_rdma_alloc_pd,
+ .rdma_dealloc_pd = &qed_rdma_free_pd,
+ .rdma_create_cq = &qed_rdma_create_cq,
+ .rdma_destroy_cq = &qed_rdma_destroy_cq,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index c94c39e..6802cd0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -94,6 +94,26 @@ struct qed_rdma_info {
enum protocol_type proto;
};
+struct qed_rdma_resize_cq_in_params {
+ u16 icid;
+ u32 cq_size;
+ bool pbl_two_level;
+ u64 pbl_ptr;
+ u16 pbl_num_pages;
+ u8 pbl_page_size_log;
+};
+
+struct qed_rdma_resize_cq_out_params {
+ u32 prod;
+ u32 cons;
+};
+
+struct qed_rdma_resize_cnq_in_params {
+ u32 cnq_id;
+ u32 pbl_page_size_log;
+ u64 pbl_ptr;
+};
+
int
qed_rdma_add_user(void *rdma_cxt,
struct qed_rdma_add_user_out_params *out_params);
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index e99470c..b559b1c 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -263,6 +263,19 @@ struct qed_rdma_register_tid_in_params {
u64 dif_runt_addr;
};
+struct qed_rdma_create_cq_in_params {
+ u32 cq_handle_lo;
+ u32 cq_handle_hi;
+ u32 cq_size;
+ u16 dpi;
+ bool pbl_two_level;
+ u64 pbl_ptr;
+ u16 pbl_num_pages;
+ u8 pbl_page_size_log;
+ u8 cnq_id;
+ u16 int_timeout;
+};
+
struct qed_rdma_create_srq_in_params {
u64 pbl_base_addr;
u64 prod_pair_addr;
@@ -271,6 +284,14 @@ struct qed_rdma_create_srq_in_params {
u16 page_size;
};
+struct qed_rdma_destroy_cq_in_params {
+ u16 icid;
+};
+
+struct qed_rdma_destroy_cq_out_params {
+ u16 num_cq_notif;
+};
+
struct qed_rdma_create_srq_out_params {
u16 srq_id;
};
@@ -339,6 +360,14 @@ struct qed_rdma_ops {
int (*rdma_get_rdma_int)(struct qed_dev *cdev,
struct qed_int_info *info);
int (*rdma_set_rdma_int)(struct qed_dev *cdev, u16 cnt);
+ int (*rdma_alloc_pd)(void *rdma_cxt, u16 *pd);
+ void (*rdma_dealloc_pd)(void *rdma_cxt, u16 pd);
+ int (*rdma_create_cq)(void *rdma_cxt,
+ struct qed_rdma_create_cq_in_params *params,
+ u16 *icid);
+ int (*rdma_destroy_cq)(void *rdma_cxt,
+ struct qed_rdma_destroy_cq_in_params *iparams,
+ struct qed_rdma_destroy_cq_out_params *oparams);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 06/11] Add support for QP verbs
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for Queue Pair verbs which adds, deletes,
modifies and queries Queue Pairs.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 15 +-
drivers/infiniband/hw/qedr/qedr.h | 126 +++
drivers/infiniband/hw/qedr/qedr_cm.h | 40 +
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 11 +
drivers/infiniband/hw/qedr/qedr_user.h | 34 +
drivers/infiniband/hw/qedr/verbs.c | 1098 ++++++++++++++++++++++++-
drivers/infiniband/hw/qedr/verbs.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_cxt.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 1211 ++++++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 71 ++
include/linux/qed/qed_roce_if.h | 144 ++++
11 files changed, 2756 insertions(+), 2 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.h
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 02f0dbb..722b895 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -52,6 +52,8 @@ uint debug;
module_param(debug, uint, 0);
MODULE_PARM_DESC(debug, "Default debug msglevel");
+#define QEDR_WQ_MULTIPLIER_DFT (3)
+
static LIST_HEAD(qedr_dev_list);
static DEFINE_SPINLOCK(qedr_devlist_lock);
@@ -100,7 +102,11 @@ static int qedr_register_device(struct qedr_dev *dev)
QEDR_UVERBS(CREATE_CQ) |
QEDR_UVERBS(RESIZE_CQ) |
QEDR_UVERBS(DESTROY_CQ) |
- QEDR_UVERBS(REQ_NOTIFY_CQ);
+ QEDR_UVERBS(REQ_NOTIFY_CQ) |
+ QEDR_UVERBS(CREATE_QP) |
+ QEDR_UVERBS(MODIFY_QP) |
+ QEDR_UVERBS(QUERY_QP) |
+ QEDR_UVERBS(DESTROY_QP);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -125,6 +131,11 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.resize_cq = qedr_resize_cq;
dev->ibdev.req_notify_cq = qedr_arm_cq;
+ dev->ibdev.create_qp = qedr_create_qp;
+ dev->ibdev.modify_qp = qedr_modify_qp;
+ dev->ibdev.query_qp = qedr_query_qp;
+ dev->ibdev.destroy_qp = qedr_destroy_qp;
+
dev->ibdev.query_pkey = qedr_query_pkey;
dev->ibdev.dma_device = &dev->pdev->dev;
@@ -622,6 +633,8 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
goto init_err;
}
+ dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
+
qedr_pci_set_atomic(dev, pdev);
rc = qedr_alloc_resources(dev);
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index b2aa8fd..1c9cc40 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -49,6 +49,9 @@
enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
QEDR_MSG_CQ = 0x20000,
+ QEDR_MSG_RQ = 0x40000,
+ QEDR_MSG_SQ = 0x80000,
+ QEDR_MSG_QP = (QEDR_MSG_SQ | QEDR_MSG_RQ),
QEDR_MSG_MR = 0x100000,
QEDR_MSG_MISC = 0x400000,
};
@@ -142,6 +145,8 @@ struct qedr_dev {
u32 dp_module;
u8 dp_level;
u8 num_hwfns;
+ uint wq_multiplier;
+
};
#define QEDR_MAX_SQ_PBL (0x8000)
@@ -271,6 +276,122 @@ struct qedr_mm {
struct list_head entry;
};
+union db_prod32 {
+ struct rdma_pwm_val16_data data;
+ u32 raw;
+};
+
+struct qedr_qp_hwq_info {
+ /* WQE Elements */
+ struct qed_chain pbl;
+ u64 p_phys_addr_tbl;
+ u32 max_sges;
+
+ /* WQE */
+ u16 prod;
+ u16 cons;
+ u16 wqe_cons;
+ u16 max_wr;
+
+ /* DB */
+ void __iomem *db;
+ union db_prod32 db_data;
+};
+
+#define QEDR_INC_SW_IDX(p_info, index) \
+ do { \
+ p_info->index = (p_info->index + 1) & \
+ qed_chain_get_capacity(p_info->pbl) \
+ } while (0)
+
+enum qedr_qp_err_bitmap {
+ QEDR_QP_ERR_SQ_FULL = 1,
+ QEDR_QP_ERR_RQ_FULL = 2,
+ QEDR_QP_ERR_BAD_SR = 4,
+ QEDR_QP_ERR_BAD_RR = 8,
+ QEDR_QP_ERR_SQ_PBL_FULL = 16,
+ QEDR_QP_ERR_RQ_PBL_FULL = 32,
+};
+
+struct qedr_qp {
+ struct ib_qp ibqp; /* must be first */
+ struct qedr_dev *dev;
+
+ struct qedr_qp_hwq_info sq;
+ struct qedr_qp_hwq_info rq;
+
+ u32 max_inline_data;
+
+ /* Lock for QP's */
+ spinlock_t q_lock;
+ struct qedr_cq *sq_cq;
+ struct qedr_cq *rq_cq;
+ struct qedr_srq *srq;
+ enum qed_roce_qp_state state;
+ u32 id;
+ struct qedr_pd *pd;
+ enum ib_qp_type qp_type;
+ struct qed_rdma_qp *qed_qp;
+ u32 qp_id;
+ u16 icid;
+ u16 mtu;
+ int sgid_idx;
+ u32 rq_psn;
+ u32 sq_psn;
+ u32 qkey;
+ u32 dest_qp_num;
+
+ /* Relevant to qps created from kernel space only (ULPs) */
+ u8 prev_wqe_size;
+ u16 wqe_cons;
+ u32 err_bitmap;
+ bool signaled;
+
+ /* SQ shadow */
+ struct {
+ u64 wr_id;
+ enum ib_wc_opcode opcode;
+ u32 bytes_len;
+ u8 wqe_size;
+ bool signaled;
+ dma_addr_t icrc_mapping;
+ u32 *icrc;
+ struct qedr_mr *mr;
+ } *wqe_wr_id;
+
+ /* RQ shadow */
+ struct {
+ u64 wr_id;
+ struct ib_sge sg_list[RDMA_MAX_SGE_PER_RQ_WQE];
+ u8 wqe_size;
+
+ u16 vlan_id;
+ int rc;
+ } *rqe_wr_id;
+
+ /* Relevant to qps created from user space only (applications) */
+ struct qedr_userq usq;
+ struct qedr_userq urq;
+};
+
+static inline int qedr_get_dmac(struct qedr_dev *dev,
+ struct ib_ah_attr *ah_attr, u8 *mac_addr)
+{
+ union ib_gid zero_sgid = { { 0 } };
+ struct in6_addr in6;
+
+ if (!memcmp(&ah_attr->grh.dgid, &zero_sgid, sizeof(union ib_gid))) {
+ DP_ERR(dev, "Local port GID not supported\n");
+ eth_zero_addr(mac_addr);
+ return -EINVAL;
+ }
+
+ memcpy(&in6, ah_attr->grh.dgid.raw, sizeof(in6));
+ ether_addr_copy(mac_addr, ah_attr->dmac);
+
+ return 0;
+}
+
static inline struct qedr_dev *get_qedr_dev(struct ib_device *ibdev)
{
return container_of(ibdev, struct qedr_dev, ibdev);
@@ -291,4 +412,9 @@ static inline struct qedr_cq *get_qedr_cq(struct ib_cq *ibcq)
{
return container_of(ibcq, struct qedr_cq, ibcq);
}
+
+static inline struct qedr_qp *get_qedr_qp(struct ib_qp *ibqp)
+{
+ return container_of(ibqp, struct qedr_qp, ibqp);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.h b/drivers/infiniband/hw/qedr/qedr_cm.h
new file mode 100644
index 0000000..b8a8b76
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_cm.h
@@ -0,0 +1,40 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef LINUX_QEDR_CM_H_
+#define LINUX_QEDR_CM_H_
+
+static inline u32 qedr_get_ipv4_from_gid(u8 *gid)
+{
+ return *(u32 *)(void *)&gid[12];
+}
+
+#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
index 84f6520..4770559 100644
--- a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -158,6 +158,17 @@ struct rdma_srq_sge {
__le32 l_key;
};
+/* Rdma doorbell data for SQ and RQ */
+struct rdma_pwm_val16_data {
+ __le16 icid;
+ __le16 value;
+};
+
+union rdma_pwm_val16_data_union {
+ struct rdma_pwm_val16_data as_struct;
+ __le32 as_dword;
+};
+
/* Rdma doorbell data for CQ */
struct rdma_pwm_val32_data {
__le16 icid;
diff --git a/drivers/infiniband/hw/qedr/qedr_user.h b/drivers/infiniband/hw/qedr/qedr_user.h
index 5832c30..40598bf 100644
--- a/drivers/infiniband/hw/qedr/qedr_user.h
+++ b/drivers/infiniband/hw/qedr/qedr_user.h
@@ -43,4 +43,38 @@ struct qedr_create_cq_uresp {
u16 icid;
};
+struct qedr_create_qp_ureq {
+ u32 qp_handle_hi;
+ u32 qp_handle_lo;
+
+ /* SQ */
+ /* user space virtual address of SQ buffer */
+ u64 sq_addr;
+
+ /* length of SQ buffer */
+ size_t sq_len;
+
+ /* RQ */
+ /* user space virtual address of RQ buffer */
+ u64 rq_addr;
+
+ /* length of RQ buffer */
+ size_t rq_len;
+};
+
+struct qedr_create_qp_uresp {
+ u32 qp_id;
+ int atomic_supported;
+
+ /* SQ */
+ u32 sq_db_offset;
+ u16 sq_icid;
+
+ /* RQ */
+ u32 rq_db_offset;
+ u16 rq_icid;
+
+ u32 rq_db2_offset;
+};
+
#endif /* __QEDR_USER_H__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 2c26957..1d16c34 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -24,6 +24,7 @@
#include "qedr.h"
#include "verbs.h"
#include "qedr_user.h"
+#include "qedr_cm.h"
#define DB_ADDR_SHIFT(addr) ((addr) << DB_PWM_ADDR_OFFSET_SHIFT)
@@ -69,7 +70,6 @@ int qedr_add_gid(struct ib_device *device, u8 port_num,
if (!context)
return -EINVAL;
-
return 0;
}
@@ -958,3 +958,1099 @@ int qedr_destroy_cq(struct ib_cq *ibcq)
return 0;
}
+
+static inline int get_gid_info_from_table(struct ib_qp *ibqp,
+ struct ib_qp_attr *attr,
+ int attr_mask,
+ struct qed_rdma_modify_qp_in_params
+ *qp_params)
+{
+ enum rdma_network_type nw_type;
+ struct ib_gid_attr gid_attr;
+ union ib_gid gid;
+ u32 ipv4_addr;
+ int rc = 0;
+ int i;
+
+ rc = ib_get_cached_gid(ibqp->device, attr->ah_attr.port_num,
+ attr->ah_attr.grh.sgid_index, &gid, &gid_attr);
+ if (!rc && !memcmp(&gid, &zgid, sizeof(gid)))
+ rc = -ENOENT;
+
+ if (!rc && gid_attr.ndev) {
+ qp_params->vlan_id = rdma_vlan_dev_vlan_id(gid_attr.ndev);
+
+ dev_put(gid_attr.ndev);
+ nw_type = ib_gid_to_network_type(gid_attr.gid_type, &gid);
+ switch (nw_type) {
+ case RDMA_NETWORK_IPV6:
+ memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
+ sizeof(qp_params->sgid));
+ memcpy(&qp_params->dgid.bytes[0],
+ &attr->ah_attr.grh.dgid,
+ sizeof(qp_params->dgid));
+ qp_params->roce_mode = ROCE_V2_IPV6;
+ SET_FIELD(qp_params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1);
+ break;
+ case RDMA_NETWORK_IB:
+ memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
+ sizeof(qp_params->sgid));
+ memcpy(&qp_params->dgid.bytes[0],
+ &attr->ah_attr.grh.dgid,
+ sizeof(qp_params->dgid));
+ qp_params->roce_mode = ROCE_V1;
+ break;
+ case RDMA_NETWORK_IPV4:
+ memset(&qp_params->sgid, 0, sizeof(qp_params->sgid));
+ memset(&qp_params->dgid, 0, sizeof(qp_params->dgid));
+ ipv4_addr = qedr_get_ipv4_from_gid(gid.raw);
+ qp_params->sgid.ipv4_addr = ipv4_addr;
+ ipv4_addr =
+ qedr_get_ipv4_from_gid(attr->ah_attr.grh.dgid.raw);
+ qp_params->dgid.ipv4_addr = ipv4_addr;
+ SET_FIELD(qp_params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1);
+ qp_params->roce_mode = ROCE_V2_IPV4;
+ break;
+ }
+ }
+ if (rc)
+ return -EINVAL;
+
+ for (i = 0; i < 4; i++) {
+ qp_params->sgid.dwords[i] = ntohl(qp_params->sgid.dwords[i]);
+ qp_params->dgid.dwords[i] = ntohl(qp_params->dgid.dwords[i]);
+ }
+
+ if (qp_params->vlan_id >= VLAN_CFI_MASK)
+ qp_params->vlan_id = 0;
+
+ return rc;
+}
+
+static void qedr_cleanup_user_sq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qedr_free_pbl(dev, &qp->usq.pbl_info, qp->usq.pbl_tbl);
+ ib_umem_release(qp->usq.umem);
+}
+
+static void qedr_cleanup_user_rq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qedr_free_pbl(dev, &qp->urq.pbl_info, qp->urq.pbl_tbl);
+ ib_umem_release(qp->urq.umem);
+}
+
+static void qedr_cleanup_kernel_sq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ dev->ops->common->chain_free(dev->cdev, &qp->sq.pbl);
+ kfree(qp->wqe_wr_id);
+}
+
+static void qedr_cleanup_kernel_rq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ dev->ops->common->chain_free(dev->cdev, &qp->rq.pbl);
+ kfree(qp->rqe_wr_id);
+}
+
+static int qedr_check_qp_attrs(struct ib_pd *ibpd, struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ struct qedr_device_attr *qattr = &dev->attr;
+
+ /* QP0... attrs->qp_type == IB_QPT_GSI */
+ if (attrs->qp_type != IB_QPT_RC && attrs->qp_type != IB_QPT_GSI) {
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "create qp: unsupported qp type=0x%x requested\n",
+ attrs->qp_type);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_send_wr > qattr->max_sqe) {
+ DP_ERR(dev,
+ "create qp: cannot create a SQ with %d elements (max_send_wr=0x%x)\n",
+ attrs->cap.max_send_wr, qattr->max_sqe);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_inline_data > qattr->max_inline) {
+ DP_ERR(dev,
+ "create qp: unsupported inline data size=0x%x requested (max_inline=0x%x)\n",
+ attrs->cap.max_inline_data, qattr->max_inline);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_send_sge > qattr->max_sge) {
+ DP_ERR(dev,
+ "create qp: unsupported send_sge=0x%x requested (max_send_sge=0x%x)\n",
+ attrs->cap.max_send_sge, qattr->max_sge);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_recv_sge > qattr->max_sge) {
+ DP_ERR(dev,
+ "create qp: unsupported recv_sge=0x%x requested (max_recv_sge=0x%x)\n",
+ attrs->cap.max_recv_sge, qattr->max_sge);
+ return -EINVAL;
+ }
+
+ /* Unprivileged user space cannot create special QP */
+ if (ibpd->uobject && attrs->qp_type == IB_QPT_GSI) {
+ DP_ERR(dev,
+ "create qp: userspace can't create special QPs of type=0x%x\n",
+ attrs->qp_type);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static void qedr_copy_rq_uresp(struct qedr_create_qp_uresp *uresp,
+ struct qedr_qp *qp)
+{
+ uresp->rq_db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
+ uresp->rq_icid = qp->icid;
+}
+
+static void qedr_copy_sq_uresp(struct qedr_create_qp_uresp *uresp,
+ struct qedr_qp *qp)
+{
+ uresp->sq_db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
+ uresp->sq_icid = qp->icid + 1;
+}
+
+static int qedr_copy_qp_uresp(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct ib_udata *udata)
+{
+ struct qedr_create_qp_uresp uresp;
+ int rc;
+
+ memset(&uresp, 0, sizeof(uresp));
+ qedr_copy_sq_uresp(&uresp, qp);
+ qedr_copy_rq_uresp(&uresp, qp);
+
+ uresp.atomic_supported = dev->atomic_cap != IB_ATOMIC_NONE;
+ uresp.qp_id = qp->qp_id;
+
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ DP_ERR(dev,
+ "create qp: failed a copy to user space with qp icid=0x%x.\n",
+ qp->icid);
+
+ return rc;
+}
+
+static void qedr_set_qp_init_params(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct qedr_pd *pd,
+ struct ib_qp_init_attr *attrs)
+{
+ qp->pd = pd;
+
+ spin_lock_init(&qp->q_lock);
+
+ qp->qp_type = attrs->qp_type;
+ qp->max_inline_data = attrs->cap.max_inline_data;
+ qp->sq.max_sges = attrs->cap.max_send_sge;
+ qp->state = QED_ROCE_QP_STATE_RESET;
+ qp->signaled = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR) ? true : false;
+ qp->sq_cq = get_qedr_cq(attrs->send_cq);
+ qp->rq_cq = get_qedr_cq(attrs->recv_cq);
+ qp->dev = dev;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "QP params:\tpd = %d, qp_type = %d, max_inline_data = %d, state = %d, signaled = %d, use_srq=%d\n",
+ pd->pd_id, qp->qp_type, qp->max_inline_data,
+ qp->state, qp->signaled, (attrs->srq) ? 1 : 0);
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "SQ params:\tsq_max_sges = %d, sq_cq_id = %d\n",
+ qp->sq.max_sges, qp->sq_cq->icid);
+ qp->rq.max_sges = attrs->cap.max_recv_sge;
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "RQ params:\trq_max_sges = %d, rq_cq_id = %d\n",
+ qp->rq.max_sges, qp->rq_cq->icid);
+}
+
+static inline void
+qedr_init_qp_user_params(struct qed_rdma_create_qp_in_params *params,
+ struct qedr_create_qp_ureq *ureq)
+{
+ /* QP handle to be written in CQE */
+ params->qp_handle_lo = ureq->qp_handle_lo;
+ params->qp_handle_hi = ureq->qp_handle_hi;
+}
+
+static inline void
+qedr_init_qp_kernel_doorbell_sq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qp->sq.db = dev->db_addr +
+ DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
+ qp->sq.db_data.data.icid = qp->icid + 1;
+}
+
+static inline void
+qedr_init_qp_kernel_doorbell_rq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qp->rq.db = dev->db_addr +
+ DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
+ qp->rq.db_data.data.icid = qp->icid;
+}
+
+static inline int
+qedr_init_qp_kernel_params_rq(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct ib_qp_init_attr *attrs)
+{
+ /* Allocate driver internal RQ array */
+ qp->rqe_wr_id = kcalloc(qp->rq.max_wr, sizeof(*qp->rqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->rqe_wr_id) {
+ DP_ERR(dev, "create qp: failed RQ shadow memory allocation\n");
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "RQ max_wr set to %d.\n", qp->rq.max_wr);
+
+ return 0;
+}
+
+static inline int
+qedr_init_qp_kernel_params_sq(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ u32 temp_max_wr;
+
+ /* Allocate driver internal SQ array */
+ temp_max_wr = attrs->cap.max_send_wr * dev->wq_multiplier;
+ temp_max_wr = min_t(u32, temp_max_wr, dev->attr.max_sqe);
+
+ /* temp_max_wr < attr->max_sqe < u16 so the casting is safe */
+ qp->sq.max_wr = (u16)temp_max_wr;
+ qp->wqe_wr_id = kcalloc(qp->sq.max_wr, sizeof(*qp->wqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->wqe_wr_id) {
+ DP_ERR(dev, "create qp: failed SQ shadow memory allocation\n");
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "SQ max_wr set to %d.\n", qp->sq.max_wr);
+
+ /* QP handle to be written in CQE */
+ params->qp_handle_lo = lower_32_bits((uintptr_t)qp);
+ params->qp_handle_hi = upper_32_bits((uintptr_t)qp);
+
+ return 0;
+}
+
+static inline int qedr_init_qp_kernel_sq(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs)
+{
+ u32 n_sq_elems, n_sq_entries;
+ int rc;
+
+ /* A single work request may take up to QEDR_MAX_SQ_WQE_SIZE elements in
+ * the ring. The ring should allow at least a single WR, even if the
+ * user requested none, due to allocation issues.
+ */
+ n_sq_entries = attrs->cap.max_send_wr;
+ n_sq_entries = min_t(u32, n_sq_entries, dev->attr.max_sqe);
+ n_sq_entries = max_t(u32, n_sq_entries, 1);
+ n_sq_elems = n_sq_entries * QEDR_MAX_SQE_ELEMENTS_PER_SQE;
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U32,
+ n_sq_elems,
+ QEDR_SQE_ELEMENT_SIZE,
+ &qp->sq.pbl);
+ if (rc) {
+ DP_ERR(dev, "failed to allocate QP %p SQ\n", qp);
+ return rc;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_SQ,
+ "SQ Pbl base addr = %llx max_send_wr=%d max_wr=%d capacity=%d, rc=%d\n",
+ qed_chain_get_pbl_phys(&qp->sq.pbl), attrs->cap.max_send_wr,
+ n_sq_entries, qed_chain_get_capacity(&qp->sq.pbl), rc);
+ return 0;
+}
+
+static inline int qedr_init_qp_kernel_rq(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs)
+{
+ u32 n_rq_elems, n_rq_entries;
+ int rc;
+
+ /* A single work request may take up to QEDR_MAX_RQ_WQE_SIZE elements in
+ * the ring. There ring should allow at least a single WR, even if the
+ * user requested none, due to allocation issues.
+ */
+ n_rq_entries = max_t(u32, attrs->cap.max_recv_wr, 1);
+ n_rq_elems = n_rq_entries * QEDR_MAX_RQE_ELEMENTS_PER_RQE;
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U32,
+ n_rq_elems,
+ QEDR_RQE_ELEMENT_SIZE,
+ &qp->rq.pbl);
+
+ if (rc) {
+ DP_ERR(dev, "failed to allocate memory for QP %p RQ\n", qp);
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_RQ,
+ "RQ Pbl base addr = %llx max_recv_wr=%d max_wr=%d capacity=%d, rc=%d\n",
+ qed_chain_get_pbl_phys(&qp->rq.pbl), attrs->cap.max_recv_wr,
+ n_rq_entries, qed_chain_get_capacity(&qp->rq.pbl), rc);
+
+ /* n_rq_entries < u16 so the casting is safe */
+ qp->rq.max_wr = (u16)n_rq_entries;
+
+ return 0;
+}
+
+static inline void
+qedr_init_qp_in_params_sq(struct qedr_dev *dev,
+ struct qedr_pd *pd,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct ib_udata *udata,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ /* QP handle to be written in an async event */
+ params->qp_handle_async_lo = lower_32_bits((uintptr_t)qp);
+ params->qp_handle_async_hi = upper_32_bits((uintptr_t)qp);
+
+ params->signal_all = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR);
+ params->fmr_and_reserved_lkey = !udata;
+ params->pd = pd->pd_id;
+ params->dpi = pd->uctx ? pd->uctx->dpi : dev->dpi;
+ params->sq_cq_id = get_qedr_cq(attrs->send_cq)->icid;
+ params->max_sq_sges = 0;
+ params->stats_queue = 0;
+
+ if (udata) {
+ params->sq_num_pages = qp->usq.pbl_info.num_pbes;
+ params->sq_pbl_ptr = qp->usq.pbl_tbl->pa;
+ } else {
+ params->sq_num_pages = qed_chain_get_page_cnt(&qp->sq.pbl);
+ params->sq_pbl_ptr = qed_chain_get_pbl_phys(&qp->sq.pbl);
+ }
+}
+
+static inline void
+qedr_init_qp_in_params_rq(struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct ib_udata *udata,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ params->rq_cq_id = get_qedr_cq(attrs->recv_cq)->icid;
+ params->srq_id = 0;
+ params->use_srq = false;
+
+ if (udata) {
+ params->rq_num_pages = qp->urq.pbl_info.num_pbes;
+ params->rq_pbl_ptr = qp->urq.pbl_tbl->pa;
+ } else {
+ params->rq_num_pages = qed_chain_get_page_cnt(&qp->rq.pbl);
+ params->rq_pbl_ptr = qed_chain_get_pbl_phys(&qp->rq.pbl);
+ }
+}
+
+static inline void qedr_qp_user_print(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "create qp: successfully created user QP. qp=%p, sq_addr=0x%llx, sq_len=%zd, rq_addr=0x%llx, rq_len=%zd\n",
+ qp, qp->usq.buf_addr, qp->usq.buf_len, qp->urq.buf_addr,
+ qp->urq.buf_len);
+}
+
+static inline int qedr_init_user_qp(struct ib_ucontext *ib_ctx,
+ struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct qedr_create_qp_ureq *ureq)
+{
+ int rc;
+
+ /* SQ - read access only (0), dma sync not required (0) */
+ rc = qedr_init_user_queue(ib_ctx, dev, &qp->usq, ureq->sq_addr,
+ ureq->sq_len, 0, 0);
+ if (rc)
+ return rc;
+
+ /* RQ - read access only (0), dma sync not required (0) */
+ rc = qedr_init_user_queue(ib_ctx, dev, &qp->urq, ureq->rq_addr,
+ ureq->rq_len, 0, 0);
+
+ if (rc)
+ qedr_cleanup_user_sq(dev, qp);
+ return rc;
+}
+
+static inline int
+qedr_init_kernel_qp(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ int rc;
+
+ rc = qedr_init_qp_kernel_sq(dev, qp, attrs);
+ if (rc) {
+ DP_ERR(dev, "failed to init kernel QP %p SQ\n", qp);
+ return rc;
+ }
+
+ rc = qedr_init_qp_kernel_params_sq(dev, qp, attrs, params);
+ if (rc) {
+ dev->ops->common->chain_free(dev->cdev, &qp->sq.pbl);
+ DP_ERR(dev, "failed to init kernel QP %p SQ params\n", qp);
+ return rc;
+ }
+
+ rc = qedr_init_qp_kernel_rq(dev, qp, attrs);
+ if (rc) {
+ qedr_cleanup_kernel_sq(dev, qp);
+ DP_ERR(dev, "failed to init kernel QP %p RQ\n", qp);
+ return rc;
+ }
+
+ rc = qedr_init_qp_kernel_params_rq(dev, qp, attrs);
+ if (rc) {
+ DP_ERR(dev, "failed to init kernel QP %p RQ params\n", qp);
+ qedr_cleanup_kernel_sq(dev, qp);
+ dev->ops->common->chain_free(dev->cdev, &qp->rq.pbl);
+ return rc;
+ }
+
+ return rc;
+}
+
+struct ib_qp *qedr_create_qp(struct ib_pd *ibpd,
+ struct ib_qp_init_attr *attrs,
+ struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qed_rdma_create_qp_out_params out_params;
+ struct qed_rdma_create_qp_in_params in_params;
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct ib_ucontext *ib_ctx = NULL;
+ struct qedr_ucontext *ctx = NULL;
+ struct qedr_create_qp_ureq ureq;
+ struct qedr_qp *qp;
+ int rc = 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "create qp: called from %s, pd=%p\n",
+ udata ? "user library" : "kernel", pd);
+
+ rc = qedr_check_qp_attrs(ibpd, dev, attrs);
+ if (rc)
+ return ERR_PTR(rc);
+
+ qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+ if (!qp) {
+ DP_ERR(dev, "create qp: failed allocating memory\n");
+ return ERR_PTR(-ENOMEM);
+ }
+
+ if (attrs->srq)
+ return ERR_PTR(-EINVAL);
+
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "create qp: sq_cq=%p, sq_icid=%d, rq_cq=%p, rq_icid=%d\n",
+ get_qedr_cq(attrs->send_cq),
+ get_qedr_cq(attrs->send_cq)->icid,
+ get_qedr_cq(attrs->recv_cq),
+ get_qedr_cq(attrs->recv_cq)->icid);
+
+ qedr_set_qp_init_params(dev, qp, pd, attrs);
+
+ memset(&in_params, 0, sizeof(in_params));
+
+ if (udata) {
+ if (!(udata && ibpd->uobject && ibpd->uobject->context))
+ goto err0;
+
+ ib_ctx = ibpd->uobject->context;
+ ctx = get_qedr_ucontext(ib_ctx);
+
+ memset(&ureq, 0, sizeof(ureq));
+ if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) {
+ DP_ERR(dev,
+ "create qp: problem copying data from user space\n");
+ goto err0;
+ }
+
+ rc = qedr_init_user_qp(ib_ctx, dev, qp, &ureq);
+ if (rc)
+ goto err0;
+
+ qedr_init_qp_user_params(&in_params, &ureq);
+ } else {
+ rc = qedr_init_kernel_qp(dev, qp, attrs, &in_params);
+ if (rc)
+ goto err0;
+ }
+
+ qedr_init_qp_in_params_sq(dev, pd, qp, attrs, udata, &in_params);
+ qedr_init_qp_in_params_rq(qp, attrs, udata, &in_params);
+
+ qp->qed_qp = dev->ops->rdma_create_qp(dev->rdma_ctx,
+ &in_params, &out_params);
+
+ if (!qp->qed_qp)
+ goto err1;
+
+ qp->qp_id = out_params.qp_id;
+ qp->icid = out_params.icid;
+ qp->ibqp.qp_num = qp->qp_id;
+
+ if (udata) {
+ rc = qedr_copy_qp_uresp(dev, qp, udata);
+ if (rc)
+ goto err2;
+
+ qedr_qp_user_print(dev, qp);
+ } else {
+ qedr_init_qp_kernel_doorbell_sq(dev, qp);
+ qedr_init_qp_kernel_doorbell_rq(dev, qp);
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "created %s space QP %p\n", udata ? "user" : "kernel", qp);
+
+ return &qp->ibqp;
+
+err2:
+ rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp);
+ if (rc)
+ DP_ERR(dev, "create qp: fatal fault. rc=%d", rc);
+err1:
+ if (udata) {
+ qedr_cleanup_user_sq(dev, qp);
+ qedr_cleanup_user_rq(dev, qp);
+ } else {
+ qedr_cleanup_kernel_sq(dev, qp);
+ qedr_cleanup_kernel_rq(dev, qp);
+ }
+
+err0:
+ kfree(qp);
+
+ return ERR_PTR(-EFAULT);
+}
+
+enum ib_qp_state qedr_get_ibqp_state(enum qed_roce_qp_state qp_state)
+{
+ switch (qp_state) {
+ case QED_ROCE_QP_STATE_RESET:
+ return IB_QPS_RESET;
+ case QED_ROCE_QP_STATE_INIT:
+ return IB_QPS_INIT;
+ case QED_ROCE_QP_STATE_RTR:
+ return IB_QPS_RTR;
+ case QED_ROCE_QP_STATE_RTS:
+ return IB_QPS_RTS;
+ case QED_ROCE_QP_STATE_SQD:
+ return IB_QPS_SQD;
+ case QED_ROCE_QP_STATE_ERR:
+ return IB_QPS_ERR;
+ case QED_ROCE_QP_STATE_SQE:
+ return IB_QPS_SQE;
+ }
+ return IB_QPS_ERR;
+}
+
+enum qed_roce_qp_state qedr_get_state_from_ibqp(enum ib_qp_state qp_state)
+{
+ switch (qp_state) {
+ case IB_QPS_RESET:
+ return QED_ROCE_QP_STATE_RESET;
+ case IB_QPS_INIT:
+ return QED_ROCE_QP_STATE_INIT;
+ case IB_QPS_RTR:
+ return QED_ROCE_QP_STATE_RTR;
+ case IB_QPS_RTS:
+ return QED_ROCE_QP_STATE_RTS;
+ case IB_QPS_SQD:
+ return QED_ROCE_QP_STATE_SQD;
+ case IB_QPS_ERR:
+ return QED_ROCE_QP_STATE_ERR;
+ default:
+ return QED_ROCE_QP_STATE_ERR;
+ }
+}
+
+static void qedr_reset_qp_hwq_info(struct qedr_qp_hwq_info *qph)
+{
+ qed_chain_reset(&qph->pbl);
+ qph->prod = 0;
+ qph->cons = 0;
+ qph->wqe_cons = 0;
+ qph->db_data.data.value = cpu_to_le16(0);
+}
+
+static int qedr_update_qp_state(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ enum qed_roce_qp_state new_state)
+{
+ int status = 0;
+
+ if (new_state == qp->state)
+ return 1;
+
+ switch (qp->state) {
+ case QED_ROCE_QP_STATE_RESET:
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_INIT:
+ qp->prev_wqe_size = 0;
+ qedr_reset_qp_hwq_info(&qp->sq);
+ qedr_reset_qp_hwq_info(&qp->rq);
+ break;
+ default:
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_INIT:
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RTR:
+ /* Update doorbell (in case post_recv was
+ * done before move to RTR)
+ */
+ wmb();
+ writel(qp->rq.db_data.raw, qp->rq.db);
+ /* Make sure write takes effect */
+ mmiowb();
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_RTR:
+ /* RTR->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RTS:
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_RTS:
+ /* RTS->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_SQD:
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_SQD:
+ /* SQD->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RTS:
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ /* ERR->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RESET:
+ break;
+ default:
+ status = -EINVAL;
+ break;
+ };
+ break;
+ default:
+ status = -EINVAL;
+ break;
+ };
+
+ return status;
+}
+
+int qedr_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_udata *udata)
+{
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_rdma_modify_qp_in_params qp_params = { 0 };
+ struct qedr_dev *dev = get_qedr_dev(&qp->dev->ibdev);
+ enum ib_qp_state old_qp_state, new_qp_state;
+ int rc = 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "modify qp: qp %p attr_mask=0x%x, state=%d", qp, attr_mask,
+ attr->qp_state);
+
+ old_qp_state = qedr_get_ibqp_state(qp->state);
+ if (attr_mask & IB_QP_STATE)
+ new_qp_state = attr->qp_state;
+ else
+ new_qp_state = old_qp_state;
+
+ if (!ib_modify_qp_is_ok
+ (old_qp_state, new_qp_state, ibqp->qp_type, attr_mask,
+ IB_LINK_LAYER_ETHERNET)) {
+ DP_ERR(dev,
+ "modify qp: invalid attribute mask=0x%x specified for\n"
+ "qpn=0x%x of type=0x%x old_qp_state=0x%x, new_qp_state=0x%x\n",
+ attr_mask, qp->qp_id, ibqp->qp_type, old_qp_state,
+ new_qp_state);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ /* Translate the masks... */
+ if (attr_mask & IB_QP_STATE) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_NEW_STATE, 1);
+ qp_params.new_state = qedr_get_state_from_ibqp(attr->qp_state);
+ }
+
+ if (attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY)
+ qp_params.sqd_async = true;
+
+ if (attr_mask & IB_QP_PKEY_INDEX) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_PKEY, 1);
+ if (attr->pkey_index >= QEDR_ROCE_PKEY_TABLE_LEN) {
+ rc = -EINVAL;
+ goto err;
+ }
+
+ qp_params.pkey = QEDR_ROCE_PKEY_DEFAULT;
+ }
+
+ if (attr_mask & IB_QP_QKEY)
+ qp->qkey = attr->qkey;
+
+ if (attr_mask & IB_QP_ACCESS_FLAGS) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN, 1);
+ qp_params.incoming_rdma_read_en = attr->qp_access_flags &
+ IB_ACCESS_REMOTE_READ;
+ qp_params.incoming_rdma_write_en = attr->qp_access_flags &
+ IB_ACCESS_REMOTE_WRITE;
+ qp_params.incoming_atomic_en = attr->qp_access_flags &
+ IB_ACCESS_REMOTE_ATOMIC;
+ }
+
+ if (attr_mask & (IB_QP_AV | IB_QP_PATH_MTU)) {
+ if (attr_mask & IB_QP_PATH_MTU) {
+ if (attr->path_mtu < IB_MTU_256 ||
+ attr->path_mtu > IB_MTU_4096) {
+ pr_err("ERROR QEDR: Only MTU sizes of 256, 512, 1024, 2048 and 4096 are supported\n");
+ rc = -EINVAL;
+ goto err;
+ }
+ qp->mtu = min(ib_mtu_enum_to_int(attr->path_mtu),
+ ib_mtu_enum_to_int(iboe_get_mtu
+ (dev->ndev->mtu)));
+ }
+
+ if (!qp->mtu) {
+ qp->mtu =
+ ib_mtu_enum_to_int(iboe_get_mtu(dev->ndev->mtu));
+ pr_err("Fixing zeroed MTU to qp->mtu = %d\n", qp->mtu);
+ }
+
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR, 1);
+
+ qp_params.traffic_class_tos = attr->ah_attr.grh.traffic_class;
+ qp_params.flow_label = attr->ah_attr.grh.flow_label;
+ qp_params.hop_limit_ttl = attr->ah_attr.grh.hop_limit;
+
+ qp->sgid_idx = attr->ah_attr.grh.sgid_index;
+
+ rc = get_gid_info_from_table(ibqp, attr, attr_mask, &qp_params);
+ if (rc) {
+ DP_ERR(dev,
+ "modify qp: problems with GID index %d (rc=%d)\n",
+ attr->ah_attr.grh.sgid_index, rc);
+ return rc;
+ }
+
+ rc = qedr_get_dmac(dev, &attr->ah_attr,
+ qp_params.remote_mac_addr);
+ if (rc)
+ return rc;
+
+ qp_params.use_local_mac = true;
+ ether_addr_copy(qp_params.local_mac_addr, dev->ndev->dev_addr);
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "dgid=%x:%x:%x:%x\n",
+ qp_params.dgid.dwords[0], qp_params.dgid.dwords[1],
+ qp_params.dgid.dwords[2], qp_params.dgid.dwords[3]);
+ DP_VERBOSE(dev, QEDR_MSG_QP, "sgid=%x:%x:%x:%x\n",
+ qp_params.sgid.dwords[0], qp_params.sgid.dwords[1],
+ qp_params.sgid.dwords[2], qp_params.sgid.dwords[3]);
+ DP_VERBOSE(dev, QEDR_MSG_QP, "remote_mac=[%x:%x:%x:%x:%x:%x]\n",
+ qp_params.remote_mac_addr[0],
+ qp_params.remote_mac_addr[1],
+ qp_params.remote_mac_addr[2],
+ qp_params.remote_mac_addr[3],
+ qp_params.remote_mac_addr[4],
+ qp_params.remote_mac_addr[5]);
+
+ qp_params.mtu = qp->mtu;
+ qp_params.lb_indication = false;
+ }
+
+ if (!qp_params.mtu) {
+ /* Stay with current MTU */
+ if (qp->mtu)
+ qp_params.mtu = qp->mtu;
+ else
+ qp_params.mtu =
+ ib_mtu_enum_to_int(iboe_get_mtu(dev->ndev->mtu));
+ }
+
+ if (attr_mask & IB_QP_TIMEOUT) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT, 1);
+
+ qp_params.ack_timeout = attr->timeout;
+ if (attr->timeout) {
+ u32 temp;
+
+ temp = 4096 * (1UL << attr->timeout) / 1000 / 1000;
+ /* FW requires [msec] */
+ qp_params.ack_timeout = temp;
+ } else {
+ /* Infinite */
+ qp_params.ack_timeout = 0;
+ }
+ }
+ if (attr_mask & IB_QP_RETRY_CNT) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RETRY_CNT, 1);
+ qp_params.retry_cnt = attr->retry_cnt;
+ }
+
+ if (attr_mask & IB_QP_RNR_RETRY) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT, 1);
+ qp_params.rnr_retry_cnt = attr->rnr_retry;
+ }
+
+ if (attr_mask & IB_QP_RQ_PSN) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RQ_PSN, 1);
+ qp_params.rq_psn = attr->rq_psn;
+ qp->rq_psn = attr->rq_psn;
+ }
+
+ if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
+ if (attr->max_rd_atomic > dev->attr.max_qp_req_rd_atomic_resc) {
+ rc = -EINVAL;
+ DP_ERR(dev,
+ "unsupported max_rd_atomic=%d, supported=%d\n",
+ attr->max_rd_atomic,
+ dev->attr.max_qp_req_rd_atomic_resc);
+ goto err;
+ }
+
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ, 1);
+ qp_params.max_rd_atomic_req = attr->max_rd_atomic;
+ }
+
+ if (attr_mask & IB_QP_MIN_RNR_TIMER) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER, 1);
+ qp_params.min_rnr_nak_timer = attr->min_rnr_timer;
+ }
+
+ if (attr_mask & IB_QP_SQ_PSN) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_SQ_PSN, 1);
+ qp_params.sq_psn = attr->sq_psn;
+ qp->sq_psn = attr->sq_psn;
+ }
+
+ if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
+ if (attr->max_dest_rd_atomic >
+ dev->attr.max_qp_resp_rd_atomic_resc) {
+ DP_ERR(dev,
+ "unsupported max_dest_rd_atomic=%d, supported=%d\n",
+ attr->max_dest_rd_atomic,
+ dev->attr.max_qp_resp_rd_atomic_resc);
+
+ rc = -EINVAL;
+ goto err;
+ }
+
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP, 1);
+ qp_params.max_rd_atomic_resp = attr->max_dest_rd_atomic;
+ }
+
+ if (attr_mask & IB_QP_DEST_QPN) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_DEST_QP, 1);
+
+ qp_params.dest_qp = attr->dest_qp_num;
+ qp->dest_qp_num = attr->dest_qp_num;
+ }
+
+ if (qp->qp_type != IB_QPT_GSI)
+ rc = dev->ops->rdma_modify_qp(dev->rdma_ctx,
+ qp->qed_qp, &qp_params);
+
+ if (attr_mask & IB_QP_STATE) {
+ if ((qp->qp_type != IB_QPT_GSI) && (!udata))
+ qedr_update_qp_state(dev, qp, qp_params.new_state);
+ qp->state = qp_params.new_state;
+ }
+
+err:
+ return rc;
+}
+
+static int qedr_to_ib_qp_acc_flags(struct qed_rdma_query_qp_out_params *params)
+{
+ int ib_qp_acc_flags = 0;
+
+ if (params->incoming_rdma_write_en)
+ ib_qp_acc_flags |= IB_ACCESS_REMOTE_WRITE;
+ if (params->incoming_rdma_read_en)
+ ib_qp_acc_flags |= IB_ACCESS_REMOTE_READ;
+ if (params->incoming_atomic_en)
+ ib_qp_acc_flags |= IB_ACCESS_REMOTE_ATOMIC;
+ ib_qp_acc_flags |= IB_ACCESS_LOCAL_WRITE;
+ return ib_qp_acc_flags;
+}
+
+int qedr_query_qp(struct ib_qp *ibqp,
+ struct ib_qp_attr *qp_attr,
+ int attr_mask, struct ib_qp_init_attr *qp_init_attr)
+{
+ struct qed_rdma_query_qp_out_params params;
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qedr_dev *dev = qp->dev;
+ int rc = 0;
+
+ memset(¶ms, 0, sizeof(params));
+
+ rc = dev->ops->rdma_query_qp(dev->rdma_ctx, qp->qed_qp, ¶ms);
+ if (rc)
+ goto err;
+
+ memset(qp_attr, 0, sizeof(*qp_attr));
+ memset(qp_init_attr, 0, sizeof(*qp_init_attr));
+
+ qp_attr->qp_state = qedr_get_ibqp_state(params.state);
+ qp_attr->cur_qp_state = qedr_get_ibqp_state(params.state);
+ qp_attr->path_mtu = iboe_get_mtu(params.mtu);
+ qp_attr->path_mig_state = IB_MIG_MIGRATED;
+ qp_attr->rq_psn = params.rq_psn;
+ qp_attr->sq_psn = params.sq_psn;
+ qp_attr->dest_qp_num = params.dest_qp;
+
+ qp_attr->qp_access_flags = qedr_to_ib_qp_acc_flags(¶ms);
+
+ qp_attr->cap.max_send_wr = qp->sq.max_wr;
+ qp_attr->cap.max_recv_wr = qp->rq.max_wr;
+ qp_attr->cap.max_send_sge = qp->sq.max_sges;
+ qp_attr->cap.max_recv_sge = qp->rq.max_sges;
+ qp_attr->cap.max_inline_data = qp->max_inline_data;
+ qp_init_attr->cap = qp_attr->cap;
+
+ memcpy(&qp_attr->ah_attr.grh.dgid.raw[0], ¶ms.dgid.bytes[0],
+ sizeof(qp_attr->ah_attr.grh.dgid.raw));
+
+ qp_attr->ah_attr.grh.flow_label = params.flow_label;
+ qp_attr->ah_attr.grh.sgid_index = qp->sgid_idx;
+ qp_attr->ah_attr.grh.hop_limit = params.hop_limit_ttl;
+ qp_attr->ah_attr.grh.traffic_class = params.traffic_class_tos;
+
+ qp_attr->ah_attr.ah_flags = IB_AH_GRH;
+ qp_attr->ah_attr.port_num = 1;
+ qp_attr->ah_attr.sl = 0;
+ qp_attr->timeout = params.timeout;
+ qp_attr->rnr_retry = params.rnr_retry;
+ qp_attr->retry_cnt = params.retry_cnt;
+ qp_attr->min_rnr_timer = params.min_rnr_nak_timer;
+ qp_attr->pkey_index = params.pkey_index;
+ qp_attr->port_num = 1;
+ qp_attr->ah_attr.src_path_bits = 0;
+ qp_attr->ah_attr.static_rate = 0;
+ qp_attr->alt_pkey_index = 0;
+ qp_attr->alt_port_num = 0;
+ qp_attr->alt_timeout = 0;
+ memset(&qp_attr->alt_ah_attr, 0, sizeof(qp_attr->alt_ah_attr));
+
+ qp_attr->sq_draining = (params.state == QED_ROCE_QP_STATE_SQD) ? 1 : 0;
+ qp_attr->max_dest_rd_atomic = params.max_dest_rd_atomic;
+ qp_attr->max_rd_atomic = params.max_rd_atomic;
+ qp_attr->en_sqd_async_notify = (params.sqd_async) ? 1 : 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "QEDR_QUERY_QP: max_inline_data=%d\n",
+ qp_attr->cap.max_inline_data);
+
+err:
+ return rc;
+}
+
+int qedr_destroy_qp(struct ib_qp *ibqp)
+{
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qedr_dev *dev = qp->dev;
+ struct ib_qp_attr attr;
+ int attr_mask = 0;
+ int rc = 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "destroy qp: destroying %p, qp type=%d\n",
+ qp, qp->qp_type);
+
+ if (qp->state != (QED_ROCE_QP_STATE_RESET | QED_ROCE_QP_STATE_ERR |
+ QED_ROCE_QP_STATE_INIT)) {
+ attr.qp_state = IB_QPS_ERR;
+ attr_mask |= IB_QP_STATE;
+
+ /* Change the QP state to ERROR */
+ qedr_modify_qp(ibqp, &attr, attr_mask, NULL);
+ }
+
+ if (qp->qp_type != IB_QPT_GSI) {
+ rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp);
+ if (rc)
+ return rc;
+ }
+
+ if (ibqp->uobject && ibqp->uobject->context) {
+ qedr_cleanup_user_sq(dev, qp);
+ qedr_cleanup_user_rq(dev, qp);
+ } else {
+ qedr_cleanup_kernel_sq(dev, qp);
+ qedr_cleanup_kernel_rq(dev, qp);
+ }
+
+ kfree(qp);
+
+ return rc;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index d5e1017..80dbac1 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -36,5 +36,12 @@ struct ib_cq *qedr_create_cq(struct ib_device *ibdev,
int qedr_resize_cq(struct ib_cq *, int cqe, struct ib_udata *);
int qedr_destroy_cq(struct ib_cq *);
int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
+struct ib_qp *qedr_create_qp(struct ib_pd *, struct ib_qp_init_attr *attrs,
+ struct ib_udata *);
+int qedr_modify_qp(struct ib_qp *, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_udata *udata);
+int qedr_query_qp(struct ib_qp *, struct ib_qp_attr *qp_attr,
+ int qp_attr_mask, struct ib_qp_init_attr *);
+int qedr_destroy_qp(struct ib_qp *ibqp);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.h b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
index d00ad05..2b8bdaa 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
@@ -176,6 +176,7 @@ u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
enum protocol_type type);
u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
enum protocol_type type);
+int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto);
#define QED_CTX_WORKING_MEM 0
#define QED_CTX_FL_MEM 1
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 7b489f0..cf92ff7 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -1114,6 +1114,1213 @@ err: dma_free_coherent(&p_hwfn->cdev->pdev->dev,
return rc;
}
+static void qed_rdma_set_fw_mac(u16 *p_fw_mac, u8 *p_qed_mac)
+{
+ p_fw_mac[0] = cpu_to_le16((p_qed_mac[0] << 8) + p_qed_mac[1]);
+ p_fw_mac[1] = cpu_to_le16((p_qed_mac[2] << 8) + p_qed_mac[3]);
+ p_fw_mac[2] = cpu_to_le16((p_qed_mac[4] << 8) + p_qed_mac[5]);
+}
+
+static void qed_rdma_copy_gids(struct qed_rdma_qp *qp, __le32 *src_gid,
+ __le32 *dst_gid)
+{
+ u32 i;
+
+ if (qp->roce_mode == ROCE_V2_IPV4) {
+ /* RoCE v2 - IPv4 only */
+ /* The IPv4 addresses shall be aligned to the highest word.
+ * The lower words must be zero.
+ */
+ memset(src_gid, 0, sizeof(union qed_gid));
+ memset(dst_gid, 0, sizeof(union qed_gid));
+ src_gid[3] = cpu_to_le32(qp->sgid.ipv4_addr);
+ dst_gid[3] = cpu_to_le32(qp->dgid.ipv4_addr);
+ } else {
+ /* RoCE, and RoCE v2 - IPv6 */
+ /* GIDs and IPv6 addresses coincide in location and size */
+ for (i = 0; i < ARRAY_SIZE(qp->sgid.dwords); i++) {
+ src_gid[i] = cpu_to_le32(qp->sgid.dwords[i]);
+ dst_gid[i] = cpu_to_le32(qp->dgid.dwords[i]);
+ }
+ }
+}
+
+static enum roce_flavor qed_roce_mode_to_flavor(enum roce_mode roce_mode)
+{
+ enum roce_flavor flavor;
+
+ switch (roce_mode) {
+ case ROCE_V1:
+ flavor = PLAIN_ROCE;
+ break;
+ case ROCE_V2_IPV4:
+ flavor = RROCE_IPV4;
+ break;
+ case ROCE_V2_IPV6:
+ flavor = ROCE_V2_IPV6;
+ break;
+ default:
+ flavor = MAX_ROCE_MODE;
+ break;
+ }
+ return flavor;
+}
+
+int qed_roce_alloc_cid(struct qed_hwfn *p_hwfn, u16 *cid)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+ u32 responder_icid;
+ u32 requester_icid;
+ int rc;
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_rdma_info->cid_map,
+ &responder_icid);
+ if (rc) {
+ spin_unlock_bh(&p_rdma_info->lock);
+ return rc;
+ }
+
+ rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_rdma_info->cid_map,
+ &requester_icid);
+
+ spin_unlock_bh(&p_rdma_info->lock);
+ if (rc)
+ goto err;
+
+ /* the two icid's should be adjacent */
+ if ((requester_icid - responder_icid) != 1) {
+ DP_NOTICE(p_hwfn, "Failed to allocate two adjacent qp's'\n");
+ rc = -EINVAL;
+ goto err;
+ }
+
+ responder_icid += qed_cxt_get_proto_cid_start(p_hwfn,
+ p_rdma_info->proto);
+ requester_icid += qed_cxt_get_proto_cid_start(p_hwfn,
+ p_rdma_info->proto);
+
+ /* If these icids require a new ILT line allocate DMA-able context for
+ * an ILT page
+ */
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, responder_icid);
+ if (rc)
+ goto err;
+
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, requester_icid);
+ if (rc)
+ goto err;
+
+ *cid = (u16)responder_icid;
+ return rc;
+
+err:
+ spin_lock_bh(&p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_rdma_info->cid_map, responder_icid);
+ qed_bmap_release_id(p_hwfn, &p_rdma_info->cid_map, requester_icid);
+
+ spin_unlock_bh(&p_rdma_info->lock);
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Allocate CID - failed, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_roce_sp_create_responder(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp)
+{
+ struct roce_create_qp_resp_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ union qed_qm_pq_params qm_params;
+ enum roce_flavor roce_flavor;
+ struct qed_spq_entry *p_ent;
+ u16 physical_queue0 = 0;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ /* Allocate DMA-able memory for IRQ */
+ qp->irq_num_pages = 1;
+ qp->irq = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ RDMA_RING_PAGE_SIZE,
+ &qp->irq_phys_addr, GFP_KERNEL);
+ if (!qp->irq) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed create responder failed: cannot allocate memory (irq). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_CREATE_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.roce_create_qp_resp;
+
+ p_ramrod->flags = 0;
+
+ roce_flavor = qed_roce_mode_to_flavor(qp->roce_mode);
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_ROCE_FLAVOR, roce_flavor);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_RDMA_RD_EN,
+ qp->incoming_rdma_read_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_RDMA_WR_EN,
+ qp->incoming_rdma_write_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_ATOMIC_EN,
+ qp->incoming_atomic_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_E2E_FLOW_CONTROL_EN,
+ qp->e2e_flow_control_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_SRQ_FLG, qp->use_srq);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_RESERVED_KEY_EN,
+ qp->fmr_and_reserved_lkey);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER,
+ qp->min_rnr_nak_timer);
+
+ p_ramrod->max_ird = qp->max_rd_atomic_resp;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->irq_num_pages = qp->irq_num_pages;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->dst_qp_id = cpu_to_le32(qp->dest_qp);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ p_ramrod->initial_psn = cpu_to_le32(qp->rq_psn);
+ p_ramrod->pd = cpu_to_le16(qp->pd);
+ p_ramrod->rq_num_pages = cpu_to_le16(qp->rq_num_pages);
+ p_ramrod->rq_pbl_addr.hi = DMA_HI_LE(qp->rq_pbl_ptr);
+ p_ramrod->rq_pbl_addr.lo = DMA_LO_LE(qp->rq_pbl_ptr);
+ p_ramrod->irq_pbl_addr.hi = DMA_HI_LE(qp->irq_phys_addr);
+ p_ramrod->irq_pbl_addr.lo = DMA_LO_LE(qp->irq_phys_addr);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ p_ramrod->qp_handle_for_async.hi = cpu_to_le32(qp->qp_handle_async.hi);
+ p_ramrod->qp_handle_for_async.lo = cpu_to_le32(qp->qp_handle_async.lo);
+ p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
+ p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
+ p_ramrod->stats_counter_id = p_hwfn->rel_pf_id;
+ p_ramrod->cq_cid = cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) |
+ qp->rq_cq_id);
+
+ memset(&qm_params, 0, sizeof(qm_params));
+ qm_params.roce.qpid = (qp->icid >> 1);
+ physical_queue0 = qed_get_qm_pq(p_hwfn, PROTOCOLID_ROCE, &qm_params);
+
+ p_ramrod->physical_queue0 = cpu_to_le16(physical_queue0);
+ p_ramrod->dpi = cpu_to_le16(qp->dpi);
+
+ qed_rdma_set_fw_mac(p_ramrod->remote_mac_addr, qp->remote_mac_addr);
+ qed_rdma_set_fw_mac(p_ramrod->local_mac_addr, qp->local_mac_addr);
+
+ p_ramrod->udp_src_port = qp->udp_src_port;
+ p_ramrod->vlan_id = cpu_to_le16(qp->vlan_id);
+ p_ramrod->srq_id.srq_idx = cpu_to_le16(qp->srq_id);
+ p_ramrod->srq_id.opaque_fid = cpu_to_le16(p_hwfn->hw_info.opaque_fid);
+
+ p_ramrod->stats_counter_id = RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) +
+ qp->stats_queue;
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d physical_queue0 = 0x%x\n",
+ rc, physical_queue0);
+
+ if (rc)
+ goto err;
+
+ qp->resp_offloaded = true;
+
+ return rc;
+
+err:
+ DP_NOTICE(p_hwfn, "create responder - failed, rc = %d\n", rc);
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->irq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->irq, qp->irq_phys_addr);
+
+ return rc;
+}
+
+static int qed_roce_sp_create_requester(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp)
+{
+ struct roce_create_qp_req_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ union qed_qm_pq_params qm_params;
+ enum roce_flavor roce_flavor;
+ struct qed_spq_entry *p_ent;
+ u16 physical_queue0 = 0;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ /* Allocate DMA-able memory for ORQ */
+ qp->orq_num_pages = 1;
+ qp->orq = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ RDMA_RING_PAGE_SIZE,
+ &qp->orq_phys_addr, GFP_KERNEL);
+ if (!qp->orq) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed create requester failed: cannot allocate memory (orq). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid + 1;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_RAMROD_CREATE_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.roce_create_qp_req;
+
+ p_ramrod->flags = 0;
+
+ roce_flavor = qed_roce_mode_to_flavor(qp->roce_mode);
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_ROCE_FLAVOR, roce_flavor);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_FMR_AND_RESERVED_EN,
+ qp->fmr_and_reserved_lkey);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_SIGNALED_COMP, qp->signal_all);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT, qp->retry_cnt);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_RNR_NAK_CNT,
+ qp->rnr_retry_cnt);
+
+ p_ramrod->max_ord = qp->max_rd_atomic_req;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->orq_num_pages = qp->orq_num_pages;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->dst_qp_id = cpu_to_le32(qp->dest_qp);
+ p_ramrod->ack_timeout_val = cpu_to_le32(qp->ack_timeout);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ p_ramrod->initial_psn = cpu_to_le32(qp->sq_psn);
+ p_ramrod->pd = cpu_to_le16(qp->pd);
+ p_ramrod->sq_num_pages = cpu_to_le16(qp->sq_num_pages);
+ p_ramrod->sq_pbl_addr.hi = DMA_HI_LE(qp->sq_pbl_ptr);
+ p_ramrod->sq_pbl_addr.lo = DMA_LO_LE(qp->sq_pbl_ptr);
+ p_ramrod->orq_pbl_addr.hi = DMA_HI_LE(qp->orq_phys_addr);
+ p_ramrod->orq_pbl_addr.lo = DMA_LO_LE(qp->orq_phys_addr);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ p_ramrod->qp_handle_for_async.hi = cpu_to_le32(qp->qp_handle_async.hi);
+ p_ramrod->qp_handle_for_async.lo = cpu_to_le32(qp->qp_handle_async.lo);
+ p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
+ p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
+ p_ramrod->stats_counter_id = p_hwfn->rel_pf_id;
+ p_ramrod->cq_cid = cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) |
+ qp->sq_cq_id);
+
+ memset(&qm_params, 0, sizeof(qm_params));
+ qm_params.roce.qpid = (qp->icid >> 1);
+ physical_queue0 = qed_get_qm_pq(p_hwfn, PROTOCOLID_ROCE, &qm_params);
+
+ p_ramrod->physical_queue0 = cpu_to_le16(physical_queue0);
+ p_ramrod->dpi = cpu_to_le16(qp->dpi);
+
+ qed_rdma_set_fw_mac(p_ramrod->remote_mac_addr, qp->remote_mac_addr);
+ qed_rdma_set_fw_mac(p_ramrod->local_mac_addr, qp->local_mac_addr);
+
+ p_ramrod->udp_src_port = qp->udp_src_port;
+ p_ramrod->vlan_id = cpu_to_le16(qp->vlan_id);
+ p_ramrod->stats_counter_id = RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) +
+ qp->stats_queue;
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+
+ if (rc)
+ goto err;
+
+ qp->req_offloaded = true;
+
+ return rc;
+
+err:
+ DP_NOTICE(p_hwfn, "Create requested - failed, rc = %d\n", rc);
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->orq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->orq, qp->orq_phys_addr);
+ return rc;
+}
+
+static int qed_roce_sp_modify_responder(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ bool move_to_err, u32 modify_flags)
+{
+ struct roce_modify_qp_resp_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (move_to_err && !qp->resp_offloaded)
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_EVENT_MODIFY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "rc = %d\n", rc);
+ return rc;
+ }
+
+ p_ramrod = &p_ent->ramrod.roce_modify_qp_resp;
+
+ p_ramrod->flags = 0;
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MOVE_TO_ERR_FLG, move_to_err);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_RD_EN,
+ qp->incoming_rdma_read_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_WR_EN,
+ qp->incoming_rdma_write_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_ATOMIC_EN,
+ qp->incoming_atomic_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_E2E_FLOW_CONTROL_EN,
+ qp->e2e_flow_control_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_OPS_EN_FLG,
+ GET_FIELD(modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_P_KEY_FLG,
+ GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_ADDRESS_VECTOR_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MAX_IRD_FLG,
+ GET_FIELD(modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER));
+
+ p_ramrod->fields = 0;
+ SET_FIELD(p_ramrod->fields,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER,
+ qp->min_rnr_nak_timer);
+
+ p_ramrod->max_ird = qp->max_rd_atomic_resp;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify responder, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_roce_sp_modify_requester(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ bool move_to_sqd,
+ bool move_to_err, u32 modify_flags)
+{
+ struct roce_modify_qp_req_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (move_to_err && !(qp->req_offloaded))
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid + 1;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_EVENT_MODIFY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "rc = %d\n", rc);
+ return rc;
+ }
+
+ p_ramrod = &p_ent->ramrod.roce_modify_qp_req;
+
+ p_ramrod->flags = 0;
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_MOVE_TO_ERR_FLG, move_to_err);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_MOVE_TO_SQD_FLG, move_to_sqd);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_EN_SQD_ASYNC_NOTIFY,
+ qp->sqd_async);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_P_KEY_FLG,
+ GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ADDRESS_VECTOR_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_MAX_ORD_FLG,
+ GET_FIELD(modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_RNR_NAK_CNT_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT_FLG,
+ GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_RETRY_CNT));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ACK_TIMEOUT_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT));
+
+ p_ramrod->fields = 0;
+ SET_FIELD(p_ramrod->fields,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT, qp->retry_cnt);
+
+ SET_FIELD(p_ramrod->fields,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_RNR_NAK_CNT,
+ qp->rnr_retry_cnt);
+
+ p_ramrod->max_ord = qp->max_rd_atomic_req;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->ack_timeout_val = cpu_to_le32(qp->ack_timeout);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify requester, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_roce_sp_destroy_qp_responder(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ u32 *num_invalidated_mw)
+{
+ struct roce_destroy_qp_resp_output_params *p_ramrod_res;
+ struct roce_destroy_qp_resp_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (!qp->resp_offloaded)
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_RAMROD_DESTROY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.roce_destroy_qp_resp;
+
+ p_ramrod_res = (struct roce_destroy_qp_resp_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_ramrod_res),
+ &ramrod_res_phys, GFP_KERNEL);
+
+ if (!p_ramrod_res) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed destroy responder failed: cannot allocate memory (ramrod). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ DMA_REGPAIR_LE(p_ramrod->output_params_addr, ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err;
+
+ *num_invalidated_mw = le32_to_cpu(p_ramrod_res->num_invalidated_mw);
+
+ /* Free IRQ - only if ramrod succeeded, in case FW is still using it */
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->irq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->irq, qp->irq_phys_addr);
+
+ qp->resp_offloaded = false;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroy responder, rc = %d\n", rc);
+
+err:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct roce_destroy_qp_resp_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ return rc;
+}
+
+static int qed_roce_sp_destroy_qp_requester(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ u32 *num_bound_mw)
+{
+ struct roce_destroy_qp_req_output_params *p_ramrod_res;
+ struct roce_destroy_qp_req_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (!qp->req_offloaded)
+ return 0;
+
+ p_ramrod_res = (struct roce_destroy_qp_req_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(*p_ramrod_res),
+ &ramrod_res_phys, GFP_KERNEL);
+ if (!p_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed destroy requester failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid + 1;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_DESTROY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.roce_destroy_qp_req;
+ p_ramrod->output_params_addr.hi = DMA_HI_LE(ramrod_res_phys);
+ p_ramrod->output_params_addr.lo = DMA_LO_LE(ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err;
+
+ *num_bound_mw = le32_to_cpu(p_ramrod_res->num_bound_mw);
+
+ /* Free ORQ - only if ramrod succeeded, in case FW is still using it */
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->orq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->orq, qp->orq_phys_addr);
+
+ qp->req_offloaded = false;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroy requester, rc = %d\n", rc);
+
+err:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_ramrod_res),
+ p_ramrod_res, ramrod_res_phys);
+
+ return rc;
+}
+
+int qed_roce_query_qp(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *out_params)
+{
+ struct roce_query_qp_resp_output_params *p_resp_ramrod_res;
+ struct roce_query_qp_req_output_params *p_req_ramrod_res;
+ struct roce_query_qp_resp_ramrod_data *p_resp_ramrod;
+ struct roce_query_qp_req_ramrod_data *p_req_ramrod;
+ struct qed_sp_init_data init_data;
+ dma_addr_t resp_ramrod_res_phys;
+ dma_addr_t req_ramrod_res_phys;
+ struct qed_spq_entry *p_ent;
+ bool rq_err_state;
+ bool sq_err_state;
+ bool sq_draining;
+ int rc = -ENOMEM;
+
+ if ((!(qp->resp_offloaded)) && (!(qp->req_offloaded))) {
+ /* We can't send ramrod to the fw since this qp wasn't offloaded
+ * to the fw yet
+ */
+ out_params->draining = false;
+ out_params->rq_psn = qp->rq_psn;
+ out_params->sq_psn = qp->sq_psn;
+ out_params->state = qp->cur_state;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "No QPs as no offload\n");
+ return 0;
+ }
+
+ if (!(qp->resp_offloaded)) {
+ DP_NOTICE(p_hwfn,
+ "The responder's qp should be offloded before requester's\n");
+ return -EINVAL;
+ }
+
+ /* Send a query responder ramrod to FW to get RQ-PSN and state */
+ p_resp_ramrod_res = (struct roce_query_qp_resp_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(*p_resp_ramrod_res),
+ &resp_ramrod_res_phys, GFP_KERNEL);
+ if (!p_resp_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed query qp failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_QUERY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err_resp;
+
+ p_resp_ramrod = &p_ent->ramrod.roce_query_qp_resp;
+ p_resp_ramrod->output_params_addr.hi = DMA_HI_LE(resp_ramrod_res_phys);
+ p_resp_ramrod->output_params_addr.lo = DMA_LO_LE(resp_ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err_resp;
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_resp_ramrod_res),
+ p_resp_ramrod_res, resp_ramrod_res_phys);
+
+ out_params->rq_psn = le32_to_cpu(p_resp_ramrod_res->psn);
+ rq_err_state = GET_FIELD(le32_to_cpu(p_resp_ramrod_res->err_flag),
+ ROCE_QUERY_QP_RESP_OUTPUT_PARAMS_ERROR_FLG);
+
+ if (!(qp->req_offloaded)) {
+ /* Don't send query qp for the requester */
+ out_params->sq_psn = qp->sq_psn;
+ out_params->draining = false;
+
+ if (rq_err_state)
+ qp->cur_state = QED_ROCE_QP_STATE_ERR;
+
+ out_params->state = qp->cur_state;
+
+ return 0;
+ }
+
+ /* Send a query requester ramrod to FW to get SQ-PSN and state */
+ p_req_ramrod_res = (struct roce_query_qp_req_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(*p_req_ramrod_res),
+ &req_ramrod_res_phys,
+ GFP_KERNEL);
+ if (!p_req_ramrod_res) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed query qp failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ init_data.cid = qp->icid + 1;
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_QUERY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err_req;
+
+ p_req_ramrod = &p_ent->ramrod.roce_query_qp_req;
+ p_req_ramrod->output_params_addr.hi = DMA_HI_LE(req_ramrod_res_phys);
+ p_req_ramrod->output_params_addr.lo = DMA_LO_LE(req_ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err_req;
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_req_ramrod_res),
+ p_req_ramrod_res, req_ramrod_res_phys);
+
+ out_params->sq_psn = le32_to_cpu(p_req_ramrod_res->psn);
+ sq_err_state = GET_FIELD(le32_to_cpu(p_req_ramrod_res->flags),
+ ROCE_QUERY_QP_REQ_OUTPUT_PARAMS_ERR_FLG);
+ sq_draining =
+ GET_FIELD(le32_to_cpu(p_req_ramrod_res->flags),
+ ROCE_QUERY_QP_REQ_OUTPUT_PARAMS_SQ_DRAINING_FLG);
+
+ out_params->draining = false;
+
+ if (rq_err_state)
+ qp->cur_state = QED_ROCE_QP_STATE_ERR;
+ else if (sq_err_state)
+ qp->cur_state = QED_ROCE_QP_STATE_SQE;
+ else if (sq_draining)
+ out_params->draining = true;
+ out_params->state = qp->cur_state;
+
+ return 0;
+
+err_req:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_req_ramrod_res),
+ p_req_ramrod_res, req_ramrod_res_phys);
+ return rc;
+err_resp:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_resp_ramrod_res),
+ p_resp_ramrod_res, resp_ramrod_res_phys);
+ return rc;
+}
+
+int qed_roce_destroy_qp(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
+{
+ u32 num_invalidated_mw = 0;
+ u32 num_bound_mw = 0;
+ u32 start_cid;
+ int rc;
+
+ /* Destroys the specified QP */
+ if ((qp->cur_state != QED_ROCE_QP_STATE_RESET) &&
+ (qp->cur_state != QED_ROCE_QP_STATE_ERR) &&
+ (qp->cur_state != QED_ROCE_QP_STATE_INIT)) {
+ DP_NOTICE(p_hwfn,
+ "QP must be in error, reset or init state before destroying it\n");
+ return -EINVAL;
+ }
+
+ rc = qed_roce_sp_destroy_qp_responder(p_hwfn, qp, &num_invalidated_mw);
+ if (rc)
+ return rc;
+
+ /* Send destroy requester ramrod */
+ rc = qed_roce_sp_destroy_qp_requester(p_hwfn, qp, &num_bound_mw);
+ if (rc)
+ return rc;
+
+ if (num_invalidated_mw != num_bound_mw) {
+ DP_NOTICE(p_hwfn,
+ "number of invalidate memory windows is different from bounded ones\n");
+ return -EINVAL;
+ }
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+
+ start_cid = qed_cxt_get_proto_cid_start(p_hwfn,
+ p_hwfn->p_rdma_info->proto);
+
+ /* Release responder's icid */
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map,
+ qp->icid - start_cid);
+
+ /* Release requester's icid */
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map,
+ qp->icid + 1 - start_cid);
+
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ return 0;
+}
+
+int qed_rdma_query_qp(void *rdma_cxt,
+ struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ /* The following fields are filled in from qp and not FW as they can't
+ * be modified by FW
+ */
+ out_params->mtu = qp->mtu;
+ out_params->dest_qp = qp->dest_qp;
+ out_params->incoming_atomic_en = qp->incoming_atomic_en;
+ out_params->e2e_flow_control_en = qp->e2e_flow_control_en;
+ out_params->incoming_rdma_read_en = qp->incoming_rdma_read_en;
+ out_params->incoming_rdma_write_en = qp->incoming_rdma_write_en;
+ out_params->dgid = qp->dgid;
+ out_params->flow_label = qp->flow_label;
+ out_params->hop_limit_ttl = qp->hop_limit_ttl;
+ out_params->traffic_class_tos = qp->traffic_class_tos;
+ out_params->timeout = qp->ack_timeout;
+ out_params->rnr_retry = qp->rnr_retry_cnt;
+ out_params->retry_cnt = qp->retry_cnt;
+ out_params->min_rnr_nak_timer = qp->min_rnr_nak_timer;
+ out_params->pkey_index = 0;
+ out_params->max_rd_atomic = qp->max_rd_atomic_req;
+ out_params->max_dest_rd_atomic = qp->max_rd_atomic_resp;
+ out_params->sqd_async = qp->sqd_async;
+
+ rc = qed_roce_query_qp(p_hwfn, qp, out_params);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Query QP, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_destroy_qp(void *rdma_cxt, struct qed_rdma_qp *qp)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ int rc = 0;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ rc = qed_roce_destroy_qp(p_hwfn, qp);
+
+ /* free qp params struct */
+ kfree(qp);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP destroyed\n");
+ return rc;
+}
+
+struct qed_rdma_qp *
+qed_rdma_create_qp(void *rdma_cxt,
+ struct qed_rdma_create_qp_in_params *in_params,
+ struct qed_rdma_create_qp_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_rdma_qp *qp;
+ u8 max_stats_queues;
+ int rc;
+
+ if (!rdma_cxt || !in_params || !out_params || !p_hwfn->p_rdma_info) {
+ DP_ERR(p_hwfn->cdev,
+ "qed roce create qp failed due to NULL entry (rdma_cxt=%p, in=%p, out=%p, roce_info=?\n",
+ rdma_cxt, in_params, out_params);
+ return NULL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "qed rdma create qp called with qp_handle = %08x%08x\n",
+ in_params->qp_handle_hi, in_params->qp_handle_lo);
+
+ /* Some sanity checks... */
+ max_stats_queues = p_hwfn->p_rdma_info->dev->max_stats_queues;
+ if (in_params->stats_queue >= max_stats_queues) {
+ DP_ERR(p_hwfn->cdev,
+ "qed rdma create qp failed due to invalid statistics queue %d. maximum is %d\n",
+ in_params->stats_queue, max_stats_queues);
+ return NULL;
+ }
+
+ qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+ if (!qp) {
+ DP_NOTICE(p_hwfn, "Failed to allocate qed_rdma_qp\n");
+ return NULL;
+ }
+
+ rc = qed_roce_alloc_cid(p_hwfn, &qp->icid);
+ qp->qpid = ((0xFF << 16) | qp->icid);
+
+ DP_INFO(p_hwfn, "ROCE qpid=%x\n", qp->qpid);
+
+ if (rc) {
+ kfree(qp);
+ return NULL;
+ }
+
+ qp->cur_state = QED_ROCE_QP_STATE_RESET;
+ qp->qp_handle.hi = cpu_to_le32(in_params->qp_handle_hi);
+ qp->qp_handle.lo = cpu_to_le32(in_params->qp_handle_lo);
+ qp->qp_handle_async.hi = cpu_to_le32(in_params->qp_handle_async_hi);
+ qp->qp_handle_async.lo = cpu_to_le32(in_params->qp_handle_async_lo);
+ qp->use_srq = in_params->use_srq;
+ qp->signal_all = in_params->signal_all;
+ qp->fmr_and_reserved_lkey = in_params->fmr_and_reserved_lkey;
+ qp->pd = in_params->pd;
+ qp->dpi = in_params->dpi;
+ qp->sq_cq_id = in_params->sq_cq_id;
+ qp->sq_num_pages = in_params->sq_num_pages;
+ qp->sq_pbl_ptr = in_params->sq_pbl_ptr;
+ qp->rq_cq_id = in_params->rq_cq_id;
+ qp->rq_num_pages = in_params->rq_num_pages;
+ qp->rq_pbl_ptr = in_params->rq_pbl_ptr;
+ qp->srq_id = in_params->srq_id;
+ qp->req_offloaded = false;
+ qp->resp_offloaded = false;
+ /* e2e_flow_control cannot be done in case of S-RQ.
