* [PATCH for-next 01/16] IB/hfi1: Do not override given pcie_pset value
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
@ 2018-01-26 14:31 ` Dennis Dalessandro
2018-01-26 14:31 ` [PATCH for-next 02/16] IB/hfi1: Fix for early release of sdma context Dennis Dalessandro
` (14 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:31 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Bartlomiej Dudek, Patel Jay P
From: Bartlomiej Dudek <bartlomiej.dudek-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
During PCIe Gen 3 transistion, pcie_pset is read and might be overridden
to a default value(i.e. 255) in do_pcie_gen3_transition() routine.
If the pcie_pset value is overridden then this new value will be used
during initialization of next adapter on a different card.
Introducing a new local variable to avoid modification of pcie_pset
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Bartlomiej Dudek <bartlomiej.dudek-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Patel Jay P <jay.p.patel-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/pcie.c | 23 ++++++++++++-----------
1 files changed, 12 insertions(+), 11 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/pcie.c b/drivers/infiniband/hw/hfi1/pcie.c
index 8c7e7a6..83d66e8 100644
--- a/drivers/infiniband/hw/hfi1/pcie.c
+++ b/drivers/infiniband/hw/hfi1/pcie.c
@@ -1034,6 +1034,7 @@ int do_pcie_gen3_transition(struct hfi1_devdata *dd)
int do_retry, retry_count = 0;
int intnum = 0;
uint default_pset;
+ uint pset = pcie_pset;
u16 target_vector, target_speed;
u16 lnkctl2, vendor;
u8 div;
@@ -1201,16 +1202,16 @@ int do_pcie_gen3_transition(struct hfi1_devdata *dd)
*
* Set Gen3EqPsetReqVec, leave other fields 0.
*/
- if (pcie_pset == UNSET_PSET)
- pcie_pset = default_pset;
- if (pcie_pset > 10) { /* valid range is 0-10, inclusive */
+ if (pset == UNSET_PSET)
+ pset = default_pset;
+ if (pset > 10) { /* valid range is 0-10, inclusive */
dd_dev_err(dd, "%s: Invalid Eq Pset %u, setting to %d\n",
- __func__, pcie_pset, default_pset);
- pcie_pset = default_pset;
+ __func__, pset, default_pset);
+ pset = default_pset;
}
- dd_dev_info(dd, "%s: using EQ Pset %u\n", __func__, pcie_pset);
+ dd_dev_info(dd, "%s: using EQ Pset %u\n", __func__, pset);
pci_write_config_dword(dd->pcidev, PCIE_CFG_REG_PL106,
- ((1 << pcie_pset) <<
+ ((1 << pset) <<
PCIE_CFG_REG_PL106_GEN3_EQ_PSET_REQ_VEC_SHIFT) |
PCIE_CFG_REG_PL106_GEN3_EQ_EVAL2MS_DISABLE_SMASK |
PCIE_CFG_REG_PL106_GEN3_EQ_PHASE23_EXIT_MODE_SMASK);
@@ -1240,10 +1241,10 @@ int do_pcie_gen3_transition(struct hfi1_devdata *dd)
/* apply static CTLE tunings */
u8 pcie_dc, pcie_lf, pcie_hf, pcie_bw;
- pcie_dc = ctle_tunings[pcie_pset][0];
- pcie_lf = ctle_tunings[pcie_pset][1];
- pcie_hf = ctle_tunings[pcie_pset][2];
- pcie_bw = ctle_tunings[pcie_pset][3];
+ pcie_dc = ctle_tunings[pset][0];
+ pcie_lf = ctle_tunings[pset][1];
+ pcie_hf = ctle_tunings[pset][2];
+ pcie_bw = ctle_tunings[pset][3];
write_gasket_interrupt(dd, intnum++, 0x0026, 0x0200 | pcie_dc);
write_gasket_interrupt(dd, intnum++, 0x0026, 0x0100 | pcie_lf);
write_gasket_interrupt(dd, intnum++, 0x0026, 0x0000 | pcie_hf);
--
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] 27+ messages in thread* [PATCH for-next 02/16] IB/hfi1: Fix for early release of sdma context
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2018-01-26 14:31 ` [PATCH for-next 01/16] IB/hfi1: Do not override given pcie_pset value Dennis Dalessandro
@ 2018-01-26 14:31 ` Dennis Dalessandro
2018-01-26 14:31 ` [PATCH for-next 03/16] IB/hfi1: Remove dependence on qp->s_hdrwords Dennis Dalessandro
` (13 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:31 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Michael J Ruhl,
Mike Marciniszyn, Alex Estrin
From: Alex Estrin <alex.estrin-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
With IRQF_SHARED flag set and CONFIG_DEBUG_SHIRQ enabled
module removal may result in panic in sdma_interrupt() routine
if associated sdma context was released before pci_free_irq();
[ 9198.939885] BUG: unable to handle kernel NULL pointer dereference at (null)
[ 9198.940514] IP: sdma_make_progress+0xa5/0x450 [hfi1]
[ 9198.941114] PGD 170bdc0067 P4D 170bdc0067 PUD 172063e067 PMD 0
[ 9198.941783] Oops: 0000 [#1] SMP
.....
[ 9198.958877] CPU: 132 PID: 64173 Comm: rmmod Tainted: G OE 4.14.0-rc4+ #1
[ 9198.961032] Hardware name: Intel Corporation S7200AP/S7200AP, BIOS S72C610.86B.01.02.0118.080620171935 08/06/2017
[ 9198.963323] task: ffff9681397f0000 task.stack: ffffae1647c40000
[ 9198.965695] RIP: 0010:sdma_make_progress+0xa5/0x450 [hfi1]
[ 9198.968082] RSP: 0018:ffffae1647c43be8 EFLAGS: 00010046
[ 9198.970503] RAX: 0000000000000000 RBX: ffff9680ce8b5ca8 RCX: 0000000000000000
[ 9198.973006] RDX: 0000000000000000 RSI: 0000000001a00d28 RDI: ffff9680ce8b5ca0
[ 9198.975546] RBP: ffffae1647c43c40 R08: ffff96814325ec00 R09: 00000000ffffffff
[ 9198.978142] R10: 000000004325e501 R11: ffff96814325ec00 R12: ffff9680ce8b5c44
[ 9198.980779] R13: ffff9680ce8b5ca0 R14: 0000000000000000 R15: ffff9680ce8b5b00
[ 9198.983462] FS: 00007f31196ba740(0000) GS:ffff96819df00000(0000) knlGS:0000000000000000
[ 9198.986231] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 9198.989036] CR2: 0000000000000000 CR3: 000000170833f000 CR4: 00000000001406e0
[ 9198.991911] Call Trace:
[ 9198.994847] sdma_engine_interrupt+0x82/0x100 [hfi1]
[ 9198.997852] sdma_interrupt+0x61/0xc0 [hfi1]
[ 9199.000852] __free_irq+0x1b3/0x2d0
[ 9199.003873] free_irq+0x35/0x70
[ 9199.006909] pci_free_irq+0x1c/0x30
[ 9199.009999] clean_up_interrupts+0x53/0xf0 [hfi1]
[ 9199.013137] hfi1_start_cleanup+0x117/0x190 [hfi1]
[ 9199.016315] postinit_cleanup+0x1d/0x270 [hfi1]
[ 9199.019529] remove_one+0x1f3/0x210 [hfi1]
[ 9199.022738] pci_device_remove+0x39/0xc0
[ 9199.025974] device_release_driver_internal+0x141/0x210
[ 9199.029268] driver_detach+0x3f/0x80
[ 9199.032580] bus_remove_driver+0x55/0xd0
[ 9199.035931] driver_unregister+0x2c/0x50
[ 9199.039321] pci_unregister_driver+0x2a/0xa0
[ 9199.042755] hfi1_mod_cleanup+0x10/0xb50 [hfi1]
[ 9199.046196] SyS_delete_module+0x171/0x250
...
Fix by exporting sdma_clean() and removing from sdma_exit().
sdma_exit() now just manipulates the engine state,
leaving the memory free to sdma_clean() which is now called
just before the dd is freed.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Michael J Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Alex Estrin <alex.estrin-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/init.c | 1 +
drivers/infiniband/hw/hfi1/sdma.c | 13 +++++++------
drivers/infiniband/hw/hfi1/sdma.h | 1 +
3 files changed, 9 insertions(+), 6 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index 9b12826..a618eef 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -1218,6 +1218,7 @@ static void __hfi1_free_devdata(struct kobject *kobj)
free_percpu(dd->rcv_limit);
free_percpu(dd->send_schedule);
free_percpu(dd->tx_opstats);
+ sdma_clean(dd, dd->num_sdma);
rvt_dealloc_device(&dd->verbs_dev.rdi);
}
diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c
index 31c8f89..37424a8 100644
--- a/drivers/infiniband/hw/hfi1/sdma.c
+++ b/drivers/infiniband/hw/hfi1/sdma.c
@@ -1276,13 +1276,15 @@ int sdma_map_init(struct hfi1_devdata *dd, u8 port, u8 num_vls, u8 *vl_engines)
return -ENOMEM;
}
-/*
- * Clean up allocated memory.