+ * Refer to 9.7.7.2 End-to-End Flow Control section of IB spec
+ */
+ qp->e2e_flow_control_en = qp->use_srq ? false : true;
+ qp->stats_queue = in_params->stats_queue;
+
+ out_params->icid = qp->icid;
+ out_params->qp_id = qp->qpid;
+
+ /* max_sq_sges future use only */
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Create QP, rc = %d\n", rc);
+ return qp;
+}
+
+static int qed_roce_modify_qp(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ enum qed_roce_qp_state prev_state,
+ struct qed_rdma_modify_qp_in_params *params)
+{
+ u32 num_invalidated_mw = 0, num_bound_mw = 0;
+ int rc = 0;
+
+ /* Perform additional operations according to the current state and the
+ * next state
+ */
+ if (((prev_state == QED_ROCE_QP_STATE_INIT) ||
+ (prev_state == QED_ROCE_QP_STATE_RESET)) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTR)) {
+ /* Init->RTR or Reset->RTR */
+ rc = qed_roce_sp_create_responder(p_hwfn, qp);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_RTR) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
+ /* RTR-> RTS */
+ rc = qed_roce_sp_create_requester(p_hwfn, qp);
+ if (rc)
+ return rc;
+
+ /* Send modify responder ramrod */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_RTS) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
+ /* RTS->RTS */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_RTS) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_SQD)) {
+ /* RTS->SQD */
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, true, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_SQD) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_SQD)) {
+ /* SQD->SQD */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_SQD) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
+ /* SQD->RTS */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
+ params->modify_flags);
+
+ return rc;
+ } else if (qp->cur_state == QED_ROCE_QP_STATE_ERR ||
+ qp->cur_state == QED_ROCE_QP_STATE_SQE) {
+ /* ->ERR */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, true,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, true,
+ params->modify_flags);
+ return rc;
+ } else if (qp->cur_state == QED_ROCE_QP_STATE_RESET) {
+ /* Any state -> RESET */
+
+ rc = qed_roce_sp_destroy_qp_responder(p_hwfn, qp,
+ &num_invalidated_mw);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_destroy_qp_requester(p_hwfn, qp,
+ &num_bound_mw);
+
+ if (num_invalidated_mw != num_bound_mw) {
+ DP_NOTICE(p_hwfn,
+ "number of invalidate memory windows is different from bounded ones\n");
+ return -EINVAL;
+ }
+ } else {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "0\n");
+ }
+
+ return rc;
+}
+
+int qed_rdma_modify_qp(void *rdma_cxt,
+ struct qed_rdma_qp *qp,
+ struct qed_rdma_modify_qp_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ enum qed_roce_qp_state prev_state;
+ int rc = 0;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x params->new_state=%d\n",
+ qp->icid, params->new_state);
+
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN)) {
+ qp->incoming_rdma_read_en = params->incoming_rdma_read_en;
+ qp->incoming_rdma_write_en = params->incoming_rdma_write_en;
+ qp->incoming_atomic_en = params->incoming_atomic_en;
+ }
+
+ /* Update QP structure with the updated values */
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_ROCE_MODE))
+ qp->roce_mode = params->roce_mode;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY))
+ qp->pkey = params->pkey;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_E2E_FLOW_CONTROL_EN))
+ qp->e2e_flow_control_en = params->e2e_flow_control_en;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_DEST_QP))
+ qp->dest_qp = params->dest_qp;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR)) {
+ /* Indicates that the following parameters have changed:
+ * Traffic class, flow label, hop limit, source GID,
+ * destination GID, loopback indicator
+ */
+ qp->traffic_class_tos = params->traffic_class_tos;
+ qp->flow_label = params->flow_label;
+ qp->hop_limit_ttl = params->hop_limit_ttl;
+
+ qp->sgid = params->sgid;
+ qp->dgid = params->dgid;
+ qp->udp_src_port = 0;
+ qp->vlan_id = params->vlan_id;
+ qp->mtu = params->mtu;
+ qp->lb_indication = params->lb_indication;
+ memcpy((u8 *)&qp->remote_mac_addr[0],
+ (u8 *)¶ms->remote_mac_addr[0], ETH_ALEN);
+ if (params->use_local_mac) {
+ memcpy((u8 *)&qp->local_mac_addr[0],
+ (u8 *)¶ms->local_mac_addr[0], ETH_ALEN);
+ } else {
+ memcpy((u8 *)&qp->local_mac_addr[0],
+ (u8 *)&p_hwfn->hw_info.hw_mac_addr, ETH_ALEN);
+ }
+ }
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_RQ_PSN))
+ qp->rq_psn = params->rq_psn;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_SQ_PSN))
+ qp->sq_psn = params->sq_psn;
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ))
+ qp->max_rd_atomic_req = params->max_rd_atomic_req;
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP))
+ qp->max_rd_atomic_resp = params->max_rd_atomic_resp;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT))
+ qp->ack_timeout = params->ack_timeout;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_RETRY_CNT))
+ qp->retry_cnt = params->retry_cnt;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT))
+ qp->rnr_retry_cnt = params->rnr_retry_cnt;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER))
+ qp->min_rnr_nak_timer = params->min_rnr_nak_timer;
+
+ qp->sqd_async = params->sqd_async;
+
+ prev_state = qp->cur_state;
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_NEW_STATE)) {
+ qp->cur_state = params->new_state;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "qp->cur_state=%d\n",
+ qp->cur_state);
+ }
+
+ rc = qed_roce_modify_qp(p_hwfn, qp, prev_state, params);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify QP, rc = %d\n", rc);
+ return rc;
+}
+
static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
{
return QED_LEADING_HWFN(cdev);
@@ -1208,6 +2415,10 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_dealloc_pd = &qed_rdma_free_pd,
.rdma_create_cq = &qed_rdma_create_cq,
.rdma_destroy_cq = &qed_rdma_destroy_cq,
+ .rdma_create_qp = &qed_rdma_create_qp,
+ .rdma_modify_qp = &qed_rdma_modify_qp,
+ .rdma_query_qp = &qed_rdma_query_qp,
+ .rdma_destroy_qp = &qed_rdma_destroy_qp,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 6802cd0..71b37fc 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -114,6 +114,72 @@ struct qed_rdma_resize_cnq_in_params {
u64 pbl_ptr;
};
+struct qed_rdma_qp {
+ struct regpair qp_handle;
+ struct regpair qp_handle_async;
+ u32 qpid;
+ u16 icid;
+ enum qed_roce_qp_state cur_state;
+ bool use_srq;
+ bool signal_all;
+ bool fmr_and_reserved_lkey;
+
+ bool incoming_rdma_read_en;
+ bool incoming_rdma_write_en;
+ bool incoming_atomic_en;
+ bool e2e_flow_control_en;
+
+ u16 pd;
+ u16 pkey;
+ u32 dest_qp;
+ u16 mtu;
+ u16 srq_id;
+ u8 traffic_class_tos;
+ u8 hop_limit_ttl;
+ u16 dpi;
+ u32 flow_label;
+ bool lb_indication;
+ u16 vlan_id;
+ u32 ack_timeout;
+ u8 retry_cnt;
+ u8 rnr_retry_cnt;
+ u8 min_rnr_nak_timer;
+ bool sqd_async;
+ union qed_gid sgid;
+ union qed_gid dgid;
+ enum roce_mode roce_mode;
+ u16 udp_src_port;
+ u8 stats_queue;
+
+ /* requeseter */
+ u8 max_rd_atomic_req;
+ u32 sq_psn;
+ u16 sq_cq_id;
+ u16 sq_num_pages;
+ dma_addr_t sq_pbl_ptr;
+ void *orq;
+ dma_addr_t orq_phys_addr;
+ u8 orq_num_pages;
+ bool req_offloaded;
+
+ /* responder */
+ u8 max_rd_atomic_resp;
+ u32 rq_psn;
+ u16 rq_cq_id;
+ u16 rq_num_pages;
+ dma_addr_t rq_pbl_ptr;
+ void *irq;
+ dma_addr_t irq_phys_addr;
+ u8 irq_num_pages;
+ bool resp_offloaded;
+
+ u8 remote_mac_addr[6];
+ u8 local_mac_addr[6];
+
+ void *shared_queue;
+ dma_addr_t shared_queue_phys_addr;
+};
+
int
qed_rdma_add_user(void *rdma_cxt,
struct qed_rdma_add_user_out_params *out_params);
@@ -135,4 +201,9 @@ void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 cnq_index, u16 prod);
void qed_rdma_resc_free(struct qed_hwfn *p_hwfn);
void qed_async_roce_event(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe);
+int qed_rdma_destroy_qp(void *rdma_cxt, struct qed_rdma_qp *qp);
+int qed_rdma_modify_qp(void *rdma_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_modify_qp_in_params *params);
+int qed_rdma_query_qp(void *rdma_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *out_params);
#endif
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index b559b1c..02321e3 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -43,6 +43,17 @@
#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
/* rdma interface */
+
+enum qed_roce_qp_state {
+ QED_ROCE_QP_STATE_RESET,
+ QED_ROCE_QP_STATE_INIT,
+ QED_ROCE_QP_STATE_RTR,
+ QED_ROCE_QP_STATE_RTS,
+ QED_ROCE_QP_STATE_SQD,
+ QED_ROCE_QP_STATE_ERR,
+ QED_ROCE_QP_STATE_SQE
+};
+
enum qed_rdma_tid_type {
QED_RDMA_TID_REGISTERED_MR,
QED_RDMA_TID_FMR,
@@ -292,6 +303,128 @@ struct qed_rdma_destroy_cq_out_params {
u16 num_cq_notif;
};
+struct qed_rdma_create_qp_in_params {
+ u32 qp_handle_lo;
+ u32 qp_handle_hi;
+ u32 qp_handle_async_lo;
+ u32 qp_handle_async_hi;
+ bool use_srq;
+ bool signal_all;
+ bool fmr_and_reserved_lkey;
+ u16 pd;
+ u16 dpi;
+ u16 sq_cq_id;
+ u16 sq_num_pages;
+ u64 sq_pbl_ptr;
+ u8 max_sq_sges;
+ u16 rq_cq_id;
+ u16 rq_num_pages;
+ u64 rq_pbl_ptr;
+ u16 srq_id;
+ u8 stats_queue;
+};
+
+struct qed_rdma_create_qp_out_params {
+ u32 qp_id;
+ u16 icid;
+ void *rq_pbl_virt;
+ dma_addr_t rq_pbl_phys;
+ void *sq_pbl_virt;
+ dma_addr_t sq_pbl_phys;
+};
+
+struct qed_rdma_modify_qp_in_params {
+ u32 modify_flags;
+#define QED_RDMA_MODIFY_QP_VALID_NEW_STATE_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_NEW_STATE_SHIFT 0
+#define QED_ROCE_MODIFY_QP_VALID_PKEY_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_PKEY_SHIFT 1
+#define QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN_SHIFT 2
+#define QED_ROCE_MODIFY_QP_VALID_DEST_QP_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_DEST_QP_SHIFT 3
+#define QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR_SHIFT 4
+#define QED_ROCE_MODIFY_QP_VALID_RQ_PSN_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_RQ_PSN_SHIFT 5
+#define QED_ROCE_MODIFY_QP_VALID_SQ_PSN_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_SQ_PSN_SHIFT 6
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ_SHIFT 7
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP_SHIFT 8
+#define QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT_SHIFT 9
+#define QED_ROCE_MODIFY_QP_VALID_RETRY_CNT_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_RETRY_CNT_SHIFT 10
+#define QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT_SHIFT 11
+#define QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER_SHIFT 12
+#define QED_ROCE_MODIFY_QP_VALID_E2E_FLOW_CONTROL_EN_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_E2E_FLOW_CONTROL_EN_SHIFT 13
+#define QED_ROCE_MODIFY_QP_VALID_ROCE_MODE_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_ROCE_MODE_SHIFT 14
+
+ enum qed_roce_qp_state new_state;
+ u16 pkey;
+ bool incoming_rdma_read_en;
+ bool incoming_rdma_write_en;
+ bool incoming_atomic_en;
+ bool e2e_flow_control_en;
+ u32 dest_qp;
+ bool lb_indication;
+ u16 mtu;
+ u8 traffic_class_tos;
+ u8 hop_limit_ttl;
+ u32 flow_label;
+ union qed_gid sgid;
+ union qed_gid dgid;
+ u16 udp_src_port;
+
+ u16 vlan_id;
+
+ u32 rq_psn;
+ u32 sq_psn;
+ u8 max_rd_atomic_resp;
+ u8 max_rd_atomic_req;
+ u32 ack_timeout;
+ u8 retry_cnt;
+ u8 rnr_retry_cnt;
+ u8 min_rnr_nak_timer;
+ bool sqd_async;
+ u8 remote_mac_addr[6];
+ u8 local_mac_addr[6];
+ bool use_local_mac;
+ enum roce_mode roce_mode;
+};
+
+struct qed_rdma_query_qp_out_params {
+ enum qed_roce_qp_state state;
+ u32 rq_psn;
+ u32 sq_psn;
+ bool draining;
+ u16 mtu;
+ u32 dest_qp;
+ bool incoming_rdma_read_en;
+ bool incoming_rdma_write_en;
+ bool incoming_atomic_en;
+ bool e2e_flow_control_en;
+ union qed_gid sgid;
+ union qed_gid dgid;
+ u32 flow_label;
+ u8 hop_limit_ttl;
+ u8 traffic_class_tos;
+ u32 timeout;
+ u8 rnr_retry;
+ u8 retry_cnt;
+ u8 min_rnr_nak_timer;
+ u16 pkey_index;
+ u8 max_rd_atomic;
+ u8 max_dest_rd_atomic;
+ bool sqd_async;
+};
+
struct qed_rdma_create_srq_out_params {
u16 srq_id;
};
@@ -368,6 +501,17 @@ struct qed_rdma_ops {
int (*rdma_destroy_cq)(void *rdma_cxt,
struct qed_rdma_destroy_cq_in_params *iparams,
struct qed_rdma_destroy_cq_out_params *oparams);
+ struct qed_rdma_qp *
+ (*rdma_create_qp)(void *rdma_cxt,
+ struct qed_rdma_create_qp_in_params *iparams,
+ struct qed_rdma_create_qp_out_params *oparams);
+
+ int (*rdma_modify_qp)(void *roce_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_modify_qp_in_params *iparams);
+
+ int (*rdma_query_qp)(void *rdma_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *oparams);
+ int (*rdma_destroy_qp)(void *rdma_cxt, struct qed_rdma_qp *qp);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 06/11] Add support for QP verbs
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for Queue Pair verbs which adds, deletes,
modifies and queries Queue Pairs.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 15 +-
drivers/infiniband/hw/qedr/qedr.h | 126 +++
drivers/infiniband/hw/qedr/qedr_cm.h | 40 +
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 11 +
drivers/infiniband/hw/qedr/qedr_user.h | 34 +
drivers/infiniband/hw/qedr/verbs.c | 1098 ++++++++++++++++++++++++-
drivers/infiniband/hw/qedr/verbs.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_cxt.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 1211 ++++++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 71 ++
include/linux/qed/qed_roce_if.h | 144 ++++
11 files changed, 2756 insertions(+), 2 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.h
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 02f0dbb..722b895 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -52,6 +52,8 @@ uint debug;
module_param(debug, uint, 0);
MODULE_PARM_DESC(debug, "Default debug msglevel");
+#define QEDR_WQ_MULTIPLIER_DFT (3)
+
static LIST_HEAD(qedr_dev_list);
static DEFINE_SPINLOCK(qedr_devlist_lock);
@@ -100,7 +102,11 @@ static int qedr_register_device(struct qedr_dev *dev)
QEDR_UVERBS(CREATE_CQ) |
QEDR_UVERBS(RESIZE_CQ) |
QEDR_UVERBS(DESTROY_CQ) |
- QEDR_UVERBS(REQ_NOTIFY_CQ);
+ QEDR_UVERBS(REQ_NOTIFY_CQ) |
+ QEDR_UVERBS(CREATE_QP) |
+ QEDR_UVERBS(MODIFY_QP) |
+ QEDR_UVERBS(QUERY_QP) |
+ QEDR_UVERBS(DESTROY_QP);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -125,6 +131,11 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.resize_cq = qedr_resize_cq;
dev->ibdev.req_notify_cq = qedr_arm_cq;
+ dev->ibdev.create_qp = qedr_create_qp;
+ dev->ibdev.modify_qp = qedr_modify_qp;
+ dev->ibdev.query_qp = qedr_query_qp;
+ dev->ibdev.destroy_qp = qedr_destroy_qp;
+
dev->ibdev.query_pkey = qedr_query_pkey;
dev->ibdev.dma_device = &dev->pdev->dev;
@@ -622,6 +633,8 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
goto init_err;
}
+ dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
+
qedr_pci_set_atomic(dev, pdev);
rc = qedr_alloc_resources(dev);
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index b2aa8fd..1c9cc40 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -49,6 +49,9 @@
enum DP_QEDR_MODULE {
QEDR_MSG_INIT = 0x10000,
QEDR_MSG_CQ = 0x20000,
+ QEDR_MSG_RQ = 0x40000,
+ QEDR_MSG_SQ = 0x80000,
+ QEDR_MSG_QP = (QEDR_MSG_SQ | QEDR_MSG_RQ),
QEDR_MSG_MR = 0x100000,
QEDR_MSG_MISC = 0x400000,
};
@@ -142,6 +145,8 @@ struct qedr_dev {
u32 dp_module;
u8 dp_level;
u8 num_hwfns;
+ uint wq_multiplier;
+
};
#define QEDR_MAX_SQ_PBL (0x8000)
@@ -271,6 +276,122 @@ struct qedr_mm {
struct list_head entry;
};
+union db_prod32 {
+ struct rdma_pwm_val16_data data;
+ u32 raw;
+};
+
+struct qedr_qp_hwq_info {
+ /* WQE Elements */
+ struct qed_chain pbl;
+ u64 p_phys_addr_tbl;
+ u32 max_sges;
+
+ /* WQE */
+ u16 prod;
+ u16 cons;
+ u16 wqe_cons;
+ u16 max_wr;
+
+ /* DB */
+ void __iomem *db;
+ union db_prod32 db_data;
+};
+
+#define QEDR_INC_SW_IDX(p_info, index) \
+ do { \
+ p_info->index = (p_info->index + 1) & \
+ qed_chain_get_capacity(p_info->pbl) \
+ } while (0)
+
+enum qedr_qp_err_bitmap {
+ QEDR_QP_ERR_SQ_FULL = 1,
+ QEDR_QP_ERR_RQ_FULL = 2,
+ QEDR_QP_ERR_BAD_SR = 4,
+ QEDR_QP_ERR_BAD_RR = 8,
+ QEDR_QP_ERR_SQ_PBL_FULL = 16,
+ QEDR_QP_ERR_RQ_PBL_FULL = 32,
+};
+
+struct qedr_qp {
+ struct ib_qp ibqp; /* must be first */
+ struct qedr_dev *dev;
+
+ struct qedr_qp_hwq_info sq;
+ struct qedr_qp_hwq_info rq;
+
+ u32 max_inline_data;
+
+ /* Lock for QP's */
+ spinlock_t q_lock;
+ struct qedr_cq *sq_cq;
+ struct qedr_cq *rq_cq;
+ struct qedr_srq *srq;
+ enum qed_roce_qp_state state;
+ u32 id;
+ struct qedr_pd *pd;
+ enum ib_qp_type qp_type;
+ struct qed_rdma_qp *qed_qp;
+ u32 qp_id;
+ u16 icid;
+ u16 mtu;
+ int sgid_idx;
+ u32 rq_psn;
+ u32 sq_psn;
+ u32 qkey;
+ u32 dest_qp_num;
+
+ /* Relevant to qps created from kernel space only (ULPs) */
+ u8 prev_wqe_size;
+ u16 wqe_cons;
+ u32 err_bitmap;
+ bool signaled;
+
+ /* SQ shadow */
+ struct {
+ u64 wr_id;
+ enum ib_wc_opcode opcode;
+ u32 bytes_len;
+ u8 wqe_size;
+ bool signaled;
+ dma_addr_t icrc_mapping;
+ u32 *icrc;
+ struct qedr_mr *mr;
+ } *wqe_wr_id;
+
+ /* RQ shadow */
+ struct {
+ u64 wr_id;
+ struct ib_sge sg_list[RDMA_MAX_SGE_PER_RQ_WQE];
+ u8 wqe_size;
+
+ u16 vlan_id;
+ int rc;
+ } *rqe_wr_id;
+
+ /* Relevant to qps created from user space only (applications) */
+ struct qedr_userq usq;
+ struct qedr_userq urq;
+};
+
+static inline int qedr_get_dmac(struct qedr_dev *dev,
+ struct ib_ah_attr *ah_attr, u8 *mac_addr)
+{
+ union ib_gid zero_sgid = { { 0 } };
+ struct in6_addr in6;
+
+ if (!memcmp(&ah_attr->grh.dgid, &zero_sgid, sizeof(union ib_gid))) {
+ DP_ERR(dev, "Local port GID not supported\n");
+ eth_zero_addr(mac_addr);
+ return -EINVAL;
+ }
+
+ memcpy(&in6, ah_attr->grh.dgid.raw, sizeof(in6));
+ ether_addr_copy(mac_addr, ah_attr->dmac);
+
+ return 0;
+}
+
static inline struct qedr_dev *get_qedr_dev(struct ib_device *ibdev)
{
return container_of(ibdev, struct qedr_dev, ibdev);
@@ -291,4 +412,9 @@ static inline struct qedr_cq *get_qedr_cq(struct ib_cq *ibcq)
{
return container_of(ibcq, struct qedr_cq, ibcq);
}
+
+static inline struct qedr_qp *get_qedr_qp(struct ib_qp *ibqp)
+{
+ return container_of(ibqp, struct qedr_qp, ibqp);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.h b/drivers/infiniband/hw/qedr/qedr_cm.h
new file mode 100644
index 0000000..b8a8b76
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_cm.h
@@ -0,0 +1,40 @@
+/* QLogic qedr NIC Driver
+ * Copyright (c) 2015-2016 QLogic Corporation
+ *
+ * This software is available to you under a choice of one of two
+ * licenses. You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ * Redistribution and use in source and binary forms, with or
+ * without modification, are permitted provided that the following
+ * conditions are met:
+ *
+ * - Redistributions of source code must retain the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer.
+ *
+ * - Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and /or other materials
+ * provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef LINUX_QEDR_CM_H_
+#define LINUX_QEDR_CM_H_
+
+static inline u32 qedr_get_ipv4_from_gid(u8 *gid)
+{
+ return *(u32 *)(void *)&gid[12];
+}
+
+#endif
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
index 84f6520..4770559 100644
--- a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -158,6 +158,17 @@ struct rdma_srq_sge {
__le32 l_key;
};
+/* Rdma doorbell data for SQ and RQ */
+struct rdma_pwm_val16_data {
+ __le16 icid;
+ __le16 value;
+};
+
+union rdma_pwm_val16_data_union {
+ struct rdma_pwm_val16_data as_struct;
+ __le32 as_dword;
+};
+
/* Rdma doorbell data for CQ */
struct rdma_pwm_val32_data {
__le16 icid;
diff --git a/drivers/infiniband/hw/qedr/qedr_user.h b/drivers/infiniband/hw/qedr/qedr_user.h
index 5832c30..40598bf 100644
--- a/drivers/infiniband/hw/qedr/qedr_user.h
+++ b/drivers/infiniband/hw/qedr/qedr_user.h
@@ -43,4 +43,38 @@ struct qedr_create_cq_uresp {
u16 icid;
};
+struct qedr_create_qp_ureq {
+ u32 qp_handle_hi;
+ u32 qp_handle_lo;
+
+ /* SQ */
+ /* user space virtual address of SQ buffer */
+ u64 sq_addr;
+
+ /* length of SQ buffer */
+ size_t sq_len;
+
+ /* RQ */
+ /* user space virtual address of RQ buffer */
+ u64 rq_addr;
+
+ /* length of RQ buffer */
+ size_t rq_len;
+};
+
+struct qedr_create_qp_uresp {
+ u32 qp_id;
+ int atomic_supported;
+
+ /* SQ */
+ u32 sq_db_offset;
+ u16 sq_icid;
+
+ /* RQ */
+ u32 rq_db_offset;
+ u16 rq_icid;
+
+ u32 rq_db2_offset;
+};
+
#endif /* __QEDR_USER_H__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 2c26957..1d16c34 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -24,6 +24,7 @@
#include "qedr.h"
#include "verbs.h"
#include "qedr_user.h"
+#include "qedr_cm.h"
#define DB_ADDR_SHIFT(addr) ((addr) << DB_PWM_ADDR_OFFSET_SHIFT)
@@ -69,7 +70,6 @@ int qedr_add_gid(struct ib_device *device, u8 port_num,
if (!context)
return -EINVAL;
-
return 0;
}
@@ -958,3 +958,1099 @@ int qedr_destroy_cq(struct ib_cq *ibcq)
return 0;
}
+
+static inline int get_gid_info_from_table(struct ib_qp *ibqp,
+ struct ib_qp_attr *attr,
+ int attr_mask,
+ struct qed_rdma_modify_qp_in_params
+ *qp_params)
+{
+ enum rdma_network_type nw_type;
+ struct ib_gid_attr gid_attr;
+ union ib_gid gid;
+ u32 ipv4_addr;
+ int rc = 0;
+ int i;
+
+ rc = ib_get_cached_gid(ibqp->device, attr->ah_attr.port_num,
+ attr->ah_attr.grh.sgid_index, &gid, &gid_attr);
+ if (!rc && !memcmp(&gid, &zgid, sizeof(gid)))
+ rc = -ENOENT;
+
+ if (!rc && gid_attr.ndev) {
+ qp_params->vlan_id = rdma_vlan_dev_vlan_id(gid_attr.ndev);
+
+ dev_put(gid_attr.ndev);
+ nw_type = ib_gid_to_network_type(gid_attr.gid_type, &gid);
+ switch (nw_type) {
+ case RDMA_NETWORK_IPV6:
+ memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
+ sizeof(qp_params->sgid));
+ memcpy(&qp_params->dgid.bytes[0],
+ &attr->ah_attr.grh.dgid,
+ sizeof(qp_params->dgid));
+ qp_params->roce_mode = ROCE_V2_IPV6;
+ SET_FIELD(qp_params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1);
+ break;
+ case RDMA_NETWORK_IB:
+ memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
+ sizeof(qp_params->sgid));
+ memcpy(&qp_params->dgid.bytes[0],
+ &attr->ah_attr.grh.dgid,
+ sizeof(qp_params->dgid));
+ qp_params->roce_mode = ROCE_V1;
+ break;
+ case RDMA_NETWORK_IPV4:
+ memset(&qp_params->sgid, 0, sizeof(qp_params->sgid));
+ memset(&qp_params->dgid, 0, sizeof(qp_params->dgid));
+ ipv4_addr = qedr_get_ipv4_from_gid(gid.raw);
+ qp_params->sgid.ipv4_addr = ipv4_addr;
+ ipv4_addr =
+ qedr_get_ipv4_from_gid(attr->ah_attr.grh.dgid.raw);
+ qp_params->dgid.ipv4_addr = ipv4_addr;
+ SET_FIELD(qp_params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1);
+ qp_params->roce_mode = ROCE_V2_IPV4;
+ break;
+ }
+ }
+ if (rc)
+ return -EINVAL;
+
+ for (i = 0; i < 4; i++) {
+ qp_params->sgid.dwords[i] = ntohl(qp_params->sgid.dwords[i]);
+ qp_params->dgid.dwords[i] = ntohl(qp_params->dgid.dwords[i]);
+ }
+
+ if (qp_params->vlan_id >= VLAN_CFI_MASK)
+ qp_params->vlan_id = 0;
+
+ return rc;
+}
+
+static void qedr_cleanup_user_sq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qedr_free_pbl(dev, &qp->usq.pbl_info, qp->usq.pbl_tbl);
+ ib_umem_release(qp->usq.umem);
+}
+
+static void qedr_cleanup_user_rq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qedr_free_pbl(dev, &qp->urq.pbl_info, qp->urq.pbl_tbl);
+ ib_umem_release(qp->urq.umem);
+}
+
+static void qedr_cleanup_kernel_sq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ dev->ops->common->chain_free(dev->cdev, &qp->sq.pbl);
+ kfree(qp->wqe_wr_id);
+}
+
+static void qedr_cleanup_kernel_rq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ dev->ops->common->chain_free(dev->cdev, &qp->rq.pbl);
+ kfree(qp->rqe_wr_id);
+}
+
+static int qedr_check_qp_attrs(struct ib_pd *ibpd, struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ struct qedr_device_attr *qattr = &dev->attr;
+
+ /* QP0... attrs->qp_type == IB_QPT_GSI */
+ if (attrs->qp_type != IB_QPT_RC && attrs->qp_type != IB_QPT_GSI) {
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "create qp: unsupported qp type=0x%x requested\n",
+ attrs->qp_type);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_send_wr > qattr->max_sqe) {
+ DP_ERR(dev,
+ "create qp: cannot create a SQ with %d elements (max_send_wr=0x%x)\n",
+ attrs->cap.max_send_wr, qattr->max_sqe);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_inline_data > qattr->max_inline) {
+ DP_ERR(dev,
+ "create qp: unsupported inline data size=0x%x requested (max_inline=0x%x)\n",
+ attrs->cap.max_inline_data, qattr->max_inline);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_send_sge > qattr->max_sge) {
+ DP_ERR(dev,
+ "create qp: unsupported send_sge=0x%x requested (max_send_sge=0x%x)\n",
+ attrs->cap.max_send_sge, qattr->max_sge);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_recv_sge > qattr->max_sge) {
+ DP_ERR(dev,
+ "create qp: unsupported recv_sge=0x%x requested (max_recv_sge=0x%x)\n",
+ attrs->cap.max_recv_sge, qattr->max_sge);
+ return -EINVAL;
+ }
+
+ /* Unprivileged user space cannot create special QP */
+ if (ibpd->uobject && attrs->qp_type == IB_QPT_GSI) {
+ DP_ERR(dev,
+ "create qp: userspace can't create special QPs of type=0x%x\n",
+ attrs->qp_type);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static void qedr_copy_rq_uresp(struct qedr_create_qp_uresp *uresp,
+ struct qedr_qp *qp)
+{
+ uresp->rq_db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
+ uresp->rq_icid = qp->icid;
+}
+
+static void qedr_copy_sq_uresp(struct qedr_create_qp_uresp *uresp,
+ struct qedr_qp *qp)
+{
+ uresp->sq_db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
+ uresp->sq_icid = qp->icid + 1;
+}
+
+static int qedr_copy_qp_uresp(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct ib_udata *udata)
+{
+ struct qedr_create_qp_uresp uresp;
+ int rc;
+
+ memset(&uresp, 0, sizeof(uresp));
+ qedr_copy_sq_uresp(&uresp, qp);
+ qedr_copy_rq_uresp(&uresp, qp);
+
+ uresp.atomic_supported = dev->atomic_cap != IB_ATOMIC_NONE;
+ uresp.qp_id = qp->qp_id;
+
+ rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+ if (rc)
+ DP_ERR(dev,
+ "create qp: failed a copy to user space with qp icid=0x%x.\n",
+ qp->icid);
+
+ return rc;
+}
+
+static void qedr_set_qp_init_params(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct qedr_pd *pd,
+ struct ib_qp_init_attr *attrs)
+{
+ qp->pd = pd;
+
+ spin_lock_init(&qp->q_lock);
+
+ qp->qp_type = attrs->qp_type;
+ qp->max_inline_data = attrs->cap.max_inline_data;
+ qp->sq.max_sges = attrs->cap.max_send_sge;
+ qp->state = QED_ROCE_QP_STATE_RESET;
+ qp->signaled = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR) ? true : false;
+ qp->sq_cq = get_qedr_cq(attrs->send_cq);
+ qp->rq_cq = get_qedr_cq(attrs->recv_cq);
+ qp->dev = dev;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "QP params:\tpd = %d, qp_type = %d, max_inline_data = %d, state = %d, signaled = %d, use_srq=%d\n",
+ pd->pd_id, qp->qp_type, qp->max_inline_data,
+ qp->state, qp->signaled, (attrs->srq) ? 1 : 0);
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "SQ params:\tsq_max_sges = %d, sq_cq_id = %d\n",
+ qp->sq.max_sges, qp->sq_cq->icid);
+ qp->rq.max_sges = attrs->cap.max_recv_sge;
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "RQ params:\trq_max_sges = %d, rq_cq_id = %d\n",
+ qp->rq.max_sges, qp->rq_cq->icid);
+}
+
+static inline void
+qedr_init_qp_user_params(struct qed_rdma_create_qp_in_params *params,
+ struct qedr_create_qp_ureq *ureq)
+{
+ /* QP handle to be written in CQE */
+ params->qp_handle_lo = ureq->qp_handle_lo;
+ params->qp_handle_hi = ureq->qp_handle_hi;
+}
+
+static inline void
+qedr_init_qp_kernel_doorbell_sq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qp->sq.db = dev->db_addr +
+ DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
+ qp->sq.db_data.data.icid = qp->icid + 1;
+}
+
+static inline void
+qedr_init_qp_kernel_doorbell_rq(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ qp->rq.db = dev->db_addr +
+ DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
+ qp->rq.db_data.data.icid = qp->icid;
+}
+
+static inline int
+qedr_init_qp_kernel_params_rq(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct ib_qp_init_attr *attrs)
+{
+ /* Allocate driver internal RQ array */
+ qp->rqe_wr_id = kcalloc(qp->rq.max_wr, sizeof(*qp->rqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->rqe_wr_id) {
+ DP_ERR(dev, "create qp: failed RQ shadow memory allocation\n");
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "RQ max_wr set to %d.\n", qp->rq.max_wr);
+
+ return 0;
+}
+
+static inline int
+qedr_init_qp_kernel_params_sq(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ u32 temp_max_wr;
+
+ /* Allocate driver internal SQ array */
+ temp_max_wr = attrs->cap.max_send_wr * dev->wq_multiplier;
+ temp_max_wr = min_t(u32, temp_max_wr, dev->attr.max_sqe);
+
+ /* temp_max_wr < attr->max_sqe < u16 so the casting is safe */
+ qp->sq.max_wr = (u16)temp_max_wr;
+ qp->wqe_wr_id = kcalloc(qp->sq.max_wr, sizeof(*qp->wqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->wqe_wr_id) {
+ DP_ERR(dev, "create qp: failed SQ shadow memory allocation\n");
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "SQ max_wr set to %d.\n", qp->sq.max_wr);
+
+ /* QP handle to be written in CQE */
+ params->qp_handle_lo = lower_32_bits((uintptr_t)qp);
+ params->qp_handle_hi = upper_32_bits((uintptr_t)qp);
+
+ return 0;
+}
+
+static inline int qedr_init_qp_kernel_sq(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs)
+{
+ u32 n_sq_elems, n_sq_entries;
+ int rc;
+
+ /* A single work request may take up to QEDR_MAX_SQ_WQE_SIZE elements in
+ * the ring. The ring should allow at least a single WR, even if the
+ * user requested none, due to allocation issues.
+ */
+ n_sq_entries = attrs->cap.max_send_wr;
+ n_sq_entries = min_t(u32, n_sq_entries, dev->attr.max_sqe);
+ n_sq_entries = max_t(u32, n_sq_entries, 1);
+ n_sq_elems = n_sq_entries * QEDR_MAX_SQE_ELEMENTS_PER_SQE;
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U32,
+ n_sq_elems,
+ QEDR_SQE_ELEMENT_SIZE,
+ &qp->sq.pbl);
+ if (rc) {
+ DP_ERR(dev, "failed to allocate QP %p SQ\n", qp);
+ return rc;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_SQ,
+ "SQ Pbl base addr = %llx max_send_wr=%d max_wr=%d capacity=%d, rc=%d\n",
+ qed_chain_get_pbl_phys(&qp->sq.pbl), attrs->cap.max_send_wr,
+ n_sq_entries, qed_chain_get_capacity(&qp->sq.pbl), rc);
+ return 0;
+}
+
+static inline int qedr_init_qp_kernel_rq(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs)
+{
+ u32 n_rq_elems, n_rq_entries;
+ int rc;
+
+ /* A single work request may take up to QEDR_MAX_RQ_WQE_SIZE elements in
+ * the ring. There ring should allow at least a single WR, even if the
+ * user requested none, due to allocation issues.
+ */
+ n_rq_entries = max_t(u32, attrs->cap.max_recv_wr, 1);
+ n_rq_elems = n_rq_entries * QEDR_MAX_RQE_ELEMENTS_PER_RQE;
+ rc = dev->ops->common->chain_alloc(dev->cdev,
+ QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+ QED_CHAIN_MODE_PBL,
+ QED_CHAIN_CNT_TYPE_U32,
+ n_rq_elems,
+ QEDR_RQE_ELEMENT_SIZE,
+ &qp->rq.pbl);
+
+ if (rc) {
+ DP_ERR(dev, "failed to allocate memory for QP %p RQ\n", qp);
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_RQ,
+ "RQ Pbl base addr = %llx max_recv_wr=%d max_wr=%d capacity=%d, rc=%d\n",
+ qed_chain_get_pbl_phys(&qp->rq.pbl), attrs->cap.max_recv_wr,
+ n_rq_entries, qed_chain_get_capacity(&qp->rq.pbl), rc);
+
+ /* n_rq_entries < u16 so the casting is safe */
+ qp->rq.max_wr = (u16)n_rq_entries;
+
+ return 0;
+}
+
+static inline void
+qedr_init_qp_in_params_sq(struct qedr_dev *dev,
+ struct qedr_pd *pd,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct ib_udata *udata,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ /* QP handle to be written in an async event */
+ params->qp_handle_async_lo = lower_32_bits((uintptr_t)qp);
+ params->qp_handle_async_hi = upper_32_bits((uintptr_t)qp);
+
+ params->signal_all = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR);
+ params->fmr_and_reserved_lkey = !udata;
+ params->pd = pd->pd_id;
+ params->dpi = pd->uctx ? pd->uctx->dpi : dev->dpi;
+ params->sq_cq_id = get_qedr_cq(attrs->send_cq)->icid;
+ params->max_sq_sges = 0;
+ params->stats_queue = 0;
+
+ if (udata) {
+ params->sq_num_pages = qp->usq.pbl_info.num_pbes;
+ params->sq_pbl_ptr = qp->usq.pbl_tbl->pa;
+ } else {
+ params->sq_num_pages = qed_chain_get_page_cnt(&qp->sq.pbl);
+ params->sq_pbl_ptr = qed_chain_get_pbl_phys(&qp->sq.pbl);
+ }
+}
+
+static inline void
+qedr_init_qp_in_params_rq(struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct ib_udata *udata,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ params->rq_cq_id = get_qedr_cq(attrs->recv_cq)->icid;
+ params->srq_id = 0;
+ params->use_srq = false;
+
+ if (udata) {
+ params->rq_num_pages = qp->urq.pbl_info.num_pbes;
+ params->rq_pbl_ptr = qp->urq.pbl_tbl->pa;
+ } else {
+ params->rq_num_pages = qed_chain_get_page_cnt(&qp->rq.pbl);
+ params->rq_pbl_ptr = qed_chain_get_pbl_phys(&qp->rq.pbl);
+ }
+}
+
+static inline void qedr_qp_user_print(struct qedr_dev *dev, struct qedr_qp *qp)
+{
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "create qp: successfully created user QP. qp=%p, sq_addr=0x%llx, sq_len=%zd, rq_addr=0x%llx, rq_len=%zd\n",
+ qp, qp->usq.buf_addr, qp->usq.buf_len, qp->urq.buf_addr,
+ qp->urq.buf_len);
+}
+
+static inline int qedr_init_user_qp(struct ib_ucontext *ib_ctx,
+ struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct qedr_create_qp_ureq *ureq)
+{
+ int rc;
+
+ /* SQ - read access only (0), dma sync not required (0) */
+ rc = qedr_init_user_queue(ib_ctx, dev, &qp->usq, ureq->sq_addr,
+ ureq->sq_len, 0, 0);
+ if (rc)
+ return rc;
+
+ /* RQ - read access only (0), dma sync not required (0) */
+ rc = qedr_init_user_queue(ib_ctx, dev, &qp->urq, ureq->rq_addr,
+ ureq->rq_len, 0, 0);
+
+ if (rc)
+ qedr_cleanup_user_sq(dev, qp);
+ return rc;
+}
+
+static inline int
+qedr_init_kernel_qp(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs,
+ struct qed_rdma_create_qp_in_params *params)
+{
+ int rc;
+
+ rc = qedr_init_qp_kernel_sq(dev, qp, attrs);
+ if (rc) {
+ DP_ERR(dev, "failed to init kernel QP %p SQ\n", qp);
+ return rc;
+ }
+
+ rc = qedr_init_qp_kernel_params_sq(dev, qp, attrs, params);
+ if (rc) {
+ dev->ops->common->chain_free(dev->cdev, &qp->sq.pbl);
+ DP_ERR(dev, "failed to init kernel QP %p SQ params\n", qp);
+ return rc;
+ }
+
+ rc = qedr_init_qp_kernel_rq(dev, qp, attrs);
+ if (rc) {
+ qedr_cleanup_kernel_sq(dev, qp);
+ DP_ERR(dev, "failed to init kernel QP %p RQ\n", qp);
+ return rc;
+ }
+
+ rc = qedr_init_qp_kernel_params_rq(dev, qp, attrs);
+ if (rc) {
+ DP_ERR(dev, "failed to init kernel QP %p RQ params\n", qp);
+ qedr_cleanup_kernel_sq(dev, qp);
+ dev->ops->common->chain_free(dev->cdev, &qp->rq.pbl);
+ return rc;
+ }
+
+ return rc;
+}
+
+struct ib_qp *qedr_create_qp(struct ib_pd *ibpd,
+ struct ib_qp_init_attr *attrs,
+ struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qed_rdma_create_qp_out_params out_params;
+ struct qed_rdma_create_qp_in_params in_params;
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct ib_ucontext *ib_ctx = NULL;
+ struct qedr_ucontext *ctx = NULL;
+ struct qedr_create_qp_ureq ureq;
+ struct qedr_qp *qp;
+ int rc = 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "create qp: called from %s, pd=%p\n",
+ udata ? "user library" : "kernel", pd);
+
+ rc = qedr_check_qp_attrs(ibpd, dev, attrs);
+ if (rc)
+ return ERR_PTR(rc);
+
+ qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+ if (!qp) {
+ DP_ERR(dev, "create qp: failed allocating memory\n");
+ return ERR_PTR(-ENOMEM);
+ }
+
+ if (attrs->srq)
+ return ERR_PTR(-EINVAL);
+
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "create qp: sq_cq=%p, sq_icid=%d, rq_cq=%p, rq_icid=%d\n",
+ get_qedr_cq(attrs->send_cq),
+ get_qedr_cq(attrs->send_cq)->icid,
+ get_qedr_cq(attrs->recv_cq),
+ get_qedr_cq(attrs->recv_cq)->icid);
+
+ qedr_set_qp_init_params(dev, qp, pd, attrs);
+
+ memset(&in_params, 0, sizeof(in_params));
+
+ if (udata) {
+ if (!(udata && ibpd->uobject && ibpd->uobject->context))
+ goto err0;
+
+ ib_ctx = ibpd->uobject->context;
+ ctx = get_qedr_ucontext(ib_ctx);
+
+ memset(&ureq, 0, sizeof(ureq));
+ if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) {
+ DP_ERR(dev,
+ "create qp: problem copying data from user space\n");
+ goto err0;
+ }
+
+ rc = qedr_init_user_qp(ib_ctx, dev, qp, &ureq);
+ if (rc)
+ goto err0;
+
+ qedr_init_qp_user_params(&in_params, &ureq);
+ } else {
+ rc = qedr_init_kernel_qp(dev, qp, attrs, &in_params);
+ if (rc)
+ goto err0;
+ }
+
+ qedr_init_qp_in_params_sq(dev, pd, qp, attrs, udata, &in_params);
+ qedr_init_qp_in_params_rq(qp, attrs, udata, &in_params);
+
+ qp->qed_qp = dev->ops->rdma_create_qp(dev->rdma_ctx,
+ &in_params, &out_params);
+
+ if (!qp->qed_qp)
+ goto err1;
+
+ qp->qp_id = out_params.qp_id;
+ qp->icid = out_params.icid;
+ qp->ibqp.qp_num = qp->qp_id;
+
+ if (udata) {
+ rc = qedr_copy_qp_uresp(dev, qp, udata);
+ if (rc)
+ goto err2;
+
+ qedr_qp_user_print(dev, qp);
+ } else {
+ qedr_init_qp_kernel_doorbell_sq(dev, qp);
+ qedr_init_qp_kernel_doorbell_rq(dev, qp);
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "created %s space QP %p\n", udata ? "user" : "kernel", qp);
+
+ return &qp->ibqp;
+
+err2:
+ rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp);
+ if (rc)
+ DP_ERR(dev, "create qp: fatal fault. rc=%d", rc);
+err1:
+ if (udata) {
+ qedr_cleanup_user_sq(dev, qp);
+ qedr_cleanup_user_rq(dev, qp);
+ } else {
+ qedr_cleanup_kernel_sq(dev, qp);
+ qedr_cleanup_kernel_rq(dev, qp);
+ }
+
+err0:
+ kfree(qp);
+
+ return ERR_PTR(-EFAULT);
+}
+
+enum ib_qp_state qedr_get_ibqp_state(enum qed_roce_qp_state qp_state)
+{
+ switch (qp_state) {
+ case QED_ROCE_QP_STATE_RESET:
+ return IB_QPS_RESET;
+ case QED_ROCE_QP_STATE_INIT:
+ return IB_QPS_INIT;
+ case QED_ROCE_QP_STATE_RTR:
+ return IB_QPS_RTR;
+ case QED_ROCE_QP_STATE_RTS:
+ return IB_QPS_RTS;
+ case QED_ROCE_QP_STATE_SQD:
+ return IB_QPS_SQD;
+ case QED_ROCE_QP_STATE_ERR:
+ return IB_QPS_ERR;
+ case QED_ROCE_QP_STATE_SQE:
+ return IB_QPS_SQE;
+ }
+ return IB_QPS_ERR;
+}
+
+enum qed_roce_qp_state qedr_get_state_from_ibqp(enum ib_qp_state qp_state)
+{
+ switch (qp_state) {
+ case IB_QPS_RESET:
+ return QED_ROCE_QP_STATE_RESET;
+ case IB_QPS_INIT:
+ return QED_ROCE_QP_STATE_INIT;
+ case IB_QPS_RTR:
+ return QED_ROCE_QP_STATE_RTR;
+ case IB_QPS_RTS:
+ return QED_ROCE_QP_STATE_RTS;
+ case IB_QPS_SQD:
+ return QED_ROCE_QP_STATE_SQD;
+ case IB_QPS_ERR:
+ return QED_ROCE_QP_STATE_ERR;
+ default:
+ return QED_ROCE_QP_STATE_ERR;
+ }
+}
+
+static void qedr_reset_qp_hwq_info(struct qedr_qp_hwq_info *qph)
+{
+ qed_chain_reset(&qph->pbl);
+ qph->prod = 0;
+ qph->cons = 0;
+ qph->wqe_cons = 0;
+ qph->db_data.data.value = cpu_to_le16(0);
+}
+
+static int qedr_update_qp_state(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ enum qed_roce_qp_state new_state)
+{
+ int status = 0;
+
+ if (new_state == qp->state)
+ return 1;
+
+ switch (qp->state) {
+ case QED_ROCE_QP_STATE_RESET:
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_INIT:
+ qp->prev_wqe_size = 0;
+ qedr_reset_qp_hwq_info(&qp->sq);
+ qedr_reset_qp_hwq_info(&qp->rq);
+ break;
+ default:
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_INIT:
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RTR:
+ /* Update doorbell (in case post_recv was
+ * done before move to RTR)
+ */
+ wmb();
+ writel(qp->rq.db_data.raw, qp->rq.db);
+ /* Make sure write takes effect */
+ mmiowb();
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_RTR:
+ /* RTR->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RTS:
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_RTS:
+ /* RTS->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_SQD:
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_SQD:
+ /* SQD->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RTS:
+ case QED_ROCE_QP_STATE_ERR:
+ break;
+ default:
+ /* Invalid state change. */
+ status = -EINVAL;
+ break;
+ };
+ break;
+ case QED_ROCE_QP_STATE_ERR:
+ /* ERR->XXX */
+ switch (new_state) {
+ case QED_ROCE_QP_STATE_RESET:
+ break;
+ default:
+ status = -EINVAL;
+ break;
+ };
+ break;
+ default:
+ status = -EINVAL;
+ break;
+ };
+
+ return status;
+}
+
+int qedr_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_udata *udata)
+{
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_rdma_modify_qp_in_params qp_params = { 0 };
+ struct qedr_dev *dev = get_qedr_dev(&qp->dev->ibdev);
+ enum ib_qp_state old_qp_state, new_qp_state;
+ int rc = 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP,
+ "modify qp: qp %p attr_mask=0x%x, state=%d", qp, attr_mask,
+ attr->qp_state);
+
+ old_qp_state = qedr_get_ibqp_state(qp->state);
+ if (attr_mask & IB_QP_STATE)
+ new_qp_state = attr->qp_state;
+ else
+ new_qp_state = old_qp_state;
+
+ if (!ib_modify_qp_is_ok
+ (old_qp_state, new_qp_state, ibqp->qp_type, attr_mask,
+ IB_LINK_LAYER_ETHERNET)) {
+ DP_ERR(dev,
+ "modify qp: invalid attribute mask=0x%x specified for\n"
+ "qpn=0x%x of type=0x%x old_qp_state=0x%x, new_qp_state=0x%x\n",
+ attr_mask, qp->qp_id, ibqp->qp_type, old_qp_state,
+ new_qp_state);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ /* Translate the masks... */
+ if (attr_mask & IB_QP_STATE) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_NEW_STATE, 1);
+ qp_params.new_state = qedr_get_state_from_ibqp(attr->qp_state);
+ }
+
+ if (attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY)
+ qp_params.sqd_async = true;
+
+ if (attr_mask & IB_QP_PKEY_INDEX) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_PKEY, 1);
+ if (attr->pkey_index >= QEDR_ROCE_PKEY_TABLE_LEN) {
+ rc = -EINVAL;
+ goto err;
+ }
+
+ qp_params.pkey = QEDR_ROCE_PKEY_DEFAULT;
+ }
+
+ if (attr_mask & IB_QP_QKEY)
+ qp->qkey = attr->qkey;
+
+ if (attr_mask & IB_QP_ACCESS_FLAGS) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN, 1);
+ qp_params.incoming_rdma_read_en = attr->qp_access_flags &
+ IB_ACCESS_REMOTE_READ;
+ qp_params.incoming_rdma_write_en = attr->qp_access_flags &
+ IB_ACCESS_REMOTE_WRITE;
+ qp_params.incoming_atomic_en = attr->qp_access_flags &
+ IB_ACCESS_REMOTE_ATOMIC;
+ }
+
+ if (attr_mask & (IB_QP_AV | IB_QP_PATH_MTU)) {
+ if (attr_mask & IB_QP_PATH_MTU) {
+ if (attr->path_mtu < IB_MTU_256 ||
+ attr->path_mtu > IB_MTU_4096) {
+ pr_err("ERROR QEDR: Only MTU sizes of 256, 512, 1024, 2048 and 4096 are supported\n");
+ rc = -EINVAL;
+ goto err;
+ }
+ qp->mtu = min(ib_mtu_enum_to_int(attr->path_mtu),
+ ib_mtu_enum_to_int(iboe_get_mtu
+ (dev->ndev->mtu)));
+ }
+
+ if (!qp->mtu) {
+ qp->mtu =
+ ib_mtu_enum_to_int(iboe_get_mtu(dev->ndev->mtu));
+ pr_err("Fixing zeroed MTU to qp->mtu = %d\n", qp->mtu);
+ }
+
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR, 1);
+
+ qp_params.traffic_class_tos = attr->ah_attr.grh.traffic_class;
+ qp_params.flow_label = attr->ah_attr.grh.flow_label;
+ qp_params.hop_limit_ttl = attr->ah_attr.grh.hop_limit;
+
+ qp->sgid_idx = attr->ah_attr.grh.sgid_index;
+
+ rc = get_gid_info_from_table(ibqp, attr, attr_mask, &qp_params);
+ if (rc) {
+ DP_ERR(dev,
+ "modify qp: problems with GID index %d (rc=%d)\n",
+ attr->ah_attr.grh.sgid_index, rc);
+ return rc;
+ }
+
+ rc = qedr_get_dmac(dev, &attr->ah_attr,
+ qp_params.remote_mac_addr);
+ if (rc)
+ return rc;
+
+ qp_params.use_local_mac = true;
+ ether_addr_copy(qp_params.local_mac_addr, dev->ndev->dev_addr);
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "dgid=%x:%x:%x:%x\n",
+ qp_params.dgid.dwords[0], qp_params.dgid.dwords[1],
+ qp_params.dgid.dwords[2], qp_params.dgid.dwords[3]);
+ DP_VERBOSE(dev, QEDR_MSG_QP, "sgid=%x:%x:%x:%x\n",
+ qp_params.sgid.dwords[0], qp_params.sgid.dwords[1],
+ qp_params.sgid.dwords[2], qp_params.sgid.dwords[3]);
+ DP_VERBOSE(dev, QEDR_MSG_QP, "remote_mac=[%x:%x:%x:%x:%x:%x]\n",
+ qp_params.remote_mac_addr[0],
+ qp_params.remote_mac_addr[1],
+ qp_params.remote_mac_addr[2],
+ qp_params.remote_mac_addr[3],
+ qp_params.remote_mac_addr[4],
+ qp_params.remote_mac_addr[5]);
+
+ qp_params.mtu = qp->mtu;
+ qp_params.lb_indication = false;
+ }
+
+ if (!qp_params.mtu) {
+ /* Stay with current MTU */
+ if (qp->mtu)
+ qp_params.mtu = qp->mtu;
+ else
+ qp_params.mtu =
+ ib_mtu_enum_to_int(iboe_get_mtu(dev->ndev->mtu));
+ }
+
+ if (attr_mask & IB_QP_TIMEOUT) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT, 1);
+
+ qp_params.ack_timeout = attr->timeout;
+ if (attr->timeout) {
+ u32 temp;
+
+ temp = 4096 * (1UL << attr->timeout) / 1000 / 1000;
+ /* FW requires [msec] */
+ qp_params.ack_timeout = temp;
+ } else {
+ /* Infinite */
+ qp_params.ack_timeout = 0;
+ }
+ }
+ if (attr_mask & IB_QP_RETRY_CNT) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RETRY_CNT, 1);
+ qp_params.retry_cnt = attr->retry_cnt;
+ }
+
+ if (attr_mask & IB_QP_RNR_RETRY) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT, 1);
+ qp_params.rnr_retry_cnt = attr->rnr_retry;
+ }
+
+ if (attr_mask & IB_QP_RQ_PSN) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RQ_PSN, 1);
+ qp_params.rq_psn = attr->rq_psn;
+ qp->rq_psn = attr->rq_psn;
+ }
+
+ if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
+ if (attr->max_rd_atomic > dev->attr.max_qp_req_rd_atomic_resc) {
+ rc = -EINVAL;
+ DP_ERR(dev,
+ "unsupported max_rd_atomic=%d, supported=%d\n",
+ attr->max_rd_atomic,
+ dev->attr.max_qp_req_rd_atomic_resc);
+ goto err;
+ }
+
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ, 1);
+ qp_params.max_rd_atomic_req = attr->max_rd_atomic;
+ }
+
+ if (attr_mask & IB_QP_MIN_RNR_TIMER) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER, 1);
+ qp_params.min_rnr_nak_timer = attr->min_rnr_timer;
+ }
+
+ if (attr_mask & IB_QP_SQ_PSN) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_SQ_PSN, 1);
+ qp_params.sq_psn = attr->sq_psn;
+ qp->sq_psn = attr->sq_psn;
+ }
+
+ if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
+ if (attr->max_dest_rd_atomic >
+ dev->attr.max_qp_resp_rd_atomic_resc) {
+ DP_ERR(dev,
+ "unsupported max_dest_rd_atomic=%d, supported=%d\n",
+ attr->max_dest_rd_atomic,
+ dev->attr.max_qp_resp_rd_atomic_resc);
+
+ rc = -EINVAL;
+ goto err;
+ }
+
+ SET_FIELD(qp_params.modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP, 1);
+ qp_params.max_rd_atomic_resp = attr->max_dest_rd_atomic;
+ }
+
+ if (attr_mask & IB_QP_DEST_QPN) {
+ SET_FIELD(qp_params.modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_DEST_QP, 1);
+
+ qp_params.dest_qp = attr->dest_qp_num;
+ qp->dest_qp_num = attr->dest_qp_num;
+ }
+
+ if (qp->qp_type != IB_QPT_GSI)
+ rc = dev->ops->rdma_modify_qp(dev->rdma_ctx,
+ qp->qed_qp, &qp_params);
+
+ if (attr_mask & IB_QP_STATE) {
+ if ((qp->qp_type != IB_QPT_GSI) && (!udata))
+ qedr_update_qp_state(dev, qp, qp_params.new_state);
+ qp->state = qp_params.new_state;
+ }
+
+err:
+ return rc;
+}
+
+static int qedr_to_ib_qp_acc_flags(struct qed_rdma_query_qp_out_params *params)
+{
+ int ib_qp_acc_flags = 0;
+
+ if (params->incoming_rdma_write_en)
+ ib_qp_acc_flags |= IB_ACCESS_REMOTE_WRITE;
+ if (params->incoming_rdma_read_en)
+ ib_qp_acc_flags |= IB_ACCESS_REMOTE_READ;
+ if (params->incoming_atomic_en)
+ ib_qp_acc_flags |= IB_ACCESS_REMOTE_ATOMIC;
+ ib_qp_acc_flags |= IB_ACCESS_LOCAL_WRITE;
+ return ib_qp_acc_flags;
+}
+
+int qedr_query_qp(struct ib_qp *ibqp,
+ struct ib_qp_attr *qp_attr,
+ int attr_mask, struct ib_qp_init_attr *qp_init_attr)
+{
+ struct qed_rdma_query_qp_out_params params;
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qedr_dev *dev = qp->dev;
+ int rc = 0;
+
+ memset(¶ms, 0, sizeof(params));
+
+ rc = dev->ops->rdma_query_qp(dev->rdma_ctx, qp->qed_qp, ¶ms);
+ if (rc)
+ goto err;
+
+ memset(qp_attr, 0, sizeof(*qp_attr));
+ memset(qp_init_attr, 0, sizeof(*qp_init_attr));
+
+ qp_attr->qp_state = qedr_get_ibqp_state(params.state);
+ qp_attr->cur_qp_state = qedr_get_ibqp_state(params.state);
+ qp_attr->path_mtu = iboe_get_mtu(params.mtu);
+ qp_attr->path_mig_state = IB_MIG_MIGRATED;
+ qp_attr->rq_psn = params.rq_psn;
+ qp_attr->sq_psn = params.sq_psn;
+ qp_attr->dest_qp_num = params.dest_qp;
+
+ qp_attr->qp_access_flags = qedr_to_ib_qp_acc_flags(¶ms);
+
+ qp_attr->cap.max_send_wr = qp->sq.max_wr;
+ qp_attr->cap.max_recv_wr = qp->rq.max_wr;
+ qp_attr->cap.max_send_sge = qp->sq.max_sges;
+ qp_attr->cap.max_recv_sge = qp->rq.max_sges;
+ qp_attr->cap.max_inline_data = qp->max_inline_data;
+ qp_init_attr->cap = qp_attr->cap;
+
+ memcpy(&qp_attr->ah_attr.grh.dgid.raw[0], ¶ms.dgid.bytes[0],
+ sizeof(qp_attr->ah_attr.grh.dgid.raw));
+
+ qp_attr->ah_attr.grh.flow_label = params.flow_label;
+ qp_attr->ah_attr.grh.sgid_index = qp->sgid_idx;
+ qp_attr->ah_attr.grh.hop_limit = params.hop_limit_ttl;
+ qp_attr->ah_attr.grh.traffic_class = params.traffic_class_tos;
+
+ qp_attr->ah_attr.ah_flags = IB_AH_GRH;
+ qp_attr->ah_attr.port_num = 1;
+ qp_attr->ah_attr.sl = 0;
+ qp_attr->timeout = params.timeout;
+ qp_attr->rnr_retry = params.rnr_retry;
+ qp_attr->retry_cnt = params.retry_cnt;
+ qp_attr->min_rnr_timer = params.min_rnr_nak_timer;
+ qp_attr->pkey_index = params.pkey_index;
+ qp_attr->port_num = 1;
+ qp_attr->ah_attr.src_path_bits = 0;
+ qp_attr->ah_attr.static_rate = 0;
+ qp_attr->alt_pkey_index = 0;
+ qp_attr->alt_port_num = 0;
+ qp_attr->alt_timeout = 0;
+ memset(&qp_attr->alt_ah_attr, 0, sizeof(qp_attr->alt_ah_attr));
+
+ qp_attr->sq_draining = (params.state == QED_ROCE_QP_STATE_SQD) ? 1 : 0;
+ qp_attr->max_dest_rd_atomic = params.max_dest_rd_atomic;
+ qp_attr->max_rd_atomic = params.max_rd_atomic;
+ qp_attr->en_sqd_async_notify = (params.sqd_async) ? 1 : 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "QEDR_QUERY_QP: max_inline_data=%d\n",
+ qp_attr->cap.max_inline_data);
+
+err:
+ return rc;
+}
+
+int qedr_destroy_qp(struct ib_qp *ibqp)
+{
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qedr_dev *dev = qp->dev;
+ struct ib_qp_attr attr;
+ int attr_mask = 0;
+ int rc = 0;
+
+ DP_VERBOSE(dev, QEDR_MSG_QP, "destroy qp: destroying %p, qp type=%d\n",
+ qp, qp->qp_type);
+
+ if (qp->state != (QED_ROCE_QP_STATE_RESET | QED_ROCE_QP_STATE_ERR |
+ QED_ROCE_QP_STATE_INIT)) {
+ attr.qp_state = IB_QPS_ERR;
+ attr_mask |= IB_QP_STATE;
+
+ /* Change the QP state to ERROR */
+ qedr_modify_qp(ibqp, &attr, attr_mask, NULL);
+ }
+
+ if (qp->qp_type != IB_QPT_GSI) {
+ rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp);
+ if (rc)
+ return rc;
+ }
+
+ if (ibqp->uobject && ibqp->uobject->context) {
+ qedr_cleanup_user_sq(dev, qp);
+ qedr_cleanup_user_rq(dev, qp);
+ } else {
+ qedr_cleanup_kernel_sq(dev, qp);
+ qedr_cleanup_kernel_rq(dev, qp);
+ }
+
+ kfree(qp);
+
+ return rc;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index d5e1017..80dbac1 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -36,5 +36,12 @@ struct ib_cq *qedr_create_cq(struct ib_device *ibdev,
int qedr_resize_cq(struct ib_cq *, int cqe, struct ib_udata *);
int qedr_destroy_cq(struct ib_cq *);
int qedr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
+struct ib_qp *qedr_create_qp(struct ib_pd *, struct ib_qp_init_attr *attrs,
+ struct ib_udata *);
+int qedr_modify_qp(struct ib_qp *, struct ib_qp_attr *attr,
+ int attr_mask, struct ib_udata *udata);
+int qedr_query_qp(struct ib_qp *, struct ib_qp_attr *qp_attr,
+ int qp_attr_mask, struct ib_qp_init_attr *);
+int qedr_destroy_qp(struct ib_qp *ibqp);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_cxt.h b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
index d00ad05..2b8bdaa 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_cxt.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_cxt.h
@@ -176,6 +176,7 @@ u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
enum protocol_type type);
u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
enum protocol_type type);
+int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto);
#define QED_CTX_WORKING_MEM 0
#define QED_CTX_FL_MEM 1
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 7b489f0..cf92ff7 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -1114,6 +1114,1213 @@ err: dma_free_coherent(&p_hwfn->cdev->pdev->dev,
return rc;
}
+static void qed_rdma_set_fw_mac(u16 *p_fw_mac, u8 *p_qed_mac)
+{
+ p_fw_mac[0] = cpu_to_le16((p_qed_mac[0] << 8) + p_qed_mac[1]);
+ p_fw_mac[1] = cpu_to_le16((p_qed_mac[2] << 8) + p_qed_mac[3]);
+ p_fw_mac[2] = cpu_to_le16((p_qed_mac[4] << 8) + p_qed_mac[5]);
+}
+
+static void qed_rdma_copy_gids(struct qed_rdma_qp *qp, __le32 *src_gid,
+ __le32 *dst_gid)
+{
+ u32 i;
+
+ if (qp->roce_mode == ROCE_V2_IPV4) {
+ /* RoCE v2 - IPv4 only */
+ /* The IPv4 addresses shall be aligned to the highest word.
+ * The lower words must be zero.
+ */
+ memset(src_gid, 0, sizeof(union qed_gid));
+ memset(dst_gid, 0, sizeof(union qed_gid));
+ src_gid[3] = cpu_to_le32(qp->sgid.ipv4_addr);
+ dst_gid[3] = cpu_to_le32(qp->dgid.ipv4_addr);
+ } else {
+ /* RoCE, and RoCE v2 - IPv6 */
+ /* GIDs and IPv6 addresses coincide in location and size */
+ for (i = 0; i < ARRAY_SIZE(qp->sgid.dwords); i++) {
+ src_gid[i] = cpu_to_le32(qp->sgid.dwords[i]);
+ dst_gid[i] = cpu_to_le32(qp->dgid.dwords[i]);
+ }
+ }
+}
+
+static enum roce_flavor qed_roce_mode_to_flavor(enum roce_mode roce_mode)
+{
+ enum roce_flavor flavor;
+
+ switch (roce_mode) {
+ case ROCE_V1:
+ flavor = PLAIN_ROCE;
+ break;
+ case ROCE_V2_IPV4:
+ flavor = RROCE_IPV4;
+ break;
+ case ROCE_V2_IPV6:
+ flavor = ROCE_V2_IPV6;
+ break;
+ default:
+ flavor = MAX_ROCE_MODE;
+ break;
+ }
+ return flavor;
+}
+
+int qed_roce_alloc_cid(struct qed_hwfn *p_hwfn, u16 *cid)
+{
+ struct qed_rdma_info *p_rdma_info = p_hwfn->p_rdma_info;
+ u32 responder_icid;
+ u32 requester_icid;
+ int rc;
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_rdma_info->cid_map,
+ &responder_icid);
+ if (rc) {
+ spin_unlock_bh(&p_rdma_info->lock);
+ return rc;
+ }
+
+ rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_rdma_info->cid_map,
+ &requester_icid);
+
+ spin_unlock_bh(&p_rdma_info->lock);
+ if (rc)
+ goto err;
+
+ /* the two icid's should be adjacent */
+ if ((requester_icid - responder_icid) != 1) {
+ DP_NOTICE(p_hwfn, "Failed to allocate two adjacent qp's'\n");
+ rc = -EINVAL;
+ goto err;
+ }
+
+ responder_icid += qed_cxt_get_proto_cid_start(p_hwfn,
+ p_rdma_info->proto);
+ requester_icid += qed_cxt_get_proto_cid_start(p_hwfn,
+ p_rdma_info->proto);
+
+ /* If these icids require a new ILT line allocate DMA-able context for
+ * an ILT page
+ */
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, responder_icid);
+ if (rc)
+ goto err;
+
+ rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, requester_icid);
+ if (rc)
+ goto err;
+
+ *cid = (u16)responder_icid;
+ return rc;
+
+err:
+ spin_lock_bh(&p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_rdma_info->cid_map, responder_icid);
+ qed_bmap_release_id(p_hwfn, &p_rdma_info->cid_map, requester_icid);
+
+ spin_unlock_bh(&p_rdma_info->lock);
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Allocate CID - failed, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_roce_sp_create_responder(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp)
+{
+ struct roce_create_qp_resp_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ union qed_qm_pq_params qm_params;
+ enum roce_flavor roce_flavor;
+ struct qed_spq_entry *p_ent;
+ u16 physical_queue0 = 0;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ /* Allocate DMA-able memory for IRQ */
+ qp->irq_num_pages = 1;
+ qp->irq = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ RDMA_RING_PAGE_SIZE,
+ &qp->irq_phys_addr, GFP_KERNEL);
+ if (!qp->irq) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed create responder failed: cannot allocate memory (irq). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_CREATE_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.roce_create_qp_resp;
+
+ p_ramrod->flags = 0;
+
+ roce_flavor = qed_roce_mode_to_flavor(qp->roce_mode);
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_ROCE_FLAVOR, roce_flavor);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_RDMA_RD_EN,
+ qp->incoming_rdma_read_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_RDMA_WR_EN,
+ qp->incoming_rdma_write_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_ATOMIC_EN,
+ qp->incoming_atomic_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_E2E_FLOW_CONTROL_EN,
+ qp->e2e_flow_control_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_SRQ_FLG, qp->use_srq);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_RESERVED_KEY_EN,
+ qp->fmr_and_reserved_lkey);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER,
+ qp->min_rnr_nak_timer);
+
+ p_ramrod->max_ird = qp->max_rd_atomic_resp;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->irq_num_pages = qp->irq_num_pages;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->dst_qp_id = cpu_to_le32(qp->dest_qp);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ p_ramrod->initial_psn = cpu_to_le32(qp->rq_psn);
+ p_ramrod->pd = cpu_to_le16(qp->pd);
+ p_ramrod->rq_num_pages = cpu_to_le16(qp->rq_num_pages);
+ p_ramrod->rq_pbl_addr.hi = DMA_HI_LE(qp->rq_pbl_ptr);
+ p_ramrod->rq_pbl_addr.lo = DMA_LO_LE(qp->rq_pbl_ptr);
+ p_ramrod->irq_pbl_addr.hi = DMA_HI_LE(qp->irq_phys_addr);
+ p_ramrod->irq_pbl_addr.lo = DMA_LO_LE(qp->irq_phys_addr);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ p_ramrod->qp_handle_for_async.hi = cpu_to_le32(qp->qp_handle_async.hi);
+ p_ramrod->qp_handle_for_async.lo = cpu_to_le32(qp->qp_handle_async.lo);
+ p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
+ p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
+ p_ramrod->stats_counter_id = p_hwfn->rel_pf_id;
+ p_ramrod->cq_cid = cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) |
+ qp->rq_cq_id);
+
+ memset(&qm_params, 0, sizeof(qm_params));
+ qm_params.roce.qpid = (qp->icid >> 1);
+ physical_queue0 = qed_get_qm_pq(p_hwfn, PROTOCOLID_ROCE, &qm_params);
+
+ p_ramrod->physical_queue0 = cpu_to_le16(physical_queue0);
+ p_ramrod->dpi = cpu_to_le16(qp->dpi);
+
+ qed_rdma_set_fw_mac(p_ramrod->remote_mac_addr, qp->remote_mac_addr);
+ qed_rdma_set_fw_mac(p_ramrod->local_mac_addr, qp->local_mac_addr);
+
+ p_ramrod->udp_src_port = qp->udp_src_port;
+ p_ramrod->vlan_id = cpu_to_le16(qp->vlan_id);
+ p_ramrod->srq_id.srq_idx = cpu_to_le16(qp->srq_id);
+ p_ramrod->srq_id.opaque_fid = cpu_to_le16(p_hwfn->hw_info.opaque_fid);
+
+ p_ramrod->stats_counter_id = RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) +
+ qp->stats_queue;
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d physical_queue0 = 0x%x\n",
+ rc, physical_queue0);
+
+ if (rc)
+ goto err;
+
+ qp->resp_offloaded = true;
+
+ return rc;
+
+err:
+ DP_NOTICE(p_hwfn, "create responder - failed, rc = %d\n", rc);
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->irq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->irq, qp->irq_phys_addr);
+
+ return rc;
+}
+
+static int qed_roce_sp_create_requester(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp)
+{
+ struct roce_create_qp_req_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ union qed_qm_pq_params qm_params;
+ enum roce_flavor roce_flavor;
+ struct qed_spq_entry *p_ent;
+ u16 physical_queue0 = 0;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ /* Allocate DMA-able memory for ORQ */
+ qp->orq_num_pages = 1;
+ qp->orq = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ RDMA_RING_PAGE_SIZE,
+ &qp->orq_phys_addr, GFP_KERNEL);
+ if (!qp->orq) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed create requester failed: cannot allocate memory (orq). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid + 1;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_RAMROD_CREATE_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.roce_create_qp_req;
+
+ p_ramrod->flags = 0;
+
+ roce_flavor = qed_roce_mode_to_flavor(qp->roce_mode);
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_ROCE_FLAVOR, roce_flavor);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_FMR_AND_RESERVED_EN,
+ qp->fmr_and_reserved_lkey);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_SIGNALED_COMP, qp->signal_all);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT, qp->retry_cnt);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_REQ_RAMROD_DATA_RNR_NAK_CNT,
+ qp->rnr_retry_cnt);
+
+ p_ramrod->max_ord = qp->max_rd_atomic_req;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->orq_num_pages = qp->orq_num_pages;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->dst_qp_id = cpu_to_le32(qp->dest_qp);
+ p_ramrod->ack_timeout_val = cpu_to_le32(qp->ack_timeout);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ p_ramrod->initial_psn = cpu_to_le32(qp->sq_psn);
+ p_ramrod->pd = cpu_to_le16(qp->pd);
+ p_ramrod->sq_num_pages = cpu_to_le16(qp->sq_num_pages);
+ p_ramrod->sq_pbl_addr.hi = DMA_HI_LE(qp->sq_pbl_ptr);
+ p_ramrod->sq_pbl_addr.lo = DMA_LO_LE(qp->sq_pbl_ptr);
+ p_ramrod->orq_pbl_addr.hi = DMA_HI_LE(qp->orq_phys_addr);
+ p_ramrod->orq_pbl_addr.lo = DMA_LO_LE(qp->orq_phys_addr);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ p_ramrod->qp_handle_for_async.hi = cpu_to_le32(qp->qp_handle_async.hi);
+ p_ramrod->qp_handle_for_async.lo = cpu_to_le32(qp->qp_handle_async.lo);
+ p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
+ p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
+ p_ramrod->stats_counter_id = p_hwfn->rel_pf_id;
+ p_ramrod->cq_cid = cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) |
+ qp->sq_cq_id);
+
+ memset(&qm_params, 0, sizeof(qm_params));
+ qm_params.roce.qpid = (qp->icid >> 1);
+ physical_queue0 = qed_get_qm_pq(p_hwfn, PROTOCOLID_ROCE, &qm_params);
+
+ p_ramrod->physical_queue0 = cpu_to_le16(physical_queue0);
+ p_ramrod->dpi = cpu_to_le16(qp->dpi);
+
+ qed_rdma_set_fw_mac(p_ramrod->remote_mac_addr, qp->remote_mac_addr);
+ qed_rdma_set_fw_mac(p_ramrod->local_mac_addr, qp->local_mac_addr);
+
+ p_ramrod->udp_src_port = qp->udp_src_port;
+ p_ramrod->vlan_id = cpu_to_le16(qp->vlan_id);
+ p_ramrod->stats_counter_id = RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) +
+ qp->stats_queue;
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+
+ if (rc)
+ goto err;
+
+ qp->req_offloaded = true;
+
+ return rc;
+
+err:
+ DP_NOTICE(p_hwfn, "Create requested - failed, rc = %d\n", rc);
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->orq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->orq, qp->orq_phys_addr);
+ return rc;
+}
+
+static int qed_roce_sp_modify_responder(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ bool move_to_err, u32 modify_flags)
+{
+ struct roce_modify_qp_resp_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (move_to_err && !qp->resp_offloaded)
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_EVENT_MODIFY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "rc = %d\n", rc);
+ return rc;
+ }
+
+ p_ramrod = &p_ent->ramrod.roce_modify_qp_resp;
+
+ p_ramrod->flags = 0;
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MOVE_TO_ERR_FLG, move_to_err);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_RD_EN,
+ qp->incoming_rdma_read_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_WR_EN,
+ qp->incoming_rdma_write_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_ATOMIC_EN,
+ qp->incoming_atomic_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_CREATE_QP_RESP_RAMROD_DATA_E2E_FLOW_CONTROL_EN,
+ qp->e2e_flow_control_en);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_RDMA_OPS_EN_FLG,
+ GET_FIELD(modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_P_KEY_FLG,
+ GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_ADDRESS_VECTOR_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MAX_IRD_FLG,
+ GET_FIELD(modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER));
+
+ p_ramrod->fields = 0;
+ SET_FIELD(p_ramrod->fields,
+ ROCE_MODIFY_QP_RESP_RAMROD_DATA_MIN_RNR_NAK_TIMER,
+ qp->min_rnr_nak_timer);
+
+ p_ramrod->max_ird = qp->max_rd_atomic_resp;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify responder, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_roce_sp_modify_requester(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ bool move_to_sqd,
+ bool move_to_err, u32 modify_flags)
+{
+ struct roce_modify_qp_req_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (move_to_err && !(qp->req_offloaded))
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid + 1;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_EVENT_MODIFY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc) {
+ DP_NOTICE(p_hwfn, "rc = %d\n", rc);
+ return rc;
+ }
+
+ p_ramrod = &p_ent->ramrod.roce_modify_qp_req;
+
+ p_ramrod->flags = 0;
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_MOVE_TO_ERR_FLG, move_to_err);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_MOVE_TO_SQD_FLG, move_to_sqd);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_EN_SQD_ASYNC_NOTIFY,
+ qp->sqd_async);
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_P_KEY_FLG,
+ GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ADDRESS_VECTOR_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_MAX_ORD_FLG,
+ GET_FIELD(modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_RNR_NAK_CNT_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT_FLG,
+ GET_FIELD(modify_flags, QED_ROCE_MODIFY_QP_VALID_RETRY_CNT));
+
+ SET_FIELD(p_ramrod->flags,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ACK_TIMEOUT_FLG,
+ GET_FIELD(modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT));
+
+ p_ramrod->fields = 0;
+ SET_FIELD(p_ramrod->fields,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_ERR_RETRY_CNT, qp->retry_cnt);
+
+ SET_FIELD(p_ramrod->fields,
+ ROCE_MODIFY_QP_REQ_RAMROD_DATA_RNR_NAK_CNT,
+ qp->rnr_retry_cnt);
+
+ p_ramrod->max_ord = qp->max_rd_atomic_req;
+ p_ramrod->traffic_class = qp->traffic_class_tos;
+ p_ramrod->hop_limit = qp->hop_limit_ttl;
+ p_ramrod->p_key = cpu_to_le16(qp->pkey);
+ p_ramrod->flow_label = cpu_to_le32(qp->flow_label);
+ p_ramrod->ack_timeout_val = cpu_to_le32(qp->ack_timeout);
+ p_ramrod->mtu = cpu_to_le16(qp->mtu);
+ qed_rdma_copy_gids(qp, p_ramrod->src_gid, p_ramrod->dst_gid);
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify requester, rc = %d\n", rc);
+ return rc;
+}
+
+static int qed_roce_sp_destroy_qp_responder(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ u32 *num_invalidated_mw)
+{
+ struct roce_destroy_qp_resp_output_params *p_ramrod_res;
+ struct roce_destroy_qp_resp_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (!qp->resp_offloaded)
+ return 0;
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ ROCE_RAMROD_DESTROY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ return rc;
+
+ p_ramrod = &p_ent->ramrod.roce_destroy_qp_resp;
+
+ p_ramrod_res = (struct roce_destroy_qp_resp_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_ramrod_res),
+ &ramrod_res_phys, GFP_KERNEL);
+
+ if (!p_ramrod_res) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed destroy responder failed: cannot allocate memory (ramrod). rc = %d\n",
+ rc);
+ return rc;
+ }
+
+ DMA_REGPAIR_LE(p_ramrod->output_params_addr, ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err;
+
+ *num_invalidated_mw = le32_to_cpu(p_ramrod_res->num_invalidated_mw);
+
+ /* Free IRQ - only if ramrod succeeded, in case FW is still using it */
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->irq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->irq, qp->irq_phys_addr);
+
+ qp->resp_offloaded = false;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroy responder, rc = %d\n", rc);
+
+err:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(struct roce_destroy_qp_resp_output_params),
+ p_ramrod_res, ramrod_res_phys);
+
+ return rc;
+}
+
+static int qed_roce_sp_destroy_qp_requester(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ u32 *num_bound_mw)
+{
+ struct roce_destroy_qp_req_output_params *p_ramrod_res;
+ struct roce_destroy_qp_req_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ dma_addr_t ramrod_res_phys;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ if (!qp->req_offloaded)
+ return 0;
+
+ p_ramrod_res = (struct roce_destroy_qp_req_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(*p_ramrod_res),
+ &ramrod_res_phys, GFP_KERNEL);
+ if (!p_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed destroy requester failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid + 1;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_DESTROY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err;
+
+ p_ramrod = &p_ent->ramrod.roce_destroy_qp_req;
+ p_ramrod->output_params_addr.hi = DMA_HI_LE(ramrod_res_phys);
+ p_ramrod->output_params_addr.lo = DMA_LO_LE(ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err;
+
+ *num_bound_mw = le32_to_cpu(p_ramrod_res->num_bound_mw);
+
+ /* Free ORQ - only if ramrod succeeded, in case FW is still using it */
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+ qp->orq_num_pages * RDMA_RING_PAGE_SIZE,
+ qp->orq, qp->orq_phys_addr);
+
+ qp->req_offloaded = false;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Destroy requester, rc = %d\n", rc);
+
+err:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_ramrod_res),
+ p_ramrod_res, ramrod_res_phys);
+
+ return rc;
+}
+
+int qed_roce_query_qp(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *out_params)
+{
+ struct roce_query_qp_resp_output_params *p_resp_ramrod_res;
+ struct roce_query_qp_req_output_params *p_req_ramrod_res;
+ struct roce_query_qp_resp_ramrod_data *p_resp_ramrod;
+ struct roce_query_qp_req_ramrod_data *p_req_ramrod;
+ struct qed_sp_init_data init_data;
+ dma_addr_t resp_ramrod_res_phys;
+ dma_addr_t req_ramrod_res_phys;
+ struct qed_spq_entry *p_ent;
+ bool rq_err_state;
+ bool sq_err_state;
+ bool sq_draining;
+ int rc = -ENOMEM;
+
+ if ((!(qp->resp_offloaded)) && (!(qp->req_offloaded))) {
+ /* We can't send ramrod to the fw since this qp wasn't offloaded
+ * to the fw yet
+ */
+ out_params->draining = false;
+ out_params->rq_psn = qp->rq_psn;
+ out_params->sq_psn = qp->sq_psn;
+ out_params->state = qp->cur_state;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "No QPs as no offload\n");
+ return 0;
+ }
+
+ if (!(qp->resp_offloaded)) {
+ DP_NOTICE(p_hwfn,
+ "The responder's qp should be offloded before requester's\n");
+ return -EINVAL;
+ }
+
+ /* Send a query responder ramrod to FW to get RQ-PSN and state */
+ p_resp_ramrod_res = (struct roce_query_qp_resp_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(*p_resp_ramrod_res),
+ &resp_ramrod_res_phys, GFP_KERNEL);
+ if (!p_resp_ramrod_res) {
+ DP_NOTICE(p_hwfn,
+ "qed query qp failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.cid = qp->icid;
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_QUERY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err_resp;
+
+ p_resp_ramrod = &p_ent->ramrod.roce_query_qp_resp;
+ p_resp_ramrod->output_params_addr.hi = DMA_HI_LE(resp_ramrod_res_phys);
+ p_resp_ramrod->output_params_addr.lo = DMA_LO_LE(resp_ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err_resp;
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_resp_ramrod_res),
+ p_resp_ramrod_res, resp_ramrod_res_phys);
+
+ out_params->rq_psn = le32_to_cpu(p_resp_ramrod_res->psn);
+ rq_err_state = GET_FIELD(le32_to_cpu(p_resp_ramrod_res->err_flag),
+ ROCE_QUERY_QP_RESP_OUTPUT_PARAMS_ERROR_FLG);
+
+ if (!(qp->req_offloaded)) {
+ /* Don't send query qp for the requester */
+ out_params->sq_psn = qp->sq_psn;
+ out_params->draining = false;
+
+ if (rq_err_state)
+ qp->cur_state = QED_ROCE_QP_STATE_ERR;
+
+ out_params->state = qp->cur_state;
+
+ return 0;
+ }
+
+ /* Send a query requester ramrod to FW to get SQ-PSN and state */
+ p_req_ramrod_res = (struct roce_query_qp_req_output_params *)
+ dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+ sizeof(*p_req_ramrod_res),
+ &req_ramrod_res_phys,
+ GFP_KERNEL);
+ if (!p_req_ramrod_res) {
+ rc = -ENOMEM;
+ DP_NOTICE(p_hwfn,
+ "qed query qp failed: cannot allocate memory (ramrod)\n");
+ return rc;
+ }
+
+ /* Get SPQ entry */
+ init_data.cid = qp->icid + 1;
+ rc = qed_sp_init_request(p_hwfn, &p_ent, ROCE_RAMROD_QUERY_QP,
+ PROTOCOLID_ROCE, &init_data);
+ if (rc)
+ goto err_req;
+
+ p_req_ramrod = &p_ent->ramrod.roce_query_qp_req;
+ p_req_ramrod->output_params_addr.hi = DMA_HI_LE(req_ramrod_res_phys);
+ p_req_ramrod->output_params_addr.lo = DMA_LO_LE(req_ramrod_res_phys);
+
+ rc = qed_spq_post(p_hwfn, p_ent, NULL);
+ if (rc)
+ goto err_req;
+
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_req_ramrod_res),
+ p_req_ramrod_res, req_ramrod_res_phys);
+
+ out_params->sq_psn = le32_to_cpu(p_req_ramrod_res->psn);
+ sq_err_state = GET_FIELD(le32_to_cpu(p_req_ramrod_res->flags),
+ ROCE_QUERY_QP_REQ_OUTPUT_PARAMS_ERR_FLG);
+ sq_draining =
+ GET_FIELD(le32_to_cpu(p_req_ramrod_res->flags),
+ ROCE_QUERY_QP_REQ_OUTPUT_PARAMS_SQ_DRAINING_FLG);
+
+ out_params->draining = false;
+
+ if (rq_err_state)
+ qp->cur_state = QED_ROCE_QP_STATE_ERR;
+ else if (sq_err_state)
+ qp->cur_state = QED_ROCE_QP_STATE_SQE;
+ else if (sq_draining)
+ out_params->draining = true;
+ out_params->state = qp->cur_state;
+
+ return 0;
+
+err_req:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_req_ramrod_res),
+ p_req_ramrod_res, req_ramrod_res_phys);
+ return rc;
+err_resp:
+ dma_free_coherent(&p_hwfn->cdev->pdev->dev, sizeof(*p_resp_ramrod_res),
+ p_resp_ramrod_res, resp_ramrod_res_phys);
+ return rc;
+}
+
+int qed_roce_destroy_qp(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
+{
+ u32 num_invalidated_mw = 0;
+ u32 num_bound_mw = 0;
+ u32 start_cid;
+ int rc;
+
+ /* Destroys the specified QP */
+ if ((qp->cur_state != QED_ROCE_QP_STATE_RESET) &&
+ (qp->cur_state != QED_ROCE_QP_STATE_ERR) &&
+ (qp->cur_state != QED_ROCE_QP_STATE_INIT)) {
+ DP_NOTICE(p_hwfn,
+ "QP must be in error, reset or init state before destroying it\n");
+ return -EINVAL;
+ }
+
+ rc = qed_roce_sp_destroy_qp_responder(p_hwfn, qp, &num_invalidated_mw);
+ if (rc)
+ return rc;
+
+ /* Send destroy requester ramrod */
+ rc = qed_roce_sp_destroy_qp_requester(p_hwfn, qp, &num_bound_mw);
+ if (rc)
+ return rc;
+
+ if (num_invalidated_mw != num_bound_mw) {
+ DP_NOTICE(p_hwfn,
+ "number of invalidate memory windows is different from bounded ones\n");
+ return -EINVAL;
+ }
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+
+ start_cid = qed_cxt_get_proto_cid_start(p_hwfn,
+ p_hwfn->p_rdma_info->proto);
+
+ /* Release responder's icid */
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map,
+ qp->icid - start_cid);
+
+ /* Release requester's icid */
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map,
+ qp->icid + 1 - start_cid);
+
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+ return 0;
+}
+
+int qed_rdma_query_qp(void *rdma_cxt,
+ struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ /* The following fields are filled in from qp and not FW as they can't
+ * be modified by FW
+ */
+ out_params->mtu = qp->mtu;
+ out_params->dest_qp = qp->dest_qp;
+ out_params->incoming_atomic_en = qp->incoming_atomic_en;
+ out_params->e2e_flow_control_en = qp->e2e_flow_control_en;
+ out_params->incoming_rdma_read_en = qp->incoming_rdma_read_en;
+ out_params->incoming_rdma_write_en = qp->incoming_rdma_write_en;
+ out_params->dgid = qp->dgid;
+ out_params->flow_label = qp->flow_label;
+ out_params->hop_limit_ttl = qp->hop_limit_ttl;
+ out_params->traffic_class_tos = qp->traffic_class_tos;
+ out_params->timeout = qp->ack_timeout;
+ out_params->rnr_retry = qp->rnr_retry_cnt;
+ out_params->retry_cnt = qp->retry_cnt;
+ out_params->min_rnr_nak_timer = qp->min_rnr_nak_timer;
+ out_params->pkey_index = 0;
+ out_params->max_rd_atomic = qp->max_rd_atomic_req;
+ out_params->max_dest_rd_atomic = qp->max_rd_atomic_resp;
+ out_params->sqd_async = qp->sqd_async;
+
+ rc = qed_roce_query_qp(p_hwfn, qp, out_params);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Query QP, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_destroy_qp(void *rdma_cxt, struct qed_rdma_qp *qp)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ int rc = 0;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x\n", qp->icid);
+
+ rc = qed_roce_destroy_qp(p_hwfn, qp);
+
+ /* free qp params struct */
+ kfree(qp);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP destroyed\n");
+ return rc;
+}
+
+struct qed_rdma_qp *
+qed_rdma_create_qp(void *rdma_cxt,
+ struct qed_rdma_create_qp_in_params *in_params,
+ struct qed_rdma_create_qp_out_params *out_params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct qed_rdma_qp *qp;
+ u8 max_stats_queues;
+ int rc;
+
+ if (!rdma_cxt || !in_params || !out_params || !p_hwfn->p_rdma_info) {
+ DP_ERR(p_hwfn->cdev,
+ "qed roce create qp failed due to NULL entry (rdma_cxt=%p, in=%p, out=%p, roce_info=?\n",
+ rdma_cxt, in_params, out_params);
+ return NULL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "qed rdma create qp called with qp_handle = %08x%08x\n",
+ in_params->qp_handle_hi, in_params->qp_handle_lo);
+
+ /* Some sanity checks... */
+ max_stats_queues = p_hwfn->p_rdma_info->dev->max_stats_queues;
+ if (in_params->stats_queue >= max_stats_queues) {
+ DP_ERR(p_hwfn->cdev,
+ "qed rdma create qp failed due to invalid statistics queue %d. maximum is %d\n",
+ in_params->stats_queue, max_stats_queues);
+ return NULL;
+ }
+
+ qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+ if (!qp) {
+ DP_NOTICE(p_hwfn, "Failed to allocate qed_rdma_qp\n");
+ return NULL;
+ }
+
+ rc = qed_roce_alloc_cid(p_hwfn, &qp->icid);
+ qp->qpid = ((0xFF << 16) | qp->icid);
+
+ DP_INFO(p_hwfn, "ROCE qpid=%x\n", qp->qpid);
+
+ if (rc) {
+ kfree(qp);
+ return NULL;
+ }
+
+ qp->cur_state = QED_ROCE_QP_STATE_RESET;
+ qp->qp_handle.hi = cpu_to_le32(in_params->qp_handle_hi);
+ qp->qp_handle.lo = cpu_to_le32(in_params->qp_handle_lo);
+ qp->qp_handle_async.hi = cpu_to_le32(in_params->qp_handle_async_hi);
+ qp->qp_handle_async.lo = cpu_to_le32(in_params->qp_handle_async_lo);
+ qp->use_srq = in_params->use_srq;
+ qp->signal_all = in_params->signal_all;
+ qp->fmr_and_reserved_lkey = in_params->fmr_and_reserved_lkey;
+ qp->pd = in_params->pd;
+ qp->dpi = in_params->dpi;
+ qp->sq_cq_id = in_params->sq_cq_id;
+ qp->sq_num_pages = in_params->sq_num_pages;
+ qp->sq_pbl_ptr = in_params->sq_pbl_ptr;
+ qp->rq_cq_id = in_params->rq_cq_id;
+ qp->rq_num_pages = in_params->rq_num_pages;
+ qp->rq_pbl_ptr = in_params->rq_pbl_ptr;
+ qp->srq_id = in_params->srq_id;
+ qp->req_offloaded = false;
+ qp->resp_offloaded = false;
+ /* e2e_flow_control cannot be done in case of S-RQ.
+ * Refer to 9.7.7.2 End-to-End Flow Control section of IB spec
+ */
+ qp->e2e_flow_control_en = qp->use_srq ? false : true;
+ qp->stats_queue = in_params->stats_queue;
+
+ out_params->icid = qp->icid;
+ out_params->qp_id = qp->qpid;
+
+ /* max_sq_sges future use only */
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Create QP, rc = %d\n", rc);
+ return qp;
+}
+
+static int qed_roce_modify_qp(struct qed_hwfn *p_hwfn,
+ struct qed_rdma_qp *qp,
+ enum qed_roce_qp_state prev_state,
+ struct qed_rdma_modify_qp_in_params *params)
+{
+ u32 num_invalidated_mw = 0, num_bound_mw = 0;
+ int rc = 0;
+
+ /* Perform additional operations according to the current state and the
+ * next state
+ */
+ if (((prev_state == QED_ROCE_QP_STATE_INIT) ||
+ (prev_state == QED_ROCE_QP_STATE_RESET)) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTR)) {
+ /* Init->RTR or Reset->RTR */
+ rc = qed_roce_sp_create_responder(p_hwfn, qp);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_RTR) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
+ /* RTR-> RTS */
+ rc = qed_roce_sp_create_requester(p_hwfn, qp);
+ if (rc)
+ return rc;
+
+ /* Send modify responder ramrod */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_RTS) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
+ /* RTS->RTS */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_RTS) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_SQD)) {
+ /* RTS->SQD */
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, true, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_SQD) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_SQD)) {
+ /* SQD->SQD */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
+ params->modify_flags);
+ return rc;
+ } else if ((prev_state == QED_ROCE_QP_STATE_SQD) &&
+ (qp->cur_state == QED_ROCE_QP_STATE_RTS)) {
+ /* SQD->RTS */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, false,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, false,
+ params->modify_flags);
+
+ return rc;
+ } else if (qp->cur_state == QED_ROCE_QP_STATE_ERR ||
+ qp->cur_state == QED_ROCE_QP_STATE_SQE) {
+ /* ->ERR */
+ rc = qed_roce_sp_modify_responder(p_hwfn, qp, true,
+ params->modify_flags);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_modify_requester(p_hwfn, qp, false, true,
+ params->modify_flags);
+ return rc;
+ } else if (qp->cur_state == QED_ROCE_QP_STATE_RESET) {
+ /* Any state -> RESET */
+
+ rc = qed_roce_sp_destroy_qp_responder(p_hwfn, qp,
+ &num_invalidated_mw);
+ if (rc)
+ return rc;
+
+ rc = qed_roce_sp_destroy_qp_requester(p_hwfn, qp,
+ &num_bound_mw);
+
+ if (num_invalidated_mw != num_bound_mw) {
+ DP_NOTICE(p_hwfn,
+ "number of invalidate memory windows is different from bounded ones\n");
+ return -EINVAL;
+ }
+ } else {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "0\n");
+ }
+
+ return rc;
+}
+
+int qed_rdma_modify_qp(void *rdma_cxt,
+ struct qed_rdma_qp *qp,
+ struct qed_rdma_modify_qp_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ enum qed_roce_qp_state prev_state;
+ int rc = 0;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "icid = %08x params->new_state=%d\n",
+ qp->icid, params->new_state);
+
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN)) {
+ qp->incoming_rdma_read_en = params->incoming_rdma_read_en;
+ qp->incoming_rdma_write_en = params->incoming_rdma_write_en;
+ qp->incoming_atomic_en = params->incoming_atomic_en;
+ }
+
+ /* Update QP structure with the updated values */
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_ROCE_MODE))
+ qp->roce_mode = params->roce_mode;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_PKEY))
+ qp->pkey = params->pkey;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_E2E_FLOW_CONTROL_EN))
+ qp->e2e_flow_control_en = params->e2e_flow_control_en;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_DEST_QP))
+ qp->dest_qp = params->dest_qp;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR)) {
+ /* Indicates that the following parameters have changed:
+ * Traffic class, flow label, hop limit, source GID,
+ * destination GID, loopback indicator
+ */
+ qp->traffic_class_tos = params->traffic_class_tos;
+ qp->flow_label = params->flow_label;
+ qp->hop_limit_ttl = params->hop_limit_ttl;
+
+ qp->sgid = params->sgid;
+ qp->dgid = params->dgid;
+ qp->udp_src_port = 0;
+ qp->vlan_id = params->vlan_id;
+ qp->mtu = params->mtu;
+ qp->lb_indication = params->lb_indication;
+ memcpy((u8 *)&qp->remote_mac_addr[0],
+ (u8 *)¶ms->remote_mac_addr[0], ETH_ALEN);
+ if (params->use_local_mac) {
+ memcpy((u8 *)&qp->local_mac_addr[0],
+ (u8 *)¶ms->local_mac_addr[0], ETH_ALEN);
+ } else {
+ memcpy((u8 *)&qp->local_mac_addr[0],
+ (u8 *)&p_hwfn->hw_info.hw_mac_addr, ETH_ALEN);
+ }
+ }
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_RQ_PSN))
+ qp->rq_psn = params->rq_psn;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_SQ_PSN))
+ qp->sq_psn = params->sq_psn;
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ))
+ qp->max_rd_atomic_req = params->max_rd_atomic_req;
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP))
+ qp->max_rd_atomic_resp = params->max_rd_atomic_resp;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT))
+ qp->ack_timeout = params->ack_timeout;
+ if (GET_FIELD(params->modify_flags, QED_ROCE_MODIFY_QP_VALID_RETRY_CNT))
+ qp->retry_cnt = params->retry_cnt;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT))
+ qp->rnr_retry_cnt = params->rnr_retry_cnt;
+ if (GET_FIELD(params->modify_flags,
+ QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER))
+ qp->min_rnr_nak_timer = params->min_rnr_nak_timer;
+
+ qp->sqd_async = params->sqd_async;
+
+ prev_state = qp->cur_state;
+ if (GET_FIELD(params->modify_flags,
+ QED_RDMA_MODIFY_QP_VALID_NEW_STATE)) {
+ qp->cur_state = params->new_state;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "qp->cur_state=%d\n",
+ qp->cur_state);
+ }
+
+ rc = qed_roce_modify_qp(p_hwfn, qp, prev_state, params);
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Modify QP, rc = %d\n", rc);
+ return rc;
+}
+
static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
{
return QED_LEADING_HWFN(cdev);
@@ -1208,6 +2415,10 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_dealloc_pd = &qed_rdma_free_pd,
.rdma_create_cq = &qed_rdma_create_cq,
.rdma_destroy_cq = &qed_rdma_destroy_cq,
+ .rdma_create_qp = &qed_rdma_create_qp,
+ .rdma_modify_qp = &qed_rdma_modify_qp,
+ .rdma_query_qp = &qed_rdma_query_qp,
+ .rdma_destroy_qp = &qed_rdma_destroy_qp,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 6802cd0..71b37fc 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -114,6 +114,72 @@ struct qed_rdma_resize_cnq_in_params {
u64 pbl_ptr;
};
+struct qed_rdma_qp {
+ struct regpair qp_handle;
+ struct regpair qp_handle_async;
+ u32 qpid;
+ u16 icid;
+ enum qed_roce_qp_state cur_state;
+ bool use_srq;
+ bool signal_all;
+ bool fmr_and_reserved_lkey;
+
+ bool incoming_rdma_read_en;
+ bool incoming_rdma_write_en;
+ bool incoming_atomic_en;
+ bool e2e_flow_control_en;
+
+ u16 pd;
+ u16 pkey;
+ u32 dest_qp;
+ u16 mtu;
+ u16 srq_id;
+ u8 traffic_class_tos;
+ u8 hop_limit_ttl;
+ u16 dpi;
+ u32 flow_label;
+ bool lb_indication;
+ u16 vlan_id;
+ u32 ack_timeout;
+ u8 retry_cnt;
+ u8 rnr_retry_cnt;
+ u8 min_rnr_nak_timer;
+ bool sqd_async;
+ union qed_gid sgid;
+ union qed_gid dgid;
+ enum roce_mode roce_mode;
+ u16 udp_src_port;
+ u8 stats_queue;
+
+ /* requeseter */
+ u8 max_rd_atomic_req;
+ u32 sq_psn;
+ u16 sq_cq_id;
+ u16 sq_num_pages;
+ dma_addr_t sq_pbl_ptr;
+ void *orq;
+ dma_addr_t orq_phys_addr;
+ u8 orq_num_pages;
+ bool req_offloaded;
+
+ /* responder */
+ u8 max_rd_atomic_resp;
+ u32 rq_psn;
+ u16 rq_cq_id;
+ u16 rq_num_pages;
+ dma_addr_t rq_pbl_ptr;
+ void *irq;
+ dma_addr_t irq_phys_addr;
+ u8 irq_num_pages;
+ bool resp_offloaded;
+
+ u8 remote_mac_addr[6];
+ u8 local_mac_addr[6];
+
+ void *shared_queue;
+ dma_addr_t shared_queue_phys_addr;
+};
+
int
qed_rdma_add_user(void *rdma_cxt,
struct qed_rdma_add_user_out_params *out_params);
@@ -135,4 +201,9 @@ void qed_rdma_cnq_prod_update(void *rdma_cxt, u8 cnq_index, u16 prod);
void qed_rdma_resc_free(struct qed_hwfn *p_hwfn);
void qed_async_roce_event(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe);
+int qed_rdma_destroy_qp(void *rdma_cxt, struct qed_rdma_qp *qp);
+int qed_rdma_modify_qp(void *rdma_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_modify_qp_in_params *params);
+int qed_rdma_query_qp(void *rdma_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *out_params);
#endif
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index b559b1c..02321e3 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -43,6 +43,17 @@
#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
/* rdma interface */
+
+enum qed_roce_qp_state {
+ QED_ROCE_QP_STATE_RESET,
+ QED_ROCE_QP_STATE_INIT,
+ QED_ROCE_QP_STATE_RTR,
+ QED_ROCE_QP_STATE_RTS,
+ QED_ROCE_QP_STATE_SQD,
+ QED_ROCE_QP_STATE_ERR,
+ QED_ROCE_QP_STATE_SQE
+};
+
enum qed_rdma_tid_type {
QED_RDMA_TID_REGISTERED_MR,
QED_RDMA_TID_FMR,
@@ -292,6 +303,128 @@ struct qed_rdma_destroy_cq_out_params {
u16 num_cq_notif;
};
+struct qed_rdma_create_qp_in_params {
+ u32 qp_handle_lo;
+ u32 qp_handle_hi;
+ u32 qp_handle_async_lo;
+ u32 qp_handle_async_hi;
+ bool use_srq;
+ bool signal_all;
+ bool fmr_and_reserved_lkey;
+ u16 pd;
+ u16 dpi;
+ u16 sq_cq_id;
+ u16 sq_num_pages;
+ u64 sq_pbl_ptr;
+ u8 max_sq_sges;
+ u16 rq_cq_id;
+ u16 rq_num_pages;
+ u64 rq_pbl_ptr;
+ u16 srq_id;
+ u8 stats_queue;
+};
+
+struct qed_rdma_create_qp_out_params {
+ u32 qp_id;
+ u16 icid;
+ void *rq_pbl_virt;
+ dma_addr_t rq_pbl_phys;
+ void *sq_pbl_virt;
+ dma_addr_t sq_pbl_phys;
+};
+
+struct qed_rdma_modify_qp_in_params {
+ u32 modify_flags;
+#define QED_RDMA_MODIFY_QP_VALID_NEW_STATE_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_NEW_STATE_SHIFT 0
+#define QED_ROCE_MODIFY_QP_VALID_PKEY_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_PKEY_SHIFT 1
+#define QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN_SHIFT 2
+#define QED_ROCE_MODIFY_QP_VALID_DEST_QP_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_DEST_QP_SHIFT 3
+#define QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR_SHIFT 4
+#define QED_ROCE_MODIFY_QP_VALID_RQ_PSN_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_RQ_PSN_SHIFT 5
+#define QED_ROCE_MODIFY_QP_VALID_SQ_PSN_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_SQ_PSN_SHIFT 6
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ_SHIFT 7
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP_MASK 0x1
+#define QED_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP_SHIFT 8
+#define QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT_SHIFT 9
+#define QED_ROCE_MODIFY_QP_VALID_RETRY_CNT_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_RETRY_CNT_SHIFT 10
+#define QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT_SHIFT 11
+#define QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER_SHIFT 12
+#define QED_ROCE_MODIFY_QP_VALID_E2E_FLOW_CONTROL_EN_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_E2E_FLOW_CONTROL_EN_SHIFT 13
+#define QED_ROCE_MODIFY_QP_VALID_ROCE_MODE_MASK 0x1
+#define QED_ROCE_MODIFY_QP_VALID_ROCE_MODE_SHIFT 14
+
+ enum qed_roce_qp_state new_state;
+ u16 pkey;
+ bool incoming_rdma_read_en;
+ bool incoming_rdma_write_en;
+ bool incoming_atomic_en;
+ bool e2e_flow_control_en;
+ u32 dest_qp;
+ bool lb_indication;
+ u16 mtu;
+ u8 traffic_class_tos;
+ u8 hop_limit_ttl;
+ u32 flow_label;
+ union qed_gid sgid;
+ union qed_gid dgid;
+ u16 udp_src_port;
+
+ u16 vlan_id;
+
+ u32 rq_psn;
+ u32 sq_psn;
+ u8 max_rd_atomic_resp;
+ u8 max_rd_atomic_req;
+ u32 ack_timeout;
+ u8 retry_cnt;
+ u8 rnr_retry_cnt;
+ u8 min_rnr_nak_timer;
+ bool sqd_async;
+ u8 remote_mac_addr[6];
+ u8 local_mac_addr[6];
+ bool use_local_mac;
+ enum roce_mode roce_mode;
+};
+
+struct qed_rdma_query_qp_out_params {
+ enum qed_roce_qp_state state;
+ u32 rq_psn;
+ u32 sq_psn;
+ bool draining;
+ u16 mtu;
+ u32 dest_qp;
+ bool incoming_rdma_read_en;
+ bool incoming_rdma_write_en;
+ bool incoming_atomic_en;
+ bool e2e_flow_control_en;
+ union qed_gid sgid;
+ union qed_gid dgid;
+ u32 flow_label;
+ u8 hop_limit_ttl;
+ u8 traffic_class_tos;
+ u32 timeout;
+ u8 rnr_retry;
+ u8 retry_cnt;
+ u8 min_rnr_nak_timer;
+ u16 pkey_index;
+ u8 max_rd_atomic;
+ u8 max_dest_rd_atomic;
+ bool sqd_async;
+};
+
struct qed_rdma_create_srq_out_params {
u16 srq_id;
};
@@ -368,6 +501,17 @@ struct qed_rdma_ops {
int (*rdma_destroy_cq)(void *rdma_cxt,
struct qed_rdma_destroy_cq_in_params *iparams,
struct qed_rdma_destroy_cq_out_params *oparams);
+ struct qed_rdma_qp *
+ (*rdma_create_qp)(void *rdma_cxt,
+ struct qed_rdma_create_qp_in_params *iparams,
+ struct qed_rdma_create_qp_out_params *oparams);
+
+ int (*rdma_modify_qp)(void *roce_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_modify_qp_in_params *iparams);
+
+ int (*rdma_query_qp)(void *rdma_cxt, struct qed_rdma_qp *qp,
+ struct qed_rdma_query_qp_out_params *oparams);
+ int (*rdma_destroy_qp)(void *rdma_cxt, struct qed_rdma_qp *qp);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 07/11] Add support for memory registeration verbs
[not found] ` <1473696465-27986-1-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` Ram Amrani
` (4 subsequent siblings)
5 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
Ram Amrani
Add support for user, dma and memory regions registration.
Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/qedr/main.c | 13 +-
drivers/infiniband/hw/qedr/qedr.h | 45 ++++
drivers/infiniband/hw/qedr/verbs.c | 385 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 14 ++
drivers/net/ethernet/qlogic/qed/qed_roce.c | 240 ++++++++++++++++++
include/linux/qed/qed_roce_if.h | 6 +
6 files changed, 702 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 722b895..3d4feb7 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -106,7 +106,9 @@ static int qedr_register_device(struct qedr_dev *dev)
QEDR_UVERBS(CREATE_QP) |
QEDR_UVERBS(MODIFY_QP) |
QEDR_UVERBS(QUERY_QP) |
- QEDR_UVERBS(DESTROY_QP);
+ QEDR_UVERBS(DESTROY_QP) |
+ QEDR_UVERBS(REG_MR) |
+ QEDR_UVERBS(DEREG_MR);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -138,6 +140,15 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.query_pkey = qedr_query_pkey;
+ dev->ibdev.create_ah = qedr_create_ah;
+ dev->ibdev.destroy_ah = qedr_destroy_ah;
+
+ dev->ibdev.get_dma_mr = qedr_get_dma_mr;
+ dev->ibdev.dereg_mr = qedr_dereg_mr;
+ dev->ibdev.reg_user_mr = qedr_reg_user_mr;
+ dev->ibdev.alloc_mr = qedr_alloc_mr;
+ dev->ibdev.map_mr_sg = qedr_map_mr_sg;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 1c9cc40..3cd4e52 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -374,6 +374,41 @@ struct qedr_qp {
struct qedr_userq urq;
};
+struct qedr_ah {
+ struct ib_ah ibah;
+ struct ib_ah_attr attr;
+};
+
+enum qedr_mr_type {
+ QEDR_MR_USER,
+ QEDR_MR_KERNEL,
+ QEDR_MR_DMA,
+ QEDR_MR_FRMR
+};
+
+struct mr_info {
+ struct qedr_pbl *pbl_table;
+ struct qedr_pbl_info pbl_info;
+ struct list_head free_pbl_list;
+ struct list_head inuse_pbl_list;
+ u32 completed;
+ u32 completed_handled;
+};
+
+struct qedr_mr {
+ struct ib_mr ibmr;
+ struct ib_umem *umem;
+
+ struct qed_rdma_register_tid_in_params hw_mr;
+ enum qedr_mr_type type;
+
+ struct qedr_dev *dev;
+ struct mr_info info;
+
+ u64 *pages;
+ u32 npages;
+};
+
static inline int qedr_get_dmac(struct qedr_dev *dev,
struct ib_ah_attr *ah_attr, u8 *mac_addr)
{
@@ -417,4 +452,14 @@ static inline struct qedr_qp *get_qedr_qp(struct ib_qp *ibqp)
{
return container_of(ibqp, struct qedr_qp, ibqp);
}
+
+static inline struct qedr_ah *get_qedr_ah(struct ib_ah *ibah)
+{
+ return container_of(ibah, struct qedr_ah, ibah);
+}
+
+static inline struct qedr_mr *get_qedr_mr(struct ib_mr *ibmr)
+{
+ return container_of(ibmr, struct qedr_mr, ibmr);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 1d16c34..3d6045c 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -2054,3 +2054,388 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
return rc;
}
+
+struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr)
+{
+ struct qedr_ah *ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
+
+ ah->attr = *attr;
+
+ return &ah->ibah;
+}
+
+int qedr_destroy_ah(struct ib_ah *ibah)
+{
+ struct qedr_ah *ah = get_qedr_ah(ibah);
+
+ kfree(ah);
+ return 0;
+}
+
+static void free_mr_info(struct qedr_dev *dev, struct mr_info *info)
+{
+ struct qedr_pbl *pbl, *tmp;
+
+ if (info->pbl_table)
+ list_add_tail(&info->pbl_table->list_entry,
+ &info->free_pbl_list);
+
+ if (!list_empty(&info->inuse_pbl_list))
+ list_splice(&info->inuse_pbl_list, &info->free_pbl_list);
+
+ list_for_each_entry_safe(pbl, tmp, &info->free_pbl_list, list_entry) {
+ list_del(&pbl->list_entry);
+ qedr_free_pbl(dev, &info->pbl_info, pbl);
+ }
+}
+
+static int init_mr_info(struct qedr_dev *dev, struct mr_info *info,
+ size_t page_list_len, bool two_layered)
+{
+ struct qedr_pbl *tmp;
+ int rc;
+
+ INIT_LIST_HEAD(&info->free_pbl_list);
+ INIT_LIST_HEAD(&info->inuse_pbl_list);
+
+ rc = qedr_prepare_pbl_tbl(dev, &info->pbl_info,
+ page_list_len, two_layered);
+ if (rc)
+ goto done;
+
+ info->pbl_table = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
+ if (!info->pbl_table) {
+ rc = -ENOMEM;
+ goto done;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "pbl_table_pa = %pa\n",
+ &info->pbl_table->pa);
+
+ /* in usual case we use 2 PBLs, so we add one to free
+ * list and allocating another one
+ */
+ tmp = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
+ if (!tmp) {
+ DP_VERBOSE(dev, QEDR_MSG_MR, "Extra PBL is not allocated\n");
+ goto done;
+ }
+
+ list_add_tail(&tmp->list_entry, &info->free_pbl_list);
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "extra pbl_table_pa = %pa\n", &tmp->pa);
+
+done:
+ if (rc)
+ free_mr_info(dev, info);
+
+ return rc;
+}
+
+struct ib_mr *qedr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len,
+ u64 usr_addr, int acc, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_mr *mr;
+ struct qedr_pd *pd;
+ int rc = -ENOMEM;
+
+ pd = get_qedr_pd(ibpd);
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "qedr_register user mr pd = %d start = %lld, len = %lld, usr_addr = %lld, acc = %d\n",
+ pd->pd_id, start, len, usr_addr, acc);
+
+ if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE))
+ return ERR_PTR(-EINVAL);
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(rc);
+
+ mr->type = QEDR_MR_USER;
+
+ mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0);
+ if (IS_ERR(mr->umem)) {
+ rc = -EFAULT;
+ goto err0;
+ }
+
+ rc = init_mr_info(dev, &mr->info, ib_umem_page_count(mr->umem), 1);
+ if (rc)
+ goto err1;
+
+ qedr_populate_pbls(dev, mr->umem, mr->info.pbl_table,
+ &mr->info.pbl_info);
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ /* Index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR;
+ mr->hw_mr.key = 0;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+ mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+ mr->hw_mr.mw_bind = false;
+ mr->hw_mr.pbl_ptr = mr->info.pbl_table[0].pa;
+ mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
+ mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
+ mr->hw_mr.page_size_log = ilog2(mr->umem->page_size);
+ mr->hw_mr.fbo = ib_umem_offset(mr->umem);
+ mr->hw_mr.length = len;
+ mr->hw_mr.vaddr = usr_addr;
+ mr->hw_mr.zbva = false;
+ mr->hw_mr.phy_mr = false;
+ mr->hw_mr.dma_mr = false;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err2;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
+ mr->hw_mr.remote_atomic)
+ mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "register user mr lkey: %x\n",
+ mr->ibmr.lkey);
+ return &mr->ibmr;
+
+err2:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err1:
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+err0:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
+
+int qedr_dereg_mr(struct ib_mr *ib_mr)
+{
+ struct qedr_mr *mr = get_qedr_mr(ib_mr);
+ struct qedr_dev *dev = get_qedr_dev(ib_mr->device);
+ int rc = 0;
+
+ rc = dev->ops->rdma_deregister_tid(dev->rdma_ctx, mr->hw_mr.itid);
+ if (rc)
+ return rc;
+
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+
+ if ((mr->type != QEDR_MR_DMA) && (mr->type != QEDR_MR_FRMR))
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+
+ /* it could be user registered memory. */
+ if (mr->umem)
+ ib_umem_release(mr->umem);
+
+ kfree(mr->pages);
+ kfree(mr);
+
+ return rc;
+}
+
+struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int max_page_list_len)
+{
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_mr *mr;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd->pd_id,
+ max_page_list_len);
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(rc);
+
+ mr->dev = dev;
+ mr->type = QEDR_MR_FRMR;
+
+ rc = init_mr_info(dev, &mr->info, max_page_list_len, 1);
+ if (rc)
+ goto err0;
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err0;
+ }
+
+ /* Index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_FMR;
+ mr->hw_mr.key = 0;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = 0;
+ mr->hw_mr.remote_read = 0;
+ mr->hw_mr.remote_write = 0;
+ mr->hw_mr.remote_atomic = 0;
+ mr->hw_mr.mw_bind = false;
+ mr->hw_mr.pbl_ptr = 0;
+ mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
+ mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
+ mr->hw_mr.fbo = 0;
+ mr->hw_mr.length = 0;
+ mr->hw_mr.vaddr = 0;
+ mr->hw_mr.zbva = false;
+ mr->hw_mr.phy_mr = true;
+ mr->hw_mr.dma_mr = false;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ mr->ibmr.rkey = mr->ibmr.lkey;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "alloc frmr: %x\n", mr->ibmr.lkey);
+ return mr;
+
+err1:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err0:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
+
+struct ib_mr *qedr_alloc_mr(struct ib_pd *ibpd,
+ enum ib_mr_type mr_type, u32 max_num_sg)
+{
+ struct qedr_dev *dev;
+ struct qedr_mr *mr;
+
+ if (mr_type != IB_MR_TYPE_MEM_REG)
+ return ERR_PTR(-EINVAL);
+
+ mr = __qedr_alloc_mr(ibpd, max_num_sg);
+
+ if (IS_ERR(mr))
+ return ERR_PTR(-EINVAL);
+
+ dev = mr->dev;
+ mr->pages = kcalloc(max_num_sg, sizeof(u64), GFP_KERNEL);
+ if (!mr->pages) {
+ dev->ops->rdma_deregister_tid(dev->rdma_ctx, mr->hw_mr.itid);
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+ return ERR_PTR(-EINVAL);
+ }
+
+ return &mr->ibmr;
+}
+
+static int qedr_set_page(struct ib_mr *ibmr, u64 addr)
+{
+ struct qedr_mr *mr = get_qedr_mr(ibmr);
+
+ if (unlikely(mr->npages == mr->info.pbl_info.num_pbes)) {
+ DP_ERR(mr->dev, "qedr_set_page failes when %d\n", mr->npages);
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(mr->dev, QEDR_MSG_MR, "qedr_set_page pages[%d] = 0x%llx\n\n",
+ mr->npages, addr);
+
+ mr->pages[mr->npages++] = addr;
+
+ return 0;
+}
+
+static void handle_completed_mrs(struct qedr_dev *dev, struct mr_info *info)
+{
+ int work = info->completed - info->completed_handled - 1;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "Special FMR work = %d\n", work);
+ while (work-- > 0 && !list_empty(&info->inuse_pbl_list)) {
+ struct qedr_pbl *pbl;
+
+ /* Free all the page list that are possible to be freed
+ * (all the ones that were invalidated), under the assumption
+ * that if an FMR was completed successfully that means that
+ * if there was an invalidate operation before it also ended
+ */
+ pbl = list_first_entry(&info->inuse_pbl_list,
+ struct qedr_pbl, list_entry);
+ list_del(&pbl->list_entry);
+ list_add_tail(&pbl->list_entry, &info->free_pbl_list);
+ info->completed_handled++;
+ }
+}
+
+int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset)
+{
+ struct qedr_mr *mr = get_qedr_mr(ibmr);
+
+ mr->npages = 0;
+
+ handle_completed_mrs(mr->dev, &mr->info);
+ return ib_sg_to_pages(ibmr, sg, sg_nents, NULL, qedr_set_page);
+}
+
+struct ib_mr *qedr_get_dma_mr(struct ib_pd *ibpd, int acc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct qedr_mr *mr;
+ int rc;
+
+ if (acc & IB_ACCESS_MW_BIND) {
+ DP_ERR(dev, "Unsupported access flags received for dma mr\n");
+ return ERR_PTR(-EINVAL);
+ }
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(-ENOMEM);
+
+ mr->type = QEDR_MR_DMA;
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ /* index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+ mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+ mr->hw_mr.dma_mr = true;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err2;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
+ mr->hw_mr.remote_atomic)
+ mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "get dma mr: lkey = %x\n", mr->ibmr.lkey);
+ return &mr->ibmr;
+
+err2:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err1:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index 80dbac1..e7cb3fd 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -44,4 +44,18 @@ int qedr_query_qp(struct ib_qp *, struct ib_qp_attr *qp_attr,
int qp_attr_mask, struct ib_qp_init_attr *);
int qedr_destroy_qp(struct ib_qp *ibqp);
+struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr);
+int qedr_destroy_ah(struct ib_ah *ibah);
+
+int qedr_dereg_mr(struct ib_mr *);
+struct ib_mr *qedr_get_dma_mr(struct ib_pd *, int acc);
+
+struct ib_mr *qedr_reg_user_mr(struct ib_pd *, u64 start, u64 length,
+ u64 virt, int acc, struct ib_udata *);
+
+int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset);
+
+struct ib_mr *qedr_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+ u32 max_num_sg);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index cf92ff7..04e33ee 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -690,6 +690,17 @@ struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
return p_hwfn->p_rdma_info->dev;
}
+void qed_rdma_free_tid(void *rdma_cxt, u32 itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->tid_map, itid);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
int qed_rdma_alloc_tid(void *rdma_cxt, u32 *itid)
{
struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
@@ -2321,6 +2332,231 @@ int qed_rdma_modify_qp(void *rdma_cxt,
return rc;
}
+int qed_rdma_register_tid(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_register_tid_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ enum rdma_tid_type tid_type;
+ u8 fw_return_code;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", params->itid);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_REGISTER_MR,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (p_hwfn->p_rdma_info->last_tid < params->itid)
+ p_hwfn->p_rdma_info->last_tid = params->itid;
+
+ p_ramrod = &p_ent->ramrod.rdma_register_tid;
+
+ p_ramrod->flags = 0;
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_TWO_LEVEL_PBL,
+ params->pbl_two_level);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_ZERO_BASED, params->zbva);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_PHY_MR, params->phy_mr);
+
+ /* Don't initialize D/C field, as it may override other bits. */
+ if (!(params->tid_type == QED_RDMA_TID_FMR) && !(params->dma_mr))
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_PAGE_SIZE_LOG,
+ params->page_size_log - 12);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_MAX_ID,
+ p_hwfn->p_rdma_info->last_tid);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_READ,
+ params->remote_read);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_WRITE,
+ params->remote_write);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_ATOMIC,
+ params->remote_atomic);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_WRITE,
+ params->local_write);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_READ, params->local_read);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_ENABLE_MW_BIND,
+ params->mw_bind);
+
+ SET_FIELD(p_ramrod->flags1,
+ RDMA_REGISTER_TID_RAMROD_DATA_PBL_PAGE_SIZE_LOG,
+ params->pbl_page_size_log - 12);
+
+ SET_FIELD(p_ramrod->flags2,
+ RDMA_REGISTER_TID_RAMROD_DATA_DMA_MR, params->dma_mr);
+
+ switch (params->tid_type) {
+ case QED_RDMA_TID_REGISTERED_MR:
+ tid_type = RDMA_TID_REGISTERED_MR;
+ break;
+ case QED_RDMA_TID_FMR:
+ tid_type = RDMA_TID_FMR;
+ break;
+ case QED_RDMA_TID_MW_TYPE1:
+ tid_type = RDMA_TID_MW_TYPE1;
+ break;
+ case QED_RDMA_TID_MW_TYPE2A:
+ tid_type = RDMA_TID_MW_TYPE2A;
+ break;
+ default:
+ rc = -EINVAL;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+ SET_FIELD(p_ramrod->flags1,
+ RDMA_REGISTER_TID_RAMROD_DATA_TID_TYPE, tid_type);
+
+ p_ramrod->itid = cpu_to_le32(params->itid);
+ p_ramrod->key = params->key;
+ p_ramrod->pd = cpu_to_le16(params->pd);
+ p_ramrod->length_hi = (u8)(params->length >> 32);
+ p_ramrod->length_lo = DMA_LO_LE(params->length);
+ if (params->zbva) {
+ /* Lower 32 bits of the registered MR address.
+ * In case of zero based MR, will hold FBO
+ */
+ p_ramrod->va.hi = 0;
+ p_ramrod->va.lo = cpu_to_le32(params->fbo);
+ } else {
+ DMA_REGPAIR_LE(p_ramrod->va, params->vaddr);
+ }
+ DMA_REGPAIR_LE(p_ramrod->pbl_base, params->pbl_ptr);
+
+ /* DIF */
+ if (params->dif_enabled) {
+ SET_FIELD(p_ramrod->flags2,
+ RDMA_REGISTER_TID_RAMROD_DATA_DIF_ON_HOST_FLG, 1);
+ DMA_REGPAIR_LE(p_ramrod->dif_error_addr,
+ params->dif_error_addr);
+ DMA_REGPAIR_LE(p_ramrod->dif_runt_addr, params->dif_runt_addr);
+ }
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Register TID, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_deregister_tid(void *rdma_cxt, u32 itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_deregister_tid_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ struct qed_ptt *p_ptt;
+ u8 fw_return_code;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_DEREGISTER_MR,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ p_ramrod = &p_ent->ramrod.rdma_deregister_tid;
+ p_ramrod->itid = cpu_to_le32(itid);
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (fw_return_code == RDMA_RETURN_DEREGISTER_MR_BAD_STATE_ERR) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ return -EINVAL;
+ } else if (fw_return_code == RDMA_RETURN_NIG_DRAIN_REQ) {
+ /* Bit indicating that the TID is in use and a nig drain is
+ * required before sending the ramrod again
+ */
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ rc = -EBUSY;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to acquire PTT\n");
+ return rc;
+ }
+
+ rc = qed_mcp_drain(p_hwfn, p_ptt);
+ if (rc) {
+ qed_ptt_release(p_hwfn, p_ptt);
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Drain failed\n");
+ return rc;
+ }
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ /* Resend the ramrod */
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_DEREGISTER_MR,
+ p_hwfn->p_rdma_info->proto,
+ &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to init sp-element\n");
+ return rc;
+ }
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Ramrod failed\n");
+ return rc;
+ }
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n",
+ fw_return_code);
+ return rc;
+ }
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "De-registered TID, rc = %d\n", rc);
+ return rc;
+}
+
static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
{
return QED_LEADING_HWFN(cdev);
@@ -2419,6 +2655,10 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_modify_qp = &qed_rdma_modify_qp,
.rdma_query_qp = &qed_rdma_query_qp,
.rdma_destroy_qp = &qed_rdma_destroy_qp,
+ .rdma_alloc_tid = &qed_rdma_alloc_tid,
+ .rdma_free_tid = &qed_rdma_free_tid,
+ .rdma_register_tid = &qed_rdma_register_tid,
+ .rdma_deregister_tid = &qed_rdma_deregister_tid,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 02321e3..0b6df6e 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -512,6 +512,12 @@ struct qed_rdma_ops {
int (*rdma_query_qp)(void *rdma_cxt, struct qed_rdma_qp *qp,
struct qed_rdma_query_qp_out_params *oparams);
int (*rdma_destroy_qp)(void *rdma_cxt, struct qed_rdma_qp *qp);
+ int
+ (*rdma_register_tid)(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *iparams);
+ int (*rdma_deregister_tid)(void *rdma_cxt, u32 itid);
+ int (*rdma_alloc_tid)(void *rdma_cxt, u32 *itid);
+ void (*rdma_free_tid)(void *rdma_cxt, u32 itid);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 07/11] Add support for memory registeration verbs
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
Ram Amrani
Add support for user, dma and memory regions registration.
Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
---
drivers/infiniband/hw/qedr/main.c | 13 +-
drivers/infiniband/hw/qedr/qedr.h | 45 ++++
drivers/infiniband/hw/qedr/verbs.c | 385 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 14 ++
drivers/net/ethernet/qlogic/qed/qed_roce.c | 240 ++++++++++++++++++
include/linux/qed/qed_roce_if.h | 6 +
6 files changed, 702 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 722b895..3d4feb7 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -106,7 +106,9 @@ static int qedr_register_device(struct qedr_dev *dev)
QEDR_UVERBS(CREATE_QP) |
QEDR_UVERBS(MODIFY_QP) |
QEDR_UVERBS(QUERY_QP) |
- QEDR_UVERBS(DESTROY_QP);
+ QEDR_UVERBS(DESTROY_QP) |
+ QEDR_UVERBS(REG_MR) |
+ QEDR_UVERBS(DEREG_MR);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -138,6 +140,15 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.query_pkey = qedr_query_pkey;
+ dev->ibdev.create_ah = qedr_create_ah;
+ dev->ibdev.destroy_ah = qedr_destroy_ah;
+
+ dev->ibdev.get_dma_mr = qedr_get_dma_mr;
+ dev->ibdev.dereg_mr = qedr_dereg_mr;
+ dev->ibdev.reg_user_mr = qedr_reg_user_mr;
+ dev->ibdev.alloc_mr = qedr_alloc_mr;
+ dev->ibdev.map_mr_sg = qedr_map_mr_sg;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 1c9cc40..3cd4e52 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -374,6 +374,41 @@ struct qedr_qp {
struct qedr_userq urq;
};
+struct qedr_ah {
+ struct ib_ah ibah;
+ struct ib_ah_attr attr;
+};
+
+enum qedr_mr_type {
+ QEDR_MR_USER,
+ QEDR_MR_KERNEL,
+ QEDR_MR_DMA,
+ QEDR_MR_FRMR
+};
+
+struct mr_info {
+ struct qedr_pbl *pbl_table;
+ struct qedr_pbl_info pbl_info;
+ struct list_head free_pbl_list;
+ struct list_head inuse_pbl_list;
+ u32 completed;
+ u32 completed_handled;
+};
+
+struct qedr_mr {
+ struct ib_mr ibmr;
+ struct ib_umem *umem;
+
+ struct qed_rdma_register_tid_in_params hw_mr;
+ enum qedr_mr_type type;
+
+ struct qedr_dev *dev;
+ struct mr_info info;
+
+ u64 *pages;
+ u32 npages;
+};
+
static inline int qedr_get_dmac(struct qedr_dev *dev,
struct ib_ah_attr *ah_attr, u8 *mac_addr)
{
@@ -417,4 +452,14 @@ static inline struct qedr_qp *get_qedr_qp(struct ib_qp *ibqp)
{
return container_of(ibqp, struct qedr_qp, ibqp);
}
+
+static inline struct qedr_ah *get_qedr_ah(struct ib_ah *ibah)
+{
+ return container_of(ibah, struct qedr_ah, ibah);
+}
+
+static inline struct qedr_mr *get_qedr_mr(struct ib_mr *ibmr)
+{
+ return container_of(ibmr, struct qedr_mr, ibmr);
+}
#endif
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 1d16c34..3d6045c 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -2054,3 +2054,388 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
return rc;
}
+
+struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr)
+{
+ struct qedr_ah *ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
+
+ ah->attr = *attr;
+
+ return &ah->ibah;
+}
+
+int qedr_destroy_ah(struct ib_ah *ibah)
+{
+ struct qedr_ah *ah = get_qedr_ah(ibah);
+
+ kfree(ah);
+ return 0;
+}
+
+static void free_mr_info(struct qedr_dev *dev, struct mr_info *info)
+{
+ struct qedr_pbl *pbl, *tmp;
+
+ if (info->pbl_table)
+ list_add_tail(&info->pbl_table->list_entry,
+ &info->free_pbl_list);
+
+ if (!list_empty(&info->inuse_pbl_list))
+ list_splice(&info->inuse_pbl_list, &info->free_pbl_list);
+
+ list_for_each_entry_safe(pbl, tmp, &info->free_pbl_list, list_entry) {
+ list_del(&pbl->list_entry);
+ qedr_free_pbl(dev, &info->pbl_info, pbl);
+ }
+}
+
+static int init_mr_info(struct qedr_dev *dev, struct mr_info *info,
+ size_t page_list_len, bool two_layered)
+{
+ struct qedr_pbl *tmp;
+ int rc;
+
+ INIT_LIST_HEAD(&info->free_pbl_list);
+ INIT_LIST_HEAD(&info->inuse_pbl_list);
+
+ rc = qedr_prepare_pbl_tbl(dev, &info->pbl_info,
+ page_list_len, two_layered);
+ if (rc)
+ goto done;
+
+ info->pbl_table = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
+ if (!info->pbl_table) {
+ rc = -ENOMEM;
+ goto done;
+ }
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "pbl_table_pa = %pa\n",
+ &info->pbl_table->pa);
+
+ /* in usual case we use 2 PBLs, so we add one to free
+ * list and allocating another one
+ */
+ tmp = qedr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
+ if (!tmp) {
+ DP_VERBOSE(dev, QEDR_MSG_MR, "Extra PBL is not allocated\n");
+ goto done;
+ }
+
+ list_add_tail(&tmp->list_entry, &info->free_pbl_list);
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "extra pbl_table_pa = %pa\n", &tmp->pa);
+
+done:
+ if (rc)
+ free_mr_info(dev, info);
+
+ return rc;
+}
+
+struct ib_mr *qedr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len,
+ u64 usr_addr, int acc, struct ib_udata *udata)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_mr *mr;
+ struct qedr_pd *pd;
+ int rc = -ENOMEM;
+
+ pd = get_qedr_pd(ibpd);
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "qedr_register user mr pd = %d start = %lld, len = %lld, usr_addr = %lld, acc = %d\n",
+ pd->pd_id, start, len, usr_addr, acc);
+
+ if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE))
+ return ERR_PTR(-EINVAL);
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(rc);
+
+ mr->type = QEDR_MR_USER;
+
+ mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0);
+ if (IS_ERR(mr->umem)) {
+ rc = -EFAULT;
+ goto err0;
+ }
+
+ rc = init_mr_info(dev, &mr->info, ib_umem_page_count(mr->umem), 1);
+ if (rc)
+ goto err1;
+
+ qedr_populate_pbls(dev, mr->umem, mr->info.pbl_table,
+ &mr->info.pbl_info);
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ /* Index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR;
+ mr->hw_mr.key = 0;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+ mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+ mr->hw_mr.mw_bind = false;
+ mr->hw_mr.pbl_ptr = mr->info.pbl_table[0].pa;
+ mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
+ mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
+ mr->hw_mr.page_size_log = ilog2(mr->umem->page_size);
+ mr->hw_mr.fbo = ib_umem_offset(mr->umem);
+ mr->hw_mr.length = len;
+ mr->hw_mr.vaddr = usr_addr;
+ mr->hw_mr.zbva = false;
+ mr->hw_mr.phy_mr = false;
+ mr->hw_mr.dma_mr = false;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err2;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
+ mr->hw_mr.remote_atomic)
+ mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "register user mr lkey: %x\n",
+ mr->ibmr.lkey);
+ return &mr->ibmr;
+
+err2:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err1:
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+err0:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
+
+int qedr_dereg_mr(struct ib_mr *ib_mr)
+{
+ struct qedr_mr *mr = get_qedr_mr(ib_mr);
+ struct qedr_dev *dev = get_qedr_dev(ib_mr->device);
+ int rc = 0;
+
+ rc = dev->ops->rdma_deregister_tid(dev->rdma_ctx, mr->hw_mr.itid);
+ if (rc)
+ return rc;
+
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+
+ if ((mr->type != QEDR_MR_DMA) && (mr->type != QEDR_MR_FRMR))
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+
+ /* it could be user registered memory. */
+ if (mr->umem)
+ ib_umem_release(mr->umem);
+
+ kfree(mr->pages);
+ kfree(mr);
+
+ return rc;
+}
+
+struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int max_page_list_len)
+{
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_mr *mr;
+ int rc = -ENOMEM;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd->pd_id,
+ max_page_list_len);
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(rc);
+
+ mr->dev = dev;
+ mr->type = QEDR_MR_FRMR;
+
+ rc = init_mr_info(dev, &mr->info, max_page_list_len, 1);
+ if (rc)
+ goto err0;
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err0;
+ }
+
+ /* Index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_FMR;
+ mr->hw_mr.key = 0;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = 0;
+ mr->hw_mr.remote_read = 0;
+ mr->hw_mr.remote_write = 0;
+ mr->hw_mr.remote_atomic = 0;
+ mr->hw_mr.mw_bind = false;
+ mr->hw_mr.pbl_ptr = 0;
+ mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
+ mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
+ mr->hw_mr.fbo = 0;
+ mr->hw_mr.length = 0;
+ mr->hw_mr.vaddr = 0;
+ mr->hw_mr.zbva = false;
+ mr->hw_mr.phy_mr = true;
+ mr->hw_mr.dma_mr = false;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ mr->ibmr.rkey = mr->ibmr.lkey;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "alloc frmr: %x\n", mr->ibmr.lkey);
+ return mr;
+
+err1:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err0:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
+
+struct ib_mr *qedr_alloc_mr(struct ib_pd *ibpd,
+ enum ib_mr_type mr_type, u32 max_num_sg)
+{
+ struct qedr_dev *dev;
+ struct qedr_mr *mr;
+
+ if (mr_type != IB_MR_TYPE_MEM_REG)
+ return ERR_PTR(-EINVAL);
+
+ mr = __qedr_alloc_mr(ibpd, max_num_sg);
+
+ if (IS_ERR(mr))
+ return ERR_PTR(-EINVAL);
+
+ dev = mr->dev;
+ mr->pages = kcalloc(max_num_sg, sizeof(u64), GFP_KERNEL);
+ if (!mr->pages) {
+ dev->ops->rdma_deregister_tid(dev->rdma_ctx, mr->hw_mr.itid);
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+ qedr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
+ return ERR_PTR(-EINVAL);
+ }
+
+ return &mr->ibmr;
+}
+
+static int qedr_set_page(struct ib_mr *ibmr, u64 addr)
+{
+ struct qedr_mr *mr = get_qedr_mr(ibmr);
+
+ if (unlikely(mr->npages == mr->info.pbl_info.num_pbes)) {
+ DP_ERR(mr->dev, "qedr_set_page failes when %d\n", mr->npages);
+ return -ENOMEM;
+ }
+
+ DP_VERBOSE(mr->dev, QEDR_MSG_MR, "qedr_set_page pages[%d] = 0x%llx\n\n",
+ mr->npages, addr);
+
+ mr->pages[mr->npages++] = addr;
+
+ return 0;
+}
+
+static void handle_completed_mrs(struct qedr_dev *dev, struct mr_info *info)
+{
+ int work = info->completed - info->completed_handled - 1;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "Special FMR work = %d\n", work);
+ while (work-- > 0 && !list_empty(&info->inuse_pbl_list)) {
+ struct qedr_pbl *pbl;
+
+ /* Free all the page list that are possible to be freed
+ * (all the ones that were invalidated), under the assumption
+ * that if an FMR was completed successfully that means that
+ * if there was an invalidate operation before it also ended
+ */
+ pbl = list_first_entry(&info->inuse_pbl_list,
+ struct qedr_pbl, list_entry);
+ list_del(&pbl->list_entry);
+ list_add_tail(&pbl->list_entry, &info->free_pbl_list);
+ info->completed_handled++;
+ }
+}
+
+int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset)
+{
+ struct qedr_mr *mr = get_qedr_mr(ibmr);
+
+ mr->npages = 0;
+
+ handle_completed_mrs(mr->dev, &mr->info);
+ return ib_sg_to_pages(ibmr, sg, sg_nents, NULL, qedr_set_page);
+}
+
+struct ib_mr *qedr_get_dma_mr(struct ib_pd *ibpd, int acc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+ struct qedr_pd *pd = get_qedr_pd(ibpd);
+ struct qedr_mr *mr;
+ int rc;
+
+ if (acc & IB_ACCESS_MW_BIND) {
+ DP_ERR(dev, "Unsupported access flags received for dma mr\n");
+ return ERR_PTR(-EINVAL);
+ }
+
+ mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+ if (!mr)
+ return ERR_PTR(-ENOMEM);
+
+ mr->type = QEDR_MR_DMA;
+
+ rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
+ if (rc) {
+ DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
+ goto err1;
+ }
+
+ /* index only, 18 bit long, lkey = itid << 8 | key */
+ mr->hw_mr.tid_type = QED_RDMA_TID_REGISTERED_MR;
+ mr->hw_mr.pd = pd->pd_id;
+ mr->hw_mr.local_read = 1;
+ mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+ mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+ mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+ mr->hw_mr.dma_mr = true;
+
+ rc = dev->ops->rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
+ if (rc) {
+ DP_ERR(dev, "roce register tid returned an error %d\n", rc);
+ goto err2;
+ }
+
+ mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+ if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
+ mr->hw_mr.remote_atomic)
+ mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
+
+ DP_VERBOSE(dev, QEDR_MSG_MR, "get dma mr: lkey = %x\n", mr->ibmr.lkey);
+ return &mr->ibmr;
+
+err2:
+ dev->ops->rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
+err1:
+ kfree(mr);
+ return ERR_PTR(rc);
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index 80dbac1..e7cb3fd 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -44,4 +44,18 @@ int qedr_query_qp(struct ib_qp *, struct ib_qp_attr *qp_attr,
int qp_attr_mask, struct ib_qp_init_attr *);
int qedr_destroy_qp(struct ib_qp *ibqp);
+struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr);
+int qedr_destroy_ah(struct ib_ah *ibah);
+
+int qedr_dereg_mr(struct ib_mr *);
+struct ib_mr *qedr_get_dma_mr(struct ib_pd *, int acc);
+
+struct ib_mr *qedr_reg_user_mr(struct ib_pd *, u64 start, u64 length,
+ u64 virt, int acc, struct ib_udata *);
+
+int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+ int sg_nents, unsigned int *sg_offset);
+
+struct ib_mr *qedr_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
+ u32 max_num_sg);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index cf92ff7..04e33ee 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -690,6 +690,17 @@ struct qed_rdma_device *qed_rdma_query_device(void *rdma_cxt)
return p_hwfn->p_rdma_info->dev;
}
+void qed_rdma_free_tid(void *rdma_cxt, u32 itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
+
+ spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+ qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->tid_map, itid);
+ spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
int qed_rdma_alloc_tid(void *rdma_cxt, u32 *itid)
{
struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
@@ -2321,6 +2332,231 @@ int qed_rdma_modify_qp(void *rdma_cxt,
return rc;
}
+int qed_rdma_register_tid(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *params)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_register_tid_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ enum rdma_tid_type tid_type;
+ u8 fw_return_code;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", params->itid);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_REGISTER_MR,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (p_hwfn->p_rdma_info->last_tid < params->itid)
+ p_hwfn->p_rdma_info->last_tid = params->itid;
+
+ p_ramrod = &p_ent->ramrod.rdma_register_tid;
+
+ p_ramrod->flags = 0;
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_TWO_LEVEL_PBL,
+ params->pbl_two_level);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_ZERO_BASED, params->zbva);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_PHY_MR, params->phy_mr);
+
+ /* Don't initialize D/C field, as it may override other bits. */
+ if (!(params->tid_type == QED_RDMA_TID_FMR) && !(params->dma_mr))
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_PAGE_SIZE_LOG,
+ params->page_size_log - 12);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_MAX_ID,
+ p_hwfn->p_rdma_info->last_tid);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_READ,
+ params->remote_read);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_WRITE,
+ params->remote_write);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_REMOTE_ATOMIC,
+ params->remote_atomic);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_WRITE,
+ params->local_write);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_LOCAL_READ, params->local_read);
+
+ SET_FIELD(p_ramrod->flags,
+ RDMA_REGISTER_TID_RAMROD_DATA_ENABLE_MW_BIND,
+ params->mw_bind);
+
+ SET_FIELD(p_ramrod->flags1,
+ RDMA_REGISTER_TID_RAMROD_DATA_PBL_PAGE_SIZE_LOG,
+ params->pbl_page_size_log - 12);
+
+ SET_FIELD(p_ramrod->flags2,
+ RDMA_REGISTER_TID_RAMROD_DATA_DMA_MR, params->dma_mr);
+
+ switch (params->tid_type) {
+ case QED_RDMA_TID_REGISTERED_MR:
+ tid_type = RDMA_TID_REGISTERED_MR;
+ break;
+ case QED_RDMA_TID_FMR:
+ tid_type = RDMA_TID_FMR;
+ break;
+ case QED_RDMA_TID_MW_TYPE1:
+ tid_type = RDMA_TID_MW_TYPE1;
+ break;
+ case QED_RDMA_TID_MW_TYPE2A:
+ tid_type = RDMA_TID_MW_TYPE2A;
+ break;
+ default:
+ rc = -EINVAL;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+ SET_FIELD(p_ramrod->flags1,
+ RDMA_REGISTER_TID_RAMROD_DATA_TID_TYPE, tid_type);
+
+ p_ramrod->itid = cpu_to_le32(params->itid);
+ p_ramrod->key = params->key;
+ p_ramrod->pd = cpu_to_le16(params->pd);
+ p_ramrod->length_hi = (u8)(params->length >> 32);
+ p_ramrod->length_lo = DMA_LO_LE(params->length);
+ if (params->zbva) {
+ /* Lower 32 bits of the registered MR address.