- *
- * This routine is can be called regardless of the success of sdma_init()
+/**
+ * sdma_clean() Clean up allocated memory
+ * @dd: struct hfi1_devdata
+ * @num_engines: num sdma engines
*
+ * This routine can be called regardless of the success of
+ * sdma_init()
*/
-static void sdma_clean(struct hfi1_devdata *dd, size_t num_engines)
+void sdma_clean(struct hfi1_devdata *dd, size_t num_engines)
{
size_t i;
struct sdma_engine *sde;
@@ -1618,7 +1620,6 @@ void sdma_exit(struct hfi1_devdata *dd)
*/
sdma_finalput(&sde->state);
}
- sdma_clean(dd, dd->num_sdma);
}
/*
diff --git a/drivers/infiniband/hw/hfi1/sdma.h b/drivers/infiniband/hw/hfi1/sdma.h
index 374c597..46c775f 100644
--- a/drivers/infiniband/hw/hfi1/sdma.h
+++ b/drivers/infiniband/hw/hfi1/sdma.h
@@ -420,6 +420,7 @@ struct sdma_engine {
int sdma_init(struct hfi1_devdata *dd, u8 port);
void sdma_start(struct hfi1_devdata *dd);
void sdma_exit(struct hfi1_devdata *dd);
+void sdma_clean(struct hfi1_devdata *dd, size_t num_engines);
void sdma_all_running(struct hfi1_devdata *dd);
void sdma_all_idle(struct hfi1_devdata *dd);
void sdma_freeze_notify(struct hfi1_devdata *dd, int go_idle);
--
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] 27+ messages in thread* [PATCH for-next 03/16] IB/hfi1: Remove dependence on qp->s_hdrwords
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2018-01-26 14:31 ` [PATCH for-next 01/16] IB/hfi1: Do not override given pcie_pset value Dennis Dalessandro
2018-01-26 14:31 ` [PATCH for-next 02/16] IB/hfi1: Fix for early release of sdma context Dennis Dalessandro
@ 2018-01-26 14:31 ` Dennis Dalessandro
2018-01-26 14:31 ` [PATCH for-next 04/16] IB/hfi1: Remove blind constants from 16B update Dennis Dalessandro
` (12 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:31 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Don Hiatt, Mitko Haralanov,
Mike Marciniszyn
From: Mitko Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
The s_hdrwords variable was used to indicate whether a
packet was already built on a previous iteration of the
send engine. This variable assumed the protection of the
QP's RVT_S_BUSY flag, which was required since the the
QP's s_lock was dropped just prior to the packet being
queued on the one of the egress mechanisms.
Support for multiple send engine instantiations require
that the field not be used due to concurency issues.
The ps.txreq signals the "already built" without the
potential concurency issues.
Fix by getting rid of all s_hdrword usage. A wrapper
is added to test for the already built case that used to
use s_hdrwords.
What used to be stored in s_hdrwords is now in the txreq.
The PBC is not counted, but is added in the pio/sdma code
paths prior to posting the packet.
Reviewed-by: Don Hiatt <don.hiatt-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mitko Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/iowait.h | 9 +++++++++
drivers/infiniband/hw/hfi1/qp.c | 4 +---
drivers/infiniband/hw/hfi1/qp.h | 13 +++++++++++++
drivers/infiniband/hw/hfi1/rc.c | 10 ++--------
drivers/infiniband/hw/hfi1/ruc.c | 23 +++++++++++------------
drivers/infiniband/hw/hfi1/uc.c | 6 +-----
drivers/infiniband/hw/hfi1/ud.c | 27 ++++++++++++---------------
drivers/infiniband/hw/hfi1/verbs.c | 10 +++++-----
drivers/infiniband/hw/hfi1/verbs.h | 11 -----------
drivers/infiniband/hw/hfi1/verbs_txreq.h | 7 +++++++
10 files changed, 61 insertions(+), 59 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/iowait.h b/drivers/infiniband/hw/hfi1/iowait.h
index 591697d..3d9c32c 100644
--- a/drivers/infiniband/hw/hfi1/iowait.h
+++ b/drivers/infiniband/hw/hfi1/iowait.h
@@ -371,4 +371,13 @@ static inline void iowait_starve_find_max(struct iowait *w, u8 *max,
}
}
+/**
+ * iowait_packet_queued() - determine if a packet is already built
+ * @wait: the wait structure
+ */
+static inline bool iowait_packet_queued(struct iowait *wait)
+{
+ return !list_empty(&wait->tx_head);
+}
+
#endif
diff --git a/drivers/infiniband/hw/hfi1/qp.c b/drivers/infiniband/hw/hfi1/qp.c
index 5507910..d30dd1a 100644
--- a/drivers/infiniband/hw/hfi1/qp.c
+++ b/drivers/infiniband/hw/hfi1/qp.c
@@ -565,7 +565,7 @@ void qp_iter_print(struct seq_file *s, struct rvt_qp_iter *iter)
if (qp->s_ack_queue)
e = &qp->s_ack_queue[qp->s_tail_ack_queue];
seq_printf(s,
- "N %d %s QP %x R %u %s %u %u %u f=%x %u %u %u %u %u %u SPSN %x %x %x %x %x RPSN %x S(%u %u %u %u %u %u %u) R(%u %u %u) RQP %x LID %x SL %u MTU %u %u %u %u %u SDE %p,%u SC %p,%u SCQ %u %u PID %d OS %x %x E %x %x %x RNR %d %s %d\n",
+ "N %d %s QP %x R %u %s %u %u f=%x %u %u %u %u %u %u SPSN %x %x %x %x %x RPSN %x S(%u %u %u %u %u %u %u) R(%u %u %u) RQP %x LID %x SL %u MTU %u %u %u %u %u SDE %p,%u SC %p,%u SCQ %u %u PID %d OS %x %x E %x %x %x RNR %d %s %d\n",
iter->n,
qp_idle(qp) ? "I" : "B",
qp->ibqp.qp_num,
@@ -573,7 +573,6 @@ void qp_iter_print(struct seq_file *s, struct rvt_qp_iter *iter)
qp_type_str[qp->ibqp.qp_type],
qp->state,
wqe ? wqe->wr.opcode : 0,
- qp->s_hdrwords,
qp->s_flags,
iowait_sdma_pending(&priv->s_iowait),
iowait_pio_pending(&priv->s_iowait),
@@ -795,7 +794,6 @@ void notify_error_qp(struct rvt_qp *qp)
}
if (!(qp->s_flags & RVT_S_BUSY)) {
- qp->s_hdrwords = 0;
if (qp->s_rdma_mr) {
rvt_put_mr(qp->s_rdma_mr);
qp->s_rdma_mr = NULL;
diff --git a/drivers/infiniband/hw/hfi1/qp.h b/drivers/infiniband/hw/hfi1/qp.h
index c06d2f8..b2d4cba 100644
--- a/drivers/infiniband/hw/hfi1/qp.h
+++ b/drivers/infiniband/hw/hfi1/qp.h
@@ -51,12 +51,25 @@
#include <rdma/rdmavt_qp.h>
#include "verbs.h"
#include "sdma.h"
+#include "verbs_txreq.h"
extern unsigned int hfi1_qp_table_size;
extern const struct rvt_operation_params hfi1_post_parms[];
/*
+ * Send if not busy or waiting for I/O and either
+ * a RC response is pending or we can process send work requests.
+ */
+static inline int hfi1_send_ok(struct rvt_qp *qp)
+{
+ return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) &&
+ (verbs_txreq_queued(qp) ||
+ (qp->s_flags & RVT_S_RESP_PENDING) ||
+ !(qp->s_flags & RVT_S_ANY_WAIT_SEND));
+}
+
+/*
* free_ahg - clear ahg from QP
*/
static inline void clear_ahg(struct rvt_qp *qp)
diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c
index 68d5c3c..524c12f 100644
--- a/drivers/infiniband/hw/hfi1/rc.c
+++ b/drivers/infiniband/hw/hfi1/rc.c
@@ -226,12 +226,10 @@ static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp,
bth2 = mask_psn(qp->s_ack_psn);
}
qp->s_rdma_ack_cnt++;
- qp->s_hdrwords = hwords;
ps->s_txreq->sde = priv->s_sde;
ps->s_txreq->s_cur_size = len;
+ ps->s_txreq->hdr_dwords = hwords;
hfi1_make_ruc_header(qp, ohdr, bth0, bth2, middle, ps);
- /* pbc */
- ps->s_txreq->hdr_dwords = qp->s_hdrwords + 2;
return 1;
bail:
@@ -387,7 +385,6 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
: IB_WC_SUCCESS);
if (local_ops)
atomic_dec(&qp->local_ops_pending);
- qp->s_hdrwords = 0;
goto done_free_tx;
}
@@ -690,7 +687,7 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
bth2 |= IB_BTH_REQ_ACK;
}
qp->s_len -= len;
- qp->s_hdrwords = hwords;
+ ps->s_txreq->hdr_dwords = hwords;
ps->s_txreq->sde = priv->s_sde;
ps->s_txreq->ss = ss;
ps->s_txreq->s_cur_size = len;
@@ -701,8 +698,6 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
bth2,
middle,
ps);
- /* pbc */
- ps->s_txreq->hdr_dwords = qp->s_hdrwords + 2;
return 1;
done_free_tx:
@@ -716,7 +711,6 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
bail_no_tx:
ps->s_txreq = NULL;
qp->s_flags &= ~RVT_S_BUSY;
- qp->s_hdrwords = 0;
return 0;
}
diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c
index 2c7fc6e..0cced9a 100644
--- a/drivers/infiniband/hw/hfi1/ruc.c
+++ b/drivers/infiniband/hw/hfi1/ruc.c
@@ -757,8 +757,9 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
u32 slid;
u16 pkey = hfi1_get_pkey(ibp, qp->s_pkey_index);
u8 l4 = OPA_16B_L4_IB_LOCAL;
- u8 extra_bytes = hfi1_get_16b_padding((qp->s_hdrwords << 2),
- ps->s_txreq->s_cur_size);
+ u8 extra_bytes = hfi1_get_16b_padding(
+ (ps->s_txreq->hdr_dwords << 2),
+ ps->s_txreq->s_cur_size);
u32 nwords = SIZE_OF_CRC + ((ps->s_txreq->s_cur_size +
extra_bytes + SIZE_OF_LT) >> 2);
u8 becn = 0;
@@ -778,9 +779,9 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
grd->sgid_index = 0;
grh = &ps->s_txreq->phdr.hdr.opah.u.l.grh;
l4 = OPA_16B_L4_IB_GLOBAL;
- hdrwords = qp->s_hdrwords - 4;
- qp->s_hdrwords += hfi1_make_grh(ibp, grh, grd,
- hdrwords, nwords);
+ hdrwords = ps->s_txreq->hdr_dwords - 4;
+ ps->s_txreq->hdr_dwords += hfi1_make_grh(ibp, grh, grd,
+ hdrwords, nwords);
middle = 0;
}
@@ -814,7 +815,7 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
slid,
opa_get_lid(rdma_ah_get_dlid(&qp->remote_ah_attr),
16B),
- (qp->s_hdrwords + nwords) >> 1,
+ (ps->s_txreq->hdr_dwords + nwords) >> 1,
pkey, becn, 0, l4, priv->s_sc);
}
@@ -834,10 +835,10 @@ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp,
if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)) {
struct ib_grh *grh = &ps->s_txreq->phdr.hdr.ibh.u.l.grh;
- int hdrwords = qp->s_hdrwords - 2;
+ int hdrwords = ps->s_txreq->hdr_dwords - 2;
lrh0 = HFI1_LRH_GRH;
- qp->s_hdrwords +=
+ ps->s_txreq->hdr_dwords +=
hfi1_make_grh(ibp, grh,
rdma_ah_read_grh(&qp->remote_ah_attr),
hdrwords, nwords);
@@ -866,7 +867,7 @@ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp,
hfi1_make_ruc_bth(qp, ohdr, bth0, bth1, bth2);
hfi1_make_ib_hdr(&ps->s_txreq->phdr.hdr.ibh,
lrh0,
- qp->s_hdrwords + nwords,
+ ps->s_txreq->hdr_dwords + nwords,
opa_get_lid(rdma_ah_get_dlid(&qp->remote_ah_attr), 9B),
ppd_from_ibp(ibp)->lid |
rdma_ah_get_path_bits(&qp->remote_ah_attr));
@@ -1031,7 +1032,7 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread)
ps.s_txreq = get_waiting_verbs_txreq(qp);
do {
/* Check for a constructed packet to be sent. */
- if (qp->s_hdrwords != 0) {
+ if (ps.s_txreq) {
spin_unlock_irqrestore(&qp->s_lock, ps.flags);
/*
* If the packet cannot be sent now, return and
@@ -1039,8 +1040,6 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread)
*/
if (hfi1_verbs_send(qp, &ps))
return;
- /* Record that s_ahg is empty. */
- qp->s_hdrwords = 0;
/* allow other tasks to run */
if (schedule_send_yield(qp, &ps))
return;
diff --git a/drivers/infiniband/hw/hfi1/uc.c b/drivers/infiniband/hw/hfi1/uc.c
index 991bbee..fef69d9 100644
--- a/drivers/infiniband/hw/hfi1/uc.c
+++ b/drivers/infiniband/hw/hfi1/uc.c
@@ -146,7 +146,6 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
: IB_WC_SUCCESS);
if (local_ops)
atomic_dec(&qp->local_ops_pending);
- qp->s_hdrwords = 0;
goto done_free_tx;
}
/*
@@ -269,14 +268,12 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
break;
}
qp->s_len -= len;
- qp->s_hdrwords = hwords;
+ ps->s_txreq->hdr_dwords = hwords;
ps->s_txreq->sde = priv->s_sde;
ps->s_txreq->ss = &qp->s_sge;
ps->s_txreq->s_cur_size = len;
hfi1_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24),
mask_psn(qp->s_psn++), middle, ps);
- /* pbc */
- ps->s_txreq->hdr_dwords = qp->s_hdrwords + 2;
return 1;
done_free_tx:
@@ -290,7 +287,6 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
bail_no_tx:
ps->s_txreq = NULL;
qp->s_flags &= ~RVT_S_BUSY;
- qp->s_hdrwords = 0;
return 0;
}
diff --git a/drivers/infiniband/hw/hfi1/ud.c b/drivers/infiniband/hw/hfi1/ud.c
index beb5091..6b5cd3a 100644
--- a/drivers/infiniband/hw/hfi1/ud.c
+++ b/drivers/infiniband/hw/hfi1/ud.c
@@ -340,15 +340,15 @@ void hfi1_make_ud_req_9B(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
extra_bytes = -wqe->length & 3;
nwords = ((wqe->length + extra_bytes) >> 2) + SIZE_OF_CRC;
/* header size in dwords LRH+BTH+DETH = (8+12+8)/4. */
- qp->s_hdrwords = 7;
+ ps->s_txreq->hdr_dwords = 7;
if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM)
- qp->s_hdrwords++;
+ ps->s_txreq->hdr_dwords++;
if (rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH) {
grh = &ps->s_txreq->phdr.hdr.ibh.u.l.grh;
- qp->s_hdrwords += hfi1_make_grh(ibp, grh,
- rdma_ah_read_grh(ah_attr),
- qp->s_hdrwords - 2, nwords);
+ ps->s_txreq->hdr_dwords +=
+ hfi1_make_grh(ibp, grh, rdma_ah_read_grh(ah_attr),
+ ps->s_txreq->hdr_dwords - 2, nwords);
lrh0 = HFI1_LRH_GRH;
ohdr = &ps->s_txreq->phdr.hdr.ibh.u.l.oth;
} else {
@@ -381,7 +381,7 @@ void hfi1_make_ud_req_9B(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
}
}
hfi1_make_bth_deth(qp, wqe, ohdr, &pkey, extra_bytes, false);
- len = qp->s_hdrwords + nwords;
+ len = ps->s_txreq->hdr_dwords + nwords;
/* Setup the packet */
ps->s_txreq->phdr.hdr.hdr_type = HFI1_PKT_TYPE_9B;
@@ -405,12 +405,12 @@ void hfi1_make_ud_req_16B(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
ppd = ppd_from_ibp(ibp);
ah_attr = &ibah_to_rvtah(wqe->ud_wr.ah)->attr;
/* header size in dwords 16B LRH+BTH+DETH = (16+12+8)/4. */
- qp->s_hdrwords = 9;
+ ps->s_txreq->hdr_dwords = 9;
if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM)
- qp->s_hdrwords++;
+ ps->s_txreq->hdr_dwords++;
/* SW provides space for CRC and LT for bypass packets. */
- extra_bytes = hfi1_get_16b_padding((qp->s_hdrwords << 2),
+ extra_bytes = hfi1_get_16b_padding((ps->s_txreq->hdr_dwords << 2),
wqe->length);
nwords = ((wqe->length + extra_bytes + SIZE_OF_LT) >> 2) + SIZE_OF_CRC;
@@ -428,8 +428,8 @@ void hfi1_make_ud_req_16B(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
grd->sgid_index = 0;
}
grh = &ps->s_txreq->phdr.hdr.opah.u.l.grh;
- qp->s_hdrwords += hfi1_make_grh(ibp, grh, grd,
- qp->s_hdrwords - 4, nwords);
+ ps->s_txreq->hdr_dwords += hfi1_make_grh(ibp, grh, grd,
+ ps->s_txreq->hdr_dwords - 4, nwords);
ohdr = &ps->s_txreq->phdr.hdr.opah.u.l.oth;
l4 = OPA_16B_L4_IB_GLOBAL;
} else {
@@ -452,7 +452,7 @@ void hfi1_make_ud_req_16B(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
hfi1_make_bth_deth(qp, wqe, ohdr, &pkey, extra_bytes, true);
/* Convert dwords to flits */
- len = (qp->s_hdrwords + nwords) >> 1;
+ len = (ps->s_txreq->hdr_dwords + nwords) >> 1;
/* Setup the packet */
ps->s_txreq->phdr.hdr.hdr_type = HFI1_PKT_TYPE_16B;
@@ -564,8 +564,6 @@ int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
priv->s_ahg->ahgcount = 0;
priv->s_ahg->ahgidx = 0;
priv->s_ahg->tx_flags = 0;
- /* pbc */
- ps->s_txreq->hdr_dwords = qp->s_hdrwords + 2;
return 1;
@@ -580,7 +578,6 @@ int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
bail_no_tx:
ps->s_txreq = NULL;
qp->s_flags &= ~RVT_S_BUSY;
- qp->s_hdrwords = 0;
return 0;
}
diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c
index b8776a3..471d55c 100644
--- a/drivers/infiniband/hw/hfi1/verbs.c
+++ b/drivers/infiniband/hw/hfi1/verbs.c
@@ -835,7 +835,7 @@ static int build_verbs_tx_desc(
{
int ret = 0;
struct hfi1_sdma_header *phdr = &tx->phdr;
- u16 hdrbytes = tx->hdr_dwords << 2;
+ u16 hdrbytes = (tx->hdr_dwords + sizeof(pbc) / 4) << 2;
u8 extra_bytes = 0;
if (tx->phdr.hdr.hdr_type) {
@@ -901,7 +901,7 @@ int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
{
struct hfi1_qp_priv *priv = qp->priv;
struct hfi1_ahg_info *ahg_info = priv->s_ahg;
- u32 hdrwords = qp->s_hdrwords;
+ u32 hdrwords = ps->s_txreq->hdr_dwords;
u32 len = ps->s_txreq->s_cur_size;
u32 plen;
struct hfi1_ibdev *dev = ps->dev;
@@ -919,7 +919,7 @@ int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
} else {
dwords = (len + 3) >> 2;
}
- plen = hdrwords + dwords + 2;
+ plen = hdrwords + dwords + sizeof(pbc) / 4;
tx = ps->s_txreq;
if (!sdma_txreq_built(&tx->txreq)) {
@@ -1038,7 +1038,7 @@ int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
u64 pbc)
{
struct hfi1_qp_priv *priv = qp->priv;
- u32 hdrwords = qp->s_hdrwords;
+ u32 hdrwords = ps->s_txreq->hdr_dwords;
struct rvt_sge_state *ss = ps->s_txreq->ss;
u32 len = ps->s_txreq->s_cur_size;
u32 dwords;
@@ -1064,7 +1064,7 @@ int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
dwords = (len + 3) >> 2;
hdr = (u32 *)&ps->s_txreq->phdr.hdr.ibh;
}
- plen = hdrwords + dwords + 2;
+ plen = hdrwords + dwords + sizeof(pbc) / 4;
/* only RC/UC use complete */
switch (qp->ibqp.qp_type) {
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index 87d1285..c2aeb32 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -246,17 +246,6 @@ struct hfi1_ibdev {
}
/*
- * Send if not busy or waiting for I/O and either
- * a RC response is pending or we can process send work requests.