+ * In case of zero based MR, will hold FBO
+ */
+ p_ramrod->va.hi = 0;
+ p_ramrod->va.lo = cpu_to_le32(params->fbo);
+ } else {
+ DMA_REGPAIR_LE(p_ramrod->va, params->vaddr);
+ }
+ DMA_REGPAIR_LE(p_ramrod->pbl_base, params->pbl_ptr);
+
+ /* DIF */
+ if (params->dif_enabled) {
+ SET_FIELD(p_ramrod->flags2,
+ RDMA_REGISTER_TID_RAMROD_DATA_DIF_ON_HOST_FLG, 1);
+ DMA_REGPAIR_LE(p_ramrod->dif_error_addr,
+ params->dif_error_addr);
+ DMA_REGPAIR_LE(p_ramrod->dif_runt_addr, params->dif_runt_addr);
+ }
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ return -EINVAL;
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Register TID, rc = %d\n", rc);
+ return rc;
+}
+
+int qed_rdma_deregister_tid(void *rdma_cxt, u32 itid)
+{
+ struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+ struct rdma_deregister_tid_ramrod_data *p_ramrod;
+ struct qed_sp_init_data init_data;
+ struct qed_spq_entry *p_ent;
+ struct qed_ptt *p_ptt;
+ u8 fw_return_code;
+ int rc;
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "itid = %08x\n", itid);
+
+ /* Get SPQ entry */
+ memset(&init_data, 0, sizeof(init_data));
+ init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+ init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+ rc = qed_sp_init_request(p_hwfn, &p_ent, RDMA_RAMROD_DEREGISTER_MR,
+ p_hwfn->p_rdma_info->proto, &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ p_ramrod = &p_ent->ramrod.rdma_deregister_tid;
+ p_ramrod->itid = cpu_to_le32(itid);
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = %d\n", rc);
+ return rc;
+ }
+
+ if (fw_return_code == RDMA_RETURN_DEREGISTER_MR_BAD_STATE_ERR) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n", fw_return_code);
+ return -EINVAL;
+ } else if (fw_return_code == RDMA_RETURN_NIG_DRAIN_REQ) {
+ /* Bit indicating that the TID is in use and a nig drain is
+ * required before sending the ramrod again
+ */
+ p_ptt = qed_ptt_acquire(p_hwfn);
+ if (!p_ptt) {
+ rc = -EBUSY;
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to acquire PTT\n");
+ return rc;
+ }
+
+ rc = qed_mcp_drain(p_hwfn, p_ptt);
+ if (rc) {
+ qed_ptt_release(p_hwfn, p_ptt);
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Drain failed\n");
+ return rc;
+ }
+
+ qed_ptt_release(p_hwfn, p_ptt);
+
+ /* Resend the ramrod */
+ rc = qed_sp_init_request(p_hwfn, &p_ent,
+ RDMA_RAMROD_DEREGISTER_MR,
+ p_hwfn->p_rdma_info->proto,
+ &init_data);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Failed to init sp-element\n");
+ return rc;
+ }
+
+ rc = qed_spq_post(p_hwfn, p_ent, &fw_return_code);
+ if (rc) {
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+ "Ramrod failed\n");
+ return rc;
+ }
+
+ if (fw_return_code != RDMA_RETURN_OK) {
+ DP_NOTICE(p_hwfn, "fw_return_code = %d\n",
+ fw_return_code);
+ return rc;
+ }
+ }
+
+ DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "De-registered TID, rc = %d\n", rc);
+ return rc;
+}
+
static void *qed_rdma_get_rdma_ctx(struct qed_dev *cdev)
{
return QED_LEADING_HWFN(cdev);
@@ -2419,6 +2655,10 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_modify_qp = &qed_rdma_modify_qp,
.rdma_query_qp = &qed_rdma_query_qp,
.rdma_destroy_qp = &qed_rdma_destroy_qp,
+ .rdma_alloc_tid = &qed_rdma_alloc_tid,
+ .rdma_free_tid = &qed_rdma_free_tid,
+ .rdma_register_tid = &qed_rdma_register_tid,
+ .rdma_deregister_tid = &qed_rdma_deregister_tid,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 02321e3..0b6df6e 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -512,6 +512,12 @@ struct qed_rdma_ops {
int (*rdma_query_qp)(void *rdma_cxt, struct qed_rdma_qp *qp,
struct qed_rdma_query_qp_out_params *oparams);
int (*rdma_destroy_qp)(void *rdma_cxt, struct qed_rdma_qp *qp);
+ int
+ (*rdma_register_tid)(void *rdma_cxt,
+ struct qed_rdma_register_tid_in_params *iparams);
+ int (*rdma_deregister_tid)(void *rdma_cxt, u32 itid);
+ int (*rdma_alloc_tid)(void *rdma_cxt, u32 *itid);
+ void (*rdma_free_tid)(void *rdma_cxt, u32 itid);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 08/11] Add support for data path
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Implement fastpath verbs like ib_send_post, ib_post_recv and ib_poll_cq.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 9 +-
drivers/infiniband/hw/qedr/qedr.h | 19 +
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 562 +++++++++++++++
drivers/infiniband/hw/qedr/verbs.c | 1062 ++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 5 +
5 files changed, 1656 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 3d4feb7..42206ed 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -108,7 +108,10 @@ static int qedr_register_device(struct qedr_dev *dev)
QEDR_UVERBS(QUERY_QP) |
QEDR_UVERBS(DESTROY_QP) |
QEDR_UVERBS(REG_MR) |
- QEDR_UVERBS(DEREG_MR);
+ QEDR_UVERBS(DEREG_MR) |
+ QEDR_UVERBS(POLL_CQ) |
+ QEDR_UVERBS(POST_SEND) |
+ QEDR_UVERBS(POST_RECV);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -149,6 +152,10 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.alloc_mr = qedr_alloc_mr;
dev->ibdev.map_mr_sg = qedr_map_mr_sg;
+ dev->ibdev.poll_cq = qedr_poll_cq;
+ dev->ibdev.post_send = qedr_post_send;
+ dev->ibdev.post_recv = qedr_post_recv;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 3cd4e52..bc218d2 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -409,6 +409,25 @@ struct qedr_mr {
u32 npages;
};
+#define SET_FIELD2(value, name, flag) ((value) |= ((flag) << (name ## _SHIFT)))
+
+#define QEDR_RESP_IMM (RDMA_CQE_RESPONDER_IMM_FLG_MASK << \
+ RDMA_CQE_RESPONDER_IMM_FLG_SHIFT)
+#define QEDR_RESP_RDMA (RDMA_CQE_RESPONDER_RDMA_FLG_MASK << \
+ RDMA_CQE_RESPONDER_RDMA_FLG_SHIFT)
+#define QEDR_RESP_RDMA_IMM (QEDR_RESP_IMM | QEDR_RESP_RDMA)
+
+static inline void qedr_inc_sw_cons(struct qedr_qp_hwq_info *info)
+{
+ info->cons = (info->cons + 1) % info->max_wr;
+ info->wqe_cons++;
+}
+
+static inline void qedr_inc_sw_prod(struct qedr_qp_hwq_info *info)
+{
+ info->prod = (info->prod + 1) % info->max_wr;
+}
+
static inline int qedr_get_dmac(struct qedr_dev *dev,
struct ib_ah_attr *ah_attr, u8 *mac_addr)
{
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
index 4770559..5c98d20 100644
--- a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -150,6 +150,12 @@ struct rdma_rq_sge {
struct regpair addr;
__le32 length;
__le32 flags;
+#define RDMA_RQ_SGE_L_KEY_MASK 0x3FFFFFF
+#define RDMA_RQ_SGE_L_KEY_SHIFT 0
+#define RDMA_RQ_SGE_NUM_SGES_MASK 0x7
+#define RDMA_RQ_SGE_NUM_SGES_SHIFT 26
+#define RDMA_RQ_SGE_RESERVED0_MASK 0x7
+#define RDMA_RQ_SGE_RESERVED0_SHIFT 29
};
struct rdma_srq_sge {
@@ -183,4 +189,560 @@ struct rdma_pwm_val32_data {
__le32 value;
};
+/* DIF Block size options */
+enum rdma_dif_block_size {
+ RDMA_DIF_BLOCK_512 = 0,
+ RDMA_DIF_BLOCK_4096 = 1,
+ MAX_RDMA_DIF_BLOCK_SIZE
+};
+
+/* DIF CRC initial value */
+enum rdma_dif_crc_seed {
+ RDMA_DIF_CRC_SEED_0000 = 0,
+ RDMA_DIF_CRC_SEED_FFFF = 1,
+ MAX_RDMA_DIF_CRC_SEED
+};
+
+/* RDMA DIF Error Result Structure */
+struct rdma_dif_error_result {
+ __le32 error_intervals;
+ __le32 dif_error_1st_interval;
+ u8 flags;
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_CRC_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_CRC_SHIFT 0
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_APP_TAG_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_APP_TAG_SHIFT 1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_REF_TAG_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_REF_TAG_SHIFT 2
+#define RDMA_DIF_ERROR_RESULT_RESERVED0_MASK 0xF
+#define RDMA_DIF_ERROR_RESULT_RESERVED0_SHIFT 3
+#define RDMA_DIF_ERROR_RESULT_TOGGLE_BIT_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_TOGGLE_BIT_SHIFT 7
+ u8 reserved1[55];
+};
+
+/* DIF IO direction */
+enum rdma_dif_io_direction_flg {
+ RDMA_DIF_DIR_RX = 0,
+ RDMA_DIF_DIR_TX = 1,
+ MAX_RDMA_DIF_IO_DIRECTION_FLG
+};
+
+/* RDMA DIF Runt Result Structure */
+struct rdma_dif_runt_result {
+ __le16 guard_tag;
+ __le16 reserved[3];
+};
+
+/* Memory window type enumeration */
+enum rdma_mw_type {
+ RDMA_MW_TYPE_1,
+ RDMA_MW_TYPE_2A,
+ MAX_RDMA_MW_TYPE
+};
+
+struct rdma_sq_atomic_wqe {
+ __le32 reserved1;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_ATOMIC_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_ATOMIC_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_ATOMIC_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_ATOMIC_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_ATOMIC_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_ATOMIC_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_ATOMIC_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_ATOMIC_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ struct regpair remote_va;
+ __le32 r_key;
+ __le32 reserved2;
+ struct regpair cmp_data;
+ struct regpair swap_data;
+};
+
+/* First element (16 bytes) of atomic wqe */
+struct rdma_sq_atomic_wqe_1st {
+ __le32 reserved1;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_ATOMIC_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_ATOMIC_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_ATOMIC_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_ATOMIC_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_ATOMIC_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_ATOMIC_WQE_1ST_RESERVED0_MASK 0x7
+#define RDMA_SQ_ATOMIC_WQE_1ST_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of atomic wqe */
+struct rdma_sq_atomic_wqe_2nd {
+ struct regpair remote_va;
+ __le32 r_key;
+ __le32 reserved2;
+};
+
+/* Third element (16 bytes) of atomic wqe */
+struct rdma_sq_atomic_wqe_3rd {
+ struct regpair cmp_data;
+ struct regpair swap_data;
+};
+
+struct rdma_sq_bind_wqe {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_BIND_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_BIND_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_BIND_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_BIND_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_BIND_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_BIND_WQE_RESERVED0_MASK 0x7
+#define RDMA_SQ_BIND_WQE_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ u8 bind_ctrl;
+#define RDMA_SQ_BIND_WQE_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_BIND_WQE_ZERO_BASED_SHIFT 0
+#define RDMA_SQ_BIND_WQE_MW_TYPE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_MW_TYPE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_RESERVED1_MASK 0x3F
+#define RDMA_SQ_BIND_WQE_RESERVED1_SHIFT 2
+ u8 access_ctrl;
+#define RDMA_SQ_BIND_WQE_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_BIND_WQE_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_BIND_WQE_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_BIND_WQE_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_BIND_WQE_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_BIND_WQE_RESERVED2_MASK 0x7
+#define RDMA_SQ_BIND_WQE_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ __le32 parent_l_key;
+ __le32 reserved4;
+};
+
+/* First element (16 bytes) of bind wqe */
+struct rdma_sq_bind_wqe_1st {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_BIND_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_BIND_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_BIND_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_BIND_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_BIND_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_BIND_WQE_1ST_RESERVED0_MASK 0x7
+#define RDMA_SQ_BIND_WQE_1ST_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of bind wqe */
+struct rdma_sq_bind_wqe_2nd {
+ u8 bind_ctrl;
+#define RDMA_SQ_BIND_WQE_2ND_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_ZERO_BASED_SHIFT 0
+#define RDMA_SQ_BIND_WQE_2ND_MW_TYPE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_MW_TYPE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED1_MASK 0x3F
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED1_SHIFT 2
+ u8 access_ctrl;
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_2ND_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED2_MASK 0x7
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ __le32 parent_l_key;
+ __le32 reserved4;
+};
+
+/* Structure with only the SQ WQE common
+ * fields. Size is of one SQ element (16B)
+ */
+struct rdma_sq_common_wqe {
+ __le32 reserved1[3];
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_COMMON_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_COMMON_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_COMMON_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_COMMON_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_COMMON_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_COMMON_WQE_RESERVED0_MASK 0x7
+#define RDMA_SQ_COMMON_WQE_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+struct rdma_sq_fmr_wqe {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_FMR_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_FMR_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_FMR_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_FMR_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ u8 fmr_ctrl;
+#define RDMA_SQ_FMR_WQE_PAGE_SIZE_LOG_MASK 0x1F
+#define RDMA_SQ_FMR_WQE_PAGE_SIZE_LOG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_ZERO_BASED_SHIFT 5
+#define RDMA_SQ_FMR_WQE_BIND_EN_MASK 0x1
+#define RDMA_SQ_FMR_WQE_BIND_EN_SHIFT 6
+#define RDMA_SQ_FMR_WQE_RESERVED1_MASK 0x1
+#define RDMA_SQ_FMR_WQE_RESERVED1_SHIFT 7
+ u8 access_ctrl;
+#define RDMA_SQ_FMR_WQE_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_FMR_WQE_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_FMR_WQE_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_FMR_WQE_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_FMR_WQE_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_FMR_WQE_RESERVED2_MASK 0x7
+#define RDMA_SQ_FMR_WQE_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ struct regpair pbl_addr;
+ __le32 dif_base_ref_tag;
+ __le16 dif_app_tag;
+ __le16 dif_app_tag_mask;
+ __le16 dif_runt_crc_value;
+ __le16 dif_flags;
+#define RDMA_SQ_FMR_WQE_DIF_IO_DIRECTION_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_IO_DIRECTION_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_BLOCK_SIZE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_DIF_RUNT_VALID_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_RUNT_VALID_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_CRC_GUARD_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_CRC_GUARD_SHIFT 3
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_REF_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_REF_TAG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_APP_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_APP_TAG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_DIF_CRC_SEED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_CRC_SEED_SHIFT 6
+#define RDMA_SQ_FMR_WQE_RESERVED4_MASK 0x1FF
+#define RDMA_SQ_FMR_WQE_RESERVED4_SHIFT 7
+ __le32 Reserved5;
+};
+
+/* First element (16 bytes) of fmr wqe */
+struct rdma_sq_fmr_wqe_1st {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_FMR_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_FMR_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_FMR_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_1ST_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_1ST_RESERVED0_MASK 0x3
+#define RDMA_SQ_FMR_WQE_1ST_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of fmr wqe */
+struct rdma_sq_fmr_wqe_2nd {
+ u8 fmr_ctrl;
+#define RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG_MASK 0x1F
+#define RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_2ND_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_ZERO_BASED_SHIFT 5
+#define RDMA_SQ_FMR_WQE_2ND_BIND_EN_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_BIND_EN_SHIFT 6
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED1_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED1_SHIFT 7
+ u8 access_ctrl;
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED2_MASK 0x7
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ struct regpair pbl_addr;
+};
+
+/* Third element (16 bytes) of fmr wqe */
+struct rdma_sq_fmr_wqe_3rd {
+ __le32 dif_base_ref_tag;
+ __le16 dif_app_tag;
+ __le16 dif_app_tag_mask;
+ __le16 dif_runt_crc_value;
+ __le16 dif_flags;
+#define RDMA_SQ_FMR_WQE_3RD_DIF_IO_DIRECTION_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_IO_DIRECTION_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_3RD_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_BLOCK_SIZE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_RUNT_VALID_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_RUNT_VALID_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_CRC_GUARD_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_CRC_GUARD_SHIFT 3
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_REF_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_REF_TAG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_APP_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_APP_TAG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_3RD_DIF_CRC_SEED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_CRC_SEED_SHIFT 6
+#define RDMA_SQ_FMR_WQE_3RD_RESERVED4_MASK 0x1FF
+#define RDMA_SQ_FMR_WQE_3RD_RESERVED4_SHIFT 7
+ __le32 Reserved5;
+};
+
+struct rdma_sq_local_inv_wqe {
+ struct regpair reserved;
+ __le32 inv_l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_LOCAL_INV_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_LOCAL_INV_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_LOCAL_INV_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_LOCAL_INV_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_LOCAL_INV_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_LOCAL_INV_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_LOCAL_INV_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_LOCAL_INV_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+struct rdma_sq_rdma_wqe {
+ __le32 imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_RDMA_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_RDMA_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_RDMA_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_RDMA_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_RDMA_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ struct regpair remote_va;
+ __le32 r_key;
+ u8 dif_flags;
+#define RDMA_SQ_RDMA_WQE_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_BLOCK_SIZE_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_DIF_FIRST_RDMA_IN_IO_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_FIRST_RDMA_IN_IO_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_DIF_LAST_RDMA_IN_IO_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_LAST_RDMA_IN_IO_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_RESERVED1_MASK 0x1F
+#define RDMA_SQ_RDMA_WQE_RESERVED1_SHIFT 3
+ u8 reserved2[3];
+};
+
+/* First element (16 bytes) of rdma wqe */
+struct rdma_sq_rdma_wqe_1st {
+ __le32 imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_RDMA_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_RDMA_WQE_1ST_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_RDMA_WQE_1ST_RESERVED0_MASK 0x3
+#define RDMA_SQ_RDMA_WQE_1ST_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of rdma wqe */
+struct rdma_sq_rdma_wqe_2nd {
+ struct regpair remote_va;
+ __le32 r_key;
+ u8 dif_flags;
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_BLOCK_SIZE_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_FIRST_SEGMENT_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_FIRST_SEGMENT_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_LAST_SEGMENT_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_LAST_SEGMENT_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_2ND_RESERVED1_MASK 0x1F
+#define RDMA_SQ_RDMA_WQE_2ND_RESERVED1_SHIFT 3
+ u8 reserved2[3];
+};
+
+/* SQ WQE req type enumeration */
+enum rdma_sq_req_type {
+ RDMA_SQ_REQ_TYPE_SEND,
+ RDMA_SQ_REQ_TYPE_SEND_WITH_IMM,
+ RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE,
+ RDMA_SQ_REQ_TYPE_RDMA_WR,
+ RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM,
+ RDMA_SQ_REQ_TYPE_RDMA_RD,
+ RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP,
+ RDMA_SQ_REQ_TYPE_ATOMIC_ADD,
+ RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE,
+ RDMA_SQ_REQ_TYPE_FAST_MR,
+ RDMA_SQ_REQ_TYPE_BIND,
+ RDMA_SQ_REQ_TYPE_INVALID,
+ MAX_RDMA_SQ_REQ_TYPE
+};
+
+struct rdma_sq_send_wqe {
+ __le32 inv_key_or_imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_SEND_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_SEND_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_SEND_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_SEND_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_SEND_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_SEND_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_SEND_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_SEND_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ __le32 reserved1[4];
+};
+
+struct rdma_sq_send_wqe_1st {
+ __le32 inv_key_or_imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_SEND_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_SEND_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_SEND_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_SEND_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_SEND_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_SEND_WQE_1ST_RESERVED0_MASK 0x7
+#define RDMA_SQ_SEND_WQE_1ST_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+struct rdma_sq_send_wqe_2st {
+ __le32 reserved1[4];
+};
+
#endif /* __QED_HSI_RDMA__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 3d6045c..979555e 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -2439,3 +2439,1065 @@ err1:
kfree(mr);
return ERR_PTR(rc);
}
+
+static inline int qedr_wq_is_full(struct qedr_qp_hwq_info *wq)
+{
+ return (((wq->prod + 1) % wq->max_wr) == wq->cons);
+}
+
+static int sge_data_len(struct ib_sge *sg_list, int num_sge)
+{
+ int i, len = 0;
+
+ for (i = 0; i < num_sge; i++)
+ len += sg_list[i].length;
+
+ return len;
+}
+
+static void swap_wqe_data64(u64 *p)
+{
+ int i;
+
+ for (i = 0; i < QEDR_SQE_ELEMENT_SIZE / sizeof(u64); i++, p++)
+ *p = cpu_to_be64(cpu_to_le64(*p));
+}
+
+static u32 qedr_prepare_sq_inline_data(struct qedr_dev *dev,
+ struct qedr_qp *qp, u8 *wqe_size,
+ struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr, u8 *bits,
+ u8 bit)
+{
+ u32 data_size = sge_data_len(wr->sg_list, wr->num_sge);
+ char *seg_prt, *wqe;
+ int i, seg_siz;
+
+ if (data_size > ROCE_REQ_MAX_INLINE_DATA_SIZE) {
+ DP_ERR(dev, "Too much inline data in WR: %d\n", data_size);
+ *bad_wr = wr;
+ return 0;
+ }
+
+ if (!data_size)
+ return data_size;
+
+ *bits |= bit;
+
+ seg_prt = NULL;
+ wqe = NULL;
+ seg_siz = 0;
+
+ /* Copy data inline */
+ for (i = 0; i < wr->num_sge; i++) {
+ u32 len = wr->sg_list[i].length;
+ void *src = (void *)(uintptr_t)wr->sg_list[i].addr;
+
+ while (len > 0) {
+ u32 cur;
+
+ /* New segment required */
+ if (!seg_siz) {
+ wqe = (char *)qed_chain_produce(&qp->sq.pbl);
+ seg_prt = wqe;
+ seg_siz = sizeof(struct rdma_sq_common_wqe);
+ (*wqe_size)++;
+ }
+
+ /* Calculate currently allowed length */
+ cur = min_t(u32, len, seg_siz);
+ memcpy(seg_prt, src, cur);
+
+ /* Update segment variables */
+ seg_prt += cur;
+ seg_siz -= cur;
+
+ /* Update sge variables */
+ src += cur;
+ len -= cur;
+
+ /* Wwap fully-completed segments */
+ if (!seg_siz)
+ swap_wqe_data64((u64 *)wqe);
+ }
+ }
+
+ /* swap last not completed segment */
+ if (seg_siz)
+ swap_wqe_data64((u64 *)wqe);
+
+ return data_size;
+}
+
+#define RQ_SGE_SET(sge, vaddr, vlength, vflags) \
+ do { \
+ DMA_REGPAIR_LE(sge->addr, vaddr); \
+ (sge)->length = cpu_to_le32(vlength); \
+ (sge)->flags = cpu_to_le32(vflags); \
+ } while (0)
+
+#define SRQ_HDR_SET(hdr, vwr_id, num_sge) \
+ do { \
+ DMA_REGPAIR_LE(hdr->wr_id, vwr_id); \
+ (hdr)->num_sges = num_sge; \
+ } while (0)
+
+#define SRQ_SGE_SET(sge, vaddr, vlength, vlkey) \
+ do { \
+ DMA_REGPAIR_LE(sge->addr, vaddr); \
+ (sge)->length = cpu_to_le32(vlength); \
+ (sge)->l_key = cpu_to_le32(vlkey); \
+ } while (0)
+
+static u32 qedr_prepare_sq_sges(struct qedr_qp *qp, u8 *wqe_size,
+ struct ib_send_wr *wr)
+{
+ u32 data_size = 0;
+ int i;
+
+ for (i = 0; i < wr->num_sge; i++) {
+ struct rdma_sq_sge *sge = qed_chain_produce(&qp->sq.pbl);
+
+ DMA_REGPAIR_LE(sge->addr, wr->sg_list[i].addr);
+ sge->l_key = cpu_to_le32(wr->sg_list[i].lkey);
+ sge->length = cpu_to_le32(wr->sg_list[i].length);
+ data_size += wr->sg_list[i].length;
+ }
+
+ if (wqe_size)
+ *wqe_size += wr->num_sge;
+
+ return data_size;
+}
+
+static u32 qedr_prepare_sq_rdma_data(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct rdma_sq_rdma_wqe_1st *rwqe,
+ struct rdma_sq_rdma_wqe_2nd *rwqe2,
+ struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ rwqe2->r_key = cpu_to_le32(rdma_wr(wr)->rkey);
+ DMA_REGPAIR_LE(rwqe2->remote_va, rdma_wr(wr)->remote_addr);
+
+ if (wr->send_flags & IB_SEND_INLINE) {
+ u8 flags = 0;
+
+ SET_FIELD2(flags, RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG, 1);
+ return qedr_prepare_sq_inline_data(dev, qp, &rwqe->wqe_size, wr,
+ bad_wr, &rwqe->flags, flags);
+ }
+
+ return qedr_prepare_sq_sges(qp, &rwqe->wqe_size, wr);
+}
+
+static u32 qedr_prepare_sq_send_data(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct rdma_sq_send_wqe_1st *swqe,
+ struct rdma_sq_send_wqe_2st *swqe2,
+ struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ memset(swqe2, 0, sizeof(*swqe2));
+ if (wr->send_flags & IB_SEND_INLINE) {
+ u8 flags = 0;
+
+ SET_FIELD2(flags, RDMA_SQ_SEND_WQE_INLINE_FLG, 1);
+ return qedr_prepare_sq_inline_data(dev, qp, &swqe->wqe_size, wr,
+ bad_wr, &swqe->flags, flags);
+ }
+
+ return qedr_prepare_sq_sges(qp, &swqe->wqe_size, wr);
+}
+
+static int qedr_prepare_reg(struct qedr_qp *qp,
+ struct rdma_sq_fmr_wqe_1st *fwqe1,
+ struct ib_reg_wr *wr)
+{
+ struct qedr_mr *mr = get_qedr_mr(wr->mr);
+ struct qedr_pbl *pbl_table = mr->info.pbl_table;
+ struct rdma_sq_fmr_wqe_2nd *fwqe2;
+ int num_pbes, total_num_pbes = 0;
+ struct qedr_dev *dev = qp->dev;
+ struct regpair *pbe;
+ int i;
+
+ fwqe2 = (struct rdma_sq_fmr_wqe_2nd *)qed_chain_produce(&qp->sq.pbl);
+ fwqe1->addr.hi = upper_32_bits(mr->ibmr.iova);
+ fwqe1->addr.lo = lower_32_bits(mr->ibmr.iova);
+ fwqe1->l_key = wr->key;
+
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_READ,
+ !!(wr->access & IB_ACCESS_REMOTE_READ));
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE,
+ !!(wr->access & IB_ACCESS_REMOTE_WRITE));
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC,
+ !!(wr->access & IB_ACCESS_REMOTE_ATOMIC));
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_READ, 1);
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE,
+ !!(wr->access & IB_ACCESS_LOCAL_WRITE));
+ fwqe2->fmr_ctrl = 0;
+
+ SET_FIELD2(fwqe2->fmr_ctrl, RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG,
+ ilog2(mr->ibmr.page_size) - 12);
+
+ fwqe2->length_hi = 0;
+ fwqe2->length_lo = mr->ibmr.length;
+ fwqe2->pbl_addr.hi = upper_32_bits(pbl_table->pa);
+ fwqe2->pbl_addr.lo = lower_32_bits(pbl_table->pa);
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "Prepare fmr...rkey=%x addr=%x:%x length = %x pbl_addr %x:%x\n",
+ wr->key, fwqe1->addr.hi, fwqe1->addr.lo,
+ fwqe2->length_lo, fwqe2->pbl_addr.hi, fwqe2->pbl_addr.lo);
+
+ pbe = (struct regpair *)pbl_table->va;
+ num_pbes = 0;
+
+ for (i = 0; i < mr->npages &&
+ (total_num_pbes != mr->info.pbl_info.num_pbes); i++) {
+ u64 buf_addr = mr->pages[i];
+
+ pbe->lo = cpu_to_le32((u32)buf_addr);
+ pbe->hi = cpu_to_le32((u32)upper_32_bits(buf_addr));
+
+ num_pbes += 1;
+ pbe++;
+ total_num_pbes++;
+
+ /* If the given pbl is full storing the pbes,
+ * move to next pbl.
+ */
+ if (num_pbes == (mr->info.pbl_info.pbl_size / sizeof(u64))) {
+ pbl_table++;
+ pbe = (struct regpair *)pbl_table->va;
+ num_pbes = 0;
+ }
+ }
+
+ qp->wqe_wr_id[qp->sq.prod].mr = mr;
+
+ return 0;
+}
+
+enum ib_wc_opcode qedr_ib_to_wc_opcode(enum ib_wr_opcode opcode)
+{
+ switch (opcode) {
+ case IB_WR_RDMA_WRITE:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ return IB_WC_RDMA_WRITE;
+ case IB_WR_SEND_WITH_IMM:
+ case IB_WR_SEND:
+ case IB_WR_SEND_WITH_INV:
+ return IB_WC_SEND;
+ case IB_WR_RDMA_READ:
+ return IB_WC_RDMA_READ;
+ case IB_WR_ATOMIC_CMP_AND_SWP:
+ return IB_WC_COMP_SWAP;
+ case IB_WR_ATOMIC_FETCH_AND_ADD:
+ return IB_WC_FETCH_ADD;
+ case IB_WR_REG_MR:
+ return IB_WC_REG_MR;
+ case IB_WR_LOCAL_INV:
+ return IB_WC_LOCAL_INV;
+ default:
+ return IB_WC_SEND;
+ }
+}
+
+inline bool qedr_can_post_send(struct qedr_qp *qp, struct ib_send_wr *wr)
+{
+ int wq_is_full, err_wr, pbl_is_full;
+ struct qedr_dev *dev = qp->dev;
+
+ /* prevent SQ overflow and/or processing of a bad WR */
+ err_wr = wr->num_sge > qp->sq.max_sges;
+ wq_is_full = qedr_wq_is_full(&qp->sq);
+ pbl_is_full = qed_chain_get_elem_left_u32(&qp->sq.pbl) <
+ QEDR_MAX_SQE_ELEMENTS_PER_SQE;
+ if (wq_is_full || err_wr || pbl_is_full) {
+ if (wq_is_full && !(qp->err_bitmap & QEDR_QP_ERR_SQ_FULL)) {
+ DP_ERR(dev,
+ "error: WQ is full. Post send on QP %p failed (this error appears only once)\n",
+ qp);
+ qp->err_bitmap |= QEDR_QP_ERR_SQ_FULL;
+ }
+
+ if (err_wr && !(qp->err_bitmap & QEDR_QP_ERR_BAD_SR)) {
+ DP_ERR(dev,
+ "error: WR is bad. Post send on QP %p failed (this error appears only once)\n",
+ qp);
+ qp->err_bitmap |= QEDR_QP_ERR_BAD_SR;
+ }
+
+ if (pbl_is_full &&
+ !(qp->err_bitmap & QEDR_QP_ERR_SQ_PBL_FULL)) {
+ DP_ERR(dev,
+ "error: WQ PBL is full. Post send on QP %p failed (this error appears only once)\n",
+ qp);
+ qp->err_bitmap |= QEDR_QP_ERR_SQ_PBL_FULL;
+ }
+ return false;
+ }
+ return true;
+}
+
+int __qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibqp->device);
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct rdma_sq_atomic_wqe_1st *awqe1;
+ struct rdma_sq_atomic_wqe_2nd *awqe2;
+ struct rdma_sq_atomic_wqe_3rd *awqe3;
+ struct rdma_sq_send_wqe_2st *swqe2;
+ struct rdma_sq_local_inv_wqe *iwqe;
+ struct rdma_sq_rdma_wqe_2nd *rwqe2;
+ struct rdma_sq_send_wqe_1st *swqe;
+ struct rdma_sq_rdma_wqe_1st *rwqe;
+ struct rdma_sq_fmr_wqe_1st *fwqe1;
+ struct rdma_sq_common_wqe *wqe;
+ u32 length;
+ int rc = 0;
+ bool comp;
+
+ if (!qedr_can_post_send(qp, wr)) {
+ *bad_wr = wr;
+ return -ENOMEM;
+ }
+
+ wqe = qed_chain_produce(&qp->sq.pbl);
+ qp->wqe_wr_id[qp->sq.prod].signaled =
+ !!(wr->send_flags & IB_SEND_SIGNALED) || qp->signaled;
+
+ wqe->flags = 0;
+ SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_SE_FLG,
+ !!(wr->send_flags & IB_SEND_SOLICITED));
+ comp = (!!(wr->send_flags & IB_SEND_SIGNALED)) || qp->signaled;
+ SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_COMP_FLG, comp);
+ SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_RD_FENCE_FLG,
+ !!(wr->send_flags & IB_SEND_FENCE));
+ wqe->prev_wqe_size = qp->prev_wqe_size;
+
+ qp->wqe_wr_id[qp->sq.prod].opcode = qedr_ib_to_wc_opcode(wr->opcode);
+
+ switch (wr->opcode) {
+ case IB_WR_SEND_WITH_IMM:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_IMM;
+ swqe = (struct rdma_sq_send_wqe_1st *)wqe;
+ swqe->wqe_size = 2;
+ swqe2 = qed_chain_produce(&qp->sq.pbl);
+
+ swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.imm_data);
+ length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2,
+ wr, bad_wr);
+ swqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
+ qp->prev_wqe_size = swqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
+ break;
+ case IB_WR_SEND:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_SEND;
+ swqe = (struct rdma_sq_send_wqe_1st *)wqe;
+
+ swqe->wqe_size = 2;
+ swqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2,
+ wr, bad_wr);
+ swqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
+ qp->prev_wqe_size = swqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
+ break;
+ case IB_WR_SEND_WITH_INV:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE;
+ swqe = (struct rdma_sq_send_wqe_1st *)wqe;
+ swqe2 = qed_chain_produce(&qp->sq.pbl);
+ swqe->wqe_size = 2;
+ swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.invalidate_rkey);
+ length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2,
+ wr, bad_wr);
+ swqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
+ qp->prev_wqe_size = swqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
+ break;
+
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM;
+ rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
+
+ rwqe->wqe_size = 2;
+ rwqe->imm_data = htonl(cpu_to_le32(wr->ex.imm_data));
+ rwqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2,
+ wr, bad_wr);
+ rwqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
+ qp->prev_wqe_size = rwqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
+ break;
+ case IB_WR_RDMA_WRITE:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR;
+ rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
+
+ rwqe->wqe_size = 2;
+ rwqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2,
+ wr, bad_wr);
+ rwqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
+ qp->prev_wqe_size = rwqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
+ break;
+ case IB_WR_RDMA_READ_WITH_INV:
+ DP_ERR(dev,
+ "RDMA READ WITH INVALIDATE not supported\n");
+ *bad_wr = wr;
+ rc = -EINVAL;
+ break;
+
+ case IB_WR_RDMA_READ:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_RD;
+ rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
+
+ rwqe->wqe_size = 2;
+ rwqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2,
+ wr, bad_wr);
+ rwqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
+ qp->prev_wqe_size = rwqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
+ break;
+
+ case IB_WR_ATOMIC_CMP_AND_SWP:
+ case IB_WR_ATOMIC_FETCH_AND_ADD:
+ awqe1 = (struct rdma_sq_atomic_wqe_1st *)wqe;
+ awqe1->wqe_size = 4;
+
+ awqe2 = qed_chain_produce(&qp->sq.pbl);
+ DMA_REGPAIR_LE(awqe2->remote_va, atomic_wr(wr)->remote_addr);
+ awqe2->r_key = cpu_to_le32(atomic_wr(wr)->rkey);
+
+ awqe3 = qed_chain_produce(&qp->sq.pbl);
+
+ if (wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
+ wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_ADD;
+ DMA_REGPAIR_LE(awqe3->swap_data,
+ atomic_wr(wr)->compare_add);
+ } else {
+ wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP;
+ DMA_REGPAIR_LE(awqe3->swap_data,
+ atomic_wr(wr)->swap);
+ DMA_REGPAIR_LE(awqe3->cmp_data,
+ atomic_wr(wr)->compare_add);
+ }
+
+ qedr_prepare_sq_sges(qp, NULL, wr);
+
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = awqe1->wqe_size;
+ qp->prev_wqe_size = awqe1->wqe_size;
+ break;
+
+ case IB_WR_LOCAL_INV:
+ iwqe = (struct rdma_sq_local_inv_wqe *)wqe;
+ iwqe->wqe_size = 1;
+
+ iwqe->req_type = RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE;
+ iwqe->inv_l_key = wr->ex.invalidate_rkey;
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = iwqe->wqe_size;
+ qp->prev_wqe_size = iwqe->wqe_size;
+ break;
+ case IB_WR_REG_MR:
+ DP_VERBOSE(dev, QEDR_MSG_CQ, "REG_MR\n");
+ wqe->req_type = RDMA_SQ_REQ_TYPE_FAST_MR;
+ fwqe1 = (struct rdma_sq_fmr_wqe_1st *)wqe;
+ fwqe1->wqe_size = 2;
+
+ rc = qedr_prepare_reg(qp, fwqe1, reg_wr(wr));
+ if (rc) {
+ DP_ERR(dev, "IB_REG_MR failed rc=%d\n", rc);
+ *bad_wr = wr;
+ break;
+ }
+
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = fwqe1->wqe_size;
+ qp->prev_wqe_size = fwqe1->wqe_size;
+ break;
+ default:
+ DP_ERR(dev, "invalid opcode 0x%x!\n", wr->opcode);
+ rc = -EINVAL;
+ *bad_wr = wr;
+ break;
+ }
+
+ if (*bad_wr) {
+ u16 value;
+
+ /* Restore prod to its position before
+ * this WR was processed
+ */
+ value = le16_to_cpu(qp->sq.db_data.data.value);
+ qed_chain_set_prod(&qp->sq.pbl, value, wqe);
+
+ /* Restore prev_wqe_size */
+ qp->prev_wqe_size = wqe->prev_wqe_size;
+ rc = -EINVAL;
+ DP_ERR(dev, "POST SEND FAILED\n");
+ }
+
+ return rc;
+}
+
+int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibqp->device);
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ unsigned long flags;
+ int rc = 0;
+
+ *bad_wr = NULL;
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
+ (qp->state == QED_ROCE_QP_STATE_ERR)) {
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ *bad_wr = wr;
+ DP_VERBOSE(dev, QEDR_MSG_CQ,
+ "QP in wrong state! QP icid=0x%x state %d\n",
+ qp->icid, qp->state);
+ return -EINVAL;
+ }
+
+ if (!wr) {
+ DP_ERR(dev, "Got an empty post send.\n");
+ return -EINVAL;
+ }
+
+ while (wr) {
+ rc = __qedr_post_send(ibqp, wr, bad_wr);
+ if (rc)
+ break;
+
+ qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id;
+
+ qedr_inc_sw_prod(&qp->sq);
+
+ qp->sq.db_data.data.value++;
+
+ wr = wr->next;
+ }
+
+ /* Trigger doorbell
+ * If there was a failure in the first WR then it will be triggered in
+ * vane. However this is not harmful (as long as the producer value is
+ * unchanged). For performance reasons we avoid checking for this
+ * redundant doorbell.
+ */
+ wmb();
+ writel(qp->sq.db_data.raw, qp->sq.db);
+
+ /* Make sure write sticks */
+ mmiowb();
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ return rc;
+}
+
+int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr)
+{
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qedr_dev *dev = qp->dev;
+ unsigned long flags;
+ int status = 0;
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
+ (qp->state == QED_ROCE_QP_STATE_ERR)) {
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ *bad_wr = wr;
+ return -EINVAL;
+ }
+
+ while (wr) {
+ int i;
+
+ if (qed_chain_get_elem_left_u32(&qp->rq.pbl) <
+ QEDR_MAX_RQE_ELEMENTS_PER_RQE ||
+ wr->num_sge > qp->rq.max_sges) {
+ DP_ERR(dev, "Can't post WR (%d < %d) || (%d > %d)\n",
+ qed_chain_get_elem_left_u32(&qp->rq.pbl),
+ QEDR_MAX_RQE_ELEMENTS_PER_RQE, wr->num_sge,
+ qp->rq.max_sges);
+ status = -ENOMEM;
+ *bad_wr = wr;
+ break;
+ }
+ for (i = 0; i < wr->num_sge; i++) {
+ u32 flags = 0;
+ struct rdma_rq_sge *rqe =
+ qed_chain_produce(&qp->rq.pbl);
+
+ /* First one must include the number
+ * of SGE in the list
+ */
+ if (!i)
+ SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES,
+ wr->num_sge);
+
+ SET_FIELD(flags, RDMA_RQ_SGE_L_KEY,
+ wr->sg_list[i].lkey);
+
+ RQ_SGE_SET(rqe, wr->sg_list[i].addr,
+ wr->sg_list[i].length, flags);
+ }
+
+ /* Special case of no sges. FW requires between 1-4 sges...
+ * in this case we need to post 1 sge with length zero. this is
+ * because rdma write with immediate consumes an RQ.
+ */
+ if (!wr->num_sge) {
+ u32 flags = 0;
+ struct rdma_rq_sge *rqe =
+ qed_chain_produce(&qp->rq.pbl);
+
+ /* First one must include the number
+ * of SGE in the list
+ */
+ SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, 0);
+ SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, 1);
+
+ RQ_SGE_SET(rqe, 0, 0, flags);
+ i = 1;
+ }
+
+ qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id;
+ qp->rqe_wr_id[qp->rq.prod].wqe_size = i;
+
+ qedr_inc_sw_prod(&qp->rq);
+
+ /* Flush all the writes before signalling doorbell */
+ wmb();
+
+ qp->rq.db_data.data.value++;
+
+ writel(qp->rq.db_data.raw, qp->rq.db);
+
+ /* Make sure write sticks */
+ mmiowb();
+
+ wr = wr->next;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ return status;
+}
+
+static int is_valid_cqe(struct qedr_cq *cq, union rdma_cqe *cqe)
+{
+ struct rdma_cqe_requester *resp_cqe = &cqe->req;
+
+ return (resp_cqe->flags & RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK) ==
+ cq->pbl_toggle;
+}
+
+static struct qedr_qp *cqe_get_qp(union rdma_cqe *cqe)
+{
+ struct rdma_cqe_requester *resp_cqe = &cqe->req;
+ struct qedr_qp *qp;
+
+ qp = (struct qedr_qp *)(uintptr_t)HILO_GEN(resp_cqe->qp_handle.hi,
+ resp_cqe->qp_handle.lo,
+ u64);
+ return qp;
+}
+
+static enum rdma_cqe_type cqe_get_type(union rdma_cqe *cqe)
+{
+ struct rdma_cqe_requester *resp_cqe = &cqe->req;
+
+ return GET_FIELD(resp_cqe->flags, RDMA_CQE_REQUESTER_TYPE);
+}
+
+/* Return latest CQE (needs processing) */
+static union rdma_cqe *get_cqe(struct qedr_cq *cq)
+{
+ return cq->latest_cqe;
+}
+
+/* In fmr we need to increase the number of fmr completed counter for the fmr
+ * algorithm determining whether we can free a pbl or not.
+ * we need to perform this whether the work request was signaled or not. for
+ * this purpose we call this function from the condition that checks if a wr
+ * should be skipped, to make sure we don't miss it ( possibly this fmr
+ * operation was not signalted)
+ */
+static inline void qedr_chk_if_fmr(struct qedr_qp *qp)
+{
+ if (qp->wqe_wr_id[qp->sq.cons].opcode == IB_WC_REG_MR)
+ qp->wqe_wr_id[qp->sq.cons].mr->info.completed++;
+}
+
+static int process_req(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, int num_entries,
+ struct ib_wc *wc, u16 hw_cons, enum ib_wc_status status,
+ int force)
+{
+ u16 cnt = 0;
+
+ while (num_entries && qp->sq.wqe_cons != hw_cons) {
+ if (!qp->wqe_wr_id[qp->sq.cons].signaled && !force) {
+ qedr_chk_if_fmr(qp);
+ /* skip WC */
+ goto next_cqe;
+ }
+
+ /* fill WC */
+ wc->status = status;
+ wc->wc_flags = 0;
+ wc->src_qp = qp->id;
+ wc->qp = &qp->ibqp;
+
+ wc->wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id;
+ wc->opcode = qp->wqe_wr_id[qp->sq.cons].opcode;
+
+ switch (wc->opcode) {
+ case IB_WC_RDMA_WRITE:
+ wc->byte_len = qp->wqe_wr_id[qp->sq.cons].bytes_len;
+ break;
+ case IB_WC_COMP_SWAP:
+ case IB_WC_FETCH_ADD:
+ wc->byte_len = 8;
+ break;
+ case IB_WC_REG_MR:
+ qp->wqe_wr_id[qp->sq.cons].mr->info.completed++;
+ break;
+ default:
+ break;
+ }
+
+ num_entries--;
+ wc++;
+ cnt++;
+next_cqe:
+ while (qp->wqe_wr_id[qp->sq.cons].wqe_size--)
+ qed_chain_consume(&qp->sq.pbl);
+ qedr_inc_sw_cons(&qp->sq);
+ }
+
+ return cnt;
+}
+
+static int qedr_poll_cq_req(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct qedr_cq *cq,
+ int num_entries, struct ib_wc *wc,
+ struct rdma_cqe_requester *req)
+{
+ int cnt = 0;
+
+ switch (req->status) {
+ case RDMA_CQE_REQ_STS_OK:
+ cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons,
+ IB_WC_SUCCESS, 0);
+ break;
+ case RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons,
+ IB_WC_WR_FLUSH_ERR, 0);
+ break;
+ default:
+ /* process all WQE before the cosumer */
+ qp->state = QED_ROCE_QP_STATE_ERR;
+ cnt = process_req(dev, qp, cq, num_entries, wc,
+ req->sq_cons - 1, IB_WC_SUCCESS, 0);
+ wc += cnt;
+ /* if we have extra WC fill it with actual error info */
+ if (cnt < num_entries) {
+ enum ib_wc_status wc_status;
+
+ switch (req->status) {
+ case RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_BAD_RESP_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_LOC_LEN_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_LOC_QP_OP_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_LOC_PROT_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_MW_BIND_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_REM_INV_REQ_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_REM_ACCESS_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_REM_OP_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_RNR_RETRY_EXC_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with ROCE_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_RETRY_EXC_ERR;
+ break;
+ default:
+ DP_ERR(dev,
+ "Error: POLL CQ with IB_WC_GENERAL_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_GENERAL_ERR;
+ }
+ cnt += process_req(dev, qp, cq, 1, wc, req->sq_cons,
+ wc_status, 1);
+ }
+ }
+
+ return cnt;
+}
+
+static void __process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, struct ib_wc *wc,
+ struct rdma_cqe_responder *resp, u64 wr_id)
+{
+ enum ib_wc_status wc_status = IB_WC_SUCCESS;
+ u8 flags;
+
+ wc->opcode = IB_WC_RECV;
+ wc->wc_flags = 0;
+
+ switch (resp->status) {
+ case RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR:
+ wc_status = IB_WC_LOC_ACCESS_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR:
+ wc_status = IB_WC_LOC_LEN_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR:
+ wc_status = IB_WC_LOC_QP_OP_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR:
+ wc_status = IB_WC_LOC_PROT_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR:
+ wc_status = IB_WC_MW_BIND_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR:
+ wc_status = IB_WC_REM_INV_RD_REQ_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_OK:
+ wc_status = IB_WC_SUCCESS;
+ wc->byte_len = le32_to_cpu(resp->length);
+
+ flags = resp->flags & QEDR_RESP_RDMA_IMM;
+
+ switch (flags) {
+ case QEDR_RESP_RDMA_IMM:
+ /* update opcode */
+ wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+ /* fall to set imm data */
+ case QEDR_RESP_IMM:
+ wc->ex.imm_data =
+ le32_to_cpu(resp->imm_data_or_inv_r_Key);
+ wc->wc_flags |= IB_WC_WITH_IMM;
+ break;
+ case QEDR_RESP_RDMA:
+ DP_ERR(dev, "Invalid flags detected\n");
+ break;
+ default:
+ DP_ERR(dev, "Invalid opcode\n");
+ }
+
+ break;
+ default:
+ wc->status = IB_WC_GENERAL_ERR;
+ DP_ERR(dev, "Invalid CQE status detected\n");
+ }
+
+ /* fill WC */
+ wc->status = wc_status;
+ wc->src_qp = qp->id;
+ wc->qp = &qp->ibqp;
+ wc->wr_id = wr_id;
+}
+
+static int process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, struct ib_wc *wc,
+ struct rdma_cqe_responder *resp)
+{
+ u64 wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
+
+ __process_resp_one(dev, qp, cq, wc, resp, wr_id);
+
+ while (qp->rqe_wr_id[qp->rq.cons].wqe_size--)
+ qed_chain_consume(&qp->rq.pbl);
+ qedr_inc_sw_cons(&qp->rq);
+
+ return 1;
+}
+
+static int process_resp_flush(struct qedr_qp *qp, struct qedr_cq *cq,
+ int num_entries, struct ib_wc *wc, u16 hw_cons)
+{
+ u16 cnt = 0;
+
+ while (num_entries && qp->rq.wqe_cons != hw_cons) {
+ /* fill WC */
+ wc->status = IB_WC_WR_FLUSH_ERR;
+ wc->wc_flags = 0;
+ wc->src_qp = qp->id;
+ wc->byte_len = 0;
+ wc->wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
+ wc->qp = &qp->ibqp;
+ num_entries--;
+ wc++;
+ cnt++;
+ while (qp->rqe_wr_id[qp->rq.cons].wqe_size--)
+ qed_chain_consume(&qp->rq.pbl);
+ qedr_inc_sw_cons(&qp->rq);
+ }
+
+ return cnt;
+}
+
+static void try_consume_resp_cqe(struct qedr_cq *cq, struct qedr_qp *qp,
+ struct rdma_cqe_responder *resp, int *update)
+{
+ if (le16_to_cpu(resp->rq_cons) == qp->rq.wqe_cons) {
+ consume_cqe(cq);
+ *update |= 1;
+ }
+}
+
+static int qedr_poll_cq_resp(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, int num_entries,
+ struct ib_wc *wc, struct rdma_cqe_responder *resp,
+ int *update)
+{
+ int cnt;
+
+ if (resp->status == RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR) {
+ cnt = process_resp_flush(qp, cq, num_entries, wc,
+ resp->rq_cons);
+ try_consume_resp_cqe(cq, qp, resp, update);
+ } else {
+ cnt = process_resp_one(dev, qp, cq, wc, resp);
+ consume_cqe(cq);
+ *update |= 1;
+ }
+
+ return cnt;
+}
+
+static void try_consume_req_cqe(struct qedr_cq *cq, struct qedr_qp *qp,
+ struct rdma_cqe_requester *req, int *update)
+{
+ if (le16_to_cpu(req->sq_cons) == qp->sq.wqe_cons) {
+ consume_cqe(cq);
+ *update |= 1;
+ }
+}
+
+int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+ union rdma_cqe *cqe = cq->latest_cqe;
+ u32 old_cons, new_cons;
+ unsigned long flags;
+ int update = 0;
+ int done = 0;
+
+ spin_lock_irqsave(&cq->cq_lock, flags);
+ old_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
+ while (num_entries && is_valid_cqe(cq, cqe)) {
+ struct qedr_qp *qp;
+ int cnt = 0;
+
+ /* prevent speculative reads of any field of CQE */
+ rmb();
+
+ qp = cqe_get_qp(cqe);
+ if (!qp) {
+ WARN(1, "Error: CQE QP pointer is NULL. CQE=%p\n", cqe);
+ break;
+ }
+
+ wc->qp = &qp->ibqp;
+
+ switch (cqe_get_type(cqe)) {
+ case RDMA_CQE_TYPE_REQUESTER:
+ cnt = qedr_poll_cq_req(dev, qp, cq, num_entries, wc,
+ &cqe->req);
+ try_consume_req_cqe(cq, qp, &cqe->req, &update);
+ break;
+ case RDMA_CQE_TYPE_RESPONDER_RQ:
+ cnt = qedr_poll_cq_resp(dev, qp, cq, num_entries, wc,
+ &cqe->resp, &update);
+ break;
+ case RDMA_CQE_TYPE_INVALID:
+ default:
+ DP_ERR(dev, "Error: invalid CQE type = %d\n",
+ cqe_get_type(cqe));
+ }
+ num_entries -= cnt;
+ wc += cnt;
+ done += cnt;
+
+ cqe = get_cqe(cq);
+ }
+ new_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
+
+ cq->cq_cons += new_cons - old_cons;
+
+ if (update)
+ /* doorbell notifies abount latest VALID entry,
+ * but chain already point to the next INVALID one
+ */
+ doorbell_cq(cq, cq->cq_cons - 1, cq->arm_flags);
+
+ spin_unlock_irqrestore(&cq->cq_lock, flags);
+ return done;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index e7cb3fd..9eb2722b 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -58,4 +58,9 @@ int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
struct ib_mr *qedr_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
u32 max_num_sg);
+int qedr_poll_cq(struct ib_cq *, int num_entries, struct ib_wc *wc);
+int qedr_post_send(struct ib_qp *, struct ib_send_wr *,
+ struct ib_send_wr **bad_wr);
+int qedr_post_recv(struct ib_qp *, struct ib_recv_wr *,
+ struct ib_recv_wr **bad_wr);
#endif
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 08/11] Add support for data path
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Implement fastpath verbs like ib_send_post, ib_post_recv and ib_poll_cq.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 9 +-
drivers/infiniband/hw/qedr/qedr.h | 19 +
drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 562 +++++++++++++++
drivers/infiniband/hw/qedr/verbs.c | 1062 ++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 5 +
5 files changed, 1656 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 3d4feb7..42206ed 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -108,7 +108,10 @@ static int qedr_register_device(struct qedr_dev *dev)
QEDR_UVERBS(QUERY_QP) |
QEDR_UVERBS(DESTROY_QP) |
QEDR_UVERBS(REG_MR) |
- QEDR_UVERBS(DEREG_MR);
+ QEDR_UVERBS(DEREG_MR) |
+ QEDR_UVERBS(POLL_CQ) |
+ QEDR_UVERBS(POST_SEND) |
+ QEDR_UVERBS(POST_RECV);
dev->ibdev.phys_port_cnt = 1;
dev->ibdev.num_comp_vectors = dev->num_cnq;
@@ -149,6 +152,10 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.alloc_mr = qedr_alloc_mr;
dev->ibdev.map_mr_sg = qedr_map_mr_sg;
+ dev->ibdev.poll_cq = qedr_poll_cq;
+ dev->ibdev.post_send = qedr_post_send;
+ dev->ibdev.post_recv = qedr_post_recv;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 3cd4e52..bc218d2 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -409,6 +409,25 @@ struct qedr_mr {
u32 npages;
};
+#define SET_FIELD2(value, name, flag) ((value) |= ((flag) << (name ## _SHIFT)))
+
+#define QEDR_RESP_IMM (RDMA_CQE_RESPONDER_IMM_FLG_MASK << \
+ RDMA_CQE_RESPONDER_IMM_FLG_SHIFT)
+#define QEDR_RESP_RDMA (RDMA_CQE_RESPONDER_RDMA_FLG_MASK << \
+ RDMA_CQE_RESPONDER_RDMA_FLG_SHIFT)
+#define QEDR_RESP_RDMA_IMM (QEDR_RESP_IMM | QEDR_RESP_RDMA)
+
+static inline void qedr_inc_sw_cons(struct qedr_qp_hwq_info *info)
+{
+ info->cons = (info->cons + 1) % info->max_wr;
+ info->wqe_cons++;
+}
+
+static inline void qedr_inc_sw_prod(struct qedr_qp_hwq_info *info)
+{
+ info->prod = (info->prod + 1) % info->max_wr;
+}
+
static inline int qedr_get_dmac(struct qedr_dev *dev,
struct ib_ah_attr *ah_attr, u8 *mac_addr)
{
diff --git a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
index 4770559..5c98d20 100644
--- a/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
+++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
@@ -150,6 +150,12 @@ struct rdma_rq_sge {
struct regpair addr;
__le32 length;
__le32 flags;
+#define RDMA_RQ_SGE_L_KEY_MASK 0x3FFFFFF
+#define RDMA_RQ_SGE_L_KEY_SHIFT 0
+#define RDMA_RQ_SGE_NUM_SGES_MASK 0x7
+#define RDMA_RQ_SGE_NUM_SGES_SHIFT 26
+#define RDMA_RQ_SGE_RESERVED0_MASK 0x7
+#define RDMA_RQ_SGE_RESERVED0_SHIFT 29
};
struct rdma_srq_sge {
@@ -183,4 +189,560 @@ struct rdma_pwm_val32_data {
__le32 value;
};
+/* DIF Block size options */
+enum rdma_dif_block_size {
+ RDMA_DIF_BLOCK_512 = 0,
+ RDMA_DIF_BLOCK_4096 = 1,
+ MAX_RDMA_DIF_BLOCK_SIZE
+};
+
+/* DIF CRC initial value */
+enum rdma_dif_crc_seed {
+ RDMA_DIF_CRC_SEED_0000 = 0,
+ RDMA_DIF_CRC_SEED_FFFF = 1,
+ MAX_RDMA_DIF_CRC_SEED
+};
+
+/* RDMA DIF Error Result Structure */
+struct rdma_dif_error_result {
+ __le32 error_intervals;
+ __le32 dif_error_1st_interval;
+ u8 flags;
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_CRC_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_CRC_SHIFT 0
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_APP_TAG_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_APP_TAG_SHIFT 1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_REF_TAG_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_DIF_ERROR_TYPE_REF_TAG_SHIFT 2
+#define RDMA_DIF_ERROR_RESULT_RESERVED0_MASK 0xF
+#define RDMA_DIF_ERROR_RESULT_RESERVED0_SHIFT 3
+#define RDMA_DIF_ERROR_RESULT_TOGGLE_BIT_MASK 0x1
+#define RDMA_DIF_ERROR_RESULT_TOGGLE_BIT_SHIFT 7
+ u8 reserved1[55];
+};
+
+/* DIF IO direction */
+enum rdma_dif_io_direction_flg {
+ RDMA_DIF_DIR_RX = 0,
+ RDMA_DIF_DIR_TX = 1,
+ MAX_RDMA_DIF_IO_DIRECTION_FLG
+};
+
+/* RDMA DIF Runt Result Structure */
+struct rdma_dif_runt_result {
+ __le16 guard_tag;
+ __le16 reserved[3];
+};
+
+/* Memory window type enumeration */
+enum rdma_mw_type {
+ RDMA_MW_TYPE_1,
+ RDMA_MW_TYPE_2A,
+ MAX_RDMA_MW_TYPE
+};
+
+struct rdma_sq_atomic_wqe {
+ __le32 reserved1;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_ATOMIC_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_ATOMIC_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_ATOMIC_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_ATOMIC_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_ATOMIC_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_ATOMIC_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_ATOMIC_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_ATOMIC_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ struct regpair remote_va;
+ __le32 r_key;
+ __le32 reserved2;
+ struct regpair cmp_data;
+ struct regpair swap_data;
+};
+
+/* First element (16 bytes) of atomic wqe */
+struct rdma_sq_atomic_wqe_1st {
+ __le32 reserved1;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_ATOMIC_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_ATOMIC_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_ATOMIC_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_ATOMIC_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_ATOMIC_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_ATOMIC_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_ATOMIC_WQE_1ST_RESERVED0_MASK 0x7
+#define RDMA_SQ_ATOMIC_WQE_1ST_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of atomic wqe */
+struct rdma_sq_atomic_wqe_2nd {
+ struct regpair remote_va;
+ __le32 r_key;
+ __le32 reserved2;
+};
+
+/* Third element (16 bytes) of atomic wqe */
+struct rdma_sq_atomic_wqe_3rd {
+ struct regpair cmp_data;
+ struct regpair swap_data;
+};
+
+struct rdma_sq_bind_wqe {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_BIND_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_BIND_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_BIND_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_BIND_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_BIND_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_BIND_WQE_RESERVED0_MASK 0x7
+#define RDMA_SQ_BIND_WQE_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ u8 bind_ctrl;
+#define RDMA_SQ_BIND_WQE_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_BIND_WQE_ZERO_BASED_SHIFT 0
+#define RDMA_SQ_BIND_WQE_MW_TYPE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_MW_TYPE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_RESERVED1_MASK 0x3F
+#define RDMA_SQ_BIND_WQE_RESERVED1_SHIFT 2
+ u8 access_ctrl;
+#define RDMA_SQ_BIND_WQE_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_BIND_WQE_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_BIND_WQE_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_BIND_WQE_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_BIND_WQE_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_BIND_WQE_RESERVED2_MASK 0x7
+#define RDMA_SQ_BIND_WQE_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ __le32 parent_l_key;
+ __le32 reserved4;
+};
+
+/* First element (16 bytes) of bind wqe */
+struct rdma_sq_bind_wqe_1st {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_BIND_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_BIND_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_BIND_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_BIND_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_BIND_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_BIND_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_BIND_WQE_1ST_RESERVED0_MASK 0x7
+#define RDMA_SQ_BIND_WQE_1ST_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of bind wqe */
+struct rdma_sq_bind_wqe_2nd {
+ u8 bind_ctrl;
+#define RDMA_SQ_BIND_WQE_2ND_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_ZERO_BASED_SHIFT 0
+#define RDMA_SQ_BIND_WQE_2ND_MW_TYPE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_MW_TYPE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED1_MASK 0x3F
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED1_SHIFT 2
+ u8 access_ctrl;
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_BIND_WQE_2ND_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_BIND_WQE_2ND_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED2_MASK 0x7
+#define RDMA_SQ_BIND_WQE_2ND_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ __le32 parent_l_key;
+ __le32 reserved4;
+};
+
+/* Structure with only the SQ WQE common
+ * fields. Size is of one SQ element (16B)
+ */
+struct rdma_sq_common_wqe {
+ __le32 reserved1[3];
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_COMMON_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_COMMON_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_COMMON_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_COMMON_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_COMMON_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_COMMON_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_COMMON_WQE_RESERVED0_MASK 0x7
+#define RDMA_SQ_COMMON_WQE_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+struct rdma_sq_fmr_wqe {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_FMR_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_FMR_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_FMR_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_FMR_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ u8 fmr_ctrl;
+#define RDMA_SQ_FMR_WQE_PAGE_SIZE_LOG_MASK 0x1F
+#define RDMA_SQ_FMR_WQE_PAGE_SIZE_LOG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_ZERO_BASED_SHIFT 5
+#define RDMA_SQ_FMR_WQE_BIND_EN_MASK 0x1
+#define RDMA_SQ_FMR_WQE_BIND_EN_SHIFT 6
+#define RDMA_SQ_FMR_WQE_RESERVED1_MASK 0x1
+#define RDMA_SQ_FMR_WQE_RESERVED1_SHIFT 7
+ u8 access_ctrl;
+#define RDMA_SQ_FMR_WQE_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_FMR_WQE_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_FMR_WQE_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_FMR_WQE_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_FMR_WQE_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_FMR_WQE_RESERVED2_MASK 0x7
+#define RDMA_SQ_FMR_WQE_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ struct regpair pbl_addr;
+ __le32 dif_base_ref_tag;
+ __le16 dif_app_tag;
+ __le16 dif_app_tag_mask;
+ __le16 dif_runt_crc_value;
+ __le16 dif_flags;
+#define RDMA_SQ_FMR_WQE_DIF_IO_DIRECTION_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_IO_DIRECTION_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_BLOCK_SIZE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_DIF_RUNT_VALID_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_RUNT_VALID_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_CRC_GUARD_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_CRC_GUARD_SHIFT 3
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_REF_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_REF_TAG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_APP_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_VALIDATE_APP_TAG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_DIF_CRC_SEED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_DIF_CRC_SEED_SHIFT 6
+#define RDMA_SQ_FMR_WQE_RESERVED4_MASK 0x1FF
+#define RDMA_SQ_FMR_WQE_RESERVED4_SHIFT 7
+ __le32 Reserved5;
+};
+
+/* First element (16 bytes) of fmr wqe */
+struct rdma_sq_fmr_wqe_1st {
+ struct regpair addr;
+ __le32 l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_FMR_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_FMR_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_FMR_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_1ST_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_1ST_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_1ST_RESERVED0_MASK 0x3
+#define RDMA_SQ_FMR_WQE_1ST_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of fmr wqe */
+struct rdma_sq_fmr_wqe_2nd {
+ u8 fmr_ctrl;
+#define RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG_MASK 0x1F
+#define RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_2ND_ZERO_BASED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_ZERO_BASED_SHIFT 5
+#define RDMA_SQ_FMR_WQE_2ND_BIND_EN_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_BIND_EN_SHIFT 6
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED1_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED1_SHIFT 7
+ u8 access_ctrl;
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_READ_SHIFT 0
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC_SHIFT 2
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_READ_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_READ_SHIFT 3
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE_SHIFT 4
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED2_MASK 0x7
+#define RDMA_SQ_FMR_WQE_2ND_RESERVED2_SHIFT 5
+ u8 reserved3;
+ u8 length_hi;
+ __le32 length_lo;
+ struct regpair pbl_addr;
+};
+
+/* Third element (16 bytes) of fmr wqe */
+struct rdma_sq_fmr_wqe_3rd {
+ __le32 dif_base_ref_tag;
+ __le16 dif_app_tag;
+ __le16 dif_app_tag_mask;
+ __le16 dif_runt_crc_value;
+ __le16 dif_flags;
+#define RDMA_SQ_FMR_WQE_3RD_DIF_IO_DIRECTION_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_IO_DIRECTION_FLG_SHIFT 0
+#define RDMA_SQ_FMR_WQE_3RD_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_BLOCK_SIZE_SHIFT 1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_RUNT_VALID_FLG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_RUNT_VALID_FLG_SHIFT 2
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_CRC_GUARD_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_CRC_GUARD_SHIFT 3
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_REF_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_REF_TAG_SHIFT 4
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_APP_TAG_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_VALIDATE_APP_TAG_SHIFT 5
+#define RDMA_SQ_FMR_WQE_3RD_DIF_CRC_SEED_MASK 0x1
+#define RDMA_SQ_FMR_WQE_3RD_DIF_CRC_SEED_SHIFT 6
+#define RDMA_SQ_FMR_WQE_3RD_RESERVED4_MASK 0x1FF
+#define RDMA_SQ_FMR_WQE_3RD_RESERVED4_SHIFT 7
+ __le32 Reserved5;
+};
+
+struct rdma_sq_local_inv_wqe {
+ struct regpair reserved;
+ __le32 inv_l_key;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_LOCAL_INV_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_LOCAL_INV_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_LOCAL_INV_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_LOCAL_INV_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_LOCAL_INV_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_LOCAL_INV_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_LOCAL_INV_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_LOCAL_INV_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_LOCAL_INV_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+struct rdma_sq_rdma_wqe {
+ __le32 imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_RDMA_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_RDMA_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_RDMA_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_RDMA_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_RDMA_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ struct regpair remote_va;
+ __le32 r_key;
+ u8 dif_flags;
+#define RDMA_SQ_RDMA_WQE_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_BLOCK_SIZE_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_DIF_FIRST_RDMA_IN_IO_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_FIRST_RDMA_IN_IO_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_DIF_LAST_RDMA_IN_IO_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_DIF_LAST_RDMA_IN_IO_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_RESERVED1_MASK 0x1F
+#define RDMA_SQ_RDMA_WQE_RESERVED1_SHIFT 3
+ u8 reserved2[3];
+};
+
+/* First element (16 bytes) of rdma wqe */
+struct rdma_sq_rdma_wqe_1st {
+ __le32 imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_RDMA_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_RDMA_WQE_1ST_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_1ST_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_RDMA_WQE_1ST_RESERVED0_MASK 0x3
+#define RDMA_SQ_RDMA_WQE_1ST_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+/* Second element (16 bytes) of rdma wqe */
+struct rdma_sq_rdma_wqe_2nd {
+ struct regpair remote_va;
+ __le32 r_key;
+ u8 dif_flags;
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_BLOCK_SIZE_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_BLOCK_SIZE_SHIFT 0
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_FIRST_SEGMENT_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_FIRST_SEGMENT_FLG_SHIFT 1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_LAST_SEGMENT_FLG_MASK 0x1
+#define RDMA_SQ_RDMA_WQE_2ND_DIF_LAST_SEGMENT_FLG_SHIFT 2
+#define RDMA_SQ_RDMA_WQE_2ND_RESERVED1_MASK 0x1F
+#define RDMA_SQ_RDMA_WQE_2ND_RESERVED1_SHIFT 3
+ u8 reserved2[3];
+};
+
+/* SQ WQE req type enumeration */
+enum rdma_sq_req_type {
+ RDMA_SQ_REQ_TYPE_SEND,
+ RDMA_SQ_REQ_TYPE_SEND_WITH_IMM,
+ RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE,
+ RDMA_SQ_REQ_TYPE_RDMA_WR,
+ RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM,
+ RDMA_SQ_REQ_TYPE_RDMA_RD,
+ RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP,
+ RDMA_SQ_REQ_TYPE_ATOMIC_ADD,
+ RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE,
+ RDMA_SQ_REQ_TYPE_FAST_MR,
+ RDMA_SQ_REQ_TYPE_BIND,
+ RDMA_SQ_REQ_TYPE_INVALID,
+ MAX_RDMA_SQ_REQ_TYPE
+};
+
+struct rdma_sq_send_wqe {
+ __le32 inv_key_or_imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_SEND_WQE_COMP_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_COMP_FLG_SHIFT 0
+#define RDMA_SQ_SEND_WQE_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_SEND_WQE_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_SEND_WQE_SE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_SE_FLG_SHIFT 3
+#define RDMA_SQ_SEND_WQE_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_SEND_WQE_DIF_ON_HOST_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_DIF_ON_HOST_FLG_SHIFT 5
+#define RDMA_SQ_SEND_WQE_RESERVED0_MASK 0x3
+#define RDMA_SQ_SEND_WQE_RESERVED0_SHIFT 6
+ u8 wqe_size;
+ u8 prev_wqe_size;
+ __le32 reserved1[4];
+};
+
+struct rdma_sq_send_wqe_1st {
+ __le32 inv_key_or_imm_data;
+ __le32 length;
+ __le32 xrc_srq;
+ u8 req_type;
+ u8 flags;
+#define RDMA_SQ_SEND_WQE_1ST_COMP_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_COMP_FLG_SHIFT 0
+#define RDMA_SQ_SEND_WQE_1ST_RD_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_RD_FENCE_FLG_SHIFT 1
+#define RDMA_SQ_SEND_WQE_1ST_INV_FENCE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_INV_FENCE_FLG_SHIFT 2
+#define RDMA_SQ_SEND_WQE_1ST_SE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_SE_FLG_SHIFT 3
+#define RDMA_SQ_SEND_WQE_1ST_INLINE_FLG_MASK 0x1
+#define RDMA_SQ_SEND_WQE_1ST_INLINE_FLG_SHIFT 4
+#define RDMA_SQ_SEND_WQE_1ST_RESERVED0_MASK 0x7
+#define RDMA_SQ_SEND_WQE_1ST_RESERVED0_SHIFT 5
+ u8 wqe_size;
+ u8 prev_wqe_size;
+};
+
+struct rdma_sq_send_wqe_2st {
+ __le32 reserved1[4];
+};
+
#endif /* __QED_HSI_RDMA__ */
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 3d6045c..979555e 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -2439,3 +2439,1065 @@ err1:
kfree(mr);
return ERR_PTR(rc);
}
+
+static inline int qedr_wq_is_full(struct qedr_qp_hwq_info *wq)
+{
+ return (((wq->prod + 1) % wq->max_wr) == wq->cons);
+}
+
+static int sge_data_len(struct ib_sge *sg_list, int num_sge)
+{
+ int i, len = 0;
+
+ for (i = 0; i < num_sge; i++)
+ len += sg_list[i].length;
+
+ return len;
+}
+
+static void swap_wqe_data64(u64 *p)
+{
+ int i;
+
+ for (i = 0; i < QEDR_SQE_ELEMENT_SIZE / sizeof(u64); i++, p++)
+ *p = cpu_to_be64(cpu_to_le64(*p));
+}
+
+static u32 qedr_prepare_sq_inline_data(struct qedr_dev *dev,
+ struct qedr_qp *qp, u8 *wqe_size,
+ struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr, u8 *bits,
+ u8 bit)
+{
+ u32 data_size = sge_data_len(wr->sg_list, wr->num_sge);
+ char *seg_prt, *wqe;
+ int i, seg_siz;
+
+ if (data_size > ROCE_REQ_MAX_INLINE_DATA_SIZE) {
+ DP_ERR(dev, "Too much inline data in WR: %d\n", data_size);
+ *bad_wr = wr;
+ return 0;
+ }
+
+ if (!data_size)
+ return data_size;
+
+ *bits |= bit;
+
+ seg_prt = NULL;
+ wqe = NULL;
+ seg_siz = 0;
+
+ /* Copy data inline */
+ for (i = 0; i < wr->num_sge; i++) {
+ u32 len = wr->sg_list[i].length;
+ void *src = (void *)(uintptr_t)wr->sg_list[i].addr;
+
+ while (len > 0) {
+ u32 cur;
+
+ /* New segment required */
+ if (!seg_siz) {
+ wqe = (char *)qed_chain_produce(&qp->sq.pbl);
+ seg_prt = wqe;
+ seg_siz = sizeof(struct rdma_sq_common_wqe);
+ (*wqe_size)++;
+ }
+
+ /* Calculate currently allowed length */
+ cur = min_t(u32, len, seg_siz);
+ memcpy(seg_prt, src, cur);
+
+ /* Update segment variables */
+ seg_prt += cur;
+ seg_siz -= cur;
+
+ /* Update sge variables */
+ src += cur;
+ len -= cur;
+
+ /* Wwap fully-completed segments */
+ if (!seg_siz)
+ swap_wqe_data64((u64 *)wqe);
+ }
+ }
+
+ /* swap last not completed segment */
+ if (seg_siz)
+ swap_wqe_data64((u64 *)wqe);
+
+ return data_size;
+}
+
+#define RQ_SGE_SET(sge, vaddr, vlength, vflags) \
+ do { \
+ DMA_REGPAIR_LE(sge->addr, vaddr); \
+ (sge)->length = cpu_to_le32(vlength); \
+ (sge)->flags = cpu_to_le32(vflags); \
+ } while (0)
+
+#define SRQ_HDR_SET(hdr, vwr_id, num_sge) \
+ do { \
+ DMA_REGPAIR_LE(hdr->wr_id, vwr_id); \
+ (hdr)->num_sges = num_sge; \
+ } while (0)
+
+#define SRQ_SGE_SET(sge, vaddr, vlength, vlkey) \
+ do { \
+ DMA_REGPAIR_LE(sge->addr, vaddr); \
+ (sge)->length = cpu_to_le32(vlength); \
+ (sge)->l_key = cpu_to_le32(vlkey); \
+ } while (0)
+
+static u32 qedr_prepare_sq_sges(struct qedr_qp *qp, u8 *wqe_size,
+ struct ib_send_wr *wr)
+{
+ u32 data_size = 0;
+ int i;
+
+ for (i = 0; i < wr->num_sge; i++) {
+ struct rdma_sq_sge *sge = qed_chain_produce(&qp->sq.pbl);
+
+ DMA_REGPAIR_LE(sge->addr, wr->sg_list[i].addr);
+ sge->l_key = cpu_to_le32(wr->sg_list[i].lkey);
+ sge->length = cpu_to_le32(wr->sg_list[i].length);
+ data_size += wr->sg_list[i].length;
+ }
+
+ if (wqe_size)
+ *wqe_size += wr->num_sge;
+
+ return data_size;
+}
+
+static u32 qedr_prepare_sq_rdma_data(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct rdma_sq_rdma_wqe_1st *rwqe,
+ struct rdma_sq_rdma_wqe_2nd *rwqe2,
+ struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ rwqe2->r_key = cpu_to_le32(rdma_wr(wr)->rkey);
+ DMA_REGPAIR_LE(rwqe2->remote_va, rdma_wr(wr)->remote_addr);
+
+ if (wr->send_flags & IB_SEND_INLINE) {
+ u8 flags = 0;
+
+ SET_FIELD2(flags, RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG, 1);
+ return qedr_prepare_sq_inline_data(dev, qp, &rwqe->wqe_size, wr,
+ bad_wr, &rwqe->flags, flags);
+ }
+
+ return qedr_prepare_sq_sges(qp, &rwqe->wqe_size, wr);
+}
+
+static u32 qedr_prepare_sq_send_data(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct rdma_sq_send_wqe_1st *swqe,
+ struct rdma_sq_send_wqe_2st *swqe2,
+ struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ memset(swqe2, 0, sizeof(*swqe2));
+ if (wr->send_flags & IB_SEND_INLINE) {
+ u8 flags = 0;
+
+ SET_FIELD2(flags, RDMA_SQ_SEND_WQE_INLINE_FLG, 1);
+ return qedr_prepare_sq_inline_data(dev, qp, &swqe->wqe_size, wr,
+ bad_wr, &swqe->flags, flags);
+ }
+
+ return qedr_prepare_sq_sges(qp, &swqe->wqe_size, wr);
+}
+
+static int qedr_prepare_reg(struct qedr_qp *qp,
+ struct rdma_sq_fmr_wqe_1st *fwqe1,
+ struct ib_reg_wr *wr)
+{
+ struct qedr_mr *mr = get_qedr_mr(wr->mr);
+ struct qedr_pbl *pbl_table = mr->info.pbl_table;
+ struct rdma_sq_fmr_wqe_2nd *fwqe2;
+ int num_pbes, total_num_pbes = 0;
+ struct qedr_dev *dev = qp->dev;
+ struct regpair *pbe;
+ int i;
+
+ fwqe2 = (struct rdma_sq_fmr_wqe_2nd *)qed_chain_produce(&qp->sq.pbl);
+ fwqe1->addr.hi = upper_32_bits(mr->ibmr.iova);
+ fwqe1->addr.lo = lower_32_bits(mr->ibmr.iova);
+ fwqe1->l_key = wr->key;
+
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_READ,
+ !!(wr->access & IB_ACCESS_REMOTE_READ));
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE,
+ !!(wr->access & IB_ACCESS_REMOTE_WRITE));
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC,
+ !!(wr->access & IB_ACCESS_REMOTE_ATOMIC));
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_READ, 1);
+ SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE,
+ !!(wr->access & IB_ACCESS_LOCAL_WRITE));
+ fwqe2->fmr_ctrl = 0;
+
+ SET_FIELD2(fwqe2->fmr_ctrl, RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG,
+ ilog2(mr->ibmr.page_size) - 12);
+
+ fwqe2->length_hi = 0;
+ fwqe2->length_lo = mr->ibmr.length;
+ fwqe2->pbl_addr.hi = upper_32_bits(pbl_table->pa);
+ fwqe2->pbl_addr.lo = lower_32_bits(pbl_table->pa);
+ DP_VERBOSE(dev, QEDR_MSG_MR,
+ "Prepare fmr...rkey=%x addr=%x:%x length = %x pbl_addr %x:%x\n",
+ wr->key, fwqe1->addr.hi, fwqe1->addr.lo,
+ fwqe2->length_lo, fwqe2->pbl_addr.hi, fwqe2->pbl_addr.lo);
+
+ pbe = (struct regpair *)pbl_table->va;
+ num_pbes = 0;
+
+ for (i = 0; i < mr->npages &&
+ (total_num_pbes != mr->info.pbl_info.num_pbes); i++) {
+ u64 buf_addr = mr->pages[i];
+
+ pbe->lo = cpu_to_le32((u32)buf_addr);
+ pbe->hi = cpu_to_le32((u32)upper_32_bits(buf_addr));
+
+ num_pbes += 1;
+ pbe++;
+ total_num_pbes++;
+
+ /* If the given pbl is full storing the pbes,
+ * move to next pbl.