- */
-static inline int hfi1_send_ok(struct rvt_qp *qp)
-{
- return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) &&
- (qp->s_hdrwords || (qp->s_flags & RVT_S_RESP_PENDING) ||
- !(qp->s_flags & RVT_S_ANY_WAIT_SEND));
-}
-
-/*
* This must be called with s_lock held.
*/
void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl,
diff --git a/drivers/infiniband/hw/hfi1/verbs_txreq.h b/drivers/infiniband/hw/hfi1/verbs_txreq.h
index cec7a4b..729244c 100644
--- a/drivers/infiniband/hw/hfi1/verbs_txreq.h
+++ b/drivers/infiniband/hw/hfi1/verbs_txreq.h
@@ -113,6 +113,13 @@ struct verbs_txreq *__get_txreq(struct hfi1_ibdev *dev,
return NULL;
}
+static inline bool verbs_txreq_queued(struct rvt_qp *qp)
+{
+ struct hfi1_qp_priv *priv = qp->priv;
+
+ return iowait_packet_queued(&priv->s_iowait);
+}
+
void hfi1_put_txreq(struct verbs_txreq *tx);
int verbs_txreq_init(struct hfi1_ibdev *dev);
void verbs_txreq_exit(struct hfi1_ibdev *dev);
--
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] 27+ messages in thread* [PATCH for-next 04/16] IB/hfi1: Remove blind constants from 16B update
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (2 preceding siblings ...)
2018-01-26 14:31 ` [PATCH for-next 03/16] IB/hfi1: Remove dependence on qp->s_hdrwords Dennis Dalessandro
@ 2018-01-26 14:31 ` Dennis Dalessandro
2018-01-26 14:31 ` [PATCH for-next 05/16] IB/hfi1: Convert PortXmitWait/PortVLXmitWait counters to flit times Dennis Dalessandro
` (11 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:31 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Don Hiatt, Mike Marciniszyn
From: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
These values were introduced as part of the 16B code to
account for the varying size of the LRH between the differing
packet formats.
Replace the blind constants with defines based on FIELD_SIZEOF()
calls.
Fixes: 5b6cabb0db77 ("IB/hfi1: Add 16B RC/UC support")
Reviewed-by: Don Hiatt <don.hiatt-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/driver.c | 6 ++++--
drivers/infiniband/hw/hfi1/ruc.c | 13 ++++++-------
drivers/infiniband/hw/hfi1/ud.c | 15 ++++++++++-----
drivers/infiniband/hw/hfi1/verbs.h | 5 +++++
4 files changed, 25 insertions(+), 14 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index 067b29f..e6a318c 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -1492,8 +1492,10 @@ static int hfi1_setup_bypass_packet(struct hfi1_packet *packet)
/* Query commonly used fields from packet header */
packet->opcode = ib_bth_get_opcode(packet->ohdr);
- packet->hlen = hdr_len_by_opcode[packet->opcode] + 8 + grh_len;
- packet->payload = packet->ebuf + packet->hlen - (4 * sizeof(u32));
+ /* hdr_len_by_opcode already has an IB LRH factored in */
+ packet->hlen = hdr_len_by_opcode[packet->opcode] +
+ (LRH_16B_BYTES - LRH_9B_BYTES) + grh_len;
+ packet->payload = packet->ebuf + packet->hlen - LRH_16B_BYTES;
packet->slid = hfi1_16B_get_slid(packet->hdr);
packet->dlid = hfi1_16B_get_dlid(packet->hdr);
if (unlikely(hfi1_is_16B_mcast(packet->dlid)))
diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c
index 0cced9a..1033068 100644
--- a/drivers/infiniband/hw/hfi1/ruc.c
+++ b/drivers/infiniband/hw/hfi1/ruc.c
@@ -769,8 +769,6 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
struct ib_grh *grh;
struct ib_global_route *grd =
rdma_ah_retrieve_grh(&qp->remote_ah_attr);
- int hdrwords;
-
/*
* Ensure OPA GIDs are transformed to IB gids
* before creating the GRH.
@@ -779,9 +777,10 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
grd->sgid_index = 0;
grh = &ps->s_txreq->phdr.hdr.opah.u.l.grh;
l4 = OPA_16B_L4_IB_GLOBAL;
- hdrwords = ps->s_txreq->hdr_dwords - 4;
- ps->s_txreq->hdr_dwords += hfi1_make_grh(ibp, grh, grd,
- hdrwords, nwords);
+ ps->s_txreq->hdr_dwords +=
+ hfi1_make_grh(ibp, grh, grd,
+ ps->s_txreq->hdr_dwords - LRH_16B_DWORDS,
+ nwords);
middle = 0;
}
@@ -835,13 +834,13 @@ static inline void hfi1_make_ruc_header_9B(struct rvt_qp *qp,
if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH)) {
struct ib_grh *grh = &ps->s_txreq->phdr.hdr.ibh.u.l.grh;
- int hdrwords = ps->s_txreq->hdr_dwords - 2;
lrh0 = HFI1_LRH_GRH;
ps->s_txreq->hdr_dwords +=
hfi1_make_grh(ibp, grh,
rdma_ah_read_grh(&qp->remote_ah_attr),
- hdrwords, nwords);
+ ps->s_txreq->hdr_dwords - LRH_9B_DWORDS,
+ nwords);
middle = 0;
}
lrh0 |= (priv->s_sc & 0xf) << 12 |
diff --git a/drivers/infiniband/hw/hfi1/ud.c b/drivers/infiniband/hw/hfi1/ud.c
index 6b5cd3a..f1ba510 100644
--- a/drivers/infiniband/hw/hfi1/ud.c
+++ b/drivers/infiniband/hw/hfi1/ud.c
@@ -348,7 +348,8 @@ void hfi1_make_ud_req_9B(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
grh = &ps->s_txreq->phdr.hdr.ibh.u.l.grh;
ps->s_txreq->hdr_dwords +=
hfi1_make_grh(ibp, grh, rdma_ah_read_grh(ah_attr),
- ps->s_txreq->hdr_dwords - 2, nwords);
+ ps->s_txreq->hdr_dwords - LRH_9B_DWORDS,
+ nwords);
lrh0 = HFI1_LRH_GRH;
ohdr = &ps->s_txreq->phdr.hdr.ibh.u.l.oth;
} else {
@@ -428,8 +429,10 @@ void hfi1_make_ud_req_16B(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
grd->sgid_index = 0;
}
grh = &ps->s_txreq->phdr.hdr.opah.u.l.grh;
- ps->s_txreq->hdr_dwords += hfi1_make_grh(ibp, grh, grd,
- ps->s_txreq->hdr_dwords - 4, nwords);
+ ps->s_txreq->hdr_dwords += hfi1_make_grh(
+ ibp, grh, grd,
+ ps->s_txreq->hdr_dwords - LRH_16B_DWORDS,
+ nwords);
ohdr = &ps->s_txreq->phdr.hdr.opah.u.l.oth;
l4 = OPA_16B_L4_IB_GLOBAL;
} else {
@@ -648,7 +651,8 @@ void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp,
struct ib_grh *grh = &hdr.u.l.grh;
grh->version_tclass_flow = old_grh->version_tclass_flow;
- grh->paylen = cpu_to_be16((hwords - 4 + nwords) << 2);
+ grh->paylen = cpu_to_be16(
+ (hwords - LRH_16B_DWORDS + nwords) << 2);
grh->hop_limit = 0xff;
grh->sgid = old_grh->dgid;
grh->dgid = old_grh->sgid;
@@ -702,7 +706,8 @@ void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
struct ib_grh *grh = &hdr.u.l.grh;
grh->version_tclass_flow = old_grh->version_tclass_flow;
- grh->paylen = cpu_to_be16((hwords - 2 + SIZE_OF_CRC) << 2);
+ grh->paylen = cpu_to_be16(
+ (hwords - LRH_9B_DWORDS + SIZE_OF_CRC) << 2);
grh->hop_limit = 0xff;
grh->sgid = old_grh->dgid;
grh->dgid = old_grh->sgid;
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index c2aeb32..26c1ed6 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -105,6 +105,11 @@ enum {
HFI1_HAS_GRH = (1 << 0),
};
+#define LRH_16B_BYTES (FIELD_SIZEOF(struct hfi1_16b_header, lrh))
+#define LRH_16B_DWORDS (LRH_16B_BYTES / sizeof(u32))
+#define LRH_9B_BYTES (FIELD_SIZEOF(struct ib_header, lrh))
+#define LRH_9B_DWORDS (LRH_9B_BYTES / sizeof(u32))
+
struct hfi1_16b_header {
u32 lrh[4];
union {
--
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] 27+ messages in thread* [PATCH for-next 05/16] IB/hfi1: Convert PortXmitWait/PortVLXmitWait counters to flit times
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (3 preceding siblings ...)
2018-01-26 14:31 ` [PATCH for-next 04/16] IB/hfi1: Remove blind constants from 16B update Dennis Dalessandro
@ 2018-01-26 14:31 ` Dennis Dalessandro
2018-01-26 14:32 ` [PATCH for-next 06/16] IB/hfi1: hfi1_open_file() missing kobject_put in err path Dennis Dalessandro
` (10 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:31 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Michael J. Ruhl, Kamenee Arumugam
From: Kamenee Arumugam <kamenee.arumugam-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
HFI's counters SendWaitCnt and SendWaitVlCnt are in units
of TXE cycle time (at 805MHz). OPA counters PortXmitWait and
PortVLXmtWait are in units of flit times.
Convert the counter values to flit units using following
conversion formula:
PortXmitWait =
SendWaitCnt * 2 * (4 /link_width) * (25 Gbps /link_speed)
PortVLXmitWait =
SendWaitVLCnt * 2 * (4 /link_width) * (25 Gbps /link_speed)
At link up or downgrade events, the link width can change. To ensure
accurate counter calculations, sample the counters after the events,
during counter requests, and then aggregate the OPA counters.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/chip.c | 64 +++++++++++++++++--
drivers/infiniband/hw/hfi1/chip.h | 4 +
drivers/infiniband/hw/hfi1/hfi.h | 7 ++
drivers/infiniband/hw/hfi1/init.c | 9 +++
drivers/infiniband/hw/hfi1/mad.c | 127 ++++++++++++++++++++++++++++++++++---
drivers/infiniband/hw/hfi1/mad.h | 47 +++++++++++++-
6 files changed, 239 insertions(+), 19 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index 6660f92..a986afe 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -1083,6 +1083,7 @@ static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
static void clear_full_mgmt_pkey(struct hfi1_pportdata *ppd);
static int wait_link_transfer_active(struct hfi1_devdata *dd, int wait_ms);
static void clear_rsm_rule(struct hfi1_devdata *dd, u8 rule_index);
+static void update_xmit_counters(struct hfi1_pportdata *ppd, u16 link_width);
/*
* Error interrupt table entry. This is used as input to the interrupt
@@ -6905,6 +6906,32 @@ void handle_freeze(struct work_struct *work)
/* no longer frozen */
}
+/**
+ * update_xmit_counters - update PortXmitWait/PortVlXmitWait
+ * counters.
+ * @ppd: info of physical Hfi port
+ * @link_width: new link width after link up or downgrade
+ *
+ * Update the PortXmitWait and PortVlXmitWait counters after
+ * a link up or downgrade event to reflect a link width change.
+ */
+static void update_xmit_counters(struct hfi1_pportdata *ppd, u16 link_width)
+{
+ int i;
+ u16 tx_width;
+ u16 link_speed;
+
+ tx_width = tx_link_width(link_width);
+ link_speed = get_link_speed(ppd->link_speed_active);
+
+ /*
+ * There are C_VL_COUNT number of PortVLXmitWait counters.
+ * Adding 1 to C_VL_COUNT to include the PortXmitWait counter.
+ */
+ for (i = 0; i < C_VL_COUNT + 1; i++)
+ get_xmit_wait_counters(ppd, tx_width, link_speed, i);
+}
+
/*
* Handle a link up interrupt from the 8051.
*
@@ -7526,18 +7553,29 @@ void handle_verify_cap(struct work_struct *work)
set_link_state(ppd, HLS_GOING_UP);
}
-/*
- * Apply the link width downgrade enabled policy against the current active
- * link widths.
+/**
+ * apply_link_downgrade_policy - Apply the link width downgrade enabled
+ * policy against the current active link widths.
+ * @ppd: info of physical Hfi port
+ * @refresh_widths: True indicates link downgrade event
+ * @return: True indicates a successful link downgrade. False indicates
+ * link downgrade event failed and the link will bounce back to
+ * default link width.
*
- * Called when the enabled policy changes or the active link widths change.
+ * Called when the enabled policy changes or the active link widths
+ * change.
+ * Refresh_widths indicates that a link downgrade occurred. The
+ * link_downgraded variable is set by refresh_widths and
+ * determines the success/failure of the policy application.
*/
-void apply_link_downgrade_policy(struct hfi1_pportdata *ppd, int refresh_widths)
+bool apply_link_downgrade_policy(struct hfi1_pportdata *ppd,
+ bool refresh_widths)
{
int do_bounce = 0;
int tries;
u16 lwde;
u16 tx, rx;
+ bool link_downgraded = refresh_widths;
/* use the hls lock to avoid a race with actual link up */
tries = 0;
@@ -7571,6 +7609,7 @@ void apply_link_downgrade_policy(struct hfi1_pportdata *ppd, int refresh_widths)
ppd->link_width_downgrade_rx_active == 0) {
/* the 8051 reported a dead link as a downgrade */
dd_dev_err(ppd->dd, "Link downgrade is really a link down, ignoring\n");
+ link_downgraded = false;
} else if (lwde == 0) {
/* downgrade is disabled */
@@ -7587,6 +7626,7 @@ void apply_link_downgrade_policy(struct hfi1_pportdata *ppd, int refresh_widths)
ppd->link_width_downgrade_tx_active,
ppd->link_width_downgrade_rx_active);
do_bounce = 1;
+ link_downgraded = false;
}
} else if ((lwde & ppd->link_width_downgrade_tx_active) == 0 ||
(lwde & ppd->link_width_downgrade_rx_active) == 0) {
@@ -7598,6 +7638,7 @@ void apply_link_downgrade_policy(struct hfi1_pportdata *ppd, int refresh_widths)
lwde, ppd->link_width_downgrade_tx_active,
ppd->link_width_downgrade_rx_active);
do_bounce = 1;
+ link_downgraded = false;
}
done:
@@ -7609,6 +7650,8 @@ void apply_link_downgrade_policy(struct hfi1_pportdata *ppd, int refresh_widths)
set_link_state(ppd, HLS_DN_OFFLINE);
start_link(ppd);
}
+
+ return link_downgraded;
}
/*
@@ -7622,7 +7665,8 @@ void handle_link_downgrade(struct work_struct *work)
link_downgrade_work);
dd_dev_info(ppd->dd, "8051: Link width downgrade\n");
- apply_link_downgrade_policy(ppd, 1);
+ if (apply_link_downgrade_policy(ppd, true))
+ update_xmit_counters(ppd, ppd->link_width_downgrade_tx_active);
}
static char *dcc_err_string(char *buf, int buf_len, u64 flags)
@@ -10597,6 +10641,14 @@ int set_link_state(struct hfi1_pportdata *ppd, u32 state)
add_rcvctrl(dd, RCV_CTRL_RCV_PORT_ENABLE_SMASK);
handle_linkup_change(dd, 1);
+
+ /*
+ * After link up, a new link width will have been set.
+ * Update the xmit counters with regards to the new
+ * link width.
+ */
+ update_xmit_counters(ppd, ppd->link_width_active);
+
ppd->host_link_state = HLS_UP_INIT;
update_statusp(ppd, IB_PORT_INIT);
break;
diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h
index 21fca8e..c0d70f2 100644
--- a/drivers/infiniband/hw/hfi1/chip.h
+++ b/drivers/infiniband/hw/hfi1/chip.h
@@ -736,8 +736,8 @@ void set_link_down_reason(struct hfi1_pportdata *ppd, u8 lcl_reason,
int start_link(struct hfi1_pportdata *ppd);
int bringup_serdes(struct hfi1_pportdata *ppd);
void set_intr_state(struct hfi1_devdata *dd, u32 enable);
-void apply_link_downgrade_policy(struct hfi1_pportdata *ppd,
- int refresh_widths);
+bool apply_link_downgrade_policy(struct hfi1_pportdata *ppd,
+ bool refresh_widths);
void update_usrhead(struct hfi1_ctxtdata *rcd, u32 hd, u32 updegr, u32 egrhd,
u32 intr_adjust, u32 npkts);
int stop_drain_data_vls(struct hfi1_devdata *dd);
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index b42c222..8aea811 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -858,6 +858,13 @@ struct hfi1_pportdata {
struct work_struct linkstate_active_work;
/* Does this port need to prescan for FECNs */
bool cc_prescan;
+ /*
+ * Sample sendWaitCnt & sendWaitVlCnt during link transition
+ * and counter request.
+ */
+ u64 port_vl_xmit_wait_last[C_VL_COUNT + 1];
+ u16 prev_link_width;
+ u64 vl_xmit_flit_cnt[C_VL_COUNT + 1];
};
typedef int (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index a618eef..d89cd86 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -637,6 +637,15 @@ void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
ppd->dd = dd;
ppd->hw_pidx = hw_pidx;
ppd->port = port; /* IB port number, not index */
+ ppd->prev_link_width = LINK_WIDTH_DEFAULT;
+ /*
+ * There are C_VL_COUNT number of PortVLXmitWait counters.
+ * Adding 1 to C_VL_COUNT to include the PortXmitWait counter.