+ */
+ if (num_pbes == (mr->info.pbl_info.pbl_size / sizeof(u64))) {
+ pbl_table++;
+ pbe = (struct regpair *)pbl_table->va;
+ num_pbes = 0;
+ }
+ }
+
+ qp->wqe_wr_id[qp->sq.prod].mr = mr;
+
+ return 0;
+}
+
+enum ib_wc_opcode qedr_ib_to_wc_opcode(enum ib_wr_opcode opcode)
+{
+ switch (opcode) {
+ case IB_WR_RDMA_WRITE:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ return IB_WC_RDMA_WRITE;
+ case IB_WR_SEND_WITH_IMM:
+ case IB_WR_SEND:
+ case IB_WR_SEND_WITH_INV:
+ return IB_WC_SEND;
+ case IB_WR_RDMA_READ:
+ return IB_WC_RDMA_READ;
+ case IB_WR_ATOMIC_CMP_AND_SWP:
+ return IB_WC_COMP_SWAP;
+ case IB_WR_ATOMIC_FETCH_AND_ADD:
+ return IB_WC_FETCH_ADD;
+ case IB_WR_REG_MR:
+ return IB_WC_REG_MR;
+ case IB_WR_LOCAL_INV:
+ return IB_WC_LOCAL_INV;
+ default:
+ return IB_WC_SEND;
+ }
+}
+
+inline bool qedr_can_post_send(struct qedr_qp *qp, struct ib_send_wr *wr)
+{
+ int wq_is_full, err_wr, pbl_is_full;
+ struct qedr_dev *dev = qp->dev;
+
+ /* prevent SQ overflow and/or processing of a bad WR */
+ err_wr = wr->num_sge > qp->sq.max_sges;
+ wq_is_full = qedr_wq_is_full(&qp->sq);
+ pbl_is_full = qed_chain_get_elem_left_u32(&qp->sq.pbl) <
+ QEDR_MAX_SQE_ELEMENTS_PER_SQE;
+ if (wq_is_full || err_wr || pbl_is_full) {
+ if (wq_is_full && !(qp->err_bitmap & QEDR_QP_ERR_SQ_FULL)) {
+ DP_ERR(dev,
+ "error: WQ is full. Post send on QP %p failed (this error appears only once)\n",
+ qp);
+ qp->err_bitmap |= QEDR_QP_ERR_SQ_FULL;
+ }
+
+ if (err_wr && !(qp->err_bitmap & QEDR_QP_ERR_BAD_SR)) {
+ DP_ERR(dev,
+ "error: WR is bad. Post send on QP %p failed (this error appears only once)\n",
+ qp);
+ qp->err_bitmap |= QEDR_QP_ERR_BAD_SR;
+ }
+
+ if (pbl_is_full &&
+ !(qp->err_bitmap & QEDR_QP_ERR_SQ_PBL_FULL)) {
+ DP_ERR(dev,
+ "error: WQ PBL is full. Post send on QP %p failed (this error appears only once)\n",
+ qp);
+ qp->err_bitmap |= QEDR_QP_ERR_SQ_PBL_FULL;
+ }
+ return false;
+ }
+ return true;
+}
+
+int __qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibqp->device);
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct rdma_sq_atomic_wqe_1st *awqe1;
+ struct rdma_sq_atomic_wqe_2nd *awqe2;
+ struct rdma_sq_atomic_wqe_3rd *awqe3;
+ struct rdma_sq_send_wqe_2st *swqe2;
+ struct rdma_sq_local_inv_wqe *iwqe;
+ struct rdma_sq_rdma_wqe_2nd *rwqe2;
+ struct rdma_sq_send_wqe_1st *swqe;
+ struct rdma_sq_rdma_wqe_1st *rwqe;
+ struct rdma_sq_fmr_wqe_1st *fwqe1;
+ struct rdma_sq_common_wqe *wqe;
+ u32 length;
+ int rc = 0;
+ bool comp;
+
+ if (!qedr_can_post_send(qp, wr)) {
+ *bad_wr = wr;
+ return -ENOMEM;
+ }
+
+ wqe = qed_chain_produce(&qp->sq.pbl);
+ qp->wqe_wr_id[qp->sq.prod].signaled =
+ !!(wr->send_flags & IB_SEND_SIGNALED) || qp->signaled;
+
+ wqe->flags = 0;
+ SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_SE_FLG,
+ !!(wr->send_flags & IB_SEND_SOLICITED));
+ comp = (!!(wr->send_flags & IB_SEND_SIGNALED)) || qp->signaled;
+ SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_COMP_FLG, comp);
+ SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_RD_FENCE_FLG,
+ !!(wr->send_flags & IB_SEND_FENCE));
+ wqe->prev_wqe_size = qp->prev_wqe_size;
+
+ qp->wqe_wr_id[qp->sq.prod].opcode = qedr_ib_to_wc_opcode(wr->opcode);
+
+ switch (wr->opcode) {
+ case IB_WR_SEND_WITH_IMM:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_IMM;
+ swqe = (struct rdma_sq_send_wqe_1st *)wqe;
+ swqe->wqe_size = 2;
+ swqe2 = qed_chain_produce(&qp->sq.pbl);
+
+ swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.imm_data);
+ length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2,
+ wr, bad_wr);
+ swqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
+ qp->prev_wqe_size = swqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
+ break;
+ case IB_WR_SEND:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_SEND;
+ swqe = (struct rdma_sq_send_wqe_1st *)wqe;
+
+ swqe->wqe_size = 2;
+ swqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2,
+ wr, bad_wr);
+ swqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
+ qp->prev_wqe_size = swqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
+ break;
+ case IB_WR_SEND_WITH_INV:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE;
+ swqe = (struct rdma_sq_send_wqe_1st *)wqe;
+ swqe2 = qed_chain_produce(&qp->sq.pbl);
+ swqe->wqe_size = 2;
+ swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.invalidate_rkey);
+ length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2,
+ wr, bad_wr);
+ swqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
+ qp->prev_wqe_size = swqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
+ break;
+
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM;
+ rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
+
+ rwqe->wqe_size = 2;
+ rwqe->imm_data = htonl(cpu_to_le32(wr->ex.imm_data));
+ rwqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2,
+ wr, bad_wr);
+ rwqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
+ qp->prev_wqe_size = rwqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
+ break;
+ case IB_WR_RDMA_WRITE:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR;
+ rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
+
+ rwqe->wqe_size = 2;
+ rwqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2,
+ wr, bad_wr);
+ rwqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
+ qp->prev_wqe_size = rwqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
+ break;
+ case IB_WR_RDMA_READ_WITH_INV:
+ DP_ERR(dev,
+ "RDMA READ WITH INVALIDATE not supported\n");
+ *bad_wr = wr;
+ rc = -EINVAL;
+ break;
+
+ case IB_WR_RDMA_READ:
+ wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_RD;
+ rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
+
+ rwqe->wqe_size = 2;
+ rwqe2 = qed_chain_produce(&qp->sq.pbl);
+ length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2,
+ wr, bad_wr);
+ rwqe->length = cpu_to_le32(length);
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
+ qp->prev_wqe_size = rwqe->wqe_size;
+ qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
+ break;
+
+ case IB_WR_ATOMIC_CMP_AND_SWP:
+ case IB_WR_ATOMIC_FETCH_AND_ADD:
+ awqe1 = (struct rdma_sq_atomic_wqe_1st *)wqe;
+ awqe1->wqe_size = 4;
+
+ awqe2 = qed_chain_produce(&qp->sq.pbl);
+ DMA_REGPAIR_LE(awqe2->remote_va, atomic_wr(wr)->remote_addr);
+ awqe2->r_key = cpu_to_le32(atomic_wr(wr)->rkey);
+
+ awqe3 = qed_chain_produce(&qp->sq.pbl);
+
+ if (wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
+ wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_ADD;
+ DMA_REGPAIR_LE(awqe3->swap_data,
+ atomic_wr(wr)->compare_add);
+ } else {
+ wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP;
+ DMA_REGPAIR_LE(awqe3->swap_data,
+ atomic_wr(wr)->swap);
+ DMA_REGPAIR_LE(awqe3->cmp_data,
+ atomic_wr(wr)->compare_add);
+ }
+
+ qedr_prepare_sq_sges(qp, NULL, wr);
+
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = awqe1->wqe_size;
+ qp->prev_wqe_size = awqe1->wqe_size;
+ break;
+
+ case IB_WR_LOCAL_INV:
+ iwqe = (struct rdma_sq_local_inv_wqe *)wqe;
+ iwqe->wqe_size = 1;
+
+ iwqe->req_type = RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE;
+ iwqe->inv_l_key = wr->ex.invalidate_rkey;
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = iwqe->wqe_size;
+ qp->prev_wqe_size = iwqe->wqe_size;
+ break;
+ case IB_WR_REG_MR:
+ DP_VERBOSE(dev, QEDR_MSG_CQ, "REG_MR\n");
+ wqe->req_type = RDMA_SQ_REQ_TYPE_FAST_MR;
+ fwqe1 = (struct rdma_sq_fmr_wqe_1st *)wqe;
+ fwqe1->wqe_size = 2;
+
+ rc = qedr_prepare_reg(qp, fwqe1, reg_wr(wr));
+ if (rc) {
+ DP_ERR(dev, "IB_REG_MR failed rc=%d\n", rc);
+ *bad_wr = wr;
+ break;
+ }
+
+ qp->wqe_wr_id[qp->sq.prod].wqe_size = fwqe1->wqe_size;
+ qp->prev_wqe_size = fwqe1->wqe_size;
+ break;
+ default:
+ DP_ERR(dev, "invalid opcode 0x%x!\n", wr->opcode);
+ rc = -EINVAL;
+ *bad_wr = wr;
+ break;
+ }
+
+ if (*bad_wr) {
+ u16 value;
+
+ /* Restore prod to its position before
+ * this WR was processed
+ */
+ value = le16_to_cpu(qp->sq.db_data.data.value);
+ qed_chain_set_prod(&qp->sq.pbl, value, wqe);
+
+ /* Restore prev_wqe_size */
+ qp->prev_wqe_size = wqe->prev_wqe_size;
+ rc = -EINVAL;
+ DP_ERR(dev, "POST SEND FAILED\n");
+ }
+
+ return rc;
+}
+
+int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibqp->device);
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ unsigned long flags;
+ int rc = 0;
+
+ *bad_wr = NULL;
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
+ (qp->state == QED_ROCE_QP_STATE_ERR)) {
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ *bad_wr = wr;
+ DP_VERBOSE(dev, QEDR_MSG_CQ,
+ "QP in wrong state! QP icid=0x%x state %d\n",
+ qp->icid, qp->state);
+ return -EINVAL;
+ }
+
+ if (!wr) {
+ DP_ERR(dev, "Got an empty post send.\n");
+ return -EINVAL;
+ }
+
+ while (wr) {
+ rc = __qedr_post_send(ibqp, wr, bad_wr);
+ if (rc)
+ break;
+
+ qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id;
+
+ qedr_inc_sw_prod(&qp->sq);
+
+ qp->sq.db_data.data.value++;
+
+ wr = wr->next;
+ }
+
+ /* Trigger doorbell
+ * If there was a failure in the first WR then it will be triggered in
+ * vane. However this is not harmful (as long as the producer value is
+ * unchanged). For performance reasons we avoid checking for this
+ * redundant doorbell.
+ */
+ wmb();
+ writel(qp->sq.db_data.raw, qp->sq.db);
+
+ /* Make sure write sticks */
+ mmiowb();
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ return rc;
+}
+
+int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr)
+{
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qedr_dev *dev = qp->dev;
+ unsigned long flags;
+ int status = 0;
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
+ (qp->state == QED_ROCE_QP_STATE_ERR)) {
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ *bad_wr = wr;
+ return -EINVAL;
+ }
+
+ while (wr) {
+ int i;
+
+ if (qed_chain_get_elem_left_u32(&qp->rq.pbl) <
+ QEDR_MAX_RQE_ELEMENTS_PER_RQE ||
+ wr->num_sge > qp->rq.max_sges) {
+ DP_ERR(dev, "Can't post WR (%d < %d) || (%d > %d)\n",
+ qed_chain_get_elem_left_u32(&qp->rq.pbl),
+ QEDR_MAX_RQE_ELEMENTS_PER_RQE, wr->num_sge,
+ qp->rq.max_sges);
+ status = -ENOMEM;
+ *bad_wr = wr;
+ break;
+ }
+ for (i = 0; i < wr->num_sge; i++) {
+ u32 flags = 0;
+ struct rdma_rq_sge *rqe =
+ qed_chain_produce(&qp->rq.pbl);
+
+ /* First one must include the number
+ * of SGE in the list
+ */
+ if (!i)
+ SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES,
+ wr->num_sge);
+
+ SET_FIELD(flags, RDMA_RQ_SGE_L_KEY,
+ wr->sg_list[i].lkey);
+
+ RQ_SGE_SET(rqe, wr->sg_list[i].addr,
+ wr->sg_list[i].length, flags);
+ }
+
+ /* Special case of no sges. FW requires between 1-4 sges...
+ * in this case we need to post 1 sge with length zero. this is
+ * because rdma write with immediate consumes an RQ.
+ */
+ if (!wr->num_sge) {
+ u32 flags = 0;
+ struct rdma_rq_sge *rqe =
+ qed_chain_produce(&qp->rq.pbl);
+
+ /* First one must include the number
+ * of SGE in the list
+ */
+ SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, 0);
+ SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, 1);
+
+ RQ_SGE_SET(rqe, 0, 0, flags);
+ i = 1;
+ }
+
+ qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id;
+ qp->rqe_wr_id[qp->rq.prod].wqe_size = i;
+
+ qedr_inc_sw_prod(&qp->rq);
+
+ /* Flush all the writes before signalling doorbell */
+ wmb();
+
+ qp->rq.db_data.data.value++;
+
+ writel(qp->rq.db_data.raw, qp->rq.db);
+
+ /* Make sure write sticks */
+ mmiowb();
+
+ wr = wr->next;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ return status;
+}
+
+static int is_valid_cqe(struct qedr_cq *cq, union rdma_cqe *cqe)
+{
+ struct rdma_cqe_requester *resp_cqe = &cqe->req;
+
+ return (resp_cqe->flags & RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK) ==
+ cq->pbl_toggle;
+}
+
+static struct qedr_qp *cqe_get_qp(union rdma_cqe *cqe)
+{
+ struct rdma_cqe_requester *resp_cqe = &cqe->req;
+ struct qedr_qp *qp;
+
+ qp = (struct qedr_qp *)(uintptr_t)HILO_GEN(resp_cqe->qp_handle.hi,
+ resp_cqe->qp_handle.lo,
+ u64);
+ return qp;
+}
+
+static enum rdma_cqe_type cqe_get_type(union rdma_cqe *cqe)
+{
+ struct rdma_cqe_requester *resp_cqe = &cqe->req;
+
+ return GET_FIELD(resp_cqe->flags, RDMA_CQE_REQUESTER_TYPE);
+}
+
+/* Return latest CQE (needs processing) */
+static union rdma_cqe *get_cqe(struct qedr_cq *cq)
+{
+ return cq->latest_cqe;
+}
+
+/* In fmr we need to increase the number of fmr completed counter for the fmr
+ * algorithm determining whether we can free a pbl or not.
+ * we need to perform this whether the work request was signaled or not. for
+ * this purpose we call this function from the condition that checks if a wr
+ * should be skipped, to make sure we don't miss it ( possibly this fmr
+ * operation was not signalted)
+ */
+static inline void qedr_chk_if_fmr(struct qedr_qp *qp)
+{
+ if (qp->wqe_wr_id[qp->sq.cons].opcode == IB_WC_REG_MR)
+ qp->wqe_wr_id[qp->sq.cons].mr->info.completed++;
+}
+
+static int process_req(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, int num_entries,
+ struct ib_wc *wc, u16 hw_cons, enum ib_wc_status status,
+ int force)
+{
+ u16 cnt = 0;
+
+ while (num_entries && qp->sq.wqe_cons != hw_cons) {
+ if (!qp->wqe_wr_id[qp->sq.cons].signaled && !force) {
+ qedr_chk_if_fmr(qp);
+ /* skip WC */
+ goto next_cqe;
+ }
+
+ /* fill WC */
+ wc->status = status;
+ wc->wc_flags = 0;
+ wc->src_qp = qp->id;
+ wc->qp = &qp->ibqp;
+
+ wc->wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id;
+ wc->opcode = qp->wqe_wr_id[qp->sq.cons].opcode;
+
+ switch (wc->opcode) {
+ case IB_WC_RDMA_WRITE:
+ wc->byte_len = qp->wqe_wr_id[qp->sq.cons].bytes_len;
+ break;
+ case IB_WC_COMP_SWAP:
+ case IB_WC_FETCH_ADD:
+ wc->byte_len = 8;
+ break;
+ case IB_WC_REG_MR:
+ qp->wqe_wr_id[qp->sq.cons].mr->info.completed++;
+ break;
+ default:
+ break;
+ }
+
+ num_entries--;
+ wc++;
+ cnt++;
+next_cqe:
+ while (qp->wqe_wr_id[qp->sq.cons].wqe_size--)
+ qed_chain_consume(&qp->sq.pbl);
+ qedr_inc_sw_cons(&qp->sq);
+ }
+
+ return cnt;
+}
+
+static int qedr_poll_cq_req(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct qedr_cq *cq,
+ int num_entries, struct ib_wc *wc,
+ struct rdma_cqe_requester *req)
+{
+ int cnt = 0;
+
+ switch (req->status) {
+ case RDMA_CQE_REQ_STS_OK:
+ cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons,
+ IB_WC_SUCCESS, 0);
+ break;
+ case RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons,
+ IB_WC_WR_FLUSH_ERR, 0);
+ break;
+ default:
+ /* process all WQE before the cosumer */
+ qp->state = QED_ROCE_QP_STATE_ERR;
+ cnt = process_req(dev, qp, cq, num_entries, wc,
+ req->sq_cons - 1, IB_WC_SUCCESS, 0);
+ wc += cnt;
+ /* if we have extra WC fill it with actual error info */
+ if (cnt < num_entries) {
+ enum ib_wc_status wc_status;
+
+ switch (req->status) {
+ case RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_BAD_RESP_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_LOC_LEN_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_LOC_QP_OP_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_LOC_PROT_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_MW_BIND_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_REM_INV_REQ_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_REM_ACCESS_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_REM_OP_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_RNR_RETRY_EXC_ERR;
+ break;
+ case RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR:
+ DP_ERR(dev,
+ "Error: POLL CQ with ROCE_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_RETRY_EXC_ERR;
+ break;
+ default:
+ DP_ERR(dev,
+ "Error: POLL CQ with IB_WC_GENERAL_ERR. CQ icid=0x%x, QP icid=0x%x\n",
+ cq->icid, qp->icid);
+ wc_status = IB_WC_GENERAL_ERR;
+ }
+ cnt += process_req(dev, qp, cq, 1, wc, req->sq_cons,
+ wc_status, 1);
+ }
+ }
+
+ return cnt;
+}
+
+static void __process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, struct ib_wc *wc,
+ struct rdma_cqe_responder *resp, u64 wr_id)
+{
+ enum ib_wc_status wc_status = IB_WC_SUCCESS;
+ u8 flags;
+
+ wc->opcode = IB_WC_RECV;
+ wc->wc_flags = 0;
+
+ switch (resp->status) {
+ case RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR:
+ wc_status = IB_WC_LOC_ACCESS_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR:
+ wc_status = IB_WC_LOC_LEN_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR:
+ wc_status = IB_WC_LOC_QP_OP_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR:
+ wc_status = IB_WC_LOC_PROT_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR:
+ wc_status = IB_WC_MW_BIND_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR:
+ wc_status = IB_WC_REM_INV_RD_REQ_ERR;
+ break;
+ case RDMA_CQE_RESP_STS_OK:
+ wc_status = IB_WC_SUCCESS;
+ wc->byte_len = le32_to_cpu(resp->length);
+
+ flags = resp->flags & QEDR_RESP_RDMA_IMM;
+
+ switch (flags) {
+ case QEDR_RESP_RDMA_IMM:
+ /* update opcode */
+ wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+ /* fall to set imm data */
+ case QEDR_RESP_IMM:
+ wc->ex.imm_data =
+ le32_to_cpu(resp->imm_data_or_inv_r_Key);
+ wc->wc_flags |= IB_WC_WITH_IMM;
+ break;
+ case QEDR_RESP_RDMA:
+ DP_ERR(dev, "Invalid flags detected\n");
+ break;
+ default:
+ DP_ERR(dev, "Invalid opcode\n");
+ }
+
+ break;
+ default:
+ wc->status = IB_WC_GENERAL_ERR;
+ DP_ERR(dev, "Invalid CQE status detected\n");
+ }
+
+ /* fill WC */
+ wc->status = wc_status;
+ wc->src_qp = qp->id;
+ wc->qp = &qp->ibqp;
+ wc->wr_id = wr_id;
+}
+
+static int process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, struct ib_wc *wc,
+ struct rdma_cqe_responder *resp)
+{
+ u64 wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
+
+ __process_resp_one(dev, qp, cq, wc, resp, wr_id);
+
+ while (qp->rqe_wr_id[qp->rq.cons].wqe_size--)
+ qed_chain_consume(&qp->rq.pbl);
+ qedr_inc_sw_cons(&qp->rq);
+
+ return 1;
+}
+
+static int process_resp_flush(struct qedr_qp *qp, struct qedr_cq *cq,
+ int num_entries, struct ib_wc *wc, u16 hw_cons)
+{
+ u16 cnt = 0;
+
+ while (num_entries && qp->rq.wqe_cons != hw_cons) {
+ /* fill WC */
+ wc->status = IB_WC_WR_FLUSH_ERR;
+ wc->wc_flags = 0;
+ wc->src_qp = qp->id;
+ wc->byte_len = 0;
+ wc->wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
+ wc->qp = &qp->ibqp;
+ num_entries--;
+ wc++;
+ cnt++;
+ while (qp->rqe_wr_id[qp->rq.cons].wqe_size--)
+ qed_chain_consume(&qp->rq.pbl);
+ qedr_inc_sw_cons(&qp->rq);
+ }
+
+ return cnt;
+}
+
+static void try_consume_resp_cqe(struct qedr_cq *cq, struct qedr_qp *qp,
+ struct rdma_cqe_responder *resp, int *update)
+{
+ if (le16_to_cpu(resp->rq_cons) == qp->rq.wqe_cons) {
+ consume_cqe(cq);
+ *update |= 1;
+ }
+}
+
+static int qedr_poll_cq_resp(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct qedr_cq *cq, int num_entries,
+ struct ib_wc *wc, struct rdma_cqe_responder *resp,
+ int *update)
+{
+ int cnt;
+
+ if (resp->status == RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR) {
+ cnt = process_resp_flush(qp, cq, num_entries, wc,
+ resp->rq_cons);
+ try_consume_resp_cqe(cq, qp, resp, update);
+ } else {
+ cnt = process_resp_one(dev, qp, cq, wc, resp);
+ consume_cqe(cq);
+ *update |= 1;
+ }
+
+ return cnt;
+}
+
+static void try_consume_req_cqe(struct qedr_cq *cq, struct qedr_qp *qp,
+ struct rdma_cqe_requester *req, int *update)
+{
+ if (le16_to_cpu(req->sq_cons) == qp->sq.wqe_cons) {
+ consume_cqe(cq);
+ *update |= 1;
+ }
+}
+
+int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+ union rdma_cqe *cqe = cq->latest_cqe;
+ u32 old_cons, new_cons;
+ unsigned long flags;
+ int update = 0;
+ int done = 0;
+
+ spin_lock_irqsave(&cq->cq_lock, flags);
+ old_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
+ while (num_entries && is_valid_cqe(cq, cqe)) {
+ struct qedr_qp *qp;
+ int cnt = 0;
+
+ /* prevent speculative reads of any field of CQE */
+ rmb();
+
+ qp = cqe_get_qp(cqe);
+ if (!qp) {
+ WARN(1, "Error: CQE QP pointer is NULL. CQE=%p\n", cqe);
+ break;
+ }
+
+ wc->qp = &qp->ibqp;
+
+ switch (cqe_get_type(cqe)) {
+ case RDMA_CQE_TYPE_REQUESTER:
+ cnt = qedr_poll_cq_req(dev, qp, cq, num_entries, wc,
+ &cqe->req);
+ try_consume_req_cqe(cq, qp, &cqe->req, &update);
+ break;
+ case RDMA_CQE_TYPE_RESPONDER_RQ:
+ cnt = qedr_poll_cq_resp(dev, qp, cq, num_entries, wc,
+ &cqe->resp, &update);
+ break;
+ case RDMA_CQE_TYPE_INVALID:
+ default:
+ DP_ERR(dev, "Error: invalid CQE type = %d\n",
+ cqe_get_type(cqe));
+ }
+ num_entries -= cnt;
+ wc += cnt;
+ done += cnt;
+
+ cqe = get_cqe(cq);
+ }
+ new_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
+
+ cq->cq_cons += new_cons - old_cons;
+
+ if (update)
+ /* doorbell notifies abount latest VALID entry,
+ * but chain already point to the next INVALID one
+ */
+ doorbell_cq(cq, cq->cq_cons - 1, cq->arm_flags);
+
+ spin_unlock_irqrestore(&cq->cq_lock, flags);
+ return done;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index e7cb3fd..9eb2722b 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -58,4 +58,9 @@ int qedr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
struct ib_mr *qedr_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
u32 max_num_sg);
+int qedr_poll_cq(struct ib_cq *, int num_entries, struct ib_wc *wc);
+int qedr_post_send(struct ib_qp *, struct ib_send_wr *,
+ struct ib_send_wr **bad_wr);
+int qedr_post_recv(struct ib_qp *, struct ib_recv_wr *,
+ struct ib_recv_wr **bad_wr);
#endif
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 09/11] Add LL2 RoCE interface
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add light L2 interface for RoCE.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 34 +++-
drivers/infiniband/hw/qedr/qedr.h | 2 +-
drivers/net/ethernet/qlogic/qed/qed.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_hsi.h | 3 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 21 ++-
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 241 +++++++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 1 +
include/linux/qed/qed_roce_if.h | 79 ++++++++++
9 files changed, 385 insertions(+), 4 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 42206ed..fb0b76c 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -727,6 +727,38 @@ static void qedr_shutdown(struct qedr_dev *dev)
qedr_remove(dev);
}
+static void qedr_mac_address_change(struct qedr_dev *dev)
+{
+ union ib_gid *sgid = &dev->sgid_tbl[0];
+ u8 guid[8], mac_addr[6];
+ int rc;
+
+ /* Update SGID */
+ ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
+ guid[0] = mac_addr[0] ^ 2;
+ guid[1] = mac_addr[1];
+ guid[2] = mac_addr[2];
+ guid[3] = 0xff;
+ guid[4] = 0xfe;
+ guid[5] = mac_addr[3];
+ guid[6] = mac_addr[4];
+ guid[7] = mac_addr[5];
+ sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
+ memcpy(&sgid->raw[8], guid, sizeof(guid));
+
+ /* Update LL2 */
+ rc = dev->ops->roce_ll2_set_mac_filter(dev->cdev,
+ dev->gsi_ll2_mac_address,
+ dev->ndev->dev_addr);
+
+ ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
+
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+
+ if (rc)
+ DP_ERR(dev, "Error updating MAC filter\n");
+}
+
/* event handling via NIC driver ensures that all the NIC specific
* initialization done before RoCE driver notifies
* event to stack.
@@ -744,7 +776,7 @@ static void qedr_notify(struct qedr_dev *dev, enum qede_roce_event event)
qedr_shutdown(dev);
break;
case QEDE_CHANGE_ADDR:
- qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+ qedr_mac_address_change(dev);
break;
default:
pr_err("Event not supported\n");
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index bc218d2..7ae0180 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -146,7 +146,7 @@ struct qedr_dev {
u8 dp_level;
u8 num_hwfns;
uint wq_multiplier;
-
+ u8 gsi_ll2_mac_address[ETH_ALEN];
};
#define QEDR_MAX_SQ_PBL (0x8000)
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index 9640475..87a544f 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -426,6 +426,7 @@ struct qed_hwfn {
#endif
struct z_stream_s *stream;
+ struct qed_roce_ll2_info *ll2;
};
struct pci_params {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_hsi.h b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
index a67b355..2b4dfc6 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_hsi.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
@@ -727,6 +727,9 @@ struct core_tx_bd_flags {
#define CORE_TX_BD_FLAGS_L4_PROTOCOL_SHIFT 6
#define CORE_TX_BD_FLAGS_L4_PSEUDO_CSUM_MODE_MASK 0x1
#define CORE_TX_BD_FLAGS_L4_PSEUDO_CSUM_MODE_SHIFT 7
+#define CORE_TX_BD_FLAGS_ROCE_FLAV_MASK 0x1
+#define CORE_TX_BD_FLAGS_ROCE_FLAV_SHIFT 12
+
};
struct core_tx_bd {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index 1a96a2d..4ea6a07 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -301,7 +301,8 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
p_pkt->cookie,
p_pkt->bds_set[0].tx_frag,
- b_last_frag, b_last_packet);
+ b_last_frag,
+ b_last_packet);
}
}
@@ -1025,6 +1026,7 @@ static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum core_roce_flavor_type type,
dma_addr_t first_frag,
u16 first_frag_len)
{
@@ -1045,6 +1047,9 @@ static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
DMA_REGPAIR_LE(start_bd->addr, first_frag);
start_bd->nbytes = cpu_to_le16(first_frag_len);
+ SET_FIELD(start_bd->bd_flags.as_bitfield, CORE_TX_BD_FLAGS_ROCE_FLAV,
+ type);
+
DP_VERBOSE(p_hwfn,
(NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
"LL2 [q 0x%02x cid 0x%08x type 0x%08x] Tx Producer at [0x%04x] - set with a %04x bytes %02x BDs buffer at %08x:%08x\n",
@@ -1136,11 +1141,13 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum qed_ll2_roce_flavor_type qed_roce_flavor,
dma_addr_t first_frag,
u16 first_frag_len, void *cookie, u8 notify_fw)
{
struct qed_ll2_tx_packet *p_curp = NULL;
struct qed_ll2_info *p_ll2_conn = NULL;
+ enum core_roce_flavor_type roce_flavor;
struct qed_ll2_tx_queue *p_tx;
struct qed_chain *p_tx_chain;
unsigned long flags;
@@ -1173,6 +1180,15 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
goto out;
}
+ if (qed_roce_flavor == QED_LL2_ROCE) {
+ roce_flavor = CORE_ROCE;
+ } else if (qed_roce_flavor == QED_LL2_RROCE) {
+ roce_flavor = CORE_RROCE;
+ } else {
+ rc = -EINVAL;
+ goto out;
+ }
+
/* Prepare packet and BD, and perhaps send a doorbell to FW */
qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp,
num_of_bds, first_frag,
@@ -1180,6 +1196,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp,
num_of_bds, CORE_TX_DEST_NW,
vlan, bd_flags, l4_hdr_offset_w,
+ roce_flavor,
first_frag, first_frag_len);
qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
@@ -1624,7 +1641,7 @@ static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
cdev->ll2->handle,
1 + skb_shinfo(skb)->nr_frags,
- vlan, flags, 0, mapping,
+ vlan, flags, 0, mapping, 0,
skb->len, skb, 1);
if (rc)
goto err;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index a037c48..b3783fd 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -24,6 +24,12 @@
#define QED_MAX_NUM_OF_LL2_CONNECTIONS (4)
+enum qed_ll2_roce_flavor_type {
+ QED_LL2_ROCE,
+ QED_LL2_RROCE,
+ MAX_QED_LL2_ROCE_FLAVOR_TYPE
+};
+
enum qed_ll2_conn_type {
QED_LL2_TYPE_RESERVED,
QED_LL2_TYPE_ISCSI,
@@ -199,6 +205,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum qed_ll2_roce_flavor_type qed_roce_flavor,
dma_addr_t first_frag,
u16 first_frag_len, void *cookie, u8 notify_fw);
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 04e33ee..27095f0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -64,6 +64,7 @@
#include "qed_reg_addr.h"
#include "qed_sp.h"
#include "qed_roce.h"
+#include "qed_ll2.h"
void qed_async_roce_event(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe)
@@ -2632,6 +2633,240 @@ void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
}
+static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
+ u8 *old_mac_address,
+ u8 *new_mac_address)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_ptt *p_ptt;
+ int rc = 0;
+
+ if (!hwfn->ll2 || hwfn->ll2->handle == QED_LL2_UNUSED_HANDLE) {
+ DP_ERR(cdev,
+ "QED RoCE set MAC filter failed - roce_info/ll2 NULL\n");
+ return -EINVAL;
+ }
+
+ p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ if (!p_ptt) {
+ DP_ERR(cdev,
+ "qed roce ll2 mac filter set: failed to acquire PTT\n");
+ return -EINVAL;
+ }
+
+ mutex_lock(&hwfn->ll2->lock);
+ if (old_mac_address)
+ qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ old_mac_address);
+ if (new_mac_address)
+ rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ new_mac_address);
+ mutex_unlock(&hwfn->ll2->lock);
+
+ qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+
+ if (rc)
+ DP_ERR(cdev,
+ "qed roce ll2 mac filter set: failed to add MAC filter\n");
+
+ return rc;
+}
+
+static int qed_roce_ll2_start(struct qed_dev *cdev,
+ struct qed_roce_ll2_params *params)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2;
+ struct qed_ll2_info ll2_params;
+ int rc;
+
+ if (!params) {
+ DP_ERR(cdev, "qed roce ll2 start: failed due to NULL params\n");
+ return -EINVAL;
+ }
+ if (!params->cbs.tx_cb || !params->cbs.rx_cb) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed due to NULL tx/rx. tx_cb=%p, rx_cb=%p\n",
+ params->cbs.tx_cb, params->cbs.rx_cb);
+ return -EINVAL;
+ }
+ if (!is_valid_ether_addr(params->mac_address)) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed due to invalid Ethernet address %pM\n",
+ params->mac_address);
+ return -EINVAL;
+ }
+
+ /* Initialize */
+ roce_ll2 = kzalloc(sizeof(*roce_ll2), GFP_ATOMIC);
+ if (!roce_ll2) {
+ DP_ERR(cdev, "qed roce ll2 start: failed memory allocation\n");
+ return -ENOMEM;
+ }
+ memset(roce_ll2, 0, sizeof(*roce_ll2));
+ roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
+ roce_ll2->cbs = params->cbs;
+ roce_ll2->cb_cookie = params->cb_cookie;
+ mutex_init(&roce_ll2->lock);
+
+ memset(&ll2_params, 0, sizeof(ll2_params));
+ ll2_params.conn_type = QED_LL2_TYPE_ROCE;
+ ll2_params.mtu = params->mtu;
+ ll2_params.rx_drop_ttl0_flg = true;
+ ll2_params.rx_vlan_removal_en = false;
+ ll2_params.tx_dest = CORE_TX_DEST_NW;
+ ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
+ ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
+
+ rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
+ params->max_rx_buffers,
+ params->max_tx_buffers,
+ &roce_ll2->handle);
+ if (rc) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed to acquire LL2 connection (rc=%d)\n",
+ rc);
+ goto err;
+ }
+
+ rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle);
+ if (rc) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed to establish LL2 connection (rc=%d)\n",
+ rc);
+ goto err1;
+ }
+
+ hwfn->ll2 = roce_ll2;
+
+ rc = qed_roce_ll2_set_mac_filter(cdev, NULL, params->mac_address);
+ if (rc) {
+ hwfn->ll2 = NULL;
+ goto err2;
+ }
+ ether_addr_copy(roce_ll2->mac_address, params->mac_address);
+
+ return 0;
+
+err2:
+ qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+err1:
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+err:
+ kfree(roce_ll2);
+ return rc;
+}
+
+static int qed_roce_ll2_stop(struct qed_dev *cdev)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+ int rc;
+
+ if (!cdev) {
+ DP_ERR(cdev, "qed roce ll2 stop: invalid cdev\n");
+ return -EINVAL;
+ }
+
+ if (roce_ll2->handle == QED_LL2_UNUSED_HANDLE) {
+ DP_ERR(cdev, "qed roce ll2 stop: cannot stop an unused LL2\n");
+ return -EINVAL;
+ }
+
+ /* remove LL2 MAC address filter */
+ rc = qed_roce_ll2_set_mac_filter(cdev, roce_ll2->mac_address, NULL);
+ eth_zero_addr(roce_ll2->mac_address);
+
+ rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle);
+ if (rc)
+ DP_ERR(cdev,
+ "qed roce ll2 stop: failed to terminate LL2 connection (rc=%d)\n",
+ rc);
+
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+
+ roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
+
+ kfree(roce_ll2);
+
+ return rc;
+}
+
+static int qed_roce_ll2_tx(struct qed_dev *cdev,
+ struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_tx_params *params)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+ enum qed_ll2_roce_flavor_type qed_roce_flavor;
+ u8 flags = 0;
+ int rc;
+ int i;
+
+ if (!cdev || !pkt || !params) {
+ DP_ERR(cdev,
+ "roce ll2 tx: failed tx because one of the following is NULL - drv=%p, pkt=%p, params=%p\n",
+ cdev, pkt, params);
+ return -EINVAL;
+ }
+
+ qed_roce_flavor = (pkt->roce_mode == ROCE_V1) ? QED_LL2_ROCE
+ : QED_LL2_RROCE;
+
+ if (pkt->roce_mode == ROCE_V2_IPV4)
+ flags |= BIT(CORE_TX_BD_FLAGS_IP_CSUM_SHIFT);
+
+ /* Tx header */
+ rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev), roce_ll2->handle,
+ 1 + pkt->n_seg, 0, flags, 0,
+ qed_roce_flavor, pkt->header.baddr,
+ pkt->header.len, pkt, 1);
+ if (rc) {
+ DP_ERR(cdev, "roce ll2 tx: header failed (rc=%d)\n", rc);
+ return QED_ROCE_TX_HEAD_FAILURE;
+ }
+
+ /* Tx payload */
+ for (i = 0; i < pkt->n_seg; i++) {
+ rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle,
+ pkt->payload[i].baddr,
+ pkt->payload[i].len);
+ if (rc) {
+ /* If failed not much to do here, partial packet has
+ * been posted * we can't free memory, will need to wait
+ * for completion
+ */
+ DP_ERR(cdev,
+ "roce ll2 tx: payload failed (rc=%d)\n", rc);
+ return QED_ROCE_TX_FRAG_FAILURE;
+ }
+ }
+
+ return 0;
+}
+
+static int qed_roce_ll2_post_rx_buffer(struct qed_dev *cdev,
+ struct qed_roce_ll2_buffer *buf,
+ u64 cookie, u8 notify_fw)
+{
+ return qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
+ QED_LEADING_HWFN(cdev)->ll2->handle,
+ buf->baddr, buf->len,
+ (void *)(uintptr_t)cookie, notify_fw);
+}
+
+static int qed_roce_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+
+ return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle, stats);
+}
+
static const struct qed_rdma_ops qed_rdma_ops_pass = {
.common = &qed_common_ops_pass,
.fill_dev_info = &qed_fill_rdma_dev_info,
@@ -2659,6 +2894,12 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_free_tid = &qed_rdma_free_tid,
.rdma_register_tid = &qed_rdma_register_tid,
.rdma_deregister_tid = &qed_rdma_deregister_tid,
+ .roce_ll2_start = &qed_roce_ll2_start,
+ .roce_ll2_stop = &qed_roce_ll2_stop,
+ .roce_ll2_tx = &qed_roce_ll2_tx,
+ .roce_ll2_post_rx_buffer = &qed_roce_ll2_post_rx_buffer,
+ .roce_ll2_set_mac_filter = &qed_roce_ll2_set_mac_filter,
+ .roce_ll2_stats = &qed_roce_ll2_stats,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 71b37fc..f688a47 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -42,6 +42,7 @@
#include "qed.h"
#include "qed_dev_api.h"
#include "qed_hsi.h"
+#include "qed_ll2.h"
#define QED_RDMA_MAX_FMR (RDMA_MAX_TIDS)
#define QED_RDMA_MAX_P_KEY (1)
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 0b6df6e..53047d3 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -39,6 +39,16 @@
#include <linux/slab.h>
#include <linux/qed/qed_if.h>
#include <linux/qed/qed_ll2_if.h>
+#include <linux/qed/rdma_common.h>
+
+enum qed_roce_ll2_tx_dest {
+ /* Light L2 TX Destination to the Network */
+ QED_ROCE_LL2_TX_DEST_NW,
+
+ /* Light L2 TX Destination to the Loopback */
+ QED_ROCE_LL2_TX_DEST_LB,
+ QED_ROCE_LL2_TX_DEST_MAX
+};
#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
@@ -461,6 +471,61 @@ struct qed_rdma_counters_out_params {
#define QED_ROCE_TX_HEAD_FAILURE (1)
#define QED_ROCE_TX_FRAG_FAILURE (2)
+struct qed_roce_ll2_header {
+ void *vaddr;
+ dma_addr_t baddr;
+ size_t len;
+};
+
+struct qed_roce_ll2_buffer {
+ dma_addr_t baddr;
+ size_t len;
+};
+
+struct qed_roce_ll2_packet {
+ struct qed_roce_ll2_header header;
+ int n_seg;
+ struct qed_roce_ll2_buffer payload[RDMA_MAX_SGE_PER_SQ_WQE];
+ int roce_mode;
+ enum qed_roce_ll2_tx_dest tx_dest;
+};
+
+struct qed_roce_ll2_tx_params {
+ int reserved;
+};
+
+struct qed_roce_ll2_rx_params {
+ u16 vlan_id;
+ u8 smac[ETH_ALEN];
+ int rc;
+};
+
+struct qed_roce_ll2_cbs {
+ void (*tx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt);
+
+ void (*rx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_rx_params *params);
+};
+
+struct qed_roce_ll2_params {
+ u16 max_rx_buffers;
+ u16 max_tx_buffers;
+ u16 mtu;
+ u8 mac_address[ETH_ALEN];
+ struct qed_roce_ll2_cbs cbs;
+ void *cb_cookie;
+};
+
+struct qed_roce_ll2_info {
+ u8 handle;
+ struct qed_roce_ll2_cbs cbs;
+ u8 mac_address[ETH_ALEN];
+ void *cb_cookie;
+
+ /* Lock to protect ll2 */
+ struct mutex lock;
+};
+
enum qed_rdma_type {
QED_RDMA_TYPE_ROCE,
};
@@ -518,6 +583,20 @@ struct qed_rdma_ops {
int (*rdma_deregister_tid)(void *rdma_cxt, u32 itid);
int (*rdma_alloc_tid)(void *rdma_cxt, u32 *itid);
void (*rdma_free_tid)(void *rdma_cxt, u32 itid);
+ int (*roce_ll2_start)(struct qed_dev *cdev,
+ struct qed_roce_ll2_params *params);
+ int (*roce_ll2_stop)(struct qed_dev *cdev);
+ int (*roce_ll2_tx)(struct qed_dev *cdev,
+ struct qed_roce_ll2_packet *packet,
+ struct qed_roce_ll2_tx_params *params);
+ int (*roce_ll2_post_rx_buffer)(struct qed_dev *cdev,
+ struct qed_roce_ll2_buffer *buf,
+ u64 cookie, u8 notify_fw);
+ int (*roce_ll2_set_mac_filter)(struct qed_dev *cdev,
+ u8 *old_mac_address,
+ u8 *new_mac_address);
+ int (*roce_ll2_stats)(struct qed_dev *cdev,
+ struct qed_ll2_stats *stats);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 09/11] Add LL2 RoCE interface
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add light L2 interface for RoCE.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 34 +++-
drivers/infiniband/hw/qedr/qedr.h | 2 +-
drivers/net/ethernet/qlogic/qed/qed.h | 1 +
drivers/net/ethernet/qlogic/qed/qed_hsi.h | 3 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 21 ++-
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 7 +
drivers/net/ethernet/qlogic/qed/qed_roce.c | 241 +++++++++++++++++++++++++++++
drivers/net/ethernet/qlogic/qed/qed_roce.h | 1 +
include/linux/qed/qed_roce_if.h | 79 ++++++++++
9 files changed, 385 insertions(+), 4 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index 42206ed..fb0b76c 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -727,6 +727,38 @@ static void qedr_shutdown(struct qedr_dev *dev)
qedr_remove(dev);
}
+static void qedr_mac_address_change(struct qedr_dev *dev)
+{
+ union ib_gid *sgid = &dev->sgid_tbl[0];
+ u8 guid[8], mac_addr[6];
+ int rc;
+
+ /* Update SGID */
+ ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
+ guid[0] = mac_addr[0] ^ 2;
+ guid[1] = mac_addr[1];
+ guid[2] = mac_addr[2];
+ guid[3] = 0xff;
+ guid[4] = 0xfe;
+ guid[5] = mac_addr[3];
+ guid[6] = mac_addr[4];
+ guid[7] = mac_addr[5];
+ sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
+ memcpy(&sgid->raw[8], guid, sizeof(guid));
+
+ /* Update LL2 */
+ rc = dev->ops->roce_ll2_set_mac_filter(dev->cdev,
+ dev->gsi_ll2_mac_address,
+ dev->ndev->dev_addr);
+
+ ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
+
+ qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+
+ if (rc)
+ DP_ERR(dev, "Error updating MAC filter\n");
+}
+
/* event handling via NIC driver ensures that all the NIC specific
* initialization done before RoCE driver notifies
* event to stack.
@@ -744,7 +776,7 @@ static void qedr_notify(struct qedr_dev *dev, enum qede_roce_event event)
qedr_shutdown(dev);
break;
case QEDE_CHANGE_ADDR:
- qedr_ib_dispatch_event(dev, 1, IB_EVENT_GID_CHANGE);
+ qedr_mac_address_change(dev);
break;
default:
pr_err("Event not supported\n");
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index bc218d2..7ae0180 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -146,7 +146,7 @@ struct qedr_dev {
u8 dp_level;
u8 num_hwfns;
uint wq_multiplier;
-
+ u8 gsi_ll2_mac_address[ETH_ALEN];
};
#define QEDR_MAX_SQ_PBL (0x8000)
diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index 9640475..87a544f 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -426,6 +426,7 @@ struct qed_hwfn {
#endif
struct z_stream_s *stream;
+ struct qed_roce_ll2_info *ll2;
};
struct pci_params {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_hsi.h b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
index a67b355..2b4dfc6 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_hsi.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
@@ -727,6 +727,9 @@ struct core_tx_bd_flags {
#define CORE_TX_BD_FLAGS_L4_PROTOCOL_SHIFT 6
#define CORE_TX_BD_FLAGS_L4_PSEUDO_CSUM_MODE_MASK 0x1
#define CORE_TX_BD_FLAGS_L4_PSEUDO_CSUM_MODE_SHIFT 7
+#define CORE_TX_BD_FLAGS_ROCE_FLAV_MASK 0x1
+#define CORE_TX_BD_FLAGS_ROCE_FLAV_SHIFT 12
+
};
struct core_tx_bd {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index 1a96a2d..4ea6a07 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -301,7 +301,8 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
p_pkt->cookie,
p_pkt->bds_set[0].tx_frag,
- b_last_frag, b_last_packet);
+ b_last_frag,
+ b_last_packet);
}
}
@@ -1025,6 +1026,7 @@ static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum core_roce_flavor_type type,
dma_addr_t first_frag,
u16 first_frag_len)
{
@@ -1045,6 +1047,9 @@ static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
DMA_REGPAIR_LE(start_bd->addr, first_frag);
start_bd->nbytes = cpu_to_le16(first_frag_len);
+ SET_FIELD(start_bd->bd_flags.as_bitfield, CORE_TX_BD_FLAGS_ROCE_FLAV,
+ type);
+
DP_VERBOSE(p_hwfn,
(NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
"LL2 [q 0x%02x cid 0x%08x type 0x%08x] Tx Producer at [0x%04x] - set with a %04x bytes %02x BDs buffer at %08x:%08x\n",
@@ -1136,11 +1141,13 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum qed_ll2_roce_flavor_type qed_roce_flavor,
dma_addr_t first_frag,
u16 first_frag_len, void *cookie, u8 notify_fw)
{
struct qed_ll2_tx_packet *p_curp = NULL;
struct qed_ll2_info *p_ll2_conn = NULL;
+ enum core_roce_flavor_type roce_flavor;
struct qed_ll2_tx_queue *p_tx;
struct qed_chain *p_tx_chain;
unsigned long flags;
@@ -1173,6 +1180,15 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
goto out;
}
+ if (qed_roce_flavor == QED_LL2_ROCE) {
+ roce_flavor = CORE_ROCE;
+ } else if (qed_roce_flavor == QED_LL2_RROCE) {
+ roce_flavor = CORE_RROCE;
+ } else {
+ rc = -EINVAL;
+ goto out;
+ }
+
/* Prepare packet and BD, and perhaps send a doorbell to FW */
qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp,
num_of_bds, first_frag,
@@ -1180,6 +1196,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp,
num_of_bds, CORE_TX_DEST_NW,
vlan, bd_flags, l4_hdr_offset_w,
+ roce_flavor,
first_frag, first_frag_len);
qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
@@ -1624,7 +1641,7 @@ static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
cdev->ll2->handle,
1 + skb_shinfo(skb)->nr_frags,
- vlan, flags, 0, mapping,
+ vlan, flags, 0, mapping, 0,
skb->len, skb, 1);
if (rc)
goto err;
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index a037c48..b3783fd 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -24,6 +24,12 @@
#define QED_MAX_NUM_OF_LL2_CONNECTIONS (4)
+enum qed_ll2_roce_flavor_type {
+ QED_LL2_ROCE,
+ QED_LL2_RROCE,
+ MAX_QED_LL2_ROCE_FLAVOR_TYPE
+};
+
enum qed_ll2_conn_type {
QED_LL2_TYPE_RESERVED,
QED_LL2_TYPE_ISCSI,
@@ -199,6 +205,7 @@ int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
u16 vlan,
u8 bd_flags,
u16 l4_hdr_offset_w,
+ enum qed_ll2_roce_flavor_type qed_roce_flavor,
dma_addr_t first_frag,
u16 first_frag_len, void *cookie, u8 notify_fw);
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 04e33ee..27095f0 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -64,6 +64,7 @@
#include "qed_reg_addr.h"
#include "qed_sp.h"
#include "qed_roce.h"
+#include "qed_ll2.h"
void qed_async_roce_event(struct qed_hwfn *p_hwfn,
struct event_ring_entry *p_eqe)
@@ -2632,6 +2633,240 @@ void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
}
+static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
+ u8 *old_mac_address,
+ u8 *new_mac_address)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_ptt *p_ptt;
+ int rc = 0;
+
+ if (!hwfn->ll2 || hwfn->ll2->handle == QED_LL2_UNUSED_HANDLE) {
+ DP_ERR(cdev,
+ "QED RoCE set MAC filter failed - roce_info/ll2 NULL\n");
+ return -EINVAL;
+ }
+
+ p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+ if (!p_ptt) {
+ DP_ERR(cdev,
+ "qed roce ll2 mac filter set: failed to acquire PTT\n");
+ return -EINVAL;
+ }
+
+ mutex_lock(&hwfn->ll2->lock);
+ if (old_mac_address)
+ qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ old_mac_address);
+ if (new_mac_address)
+ rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+ new_mac_address);
+ mutex_unlock(&hwfn->ll2->lock);
+
+ qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+
+ if (rc)
+ DP_ERR(cdev,
+ "qed roce ll2 mac filter set: failed to add MAC filter\n");
+
+ return rc;
+}
+
+static int qed_roce_ll2_start(struct qed_dev *cdev,
+ struct qed_roce_ll2_params *params)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2;
+ struct qed_ll2_info ll2_params;
+ int rc;
+
+ if (!params) {
+ DP_ERR(cdev, "qed roce ll2 start: failed due to NULL params\n");
+ return -EINVAL;
+ }
+ if (!params->cbs.tx_cb || !params->cbs.rx_cb) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed due to NULL tx/rx. tx_cb=%p, rx_cb=%p\n",
+ params->cbs.tx_cb, params->cbs.rx_cb);
+ return -EINVAL;
+ }
+ if (!is_valid_ether_addr(params->mac_address)) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed due to invalid Ethernet address %pM\n",
+ params->mac_address);
+ return -EINVAL;
+ }
+
+ /* Initialize */
+ roce_ll2 = kzalloc(sizeof(*roce_ll2), GFP_ATOMIC);
+ if (!roce_ll2) {
+ DP_ERR(cdev, "qed roce ll2 start: failed memory allocation\n");
+ return -ENOMEM;
+ }
+ memset(roce_ll2, 0, sizeof(*roce_ll2));
+ roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
+ roce_ll2->cbs = params->cbs;
+ roce_ll2->cb_cookie = params->cb_cookie;
+ mutex_init(&roce_ll2->lock);
+
+ memset(&ll2_params, 0, sizeof(ll2_params));
+ ll2_params.conn_type = QED_LL2_TYPE_ROCE;
+ ll2_params.mtu = params->mtu;
+ ll2_params.rx_drop_ttl0_flg = true;
+ ll2_params.rx_vlan_removal_en = false;
+ ll2_params.tx_dest = CORE_TX_DEST_NW;
+ ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
+ ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
+
+ rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
+ params->max_rx_buffers,
+ params->max_tx_buffers,
+ &roce_ll2->handle);
+ if (rc) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed to acquire LL2 connection (rc=%d)\n",
+ rc);
+ goto err;
+ }
+
+ rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle);
+ if (rc) {
+ DP_ERR(cdev,
+ "qed roce ll2 start: failed to establish LL2 connection (rc=%d)\n",
+ rc);
+ goto err1;
+ }
+
+ hwfn->ll2 = roce_ll2;
+
+ rc = qed_roce_ll2_set_mac_filter(cdev, NULL, params->mac_address);
+ if (rc) {
+ hwfn->ll2 = NULL;
+ goto err2;
+ }
+ ether_addr_copy(roce_ll2->mac_address, params->mac_address);
+
+ return 0;
+
+err2:
+ qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+err1:
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+err:
+ kfree(roce_ll2);
+ return rc;
+}
+
+static int qed_roce_ll2_stop(struct qed_dev *cdev)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+ int rc;
+
+ if (!cdev) {
+ DP_ERR(cdev, "qed roce ll2 stop: invalid cdev\n");
+ return -EINVAL;
+ }
+
+ if (roce_ll2->handle == QED_LL2_UNUSED_HANDLE) {
+ DP_ERR(cdev, "qed roce ll2 stop: cannot stop an unused LL2\n");
+ return -EINVAL;
+ }
+
+ /* remove LL2 MAC address filter */
+ rc = qed_roce_ll2_set_mac_filter(cdev, roce_ll2->mac_address, NULL);
+ eth_zero_addr(roce_ll2->mac_address);
+
+ rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle);
+ if (rc)
+ DP_ERR(cdev,
+ "qed roce ll2 stop: failed to terminate LL2 connection (rc=%d)\n",
+ rc);
+
+ qed_ll2_release_connection(QED_LEADING_HWFN(cdev), roce_ll2->handle);
+
+ roce_ll2->handle = QED_LL2_UNUSED_HANDLE;
+
+ kfree(roce_ll2);
+
+ return rc;
+}
+
+static int qed_roce_ll2_tx(struct qed_dev *cdev,
+ struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_tx_params *params)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+ enum qed_ll2_roce_flavor_type qed_roce_flavor;
+ u8 flags = 0;
+ int rc;
+ int i;
+
+ if (!cdev || !pkt || !params) {
+ DP_ERR(cdev,
+ "roce ll2 tx: failed tx because one of the following is NULL - drv=%p, pkt=%p, params=%p\n",
+ cdev, pkt, params);
+ return -EINVAL;
+ }
+
+ qed_roce_flavor = (pkt->roce_mode == ROCE_V1) ? QED_LL2_ROCE
+ : QED_LL2_RROCE;
+
+ if (pkt->roce_mode == ROCE_V2_IPV4)
+ flags |= BIT(CORE_TX_BD_FLAGS_IP_CSUM_SHIFT);
+
+ /* Tx header */
+ rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev), roce_ll2->handle,
+ 1 + pkt->n_seg, 0, flags, 0,
+ qed_roce_flavor, pkt->header.baddr,
+ pkt->header.len, pkt, 1);
+ if (rc) {
+ DP_ERR(cdev, "roce ll2 tx: header failed (rc=%d)\n", rc);
+ return QED_ROCE_TX_HEAD_FAILURE;
+ }
+
+ /* Tx payload */
+ for (i = 0; i < pkt->n_seg; i++) {
+ rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle,
+ pkt->payload[i].baddr,
+ pkt->payload[i].len);
+ if (rc) {
+ /* If failed not much to do here, partial packet has
+ * been posted * we can't free memory, will need to wait
+ * for completion
+ */
+ DP_ERR(cdev,
+ "roce ll2 tx: payload failed (rc=%d)\n", rc);
+ return QED_ROCE_TX_FRAG_FAILURE;
+ }
+ }
+
+ return 0;
+}
+
+static int qed_roce_ll2_post_rx_buffer(struct qed_dev *cdev,
+ struct qed_roce_ll2_buffer *buf,
+ u64 cookie, u8 notify_fw)
+{
+ return qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
+ QED_LEADING_HWFN(cdev)->ll2->handle,
+ buf->baddr, buf->len,
+ (void *)(uintptr_t)cookie, notify_fw);
+}
+
+static int qed_roce_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
+{
+ struct qed_hwfn *hwfn = QED_LEADING_HWFN(cdev);
+ struct qed_roce_ll2_info *roce_ll2 = hwfn->ll2;
+
+ return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
+ roce_ll2->handle, stats);
+}
+
static const struct qed_rdma_ops qed_rdma_ops_pass = {
.common = &qed_common_ops_pass,
.fill_dev_info = &qed_fill_rdma_dev_info,
@@ -2659,6 +2894,12 @@ static const struct qed_rdma_ops qed_rdma_ops_pass = {
.rdma_free_tid = &qed_rdma_free_tid,
.rdma_register_tid = &qed_rdma_register_tid,
.rdma_deregister_tid = &qed_rdma_deregister_tid,
+ .roce_ll2_start = &qed_roce_ll2_start,
+ .roce_ll2_stop = &qed_roce_ll2_stop,
+ .roce_ll2_tx = &qed_roce_ll2_tx,
+ .roce_ll2_post_rx_buffer = &qed_roce_ll2_post_rx_buffer,
+ .roce_ll2_set_mac_filter = &qed_roce_ll2_set_mac_filter,
+ .roce_ll2_stats = &qed_roce_ll2_stats,
};
const struct qed_rdma_ops *qed_get_rdma_ops()
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.h b/drivers/net/ethernet/qlogic/qed/qed_roce.h
index 71b37fc..f688a47 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.h
@@ -42,6 +42,7 @@
#include "qed.h"
#include "qed_dev_api.h"
#include "qed_hsi.h"
+#include "qed_ll2.h"
#define QED_RDMA_MAX_FMR (RDMA_MAX_TIDS)
#define QED_RDMA_MAX_P_KEY (1)
diff --git a/include/linux/qed/qed_roce_if.h b/include/linux/qed/qed_roce_if.h
index 0b6df6e..53047d3 100644
--- a/include/linux/qed/qed_roce_if.h
+++ b/include/linux/qed/qed_roce_if.h
@@ -39,6 +39,16 @@
#include <linux/slab.h>
#include <linux/qed/qed_if.h>
#include <linux/qed/qed_ll2_if.h>
+#include <linux/qed/rdma_common.h>
+
+enum qed_roce_ll2_tx_dest {
+ /* Light L2 TX Destination to the Network */
+ QED_ROCE_LL2_TX_DEST_NW,
+
+ /* Light L2 TX Destination to the Loopback */
+ QED_ROCE_LL2_TX_DEST_LB,
+ QED_ROCE_LL2_TX_DEST_MAX
+};
#define QED_RDMA_MAX_CNQ_SIZE (0xFFFF)
@@ -461,6 +471,61 @@ struct qed_rdma_counters_out_params {
#define QED_ROCE_TX_HEAD_FAILURE (1)
#define QED_ROCE_TX_FRAG_FAILURE (2)
+struct qed_roce_ll2_header {
+ void *vaddr;
+ dma_addr_t baddr;
+ size_t len;
+};
+
+struct qed_roce_ll2_buffer {
+ dma_addr_t baddr;
+ size_t len;
+};
+
+struct qed_roce_ll2_packet {
+ struct qed_roce_ll2_header header;
+ int n_seg;
+ struct qed_roce_ll2_buffer payload[RDMA_MAX_SGE_PER_SQ_WQE];
+ int roce_mode;
+ enum qed_roce_ll2_tx_dest tx_dest;
+};
+
+struct qed_roce_ll2_tx_params {
+ int reserved;
+};
+
+struct qed_roce_ll2_rx_params {
+ u16 vlan_id;
+ u8 smac[ETH_ALEN];
+ int rc;
+};
+
+struct qed_roce_ll2_cbs {
+ void (*tx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt);
+
+ void (*rx_cb)(void *pdev, struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_rx_params *params);
+};
+
+struct qed_roce_ll2_params {
+ u16 max_rx_buffers;
+ u16 max_tx_buffers;
+ u16 mtu;
+ u8 mac_address[ETH_ALEN];
+ struct qed_roce_ll2_cbs cbs;
+ void *cb_cookie;
+};
+
+struct qed_roce_ll2_info {
+ u8 handle;
+ struct qed_roce_ll2_cbs cbs;
+ u8 mac_address[ETH_ALEN];
+ void *cb_cookie;
+
+ /* Lock to protect ll2 */
+ struct mutex lock;
+};
+
enum qed_rdma_type {
QED_RDMA_TYPE_ROCE,
};
@@ -518,6 +583,20 @@ struct qed_rdma_ops {
int (*rdma_deregister_tid)(void *rdma_cxt, u32 itid);
int (*rdma_alloc_tid)(void *rdma_cxt, u32 *itid);
void (*rdma_free_tid)(void *rdma_cxt, u32 itid);
+ int (*roce_ll2_start)(struct qed_dev *cdev,
+ struct qed_roce_ll2_params *params);
+ int (*roce_ll2_stop)(struct qed_dev *cdev);
+ int (*roce_ll2_tx)(struct qed_dev *cdev,
+ struct qed_roce_ll2_packet *packet,
+ struct qed_roce_ll2_tx_params *params);
+ int (*roce_ll2_post_rx_buffer)(struct qed_dev *cdev,
+ struct qed_roce_ll2_buffer *buf,
+ u64 cookie, u8 notify_fw);
+ int (*roce_ll2_set_mac_filter)(struct qed_dev *cdev,
+ u8 *old_mac_address,
+ u8 *new_mac_address);
+ int (*roce_ll2_stats)(struct qed_dev *cdev,
+ struct qed_ll2_stats *stats);
};
const struct qed_rdma_ops *qed_get_rdma_ops(void);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 10/11] Add GSI support
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for GSI over light L2.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/Makefile | 2 +-
drivers/infiniband/hw/qedr/qedr.h | 10 +
drivers/infiniband/hw/qedr/qedr_cm.c | 626 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/qedr_cm.h | 21 +
drivers/infiniband/hw/qedr/verbs.c | 22 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 97 ++++-
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 22 +-
drivers/net/ethernet/qlogic/qed/qed_roce.c | 66 +++
8 files changed, 853 insertions(+), 13 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.c
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
index b10f2b1..ba7067c 100644
--- a/drivers/infiniband/hw/qedr/Makefile
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
-qedr-y := main.o verbs.o
+qedr-y := main.o verbs.o qedr_cm.o
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 7ae0180..ae50d2c 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -53,6 +53,7 @@ enum DP_QEDR_MODULE {
QEDR_MSG_SQ = 0x80000,
QEDR_MSG_QP = (QEDR_MSG_SQ | QEDR_MSG_RQ),
QEDR_MSG_MR = 0x100000,
+ QEDR_MSG_GSI = 0x200000,
QEDR_MSG_MISC = 0x400000,
};
@@ -147,6 +148,10 @@ struct qedr_dev {
u8 num_hwfns;
uint wq_multiplier;
u8 gsi_ll2_mac_address[ETH_ALEN];
+ int gsi_qp_created;
+ struct qedr_cq *gsi_sqcq;
+ struct qedr_cq *gsi_rqcq;
+ struct qedr_qp *gsi_qp;
};
#define QEDR_MAX_SQ_PBL (0x8000)
@@ -245,6 +250,9 @@ struct qedr_cq {
u16 icid;
+ /* Lock to protect completion handler */
+ spinlock_t comp_handler_lock;
+
/* Lock to protect multiplem CQ's */
spinlock_t cq_lock;
u8 arm_flags;
@@ -291,6 +299,7 @@ struct qedr_qp_hwq_info {
u16 prod;
u16 cons;
u16 wqe_cons;
+ u16 gsi_cons;
u16 max_wr;
/* DB */
@@ -365,6 +374,7 @@ struct qedr_qp {
struct ib_sge sg_list[RDMA_MAX_SGE_PER_RQ_WQE];
u8 wqe_size;
+ u8 smac[ETH_ALEN];
u16 vlan_id;
int rc;
} *rqe_wr_id;
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.c b/drivers/infiniband/hw/qedr/qedr_cm.c
new file mode 100644
index 0000000..646027b
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_cm.c
@@ -0,0 +1,626 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#include <linux/dma-mapping.h>
+#include <linux/crc32.h>
+#include <linux/iommu.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/udp.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_cache.h>
+
+#include "qedr_hsi.h"
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qedr.h"
+#include "qedr_hsi.h"
+#include "verbs.h"
+#include "qedr_user.h"
+#include "qedr_hsi.h"
+#include "qedr_cm.h"
+
+void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info)
+{
+ info->gsi_cons = (info->gsi_cons + 1) % info->max_wr;
+}
+
+void qedr_store_gsi_qp_cq(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs)
+{
+ dev->gsi_qp_created = 1;
+ dev->gsi_sqcq = get_qedr_cq(attrs->send_cq);
+ dev->gsi_rqcq = get_qedr_cq(attrs->recv_cq);
+ dev->gsi_qp = qp;
+}
+
+void qedr_ll2_tx_cb(void *_qdev, struct qed_roce_ll2_packet *pkt)
+{
+ struct qedr_dev *dev = (struct qedr_dev *)_qdev;
+ struct qedr_cq *cq = dev->gsi_sqcq;
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+
+ DP_VERBOSE(dev, QEDR_MSG_GSI,
+ "LL2 TX CB: gsi_sqcq=%p, gsi_rqcq=%p, gsi_cons=%d, ibcq_comp=%s\n",
+ dev->gsi_sqcq, dev->gsi_rqcq, qp->sq.gsi_cons,
+ cq->ibcq.comp_handler ? "Yes" : "No");
+
+ dma_free_coherent(&dev->pdev->dev, pkt->header.len, pkt->header.vaddr,
+ pkt->header.baddr);
+ kfree(pkt);
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+ qedr_inc_sw_gsi_cons(&qp->sq);
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (cq->ibcq.comp_handler) {
+ spin_lock_irqsave(&cq->comp_handler_lock, flags);
+ (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+ spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+ }
+}
+
+static inline void qedr_print_hex_dump(const char *str, void *buf, size_t size)
+{
+ print_hex_dump(KERN_INFO, str, DUMP_PREFIX_OFFSET, 32, 1, buf,
+ size, false);
+}
+
+void qedr_ll2_rx_cb(void *_dev, struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_rx_params *params)
+{
+ struct qedr_dev *dev = (struct qedr_dev *)_dev;
+ struct qedr_cq *cq = dev->gsi_rqcq;
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+
+ /* CM packet dump - supported unless IOMMU is enabled */
+ if (unlikely((dev->dp_level <= QED_LEVEL_VERBOSE) &&
+ (dev->dp_module & QEDR_MSG_GSI) &&
+ (!iommu_present(&pci_bus_type))))
+ qedr_print_hex_dump("RX CM ",
+ phys_to_virt(pkt->payload[0].baddr),
+ pkt->payload[0].len + IB_GRH_BYTES);
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ qp->rqe_wr_id[qp->rq.gsi_cons].rc = params->rc;
+ qp->rqe_wr_id[qp->rq.gsi_cons].vlan_id = params->vlan_id;
+ qp->rqe_wr_id[qp->rq.gsi_cons].sg_list[0].length = pkt->payload[0].len;
+ ether_addr_copy(qp->rqe_wr_id[qp->rq.gsi_cons].smac, params->smac);
+
+ qedr_inc_sw_gsi_cons(&qp->rq);
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (cq->ibcq.comp_handler) {
+ spin_lock_irqsave(&cq->comp_handler_lock, flags);
+ (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+ spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+ }
+}
+
+static void qedr_destroy_gsi_cq(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ struct qed_rdma_destroy_cq_in_params iparams;
+ struct qed_rdma_destroy_cq_out_params oparams;
+ struct qedr_cq *cq;
+
+ cq = get_qedr_cq(attrs->send_cq);
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+
+ cq = get_qedr_cq(attrs->recv_cq);
+ /* if a dedicated recv_cq was used, delete it too */
+ if (iparams.icid != cq->icid) {
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+ }
+}
+
+static inline int qedr_check_gsi_qp_attrs(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ if (attrs->cap.max_recv_sge > QEDR_GSI_MAX_RECV_SGE) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_recv_sge is larger the max %d>%d\n",
+ attrs->cap.max_recv_sge, QEDR_GSI_MAX_RECV_SGE);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_recv_wr > QEDR_GSI_MAX_RECV_WR) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_recv_wr is too large %d>%d\n",
+ attrs->cap.max_recv_wr, QEDR_GSI_MAX_RECV_WR);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_send_wr > QEDR_GSI_MAX_SEND_WR) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_send_wr is too large %d>%d\n",
+ attrs->cap.max_send_wr, QEDR_GSI_MAX_SEND_WR);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs,
+ struct qedr_qp *qp)
+{
+ struct qed_roce_ll2_params ll2_params;
+ int rc;
+
+ rc = qedr_check_gsi_qp_attrs(dev, attrs);
+ if (rc)
+ return ERR_PTR(rc);
+
+ /* configure and start LL2 */
+ memset(&ll2_params, 0, sizeof(ll2_params));
+ ll2_params.max_tx_buffers = attrs->cap.max_send_wr;
+ ll2_params.max_rx_buffers = attrs->cap.max_recv_wr;
+ ll2_params.cbs.tx_cb = qedr_ll2_tx_cb;
+ ll2_params.cbs.rx_cb = qedr_ll2_rx_cb;
+ ll2_params.cb_cookie = (void *)dev;
+ ll2_params.mtu = dev->ndev->mtu;
+ ether_addr_copy(ll2_params.mac_address, dev->ndev->dev_addr);
+ rc = dev->ops->roce_ll2_start(dev->cdev, &ll2_params);
+ if (rc) {
+ DP_ERR(dev, "create gsi qp: failed on ll2 start. rc=%d\n", rc);
+ return ERR_PTR(rc);
+ }
+
+ /* create QP */
+ qp->ibqp.qp_num = 1;
+ qp->rq.max_wr = attrs->cap.max_recv_wr;
+ qp->sq.max_wr = attrs->cap.max_send_wr;
+
+ qp->rqe_wr_id = kcalloc(qp->rq.max_wr, sizeof(*qp->rqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->rqe_wr_id) {
+ DP_ERR(dev, "create gsi qp: failed on rqe_wr_id allocation\n");
+ goto err;
+ }
+ qp->wqe_wr_id = kcalloc(qp->sq.max_wr, sizeof(*qp->wqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->wqe_wr_id) {
+ DP_ERR(dev, "create gsi qp: failed on wqe_wr_id allocation\n");
+ goto err;
+ }
+
+ qedr_store_gsi_qp_cq(dev, qp, attrs);
+ ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
+
+ /* the GSI CQ is handled by the driver so remove it from the FW */
+ qedr_destroy_gsi_cq(dev, attrs);
+ dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI;
+ dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI;
+
+ DP_VERBOSE(dev, QEDR_MSG_GSI, "created GSI QP %p\n", qp);
+
+ return &qp->ibqp;
+
+err:
+ kfree(qp->rqe_wr_id);
+
+ rc = dev->ops->roce_ll2_stop(dev->cdev);
+ if (rc)
+ DP_ERR(dev, "create gsi qp: failed destroy on create\n");
+
+ return ERR_PTR(-ENOMEM);
+}
+
+int qedr_destroy_gsi_qp(struct qedr_dev *dev)
+{
+ int rc;
+
+ rc = dev->ops->roce_ll2_stop(dev->cdev);
+ if (rc)
+ DP_ERR(dev, "destroy gsi qp: failed (rc=%d)\n", rc);
+ else
+ DP_VERBOSE(dev, (QEDR_MSG_QP | QEDR_MSG_GSI),
+ "destroy gsi qp: success\n");
+
+ return rc;
+}
+
+#define QEDR_MAX_UD_HEADER_SIZE (100)
+#define QEDR_GSI_QPN (1)
+static inline int qedr_gsi_build_header(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_send_wr *swr,
+ struct ib_ud_header *udh,
+ int *roce_mode)
+{
+ bool has_vlan = false, has_grh_ipv6 = true;
+ struct ib_ah_attr *ah_attr = &get_qedr_ah(ud_wr(swr)->ah)->attr;
+ struct ib_global_route *grh = &ah_attr->grh;
+ union ib_gid sgid;
+ int send_size = 0;
+ u16 vlan_id = 0;
+ u16 ether_type;
+ struct ib_gid_attr sgid_attr;
+ int rc;
+ int ip_ver = 0;
+
+ bool has_udp = false;
+ int i;
+
+ send_size = 0;
+ for (i = 0; i < swr->num_sge; ++i)
+ send_size += swr->sg_list[i].length;
+
+ rc = ib_get_cached_gid(qp->ibqp.device, ah_attr->port_num,
+ grh->sgid_index, &sgid, &sgid_attr);
+ if (rc) {
+ DP_ERR(dev,
+ "gsi post send: failed to get cached GID (port=%d, ix=%d)\n",
+ ah_attr->port_num, grh->sgid_index);
+ return rc;
+ }
+
+ vlan_id = rdma_vlan_dev_vlan_id(sgid_attr.ndev);
+ if (vlan_id < VLAN_CFI_MASK)
+ has_vlan = true;
+ if (sgid_attr.ndev)
+ dev_put(sgid_attr.ndev);
+
+ if (!memcmp(&sgid, &zgid, sizeof(sgid))) {
+ DP_ERR(dev, "gsi post send: GID not found GID index %d\n",
+ ah_attr->grh.sgid_index);
+ return -ENOENT;
+ }
+
+ has_udp = (sgid_attr.gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP);
+ if (!has_udp) {
+ /* RoCE v1 */
+ ether_type = ETH_P_ROCE;
+ *roce_mode = ROCE_V1;
+ } else if (ipv6_addr_v4mapped((struct in6_addr *)&sgid)) {
+ /* RoCE v2 IPv4 */
+ ip_ver = 4;
+ ether_type = ETH_P_IP;
+ has_grh_ipv6 = false;
+ *roce_mode = ROCE_V2_IPV4;
+ } else {
+ /* RoCE v2 IPv6 */
+ ip_ver = 6;
+ ether_type = ETH_P_IPV6;
+ *roce_mode = ROCE_V2_IPV6;
+ }
+
+ rc = ib_ud_header_init(send_size, false, true, has_vlan,
+ has_grh_ipv6, ip_ver, has_udp, 0, udh);
+ if (rc) {
+ DP_ERR(dev, "gsi post send: failed to init header\n");
+ return rc;
+ }
+
+ /* ENET + VLAN headers */
+ ether_addr_copy(udh->eth.dmac_h, ah_attr->dmac);
+ ether_addr_copy(udh->eth.smac_h, dev->ndev->dev_addr);
+ if (has_vlan) {
+ udh->eth.type = htons(ETH_P_8021Q);
+ udh->vlan.tag = htons(vlan_id);
+ udh->vlan.type = htons(ether_type);
+ } else {
+ udh->eth.type = htons(ether_type);
+ }
+
+ /* BTH */
+ udh->bth.solicited_event = !!(swr->send_flags & IB_SEND_SOLICITED);
+ udh->bth.pkey = QEDR_ROCE_PKEY_DEFAULT;
+ udh->bth.destination_qpn = htonl(ud_wr(swr)->remote_qpn);
+ udh->bth.psn = htonl((qp->sq_psn++) & ((1 << 24) - 1));
+ udh->bth.opcode = IB_OPCODE_UD_SEND_ONLY;
+
+ /* DETH */
+ udh->deth.qkey = htonl(0x80010000);
+ udh->deth.source_qpn = htonl(QEDR_GSI_QPN);
+
+ if (has_grh_ipv6) {
+ /* GRH / IPv6 header */
+ udh->grh.traffic_class = grh->traffic_class;
+ udh->grh.flow_label = grh->flow_label;
+ udh->grh.hop_limit = grh->hop_limit;
+ udh->grh.destination_gid = grh->dgid;
+ memcpy(&udh->grh.source_gid.raw, &sgid.raw,
+ sizeof(udh->grh.source_gid.raw));
+ } else {
+ /* IPv4 header */
+ u32 ipv4_addr;
+
+ udh->ip4.protocol = IPPROTO_UDP;
+ udh->ip4.tos = htonl(ah_attr->grh.flow_label);
+ udh->ip4.frag_off = htons(IP_DF);
+ udh->ip4.ttl = ah_attr->grh.hop_limit;
+
+ ipv4_addr = qedr_get_ipv4_from_gid(sgid.raw);
+ udh->ip4.saddr = ipv4_addr;
+ ipv4_addr = qedr_get_ipv4_from_gid(ah_attr->grh.dgid.raw);
+ udh->ip4.daddr = ipv4_addr;
+ /* note: checksum is calculated by the device */
+ }
+
+ /* UDP */
+ if (has_udp) {
+ udh->udp.sport = htons(QEDR_ROCE_V2_UDP_SPORT);
+ udh->udp.dport = htons(ROCE_V2_UDP_DPORT);
+ udh->udp.csum = 0;
+ /* UDP length is untouched hence is zero */
+ }
+ return 0;
+}
+
+static inline int qedr_gsi_build_packet(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_send_wr *swr,
+ struct qed_roce_ll2_packet **p_packet)
+{
+ u8 ud_header_buffer[QEDR_MAX_UD_HEADER_SIZE];
+ struct qed_roce_ll2_packet *packet;
+ struct pci_dev *pdev = dev->pdev;
+ int roce_mode, header_size;
+ struct ib_ud_header udh;
+ int i, rc;
+
+ *p_packet = NULL;
+
+ rc = qedr_gsi_build_header(dev, qp, swr, &udh, &roce_mode);
+ if (rc)
+ return rc;
+
+ header_size = ib_ud_header_pack(&udh, &ud_header_buffer);
+
+ packet = kzalloc(sizeof(*packet), GFP_ATOMIC);
+ if (!packet) {
+ DP_ERR(dev, "gsi post send: failed to allocate packet\n");
+ return -ENOMEM;
+ }
+
+ packet->header.vaddr = dma_alloc_coherent(&pdev->dev, header_size,
+ &packet->header.baddr,
+ GFP_ATOMIC);
+ if (!packet->header.vaddr) {
+ DP_ERR(dev,
+ "cm header build: failed to allocate buffer for header\n");
+ kfree(packet);
+ return -ENOMEM;
+ }
+
+ if (ether_addr_equal(udh.eth.smac_h, udh.eth.dmac_h))
+ packet->tx_dest = QED_ROCE_LL2_TX_DEST_NW;
+ else
+ packet->tx_dest = QED_ROCE_LL2_TX_DEST_LB;
+
+ packet->roce_mode = roce_mode;
+ memcpy(packet->header.vaddr, ud_header_buffer, header_size);
+ packet->header.len = header_size;
+ packet->n_seg = swr->num_sge;
+ for (i = 0; i < packet->n_seg; i++) {
+ packet->payload[i].baddr = swr->sg_list[i].addr;
+ packet->payload[i].len = swr->sg_list[i].length;
+ }
+
+ *p_packet = packet;
+
+ return 0;
+}
+
+int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ struct qed_roce_ll2_packet *pkt = NULL;
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_roce_ll2_tx_params params;
+ struct qedr_dev *dev = qp->dev;
+ unsigned long flags;
+ int rc;
+
+ if (qp->state != QED_ROCE_QP_STATE_RTS) {
+ *bad_wr = wr;
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTS\n",
+ qp->state);
+ return -EINVAL;
+ }
+
+ if (wr->num_sge > RDMA_MAX_SGE_PER_SQ_WQE) {
+ DP_ERR(dev, "gsi post send: num_sge is too large (%d>%d)\n",
+ wr->num_sge, RDMA_MAX_SGE_PER_SQ_WQE);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ if (wr->opcode != IB_WR_SEND) {
+ DP_ERR(dev,
+ "gsi post send: failed due to unsupported opcode %d\n",
+ wr->opcode);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ memset(¶ms, 0, sizeof(params));
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ rc = qedr_gsi_build_packet(dev, qp, wr, &pkt);
+ if (rc) {
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ goto err;
+ }
+
+ if (unlikely((dev->dp_level <= QED_LEVEL_VERBOSE) &&
+ (dev->dp_module & QEDR_MSG_GSI)))
+ qedr_print_hex_dump("TX CM Header ", pkt->header.vaddr,
+ pkt->header.len);
+
+ rc = dev->ops->roce_ll2_tx(dev->cdev, pkt, ¶ms);
+ if (!rc) {
+ qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id;
+ qedr_inc_sw_prod(&qp->sq);
+ DP_VERBOSE(qp->dev, QEDR_MSG_GSI,
+ "gsi post send: opcode=%d, in_irq=%ld, irqs_disabled=%d, wr_id=%llx\n",
+ wr->opcode, in_irq(), irqs_disabled(), wr->wr_id);
+ } else {
+ if (rc == QED_ROCE_TX_HEAD_FAILURE) {
+ /* TX failed while posting header - release resources */
+ dma_free_coherent(&dev->pdev->dev, pkt->header.len,
+ pkt->header.vaddr, pkt->header.baddr);
+ kfree(pkt);
+ } else if (rc == QED_ROCE_TX_FRAG_FAILURE) {
+ /* NTD since TX failed while posting a fragment. We will
+ * release the resources on TX callback
+ */
+ }
+
+ DP_ERR(dev, "gsi post send: failed to transmit (rc=%d)\n", rc);
+ rc = -EAGAIN;
+ *bad_wr = wr;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (wr->next) {
+ DP_ERR(dev,
+ "gsi post send: failed second WR. Only one WR may be passed at a time\n");
+ *bad_wr = wr->next;
+ rc = -EINVAL;
+ }
+
+ return rc;
+
+err:
+ *bad_wr = wr;
+ return rc;
+}
+
+int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibqp->device);
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_roce_ll2_buffer buf;
+ unsigned long flags;
+ int status = 0;
+ int rc;
+
+ if ((qp->state != QED_ROCE_QP_STATE_RTR) &&
+ (qp->state != QED_ROCE_QP_STATE_RTS)) {
+ *bad_wr = wr;
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTR/S\n",
+ qp->state);
+ return -EINVAL;
+ }
+
+ memset(&buf, 0, sizeof(buf));
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ while (wr) {
+ if (wr->num_sge > QEDR_GSI_MAX_RECV_SGE) {
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. too many sges %d>%d\n",
+ wr->num_sge, QEDR_GSI_MAX_RECV_SGE);
+ goto err;
+ }
+
+ buf.baddr = wr->sg_list[0].addr;
+ buf.len = wr->sg_list[0].length;
+
+ rc = dev->ops->roce_ll2_post_rx_buffer(dev->cdev, &buf, 0, 1);
+ if (rc) {
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer (rc=%d)\n",
+ rc);
+ goto err;
+ }
+
+ memset(&qp->rqe_wr_id[qp->rq.prod], 0,
+ sizeof(qp->rqe_wr_id[qp->rq.prod]));
+ qp->rqe_wr_id[qp->rq.prod].sg_list[0] = wr->sg_list[0];
+ qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id;
+
+ qedr_inc_sw_prod(&qp->rq);
+
+ wr = wr->next;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ return status;
+err:
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ *bad_wr = wr;
+ return -ENOMEM;
+}
+
+int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+ int i = 0;
+
+ spin_lock_irqsave(&cq->cq_lock, flags);
+
+ while (i < num_entries && qp->rq.cons != qp->rq.gsi_cons) {
+ memset(&wc[i], 0, sizeof(*wc));
+
+ wc[i].qp = &qp->ibqp;
+ wc[i].wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
+ wc[i].opcode = IB_WC_RECV;
+ wc[i].pkey_index = 0;
+ wc[i].status = (qp->rqe_wr_id[qp->rq.cons].rc) ?
+ IB_WC_GENERAL_ERR : IB_WC_SUCCESS;
+ /* 0 - currently only one recv sg is supported */
+ wc[i].byte_len = qp->rqe_wr_id[qp->rq.cons].sg_list[0].length;
+ wc[i].wc_flags |= IB_WC_GRH | IB_WC_IP_CSUM_OK;
+ ether_addr_copy(wc[i].smac, qp->rqe_wr_id[qp->rq.cons].smac);
+ wc[i].wc_flags |= IB_WC_WITH_SMAC;
+ if (qp->rqe_wr_id[qp->rq.cons].vlan_id) {
+ wc[i].wc_flags |= IB_WC_WITH_VLAN;
+ wc[i].vlan_id = qp->rqe_wr_id[qp->rq.cons].vlan_id;
+ }
+
+ qedr_inc_sw_cons(&qp->rq);
+ i++;
+ }
+
+ while (i < num_entries && qp->sq.cons != qp->sq.gsi_cons) {
+ memset(&wc[i], 0, sizeof(*wc));
+
+ wc[i].qp = &qp->ibqp;
+ wc[i].wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id;
+ wc[i].opcode = IB_WC_SEND;
+ wc[i].status = IB_WC_SUCCESS;
+
+ qedr_inc_sw_cons(&qp->sq);
+ i++;
+ }
+
+ spin_unlock_irqrestore(&cq->cq_lock, flags);
+
+ DP_VERBOSE(dev, (QEDR_MSG_CQ | QEDR_MSG_GSI),
+ "gsi poll_cq: requested entries=%d, actual=%d, qp->rq.cons=%d, qp->rq.gsi_cons=%x, qp->sq.cons=%d, qp->sq.gsi_cons=%d, qp_num=%d\n",
+ num_entries, i, qp->rq.cons, qp->rq.gsi_cons, qp->sq.cons,
+ qp->sq.gsi_cons, qp->ibqp.qp_num);
+
+ return i;
+}
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.h b/drivers/infiniband/hw/qedr/qedr_cm.h
index b8a8b76..9ba6e15 100644
--- a/drivers/infiniband/hw/qedr/qedr_cm.h
+++ b/drivers/infiniband/hw/qedr/qedr_cm.h
@@ -32,9 +32,30 @@
#ifndef LINUX_QEDR_CM_H_
#define LINUX_QEDR_CM_H_
+#define QEDR_GSI_MAX_RECV_WR (4096)
+#define QEDR_GSI_MAX_SEND_WR (4096)
+
+#define QEDR_GSI_MAX_RECV_SGE (1) /* LL2 FW limitation */
+
+#define ETH_P_ROCE (0x8915)
+#define QEDR_ROCE_V2_UDP_SPORT (0000)
+
static inline u32 qedr_get_ipv4_from_gid(u8 *gid)
{
return *(u32 *)(void *)&gid[12];
}
+/* RDMA CM */
+int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
+int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr);
+int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr);
+struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs,
+ struct qedr_qp *qp);
+void qedr_store_gsi_qp_cq(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct ib_qp_init_attr *attrs);
+int qedr_destroy_gsi_qp(struct qedr_dev *dev);
+void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info);
#endif
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 979555e..9381bc6 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -1471,6 +1471,15 @@ struct ib_qp *qedr_create_qp(struct ib_pd *ibpd,
qedr_set_qp_init_params(dev, qp, pd, attrs);
+ if (attrs->qp_type == IB_QPT_GSI) {
+ if (udata) {
+ DP_ERR(dev,
+ "create qp: unexpected udata when creating GSI QP\n");
+ goto err0;
+ }
+ return qedr_create_gsi_qp(dev, attrs, qp);
+ }
+
memset(&in_params, 0, sizeof(in_params));
if (udata) {
@@ -2040,6 +2049,8 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp);
if (rc)
return rc;
+ } else {
+ qedr_destroy_gsi_qp(dev);
}
if (ibqp->uobject && ibqp->uobject->context) {
@@ -2959,6 +2970,10 @@ int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
*bad_wr = NULL;
+ if (qp->qp_type == IB_QPT_GSI) {
+ return qedr_gsi_post_send(ibqp, wr, bad_wr);
+ }
+
spin_lock_irqsave(&qp->q_lock, flags);
if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
@@ -3015,6 +3030,10 @@ int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
unsigned long flags;
int status = 0;
+ if (qp->qp_type == IB_QPT_GSI) {
+ return qedr_gsi_post_recv(ibqp, wr, bad_wr);
+ }
+
spin_lock_irqsave(&qp->q_lock, flags);
if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
@@ -3450,6 +3469,9 @@ int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
int update = 0;
int done = 0;
+ if (cq->cq_type == QEDR_CQ_TYPE_GSI)
+ return qedr_gsi_poll_cq(ibcq, num_entries, wc);
+
spin_lock_irqsave(&cq->cq_lock, flags);
old_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
while (num_entries && is_valid_cqe(cq, cqe)) {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index 4ea6a07..f907569 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -278,6 +278,7 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
struct qed_ll2_tx_packet *p_pkt = NULL;
struct qed_ll2_info *p_ll2_conn;
struct qed_ll2_tx_queue *p_tx;
+ dma_addr_t tx_frag;
p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
if (!p_ll2_conn)
@@ -294,15 +295,25 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
list_del(&p_pkt->list_entry);
b_last_packet = list_empty(&p_tx->active_descq);
list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+
p_tx->cur_completing_packet = *p_pkt;
p_tx->cur_completing_bd_idx = 1;
b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
-
- qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
- p_pkt->cookie,
- p_pkt->bds_set[0].tx_frag,
- b_last_frag,
- b_last_packet);
+ tx_frag = p_pkt->bds_set[0].tx_frag;
+ if (p_ll2_conn->gsi_enable)
+ qed_ll2b_release_tx_gsi_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag,
+ b_last_packet);
+ else
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag,
+ b_last_packet);
}
}
@@ -314,6 +325,7 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
struct qed_ll2_tx_packet *p_pkt;
bool b_last_frag = false;
unsigned long flags;
+ dma_addr_t tx_frag;
int rc = -EINVAL;
spin_lock_irqsave(&p_tx->lock, flags);
@@ -354,11 +366,19 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
spin_unlock_irqrestore(&p_tx->lock, flags);
- qed_ll2b_complete_tx_packet(p_hwfn,
- p_ll2_conn->my_id,
- p_pkt->cookie,
- p_pkt->bds_set[0].tx_frag,
- b_last_frag, !num_bds);
+ tx_frag = p_pkt->bds_set[0].tx_frag;
+ if (p_ll2_conn->gsi_enable)
+ qed_ll2b_complete_tx_gsi_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag, !num_bds);
+ else
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag, !num_bds);
spin_lock_irqsave(&p_tx->lock, flags);
}
@@ -369,6 +389,54 @@ out:
return rc;
}
+static int
+qed_ll2_rxq_completion_gsi(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info,
+ union core_rx_cqe_union *p_cqe,
+ unsigned long lock_flags, bool b_last_cqe)
+{
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_info->rx_queue;
+ struct qed_ll2_rx_packet *p_pkt = NULL;
+ u16 packet_length, parse_flags, vlan;
+ u32 src_mac_addrhi;
+ u16 src_mac_addrlo;
+
+ if (!list_empty(&p_rx->active_descq))
+ p_pkt = list_first_entry(&p_rx->active_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (!p_pkt) {
+ DP_NOTICE(p_hwfn,
+ "GSI Rx completion but active_descq is empty\n");
+ return -EIO;
+ }
+
+ list_del(&p_pkt->list_entry);
+ parse_flags = le16_to_cpu(p_cqe->rx_cqe_gsi.parse_flags.flags);
+ packet_length = le16_to_cpu(p_cqe->rx_cqe_gsi.data_length);
+ vlan = le16_to_cpu(p_cqe->rx_cqe_gsi.vlan);
+ src_mac_addrhi = le32_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrhi);
+ src_mac_addrlo = le16_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrlo);
+ if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
+ DP_NOTICE(p_hwfn,
+ "Mismatch between active_descq and the LL2 Rx chain\n");
+ list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+ spin_unlock_irqrestore(&p_rx->lock, lock_flags);
+ qed_ll2b_complete_rx_gsi_packet(p_hwfn,
+ p_ll2_info->my_id,
+ p_pkt->cookie,
+ p_pkt->rx_buf_addr,
+ packet_length,
+ p_cqe->rx_cqe_gsi.data_length_error,
+ parse_flags,
+ vlan,
+ src_mac_addrhi,
+ src_mac_addrlo, b_last_cqe);
+ spin_lock_irqsave(&p_rx->lock, lock_flags);
+
+ return 0;
+}
+
static int qed_ll2_rxq_completion_reg(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn,
union core_rx_cqe_union *p_cqe,
@@ -430,6 +498,10 @@ static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
DP_NOTICE(p_hwfn, "LL2 - unexpected Rx CQE slowpath\n");
rc = -EINVAL;
break;
+ case CORE_RX_CQE_TYPE_GSI_OFFLOAD:
+ rc = qed_ll2_rxq_completion_gsi(p_hwfn, p_ll2_conn,
+ cqe, flags, b_last_cqe);
+ break;
case CORE_RX_CQE_TYPE_REGULAR:
rc = qed_ll2_rxq_completion_reg(p_hwfn, p_ll2_conn,
cqe, flags, b_last_cqe);
@@ -527,6 +599,7 @@ static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
}
p_ramrod->action_on_error.error_type = action_on_error;
+ p_ramrod->gsi_offload_flag = p_ll2_conn->gsi_enable;
return qed_spq_post(p_hwfn, p_ent, NULL);
}
@@ -589,6 +662,7 @@ static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
DP_NOTICE(p_hwfn, "Unknown connection type: %d\n", conn_type);
}
+ p_ramrod->gsi_offload_flag = p_ll2_conn->gsi_enable;
return qed_spq_post(p_hwfn, p_ent, NULL);
}
@@ -775,6 +849,7 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
p_ll2_info->tx_dest = p_params->tx_dest;
p_ll2_info->ai_err_packet_too_big = p_params->ai_err_packet_too_big;
p_ll2_info->ai_err_no_buf = p_params->ai_err_no_buf;
+ p_ll2_info->gsi_enable = p_params->gsi_enable;
rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info, rx_num_desc);
if (rc)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index b3783fd..80a5dc2 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -125,6 +125,7 @@ struct qed_ll2_info {
u8 tx_stats_en;
struct qed_ll2_rx_queue rx_queue;
struct qed_ll2_tx_queue tx_queue;
+ u8 gsi_enable;
};
/**
@@ -292,5 +293,24 @@ void qed_ll2_setup(struct qed_hwfn *p_hwfn,
*/
void qed_ll2_free(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_connections);
-
+void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ u16 data_length,
+ u8 data_length_error,
+ u16 parse_flags,
+ u16 vlan,
+ u32 src_mac_addr_hi,
+ u16 src_mac_addr_lo, bool b_last_packet);
+void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet);
+void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 27095f0..cf31a48 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -2633,6 +2633,71 @@ void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
}
+void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet)
+{
+ struct qed_roce_ll2_packet *packet = cookie;
+ struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
+
+ roce_ll2->cbs.tx_cb(roce_ll2->cb_cookie, packet);
+}
+
+void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet)
+{
+ qed_ll2b_complete_tx_gsi_packet(p_hwfn, connection_handle,
+ cookie, first_frag_addr,
+ b_last_fragment, b_last_packet);
+}
+
+void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ u16 data_length,
+ u8 data_length_error,
+ u16 parse_flags,
+ u16 vlan,
+ u32 src_mac_addr_hi,
+ u16 src_mac_addr_lo, bool b_last_packet)
+{
+ struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
+ struct qed_roce_ll2_rx_params params;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ struct qed_roce_ll2_packet pkt;
+
+ DP_VERBOSE(cdev,
+ QED_MSG_LL2,
+ "roce ll2 rx complete: bus_addr=%p, len=%d, data_len_err=%d\n",
+ (void *)(uintptr_t)rx_buf_addr,
+ data_length, data_length_error);
+
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.n_seg = 1;
+ pkt.payload[0].baddr = rx_buf_addr;
+ pkt.payload[0].len = data_length;
+
+ memset(¶ms, 0, sizeof(params));
+ params.vlan_id = vlan;
+ *((u32 *)¶ms.smac[0]) = ntohl(src_mac_addr_hi);
+ *((u16 *)¶ms.smac[4]) = ntohs(src_mac_addr_lo);
+
+ if (data_length_error) {
+ DP_ERR(cdev,
+ "roce ll2 rx complete: data length error %d, length=%d\n",
+ data_length_error, data_length);
+ params.rc = -EINVAL;
+ }
+
+ roce_ll2->cbs.rx_cb(roce_ll2->cb_cookie, &pkt, ¶ms);
+}
+
static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
u8 *old_mac_address,
u8 *new_mac_address)
@@ -2717,6 +2782,7 @@ static int qed_roce_ll2_start(struct qed_dev *cdev,
ll2_params.tx_dest = CORE_TX_DEST_NW;
ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
+ ll2_params.gsi_enable = true;
rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
params->max_rx_buffers,
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 10/11] Add GSI support
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add support for GSI over light L2.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/Makefile | 2 +-
drivers/infiniband/hw/qedr/qedr.h | 10 +
drivers/infiniband/hw/qedr/qedr_cm.c | 626 +++++++++++++++++++++++++++++
drivers/infiniband/hw/qedr/qedr_cm.h | 21 +
drivers/infiniband/hw/qedr/verbs.c | 22 +
drivers/net/ethernet/qlogic/qed/qed_ll2.c | 97 ++++-
drivers/net/ethernet/qlogic/qed/qed_ll2.h | 22 +-
drivers/net/ethernet/qlogic/qed/qed_roce.c | 66 +++
8 files changed, 853 insertions(+), 13 deletions(-)
create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.c
diff --git a/drivers/infiniband/hw/qedr/Makefile b/drivers/infiniband/hw/qedr/Makefile
index b10f2b1..ba7067c 100644
--- a/drivers/infiniband/hw/qedr/Makefile
+++ b/drivers/infiniband/hw/qedr/Makefile
@@ -1,3 +1,3 @@
obj-$(CONFIG_INFINIBAND_QEDR) := qedr.o
-qedr-y := main.o verbs.o
+qedr-y := main.o verbs.o qedr_cm.o
diff --git a/drivers/infiniband/hw/qedr/qedr.h b/drivers/infiniband/hw/qedr/qedr.h
index 7ae0180..ae50d2c 100644
--- a/drivers/infiniband/hw/qedr/qedr.h
+++ b/drivers/infiniband/hw/qedr/qedr.h
@@ -53,6 +53,7 @@ enum DP_QEDR_MODULE {
QEDR_MSG_SQ = 0x80000,
QEDR_MSG_QP = (QEDR_MSG_SQ | QEDR_MSG_RQ),
QEDR_MSG_MR = 0x100000,
+ QEDR_MSG_GSI = 0x200000,
QEDR_MSG_MISC = 0x400000,
};
@@ -147,6 +148,10 @@ struct qedr_dev {
u8 num_hwfns;
uint wq_multiplier;
u8 gsi_ll2_mac_address[ETH_ALEN];
+ int gsi_qp_created;
+ struct qedr_cq *gsi_sqcq;
+ struct qedr_cq *gsi_rqcq;
+ struct qedr_qp *gsi_qp;
};
#define QEDR_MAX_SQ_PBL (0x8000)
@@ -245,6 +250,9 @@ struct qedr_cq {
u16 icid;
+ /* Lock to protect completion handler */
+ spinlock_t comp_handler_lock;
+
/* Lock to protect multiplem CQ's */
spinlock_t cq_lock;
u8 arm_flags;
@@ -291,6 +299,7 @@ struct qedr_qp_hwq_info {
u16 prod;
u16 cons;
u16 wqe_cons;
+ u16 gsi_cons;
u16 max_wr;
/* DB */
@@ -365,6 +374,7 @@ struct qedr_qp {
struct ib_sge sg_list[RDMA_MAX_SGE_PER_RQ_WQE];
u8 wqe_size;
+ u8 smac[ETH_ALEN];
u16 vlan_id;
int rc;
} *rqe_wr_id;
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.c b/drivers/infiniband/hw/qedr/qedr_cm.c
new file mode 100644
index 0000000..646027b
--- /dev/null
+++ b/drivers/infiniband/hw/qedr/qedr_cm.c
@@ -0,0 +1,626 @@
+/* QLogic qed NIC Driver
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+#include <linux/dma-mapping.h>
+#include <linux/crc32.h>
+#include <linux/iommu.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/udp.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_cache.h>
+
+#include "qedr_hsi.h"
+#include <linux/qed/qed_if.h>
+#include <linux/qed/qed_roce_if.h>
+#include "qedr.h"
+#include "qedr_hsi.h"
+#include "verbs.h"
+#include "qedr_user.h"
+#include "qedr_hsi.h"
+#include "qedr_cm.h"
+
+void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info)
+{
+ info->gsi_cons = (info->gsi_cons + 1) % info->max_wr;
+}
+
+void qedr_store_gsi_qp_cq(struct qedr_dev *dev, struct qedr_qp *qp,
+ struct ib_qp_init_attr *attrs)
+{
+ dev->gsi_qp_created = 1;
+ dev->gsi_sqcq = get_qedr_cq(attrs->send_cq);
+ dev->gsi_rqcq = get_qedr_cq(attrs->recv_cq);
+ dev->gsi_qp = qp;
+}
+
+void qedr_ll2_tx_cb(void *_qdev, struct qed_roce_ll2_packet *pkt)
+{
+ struct qedr_dev *dev = (struct qedr_dev *)_qdev;
+ struct qedr_cq *cq = dev->gsi_sqcq;
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+
+ DP_VERBOSE(dev, QEDR_MSG_GSI,
+ "LL2 TX CB: gsi_sqcq=%p, gsi_rqcq=%p, gsi_cons=%d, ibcq_comp=%s\n",
+ dev->gsi_sqcq, dev->gsi_rqcq, qp->sq.gsi_cons,
+ cq->ibcq.comp_handler ? "Yes" : "No");
+
+ dma_free_coherent(&dev->pdev->dev, pkt->header.len, pkt->header.vaddr,
+ pkt->header.baddr);
+ kfree(pkt);
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+ qedr_inc_sw_gsi_cons(&qp->sq);
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (cq->ibcq.comp_handler) {
+ spin_lock_irqsave(&cq->comp_handler_lock, flags);
+ (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+ spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+ }
+}
+
+static inline void qedr_print_hex_dump(const char *str, void *buf, size_t size)
+{
+ print_hex_dump(KERN_INFO, str, DUMP_PREFIX_OFFSET, 32, 1, buf,
+ size, false);
+}
+
+void qedr_ll2_rx_cb(void *_dev, struct qed_roce_ll2_packet *pkt,
+ struct qed_roce_ll2_rx_params *params)
+{
+ struct qedr_dev *dev = (struct qedr_dev *)_dev;
+ struct qedr_cq *cq = dev->gsi_rqcq;
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+
+ /* CM packet dump - supported unless IOMMU is enabled */
+ if (unlikely((dev->dp_level <= QED_LEVEL_VERBOSE) &&
+ (dev->dp_module & QEDR_MSG_GSI) &&
+ (!iommu_present(&pci_bus_type))))
+ qedr_print_hex_dump("RX CM ",
+ phys_to_virt(pkt->payload[0].baddr),
+ pkt->payload[0].len + IB_GRH_BYTES);
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ qp->rqe_wr_id[qp->rq.gsi_cons].rc = params->rc;
+ qp->rqe_wr_id[qp->rq.gsi_cons].vlan_id = params->vlan_id;
+ qp->rqe_wr_id[qp->rq.gsi_cons].sg_list[0].length = pkt->payload[0].len;
+ ether_addr_copy(qp->rqe_wr_id[qp->rq.gsi_cons].smac, params->smac);
+
+ qedr_inc_sw_gsi_cons(&qp->rq);
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (cq->ibcq.comp_handler) {
+ spin_lock_irqsave(&cq->comp_handler_lock, flags);
+ (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+ spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+ }
+}
+
+static void qedr_destroy_gsi_cq(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ struct qed_rdma_destroy_cq_in_params iparams;
+ struct qed_rdma_destroy_cq_out_params oparams;
+ struct qedr_cq *cq;
+
+ cq = get_qedr_cq(attrs->send_cq);
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+
+ cq = get_qedr_cq(attrs->recv_cq);
+ /* if a dedicated recv_cq was used, delete it too */
+ if (iparams.icid != cq->icid) {
+ iparams.icid = cq->icid;
+ dev->ops->rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
+ dev->ops->common->chain_free(dev->cdev, &cq->pbl);
+ }
+}
+
+static inline int qedr_check_gsi_qp_attrs(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs)
+{
+ if (attrs->cap.max_recv_sge > QEDR_GSI_MAX_RECV_SGE) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_recv_sge is larger the max %d>%d\n",
+ attrs->cap.max_recv_sge, QEDR_GSI_MAX_RECV_SGE);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_recv_wr > QEDR_GSI_MAX_RECV_WR) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_recv_wr is too large %d>%d\n",
+ attrs->cap.max_recv_wr, QEDR_GSI_MAX_RECV_WR);
+ return -EINVAL;
+ }
+
+ if (attrs->cap.max_send_wr > QEDR_GSI_MAX_SEND_WR) {
+ DP_ERR(dev,
+ " create gsi qp: failed. max_send_wr is too large %d>%d\n",
+ attrs->cap.max_send_wr, QEDR_GSI_MAX_SEND_WR);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs,
+ struct qedr_qp *qp)
+{
+ struct qed_roce_ll2_params ll2_params;
+ int rc;
+
+ rc = qedr_check_gsi_qp_attrs(dev, attrs);
+ if (rc)
+ return ERR_PTR(rc);
+
+ /* configure and start LL2 */
+ memset(&ll2_params, 0, sizeof(ll2_params));
+ ll2_params.max_tx_buffers = attrs->cap.max_send_wr;
+ ll2_params.max_rx_buffers = attrs->cap.max_recv_wr;
+ ll2_params.cbs.tx_cb = qedr_ll2_tx_cb;
+ ll2_params.cbs.rx_cb = qedr_ll2_rx_cb;
+ ll2_params.cb_cookie = (void *)dev;
+ ll2_params.mtu = dev->ndev->mtu;
+ ether_addr_copy(ll2_params.mac_address, dev->ndev->dev_addr);
+ rc = dev->ops->roce_ll2_start(dev->cdev, &ll2_params);
+ if (rc) {
+ DP_ERR(dev, "create gsi qp: failed on ll2 start. rc=%d\n", rc);
+ return ERR_PTR(rc);
+ }
+
+ /* create QP */
+ qp->ibqp.qp_num = 1;
+ qp->rq.max_wr = attrs->cap.max_recv_wr;
+ qp->sq.max_wr = attrs->cap.max_send_wr;
+
+ qp->rqe_wr_id = kcalloc(qp->rq.max_wr, sizeof(*qp->rqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->rqe_wr_id) {
+ DP_ERR(dev, "create gsi qp: failed on rqe_wr_id allocation\n");
+ goto err;
+ }
+ qp->wqe_wr_id = kcalloc(qp->sq.max_wr, sizeof(*qp->wqe_wr_id),
+ GFP_KERNEL);
+ if (!qp->wqe_wr_id) {
+ DP_ERR(dev, "create gsi qp: failed on wqe_wr_id allocation\n");
+ goto err;
+ }
+
+ qedr_store_gsi_qp_cq(dev, qp, attrs);
+ ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
+
+ /* the GSI CQ is handled by the driver so remove it from the FW */
+ qedr_destroy_gsi_cq(dev, attrs);
+ dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI;
+ dev->gsi_rqcq->cq_type = QEDR_CQ_TYPE_GSI;
+
+ DP_VERBOSE(dev, QEDR_MSG_GSI, "created GSI QP %p\n", qp);
+
+ return &qp->ibqp;
+
+err:
+ kfree(qp->rqe_wr_id);
+
+ rc = dev->ops->roce_ll2_stop(dev->cdev);
+ if (rc)
+ DP_ERR(dev, "create gsi qp: failed destroy on create\n");
+
+ return ERR_PTR(-ENOMEM);
+}
+
+int qedr_destroy_gsi_qp(struct qedr_dev *dev)
+{
+ int rc;
+
+ rc = dev->ops->roce_ll2_stop(dev->cdev);
+ if (rc)
+ DP_ERR(dev, "destroy gsi qp: failed (rc=%d)\n", rc);
+ else
+ DP_VERBOSE(dev, (QEDR_MSG_QP | QEDR_MSG_GSI),
+ "destroy gsi qp: success\n");
+
+ return rc;
+}
+
+#define QEDR_MAX_UD_HEADER_SIZE (100)
+#define QEDR_GSI_QPN (1)
+static inline int qedr_gsi_build_header(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_send_wr *swr,
+ struct ib_ud_header *udh,
+ int *roce_mode)
+{
+ bool has_vlan = false, has_grh_ipv6 = true;
+ struct ib_ah_attr *ah_attr = &get_qedr_ah(ud_wr(swr)->ah)->attr;
+ struct ib_global_route *grh = &ah_attr->grh;
+ union ib_gid sgid;
+ int send_size = 0;
+ u16 vlan_id = 0;
+ u16 ether_type;
+ struct ib_gid_attr sgid_attr;
+ int rc;
+ int ip_ver = 0;
+
+ bool has_udp = false;
+ int i;
+
+ send_size = 0;
+ for (i = 0; i < swr->num_sge; ++i)
+ send_size += swr->sg_list[i].length;
+
+ rc = ib_get_cached_gid(qp->ibqp.device, ah_attr->port_num,
+ grh->sgid_index, &sgid, &sgid_attr);
+ if (rc) {
+ DP_ERR(dev,
+ "gsi post send: failed to get cached GID (port=%d, ix=%d)\n",
+ ah_attr->port_num, grh->sgid_index);
+ return rc;
+ }
+
+ vlan_id = rdma_vlan_dev_vlan_id(sgid_attr.ndev);
+ if (vlan_id < VLAN_CFI_MASK)
+ has_vlan = true;
+ if (sgid_attr.ndev)
+ dev_put(sgid_attr.ndev);
+
+ if (!memcmp(&sgid, &zgid, sizeof(sgid))) {
+ DP_ERR(dev, "gsi post send: GID not found GID index %d\n",
+ ah_attr->grh.sgid_index);
+ return -ENOENT;
+ }
+
+ has_udp = (sgid_attr.gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP);
+ if (!has_udp) {
+ /* RoCE v1 */
+ ether_type = ETH_P_ROCE;
+ *roce_mode = ROCE_V1;
+ } else if (ipv6_addr_v4mapped((struct in6_addr *)&sgid)) {
+ /* RoCE v2 IPv4 */
+ ip_ver = 4;
+ ether_type = ETH_P_IP;
+ has_grh_ipv6 = false;
+ *roce_mode = ROCE_V2_IPV4;
+ } else {
+ /* RoCE v2 IPv6 */
+ ip_ver = 6;
+ ether_type = ETH_P_IPV6;
+ *roce_mode = ROCE_V2_IPV6;
+ }
+
+ rc = ib_ud_header_init(send_size, false, true, has_vlan,
+ has_grh_ipv6, ip_ver, has_udp, 0, udh);
+ if (rc) {
+ DP_ERR(dev, "gsi post send: failed to init header\n");
+ return rc;
+ }
+
+ /* ENET + VLAN headers */
+ ether_addr_copy(udh->eth.dmac_h, ah_attr->dmac);
+ ether_addr_copy(udh->eth.smac_h, dev->ndev->dev_addr);
+ if (has_vlan) {
+ udh->eth.type = htons(ETH_P_8021Q);
+ udh->vlan.tag = htons(vlan_id);
+ udh->vlan.type = htons(ether_type);
+ } else {
+ udh->eth.type = htons(ether_type);
+ }
+
+ /* BTH */
+ udh->bth.solicited_event = !!(swr->send_flags & IB_SEND_SOLICITED);
+ udh->bth.pkey = QEDR_ROCE_PKEY_DEFAULT;
+ udh->bth.destination_qpn = htonl(ud_wr(swr)->remote_qpn);
+ udh->bth.psn = htonl((qp->sq_psn++) & ((1 << 24) - 1));
+ udh->bth.opcode = IB_OPCODE_UD_SEND_ONLY;
+
+ /* DETH */
+ udh->deth.qkey = htonl(0x80010000);
+ udh->deth.source_qpn = htonl(QEDR_GSI_QPN);
+
+ if (has_grh_ipv6) {
+ /* GRH / IPv6 header */
+ udh->grh.traffic_class = grh->traffic_class;
+ udh->grh.flow_label = grh->flow_label;
+ udh->grh.hop_limit = grh->hop_limit;
+ udh->grh.destination_gid = grh->dgid;
+ memcpy(&udh->grh.source_gid.raw, &sgid.raw,
+ sizeof(udh->grh.source_gid.raw));
+ } else {
+ /* IPv4 header */
+ u32 ipv4_addr;
+
+ udh->ip4.protocol = IPPROTO_UDP;
+ udh->ip4.tos = htonl(ah_attr->grh.flow_label);
+ udh->ip4.frag_off = htons(IP_DF);
+ udh->ip4.ttl = ah_attr->grh.hop_limit;
+
+ ipv4_addr = qedr_get_ipv4_from_gid(sgid.raw);
+ udh->ip4.saddr = ipv4_addr;
+ ipv4_addr = qedr_get_ipv4_from_gid(ah_attr->grh.dgid.raw);
+ udh->ip4.daddr = ipv4_addr;
+ /* note: checksum is calculated by the device */
+ }
+
+ /* UDP */
+ if (has_udp) {
+ udh->udp.sport = htons(QEDR_ROCE_V2_UDP_SPORT);
+ udh->udp.dport = htons(ROCE_V2_UDP_DPORT);
+ udh->udp.csum = 0;
+ /* UDP length is untouched hence is zero */
+ }
+ return 0;
+}
+
+static inline int qedr_gsi_build_packet(struct qedr_dev *dev,
+ struct qedr_qp *qp,
+ struct ib_send_wr *swr,
+ struct qed_roce_ll2_packet **p_packet)
+{
+ u8 ud_header_buffer[QEDR_MAX_UD_HEADER_SIZE];
+ struct qed_roce_ll2_packet *packet;
+ struct pci_dev *pdev = dev->pdev;
+ int roce_mode, header_size;
+ struct ib_ud_header udh;
+ int i, rc;
+
+ *p_packet = NULL;
+
+ rc = qedr_gsi_build_header(dev, qp, swr, &udh, &roce_mode);
+ if (rc)
+ return rc;
+
+ header_size = ib_ud_header_pack(&udh, &ud_header_buffer);
+
+ packet = kzalloc(sizeof(*packet), GFP_ATOMIC);
+ if (!packet) {
+ DP_ERR(dev, "gsi post send: failed to allocate packet\n");
+ return -ENOMEM;
+ }
+
+ packet->header.vaddr = dma_alloc_coherent(&pdev->dev, header_size,
+ &packet->header.baddr,
+ GFP_ATOMIC);
+ if (!packet->header.vaddr) {
+ DP_ERR(dev,
+ "cm header build: failed to allocate buffer for header\n");
+ kfree(packet);
+ return -ENOMEM;
+ }
+
+ if (ether_addr_equal(udh.eth.smac_h, udh.eth.dmac_h))
+ packet->tx_dest = QED_ROCE_LL2_TX_DEST_NW;
+ else
+ packet->tx_dest = QED_ROCE_LL2_TX_DEST_LB;
+
+ packet->roce_mode = roce_mode;
+ memcpy(packet->header.vaddr, ud_header_buffer, header_size);
+ packet->header.len = header_size;
+ packet->n_seg = swr->num_sge;
+ for (i = 0; i < packet->n_seg; i++) {
+ packet->payload[i].baddr = swr->sg_list[i].addr;
+ packet->payload[i].len = swr->sg_list[i].length;
+ }
+
+ *p_packet = packet;
+
+ return 0;
+}
+
+int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr)
+{
+ struct qed_roce_ll2_packet *pkt = NULL;
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_roce_ll2_tx_params params;
+ struct qedr_dev *dev = qp->dev;
+ unsigned long flags;
+ int rc;
+
+ if (qp->state != QED_ROCE_QP_STATE_RTS) {
+ *bad_wr = wr;
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTS\n",
+ qp->state);
+ return -EINVAL;
+ }
+
+ if (wr->num_sge > RDMA_MAX_SGE_PER_SQ_WQE) {
+ DP_ERR(dev, "gsi post send: num_sge is too large (%d>%d)\n",
+ wr->num_sge, RDMA_MAX_SGE_PER_SQ_WQE);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ if (wr->opcode != IB_WR_SEND) {
+ DP_ERR(dev,
+ "gsi post send: failed due to unsupported opcode %d\n",
+ wr->opcode);
+ rc = -EINVAL;
+ goto err;
+ }
+
+ memset(¶ms, 0, sizeof(params));
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ rc = qedr_gsi_build_packet(dev, qp, wr, &pkt);
+ if (rc) {
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ goto err;
+ }
+
+ if (unlikely((dev->dp_level <= QED_LEVEL_VERBOSE) &&
+ (dev->dp_module & QEDR_MSG_GSI)))
+ qedr_print_hex_dump("TX CM Header ", pkt->header.vaddr,
+ pkt->header.len);
+
+ rc = dev->ops->roce_ll2_tx(dev->cdev, pkt, ¶ms);
+ if (!rc) {
+ qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id;
+ qedr_inc_sw_prod(&qp->sq);
+ DP_VERBOSE(qp->dev, QEDR_MSG_GSI,
+ "gsi post send: opcode=%d, in_irq=%ld, irqs_disabled=%d, wr_id=%llx\n",
+ wr->opcode, in_irq(), irqs_disabled(), wr->wr_id);
+ } else {
+ if (rc == QED_ROCE_TX_HEAD_FAILURE) {
+ /* TX failed while posting header - release resources */
+ dma_free_coherent(&dev->pdev->dev, pkt->header.len,
+ pkt->header.vaddr, pkt->header.baddr);
+ kfree(pkt);
+ } else if (rc == QED_ROCE_TX_FRAG_FAILURE) {
+ /* NTD since TX failed while posting a fragment. We will
+ * release the resources on TX callback
+ */
+ }
+
+ DP_ERR(dev, "gsi post send: failed to transmit (rc=%d)\n", rc);
+ rc = -EAGAIN;
+ *bad_wr = wr;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ if (wr->next) {
+ DP_ERR(dev,
+ "gsi post send: failed second WR. Only one WR may be passed at a time\n");
+ *bad_wr = wr->next;
+ rc = -EINVAL;
+ }
+
+ return rc;
+
+err:
+ *bad_wr = wr;
+ return rc;
+}
+
+int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibqp->device);
+ struct qedr_qp *qp = get_qedr_qp(ibqp);
+ struct qed_roce_ll2_buffer buf;
+ unsigned long flags;
+ int status = 0;
+ int rc;
+
+ if ((qp->state != QED_ROCE_QP_STATE_RTR) &&
+ (qp->state != QED_ROCE_QP_STATE_RTS)) {
+ *bad_wr = wr;
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. state is %d and not QED_ROCE_QP_STATE_RTR/S\n",
+ qp->state);
+ return -EINVAL;
+ }
+
+ memset(&buf, 0, sizeof(buf));
+
+ spin_lock_irqsave(&qp->q_lock, flags);
+
+ while (wr) {
+ if (wr->num_sge > QEDR_GSI_MAX_RECV_SGE) {
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer. too many sges %d>%d\n",
+ wr->num_sge, QEDR_GSI_MAX_RECV_SGE);
+ goto err;
+ }
+
+ buf.baddr = wr->sg_list[0].addr;
+ buf.len = wr->sg_list[0].length;
+
+ rc = dev->ops->roce_ll2_post_rx_buffer(dev->cdev, &buf, 0, 1);
+ if (rc) {
+ DP_ERR(dev,
+ "gsi post recv: failed to post rx buffer (rc=%d)\n",
+ rc);
+ goto err;
+ }
+
+ memset(&qp->rqe_wr_id[qp->rq.prod], 0,
+ sizeof(qp->rqe_wr_id[qp->rq.prod]));
+ qp->rqe_wr_id[qp->rq.prod].sg_list[0] = wr->sg_list[0];
+ qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id;
+
+ qedr_inc_sw_prod(&qp->rq);
+
+ wr = wr->next;
+ }
+
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+
+ return status;
+err:
+ spin_unlock_irqrestore(&qp->q_lock, flags);
+ *bad_wr = wr;
+ return -ENOMEM;
+}
+
+int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibcq->device);
+ struct qedr_cq *cq = get_qedr_cq(ibcq);
+ struct qedr_qp *qp = dev->gsi_qp;
+ unsigned long flags;
+ int i = 0;
+
+ spin_lock_irqsave(&cq->cq_lock, flags);
+
+ while (i < num_entries && qp->rq.cons != qp->rq.gsi_cons) {
+ memset(&wc[i], 0, sizeof(*wc));
+
+ wc[i].qp = &qp->ibqp;
+ wc[i].wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
+ wc[i].opcode = IB_WC_RECV;
+ wc[i].pkey_index = 0;
+ wc[i].status = (qp->rqe_wr_id[qp->rq.cons].rc) ?
+ IB_WC_GENERAL_ERR : IB_WC_SUCCESS;
+ /* 0 - currently only one recv sg is supported */
+ wc[i].byte_len = qp->rqe_wr_id[qp->rq.cons].sg_list[0].length;
+ wc[i].wc_flags |= IB_WC_GRH | IB_WC_IP_CSUM_OK;
+ ether_addr_copy(wc[i].smac, qp->rqe_wr_id[qp->rq.cons].smac);
+ wc[i].wc_flags |= IB_WC_WITH_SMAC;
+ if (qp->rqe_wr_id[qp->rq.cons].vlan_id) {
+ wc[i].wc_flags |= IB_WC_WITH_VLAN;
+ wc[i].vlan_id = qp->rqe_wr_id[qp->rq.cons].vlan_id;
+ }
+
+ qedr_inc_sw_cons(&qp->rq);
+ i++;
+ }
+
+ while (i < num_entries && qp->sq.cons != qp->sq.gsi_cons) {
+ memset(&wc[i], 0, sizeof(*wc));
+
+ wc[i].qp = &qp->ibqp;
+ wc[i].wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id;
+ wc[i].opcode = IB_WC_SEND;
+ wc[i].status = IB_WC_SUCCESS;
+
+ qedr_inc_sw_cons(&qp->sq);
+ i++;
+ }
+
+ spin_unlock_irqrestore(&cq->cq_lock, flags);
+
+ DP_VERBOSE(dev, (QEDR_MSG_CQ | QEDR_MSG_GSI),
+ "gsi poll_cq: requested entries=%d, actual=%d, qp->rq.cons=%d, qp->rq.gsi_cons=%x, qp->sq.cons=%d, qp->sq.gsi_cons=%d, qp_num=%d\n",
+ num_entries, i, qp->rq.cons, qp->rq.gsi_cons, qp->sq.cons,
+ qp->sq.gsi_cons, qp->ibqp.qp_num);
+
+ return i;
+}
diff --git a/drivers/infiniband/hw/qedr/qedr_cm.h b/drivers/infiniband/hw/qedr/qedr_cm.h
index b8a8b76..9ba6e15 100644
--- a/drivers/infiniband/hw/qedr/qedr_cm.h
+++ b/drivers/infiniband/hw/qedr/qedr_cm.h
@@ -32,9 +32,30 @@
#ifndef LINUX_QEDR_CM_H_
#define LINUX_QEDR_CM_H_
+#define QEDR_GSI_MAX_RECV_WR (4096)
+#define QEDR_GSI_MAX_SEND_WR (4096)
+
+#define QEDR_GSI_MAX_RECV_SGE (1) /* LL2 FW limitation */
+
+#define ETH_P_ROCE (0x8915)
+#define QEDR_ROCE_V2_UDP_SPORT (0000)
+
static inline u32 qedr_get_ipv4_from_gid(u8 *gid)
{
return *(u32 *)(void *)&gid[12];
}
+/* RDMA CM */
+int qedr_gsi_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
+int qedr_gsi_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr);
+int qedr_gsi_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+ struct ib_send_wr **bad_wr);
+struct ib_qp *qedr_create_gsi_qp(struct qedr_dev *dev,
+ struct ib_qp_init_attr *attrs,
+ struct qedr_qp *qp);
+void qedr_store_gsi_qp_cq(struct qedr_dev *dev,
+ struct qedr_qp *qp, struct ib_qp_init_attr *attrs);
+int qedr_destroy_gsi_qp(struct qedr_dev *dev);
+void qedr_inc_sw_gsi_cons(struct qedr_qp_hwq_info *info);
#endif
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 979555e..9381bc6 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -1471,6 +1471,15 @@ struct ib_qp *qedr_create_qp(struct ib_pd *ibpd,
qedr_set_qp_init_params(dev, qp, pd, attrs);
+ if (attrs->qp_type == IB_QPT_GSI) {
+ if (udata) {
+ DP_ERR(dev,
+ "create qp: unexpected udata when creating GSI QP\n");
+ goto err0;
+ }
+ return qedr_create_gsi_qp(dev, attrs, qp);
+ }
+
memset(&in_params, 0, sizeof(in_params));
if (udata) {
@@ -2040,6 +2049,8 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
rc = dev->ops->rdma_destroy_qp(dev->rdma_ctx, qp->qed_qp);
if (rc)
return rc;
+ } else {
+ qedr_destroy_gsi_qp(dev);
}
if (ibqp->uobject && ibqp->uobject->context) {
@@ -2959,6 +2970,10 @@ int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
*bad_wr = NULL;
+ if (qp->qp_type == IB_QPT_GSI) {
+ return qedr_gsi_post_send(ibqp, wr, bad_wr);
+ }
+
spin_lock_irqsave(&qp->q_lock, flags);
if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
@@ -3015,6 +3030,10 @@ int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
unsigned long flags;
int status = 0;
+ if (qp->qp_type == IB_QPT_GSI) {
+ return qedr_gsi_post_recv(ibqp, wr, bad_wr);
+ }
+
spin_lock_irqsave(&qp->q_lock, flags);
if ((qp->state == QED_ROCE_QP_STATE_RESET) ||
@@ -3450,6 +3469,9 @@ int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
int update = 0;
int done = 0;
+ if (cq->cq_type == QEDR_CQ_TYPE_GSI)
+ return qedr_gsi_poll_cq(ibcq, num_entries, wc);
+
spin_lock_irqsave(&cq->cq_lock, flags);
old_cons = qed_chain_get_cons_idx_u32(&cq->pbl);
while (num_entries && is_valid_cqe(cq, cqe)) {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
index 4ea6a07..f907569 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
@@ -278,6 +278,7 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
struct qed_ll2_tx_packet *p_pkt = NULL;
struct qed_ll2_info *p_ll2_conn;
struct qed_ll2_tx_queue *p_tx;
+ dma_addr_t tx_frag;
p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
if (!p_ll2_conn)
@@ -294,15 +295,25 @@ static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
list_del(&p_pkt->list_entry);
b_last_packet = list_empty(&p_tx->active_descq);
list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+
p_tx->cur_completing_packet = *p_pkt;
p_tx->cur_completing_bd_idx = 1;
b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
-
- qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
- p_pkt->cookie,
- p_pkt->bds_set[0].tx_frag,
- b_last_frag,
- b_last_packet);
+ tx_frag = p_pkt->bds_set[0].tx_frag;
+ if (p_ll2_conn->gsi_enable)
+ qed_ll2b_release_tx_gsi_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag,
+ b_last_packet);
+ else
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag,
+ b_last_packet);
}
}
@@ -314,6 +325,7 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
struct qed_ll2_tx_packet *p_pkt;
bool b_last_frag = false;
unsigned long flags;
+ dma_addr_t tx_frag;
int rc = -EINVAL;
spin_lock_irqsave(&p_tx->lock, flags);
@@ -354,11 +366,19 @@ static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
spin_unlock_irqrestore(&p_tx->lock, flags);
- qed_ll2b_complete_tx_packet(p_hwfn,
- p_ll2_conn->my_id,
- p_pkt->cookie,
- p_pkt->bds_set[0].tx_frag,
- b_last_frag, !num_bds);
+ tx_frag = p_pkt->bds_set[0].tx_frag;
+ if (p_ll2_conn->gsi_enable)
+ qed_ll2b_complete_tx_gsi_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag, !num_bds);
+ else
+ qed_ll2b_complete_tx_packet(p_hwfn,
+ p_ll2_conn->my_id,
+ p_pkt->cookie,
+ tx_frag,
+ b_last_frag, !num_bds);
spin_lock_irqsave(&p_tx->lock, flags);
}
@@ -369,6 +389,54 @@ out:
return rc;
}
+static int
+qed_ll2_rxq_completion_gsi(struct qed_hwfn *p_hwfn,
+ struct qed_ll2_info *p_ll2_info,
+ union core_rx_cqe_union *p_cqe,
+ unsigned long lock_flags, bool b_last_cqe)
+{
+ struct qed_ll2_rx_queue *p_rx = &p_ll2_info->rx_queue;
+ struct qed_ll2_rx_packet *p_pkt = NULL;
+ u16 packet_length, parse_flags, vlan;
+ u32 src_mac_addrhi;
+ u16 src_mac_addrlo;
+
+ if (!list_empty(&p_rx->active_descq))
+ p_pkt = list_first_entry(&p_rx->active_descq,
+ struct qed_ll2_rx_packet, list_entry);
+ if (!p_pkt) {
+ DP_NOTICE(p_hwfn,
+ "GSI Rx completion but active_descq is empty\n");
+ return -EIO;
+ }
+
+ list_del(&p_pkt->list_entry);
+ parse_flags = le16_to_cpu(p_cqe->rx_cqe_gsi.parse_flags.flags);
+ packet_length = le16_to_cpu(p_cqe->rx_cqe_gsi.data_length);
+ vlan = le16_to_cpu(p_cqe->rx_cqe_gsi.vlan);
+ src_mac_addrhi = le32_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrhi);
+ src_mac_addrlo = le16_to_cpu(p_cqe->rx_cqe_gsi.src_mac_addrlo);
+ if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
+ DP_NOTICE(p_hwfn,
+ "Mismatch between active_descq and the LL2 Rx chain\n");
+ list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+ spin_unlock_irqrestore(&p_rx->lock, lock_flags);
+ qed_ll2b_complete_rx_gsi_packet(p_hwfn,
+ p_ll2_info->my_id,
+ p_pkt->cookie,
+ p_pkt->rx_buf_addr,
+ packet_length,
+ p_cqe->rx_cqe_gsi.data_length_error,
+ parse_flags,
+ vlan,
+ src_mac_addrhi,
+ src_mac_addrlo, b_last_cqe);
+ spin_lock_irqsave(&p_rx->lock, lock_flags);
+
+ return 0;
+}
+
static int qed_ll2_rxq_completion_reg(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_conn,
union core_rx_cqe_union *p_cqe,
@@ -430,6 +498,10 @@ static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
DP_NOTICE(p_hwfn, "LL2 - unexpected Rx CQE slowpath\n");
rc = -EINVAL;
break;
+ case CORE_RX_CQE_TYPE_GSI_OFFLOAD:
+ rc = qed_ll2_rxq_completion_gsi(p_hwfn, p_ll2_conn,
+ cqe, flags, b_last_cqe);
+ break;
case CORE_RX_CQE_TYPE_REGULAR:
rc = qed_ll2_rxq_completion_reg(p_hwfn, p_ll2_conn,
cqe, flags, b_last_cqe);
@@ -527,6 +599,7 @@ static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
}
p_ramrod->action_on_error.error_type = action_on_error;
+ p_ramrod->gsi_offload_flag = p_ll2_conn->gsi_enable;
return qed_spq_post(p_hwfn, p_ent, NULL);
}
@@ -589,6 +662,7 @@ static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
DP_NOTICE(p_hwfn, "Unknown connection type: %d\n", conn_type);
}
+ p_ramrod->gsi_offload_flag = p_ll2_conn->gsi_enable;
return qed_spq_post(p_hwfn, p_ent, NULL);
}
@@ -775,6 +849,7 @@ int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
p_ll2_info->tx_dest = p_params->tx_dest;
p_ll2_info->ai_err_packet_too_big = p_params->ai_err_packet_too_big;
p_ll2_info->ai_err_no_buf = p_params->ai_err_no_buf;
+ p_ll2_info->gsi_enable = p_params->gsi_enable;
rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info, rx_num_desc);
if (rc)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
index b3783fd..80a5dc2 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
@@ -125,6 +125,7 @@ struct qed_ll2_info {
u8 tx_stats_en;
struct qed_ll2_rx_queue rx_queue;
struct qed_ll2_tx_queue tx_queue;
+ u8 gsi_enable;
};
/**
@@ -292,5 +293,24 @@ void qed_ll2_setup(struct qed_hwfn *p_hwfn,
*/
void qed_ll2_free(struct qed_hwfn *p_hwfn,
struct qed_ll2_info *p_ll2_connections);
-
+void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ u16 data_length,
+ u8 data_length_error,
+ u16 parse_flags,
+ u16 vlan,
+ u32 src_mac_addr_hi,
+ u16 src_mac_addr_lo, bool b_last_packet);
+void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet);
+void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet);
#endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
index 27095f0..cf31a48 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_roce.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_roce.c
@@ -2633,6 +2633,71 @@ void qed_rdma_remove_user(void *rdma_cxt, u16 dpi)
spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
}
+void qed_ll2b_complete_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet)
+{
+ struct qed_roce_ll2_packet *packet = cookie;
+ struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
+
+ roce_ll2->cbs.tx_cb(roce_ll2->cb_cookie, packet);
+}
+
+void qed_ll2b_release_tx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t first_frag_addr,
+ bool b_last_fragment, bool b_last_packet)
+{
+ qed_ll2b_complete_tx_gsi_packet(p_hwfn, connection_handle,
+ cookie, first_frag_addr,
+ b_last_fragment, b_last_packet);
+}
+
+void qed_ll2b_complete_rx_gsi_packet(struct qed_hwfn *p_hwfn,
+ u8 connection_handle,
+ void *cookie,
+ dma_addr_t rx_buf_addr,
+ u16 data_length,
+ u8 data_length_error,
+ u16 parse_flags,
+ u16 vlan,
+ u32 src_mac_addr_hi,
+ u16 src_mac_addr_lo, bool b_last_packet)
+{
+ struct qed_roce_ll2_info *roce_ll2 = p_hwfn->ll2;
+ struct qed_roce_ll2_rx_params params;
+ struct qed_dev *cdev = p_hwfn->cdev;
+ struct qed_roce_ll2_packet pkt;
+
+ DP_VERBOSE(cdev,
+ QED_MSG_LL2,
+ "roce ll2 rx complete: bus_addr=%p, len=%d, data_len_err=%d\n",
+ (void *)(uintptr_t)rx_buf_addr,
+ data_length, data_length_error);
+
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.n_seg = 1;
+ pkt.payload[0].baddr = rx_buf_addr;
+ pkt.payload[0].len = data_length;
+
+ memset(¶ms, 0, sizeof(params));
+ params.vlan_id = vlan;
+ *((u32 *)¶ms.smac[0]) = ntohl(src_mac_addr_hi);
+ *((u16 *)¶ms.smac[4]) = ntohs(src_mac_addr_lo);
+
+ if (data_length_error) {
+ DP_ERR(cdev,
+ "roce ll2 rx complete: data length error %d, length=%d\n",
+ data_length_error, data_length);
+ params.rc = -EINVAL;
+ }
+
+ roce_ll2->cbs.rx_cb(roce_ll2->cb_cookie, &pkt, ¶ms);
+}
+
static int qed_roce_ll2_set_mac_filter(struct qed_dev *cdev,
u8 *old_mac_address,
u8 *new_mac_address)
@@ -2717,6 +2782,7 @@ static int qed_roce_ll2_start(struct qed_dev *cdev,
ll2_params.tx_dest = CORE_TX_DEST_NW;
ll2_params.ai_err_packet_too_big = LL2_DROP_PACKET;
ll2_params.ai_err_no_buf = LL2_DROP_PACKET;
+ ll2_params.gsi_enable = true;
rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_params,
params->max_rx_buffers,
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 11/11] Add events support and register IB device
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:07 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add error handling support.
Register ib device with ib stack.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 107 ++++++++++++++++++++++++++++++++++++-
drivers/infiniband/hw/qedr/verbs.c | 37 +++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 9 ++++
include/linux/qed/common_hsi.h | 1 +
4 files changed, 152 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index fb0b76c..242f31d 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -86,10 +86,25 @@ static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str,
(fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
}
+static struct net_device *qedr_get_netdev(struct ib_device *dev, u8 port_num)
+{
+ struct qedr_dev *qdev;
+
+ qdev = get_qedr_dev(dev);
+ dev_hold(qdev->ndev);
+
+ /* The HW vendor's device driver must guarantee
+ * that this function returns NULL before the net device reaches
+ * NETDEV_UNREGISTER_FINAL state.
+ */
+ return qdev->ndev;
+}
+
static int qedr_register_device(struct qedr_dev *dev)
{
strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
+ dev->ibdev.node_guid = dev->attr.node_guid;
memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
dev->ibdev.owner = THIS_MODULE;
dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
@@ -156,11 +171,15 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.post_send = qedr_post_send;
dev->ibdev.post_recv = qedr_post_recv;
+ dev->ibdev.process_mad = qedr_process_mad;
+ dev->ibdev.get_port_immutable = qedr_port_immutable;
+ dev->ibdev.get_netdev = qedr_get_netdev;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
- return 0;
+ return ib_register_device(&dev->ibdev, NULL);
}
/* This function allocates fast-path status block memory */
@@ -548,6 +567,86 @@ static int qedr_set_device_attr(struct qedr_dev *dev)
return 0;
}
+void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
+{
+#define EVENT_TYPE_NOT_DEFINED 0
+#define EVENT_TYPE_CQ 1
+#define EVENT_TYPE_QP 2
+ struct qedr_dev *dev = (struct qedr_dev *)context;
+ union event_ring_data *data = fw_handle;
+ u64 roce_handle64 = ((u64)data->roce_handle.hi << 32) +
+ data->roce_handle.lo;
+ u8 event_type = EVENT_TYPE_NOT_DEFINED;
+ struct ib_event event;
+ struct ib_cq *ibcq;
+ struct ib_qp *ibqp;
+ struct qedr_cq *cq;
+ struct qedr_qp *qp;
+
+ switch (e_code) {
+ case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
+ event.event = IB_EVENT_CQ_ERR;
+ event_type = EVENT_TYPE_CQ;
+ break;
+ case ROCE_ASYNC_EVENT_SQ_DRAINED:
+ event.event = IB_EVENT_SQ_DRAINED;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
+ event.event = IB_EVENT_QP_FATAL;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
+ event.event = IB_EVENT_QP_REQ_ERR;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
+ event.event = IB_EVENT_QP_ACCESS_ERR;
+ event_type = EVENT_TYPE_QP;
+ break;
+ default:
+ DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
+ roce_handle64);
+ }
+
+ switch (event_type) {
+ case EVENT_TYPE_CQ:
+ cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
+ if (cq) {
+ ibcq = &cq->ibcq;
+ if (ibcq->event_handler) {
+ event.device = ibcq->device;
+ event.element.cq = ibcq;
+ ibcq->event_handler(&event, ibcq->cq_context);
+ }
+ } else {
+ WARN(1,
+ "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
+ roce_handle64);
+ }
+ DP_ERR(dev, "CQ event %d on hanlde %p\n", e_code, cq);
+ break;
+ case EVENT_TYPE_QP:
+ qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
+ if (qp) {
+ ibqp = &qp->ibqp;
+ if (ibqp->event_handler) {
+ event.device = ibqp->device;
+ event.element.qp = ibqp;
+ ibqp->event_handler(&event, ibqp->qp_context);
+ }
+ } else {
+ WARN(1,
+ "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
+ roce_handle64);
+ }
+ DP_ERR(dev, "QP event %d on hanlde %p\n", e_code, qp);
+ break;
+ default:
+ break;
+ }
+}
+
static int qedr_init_hw(struct qedr_dev *dev)
{
struct qed_rdma_add_user_out_params out_params;
@@ -576,6 +675,7 @@ static int qedr_init_hw(struct qedr_dev *dev)
cur_pbl->pbl_ptr = (u64)p_phys_table;
}
+ events.affiliated_event = qedr_affiliated_event;
events.context = dev;
in_params->events = &events;
@@ -675,7 +775,7 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
- goto reg_err;
+ goto sysfs_err;
spin_lock(&qedr_devlist_lock);
list_add_tail_rcu(&dev->entry, &qedr_dev_list);
@@ -684,6 +784,8 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
return dev;
+sysfs_err:
+ ib_unregister_device(&dev->ibdev);
reg_err:
qedr_sync_free_irqs(dev);
irq_err:
@@ -703,6 +805,7 @@ static void qedr_remove(struct qedr_dev *dev)
* of the registered clients.