+ */
+ for (i = 0; i < C_VL_COUNT + 1; i++) {
+ ppd->port_vl_xmit_wait_last[i] = 0;
+ ppd->vl_xmit_flit_cnt[i] = 0;
+ }
default_pkey_idx = 1;
diff --git a/drivers/infiniband/hw/hfi1/mad.c b/drivers/infiniband/hw/hfi1/mad.c
index 34547a4..e9962c6 100644
--- a/drivers/infiniband/hw/hfi1/mad.c
+++ b/drivers/infiniband/hw/hfi1/mad.c
@@ -2649,6 +2649,79 @@ static void a0_portstatus(struct hfi1_pportdata *ppd,
}
}
+/**
+ * tx_link_width - convert link width bitmask to integer
+ * value representing actual link width.
+ * @link_width: width of active link
+ * @return: return index of the bit set in link_width var
+ *
+ * The function convert and return the index of bit set
+ * that indicate the current link width.
+ */
+u16 tx_link_width(u16 link_width)
+{
+ int n = LINK_WIDTH_DEFAULT;
+ u16 tx_width = n;
+
+ while (link_width && n) {
+ if (link_width & (1 << (n - 1))) {
+ tx_width = n;
+ break;
+ }
+ n--;
+ }
+
+ return tx_width;
+}
+
+/**
+ * get_xmit_wait_counters - Convert HFI 's SendWaitCnt/SendWaitVlCnt
+ * counter in unit of TXE cycle times to flit times.
+ * @ppd: info of physical Hfi port
+ * @link_width: width of active link
+ * @link_speed: speed of active link
+ * @vl: represent VL0-VL7, VL15 for PortVLXmitWait counters request
+ * and if vl value is C_VL_COUNT, it represent SendWaitCnt
+ * counter request
+ * @return: return SendWaitCnt/SendWaitVlCnt counter value per vl.
+ *
+ * Convert SendWaitCnt/SendWaitVlCnt counter from TXE cycle times to
+ * flit times. Call this function to samples these counters. This
+ * function will calculate for previous state transition and update
+ * current state at end of function using ppd->prev_link_width and
+ * ppd->port_vl_xmit_wait_last to port_vl_xmit_wait_curr and link_width.
+ */
+u64 get_xmit_wait_counters(struct hfi1_pportdata *ppd,
+ u16 link_width, u16 link_speed, int vl)
+{
+ u64 port_vl_xmit_wait_curr;
+ u64 delta_vl_xmit_wait;
+ u64 xmit_wait_val;
+
+ if (vl > C_VL_COUNT)
+ return 0;
+ if (vl < C_VL_COUNT)
+ port_vl_xmit_wait_curr =
+ read_port_cntr(ppd, C_TX_WAIT_VL, vl);
+ else
+ port_vl_xmit_wait_curr =
+ read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL);
+
+ xmit_wait_val =
+ port_vl_xmit_wait_curr -
+ ppd->port_vl_xmit_wait_last[vl];
+ delta_vl_xmit_wait =
+ convert_xmit_counter(xmit_wait_val,
+ ppd->prev_link_width,
+ link_speed);
+
+ ppd->vl_xmit_flit_cnt[vl] += delta_vl_xmit_wait;
+ ppd->port_vl_xmit_wait_last[vl] = port_vl_xmit_wait_curr;
+ ppd->prev_link_width = link_width;
+
+ return ppd->vl_xmit_flit_cnt[vl];
+}
+
static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
struct ib_device *ibdev,
u8 port, u32 *resp_len)
@@ -2668,6 +2741,8 @@ static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
int vfi;
u64 tmp, tmp2;
+ u16 link_width;
+ u16 link_speed;
response_data_size = sizeof(struct opa_port_status_rsp) +
num_vls * sizeof(struct _vls_pctrs);
@@ -2711,8 +2786,16 @@ static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
rsp->port_multicast_rcv_pkts =
cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
CNTR_INVALID_VL));
+ /*
+ * Convert PortXmitWait counter from TXE cycle times
+ * to flit times.
+ */
+ link_width =
+ tx_link_width(ppd->link_width_downgrade_tx_active);
+ link_speed = get_link_speed(ppd->link_speed_active);
rsp->port_xmit_wait =
- cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
+ cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
+ link_speed, C_VL_COUNT));
rsp->port_rcv_fecn =
cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
rsp->port_rcv_becn =
@@ -2777,10 +2860,14 @@ static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
rsp->vls[vfi].port_vl_xmit_pkts =
cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
idx_from_vl(vl)));
-
+ /*
+ * Convert PortVlXmitWait counter from TXE cycle
+ * times to flit times.
+ */
rsp->vls[vfi].port_vl_xmit_wait =
- cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
- idx_from_vl(vl)));
+ cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
+ link_speed,
+ idx_from_vl(vl)));
rsp->vls[vfi].port_vl_rcv_fecn =
cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
@@ -2910,6 +2997,8 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
unsigned long vl;
u32 vl_select_mask;
int vfi;
+ u16 link_width;
+ u16 link_speed;
num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
@@ -2959,8 +3048,16 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
rsp->link_quality_indicator = cpu_to_be32((u32)lq);
pma_get_opa_port_dctrs(ibdev, rsp);
+ /*
+ * Convert PortXmitWait counter from TXE
+ * cycle times to flit times.
+ */
+ link_width =
+ tx_link_width(ppd->link_width_downgrade_tx_active);
+ link_speed = get_link_speed(ppd->link_speed_active);
rsp->port_xmit_wait =
- cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
+ cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
+ link_speed, C_VL_COUNT));
rsp->port_rcv_fecn =
cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
rsp->port_rcv_becn =
@@ -2996,9 +3093,14 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
idx_from_vl(vl)));
+ /*
+ * Convert PortVlXmitWait counter from TXE
+ * cycle times to flit times.
+ */
rsp->vls[vfi].port_vl_xmit_wait =
- cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
- idx_from_vl(vl)));
+ cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
+ link_speed,
+ idx_from_vl(vl)));
rsp->vls[vfi].port_vl_rcv_fecn =
cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
@@ -3416,9 +3518,11 @@ static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
if (counter_select & CS_PORT_MCAST_RCV_PKTS)
write_dev_cntr(dd, C_DC_MC_RCV_PKTS, CNTR_INVALID_VL, 0);
- if (counter_select & CS_PORT_XMIT_WAIT)
+ if (counter_select & CS_PORT_XMIT_WAIT) {
write_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL, 0);
-
+ ppd->port_vl_xmit_wait_last[C_VL_COUNT] = 0;
+ ppd->vl_xmit_flit_cnt[C_VL_COUNT] = 0;
+ }
/* ignore cs_sw_portCongestion for HFIs */
if (counter_select & CS_PORT_RCV_FECN)
@@ -3491,8 +3595,11 @@ static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
if (counter_select & CS_PORT_RCV_PKTS)
write_dev_cntr(dd, C_DC_RX_PKT_VL, idx_from_vl(vl), 0);
- if (counter_select & CS_PORT_XMIT_WAIT)
+ if (counter_select & CS_PORT_XMIT_WAIT) {
write_port_cntr(ppd, C_TX_WAIT_VL, idx_from_vl(vl), 0);
+ ppd->port_vl_xmit_wait_last[idx_from_vl(vl)] = 0;
+ ppd->vl_xmit_flit_cnt[idx_from_vl(vl)] = 0;
+ }
/* sw_port_vl_congestion is 0 for HFIs */
if (counter_select & CS_PORT_RCV_FECN)
diff --git a/drivers/infiniband/hw/hfi1/mad.h b/drivers/infiniband/hw/hfi1/mad.h
index c4938f3..2f48e69 100644
--- a/drivers/infiniband/hw/hfi1/mad.h
+++ b/drivers/infiniband/hw/hfi1/mad.h
@@ -180,6 +180,15 @@ struct opa_mad_notice_attr {
#define OPA_VLARB_PREEMPT_MATRIX 3
#define IB_PMA_PORT_COUNTERS_CONG cpu_to_be16(0xFF00)
+#define LINK_SPEED_25G 1
+#define LINK_SPEED_12_5G 2
+#define LINK_WIDTH_DEFAULT 4
+#define DECIMAL_FACTORING 1000
+/*
+ * The default link width is multiplied by 1000
+ * to get accurate value after division.
+ */
+#define FACTOR_LINK_WIDTH (LINK_WIDTH_DEFAULT * DECIMAL_FACTORING)
struct ib_pma_portcounters_cong {
u8 reserved;
@@ -429,5 +438,41 @@ struct sc2vlnt {
void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port);
void hfi1_handle_trap_timer(struct timer_list *t);
-
+u16 tx_link_width(u16 link_width);
+u64 get_xmit_wait_counters(struct hfi1_pportdata *ppd, u16 link_width,
+ u16 link_speed, int vl);
+/**
+ * get_link_speed - determine whether 12.5G or 25G speed
+ * @link_speed: the speed of active link
+ * @return: Return 2 if link speed identified as 12.5G
+ * or return 1 if link speed is 25G.
+ *
+ * The function indirectly calculate required link speed
+ * value for convert_xmit_counter function. If the link
+ * speed is 25G, the function return as 1 as it is required
+ * by xmit counter conversion formula :-( 25G / link_speed).
+ * This conversion will provide value 1 if current
+ * link speed is 25G or 2 if 12.5G.This is done to avoid
+ * 12.5 float number conversion.
+ */
+static inline u16 get_link_speed(u16 link_speed)
+{
+ return (link_speed == 1) ?
+ LINK_SPEED_12_5G : LINK_SPEED_25G;
+}
+
+/**
+ * convert_xmit_counter - calculate flit times for given xmit counter
+ * value
+ * @xmit_wait_val: current xmit counter value
+ * @link_width: width of active link
+ * @link_speed: speed of active link
+ * @return: return xmit counter value in flit times.