*/
qedr_remove_sysfiles(dev);
+ ib_unregister_device(&dev->ibdev);
spin_lock(&qedr_devlist_lock);
list_del_rcu(&dev->entry);
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 9381bc6..6055e27 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -3523,3 +3523,40 @@ int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
spin_unlock_irqrestore(&cq->cq_lock, flags);
return done;
}
+
+int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags,
+ u8 port_num,
+ const struct ib_wc *in_wc,
+ const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *mad_hdr,
+ size_t in_mad_size, struct ib_mad_hdr *out_mad,
+ size_t *out_mad_size, u16 *out_mad_pkey_index)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+
+ DP_VERBOSE(dev, QEDR_MSG_GSI,
+ "QEDR_PROCESS_MAD in_mad %x %x %x %x %x %x %x %x\n",
+ mad_hdr->attr_id, mad_hdr->base_version, mad_hdr->attr_mod,
+ mad_hdr->class_specific, mad_hdr->class_version,
+ mad_hdr->method, mad_hdr->mgmt_class, mad_hdr->status);
+ return IB_MAD_RESULT_SUCCESS;
+}
+
+int qedr_port_immutable(struct ib_device *ibdev, u8 port_num,
+ struct ib_port_immutable *immutable)
+{
+ struct ib_port_attr attr;
+ int err;
+
+ err = qedr_query_port(ibdev, port_num, &attr);
+ if (err)
+ return err;
+
+ immutable->pkey_tbl_len = attr.pkey_tbl_len;
+ immutable->gid_tbl_len = attr.gid_tbl_len;
+ immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
+ RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
+ immutable->max_mad_size = IB_MGMT_MAD_SIZE;
+
+ return 0;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index 9eb2722b..da4535e 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -63,4 +63,13 @@ int qedr_post_send(struct ib_qp *, struct ib_send_wr *,
struct ib_send_wr **bad_wr);
int qedr_post_recv(struct ib_qp *, struct ib_recv_wr *,
struct ib_recv_wr **bad_wr);
+int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags,
+ u8 port_num, const struct ib_wc *in_wc,
+ const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *in_mad,
+ size_t in_mad_size, struct ib_mad_hdr *out_mad,
+ size_t *out_mad_size, u16 *out_mad_pkey_index);
+
+int qedr_port_immutable(struct ib_device *ibdev, u8 port_num,
+ struct ib_port_immutable *immutable);
#endif
diff --git a/include/linux/qed/common_hsi.h b/include/linux/qed/common_hsi.h
index 70b30e4..83ce645 100644
--- a/include/linux/qed/common_hsi.h
+++ b/include/linux/qed/common_hsi.h
@@ -671,6 +671,7 @@ union event_ring_data {
struct iscsi_eqe_data iscsi_info;
struct malicious_vf_eqe_data malicious_vf;
struct initial_cleanup_eqe_data vf_init_cleanup;
+ struct regpair roce_handle;
};
/* Event Ring Entry */
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* [RFC 11/11] Add events support and register IB device
@ 2016-09-12 16:07 ` Ram Amrani
0 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-12 16:07 UTC (permalink / raw)
To: dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, Ram Amrani
Add error handling support.
Register ib device with ib stack.
Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
---
drivers/infiniband/hw/qedr/main.c | 107 ++++++++++++++++++++++++++++++++++++-
drivers/infiniband/hw/qedr/verbs.c | 37 +++++++++++++
drivers/infiniband/hw/qedr/verbs.h | 9 ++++
include/linux/qed/common_hsi.h | 1 +
4 files changed, 152 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
index fb0b76c..242f31d 100644
--- a/drivers/infiniband/hw/qedr/main.c
+++ b/drivers/infiniband/hw/qedr/main.c
@@ -86,10 +86,25 @@ static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str,
(fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
}
+static struct net_device *qedr_get_netdev(struct ib_device *dev, u8 port_num)
+{
+ struct qedr_dev *qdev;
+
+ qdev = get_qedr_dev(dev);
+ dev_hold(qdev->ndev);
+
+ /* The HW vendor's device driver must guarantee
+ * that this function returns NULL before the net device reaches
+ * NETDEV_UNREGISTER_FINAL state.
+ */
+ return qdev->ndev;
+}
+
static int qedr_register_device(struct qedr_dev *dev)
{
strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
+ dev->ibdev.node_guid = dev->attr.node_guid;
memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
dev->ibdev.owner = THIS_MODULE;
dev->ibdev.uverbs_abi_ver = QEDR_ABI_VERSION;
@@ -156,11 +171,15 @@ static int qedr_register_device(struct qedr_dev *dev)
dev->ibdev.post_send = qedr_post_send;
dev->ibdev.post_recv = qedr_post_recv;
+ dev->ibdev.process_mad = qedr_process_mad;
+ dev->ibdev.get_port_immutable = qedr_port_immutable;
+ dev->ibdev.get_netdev = qedr_get_netdev;
+
dev->ibdev.dma_device = &dev->pdev->dev;
dev->ibdev.get_link_layer = qedr_link_layer;
dev->ibdev.get_dev_fw_str = qedr_get_dev_fw_str;
- return 0;
+ return ib_register_device(&dev->ibdev, NULL);
}
/* This function allocates fast-path status block memory */
@@ -548,6 +567,86 @@ static int qedr_set_device_attr(struct qedr_dev *dev)
return 0;
}
+void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
+{
+#define EVENT_TYPE_NOT_DEFINED 0
+#define EVENT_TYPE_CQ 1
+#define EVENT_TYPE_QP 2
+ struct qedr_dev *dev = (struct qedr_dev *)context;
+ union event_ring_data *data = fw_handle;
+ u64 roce_handle64 = ((u64)data->roce_handle.hi << 32) +
+ data->roce_handle.lo;
+ u8 event_type = EVENT_TYPE_NOT_DEFINED;
+ struct ib_event event;
+ struct ib_cq *ibcq;
+ struct ib_qp *ibqp;
+ struct qedr_cq *cq;
+ struct qedr_qp *qp;
+
+ switch (e_code) {
+ case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
+ event.event = IB_EVENT_CQ_ERR;
+ event_type = EVENT_TYPE_CQ;
+ break;
+ case ROCE_ASYNC_EVENT_SQ_DRAINED:
+ event.event = IB_EVENT_SQ_DRAINED;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
+ event.event = IB_EVENT_QP_FATAL;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
+ event.event = IB_EVENT_QP_REQ_ERR;
+ event_type = EVENT_TYPE_QP;
+ break;
+ case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
+ event.event = IB_EVENT_QP_ACCESS_ERR;
+ event_type = EVENT_TYPE_QP;
+ break;
+ default:
+ DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
+ roce_handle64);
+ }
+
+ switch (event_type) {
+ case EVENT_TYPE_CQ:
+ cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
+ if (cq) {
+ ibcq = &cq->ibcq;
+ if (ibcq->event_handler) {
+ event.device = ibcq->device;
+ event.element.cq = ibcq;
+ ibcq->event_handler(&event, ibcq->cq_context);
+ }
+ } else {
+ WARN(1,
+ "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
+ roce_handle64);
+ }
+ DP_ERR(dev, "CQ event %d on hanlde %p\n", e_code, cq);
+ break;
+ case EVENT_TYPE_QP:
+ qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
+ if (qp) {
+ ibqp = &qp->ibqp;
+ if (ibqp->event_handler) {
+ event.device = ibqp->device;
+ event.element.qp = ibqp;
+ ibqp->event_handler(&event, ibqp->qp_context);
+ }
+ } else {
+ WARN(1,
+ "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
+ roce_handle64);
+ }
+ DP_ERR(dev, "QP event %d on hanlde %p\n", e_code, qp);
+ break;
+ default:
+ break;
+ }
+}
+
static int qedr_init_hw(struct qedr_dev *dev)
{
struct qed_rdma_add_user_out_params out_params;
@@ -576,6 +675,7 @@ static int qedr_init_hw(struct qedr_dev *dev)
cur_pbl->pbl_ptr = (u64)p_phys_table;
}
+ events.affiliated_event = qedr_affiliated_event;
events.context = dev;
in_params->events = &events;
@@ -675,7 +775,7 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
- goto reg_err;
+ goto sysfs_err;
spin_lock(&qedr_devlist_lock);
list_add_tail_rcu(&dev->entry, &qedr_dev_list);
@@ -684,6 +784,8 @@ static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
return dev;
+sysfs_err:
+ ib_unregister_device(&dev->ibdev);
reg_err:
qedr_sync_free_irqs(dev);
irq_err:
@@ -703,6 +805,7 @@ static void qedr_remove(struct qedr_dev *dev)
* of the registered clients.
*/
qedr_remove_sysfiles(dev);
+ ib_unregister_device(&dev->ibdev);
spin_lock(&qedr_devlist_lock);
list_del_rcu(&dev->entry);
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 9381bc6..6055e27 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -3523,3 +3523,40 @@ int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
spin_unlock_irqrestore(&cq->cq_lock, flags);
return done;
}
+
+int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags,
+ u8 port_num,
+ const struct ib_wc *in_wc,
+ const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *mad_hdr,
+ size_t in_mad_size, struct ib_mad_hdr *out_mad,
+ size_t *out_mad_size, u16 *out_mad_pkey_index)
+{
+ struct qedr_dev *dev = get_qedr_dev(ibdev);
+
+ DP_VERBOSE(dev, QEDR_MSG_GSI,
+ "QEDR_PROCESS_MAD in_mad %x %x %x %x %x %x %x %x\n",
+ mad_hdr->attr_id, mad_hdr->base_version, mad_hdr->attr_mod,
+ mad_hdr->class_specific, mad_hdr->class_version,
+ mad_hdr->method, mad_hdr->mgmt_class, mad_hdr->status);
+ return IB_MAD_RESULT_SUCCESS;
+}
+
+int qedr_port_immutable(struct ib_device *ibdev, u8 port_num,
+ struct ib_port_immutable *immutable)
+{
+ struct ib_port_attr attr;
+ int err;
+
+ err = qedr_query_port(ibdev, port_num, &attr);
+ if (err)
+ return err;
+
+ immutable->pkey_tbl_len = attr.pkey_tbl_len;
+ immutable->gid_tbl_len = attr.gid_tbl_len;
+ immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
+ RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
+ immutable->max_mad_size = IB_MGMT_MAD_SIZE;
+
+ return 0;
+}
diff --git a/drivers/infiniband/hw/qedr/verbs.h b/drivers/infiniband/hw/qedr/verbs.h
index 9eb2722b..da4535e 100644
--- a/drivers/infiniband/hw/qedr/verbs.h
+++ b/drivers/infiniband/hw/qedr/verbs.h
@@ -63,4 +63,13 @@ int qedr_post_send(struct ib_qp *, struct ib_send_wr *,
struct ib_send_wr **bad_wr);
int qedr_post_recv(struct ib_qp *, struct ib_recv_wr *,
struct ib_recv_wr **bad_wr);
+int qedr_process_mad(struct ib_device *ibdev, int process_mad_flags,
+ u8 port_num, const struct ib_wc *in_wc,
+ const struct ib_grh *in_grh,
+ const struct ib_mad_hdr *in_mad,
+ size_t in_mad_size, struct ib_mad_hdr *out_mad,
+ size_t *out_mad_size, u16 *out_mad_pkey_index);
+
+int qedr_port_immutable(struct ib_device *ibdev, u8 port_num,
+ struct ib_port_immutable *immutable);
#endif
diff --git a/include/linux/qed/common_hsi.h b/include/linux/qed/common_hsi.h
index 70b30e4..83ce645 100644
--- a/include/linux/qed/common_hsi.h
+++ b/include/linux/qed/common_hsi.h
@@ -671,6 +671,7 @@ union event_ring_data {
struct iscsi_eqe_data iscsi_info;
struct malicious_vf_eqe_data malicious_vf;
struct initial_cleanup_eqe_data vf_init_cleanup;
+ struct regpair roce_handle;
};
/* Event Ring Entry */
--
1.8.3.1
^ permalink raw reply related [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <1473696465-27986-1-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
` (2 preceding siblings ...)
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-12 16:39 ` Leon Romanovsky
[not found] ` <20160912163928.GK8812-2ukJVAZIZ/Y@public.gmane.org>
2016-09-12 18:05 ` Jason Gunthorpe
2016-09-13 14:23 ` Sagi Grimberg
5 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-12 16:39 UTC (permalink / raw)
To: Ram Amrani
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q,
Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
[-- Attachment #1: Type: text/plain, Size: 1004 bytes --]
On Mon, Sep 12, 2016 at 07:07:34PM +0300, Ram Amrani wrote:
> include/linux/qed/common_hsi.h | 1 +
> include/linux/qed/qed_if.h | 9 +-
> include/linux/qed/qed_ll2_if.h | 140 +
> include/linux/qed/qed_roce_if.h | 604 ++++
> include/linux/qed/qede_roce.h | 88 +
> include/linux/qed/rdma_common.h | 1 +
Something not directly related to your patches, but they brought my
attention to the fact that all these new (and old) rdma<->net devices
are polluting include/linux
Filtered output:
➜ linux-rdma git:(topic/fixes-for-4.8-2) ls -dl include/linux/*/
drwxrwxr-x 2 leonro leonro 4096 Aug 30 16:27 include/linux/hsi/
drwxrwxr-x 2 leonro leonro 4096 Sep 12 19:08 include/linux/mlx4/
drwxrwxr-x 2 leonro leonro 4096 Sep 7 15:31 include/linux/mlx5/
drwxrwxr-x 2 leonro leonro 4096 Sep 8 17:46 include/linux/qed/
Is this the right place for them?
Thanks
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <20160912163928.GK8812-2ukJVAZIZ/Y@public.gmane.org>
@ 2016-09-12 16:49 ` Parav Pandit
2016-09-12 17:39 ` Yuval Mintz
0 siblings, 1 reply; 78+ messages in thread
From: Parav Pandit @ 2016-09-12 16:49 UTC (permalink / raw)
To: Leon Romanovsky
Cc: Ram Amrani, Doug Ledford, davem-fT/PcQaiUtIeIZ0/mPfg9Q,
Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
On Mon, Sep 12, 2016 at 10:09 PM, Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
> On Mon, Sep 12, 2016 at 07:07:34PM +0300, Ram Amrani wrote:
>
>> include/linux/qed/common_hsi.h | 1 +
>> include/linux/qed/qed_if.h | 9 +-
>> include/linux/qed/qed_ll2_if.h | 140 +
>> include/linux/qed/qed_roce_if.h | 604 ++++
>> include/linux/qed/qede_roce.h | 88 +
>> include/linux/qed/rdma_common.h | 1 +
>
> Something not directly related to your patches, but they brought my
> attention to the fact that all these new (and old) rdma<->net devices
> are polluting include/linux
>
ocrdma driver includes be_roce.h located in net/ethernet/emulex/benet
location instead of include/linux/.
This file helps to bind rdma to net device or underlying hw device.
May be similar change can be done for rest of the drivers for
rdma<-->net devices?
> Filtered output:
> ➜ linux-rdma git:(topic/fixes-for-4.8-2) ls -dl include/linux/*/
> drwxrwxr-x 2 leonro leonro 4096 Aug 30 16:27 include/linux/hsi/
> drwxrwxr-x 2 leonro leonro 4096 Sep 12 19:08 include/linux/mlx4/
> drwxrwxr-x 2 leonro leonro 4096 Sep 7 15:31 include/linux/mlx5/
> drwxrwxr-x 2 leonro leonro 4096 Sep 8 17:46 include/linux/qed/
>
> Is this the right place for them?
>
> Thanks
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
2016-09-12 16:49 ` Parav Pandit
@ 2016-09-12 17:39 ` Yuval Mintz
[not found] ` <CY4PR11MB17202A542AC07CBA4A65E93697FF0-JNf6+SjKdlG0ooKL/ADlEpPPoyLQLiKMvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Yuval Mintz @ 2016-09-12 17:39 UTC (permalink / raw)
To: Parav Pandit, Leon Romanovsky
Cc: Ram Amrani, Doug Ledford, David Miller, Ariel Elior,
Michal Kalderon, Rajesh Borundia, linux-rdma, netdev
>>> include/linux/qed/common_hsi.h | 1 +
>>> include/linux/qed/qed_if.h | 9 +-
>>> include/linux/qed/qed_ll2_if.h | 140 +
>>> include/linux/qed/qed_roce_if.h | 604 ++++
>>> include/linux/qed/qede_roce.h | 88 +
>> > include/linux/qed/rdma_common.h | 1 +
>>
>> Something not directly related to your patches, but they brought my
>> attention to the fact that all these new (and old) rdma<->net devices
>> are polluting include/linux
>>
> ocrdma driver includes be_roce.h located in net/ethernet/emulex/benet
> location instead of include/linux/.
> This file helps to bind rdma to net device or underlying hw device.
> May be similar change can be done for rest of the drivers for
> rdma<-->net devices?
By adding explicit inclusion paths in the Makefile, a la
ccflags-y := -Idrivers/net/ethernet/emulex/benet ?
While this might work, I personally dislike it as I find it
counter-intuitive when going over the code -
I don't expect driver to locally modify the inclusion path.
Besides, we're going to [eventually] a whole suite of drivers based
on the qed module, some of which would reside under drivers/scsi;
Not sure it's best to have 3 or 4 different drivers privately include the
same directory under a different subsystem.
>> Filtered output:
>> ➜ linux-rdma git:(topic/fixes-for-4.8-2) ls -dl include/linux/*/
>> drwxrwxr-x 2 leonro leonro 4096 Aug 30 16:27 include/linux/hsi/
>> drwxrwxr-x 2 leonro leonro 4096 Sep 12 19:08 include/linux/mlx4/
>> drwxrwxr-x 2 leonro leonro 4096 Sep 7 15:31 include/linux/mlx5/
>> drwxrwxr-x 2 leonro leonro 4096 Sep 8 17:46 include/linux/qed/
>>
>> Is this the right place for them?
>
> Thanks
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <1473696465-27986-1-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
` (3 preceding siblings ...)
2016-09-12 16:39 ` [RFC 00/11] QLogic RDMA Driver (qedr) RFC Leon Romanovsky
@ 2016-09-12 18:05 ` Jason Gunthorpe
[not found] ` <20160912180508.GI5843-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
2016-09-13 14:23 ` Sagi Grimberg
5 siblings, 1 reply; 78+ messages in thread
From: Jason Gunthorpe @ 2016-09-12 18:05 UTC (permalink / raw)
To: Ram Amrani
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q,
Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
On Mon, Sep 12, 2016 at 07:07:34PM +0300, Ram Amrani wrote:
> drivers/infiniband/hw/qedr/main.c | 907 ++++++
> drivers/infiniband/hw/qedr/qedr.h | 494 ++++
> drivers/infiniband/hw/qedr/qedr_cm.c | 626 +++++
> drivers/infiniband/hw/qedr/qedr_cm.h | 61 +
> drivers/infiniband/hw/qedr/qedr_hsi.h | 56 +
> drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 748 +++++
> drivers/infiniband/hw/qedr/qedr_user.h | 80 +
We are requiring new uAPI headers are placed under include/uapi/rdma/,
please coordinate with Leon on the path.
Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <1473696465-27986-3-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-12 18:44 ` Mark Bloch
0 siblings, 0 replies; 78+ messages in thread
From: Mark Bloch @ 2016-09-12 18:44 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
Hi Ram,
Just a few thoughts
On 12/09/2016 19:07, Ram Amrani wrote:
> Adds a skeletal implementation of the qed* RoCE driver -
> basically the ability to communicate with the qede driver and
> receive notifications from it regarding various init/exit events.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> ---
> drivers/infiniband/Kconfig | 2 +
> drivers/infiniband/hw/Makefile | 1 +
> drivers/infiniband/hw/qedr/Kconfig | 7 +
> drivers/infiniband/hw/qedr/Makefile | 3 +
> drivers/infiniband/hw/qedr/main.c | 293 +++++++++++++++++++++++++
> drivers/infiniband/hw/qedr/qedr.h | 60 ++++++
> drivers/net/ethernet/qlogic/qede/Makefile | 1 +
> drivers/net/ethernet/qlogic/qede/qede.h | 9 +
> drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 +++++++++++++++++++++++++++
> include/linux/qed/qed_if.h | 3 +-
> include/linux/qed/qede_roce.h | 88 ++++++++
> include/uapi/linux/pci_regs.h | 3 +
> 13 files changed, 803 insertions(+), 11 deletions(-)
> create mode 100644 drivers/infiniband/hw/qedr/Kconfig
> create mode 100644 drivers/infiniband/hw/qedr/Makefile
> create mode 100644 drivers/infiniband/hw/qedr/main.c
> create mode 100644 drivers/infiniband/hw/qedr/qedr.h
> create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
> create mode 100644 include/linux/qed/qede_roce.h
[SNIP]
> +
> +MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
> +MODULE_AUTHOR("QLogic Corporation");
> +MODULE_LICENSE("Dual BSD/GPL");
> +MODULE_VERSION(QEDR_MODULE_VERSION);
> +
> +uint debug;
> +module_param(debug, uint, 0);
> +MODULE_PARM_DESC(debug, "Default debug msglevel");
Why are you adding this as a module parameter?
> +static LIST_HEAD(qedr_dev_list);
> +static DEFINE_SPINLOCK(qedr_devlist_lock);
> +
You already have a qedr_dev_list mutex in the qede_roce.c file,
why do you need this spinlock as well?
> +void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
> + enum ib_event_type type)
> +{
> + struct ib_event ibev;
> +
> + ibev.device = &dev->ibdev;
> + ibev.element.port_num = port_num;
> + ibev.event = type;
> +
> + ib_dispatch_event(&ibev);
> +}
> +
> +static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
> + u8 port_num)
> +{
> + return IB_LINK_LAYER_ETHERNET;
> +}
> +
> +static int qedr_register_device(struct qedr_dev *dev)
> +{
> + strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
> +
> + memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
> + dev->ibdev.owner = THIS_MODULE;
> +
> + dev->ibdev.get_link_layer = qedr_link_layer;
> +
> + return 0;
> +}
> +
> +/* QEDR sysfs interface */
> +static ssize_t show_rev(struct device *device, struct device_attribute *attr,
> + char *buf)
> +{
> + struct qedr_dev *dev = dev_get_drvdata(device);
> +
> + return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
> +}
> +
> +static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
> + char *buf)
> +{
> + return scnprintf(buf, PAGE_SIZE, "%s\n", "FW_VER_TO_ADD");
> +}
Ira Weiny has added a generic way to expose firmware versions in the rdma stack,
can you have please have a look at c73428230d98d1352bcc69cd8306c292a85e1e42 and
see how he converted the mlx5_ib module to use it.
> +static ssize_t show_hca_type(struct device *device,
> + struct device_attribute *attr, char *buf)
> +{
> + return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
> +}
> +
> +static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
> +static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
> +static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
> +
> +static struct device_attribute *qedr_attributes[] = {
> + &dev_attr_hw_rev,
> + &dev_attr_fw_ver,
> + &dev_attr_hca_type
> +};
> +
> +static void qedr_remove_sysfiles(struct qedr_dev *dev)
> +{
> + int i;
> +
> + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
> + device_remove_file(&dev->ibdev.dev, qedr_attributes[i]);
> +}
> +
> +void qedr_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
> +{
> + *p_dp_level = 0;
> + *p_dp_module = 0;
> +
> + if (debug & QED_LOG_VERBOSE_MASK) {
> + *p_dp_level = QED_LEVEL_VERBOSE;
> + *p_dp_module = (debug & 0x3FFFFFFF);
> + } else if (debug & QED_LOG_INFO_MASK) {
> + *p_dp_level = QED_LEVEL_INFO;
> + } else if (debug & QED_LOG_NOTICE_MASK) {
> + *p_dp_level = QED_LEVEL_NOTICE;
> + }
> +}
> +
> +static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
> +{
> + struct pci_dev *bridge;
> + u32 val;
> +
> + dev->atomic_cap = IB_ATOMIC_NONE;
> +
> + bridge = pdev->bus->self;
> + if (!bridge)
> + return;
> +
> + /* Check whether we are connected directly or via a switch */
> + while (bridge && bridge->bus->parent) {
> + DP_NOTICE(dev,
> + "Device is not connected directly to root. bridge->bus->number=%d primary=%d\n",
> + bridge->bus->number, bridge->bus->primary);
> + /* Need to check Atomic Op Routing Supported all the way to
> + * root complex.
> + */
> + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
> + if (!(val & PCI_EXP_DEVCAP2_ATOMIC_ROUTE)) {
> + pcie_capability_clear_word(pdev,
> + PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + return;
> + }
> + bridge = bridge->bus->parent->self;
> + }
> + bridge = pdev->bus->self;
> +
> + /* according to bridge capability */
> + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
> + if (val & PCI_EXP_DEVCAP2_ATOMIC_COMP64) {
> + pcie_capability_set_word(pdev, PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + dev->atomic_cap = IB_ATOMIC_GLOB;
> + } else {
> + pcie_capability_clear_word(pdev, PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + }
> +}
> +
> +static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
> + struct net_device *ndev)
> +{
> + struct qedr_dev *dev;
> + int rc = 0, i;
> +
> + dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
> + if (!dev) {
> + pr_err("Unable to allocate ib device\n");
> + return NULL;
> + }
> +
> + qedr_config_debug(debug, &dev->dp_module, &dev->dp_level);
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr add device called\n");
> +
> + dev->pdev = pdev;
> + dev->ndev = ndev;
> + dev->cdev = cdev;
> +
> + qedr_pci_set_atomic(dev, pdev);
> +
> + rc = qedr_register_device(dev);
> + if (rc) {
> + DP_ERR(dev, "Unable to allocate register device\n");
> + goto init_err;
> + }
> +
> + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
> + if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
> + goto init_err;
> +
> + spin_lock(&qedr_devlist_lock);
> + list_add_tail_rcu(&dev->entry, &qedr_dev_list);
> + spin_unlock(&qedr_devlist_lock);
> +
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
> + return dev;
> +
> +init_err:
> + ib_dealloc_device(&dev->ibdev);
> + DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
> +
> + return NULL;
> +}
> +
> +static void qedr_remove(struct qedr_dev *dev)
> +{
> + /* First unregister with stack to stop all the active traffic
> + * of the registered clients.
> + */
> + qedr_remove_sysfiles(dev);
> +
> + spin_lock(&qedr_devlist_lock);
> + list_del_rcu(&dev->entry);
> + spin_unlock(&qedr_devlist_lock);
> +
> + ib_dealloc_device(&dev->ibdev);
> +}
> +
> +static int qedr_close(struct qedr_dev *dev)
> +{
> + qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
> +
Why are you sending port number hard-coded as 1?
Mark
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
@ 2016-09-12 18:44 ` Mark Bloch
0 siblings, 0 replies; 78+ messages in thread
From: Mark Bloch @ 2016-09-12 18:44 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
Hi Ram,
Just a few thoughts
On 12/09/2016 19:07, Ram Amrani wrote:
> Adds a skeletal implementation of the qed* RoCE driver -
> basically the ability to communicate with the qede driver and
> receive notifications from it regarding various init/exit events.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> ---
> drivers/infiniband/Kconfig | 2 +
> drivers/infiniband/hw/Makefile | 1 +
> drivers/infiniband/hw/qedr/Kconfig | 7 +
> drivers/infiniband/hw/qedr/Makefile | 3 +
> drivers/infiniband/hw/qedr/main.c | 293 +++++++++++++++++++++++++
> drivers/infiniband/hw/qedr/qedr.h | 60 ++++++
> drivers/net/ethernet/qlogic/qede/Makefile | 1 +
> drivers/net/ethernet/qlogic/qede/qede.h | 9 +
> drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 +++++++++++++++++++++++++++
> include/linux/qed/qed_if.h | 3 +-
> include/linux/qed/qede_roce.h | 88 ++++++++
> include/uapi/linux/pci_regs.h | 3 +
> 13 files changed, 803 insertions(+), 11 deletions(-)
> create mode 100644 drivers/infiniband/hw/qedr/Kconfig
> create mode 100644 drivers/infiniband/hw/qedr/Makefile
> create mode 100644 drivers/infiniband/hw/qedr/main.c
> create mode 100644 drivers/infiniband/hw/qedr/qedr.h
> create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
> create mode 100644 include/linux/qed/qede_roce.h
[SNIP]
> +
> +MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
> +MODULE_AUTHOR("QLogic Corporation");
> +MODULE_LICENSE("Dual BSD/GPL");
> +MODULE_VERSION(QEDR_MODULE_VERSION);
> +
> +uint debug;
> +module_param(debug, uint, 0);
> +MODULE_PARM_DESC(debug, "Default debug msglevel");
Why are you adding this as a module parameter?
> +static LIST_HEAD(qedr_dev_list);
> +static DEFINE_SPINLOCK(qedr_devlist_lock);
> +
You already have a qedr_dev_list mutex in the qede_roce.c file,
why do you need this spinlock as well?
> +void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
> + enum ib_event_type type)
> +{
> + struct ib_event ibev;
> +
> + ibev.device = &dev->ibdev;
> + ibev.element.port_num = port_num;
> + ibev.event = type;
> +
> + ib_dispatch_event(&ibev);
> +}
> +
> +static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
> + u8 port_num)
> +{
> + return IB_LINK_LAYER_ETHERNET;
> +}
> +
> +static int qedr_register_device(struct qedr_dev *dev)
> +{
> + strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
> +
> + memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
> + dev->ibdev.owner = THIS_MODULE;
> +
> + dev->ibdev.get_link_layer = qedr_link_layer;
> +
> + return 0;
> +}
> +
> +/* QEDR sysfs interface */
> +static ssize_t show_rev(struct device *device, struct device_attribute *attr,
> + char *buf)
> +{
> + struct qedr_dev *dev = dev_get_drvdata(device);
> +
> + return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor);
> +}
> +
> +static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
> + char *buf)
> +{
> + return scnprintf(buf, PAGE_SIZE, "%s\n", "FW_VER_TO_ADD");
> +}
Ira Weiny has added a generic way to expose firmware versions in the rdma stack,
can you have please have a look at c73428230d98d1352bcc69cd8306c292a85e1e42 and
see how he converted the mlx5_ib module to use it.
> +static ssize_t show_hca_type(struct device *device,
> + struct device_attribute *attr, char *buf)
> +{
> + return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET");
> +}
> +
> +static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
> +static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
> +static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
> +
> +static struct device_attribute *qedr_attributes[] = {
> + &dev_attr_hw_rev,
> + &dev_attr_fw_ver,
> + &dev_attr_hca_type
> +};
> +
> +static void qedr_remove_sysfiles(struct qedr_dev *dev)
> +{
> + int i;
> +
> + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
> + device_remove_file(&dev->ibdev.dev, qedr_attributes[i]);
> +}
> +
> +void qedr_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
> +{
> + *p_dp_level = 0;
> + *p_dp_module = 0;
> +
> + if (debug & QED_LOG_VERBOSE_MASK) {
> + *p_dp_level = QED_LEVEL_VERBOSE;
> + *p_dp_module = (debug & 0x3FFFFFFF);
> + } else if (debug & QED_LOG_INFO_MASK) {
> + *p_dp_level = QED_LEVEL_INFO;
> + } else if (debug & QED_LOG_NOTICE_MASK) {
> + *p_dp_level = QED_LEVEL_NOTICE;
> + }
> +}
> +
> +static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
> +{
> + struct pci_dev *bridge;
> + u32 val;
> +
> + dev->atomic_cap = IB_ATOMIC_NONE;
> +
> + bridge = pdev->bus->self;
> + if (!bridge)
> + return;
> +
> + /* Check whether we are connected directly or via a switch */
> + while (bridge && bridge->bus->parent) {
> + DP_NOTICE(dev,
> + "Device is not connected directly to root. bridge->bus->number=%d primary=%d\n",
> + bridge->bus->number, bridge->bus->primary);
> + /* Need to check Atomic Op Routing Supported all the way to
> + * root complex.
> + */
> + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
> + if (!(val & PCI_EXP_DEVCAP2_ATOMIC_ROUTE)) {
> + pcie_capability_clear_word(pdev,
> + PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + return;
> + }
> + bridge = bridge->bus->parent->self;
> + }
> + bridge = pdev->bus->self;
> +
> + /* according to bridge capability */
> + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
> + if (val & PCI_EXP_DEVCAP2_ATOMIC_COMP64) {
> + pcie_capability_set_word(pdev, PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + dev->atomic_cap = IB_ATOMIC_GLOB;
> + } else {
> + pcie_capability_clear_word(pdev, PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + }
> +}
> +
> +static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
> + struct net_device *ndev)
> +{
> + struct qedr_dev *dev;
> + int rc = 0, i;
> +
> + dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
> + if (!dev) {
> + pr_err("Unable to allocate ib device\n");
> + return NULL;
> + }
> +
> + qedr_config_debug(debug, &dev->dp_module, &dev->dp_level);
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr add device called\n");
> +
> + dev->pdev = pdev;
> + dev->ndev = ndev;
> + dev->cdev = cdev;
> +
> + qedr_pci_set_atomic(dev, pdev);
> +
> + rc = qedr_register_device(dev);
> + if (rc) {
> + DP_ERR(dev, "Unable to allocate register device\n");
> + goto init_err;
> + }
> +
> + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
> + if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
> + goto init_err;
> +
> + spin_lock(&qedr_devlist_lock);
> + list_add_tail_rcu(&dev->entry, &qedr_dev_list);
> + spin_unlock(&qedr_devlist_lock);
> +
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
> + return dev;
> +
> +init_err:
> + ib_dealloc_device(&dev->ibdev);
> + DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
> +
> + return NULL;
> +}
> +
> +static void qedr_remove(struct qedr_dev *dev)
> +{
> + /* First unregister with stack to stop all the active traffic
> + * of the registered clients.
> + */
> + qedr_remove_sysfiles(dev);
> +
> + spin_lock(&qedr_devlist_lock);
> + list_del_rcu(&dev->entry);
> + spin_unlock(&qedr_devlist_lock);
> +
> + ib_dealloc_device(&dev->ibdev);
> +}
> +
> +static int qedr_close(struct qedr_dev *dev)
> +{
> + qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
> +
Why are you sending port number hard-coded as 1?
Mark
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 03/11] Add support for RoCE HW init
[not found] ` <1473696465-27986-4-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-12 18:57 ` Mark Bloch
2016-09-13 14:38 ` Sagi Grimberg
1 sibling, 0 replies; 78+ messages in thread
From: Mark Bloch @ 2016-09-12 18:57 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
On 12/09/2016 19:07, Ram Amrani wrote:
> Allocate and setup RoCE resources, interrupts and completion queues.
> Adds device attributes.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> ---
> drivers/infiniband/hw/qedr/main.c | 408 +++++++++++-
> drivers/infiniband/hw/qedr/qedr.h | 118 ++++
> drivers/infiniband/hw/qedr/qedr_hsi.h | 56 ++
> drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 96 +++
> drivers/net/ethernet/qlogic/qed/Makefile | 1 +
> drivers/net/ethernet/qlogic/qed/qed.h | 26 +-
> drivers/net/ethernet/qlogic/qed/qed_cxt.c | 6 +
> drivers/net/ethernet/qlogic/qed/qed_cxt.h | 6 +
> drivers/net/ethernet/qlogic/qed/qed_dev.c | 155 +++++
> drivers/net/ethernet/qlogic/qed/qed_main.c | 44 +-
> drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 7 +
> drivers/net/ethernet/qlogic/qed/qed_roce.c | 887 +++++++++++++++++++++++++
> drivers/net/ethernet/qlogic/qed/qed_roce.h | 117 ++++
> drivers/net/ethernet/qlogic/qed/qed_sp.h | 1 +
> drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 2 +-
> include/linux/qed/qed_if.h | 5 +-
> include/linux/qed/qed_roce_if.h | 345 ++++++++++
> 18 files changed, 2281 insertions(+), 7 deletions(-)
> create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
> create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
> create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
> create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
> create mode 100644 include/linux/qed/qed_roce_if.h
>
> diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
> index 3fe58a3..0b5274a 100644
> --- a/drivers/infiniband/hw/qedr/main.c
> +++ b/drivers/infiniband/hw/qedr/main.c
> @@ -36,6 +36,8 @@
> #include <linux/iommu.h>
> #include <net/addrconf.h>
> #include <linux/qed/qede_roce.h>
> +#include <linux/qed/qed_chain.h>
> +#include <linux/qed/qed_if.h>
> #include "qedr.h"
>
> MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
> @@ -80,6 +82,139 @@ static int qedr_register_device(struct qedr_dev *dev)
> return 0;
> }
>
> +/* This function allocates fast-path status block memory */
> +static int qedr_alloc_mem_sb(struct qedr_dev *dev,
> + struct qed_sb_info *sb_info, u16 sb_id)
> +{
> + struct status_block *sb_virt;
> + dma_addr_t sb_phys;
> + int rc;
> +
> + sb_virt = dma_alloc_coherent(&dev->pdev->dev,
> + sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
> + if (!sb_virt) {
> + pr_err("Status block allocation failed\n");
> + return -ENOMEM;
> + }
> +
> + rc = dev->ops->common->sb_init(dev->cdev, sb_info,
> + sb_virt, sb_phys, sb_id,
> + QED_SB_TYPE_CNQ);
> + if (rc) {
> + pr_err("Status block initialization failed\n");
> + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
> + sb_virt, sb_phys);
> + return rc;
> + }
> +
> + return 0;
> +}
> +
> +static void qedr_free_mem_sb(struct qedr_dev *dev,
> + struct qed_sb_info *sb_info, int sb_id)
> +{
> + if (sb_info->sb_virt) {
> + dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
> + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
> + (void *)sb_info->sb_virt, sb_info->sb_phys);
> + }
> +}
> +
> +static void qedr_free_resources(struct qedr_dev *dev)
> +{
> + int i;
> +
> + for (i = 0; i < dev->num_cnq; i++) {
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
> + }
> +
> + kfree(dev->cnq_array);
> + kfree(dev->sb_array);
> + kfree(dev->sgid_tbl);
> +}
> +
> +static int qedr_alloc_resources(struct qedr_dev *dev)
> +{
> + struct qedr_cnq *cnq;
> + __le16 *cons_pi;
> + u16 n_entries;
> + int i, rc;
> +
> + dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
> + QEDR_MAX_SGID, GFP_KERNEL);
> + if (!dev->sgid_tbl)
> + return -ENOMEM;
> +
> + spin_lock_init(&dev->sgid_lock);
> +
> + /* Allocate Status blocks for CNQ */
> + dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
> + GFP_KERNEL);
> + if (!dev->sb_array) {
> + rc = -ENOMEM;
> + goto err1;
> + }
> +
> + dev->cnq_array = kcalloc(dev->num_cnq,
> + sizeof(*dev->cnq_array), GFP_KERNEL);
> + if (!dev->cnq_array) {
> + rc = -ENOMEM;
> + goto err2;
> + }
> +
> + dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
> +
> + /* Allocate CNQ PBLs */
> + n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
> + for (i = 0; i < dev->num_cnq; i++) {
> + cnq = &dev->cnq_array[i];
> +
> + rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
> + dev->sb_start + i);
> + if (rc)
> + goto err3;
> +
> + rc = dev->ops->common->chain_alloc(dev->cdev,
> + QED_CHAIN_USE_TO_CONSUME,
> + QED_CHAIN_MODE_PBL,
> + QED_CHAIN_CNT_TYPE_U16,
> + n_entries,
> + sizeof(struct regpair *),
> + &cnq->pbl);
> + if (rc)
> + goto err4;
> +
> + cnq->dev = dev;
> + cnq->sb = &dev->sb_array[i];
> + cons_pi = dev->sb_array[i].sb_virt->pi_array;
> + cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
> + cnq->index = i;
> + sprintf(cnq->name, "qedr%d@pci:%s",
> + i, pci_name(dev->pdev));
> +
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
> + i, qed_chain_get_cons_idx(&cnq->pbl));
> + }
> +
> + return 0;
> +err4:
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + for (--i; i >= 0; i--) {
> + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + }
> +err3:
> + for (--i; i >= 0; i--)
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + kfree(dev->cnq_array);
This doesn't seem right, going from err4 to err3 i = 0, the for loop in err3
won't do anything.
> +err2:
> + kfree(dev->sb_array);
> +err1:
> + kfree(dev->sgid_tbl);
> + return rc;
> +}
> +
Mark
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 03/11] Add support for RoCE HW init
@ 2016-09-12 18:57 ` Mark Bloch
0 siblings, 0 replies; 78+ messages in thread
From: Mark Bloch @ 2016-09-12 18:57 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
On 12/09/2016 19:07, Ram Amrani wrote:
> Allocate and setup RoCE resources, interrupts and completion queues.
> Adds device attributes.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> ---
> drivers/infiniband/hw/qedr/main.c | 408 +++++++++++-
> drivers/infiniband/hw/qedr/qedr.h | 118 ++++
> drivers/infiniband/hw/qedr/qedr_hsi.h | 56 ++
> drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 96 +++
> drivers/net/ethernet/qlogic/qed/Makefile | 1 +
> drivers/net/ethernet/qlogic/qed/qed.h | 26 +-
> drivers/net/ethernet/qlogic/qed/qed_cxt.c | 6 +
> drivers/net/ethernet/qlogic/qed/qed_cxt.h | 6 +
> drivers/net/ethernet/qlogic/qed/qed_dev.c | 155 +++++
> drivers/net/ethernet/qlogic/qed/qed_main.c | 44 +-
> drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 7 +
> drivers/net/ethernet/qlogic/qed/qed_roce.c | 887 +++++++++++++++++++++++++
> drivers/net/ethernet/qlogic/qed/qed_roce.h | 117 ++++
> drivers/net/ethernet/qlogic/qed/qed_sp.h | 1 +
> drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 2 +-
> include/linux/qed/qed_if.h | 5 +-
> include/linux/qed/qed_roce_if.h | 345 ++++++++++
> 18 files changed, 2281 insertions(+), 7 deletions(-)
> create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
> create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
> create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
> create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
> create mode 100644 include/linux/qed/qed_roce_if.h
>
> diff --git a/drivers/infiniband/hw/qedr/main.c b/drivers/infiniband/hw/qedr/main.c
> index 3fe58a3..0b5274a 100644
> --- a/drivers/infiniband/hw/qedr/main.c
> +++ b/drivers/infiniband/hw/qedr/main.c
> @@ -36,6 +36,8 @@
> #include <linux/iommu.h>
> #include <net/addrconf.h>
> #include <linux/qed/qede_roce.h>
> +#include <linux/qed/qed_chain.h>
> +#include <linux/qed/qed_if.h>
> #include "qedr.h"
>
> MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
> @@ -80,6 +82,139 @@ static int qedr_register_device(struct qedr_dev *dev)
> return 0;
> }
>
> +/* This function allocates fast-path status block memory */
> +static int qedr_alloc_mem_sb(struct qedr_dev *dev,
> + struct qed_sb_info *sb_info, u16 sb_id)
> +{
> + struct status_block *sb_virt;
> + dma_addr_t sb_phys;
> + int rc;
> +
> + sb_virt = dma_alloc_coherent(&dev->pdev->dev,
> + sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
> + if (!sb_virt) {
> + pr_err("Status block allocation failed\n");
> + return -ENOMEM;
> + }
> +
> + rc = dev->ops->common->sb_init(dev->cdev, sb_info,
> + sb_virt, sb_phys, sb_id,
> + QED_SB_TYPE_CNQ);
> + if (rc) {
> + pr_err("Status block initialization failed\n");
> + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
> + sb_virt, sb_phys);
> + return rc;
> + }
> +
> + return 0;
> +}
> +
> +static void qedr_free_mem_sb(struct qedr_dev *dev,
> + struct qed_sb_info *sb_info, int sb_id)
> +{
> + if (sb_info->sb_virt) {
> + dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
> + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
> + (void *)sb_info->sb_virt, sb_info->sb_phys);
> + }
> +}
> +
> +static void qedr_free_resources(struct qedr_dev *dev)
> +{
> + int i;
> +
> + for (i = 0; i < dev->num_cnq; i++) {
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
> + }
> +
> + kfree(dev->cnq_array);
> + kfree(dev->sb_array);
> + kfree(dev->sgid_tbl);
> +}
> +
> +static int qedr_alloc_resources(struct qedr_dev *dev)
> +{
> + struct qedr_cnq *cnq;
> + __le16 *cons_pi;
> + u16 n_entries;
> + int i, rc;
> +
> + dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
> + QEDR_MAX_SGID, GFP_KERNEL);
> + if (!dev->sgid_tbl)
> + return -ENOMEM;
> +
> + spin_lock_init(&dev->sgid_lock);
> +
> + /* Allocate Status blocks for CNQ */
> + dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
> + GFP_KERNEL);
> + if (!dev->sb_array) {
> + rc = -ENOMEM;
> + goto err1;
> + }
> +
> + dev->cnq_array = kcalloc(dev->num_cnq,
> + sizeof(*dev->cnq_array), GFP_KERNEL);
> + if (!dev->cnq_array) {
> + rc = -ENOMEM;
> + goto err2;
> + }
> +
> + dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
> +
> + /* Allocate CNQ PBLs */
> + n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
> + for (i = 0; i < dev->num_cnq; i++) {
> + cnq = &dev->cnq_array[i];
> +
> + rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
> + dev->sb_start + i);
> + if (rc)
> + goto err3;
> +
> + rc = dev->ops->common->chain_alloc(dev->cdev,
> + QED_CHAIN_USE_TO_CONSUME,
> + QED_CHAIN_MODE_PBL,
> + QED_CHAIN_CNT_TYPE_U16,
> + n_entries,
> + sizeof(struct regpair *),
> + &cnq->pbl);
> + if (rc)
> + goto err4;
> +
> + cnq->dev = dev;
> + cnq->sb = &dev->sb_array[i];
> + cons_pi = dev->sb_array[i].sb_virt->pi_array;
> + cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
> + cnq->index = i;
> + sprintf(cnq->name, "qedr%d@pci:%s",
> + i, pci_name(dev->pdev));
> +
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
> + i, qed_chain_get_cons_idx(&cnq->pbl));
> + }
> +
> + return 0;
> +err4:
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + for (--i; i >= 0; i--) {
> + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + }
> +err3:
> + for (--i; i >= 0; i--)
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + kfree(dev->cnq_array);
This doesn't seem right, going from err4 to err3 i = 0, the for loop in err3
won't do anything.
> +err2:
> + kfree(dev->sb_array);
> +err1:
> + kfree(dev->sgid_tbl);
> + return rc;
> +}
> +
Mark
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <516b98c7-477a-4890-0d92-529dc32f2c4e-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
@ 2016-09-12 19:17 ` Yuval Mintz
[not found] ` <CY4PR11MB1720F9B74BBF82D0779FAC8F97FF0-JNf6+SjKdlG0ooKL/ADlEpPPoyLQLiKMvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
2016-09-13 9:22 ` Ram Amrani
1 sibling, 1 reply; 78+ messages in thread
From: Yuval Mintz @ 2016-09-12 19:17 UTC (permalink / raw)
To: Mark Bloch, Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller
Cc: Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
>> +uint debug;
>> +module_param(debug, uint, 0);
>> +MODULE_PARM_DESC(debug, "Default debug msglevel");
>Why are you adding this as a module parameter?
I believe this is mostly to follow same line as qede which also defines
'debug' module parameter for allowing easy user control of debug
prints [& specifically for probe prints, which can't be controlled
otherwise].
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <CY4PR11MB17202A542AC07CBA4A65E93697FF0-JNf6+SjKdlG0ooKL/ADlEpPPoyLQLiKMvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
@ 2016-09-13 6:05 ` Leon Romanovsky
[not found] ` <20160913060545.GN8812-2ukJVAZIZ/Y@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-13 6:05 UTC (permalink / raw)
To: Yuval Mintz
Cc: Parav Pandit, Ram Amrani, Doug Ledford, David Miller,
Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 2418 bytes --]
On Mon, Sep 12, 2016 at 05:39:35PM +0000, Yuval Mintz wrote:
> >>> include/linux/qed/common_hsi.h | 1 +
> >>> include/linux/qed/qed_if.h | 9 +-
> >>> include/linux/qed/qed_ll2_if.h | 140 +
> >>> include/linux/qed/qed_roce_if.h | 604 ++++
> >>> include/linux/qed/qede_roce.h | 88 +
> >> > include/linux/qed/rdma_common.h | 1 +
> >>
> >> Something not directly related to your patches, but they brought my
> >> attention to the fact that all these new (and old) rdma<->net devices
> >> are polluting include/linux
> >>
> > ocrdma driver includes be_roce.h located in net/ethernet/emulex/benet
> > location instead of include/linux/.
> > This file helps to bind rdma to net device or underlying hw device.
>
> > May be similar change can be done for rest of the drivers for
> > rdma<-->net devices?
>
> By adding explicit inclusion paths in the Makefile, a la
> ccflags-y := -Idrivers/net/ethernet/emulex/benet ?
>
> While this might work, I personally dislike it as I find it
> counter-intuitive when going over the code -
> I don't expect driver to locally modify the inclusion path.
> Besides, we're going to [eventually] a whole suite of drivers based
> on the qed module, some of which would reside under drivers/scsi;
> Not sure it's best to have 3 or 4 different drivers privately include the
> same directory under a different subsystem.
I agree with you that orcdma's way can be valuable for small drivers.
Orcmda has small shared headers set and doesn't need to change them rapidly
to support different devices.
I thought to place them in similar directory to include/soc/* and remove
from include/linux/. We have include/rdma/ and it looks like a good
candidate.
>
> >> Filtered output:
> >> ➜ linux-rdma git:(topic/fixes-for-4.8-2) ls -dl include/linux/*/
> >> drwxrwxr-x 2 leonro leonro 4096 Aug 30 16:27 include/linux/hsi/
> >> drwxrwxr-x 2 leonro leonro 4096 Sep 12 19:08 include/linux/mlx4/
> >> drwxrwxr-x 2 leonro leonro 4096 Sep 7 15:31 include/linux/mlx5/
> >> drwxrwxr-x 2 leonro leonro 4096 Sep 8 17:46 include/linux/qed/
> >>
> >> Is this the right place for them?
> >
> > Thanks
>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <CY4PR11MB1720F9B74BBF82D0779FAC8F97FF0-JNf6+SjKdlG0ooKL/ADlEpPPoyLQLiKMvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
@ 2016-09-13 6:38 ` Leon Romanovsky
2016-09-13 7:18 ` Mintz, Yuval
0 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-13 6:38 UTC (permalink / raw)
To: Yuval Mintz
Cc: Mark Bloch, Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
David Miller, Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 897 bytes --]
On Mon, Sep 12, 2016 at 07:17:35PM +0000, Yuval Mintz wrote:
> >> +uint debug;
> >> +module_param(debug, uint, 0);
> >> +MODULE_PARM_DESC(debug, "Default debug msglevel");
>
> >Why are you adding this as a module parameter?
>
> I believe this is mostly to follow same line as qede which also defines
> 'debug' module parameter for allowing easy user control of debug
> prints [& specifically for probe prints, which can't be controlled
> otherwise].
Can you give us an example where dynamic debug and tracing infrastructures
are not enough?
AFAIK, most of these debug module parameters are legacy copy/paste
code which is useless in real life scenarios.
Thanks
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
2016-09-12 16:07 ` Ram Amrani
(?)
(?)
@ 2016-09-13 6:46 ` Leon Romanovsky
-1 siblings, 0 replies; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-13 6:46 UTC (permalink / raw)
To: Ram Amrani
Cc: dledford, davem, Yuval.Mintz, Ariel.Elior, Michal.Kalderon,
rajesh.borundia, linux-rdma, netdev
[-- Attachment #1: Type: text/plain, Size: 607 bytes --]
On Mon, Sep 12, 2016 at 07:07:36PM +0300, Ram Amrani wrote:
> Adds a skeletal implementation of the qed* RoCE driver -
> basically the ability to communicate with the qede driver and
> receive notifications from it regarding various init/exit events.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
> Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
> ---
<...>
> +
> +#define QEDR_MODULE_VERSION "8.10.10.0"
I am strongly against module versions. You should rely on official
kernel version.
> +#define QEDR_NODE_DESC "QLogic 579xx RoCE HCA"
> +#define DP_NAME(dev) ((dev)->ibdev.name)
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <20160913060545.GN8812-2ukJVAZIZ/Y@public.gmane.org>
@ 2016-09-13 6:48 ` Mintz, Yuval
[not found] ` <BL2PR07MB23064356EA2492675AC980A78DFE0-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Mintz, Yuval @ 2016-09-13 6:48 UTC (permalink / raw)
To: Leon Romanovsky, Yuval Mintz
Cc: Parav Pandit, Ram Amrani, Doug Ledford, David Miller,
Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
>> While this might work, I personally dislike it as I find it
>> counter-intuitive when going over the code -
>> I don't expect driver to locally modify the inclusion path.
>> Besides, we're going to [eventually] a whole suite of drivers based
>> on the qed module, some of which would reside under drivers/scsi;
>> Not sure it's best to have 3 or 4 different drivers privately include the
>> same directory under a different subsystem.
> I agree with you that orcdma's way can be valuable for small drivers.
> Orcmda has small shared headers set and doesn't need to change them rapidly
> to support different devices.
> I thought to place them in similar directory to include/soc/* and remove
> from include/linux/. We have include/rdma/ and it looks like a good
> candidate.
I'm perfectly fine with relocating those to a different directory under include/,
although using 'rdma' doesn't sound like a good fit [as the headers would be
included by ethernet, scsi and rdma drivers].
Are there good existing alternatives?
Regardless, I don't believe this should be part of the initial submission,
as it would involve in relocating existing networking headers as well.
I think we can move those at leisure later on.
[We're in the middle of transitioning our e-mails from qlogic -> cavium,
so sorry if things become corrupted]--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
2016-09-13 6:38 ` Leon Romanovsky
@ 2016-09-13 7:18 ` Mintz, Yuval
2016-09-13 10:16 ` Leon Romanovsky
0 siblings, 1 reply; 78+ messages in thread
From: Mintz, Yuval @ 2016-09-13 7:18 UTC (permalink / raw)
To: Leon Romanovsky, Yuval Mintz
Cc: Mark Bloch, Ram Amrani, dledford, David Miller, Ariel Elior,
Michal Kalderon, Rajesh Borundia, linux-rdma, netdev
>> >> +uint debug;
>> >> +module_param(debug, uint, 0);
> >>> +MODULE_PARM_DESC(debug, "Default debug msglevel");
>>
>> >Why are you adding this as a module parameter?
>>
>> I believe this is mostly to follow same line as qede which also defines
> > 'debug' module parameter for allowing easy user control of debug
> > prints [& specifically for probe prints, which can't be controlled
> > otherwise].
> Can you give us an example where dynamic debug and tracing infrastructures
> are not enough?
> AFAIK, most of these debug module parameters are legacy copy/paste
> code which is useless in real life scenarios.
Define 'enough'; Using dynamic debug you can provide all the necessary
information and at an even better granularity that's achieved by suggested
infrastructure, but is harder for an end-user to use. Same goes for tracing.
The 'debug' option provides an easy grouping for prints related to a specific
area in the driver.
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 03/11] Add support for RoCE HW init
2016-09-12 18:57 ` Mark Bloch
(?)
@ 2016-09-13 8:30 ` Ram Amrani
-1 siblings, 0 replies; 78+ messages in thread
From: Ram Amrani @ 2016-09-13 8:30 UTC (permalink / raw)
To: Mark Bloch, dledford, David Miller
Cc: Yuval Mintz, Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma, netdev
Thanks Mark.
This will be fixed in next version.
Ram
-----Original Message-----
From: Mark Bloch [mailto:markb@mellanox.com]
Sent: Monday, September 12, 2016 9:58 PM
To: Ram Amrani <Ram.Amrani@qlogic.com>; dledford@redhat.com; David Miller <davem@davemloft.net>
Cc: Yuval Mintz <Yuval.Mintz@qlogic.com>; Ariel Elior <Ariel.Elior@qlogic.com>; Michal Kalderon <Michal.Kalderon@qlogic.com>; Rajesh Borundia <rajesh.borundia@qlogic.com>; linux-rdma@vger.kernel.org; netdev <netdev@vger.kernel.org>
Subject: Re: [RFC 03/11] Add support for RoCE HW init
On 12/09/2016 19:07, Ram Amrani wrote:
> Allocate and setup RoCE resources, interrupts and completion queues.
> Adds device attributes.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
> Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
> ---
> drivers/infiniband/hw/qedr/main.c | 408 +++++++++++-
> drivers/infiniband/hw/qedr/qedr.h | 118 ++++
> drivers/infiniband/hw/qedr/qedr_hsi.h | 56 ++
> drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 96 +++
> drivers/net/ethernet/qlogic/qed/Makefile | 1 +
> drivers/net/ethernet/qlogic/qed/qed.h | 26 +-
> drivers/net/ethernet/qlogic/qed/qed_cxt.c | 6 +
> drivers/net/ethernet/qlogic/qed/qed_cxt.h | 6 +
> drivers/net/ethernet/qlogic/qed/qed_dev.c | 155 +++++
> drivers/net/ethernet/qlogic/qed/qed_main.c | 44 +-
> drivers/net/ethernet/qlogic/qed/qed_reg_addr.h | 7 +
> drivers/net/ethernet/qlogic/qed/qed_roce.c | 887 +++++++++++++++++++++++++
> drivers/net/ethernet/qlogic/qed/qed_roce.h | 117 ++++
> drivers/net/ethernet/qlogic/qed/qed_sp.h | 1 +
> drivers/net/ethernet/qlogic/qed/qed_spq.c | 8 +
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 2 +-
> include/linux/qed/qed_if.h | 5 +-
> include/linux/qed/qed_roce_if.h | 345 ++++++++++
> 18 files changed, 2281 insertions(+), 7 deletions(-) create mode
> 100644 drivers/infiniband/hw/qedr/qedr_hsi.h
> create mode 100644 drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
> create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.c
> create mode 100644 drivers/net/ethernet/qlogic/qed/qed_roce.h
> create mode 100644 include/linux/qed/qed_roce_if.h
>
> diff --git a/drivers/infiniband/hw/qedr/main.c
> b/drivers/infiniband/hw/qedr/main.c
> index 3fe58a3..0b5274a 100644
> --- a/drivers/infiniband/hw/qedr/main.c
> +++ b/drivers/infiniband/hw/qedr/main.c
> @@ -36,6 +36,8 @@
> #include <linux/iommu.h>
> #include <net/addrconf.h>
> #include <linux/qed/qede_roce.h>
> +#include <linux/qed/qed_chain.h>
> +#include <linux/qed/qed_if.h>
> #include "qedr.h"
>
> MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver"); @@ -80,6 +82,139
> @@ static int qedr_register_device(struct qedr_dev *dev)
> return 0;
> }
>
> +/* This function allocates fast-path status block memory */ static
> +int qedr_alloc_mem_sb(struct qedr_dev *dev,
> + struct qed_sb_info *sb_info, u16 sb_id) {
> + struct status_block *sb_virt;
> + dma_addr_t sb_phys;
> + int rc;
> +
> + sb_virt = dma_alloc_coherent(&dev->pdev->dev,
> + sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
> + if (!sb_virt) {
> + pr_err("Status block allocation failed\n");
> + return -ENOMEM;
> + }
> +
> + rc = dev->ops->common->sb_init(dev->cdev, sb_info,
> + sb_virt, sb_phys, sb_id,
> + QED_SB_TYPE_CNQ);
> + if (rc) {
> + pr_err("Status block initialization failed\n");
> + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
> + sb_virt, sb_phys);
> + return rc;
> + }
> +
> + return 0;
> +}
> +
> +static void qedr_free_mem_sb(struct qedr_dev *dev,
> + struct qed_sb_info *sb_info, int sb_id) {
> + if (sb_info->sb_virt) {
> + dev->ops->common->sb_release(dev->cdev, sb_info, sb_id);
> + dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
> + (void *)sb_info->sb_virt, sb_info->sb_phys);
> + }
> +}
> +
> +static void qedr_free_resources(struct qedr_dev *dev) {
> + int i;
> +
> + for (i = 0; i < dev->num_cnq; i++) {
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
> + }
> +
> + kfree(dev->cnq_array);
> + kfree(dev->sb_array);
> + kfree(dev->sgid_tbl);
> +}
> +
> +static int qedr_alloc_resources(struct qedr_dev *dev) {
> + struct qedr_cnq *cnq;
> + __le16 *cons_pi;
> + u16 n_entries;
> + int i, rc;
> +
> + dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
> + QEDR_MAX_SGID, GFP_KERNEL);
> + if (!dev->sgid_tbl)
> + return -ENOMEM;
> +
> + spin_lock_init(&dev->sgid_lock);
> +
> + /* Allocate Status blocks for CNQ */
> + dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
> + GFP_KERNEL);
> + if (!dev->sb_array) {
> + rc = -ENOMEM;
> + goto err1;
> + }
> +
> + dev->cnq_array = kcalloc(dev->num_cnq,
> + sizeof(*dev->cnq_array), GFP_KERNEL);
> + if (!dev->cnq_array) {
> + rc = -ENOMEM;
> + goto err2;
> + }
> +
> + dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
> +
> + /* Allocate CNQ PBLs */
> + n_entries = min_t(u32, QED_RDMA_MAX_CNQ_SIZE, QEDR_ROCE_MAX_CNQ_SIZE);
> + for (i = 0; i < dev->num_cnq; i++) {
> + cnq = &dev->cnq_array[i];
> +
> + rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
> + dev->sb_start + i);
> + if (rc)
> + goto err3;
> +
> + rc = dev->ops->common->chain_alloc(dev->cdev,
> + QED_CHAIN_USE_TO_CONSUME,
> + QED_CHAIN_MODE_PBL,
> + QED_CHAIN_CNT_TYPE_U16,
> + n_entries,
> + sizeof(struct regpair *),
> + &cnq->pbl);
> + if (rc)
> + goto err4;
> +
> + cnq->dev = dev;
> + cnq->sb = &dev->sb_array[i];
> + cons_pi = dev->sb_array[i].sb_virt->pi_array;
> + cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
> + cnq->index = i;
> + sprintf(cnq->name, "qedr%d@pci:%s",
> + i, pci_name(dev->pdev));
> +
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
> + i, qed_chain_get_cons_idx(&cnq->pbl));
> + }
> +
> + return 0;
> +err4:
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + for (--i; i >= 0; i--) {
> + dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + }
> +err3:
> + for (--i; i >= 0; i--)
> + qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
> + kfree(dev->cnq_array);
This doesn't seem right, going from err4 to err3 i = 0, the for loop in err3 won't do anything.
> +err2:
> + kfree(dev->sb_array);
> +err1:
> + kfree(dev->sgid_tbl);
> + return rc;
> +}
> +
Mark
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <20160912180508.GI5843-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
@ 2016-09-13 8:44 ` Ram Amrani
[not found] ` <DM3PR1101MB1181DF570D4F01A0CEB1CFC7E2FE0-xYdf0wd+uoGW1Nawvih6nR68uu4wjhmwnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Ram Amrani @ 2016-09-13 8:44 UTC (permalink / raw)
To: Jason Gunthorpe
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller, Yuval Mintz,
Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
Hi Jason,
I see that "include/uapi/rdma" contains API that is common.
The qedr_user.h, that I assume you are referring to, is a qedr specific API.
For example, we issue the ib_copy_to_udata() on structures defined in the file.
So per my understanding it is in place.
Thanks,
Ram
-----Original Message-----
From: Jason Gunthorpe [mailto:jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org]
Sent: Monday, September 12, 2016 9:05 PM
To: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org; David Miller <davem-fT/PcQaiUtIeIZ0/mPfg9Q@public.gmane.org>; Yuval Mintz <Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Ariel Elior <Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Michal Kalderon <Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Rajesh Borundia <rajesh.borundia-h88ZbnxC6KBhl2p70BpVqQ@public.gmane.orgm>; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; netdev <netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>
Subject: Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
On Mon, Sep 12, 2016 at 07:07:34PM +0300, Ram Amrani wrote:
> drivers/infiniband/hw/qedr/main.c | 907 ++++++
> drivers/infiniband/hw/qedr/qedr.h | 494 ++++
> drivers/infiniband/hw/qedr/qedr_cm.c | 626 +++++
> drivers/infiniband/hw/qedr/qedr_cm.h | 61 +
> drivers/infiniband/hw/qedr/qedr_hsi.h | 56 +
> drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 748 +++++
> drivers/infiniband/hw/qedr/qedr_user.h | 80 +
We are requiring new uAPI headers are placed under include/uapi/rdma/, please coordinate with Leon on the path.
Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 02/11] Add RoCE driver framework
[not found] ` <516b98c7-477a-4890-0d92-529dc32f2c4e-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
2016-09-12 19:17 ` Yuval Mintz
@ 2016-09-13 9:22 ` Ram Amrani
[not found] ` <DM3PR1101MB1181337A65B7CEFC1A30F353E2FE0-xYdf0wd+uoGW1Nawvih6nR68uu4wjhmwnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
1 sibling, 1 reply; 78+ messages in thread
From: Ram Amrani @ 2016-09-13 9:22 UTC (permalink / raw)
To: Mark Bloch, dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller
Cc: Yuval Mintz, Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
Thanks for your comments.
See my replies in line with [Ram].
-----Original Message-----
From: Mark Bloch [mailto:markb-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org]
Sent: Monday, September 12, 2016 9:44 PM
To: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org; David Miller <davem-fT/PcQaiUtIeIZ0/mPfg9Q@public.gmane.org>
Cc: Yuval Mintz <Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Ariel Elior <Ariel.Elior@qlogic.com>; Michal Kalderon <Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; netdev <netdev-u79uwXL29TZNg+MwTxZMZA@public.gmane.orgl.org>
Subject: Re: [RFC 02/11] Add RoCE driver framework
Hi Ram,
Just a few thoughts
On 12/09/2016 19:07, Ram Amrani wrote:
> Adds a skeletal implementation of the qed* RoCE driver - basically the
> ability to communicate with the qede driver and receive notifications
> from it regarding various init/exit events.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> ---
> drivers/infiniband/Kconfig | 2 +
> drivers/infiniband/hw/Makefile | 1 +
> drivers/infiniband/hw/qedr/Kconfig | 7 +
> drivers/infiniband/hw/qedr/Makefile | 3 +
> drivers/infiniband/hw/qedr/main.c | 293 +++++++++++++++++++++++++
> drivers/infiniband/hw/qedr/qedr.h | 60 ++++++
> drivers/net/ethernet/qlogic/qede/Makefile | 1 +
> drivers/net/ethernet/qlogic/qede/qede.h | 9 +
> drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 309 +++++++++++++++++++++++++++
> include/linux/qed/qed_if.h | 3 +-
> include/linux/qed/qede_roce.h | 88 ++++++++
> include/uapi/linux/pci_regs.h | 3 +
> 13 files changed, 803 insertions(+), 11 deletions(-) create mode
> 100644 drivers/infiniband/hw/qedr/Kconfig
> create mode 100644 drivers/infiniband/hw/qedr/Makefile
> create mode 100644 drivers/infiniband/hw/qedr/main.c create mode
> 100644 drivers/infiniband/hw/qedr/qedr.h create mode 100644
> drivers/net/ethernet/qlogic/qede/qede_roce.c
> create mode 100644 include/linux/qed/qede_roce.h
[SNIP]
> +
> +MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
> +MODULE_AUTHOR("QLogic Corporation"); MODULE_LICENSE("Dual BSD/GPL");
> +MODULE_VERSION(QEDR_MODULE_VERSION);
> +
> +uint debug;
> +module_param(debug, uint, 0);
> +MODULE_PARM_DESC(debug, "Default debug msglevel");
Why are you adding this as a module parameter?
[Ram] Yuval commented on this in a previous e-mail
> +static LIST_HEAD(qedr_dev_list);
> +static DEFINE_SPINLOCK(qedr_devlist_lock);
> +
You already have a qedr_dev_list mutex in the qede_roce.c file, why do you need this spinlock as well?
[Ram] qedr_devlist_lock - a static (local) list of qedr devices maintained by qedr, protected by spinlock. Not in used in the current patches.
qedr_dev_list_lock (with '_') - a static (local) list of qedr devices maintained by qede, protected by mutex.
We'll consider removing the first as it is currently not used and/or rename them to be more distinct.
> +void qedr_ib_dispatch_event(struct qedr_dev *dev, u8 port_num,
> + enum ib_event_type type)
> +{
> + struct ib_event ibev;
> +
> + ibev.device = &dev->ibdev;
> + ibev.element.port_num = port_num;
> + ibev.event = type;
> +
> + ib_dispatch_event(&ibev);
> +}
> +
> +static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
> + u8 port_num)
> +{
> + return IB_LINK_LAYER_ETHERNET;
> +}
> +
> +static int qedr_register_device(struct qedr_dev *dev) {
> + strlcpy(dev->ibdev.name, "qedr%d", IB_DEVICE_NAME_MAX);
> +
> + memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
> + dev->ibdev.owner = THIS_MODULE;
> +
> + dev->ibdev.get_link_layer = qedr_link_layer;
> +
> + return 0;
> +}
> +
> +/* QEDR sysfs interface */
> +static ssize_t show_rev(struct device *device, struct device_attribute *attr,
> + char *buf)
> +{
> + struct qedr_dev *dev = dev_get_drvdata(device);
> +
> + return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->pdev->vendor); }
> +
> +static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
> + char *buf)
> +{
> + return scnprintf(buf, PAGE_SIZE, "%s\n", "FW_VER_TO_ADD"); }
Ira Weiny has added a generic way to expose firmware versions in the rdma stack, can you have please have a look at c73428230d98d1352bcc69cd8306c292a85e1e42 and see how he converted the mlx5_ib module to use it.
[Ram] This way is replaced to be the same as you describe in patch 0004. I'll if I can move it to this patch to avoid confusion.
> +static ssize_t show_hca_type(struct device *device,
> + struct device_attribute *attr, char *buf) {
> + return scnprintf(buf, PAGE_SIZE, "%s\n", "HCA_TYPE_TO_SET"); }
> +
> +static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL); static
> +DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL); static
> +DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
> +
> +static struct device_attribute *qedr_attributes[] = {
> + &dev_attr_hw_rev,
> + &dev_attr_fw_ver,
> + &dev_attr_hca_type
> +};
> +
> +static void qedr_remove_sysfiles(struct qedr_dev *dev) {
> + int i;
> +
> + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
> + device_remove_file(&dev->ibdev.dev, qedr_attributes[i]); }
> +
> +void qedr_config_debug(uint debug, u32 *p_dp_module, u8 *p_dp_level)
> +{
> + *p_dp_level = 0;
> + *p_dp_module = 0;
> +
> + if (debug & QED_LOG_VERBOSE_MASK) {
> + *p_dp_level = QED_LEVEL_VERBOSE;
> + *p_dp_module = (debug & 0x3FFFFFFF);
> + } else if (debug & QED_LOG_INFO_MASK) {
> + *p_dp_level = QED_LEVEL_INFO;
> + } else if (debug & QED_LOG_NOTICE_MASK) {
> + *p_dp_level = QED_LEVEL_NOTICE;
> + }
> +}
> +
> +static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev
> +*pdev) {
> + struct pci_dev *bridge;
> + u32 val;
> +
> + dev->atomic_cap = IB_ATOMIC_NONE;
> +
> + bridge = pdev->bus->self;
> + if (!bridge)
> + return;
> +
> + /* Check whether we are connected directly or via a switch */
> + while (bridge && bridge->bus->parent) {
> + DP_NOTICE(dev,
> + "Device is not connected directly to root. bridge->bus->number=%d primary=%d\n",
> + bridge->bus->number, bridge->bus->primary);
> + /* Need to check Atomic Op Routing Supported all the way to
> + * root complex.
> + */
> + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
> + if (!(val & PCI_EXP_DEVCAP2_ATOMIC_ROUTE)) {
> + pcie_capability_clear_word(pdev,
> + PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + return;
> + }
> + bridge = bridge->bus->parent->self;
> + }
> + bridge = pdev->bus->self;
> +
> + /* according to bridge capability */
> + pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &val);
> + if (val & PCI_EXP_DEVCAP2_ATOMIC_COMP64) {
> + pcie_capability_set_word(pdev, PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + dev->atomic_cap = IB_ATOMIC_GLOB;
> + } else {
> + pcie_capability_clear_word(pdev, PCI_EXP_DEVCTL2,
> + PCI_EXP_DEVCTL2_ATOMIC_REQ);
> + }
> +}
> +
> +static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
> + struct net_device *ndev)
> +{
> + struct qedr_dev *dev;
> + int rc = 0, i;
> +
> + dev = (struct qedr_dev *)ib_alloc_device(sizeof(*dev));
> + if (!dev) {
> + pr_err("Unable to allocate ib device\n");
> + return NULL;
> + }
> +
> + qedr_config_debug(debug, &dev->dp_module, &dev->dp_level);
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr add device called\n");
> +
> + dev->pdev = pdev;
> + dev->ndev = ndev;
> + dev->cdev = cdev;
> +
> + qedr_pci_set_atomic(dev, pdev);
> +
> + rc = qedr_register_device(dev);
> + if (rc) {
> + DP_ERR(dev, "Unable to allocate register device\n");
> + goto init_err;
> + }
> +
> + for (i = 0; i < ARRAY_SIZE(qedr_attributes); i++)
> + if (device_create_file(&dev->ibdev.dev, qedr_attributes[i]))
> + goto init_err;
> +
> + spin_lock(&qedr_devlist_lock);
> + list_add_tail_rcu(&dev->entry, &qedr_dev_list);
> + spin_unlock(&qedr_devlist_lock);
> +
> + DP_VERBOSE(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
> + return dev;
> +
> +init_err:
> + ib_dealloc_device(&dev->ibdev);
> + DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
> +
> + return NULL;
> +}
> +
> +static void qedr_remove(struct qedr_dev *dev) {
> + /* First unregister with stack to stop all the active traffic
> + * of the registered clients.