+ */
+static inline u64 convert_xmit_counter(u64 xmit_wait_val, u16 link_width,
+ u16 link_speed)
+{
+ return (xmit_wait_val * 2 * (FACTOR_LINK_WIDTH / link_width)
+ * link_speed) / DECIMAL_FACTORING;
+}
#endif /* _HFI1_MAD_H */
--
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] 27+ messages in thread* [PATCH for-next 06/16] IB/hfi1: hfi1_open_file() missing kobject_put in err path
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (4 preceding siblings ...)
2018-01-26 14:31 ` [PATCH for-next 05/16] IB/hfi1: Convert PortXmitWait/PortVLXmitWait counters to flit times Dennis Dalessandro
@ 2018-01-26 14:32 ` Dennis Dalessandro
[not found] ` <20180126143203.6868.85360.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2018-01-26 14:32 ` [PATCH for-next 07/16] IB/hfi1: Show fault stats in both TX and RX directions Dennis Dalessandro
` (9 subsequent siblings)
15 siblings, 1 reply; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:32 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Michael J Ruhl, Alex Estrin
From: Alex Estrin <alex.estrin-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
The dd refcount is speculatively incremented prior to allocating
the fd memory with kzalloc(). If that kzalloc() failed the dd
refcount leaks.
Fixes: e11ffbd57520 ("IB/hfi1: Do not free hfi1 cdev parent structure early")
Reviewed-by: Michael J Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Alex Estrin <alex.estrin-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/file_ops.c | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 7750a9c..d7bdf3d 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -210,6 +210,7 @@ static int hfi1_file_open(struct inode *inode, struct file *fp)
fd->dd = dd;
fp->private_data = fd;
} else {
+ kobject_put(&dd->kobj);
fp->private_data = NULL;
if (atomic_dec_and_test(&dd->user_refcount))
--
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] 27+ messages in thread* [PATCH for-next 07/16] IB/hfi1: Show fault stats in both TX and RX directions
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (5 preceding siblings ...)
2018-01-26 14:32 ` [PATCH for-next 06/16] IB/hfi1: hfi1_open_file() missing kobject_put in err path Dennis Dalessandro
@ 2018-01-26 14:32 ` Dennis Dalessandro
2018-01-26 14:32 ` [PATCH for-next 08/16] IB/hfi1: Prevent LNI hang when LCB can't obtain lanes Dennis Dalessandro
` (8 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:32 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Don Hiatt, Mitko Haralanov,
Michael J. Ruhl
From: Mitko Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
The routine which shows the fault stats checks the counters
to determine whether to show any stats based on the number of
transmitted pkts/bytes for a particular opcode.
Unfortunately, it only checked the receive counters. As a result,
if any packet faults have happened for packets egressing the HFI,
those stats would not be shown.
In order to fix this, the routine is amended to also check the
TX counters. With this change the pkt/byte counts are the sum of
both TX and RX counts for the opcode.
Fixes: 1b311f8931cf ("IB/hfi1: Add tx_opcode_stats like the opcode_stats")
Reviewed-by: Don Hiatt <don.hiatt-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mitko Haralanov <mitko.haralanov-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/debugfs.c | 9 ++++++++-
1 files changed, 8 insertions(+), 1 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/debugfs.c b/drivers/infiniband/hw/hfi1/debugfs.c
index 2e6e0c5..852173b 100644
--- a/drivers/infiniband/hw/hfi1/debugfs.c
+++ b/drivers/infiniband/hw/hfi1/debugfs.c
@@ -1,5 +1,5 @@
/*
- * Copyright(c) 2015-2017 Intel Corporation.
+ * Copyright(c) 2015-2018 Intel Corporation.
*
* This file is provided under a dual BSD/GPLv2 license. When using or
* redistributing this file, you may do so under either license.
@@ -1201,6 +1201,13 @@ static int _fault_stats_seq_show(struct seq_file *s, void *v)
}
hfi1_rcd_put(rcd);
}
+ for_each_possible_cpu(j) {
+ struct hfi1_opcode_stats_perctx *sp =
+ per_cpu_ptr(dd->tx_opstats, j);
+
+ n_packets += sp->stats[i].n_packets;
+ n_bytes += sp->stats[i].n_bytes;
+ }
if (!n_packets && !n_bytes)
return SEQ_SKIP;
if (!ibd->fault_opcode->n_rxfaults[i] &&
--
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] 27+ messages in thread* [PATCH for-next 08/16] IB/hfi1: Prevent LNI hang when LCB can't obtain lanes
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (6 preceding siblings ...)
2018-01-26 14:32 ` [PATCH for-next 07/16] IB/hfi1: Show fault stats in both TX and RX directions Dennis Dalessandro
@ 2018-01-26 14:32 ` Dennis Dalessandro
2018-01-26 14:32 ` [PATCH for-next 09/16] IB/hfi1: Compute BTH only for RDMA_WRITE_LAST/SEND_LAST packet Dennis Dalessandro
` (7 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:32 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Michael J. Ruhl,
Mike Marciniszyn, Sebastian Sanchez
From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
When the LCB isn't able to get any lanes operational on the
first transition into mission mode, the link transfer active
never happens and the LNI stays in the polling state indefinitely.
Reset LCB upon receiving an 8051 interrupt for LCB to try to obtain
lanes with firmware version 1.25.0 or later. Also, update the LCB
reset value in other parts of the code with a macro defined to make
the code more maintainable and rename functions with the link_width
label to link_mode to reflect the fact that those functions set and
read link related data not just the link width.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/chip.c | 51 +++++++++++++++++++--------
drivers/infiniband/hw/hfi1/chip.h | 15 +++++++-
drivers/infiniband/hw/hfi1/chip_registers.h | 7 +++-
3 files changed, 53 insertions(+), 20 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index a986afe..793e8b7 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -1032,8 +1032,8 @@ static void read_vc_remote_fabric(struct hfi1_devdata *dd, u8 *vau, u8 *z,
u8 *vcu, u16 *vl15buf, u8 *crc_sizes);
static void read_vc_remote_link_width(struct hfi1_devdata *dd,
u8 *remote_tx_rate, u16 *link_widths);
-static void read_vc_local_link_width(struct hfi1_devdata *dd, u8 *misc_bits,
- u8 *flag_bits, u16 *link_widths);
+static void read_vc_local_link_mode(struct hfi1_devdata *dd, u8 *misc_bits,
+ u8 *flag_bits, u16 *link_widths);
static void read_remote_device_id(struct hfi1_devdata *dd, u16 *device_id,
u8 *device_rev);
static void read_local_lni(struct hfi1_devdata *dd, u8 *enable_lane_rx);
@@ -6351,6 +6351,18 @@ static void handle_8051_request(struct hfi1_pportdata *ppd)
type);
hreq_response(dd, HREQ_NOT_SUPPORTED, 0);
break;
+ case HREQ_LCB_RESET:
+ /* Put the LCB, RX FPE and TX FPE into reset */
+ write_csr(dd, DCC_CFG_RESET, LCB_RX_FPE_TX_FPE_INTO_RESET);
+ /* Make sure the write completed */
+ (void)read_csr(dd, DCC_CFG_RESET);
+ /* Hold the reset long enough to take effect */
+ udelay(1);
+ /* Take the LCB, RX FPE and TX FPE out of reset */
+ write_csr(dd, DCC_CFG_RESET, LCB_RX_FPE_TX_FPE_OUT_OF_RESET);
+ hreq_response(dd, HREQ_SUCCESS, 0);
+
+ break;
case HREQ_CONFIG_DONE:
hreq_response(dd, HREQ_SUCCESS, 0);
break;
@@ -6461,8 +6473,7 @@ static void lcb_shutdown(struct hfi1_devdata *dd, int abort)
dd->lcb_err_en = read_csr(dd, DC_LCB_ERR_EN);
reg = read_csr(dd, DCC_CFG_RESET);
write_csr(dd, DCC_CFG_RESET, reg |
- (1ull << DCC_CFG_RESET_RESET_LCB_SHIFT) |
- (1ull << DCC_CFG_RESET_RESET_RX_FPE_SHIFT));
+ DCC_CFG_RESET_RESET_LCB | DCC_CFG_RESET_RESET_RX_FPE);
(void)read_csr(dd, DCC_CFG_RESET); /* make sure the write completed */
if (!abort) {
udelay(1); /* must hold for the longer of 16cclks or 20ns */
@@ -6527,7 +6538,7 @@ static void _dc_start(struct hfi1_devdata *dd)
__func__);
/* Take away reset for LCB and RX FPE (set in lcb_shutdown). */
- write_csr(dd, DCC_CFG_RESET, 0x10);
+ write_csr(dd, DCC_CFG_RESET, LCB_RX_FPE_TX_FPE_OUT_OF_RESET);
/* lcb_shutdown() with abort=1 does not restore these */
write_csr(dd, DC_LCB_ERR_EN, dd->lcb_err_en);
dd->dc_shutdown = 0;
@@ -7348,7 +7359,7 @@ static void get_linkup_widths(struct hfi1_devdata *dd, u16 *tx_width,
u8 misc_bits, local_flags;
u16 active_tx, active_rx;
- read_vc_local_link_width(dd, &misc_bits, &local_flags, &widths);
+ read_vc_local_link_mode(dd, &misc_bits, &local_flags, &widths);
tx = widths >> 12;
rx = (widths >> 8) & 0xf;
@@ -8820,29 +8831,29 @@ static int write_vc_local_fabric(struct hfi1_devdata *dd, u8 vau, u8 z, u8 vcu,
GENERAL_CONFIG, frame);
}
-static void read_vc_local_link_width(struct hfi1_devdata *dd, u8 *misc_bits,
- u8 *flag_bits, u16 *link_widths)
+static void read_vc_local_link_mode(struct hfi1_devdata *dd, u8 *misc_bits,
+ u8 *flag_bits, u16 *link_widths)
{
u32 frame;
- read_8051_config(dd, VERIFY_CAP_LOCAL_LINK_WIDTH, GENERAL_CONFIG,
+ read_8051_config(dd, VERIFY_CAP_LOCAL_LINK_MODE, GENERAL_CONFIG,
&frame);
*misc_bits = (frame >> MISC_CONFIG_BITS_SHIFT) & MISC_CONFIG_BITS_MASK;
*flag_bits = (frame >> LOCAL_FLAG_BITS_SHIFT) & LOCAL_FLAG_BITS_MASK;
*link_widths = (frame >> LINK_WIDTH_SHIFT) & LINK_WIDTH_MASK;
}
-static int write_vc_local_link_width(struct hfi1_devdata *dd,
- u8 misc_bits,
- u8 flag_bits,
- u16 link_widths)
+static int write_vc_local_link_mode(struct hfi1_devdata *dd,
+ u8 misc_bits,
+ u8 flag_bits,
+ u16 link_widths)
{
u32 frame;
frame = (u32)misc_bits << MISC_CONFIG_BITS_SHIFT
| (u32)flag_bits << LOCAL_FLAG_BITS_SHIFT
| (u32)link_widths << LINK_WIDTH_SHIFT;
- return load_8051_config(dd, VERIFY_CAP_LOCAL_LINK_WIDTH, GENERAL_CONFIG,
+ return load_8051_config(dd, VERIFY_CAP_LOCAL_LINK_MODE, GENERAL_CONFIG,
frame);
}
@@ -9312,8 +9323,16 @@ static int set_local_link_attributes(struct hfi1_pportdata *ppd)
if (loopback == LOOPBACK_SERDES)
misc_bits |= 1 << LOOPBACK_SERDES_CONFIG_BIT_MASK_SHIFT;
- ret = write_vc_local_link_width(dd, misc_bits, 0,
- opa_to_vc_link_widths(
+ /*
+ * An external device configuration request is used to reset the LCB
+ * to retry to obtain operational lanes when the first attempt is
+ * unsuccesful.