> + */
> + qedr_remove_sysfiles(dev);
> +
> + spin_lock(&qedr_devlist_lock);
> + list_del_rcu(&dev->entry);
> + spin_unlock(&qedr_devlist_lock);
> +
> + ib_dealloc_device(&dev->ibdev);
> +}
> +
> +static int qedr_close(struct qedr_dev *dev) {
> + qedr_ib_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
> +
Why are you sending port number hard-coded as 1?
[Ram] Our implementation always uses one port. Here's the ibv_devinfo output for example:
hca_id: qedr0
transport: InfiniBand (0)
fw_ver: 8.10.10.0
...
phys_port_cnt: 1
port: 1
state: PORT_ACTIVE (4)
...
link_layer: Ethernet
hca_id: qedr1
transport: InfiniBand (0)
fw_ver: 8.10.10.0
...
phys_port_cnt: 1
port: 1
state: PORT_ACTIVE (4)
...
link_layer: Ethernet
Mark
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
2016-09-13 7:18 ` Mintz, Yuval
@ 2016-09-13 10:16 ` Leon Romanovsky
[not found] ` <20160913101616.GT8812-2ukJVAZIZ/Y@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-13 10:16 UTC (permalink / raw)
To: Mintz, Yuval
Cc: Yuval Mintz, Mark Bloch, Ram Amrani, dledford, David Miller,
Ariel Elior, Michal Kalderon, Rajesh Borundia, linux-rdma,
netdev
[-- Attachment #1: Type: text/plain, Size: 1310 bytes --]
On Tue, Sep 13, 2016 at 07:18:01AM +0000, Mintz, Yuval wrote:
> >> >> +uint debug;
> >> >> +module_param(debug, uint, 0);
> > >>> +MODULE_PARM_DESC(debug, "Default debug msglevel");
> >>
> >> >Why are you adding this as a module parameter?
> >>
> >> I believe this is mostly to follow same line as qede which also defines
> > > 'debug' module parameter for allowing easy user control of debug
> > > prints [& specifically for probe prints, which can't be controlled
> > > otherwise].
>
> > Can you give us an example where dynamic debug and tracing infrastructures
> > are not enough?
>
> > AFAIK, most of these debug module parameters are legacy copy/paste
> > code which is useless in real life scenarios.
>
> Define 'enough'; Using dynamic debug you can provide all the necessary
> information and at an even better granularity that's achieved by suggested
> infrastructure, but is harder for an end-user to use. Same goes for tracing.
>
> The 'debug' option provides an easy grouping for prints related to a specific
> area in the driver.
It is hard to agree with you that user which knows how-to load modules
with parameters won't success to enable debug prints.
In addition, global increase in debug level for whole driver will create
printk storm in dmesg and give nothing to debuggability.
Thanks
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <DM3PR1101MB1181DF570D4F01A0CEB1CFC7E2FE0-xYdf0wd+uoGW1Nawvih6nR68uu4wjhmwnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
@ 2016-09-13 10:19 ` Leon Romanovsky
2016-09-13 15:40 ` Jason Gunthorpe
1 sibling, 0 replies; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-13 10:19 UTC (permalink / raw)
To: Ram Amrani
Cc: Jason Gunthorpe, dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller,
Yuval Mintz, Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 2178 bytes --]
On Tue, Sep 13, 2016 at 08:44:06AM +0000, Ram Amrani wrote:
> Hi Jason,
> I see that "include/uapi/rdma" contains API that is common.
> The qedr_user.h, that I assume you are referring to, is a qedr specific API.
> For example, we issue the ib_copy_to_udata() on structures defined in the file.
> So per my understanding it is in place.
1. It will be great if you avoid from top-posting.
2. In the near future, these custom UAPI header files will be placed
under include/uapi/rdma/provider/.
>
> Thanks,
> Ram
>
>
> -----Original Message-----
> From: Jason Gunthorpe [mailto:jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org]
> Sent: Monday, September 12, 2016 9:05 PM
> To: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org; David Miller <davem-fT/PcQaiUtIeIZ0/mPfg9Q@public.gmane.org>; Yuval Mintz <Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Ariel Elior <Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Michal Kalderon <Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; netdev <netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>
> Subject: Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
>
> On Mon, Sep 12, 2016 at 07:07:34PM +0300, Ram Amrani wrote:
> > drivers/infiniband/hw/qedr/main.c | 907 ++++++
> > drivers/infiniband/hw/qedr/qedr.h | 494 ++++
> > drivers/infiniband/hw/qedr/qedr_cm.c | 626 +++++
> > drivers/infiniband/hw/qedr/qedr_cm.h | 61 +
> > drivers/infiniband/hw/qedr/qedr_hsi.h | 56 +
> > drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 748 +++++
> > drivers/infiniband/hw/qedr/qedr_user.h | 80 +
>
> We are requiring new uAPI headers are placed under include/uapi/rdma/, please coordinate with Leon on the path.
>
> Jason
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <DM3PR1101MB1181337A65B7CEFC1A30F353E2FE0-xYdf0wd+uoGW1Nawvih6nR68uu4wjhmwnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
@ 2016-09-13 10:22 ` Leon Romanovsky
0 siblings, 0 replies; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-13 10:22 UTC (permalink / raw)
To: Ram Amrani
Cc: Mark Bloch, dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller,
Yuval Mintz, Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 228 bytes --]
On Tue, Sep 13, 2016 at 09:22:04AM +0000, Ram Amrani wrote:
> Thanks for your comments.
> See my replies in line with [Ram].
Please configure your email client
https://www.kernel.org/doc/Documentation/email-clients.txt
Thanks
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <BL2PR07MB23064356EA2492675AC980A78DFE0-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
@ 2016-09-13 10:39 ` Leon Romanovsky
0 siblings, 0 replies; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-13 10:39 UTC (permalink / raw)
To: Mintz, Yuval
Cc: Yuval Mintz, Parav Pandit, Ram Amrani, Doug Ledford,
David Miller, Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 1561 bytes --]
On Tue, Sep 13, 2016 at 06:48:00AM +0000, Mintz, Yuval wrote:
> >> While this might work, I personally dislike it as I find it
> >> counter-intuitive when going over the code -
> >> I don't expect driver to locally modify the inclusion path.
> >> Besides, we're going to [eventually] a whole suite of drivers based
> >> on the qed module, some of which would reside under drivers/scsi;
> >> Not sure it's best to have 3 or 4 different drivers privately include the
> >> same directory under a different subsystem.
>
> > I agree with you that orcdma's way can be valuable for small drivers.
>
> > Orcmda has small shared headers set and doesn't need to change them rapidly
> > to support different devices.
>
> > I thought to place them in similar directory to include/soc/* and remove
> > from include/linux/. We have include/rdma/ and it looks like a good
> > candidate.
>
> I'm perfectly fine with relocating those to a different directory under include/,
> although using 'rdma' doesn't sound like a good fit [as the headers would be
> included by ethernet, scsi and rdma drivers].
> Are there good existing alternatives?
There is nothing that I can name.
>
> Regardless, I don't believe this should be part of the initial submission,
> as it would involve in relocating existing networking headers as well.
> I think we can move those at leisure later on.
Sure, it is good time and context to discuss, but orthogonal to this
submission.
>
> [We're in the middle of transitioning our e-mails from qlogic -> cavium,
> so sorry if things become corrupted]
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <1473696465-27986-1-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
` (4 preceding siblings ...)
2016-09-12 18:05 ` Jason Gunthorpe
@ 2016-09-13 14:23 ` Sagi Grimberg
[not found] ` <CY1PR0701MB133732FA8478FC0B5003D97A90F10@CY1PR0701MB1337.namprd07.prod.outlook.com>
5 siblings, 1 reply; 78+ messages in thread
From: Sagi Grimberg @ 2016-09-13 14:23 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
Hey Ram and Co,
> This series introduces RoCE RDMA driver for the 579xx RDMA products by Qlogic.
> The RDMA support is added as an additional loadable module (qedr) over the Ethernet qede driver.
> The qedr module will support both RoCE and iWarp, although this series only adds RoCE support.
> The qed and qede drivers are enhanced with functionality required for RDMA support.
>
> Any review/comment is appreciated.
Was this driver tested with any of our kernel consumers (nfs/iser/nvmf)?
If so, are there known issues?
Thanks,
Sagi.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 07/11] Add support for memory registeration verbs
[not found] ` <1473696465-27986-8-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-13 14:26 ` Sagi Grimberg
2016-09-14 8:02 ` Amrani, Ram
0 siblings, 1 reply; 78+ messages in thread
From: Sagi Grimberg @ 2016-09-13 14:26 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
> +static inline struct qedr_ah *get_qedr_ah(struct ib_ah *ibah)
> +{
> + return container_of(ibah, struct qedr_ah, ibah);
> +}
Little surprising to find that here... how is the ah related
to this patch?
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 08/11] Add support for data path
[not found] ` <1473696465-27986-9-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-13 14:32 ` Sagi Grimberg
2016-09-14 10:16 ` Amrani, Ram
2016-09-15 7:24 ` Leon Romanovsky
1 sibling, 1 reply; 78+ messages in thread
From: Sagi Grimberg @ 2016-09-13 14:32 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
> + pbe = (struct regpair *)pbl_table->va;
> + num_pbes = 0;
> +
> + for (i = 0; i < mr->npages &&
> + (total_num_pbes != mr->info.pbl_info.num_pbes); i++) {
> + u64 buf_addr = mr->pages[i];
> +
> + pbe->lo = cpu_to_le32((u32)buf_addr);
> + pbe->hi = cpu_to_le32((u32)upper_32_bits(buf_addr));
Thats a shame... you could have easily set the buf_addr correctly
in qedr_set_page...
I think you could have also set the pbe directly from set_page if you
have access to pbl_table from your mr context
(and if I understand correctly I think you do, mr->info.pbl_table)...
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 03/11] Add support for RoCE HW init
[not found] ` <1473696465-27986-4-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-12 18:57 ` Mark Bloch
@ 2016-09-13 14:38 ` Sagi Grimberg
2016-09-14 10:13 ` Amrani, Ram
2016-09-19 8:45 ` Amrani, Ram
1 sibling, 2 replies; 78+ messages in thread
From: Sagi Grimberg @ 2016-09-13 14:38 UTC (permalink / raw)
To: Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
> + dev->max_sge = min_t(u32, RDMA_MAX_SGE_PER_SQ_WQE,
> + RDMA_MAX_SGE_PER_RQ_WQE);
Our kernel target mode consumers sort of rely on max_sge_rd, you need
to make sure to set it too.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 07/11] Add support for memory registeration verbs
2016-09-12 16:07 ` Ram Amrani
(?)
(?)
@ 2016-09-13 14:44 ` Sagi Grimberg
[not found] ` <7fa4a9b8-7cb1-0f83-d6e5-1055ae59bce4-NQWnxTmZq1alnMjI0IkVqw@public.gmane.org>
-1 siblings, 1 reply; 78+ messages in thread
From: Sagi Grimberg @ 2016-09-13 14:44 UTC (permalink / raw)
To: Ram Amrani, dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev
> +struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int max_page_list_len)
> +{
> + struct qedr_pd *pd = get_qedr_pd(ibpd);
> + struct qedr_dev *dev = get_qedr_dev(ibpd->device);
> + struct qedr_mr *mr;
> + int rc = -ENOMEM;
> +
> + DP_VERBOSE(dev, QEDR_MSG_MR,
> + "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd->pd_id,
> + max_page_list_len);
> +
> + mr = kzalloc(sizeof(*mr), GFP_KERNEL);
> + if (!mr)
> + return ERR_PTR(rc);
> +
> + mr->dev = dev;
> + mr->type = QEDR_MR_FRMR;
> +
> + rc = init_mr_info(dev, &mr->info, max_page_list_len, 1);
> + if (rc)
> + goto err0;
> +
> + rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
> + if (rc) {
> + DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
> + goto err0;
> + }
> +
> + /* Index only, 18 bit long, lkey = itid << 8 | key */
> + mr->hw_mr.tid_type = QED_RDMA_TID_FMR;
> + mr->hw_mr.key = 0;
> + mr->hw_mr.pd = pd->pd_id;
Do you have a real MR<->PD association in HW? If so, can you point
me to the code that binds it? If not, any reason not to expose
the local_dma_lkey?
> +struct ib_mr *qedr_get_dma_mr(struct ib_pd *ibpd, int acc)
> +{
> + struct qedr_dev *dev = get_qedr_dev(ibpd->device);
> + struct qedr_pd *pd = get_qedr_pd(ibpd);
> + struct qedr_mr *mr;
> + int rc;
> +
> + if (acc & IB_ACCESS_MW_BIND) {
> + DP_ERR(dev, "Unsupported access flags received for dma mr\n");
> + return ERR_PTR(-EINVAL);
> + }
This check looks like it really belongs in the core, it would help
everyone if you move it...
Although I know Christoph is trying to get rid of this API altogether...
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 02/11] Add RoCE driver framework
2016-09-12 16:07 ` Ram Amrani
@ 2016-09-13 14:46 ` Steve Wise
-1 siblings, 0 replies; 78+ messages in thread
From: Steve Wise @ 2016-09-13 14:46 UTC (permalink / raw)
To: 'Ram Amrani', dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev
> Adds a skeletal implementation of the qed* RoCE driver -
> basically the ability to communicate with the qede driver and
> receive notifications from it regarding various init/exit events.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
> Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
> ---
> drivers/infiniband/Kconfig | 2 +
> drivers/infiniband/hw/Makefile | 1 +
> drivers/infiniband/hw/qedr/Kconfig | 7 +
> drivers/infiniband/hw/qedr/Makefile | 3 +
> drivers/infiniband/hw/qedr/main.c | 293 +++++++++++++++++++++++++
> drivers/infiniband/hw/qedr/qedr.h | 60 ++++++
> drivers/net/ethernet/qlogic/qede/Makefile | 1 +
> drivers/net/ethernet/qlogic/qede/qede.h | 9 +
> drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 309
> +++++++++++++++++++++++++++
> include/linux/qed/qed_if.h | 3 +-
> include/linux/qed/qede_roce.h | 88 ++++++++
> include/uapi/linux/pci_regs.h | 3 +
> 13 files changed, 803 insertions(+), 11 deletions(-)
> create mode 100644 drivers/infiniband/hw/qedr/Kconfig
> create mode 100644 drivers/infiniband/hw/qedr/Makefile
> create mode 100644 drivers/infiniband/hw/qedr/main.c
> create mode 100644 drivers/infiniband/hw/qedr/qedr.h
> create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
> create mode 100644 include/linux/qed/qede_roce.h
<snip>
> @@ -189,8 +189,7 @@ static int qede_netdev_event(struct notifier_block *this,
> unsigned long event,
> struct ethtool_drvinfo drvinfo;
> struct qede_dev *edev;
>
> - /* Currently only support name change */
> - if (event != NETDEV_CHANGENAME)
> + if ((event != NETDEV_CHANGENAME) && (event !=
> NETDEV_CHANGEADDR))
nit: You don't really need the extra parens here.
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 02/11] Add RoCE driver framework
@ 2016-09-13 14:46 ` Steve Wise
0 siblings, 0 replies; 78+ messages in thread
From: Steve Wise @ 2016-09-13 14:46 UTC (permalink / raw)
To: 'Ram Amrani', dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev
> Adds a skeletal implementation of the qed* RoCE driver -
> basically the ability to communicate with the qede driver and
> receive notifications from it regarding various init/exit events.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
> Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
> ---
> drivers/infiniband/Kconfig | 2 +
> drivers/infiniband/hw/Makefile | 1 +
> drivers/infiniband/hw/qedr/Kconfig | 7 +
> drivers/infiniband/hw/qedr/Makefile | 3 +
> drivers/infiniband/hw/qedr/main.c | 293 +++++++++++++++++++++++++
> drivers/infiniband/hw/qedr/qedr.h | 60 ++++++
> drivers/net/ethernet/qlogic/qede/Makefile | 1 +
> drivers/net/ethernet/qlogic/qede/qede.h | 9 +
> drivers/net/ethernet/qlogic/qede/qede_main.c | 35 ++-
> drivers/net/ethernet/qlogic/qede/qede_roce.c | 309
> +++++++++++++++++++++++++++
> include/linux/qed/qed_if.h | 3 +-
> include/linux/qed/qede_roce.h | 88 ++++++++
> include/uapi/linux/pci_regs.h | 3 +
> 13 files changed, 803 insertions(+), 11 deletions(-)
> create mode 100644 drivers/infiniband/hw/qedr/Kconfig
> create mode 100644 drivers/infiniband/hw/qedr/Makefile
> create mode 100644 drivers/infiniband/hw/qedr/main.c
> create mode 100644 drivers/infiniband/hw/qedr/qedr.h
> create mode 100644 drivers/net/ethernet/qlogic/qede/qede_roce.c
> create mode 100644 include/linux/qed/qede_roce.h
<snip>
> @@ -189,8 +189,7 @@ static int qede_netdev_event(struct notifier_block *this,
> unsigned long event,
> struct ethtool_drvinfo drvinfo;
> struct qede_dev *edev;
>
> - /* Currently only support name change */
> - if (event != NETDEV_CHANGENAME)
> + if ((event != NETDEV_CHANGENAME) && (event !=
> NETDEV_CHANGEADDR))
nit: You don't really need the extra parens here.
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <DM3PR1101MB1181DF570D4F01A0CEB1CFC7E2FE0-xYdf0wd+uoGW1Nawvih6nR68uu4wjhmwnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2016-09-13 10:19 ` Leon Romanovsky
@ 2016-09-13 15:40 ` Jason Gunthorpe
[not found] ` <20160913154000.GA25878-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
1 sibling, 1 reply; 78+ messages in thread
From: Jason Gunthorpe @ 2016-09-13 15:40 UTC (permalink / raw)
To: Ram Amrani
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller, Yuval Mintz,
Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
On Tue, Sep 13, 2016 at 08:44:06AM +0000, Ram Amrani wrote:
> Hi Jason,
> I see that "include/uapi/rdma" contains API that is common.
It is intended to contain *all* the uapi.
> The qedr_user.h, that I assume you are referring to, is a qedr specific API.
> For example, we issue the ib_copy_to_udata() on structures defined in the file.
> So per my understanding it is in place.
Anything that is used with copy_to/from_user, ib_copy_to/from_udata,
etc, etc must be in a include/uapi header.
Any constant you might want to copy into your userspace provider must
be in include/uapi.
Avoid copying headers in your user space and use the standard kernel
names to access your driver's uapi.
Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 02/11] Add RoCE driver framework
2016-09-13 14:46 ` Steve Wise
(?)
@ 2016-09-14 7:30 ` Amrani, Ram
-1 siblings, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-14 7:30 UTC (permalink / raw)
To: Steve Wise
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
> > + if ((event != NETDEV_CHANGENAME) && (event !=
> > NETDEV_CHANGEADDR))
>
> nit: You don't really need the extra parens here.
>
Sure, thanks. Will remove.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 07/11] Add support for memory registeration verbs
2016-09-13 14:26 ` Sagi Grimberg
@ 2016-09-14 8:02 ` Amrani, Ram
0 siblings, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-14 8:02 UTC (permalink / raw)
To: Sagi Grimberg
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, dledford, davem
> > +static inline struct qedr_ah *get_qedr_ah(struct ib_ah *ibah) {
> > + return container_of(ibah, struct qedr_ah, ibah); }
>
> Little surprising to find that here... how is the ah related to this patch?
Thanks, Sagi. Will move into a proper location.
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <20160913101616.GT8812-2ukJVAZIZ/Y@public.gmane.org>
@ 2016-09-14 8:15 ` Mintz, Yuval
2016-09-14 13:00 ` Leon Romanovsky
0 siblings, 1 reply; 78+ messages in thread
From: Mintz, Yuval @ 2016-09-14 8:15 UTC (permalink / raw)
To: Leon Romanovsky
Cc: Yuval Mintz, Mark Bloch, Ram Amrani,
dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller, Ariel Elior,
Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
> > >> >> +uint debug;
> > >> >> +module_param(debug, uint, 0);
> > > >>> +MODULE_PARM_DESC(debug, "Default debug msglevel");
> > >>
> > >> >Why are you adding this as a module parameter?
> > >>
> > >> I believe this is mostly to follow same line as qede which also defines
> > > > 'debug' module parameter for allowing easy user control of debug
> > > > prints [& specifically for probe prints, which can't be controlled
> > > > otherwise].
> >
> > > Can you give us an example where dynamic debug and tracing infrastructures
> > > are not enough?
> >
> > > AFAIK, most of these debug module parameters are legacy copy/paste
> > > code which is useless in real life scenarios.
> >
> > Define 'enough'; Using dynamic debug you can provide all the necessary
> > information and at an even better granularity that's achieved by suggested
> > infrastructure, but is harder for an end-user to use. Same goes for tracing.
> >
> > The 'debug' option provides an easy grouping for prints related to a specific
> > area in the driver.
>
> It is hard to agree with you that user which knows how-to load modules
> with parameters won't success to enable debug prints.
I think you're giving too much credit to the end-user. :-D
> In addition, global increase in debug level for whole driver will create
> printk storm in dmesg and give nothing to debuggability.
So basically, what you're claiming is that ethtool 'msglvl' setting for devices
is completely obselete. While this *might* be true, we use it extensively
in our qede and qed drivers; The debug module parameter merely provides
a manner of setting the debug value prior to initial probe for all interfaces.
qedr follows the same practice.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <CY1PR0701MB133732FA8478FC0B5003D97A90F10-UpKza+2NMNLi6bjPjkn3FE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
@ 2016-09-14 8:17 ` Sagi Grimberg
0 siblings, 0 replies; 78+ messages in thread
From: Sagi Grimberg @ 2016-09-14 8:17 UTC (permalink / raw)
To: Elior, Ariel, Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
>> > SRQ is not part of the RFC (but we do have it and NVMF was tested
> with it).
>
>>
>
>> Nice, I have plans to make SRQs better usable for our ULPs so it'd be
>
>> good to have it.
>
> That’s good to know. Are there plans on implementing XRC?
>
> Right now it looks like none of the ULPS make use of it.
The problem with XRC is that it needs to be reflected on the
wire protocol for it to actually be useful for our ULPs. Not
sure if its worth the effort...
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 07/11] Add support for memory registeration verbs
[not found] ` <7fa4a9b8-7cb1-0f83-d6e5-1055ae59bce4-NQWnxTmZq1alnMjI0IkVqw@public.gmane.org>
@ 2016-09-14 8:59 ` Kalderon, Michal
2016-09-14 9:25 ` Sagi Grimberg
0 siblings, 1 reply; 78+ messages in thread
From: Kalderon, Michal @ 2016-09-14 8:59 UTC (permalink / raw)
To: Sagi Grimberg, Ram Amrani, dledford-H+wXaHxf7aLQT0dZR+AlfA,
davem-fT/PcQaiUtIeIZ0/mPfg9Q
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
> > +struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int
> > +max_page_list_len) {
> > + struct qedr_pd *pd = get_qedr_pd(ibpd);
> > + struct qedr_dev *dev = get_qedr_dev(ibpd->device);
> > + struct qedr_mr *mr;
> > + int rc = -ENOMEM;
> > +
> > + DP_VERBOSE(dev, QEDR_MSG_MR,
> > + "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd-
> >pd_id,
> > + max_page_list_len);
> > +
> > + mr = kzalloc(sizeof(*mr), GFP_KERNEL);
> > + if (!mr)
> > + return ERR_PTR(rc);
> > +
> > + mr->dev = dev;
> > + mr->type = QEDR_MR_FRMR;
> > +
> > + rc = init_mr_info(dev, &mr->info, max_page_list_len, 1);
> > + if (rc)
> > + goto err0;
> > +
> > + rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
> > + if (rc) {
> > + DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
> > + goto err0;
> > + }
> > +
> > + /* Index only, 18 bit long, lkey = itid << 8 | key */
> > + mr->hw_mr.tid_type = QED_RDMA_TID_FMR;
> > + mr->hw_mr.key = 0;
> > + mr->hw_mr.pd = pd->pd_id;
>
> Do you have a real MR<->PD association in HW? If so, can you point me to the
> code that binds it? If not, any reason not to expose the local_dma_lkey?
>
Yes, we send the pd id to the FW in function qed_rdma_register_tid. In any case, if we didn't have the association in HW
Wouldn't the local_dma_lkey be relevant only to dma_mr ? ( code snippet above refers to FMR)
> > +struct ib_mr *qedr_get_dma_mr(struct ib_pd *ibpd, int acc) {
> > + struct qedr_dev *dev = get_qedr_dev(ibpd->device);
> > + struct qedr_pd *pd = get_qedr_pd(ibpd);
> > + struct qedr_mr *mr;
> > + int rc;
> > +
> > + if (acc & IB_ACCESS_MW_BIND) {
> > + DP_ERR(dev, "Unsupported access flags received for dma
> mr\n");
> > + return ERR_PTR(-EINVAL);
> > + }
>
> This check looks like it really belongs in the core, it would help everyone if you
> move it...
>
> Although I know Christoph is trying to get rid of this API altogether...
Sure, will do.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body
> of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info at
> http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 07/11] Add support for memory registeration verbs
2016-09-14 8:59 ` Kalderon, Michal
@ 2016-09-14 9:25 ` Sagi Grimberg
2016-09-14 10:02 ` Kalderon, Michal
0 siblings, 1 reply; 78+ messages in thread
From: Sagi Grimberg @ 2016-09-14 9:25 UTC (permalink / raw)
To: Kalderon, Michal, Ram Amrani, dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev
>>> +struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int
>>> +max_page_list_len) {
>>> + struct qedr_pd *pd = get_qedr_pd(ibpd);
>>> + struct qedr_dev *dev = get_qedr_dev(ibpd->device);
>>> + struct qedr_mr *mr;
>>> + int rc = -ENOMEM;
>>> +
>>> + DP_VERBOSE(dev, QEDR_MSG_MR,
>>> + "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd-
>>> pd_id,
>>> + max_page_list_len);
>>> +
>>> + mr = kzalloc(sizeof(*mr), GFP_KERNEL);
>>> + if (!mr)
>>> + return ERR_PTR(rc);
>>> +
>>> + mr->dev = dev;
>>> + mr->type = QEDR_MR_FRMR;
>>> +
>>> + rc = init_mr_info(dev, &mr->info, max_page_list_len, 1);
>>> + if (rc)
>>> + goto err0;
>>> +
>>> + rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
>>> + if (rc) {
>>> + DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
>>> + goto err0;
>>> + }
>>> +
>>> + /* Index only, 18 bit long, lkey = itid << 8 | key */
>>> + mr->hw_mr.tid_type = QED_RDMA_TID_FMR;
>>> + mr->hw_mr.key = 0;
>>> + mr->hw_mr.pd = pd->pd_id;
>>
>> Do you have a real MR<->PD association in HW? If so, can you point me to the
>> code that binds it? If not, any reason not to expose the local_dma_lkey?
>>
> Yes, we send the pd id to the FW in function qed_rdma_register_tid.
Right, thanks.
> In any case, if we didn't have the association in HW
> Wouldn't the local_dma_lkey be relevant only to dma_mr ? ( code snippet above refers to FMR)
I was just sticking to a location in the code where you associate
MR<->PD...
The local_dma_lkey is a special key that spans the entire memory space
and unlike the notorious dma_mr, its not associated with a PD.
See the code in ib_alloc_pd(), if the device does not support a single
device local dma lkey, the core allocates a dma mr associated with
the pd. If your device has such a key, you can save a dma mr allocation
for each pd in the system.
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 07/11] Add support for memory registeration verbs
2016-09-14 9:25 ` Sagi Grimberg
@ 2016-09-14 10:02 ` Kalderon, Michal
0 siblings, 0 replies; 78+ messages in thread
From: Kalderon, Michal @ 2016-09-14 10:02 UTC (permalink / raw)
To: Sagi Grimberg, Ram Amrani, dledford, davem
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev
> >>> +struct qedr_mr *__qedr_alloc_mr(struct ib_pd *ibpd, int
> >>> +max_page_list_len) {
> >>> + struct qedr_pd *pd = get_qedr_pd(ibpd);
> >>> + struct qedr_dev *dev = get_qedr_dev(ibpd->device);
> >>> + struct qedr_mr *mr;
> >>> + int rc = -ENOMEM;
> >>> +
> >>> + DP_VERBOSE(dev, QEDR_MSG_MR,
> >>> + "qedr_alloc_frmr pd = %d max_page_list_len= %d\n", pd-
> >>> pd_id,
> >>> + max_page_list_len);
> >>> +
> >>> + mr = kzalloc(sizeof(*mr), GFP_KERNEL);
> >>> + if (!mr)
> >>> + return ERR_PTR(rc);
> >>> +
> >>> + mr->dev = dev;
> >>> + mr->type = QEDR_MR_FRMR;
> >>> +
> >>> + rc = init_mr_info(dev, &mr->info, max_page_list_len, 1);
> >>> + if (rc)
> >>> + goto err0;
> >>> +
> >>> + rc = dev->ops->rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
> >>> + if (rc) {
> >>> + DP_ERR(dev, "roce alloc tid returned an error %d\n", rc);
> >>> + goto err0;
> >>> + }
> >>> +
> >>> + /* Index only, 18 bit long, lkey = itid << 8 | key */
> >>> + mr->hw_mr.tid_type = QED_RDMA_TID_FMR;
> >>> + mr->hw_mr.key = 0;
> >>> + mr->hw_mr.pd = pd->pd_id;
> >>
> >> Do you have a real MR<->PD association in HW? If so, can you point me
> >> to the code that binds it? If not, any reason not to expose the
> local_dma_lkey?
> >>
> > Yes, we send the pd id to the FW in function qed_rdma_register_tid.
>
> Right, thanks.
>
> > In any case, if we didn't have the association in HW Wouldn't the
> > local_dma_lkey be relevant only to dma_mr ? ( code snippet above
> > refers to FMR)
>
> I was just sticking to a location in the code where you associate MR<->PD...
>
> The local_dma_lkey is a special key that spans the entire memory space and
> unlike the notorious dma_mr, its not associated with a PD.
>
> See the code in ib_alloc_pd(), if the device does not support a single device
> local dma lkey, the core allocates a dma mr associated with the pd. If your
> device has such a key, you can save a dma mr allocation for each pd in the
> system.
Managed to miss this. Our device supports such a key, we'll add support.
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 03/11] Add support for RoCE HW init
2016-09-13 14:38 ` Sagi Grimberg
@ 2016-09-14 10:13 ` Amrani, Ram
2016-09-19 8:45 ` Amrani, Ram
1 sibling, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-14 10:13 UTC (permalink / raw)
To: Sagi Grimberg
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, dledford, davem
> > + dev->max_sge = min_t(u32, RDMA_MAX_SGE_PER_SQ_WQE,
> > + RDMA_MAX_SGE_PER_RQ_WQE);
>
> Our kernel target mode consumers sort of rely on max_sge_rd, you need to
> make sure to set it too.
Good catch. Thanks!
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 08/11] Add support for data path
2016-09-13 14:32 ` Sagi Grimberg
@ 2016-09-14 10:16 ` Amrani, Ram
0 siblings, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-14 10:16 UTC (permalink / raw)
To: Sagi Grimberg
Cc: Yuval.Mintz, Ariel.Elior, Michal.Kalderon, rajesh.borundia,
linux-rdma, netdev, dledford, davem
> > + pbe = (struct regpair *)pbl_table->va;
> > + num_pbes = 0;
> > +
> > + for (i = 0; i < mr->npages &&
> > + (total_num_pbes != mr->info.pbl_info.num_pbes); i++) {
> > + u64 buf_addr = mr->pages[i];
> > +
> > + pbe->lo = cpu_to_le32((u32)buf_addr);
> > + pbe->hi = cpu_to_le32((u32)upper_32_bits(buf_addr));
>
> Thats a shame... you could have easily set the buf_addr correctly in
> qedr_set_page...
>
> I think you could have also set the pbe directly from set_page if you have access
> to pbl_table from your mr context (and if I understand correctly I think you do,
> mr->info.pbl_table)...
I see what you mean, we can surely improve here and will. Thanks.
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
2016-09-14 8:15 ` Mintz, Yuval
@ 2016-09-14 13:00 ` Leon Romanovsky
[not found] ` <20160914130032.GB26069-2ukJVAZIZ/Y@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-14 13:00 UTC (permalink / raw)
To: Mintz, Yuval
Cc: Yuval Mintz, Mark Bloch, Ram Amrani, dledford, David Miller,
Ariel Elior, Michal Kalderon, Rajesh Borundia, linux-rdma,
netdev
[-- Attachment #1: Type: text/plain, Size: 2178 bytes --]
On Wed, Sep 14, 2016 at 08:15:23AM +0000, Mintz, Yuval wrote:
> > > >> >> +uint debug;
> > > >> >> +module_param(debug, uint, 0);
> > > > >>> +MODULE_PARM_DESC(debug, "Default debug msglevel");
> > > >>
> > > >> >Why are you adding this as a module parameter?
> > > >>
> > > >> I believe this is mostly to follow same line as qede which also defines
> > > > > 'debug' module parameter for allowing easy user control of debug
> > > > > prints [& specifically for probe prints, which can't be controlled
> > > > > otherwise].
> > >
> > > > Can you give us an example where dynamic debug and tracing infrastructures
> > > > are not enough?
> > >
> > > > AFAIK, most of these debug module parameters are legacy copy/paste
> > > > code which is useless in real life scenarios.
> > >
> > > Define 'enough'; Using dynamic debug you can provide all the necessary
> > > information and at an even better granularity that's achieved by suggested
> > > infrastructure, but is harder for an end-user to use. Same goes for tracing.
> > >
> > > The 'debug' option provides an easy grouping for prints related to a specific
> > > area in the driver.
> >
> > It is hard to agree with you that user which knows how-to load modules
> > with parameters won't success to enable debug prints.
>
> I think you're giving too much credit to the end-user. :-D
>
> > In addition, global increase in debug level for whole driver will create
> > printk storm in dmesg and give nothing to debuggability.
>
> So basically, what you're claiming is that ethtool 'msglvl' setting for devices
> is completely obselete. While this *might* be true, we use it extensively
> in our qede and qed drivers; The debug module parameter merely provides
> a manner of setting the debug value prior to initial probe for all interfaces.
> qedr follows the same practice.
Thanks for this excellent example. Ethtool 'msglvl' adds this
dynamically, while your DEBUG argument works for loading module
only.
If you want dynamic prints, you have two options:
1. Add support of ethtool to whole RDMA stack.
2. Use dynamic tracing infrastructure.
Which option do you prefer?
>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <20160913154000.GA25878-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
@ 2016-09-14 14:44 ` Amrani, Ram
2016-09-14 17:17 ` Jason Gunthorpe
0 siblings, 1 reply; 78+ messages in thread
From: Amrani, Ram @ 2016-09-14 14:44 UTC (permalink / raw)
To: Jason Gunthorpe
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller, Yuval Mintz,
Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
> Anything that is used with copy_to/from_user, ib_copy_to/from_udata,
> etc, etc must be in a include/uapi header.
>
> Any constant you might want to copy into your userspace provider must
> be in include/uapi.
>
I understand you mean something like https://www.spinics.net/lists/linux-rdma/msg40414.html ([RFC rdma-next 0/3] Move mlx5 vendor specific to UAPI directory)
> Avoid copying headers in your user space and use the standard kernel
> names to access your driver's uapi.
So the user library should be able to compile against these headers in their location and not hold identical copies.
What do you mean by "standard kernel names"?
Thanks,
Ram
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
2016-09-14 14:44 ` Amrani, Ram
@ 2016-09-14 17:17 ` Jason Gunthorpe
[not found] ` <20160914171737.GH16014-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Jason Gunthorpe @ 2016-09-14 17:17 UTC (permalink / raw)
To: Amrani, Ram
Cc: dledford, David Miller, Yuval Mintz, Ariel Elior,
Michal Kalderon, Rajesh Borundia, linux-rdma, netdev
On Wed, Sep 14, 2016 at 02:44:47PM +0000, Amrani, Ram wrote:
> > Anything that is used with copy_to/from_user, ib_copy_to/from_udata,
> > etc, etc must be in a include/uapi header.
> >
> > Any constant you might want to copy into your userspace provider must
> > be in include/uapi.
> >
> I understand you mean something like
> https://www.spinics.net/lists/linux-rdma/msg40414.html ([RFC
> rdma-next 0/3] Move mlx5 vendor specific to UAPI directory)
Yes.
> > Avoid copying headers in your user space and use the standard kernel
> > names to access your driver's uapi.
> So the user library should be able to compile against these headers
> in their location and not hold identical copies.
Yes, but you will need identical copies for supporting old distros.
> What do you mean by "standard kernel names"?
By that I mean 'identical copies' do not copy the file and then
randomly change it giving things different names or putting different
content in structs.
You will want to submit your user provider to rdma-plumbing to get it
into the distros, we are planning to set it as the vehicle for code
targeting 4.9
Jason
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <20160914130032.GB26069-2ukJVAZIZ/Y@public.gmane.org>
@ 2016-09-14 18:25 ` Mintz, Yuval
[not found] ` <BL2PR07MB23066BBC4A019365A632E2FA8DF10-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Mintz, Yuval @ 2016-09-14 18:25 UTC (permalink / raw)
To: Leon Romanovsky
Cc: Yuval Mintz, Mark Bloch, Ram Amrani,
dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller, Ariel Elior,
Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
> > > > >> >> +uint debug;
> > > > >> >> +module_param(debug, uint, 0);
> > > > > >>> +MODULE_PARM_DESC(debug, "Default debug msglevel");
> > > > >>
> > > > >> >Why are you adding this as a module parameter?
> > > > >>
> > > > >> I believe this is mostly to follow same line as qede which also defines
> > > > > > 'debug' module parameter for allowing easy user control of debug
> > > > > > prints [& specifically for probe prints, which can't be controlled
> > > > > > otherwise].
> > > >
> > > > > Can you give us an example where dynamic debug and tracing infrastructures
> > > > > are not enough?
> > > >
> > > > > AFAIK, most of these debug module parameters are legacy copy/paste
> > > > > code which is useless in real life scenarios.
> > > >
> > > > Define 'enough'; Using dynamic debug you can provide all the necessary
> > > > information and at an even better granularity that's achieved by suggested
> > > > infrastructure, but is harder for an end-user to use. Same goes for tracing.
> > > >
> > > > The 'debug' option provides an easy grouping for prints related to a specific
> > > > area in the driver.
> > >
> > > It is hard to agree with you that user which knows how-to load modules
> > > with parameters won't success to enable debug prints.
> >
> > I think you're giving too much credit to the end-user. :-D
> >
> > > In addition, global increase in debug level for whole driver will create
> > > printk storm in dmesg and give nothing to debuggability.
> >
> > So basically, what you're claiming is that ethtool 'msglvl' setting for devices
> > is completely obselete. While this *might* be true, we use it extensively
> > in our qede and qed drivers; The debug module parameter merely provides
> > a manner of setting the debug value prior to initial probe for all interfaces.
> > qedr follows the same practice.
> Thanks for this excellent example. Ethtool 'msglvl' adds this
> dynamically, while your DEBUG argument works for loading module
> only.
> If you want dynamic prints, you have two options:
> 1. Add support of ethtool to whole RDMA stack.
> 2. Use dynamic tracing infrastructure.
> Which option do you prefer?
Option 3 - continuing this discussion. :-)
Perhaps I misread your intentions - I thought that by dynamic debug
you meant that all debug in RDMA should be pr_debug() based, and
therefore my objection regarding the ease with which users can
configure it.
If all you meant was 'dynamically set' as opposed to 'statically set'
then I agree that having that sort of configurability is preferable
[Even though end-user would still probably prefer a module
parameter for reproductions; As the name implies, 'debug' isn't
meant to be used in other situations].
The other thing to consider are the probe-time prints.
Problem is, you wouldn't have a control node between probe
and until after the probing would be over, so it would be a bit
hard to configure that.
You can always think of some generic method of fixing that as well
[sysfs node for the entire system for probe-time prints, perhaps?]
Do notice you would be harming user-experience of reproductions
though - as it would have to follow different mechanisms to open
debug prints of various qed* components.--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <BL2PR07MB23066BBC4A019365A632E2FA8DF10-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
@ 2016-09-15 4:37 ` Leon Romanovsky
[not found] ` <20160915043716.GD26069-2ukJVAZIZ/Y@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-15 4:37 UTC (permalink / raw)
To: Mintz, Yuval
Cc: Yuval Mintz, Mark Bloch, Ram Amrani,
dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller, Ariel Elior,
Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 4543 bytes --]
On Wed, Sep 14, 2016 at 06:25:38PM +0000, Mintz, Yuval wrote:
> > > > > >> >> +uint debug;
> > > > > >> >> +module_param(debug, uint, 0);
> > > > > > >>> +MODULE_PARM_DESC(debug, "Default debug msglevel");
> > > > > >>
> > > > > >> >Why are you adding this as a module parameter?
> > > > > >>
> > > > > >> I believe this is mostly to follow same line as qede which also defines
> > > > > > > 'debug' module parameter for allowing easy user control of debug
> > > > > > > prints [& specifically for probe prints, which can't be controlled
> > > > > > > otherwise].
> > > > >
> > > > > > Can you give us an example where dynamic debug and tracing infrastructures
> > > > > > are not enough?
> > > > >
> > > > > > AFAIK, most of these debug module parameters are legacy copy/paste
> > > > > > code which is useless in real life scenarios.
> > > > >
> > > > > Define 'enough'; Using dynamic debug you can provide all the necessary
> > > > > information and at an even better granularity that's achieved by suggested
> > > > > infrastructure, but is harder for an end-user to use. Same goes for tracing.
> > > > >
> > > > > The 'debug' option provides an easy grouping for prints related to a specific
> > > > > area in the driver.
> > > >
> > > > It is hard to agree with you that user which knows how-to load modules
> > > > with parameters won't success to enable debug prints.
> > >
> > > I think you're giving too much credit to the end-user. :-D
> > >
> > > > In addition, global increase in debug level for whole driver will create
> > > > printk storm in dmesg and give nothing to debuggability.
> > >
> > > So basically, what you're claiming is that ethtool 'msglvl' setting for devices
> > > is completely obselete. While this *might* be true, we use it extensively
> > > in our qede and qed drivers; The debug module parameter merely provides
> > > a manner of setting the debug value prior to initial probe for all interfaces.
> > > qedr follows the same practice.
>
> > Thanks for this excellent example. Ethtool 'msglvl' adds this
> > dynamically, while your DEBUG argument works for loading module
> > only.
>
> > If you want dynamic prints, you have two options:
> > 1. Add support of ethtool to whole RDMA stack.
> > 2. Use dynamic tracing infrastructure.
>
> > Which option do you prefer?
> Option 3 - continuing this discussion. :-)
Sorry,
I was under impression that you want this driver to be merged, but it
looks like It was incorrect assumption. Let's continue discussion.
>
> Perhaps I misread your intentions - I thought that by dynamic debug
> you meant that all debug in RDMA should be pr_debug() based, and
> therefore my objection regarding the ease with which users can
> configure it.
It is not for all RDMA, but in your proposed driver. You are adding this
"debug" module argument to your module.
> If all you meant was 'dynamically set' as opposed to 'statically set'
> then I agree that having that sort of configurability is preferable
> [Even though end-user would still probably prefer a module
> parameter for reproductions; As the name implies, 'debug' isn't
> meant to be used in other situations].
We are not adding code just for fun, but for a real reason, and
especially interfaces which will be visible to user.
The overall expectation from the driver's authors that they are
submitting driver which doesn't have bringup issues. For real
life scenarios, where the bugs will be reveled after some time of
usage, this global debug is useless.
>
> The other thing to consider are the probe-time prints.
> Problem is, you wouldn't have a control node between probe
> and until after the probing would be over, so it would be a bit
> hard to configure that.
> You can always think of some generic method of fixing that as well
> [sysfs node for the entire system for probe-time prints, perhaps?]
/sys/kernel/debug/tracing
/sys/kernel/debug/dynamic_debug
>
> Do notice you would be harming user-experience of reproductions
> though - as it would have to follow different mechanisms to open
> debug prints of various qed* components.
I don't understand this point at all. Do you think that it is normal to
ask user to debug your driver? Is this called "user-experience"?
And regarding the second point, the old code is not an excuse to
copy/paste bad practices.
As a summary, I didn't see in your responses any real life example where
you will need global debug level for your driver.
Thanks
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 02/11] Add RoCE driver framework
[not found] ` <20160915043716.GD26069-2ukJVAZIZ/Y@public.gmane.org>
@ 2016-09-15 5:11 ` Mintz, Yuval
[not found] ` <BL2PR07MB23060C776EDAE92B84FCFB768DF00-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Mintz, Yuval @ 2016-09-15 5:11 UTC (permalink / raw)
To: Leon Romanovsky, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: Mark Bloch, Ram Amrani, David Miller, Ariel Elior,
Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
> > > If you want dynamic prints, you have two options:
> > > 1. Add support of ethtool to whole RDMA stack.
> > > 2. Use dynamic tracing infrastructure.
> >
> > > Which option do you prefer?
> > Option 3 - continuing this discussion. :-)
>
> Sorry,
> I was under impression that you want this driver to be merged, but it looks like It
> was incorrect assumption. Let's continue discussion.
No, this is an RFC - there's no chance for *this* to merge, but this is exactly
the right time to discuss this sort of stuff.
> > Perhaps I misread your intentions - I thought that by dynamic debug
> > you meant that all debug in RDMA should be pr_debug() based, and
> > therefore my objection regarding the ease with which users can
> > configure it.
>
> It is not for all RDMA, but in your proposed driver. You are adding this "debug"
> module argument to your module.
I don't get your answer.
I made a generic remark [and actually one in favor of your arguments],
and instead of saying something meaningful you bash the driver.
> > If all you meant was 'dynamically set' as opposed to 'statically set'
> > then I agree that having that sort of configurability is preferable
> > [Even though end-user would still probably prefer a module parameter
> > for reproductions; As the name implies, 'debug' isn't meant to be used
> > in other situations].
>
> We are not adding code just for fun, but for a real reason, and especially
> interfaces which will be visible to user.
>
> The overall expectation from the driver's authors that they are submitting driver
> which doesn't have bringup issues. For real life scenarios, where the bugs will be
> reveled after some time of usage, this global debug is useless.
This has nothing to do with bringup; Real drivers are experiencing issues years after
they're productized.
> > Do notice you would be harming user-experience of reproductions though
> > - as it would have to follow different mechanisms to open debug prints
> > of various qed* components.
>
> I don't understand this point at all. Do you think that it is normal to ask user to
> debug your driver? Is this called "user-experience"?
No, I call this 'user involved in fixing the driver' - it has nothing to do with
user-experience. Sometimes user have specifics in his system that can't
be easily identified and thus lab reproductions fail, and the user assists
in the reproduction. While I never claimed this is good practice it does happen.
> As a summary, I didn't see in your responses any real life example where you will
> need global debug level for your driver.
Not sure what you you're expecting - a list of BZs /private e-mails where
user reproductions were needed?
You're basically ignoring my claims that such are used, instead wanting
"evidence". I'm not going to try and produce any such.
Doug - I think we need a definite answer from you here; Doesn't look like
this discussion would bear any fruit.
If a debug module parameter is completely unacceptable, we'd remove it
[regardless of what I think about it].
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <BL2PR07MB23060C776EDAE92B84FCFB768DF00-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
@ 2016-09-15 5:42 ` Leon Romanovsky
2016-09-20 15:04 ` Elior, Ariel
0 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-15 5:42 UTC (permalink / raw)
To: Mintz, Yuval
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, Mark Bloch, Ram Amrani,
David Miller, Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 633 bytes --]
On Thu, Sep 15, 2016 at 05:11:03AM +0000, Mintz, Yuval wrote:
> > As a summary, I didn't see in your responses any real life example where you will
> > need global debug level for your driver.
>
> Not sure what you you're expecting - a list of BZs /private e-mails where
> user reproductions were needed?
> You're basically ignoring my claims that such are used, instead wanting
> "evidence". I'm not going to try and produce any such.
I asked an example and not evidence, where "modprobe your_driver
debug=1" will be superior to "modprobe your_driver dyndbg==pmf".
https://www.kernel.org/doc/Documentation/dynamic-debug-howto.txt
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 00/11] QLogic RDMA Driver (qedr) RFC
[not found] ` <20160914171737.GH16014-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
@ 2016-09-15 5:55 ` Amrani, Ram
0 siblings, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-15 5:55 UTC (permalink / raw)
To: Jason Gunthorpe
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, David Miller, Yuval Mintz,
Ariel Elior, Michal Kalderon, Rajesh Borundia,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
> > What do you mean by "standard kernel names"?
>
> By that I mean 'identical copies' do not copy the file and then randomly change
> it giving things different names or putting different content in structs.
>
> You will want to submit your user provider to rdma-plumbing to get it into the
> distros, we are planning to set it as the vehicle for code targeting 4.9
Got it. Thanks.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 08/11] Add support for data path
[not found] ` <1473696465-27986-9-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-13 14:32 ` Sagi Grimberg
@ 2016-09-15 7:24 ` Leon Romanovsky
1 sibling, 0 replies; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-15 7:24 UTC (permalink / raw)
To: Ram Amrani
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q,
Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
[-- Attachment #1: Type: text/plain, Size: 546 bytes --]
On Mon, Sep 12, 2016 at 07:07:42PM +0300, Ram Amrani wrote:
> +++ b/drivers/infiniband/hw/qedr/qedr_hsi_rdma.h
> @@ -150,6 +150,12 @@ struct rdma_rq_sge {
> struct regpair addr;
> __le32 length;
> __le32 flags;
> +#define RDMA_RQ_SGE_L_KEY_MASK 0x3FFFFFF
> +#define RDMA_RQ_SGE_L_KEY_SHIFT 0
> +#define RDMA_RQ_SGE_NUM_SGES_MASK 0x7
> +#define RDMA_RQ_SGE_NUM_SGES_SHIFT 26
> +#define RDMA_RQ_SGE_RESERVED0_MASK 0x7
> +#define RDMA_RQ_SGE_RESERVED0_SHIFT 29
> };
It is interesting twist to mix defines and structs together.
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 09/11] Add LL2 RoCE interface
2016-09-12 16:07 ` Ram Amrani
(?)
@ 2016-09-15 10:20 ` Leon Romanovsky
[not found] ` <20160915102013.GZ26069-2ukJVAZIZ/Y@public.gmane.org>
-1 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-15 10:20 UTC (permalink / raw)
To: Ram Amrani
Cc: dledford, davem, Yuval.Mintz, Ariel.Elior, Michal.Kalderon,
rajesh.borundia, linux-rdma, netdev
[-- Attachment #1: Type: text/plain, Size: 600 bytes --]
On Mon, Sep 12, 2016 at 07:07:43PM +0300, Ram Amrani wrote:
> Add light L2 interface for RoCE.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia@qlogic.com>
> Signed-off-by: Ram Amrani <Ram.Amrani@qlogic.com>
> ---
<....>
> + DP_ERR(cdev,
> + "QED RoCE set MAC filter failed - roce_info/ll2 NULL\n");
> + return -EINVAL;
> + }
> +
> + p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
> + if (!p_ptt) {
> + DP_ERR(cdev,
> + "qed roce ll2 mac filter set: failed to acquire PTT\n");
> + return -EINVAL;
> + }
Please use single style for your debug prints QED RoCE vs. qed roce.
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 09/11] Add LL2 RoCE interface
[not found] ` <20160915102013.GZ26069-2ukJVAZIZ/Y@public.gmane.org>
@ 2016-09-15 12:07 ` Amrani, Ram
0 siblings, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-15 12:07 UTC (permalink / raw)
To: Leon Romanovsky
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q,
Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
> > + DP_ERR(cdev,
> > + "QED RoCE set MAC filter failed - roce_info/ll2 NULL\n");
> > + return -EINVAL;
> > + }
> > +
> > + p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
> > + if (!p_ptt) {
> > + DP_ERR(cdev,
> > + "qed roce ll2 mac filter set: failed to acquire PTT\n");
> > + return -EINVAL;
> > + }
>
> Please use single style for your debug prints QED RoCE vs. qed roce.
Sure. This bothers me too. Will fix. Thanks.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 06/11] Add support for QP verbs
[not found] ` <1473696465-27986-7-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
@ 2016-09-15 13:05 ` Leon Romanovsky
[not found] ` <20160915130556.GA26069-2ukJVAZIZ/Y@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-15 13:05 UTC (permalink / raw)
To: Ram Amrani
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q,
Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
[-- Attachment #1: Type: text/plain, Size: 3191 bytes --]
On Mon, Sep 12, 2016 at 07:07:40PM +0300, Ram Amrani wrote:
> Add support for Queue Pair verbs which adds, deletes,
> modifies and queries Queue Pairs.
>
> Signed-off-by: Rajesh Borundia <rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> Signed-off-by: Ram Amrani <Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
> ---
> drivers/infiniband/hw/qedr/main.c | 15 +-
> drivers/infiniband/hw/qedr/qedr.h | 126 +++
> drivers/infiniband/hw/qedr/qedr_cm.h | 40 +
> drivers/infiniband/hw/qedr/qedr_hsi_rdma.h | 11 +
> drivers/infiniband/hw/qedr/qedr_user.h | 34 +
> drivers/infiniband/hw/qedr/verbs.c | 1098 ++++++++++++++++++++++++-
> drivers/infiniband/hw/qedr/verbs.h | 7 +
> drivers/net/ethernet/qlogic/qed/qed_cxt.h | 1 +
> drivers/net/ethernet/qlogic/qed/qed_roce.c | 1211 ++++++++++++++++++++++++++++
> drivers/net/ethernet/qlogic/qed/qed_roce.h | 71 ++
> include/linux/qed/qed_roce_if.h | 144 ++++
> 11 files changed, 2756 insertions(+), 2 deletions(-)
> create mode 100644 drivers/infiniband/hw/qedr/qedr_cm.h
>
<....>
> +
> +static inline int get_gid_info_from_table(struct ib_qp *ibqp,
> + struct ib_qp_attr *attr,
> + int attr_mask,
> + struct qed_rdma_modify_qp_in_params
> + *qp_params)
> +{
> + enum rdma_network_type nw_type;
> + struct ib_gid_attr gid_attr;
> + union ib_gid gid;
> + u32 ipv4_addr;
> + int rc = 0;
> + int i;
> +
> + rc = ib_get_cached_gid(ibqp->device, attr->ah_attr.port_num,
> + attr->ah_attr.grh.sgid_index, &gid, &gid_attr);
> + if (!rc && !memcmp(&gid, &zgid, sizeof(gid)))
> + rc = -ENOENT;
> +
> + if (!rc && gid_attr.ndev) {
> + qp_params->vlan_id = rdma_vlan_dev_vlan_id(gid_attr.ndev);
> +
> + dev_put(gid_attr.ndev);
> + nw_type = ib_gid_to_network_type(gid_attr.gid_type, &gid);
> + switch (nw_type) {
> + case RDMA_NETWORK_IPV6:
> + memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
> + sizeof(qp_params->sgid));
> + memcpy(&qp_params->dgid.bytes[0],
> + &attr->ah_attr.grh.dgid,
> + sizeof(qp_params->dgid));
> + qp_params->roce_mode = ROCE_V2_IPV6;
> + SET_FIELD(qp_params->modify_flags,
> + QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1);
> + break;
> + case RDMA_NETWORK_IB:
> + memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
> + sizeof(qp_params->sgid));
> + memcpy(&qp_params->dgid.bytes[0],
> + &attr->ah_attr.grh.dgid,
> + sizeof(qp_params->dgid));
> + qp_params->roce_mode = ROCE_V1;
> + break;
> + case RDMA_NETWORK_IPV4:
> + memset(&qp_params->sgid, 0, sizeof(qp_params->sgid));
> + memset(&qp_params->dgid, 0, sizeof(qp_params->dgid));
> + ipv4_addr = qedr_get_ipv4_from_gid(gid.raw);
> + qp_params->sgid.ipv4_addr = ipv4_addr;
> + ipv4_addr =
> + qedr_get_ipv4_from_gid(attr->ah_attr.grh.dgid.raw);
> + qp_params->dgid.ipv4_addr = ipv4_addr;
> + SET_FIELD(qp_params->modify_flags,
> + QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1);
> + qp_params->roce_mode = ROCE_V2_IPV4;
> + break;
> + }
> + }
> + if (rc)
> + return -EINVAL;
I think it is better to check "rc" right after call to
ib_get_cached_gid().
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 03/11] Add support for RoCE HW init
2016-09-13 14:38 ` Sagi Grimberg
2016-09-14 10:13 ` Amrani, Ram
@ 2016-09-19 8:45 ` Amrani, Ram
1 sibling, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-19 8:45 UTC (permalink / raw)
To: Sagi Grimberg
Cc: Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA,
dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q
> > > + dev->max_sge = min_t(u32, RDMA_MAX_SGE_PER_SQ_WQE,
> > > + RDMA_MAX_SGE_PER_RQ_WQE);
> >
> > Our kernel target mode consumers sort of rely on max_sge_rd, you need
> > to make sure to set it too.
> Good catch. Thanks!
Actually, as I come to code this, I just realized that max_sge_rd is already set in the query verb itself:
+int qedr_query_device(struct ib_device *ibdev,
+ struct ib_device_attr *attr, struct ib_udata *udata)
+{
<snip>
+ attr->max_sge = qattr->max_sge;
+ attr->max_sge_rd = qattr->max_sge;
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 06/11] Add support for QP verbs
[not found] ` <20160915130556.GA26069-2ukJVAZIZ/Y@public.gmane.org>
@ 2016-09-19 9:00 ` Amrani, Ram
0 siblings, 0 replies; 78+ messages in thread
From: Amrani, Ram @ 2016-09-19 9:00 UTC (permalink / raw)
To: Leon Romanovsky
Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA, davem-fT/PcQaiUtIeIZ0/mPfg9Q,
Yuval.Mintz-h88ZbnxC6KDQT0dZR+AlfA,
Ariel.Elior-h88ZbnxC6KDQT0dZR+AlfA,
Michal.Kalderon-h88ZbnxC6KDQT0dZR+AlfA,
rajesh.borundia-h88ZbnxC6KDQT0dZR+AlfA,
linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev-u79uwXL29TY76Z2rM5mHXA
> > +
> > + rc = ib_get_cached_gid(ibqp->device, attr->ah_attr.port_num,
> > + attr->ah_attr.grh.sgid_index, &gid, &gid_attr);
> > + if (!rc && !memcmp(&gid, &zgid, sizeof(gid)))
> > + rc = -ENOENT;
> > +
> > + if (!rc && gid_attr.ndev) {
<snip>
> > + }
> > + }
> > + if (rc)
> > + return -EINVAL;
>
> I think it is better to check "rc" right after call to ib_get_cached_gid().
Somehow this e-mail slipped.
Thanks, will update.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 78+ messages in thread
* RE: [RFC 02/11] Add RoCE driver framework
2016-09-15 5:42 ` Leon Romanovsky
@ 2016-09-20 15:04 ` Elior, Ariel
[not found] ` <CY1PR0701MB13376438A5767BC844EA35C290F70-UpKza+2NMNLi6bjPjkn3FE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
0 siblings, 1 reply; 78+ messages in thread
From: Elior, Ariel @ 2016-09-20 15:04 UTC (permalink / raw)
To: Leon Romanovsky, Mintz, Yuval
Cc: dledford, Mark Bloch, Ram Amrani, David Miller, Ariel Elior,
Michal Kalderon, Rajesh Borundia, linux-rdma, netdev
> From: Leon Romanovsky [mailto:leon@kernel.org]
> On Thu, Sep 15, 2016 at 05:11:03AM +0000, Mintz, Yuval wrote:
> > > As a summary, I didn't see in your responses any real life example where you will
> > > need global debug level for your driver.
> >
> > Not sure what you you're expecting - a list of BZs /private e-mails where
> > user reproductions were needed?
> > You're basically ignoring my claims that such are used, instead wanting
> > "evidence". I'm not going to try and produce any such.
>
> I asked an example and not evidence, where "modprobe your_driver
> debug=1" will be superior to "modprobe your_driver dyndbg==pmf".
>
> https://www.kernel.org/doc/Documentation/dynamic-debug-howto.txt
Hi,
dyndbg vs module param:
Dynamic debug has two very nice features: dynamic activation/configuration and per line/file/module/format activation. The module param has neither, but it does have a few merits which I am not sure dyndbg has:
(1)
It can activate printouts according to *flow*. A lot of thought has been put into associating the right printout in our driver at the right verbosity level with the right "flow tag" (e.g. QEDR_MSG_INIT, QEDR_MSG_QP). The module parameter accepts a bitmask which allows setting any subset of these flows. This means that with the correct values for the parameter I can open only "init" printouts, or only "Memory Region" printouts, even if these cross multiple files / functions and don't share a common format. Presumably, one would claim that we could add the "flow tag" to the format to every printout according to its flow, but that would encumber the printouts, and also doesn't scale well to printouts which belong to multiple flows, where the current approach allows this (QEDR_MSG_SQ | QEDR_MSG_RQ).
(2)
As Yuval pointed out, there are users out there which have no trouble loading a driver with a module parameter, at probe or at kernel boot, but would be at a loss in mounting debugfs and dumping a matchspec script into a node there. As kernel developers, educating users is part of what we do, but it comes with a cost.
(3)
debugfs can be compiled out of the kernel in some codesize sensitive environments, or may not exist in an emergency shell or kdump kernel, whereas the module parameter would always be there.
Simply allowing our module parameter mechanism to be dynamically activated is very straightforward - we planned on adding a debugfs node for that anyway. But I would keep the module parameter for the sake of those less capable users and also for when debugfs is not available as detailed in (3) above.
I certainly see the merits of joining an existing infrastructure instead of implementing our own thing, but I would like to know if there are ways of obtaining the merits I listed for our approach via that infrastructure.
Leon, aside of commenting on the above, can you give an example of a driver which in your opinion does a good job of using dyndbg?
Thanks,
Ariel
^ permalink raw reply [flat|nested] 78+ messages in thread
* Re: [RFC 02/11] Add RoCE driver framework
[not found] ` <CY1PR0701MB13376438A5767BC844EA35C290F70-UpKza+2NMNLi6bjPjkn3FE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
@ 2016-09-20 17:03 ` Leon Romanovsky
0 siblings, 0 replies; 78+ messages in thread
From: Leon Romanovsky @ 2016-09-20 17:03 UTC (permalink / raw)
To: Elior, Ariel
Cc: Mintz, Yuval, dledford-H+wXaHxf7aLQT0dZR+AlfA, Mark Bloch,
Ram Amrani, David Miller, Ariel Elior, Michal Kalderon,
Rajesh Borundia, linux-rdma-u79uwXL29TY76Z2rM5mHXA, netdev
[-- Attachment #1: Type: text/plain, Size: 3713 bytes --]
On Tue, Sep 20, 2016 at 03:04:12PM +0000, Elior, Ariel wrote:
> > From: Leon Romanovsky [mailto:leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org]
> > On Thu, Sep 15, 2016 at 05:11:03AM +0000, Mintz, Yuval wrote:
> > > > As a summary, I didn't see in your responses any real life example where you will
> > > > need global debug level for your driver.
> > >
> > > Not sure what you you're expecting - a list of BZs /private e-mails where
> > > user reproductions were needed?
> > > You're basically ignoring my claims that such are used, instead wanting
> > > "evidence". I'm not going to try and produce any such.
> >
> > I asked an example and not evidence, where "modprobe your_driver
> > debug=1" will be superior to "modprobe your_driver dyndbg==pmf".
> >
> > https://www.kernel.org/doc/Documentation/dynamic-debug-howto.txt
> Hi,
>
> dyndbg vs module param:
> Dynamic debug has two very nice features: dynamic activation/configuration and per line/file/module/format activation. The module param has neither, but it does have a few merits which I am not sure dyndbg has:
> (1)
> It can activate printouts according to *flow*. A lot of thought has been put into associating the right printout in our driver at the right verbosity level with the right "flow tag" (e.g. QEDR_MSG_INIT, QEDR_MSG_QP). The module parameter accepts a bitmask which allows setting any subset of these flows. This means that with the correct values for the parameter I can open only "init" printouts, or only "Memory Region" printouts, even if these cross multiple files / functions and don't share a common format. Presumably, one would claim that we could add the "flow tag" to the format to every printout according to its flow, but that would encumber the printouts, and also doesn't scale well to printouts which belong to multiple flows, where the current approach allows this (QEDR_MSG_SQ | QEDR_
MSG_RQ).
Dynamic prints are enabled per-print. The best possible granularity
which you can achieve.
> (2)
> As Yuval pointed out, there are users out there which have no trouble loading a driver with a module parameter, at probe or at kernel boot, but would be at a loss in mounting debugfs and dumping a matchspec script into a node there. As kernel developers, educating users is part of what we do, but it comes with a cost.
You are free to add this parameter to your out-of-tree driver. As I said
before to Yuval your module "debug" is equal to "dyndbg". Your users can
use it and is already available in kernel.
> (3)
> debugfs can be compiled out of the kernel in some codesize sensitive environments, or may not exist in an emergency shell or kdump kernel, whereas the module parameter would always be there.
>
> Simply allowing our module parameter mechanism to be dynamically activated is very straightforward - we planned on adding a debugfs node for that anyway. But I would keep the module parameter for the sake of those less capable users and also for when debugfs is not available as detailed in (3) above.
This is an example why we don't want this module parameter, once added it
will be with us forever.
>
> I certainly see the merits of joining an existing infrastructure instead of implementing our own thing, but I would like to know if there are ways of obtaining the merits I listed for our approach via that infrastructure.
>
> Leon, aside of commenting on the above, can you give an example of a driver which in your opinion does a good job of using dyndbg?
All modern drivers in this subsystem are supporting dyndbg out of the
box.
General note, can you please configure your email to wrap lines? It is
very hard to read, follow and respond to lines with >800 symbols in them.
>
> Thanks,
> Ariel
>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]
^ permalink raw reply [flat|nested] 78+ messages in thread
end of thread, other threads:[~2016-09-20 17:03 UTC | newest]
Thread overview: 78+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-09-12 16:07 [RFC 00/11] QLogic RDMA Driver (qedr) RFC Ram Amrani
2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` [RFC 03/11] Add support for RoCE HW init Ram Amrani
2016-09-12 16:07 ` Ram Amrani
[not found] ` <1473696465-27986-4-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-12 18:57 ` Mark Bloch
2016-09-12 18:57 ` Mark Bloch
2016-09-13 8:30 ` Ram Amrani
2016-09-13 14:38 ` Sagi Grimberg
2016-09-14 10:13 ` Amrani, Ram
2016-09-19 8:45 ` Amrani, Ram
2016-09-12 16:07 ` [RFC 04/11] Add support for user context verbs Ram Amrani
2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` [RFC 05/11] Add support for PD,PKEY and CQ verbs Ram Amrani
2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` [RFC 06/11] Add support for QP verbs Ram Amrani
2016-09-12 16:07 ` Ram Amrani
[not found] ` <1473696465-27986-7-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-15 13:05 ` Leon Romanovsky
[not found] ` <20160915130556.GA26069-2ukJVAZIZ/Y@public.gmane.org>
2016-09-19 9:00 ` Amrani, Ram
2016-09-12 16:07 ` [RFC 08/11] Add support for data path Ram Amrani
2016-09-12 16:07 ` Ram Amrani
[not found] ` <1473696465-27986-9-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-13 14:32 ` Sagi Grimberg
2016-09-14 10:16 ` Amrani, Ram
2016-09-15 7:24 ` Leon Romanovsky
2016-09-12 16:07 ` [RFC 09/11] Add LL2 RoCE interface Ram Amrani
2016-09-12 16:07 ` Ram Amrani
2016-09-15 10:20 ` Leon Romanovsky
[not found] ` <20160915102013.GZ26069-2ukJVAZIZ/Y@public.gmane.org>
2016-09-15 12:07 ` Amrani, Ram
2016-09-12 16:07 ` [RFC 10/11] Add GSI support Ram Amrani
2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` [RFC 11/11] Add events support and register IB device Ram Amrani
2016-09-12 16:07 ` Ram Amrani
[not found] ` <1473696465-27986-1-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-12 16:07 ` [RFC 01/11] qed: Add LL2 Ram Amrani
2016-09-12 16:07 ` Ram Amrani
2016-09-12 16:07 ` [RFC 02/11] Add RoCE driver framework Ram Amrani
2016-09-12 16:07 ` Ram Amrani
[not found] ` <1473696465-27986-3-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-12 18:44 ` Mark Bloch
2016-09-12 18:44 ` Mark Bloch
[not found] ` <516b98c7-477a-4890-0d92-529dc32f2c4e-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
2016-09-12 19:17 ` Yuval Mintz
[not found] ` <CY4PR11MB1720F9B74BBF82D0779FAC8F97FF0-JNf6+SjKdlG0ooKL/ADlEpPPoyLQLiKMvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
2016-09-13 6:38 ` Leon Romanovsky
2016-09-13 7:18 ` Mintz, Yuval
2016-09-13 10:16 ` Leon Romanovsky
[not found] ` <20160913101616.GT8812-2ukJVAZIZ/Y@public.gmane.org>
2016-09-14 8:15 ` Mintz, Yuval
2016-09-14 13:00 ` Leon Romanovsky
[not found] ` <20160914130032.GB26069-2ukJVAZIZ/Y@public.gmane.org>
2016-09-14 18:25 ` Mintz, Yuval
[not found] ` <BL2PR07MB23066BBC4A019365A632E2FA8DF10-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
2016-09-15 4:37 ` Leon Romanovsky
[not found] ` <20160915043716.GD26069-2ukJVAZIZ/Y@public.gmane.org>
2016-09-15 5:11 ` Mintz, Yuval
[not found] ` <BL2PR07MB23060C776EDAE92B84FCFB768DF00-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
2016-09-15 5:42 ` Leon Romanovsky
2016-09-20 15:04 ` Elior, Ariel
[not found] ` <CY1PR0701MB13376438A5767BC844EA35C290F70-UpKza+2NMNLi6bjPjkn3FE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2016-09-20 17:03 ` Leon Romanovsky
2016-09-13 9:22 ` Ram Amrani
[not found] ` <DM3PR1101MB1181337A65B7CEFC1A30F353E2FE0-xYdf0wd+uoGW1Nawvih6nR68uu4wjhmwnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2016-09-13 10:22 ` Leon Romanovsky
2016-09-13 6:46 ` Leon Romanovsky
2016-09-13 14:46 ` Steve Wise
2016-09-13 14:46 ` Steve Wise
2016-09-14 7:30 ` Amrani, Ram
2016-09-12 16:07 ` [RFC 07/11] Add support for memory registeration verbs Ram Amrani
2016-09-12 16:07 ` Ram Amrani
[not found] ` <1473696465-27986-8-git-send-email-Ram.Amrani-h88ZbnxC6KDQT0dZR+AlfA@public.gmane.org>
2016-09-13 14:26 ` Sagi Grimberg
2016-09-14 8:02 ` Amrani, Ram
2016-09-13 14:44 ` Sagi Grimberg
[not found] ` <7fa4a9b8-7cb1-0f83-d6e5-1055ae59bce4-NQWnxTmZq1alnMjI0IkVqw@public.gmane.org>
2016-09-14 8:59 ` Kalderon, Michal
2016-09-14 9:25 ` Sagi Grimberg
2016-09-14 10:02 ` Kalderon, Michal
2016-09-12 16:39 ` [RFC 00/11] QLogic RDMA Driver (qedr) RFC Leon Romanovsky
[not found] ` <20160912163928.GK8812-2ukJVAZIZ/Y@public.gmane.org>
2016-09-12 16:49 ` Parav Pandit
2016-09-12 17:39 ` Yuval Mintz
[not found] ` <CY4PR11MB17202A542AC07CBA4A65E93697FF0-JNf6+SjKdlG0ooKL/ADlEpPPoyLQLiKMvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
2016-09-13 6:05 ` Leon Romanovsky
[not found] ` <20160913060545.GN8812-2ukJVAZIZ/Y@public.gmane.org>
2016-09-13 6:48 ` Mintz, Yuval
[not found] ` <BL2PR07MB23064356EA2492675AC980A78DFE0-I6Fv6QFlT9L2NWYWB7JgfuFPX92sqiQdvxpqHgZTriW3zl9H0oFU5g@public.gmane.org>
2016-09-13 10:39 ` Leon Romanovsky
2016-09-12 18:05 ` Jason Gunthorpe
[not found] ` <20160912180508.GI5843-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
2016-09-13 8:44 ` Ram Amrani
[not found] ` <DM3PR1101MB1181DF570D4F01A0CEB1CFC7E2FE0-xYdf0wd+uoGW1Nawvih6nR68uu4wjhmwnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2016-09-13 10:19 ` Leon Romanovsky
2016-09-13 15:40 ` Jason Gunthorpe
[not found] ` <20160913154000.GA25878-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
2016-09-14 14:44 ` Amrani, Ram
2016-09-14 17:17 ` Jason Gunthorpe
[not found] ` <20160914171737.GH16014-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>
2016-09-15 5:55 ` Amrani, Ram
2016-09-13 14:23 ` Sagi Grimberg
[not found] ` <CY1PR0701MB133732FA8478FC0B5003D97A90F10@CY1PR0701MB1337.namprd07.prod.outlook.com>
[not found] ` <CY1PR0701MB133732FA8478FC0B5003D97A90F10-UpKza+2NMNLi6bjPjkn3FE5OhdzP3rhOnBOFsp37pqbUKgpGm//BTAC/G2K4zDHf@public.gmane.org>
2016-09-14 8:17 ` Sagi Grimberg
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.