+ */
+ if (dd->dc8051_ver >= dc8051_ver(1, 25, 0))
+ misc_bits |= 1 << EXT_CFG_LCB_RESET_SUPPORTED_SHIFT;
+
+ ret = write_vc_local_link_mode(dd, misc_bits, 0,
+ opa_to_vc_link_widths(
ppd->link_width_enabled));
if (ret != HCMD_SUCCESS)
goto set_local_link_attributes_fail;
diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h
index c0d70f2..fdf389e 100644
--- a/drivers/infiniband/hw/hfi1/chip.h
+++ b/drivers/infiniband/hw/hfi1/chip.h
@@ -196,6 +196,15 @@
#define LSTATE_ARMED 0x3
#define LSTATE_ACTIVE 0x4
+/* DCC_CFG_RESET reset states */
+#define LCB_RX_FPE_TX_FPE_INTO_RESET (DCC_CFG_RESET_RESET_LCB | \
+ DCC_CFG_RESET_RESET_TX_FPE | \
+ DCC_CFG_RESET_RESET_RX_FPE | \
+ DCC_CFG_RESET_ENABLE_CCLK_BCC)
+ /* 0x17 */
+
+#define LCB_RX_FPE_TX_FPE_OUT_OF_RESET DCC_CFG_RESET_ENABLE_CCLK_BCC /* 0x10 */
+
/* DC8051_STS_CUR_STATE port values (physical link states) */
#define PLS_DISABLED 0x30
#define PLS_OFFLINE 0x90
@@ -283,6 +292,7 @@
#define HREQ_SET_TX_EQ_ABS 0x04
#define HREQ_SET_TX_EQ_REL 0x05
#define HREQ_ENABLE 0x06
+#define HREQ_LCB_RESET 0x07
#define HREQ_CONFIG_DONE 0xfe
#define HREQ_INTERFACE_TEST 0xff
@@ -383,7 +393,7 @@
#define TX_SETTINGS 0x06
#define VERIFY_CAP_LOCAL_PHY 0x07
#define VERIFY_CAP_LOCAL_FABRIC 0x08
-#define VERIFY_CAP_LOCAL_LINK_WIDTH 0x09
+#define VERIFY_CAP_LOCAL_LINK_MODE 0x09
#define LOCAL_DEVICE_ID 0x0a
#define RESERVED_REGISTERS 0x0b
#define LOCAL_LNI_INFO 0x0c
@@ -584,8 +594,9 @@ enum {
#define LOOPBACK_LCB 2
#define LOOPBACK_CABLE 3 /* external cable */
-/* set up serdes bit in MISC_CONFIG_BITS */
+/* set up bits in MISC_CONFIG_BITS */
#define LOOPBACK_SERDES_CONFIG_BIT_MASK_SHIFT 0
+#define EXT_CFG_LCB_RESET_SUPPORTED_SHIFT 3
/* read and write hardware registers */
u64 read_csr(const struct hfi1_devdata *dd, u32 offset);
diff --git a/drivers/infiniband/hw/hfi1/chip_registers.h b/drivers/infiniband/hw/hfi1/chip_registers.h
index 793514f..da598b5 100644
--- a/drivers/infiniband/hw/hfi1/chip_registers.h
+++ b/drivers/infiniband/hw/hfi1/chip_registers.h
@@ -97,8 +97,11 @@
#define DCC_CFG_PORT_CONFIG_MTU_CAP_SHIFT 32
#define DCC_CFG_PORT_CONFIG_MTU_CAP_SMASK 0x700000000ull
#define DCC_CFG_RESET (DCC_CSRS + 0x000000000000)
-#define DCC_CFG_RESET_RESET_LCB_SHIFT 0
-#define DCC_CFG_RESET_RESET_RX_FPE_SHIFT 2
+#define DCC_CFG_RESET_RESET_LCB BIT_ULL(0)
+#define DCC_CFG_RESET_RESET_TX_FPE BIT_ULL(1)
+#define DCC_CFG_RESET_RESET_RX_FPE BIT_ULL(2)
+#define DCC_CFG_RESET_RESET_8051 BIT_ULL(3)
+#define DCC_CFG_RESET_ENABLE_CCLK_BCC BIT_ULL(4)
#define DCC_CFG_SC_VL_TABLE_15_0 (DCC_CSRS + 0x000000000028)
#define DCC_CFG_SC_VL_TABLE_15_0_ENTRY0_SHIFT 0
#define DCC_CFG_SC_VL_TABLE_15_0_ENTRY10_SHIFT 40
--
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] 27+ messages in thread* [PATCH for-next 09/16] IB/hfi1: Compute BTH only for RDMA_WRITE_LAST/SEND_LAST packet
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (7 preceding siblings ...)
2018-01-26 14:32 ` [PATCH for-next 08/16] IB/hfi1: Prevent LNI hang when LCB can't obtain lanes Dennis Dalessandro
@ 2018-01-26 14:32 ` Dennis Dalessandro
2018-01-26 14:32 ` [PATCH for-next 10/16] IB/hfi1: Optimize packet type comparison using 9B and bypass code paths Dennis Dalessandro
` (6 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:32 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Mike Marciniszyn, Sebastian Sanchez
From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
In hfi1_rc_rcv(), BTH is computed for all packets received.
However, it's only used for packets received with opcodes
RDMA_WRITE_LAST and SEND_LAST, and it is a costly operation.
Compute BTH only in the RDMA_WRITE_LAST/SEND_LAST code path
and let the compiler handle endianness conversion for bitwise
operations.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/rc.c | 3 +--
drivers/infiniband/hw/hfi1/uc.c | 3 +--
drivers/infiniband/hw/hfi1/ud.c | 3 +--
drivers/infiniband/hw/qib/qib_rc.c | 3 +--
drivers/infiniband/hw/qib/qib_uc.c | 3 +--
drivers/infiniband/hw/qib/qib_ud.c | 3 +--
include/rdma/ib_hdrs.h | 4 ++++
7 files changed, 10 insertions(+), 12 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c
index 524c12f..70aa82b 100644
--- a/drivers/infiniband/hw/hfi1/rc.c
+++ b/drivers/infiniband/hw/hfi1/rc.c
@@ -2035,7 +2035,6 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
struct rvt_qp *qp = packet->qp;
struct hfi1_ibport *ibp = rcd_to_iport(rcd);
struct ib_other_headers *ohdr = packet->ohdr;
- u32 bth0 = be32_to_cpu(ohdr->bth[0]);
u32 opcode = packet->opcode;
u32 hdrsize = packet->hlen;
u32 psn = ib_bth_get_psn(packet->ohdr);
@@ -2233,7 +2232,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
wc.port_num = 0;
/* Signal completion event if the solicited bit is set. */
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
- (bth0 & IB_BTH_SOLICITED) != 0);
+ ib_bth_is_solicited(ohdr));
break;
case OP(RDMA_WRITE_ONLY):
diff --git a/drivers/infiniband/hw/hfi1/uc.c b/drivers/infiniband/hw/hfi1/uc.c
index fef69d9..8030c38 100644
--- a/drivers/infiniband/hw/hfi1/uc.c
+++ b/drivers/infiniband/hw/hfi1/uc.c
@@ -478,8 +478,7 @@ void hfi1_uc_rcv(struct hfi1_packet *packet)
wc.port_num = 0;
/* Signal completion event if the solicited bit is set. */
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
- (ohdr->bth[0] &
- cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+ ib_bth_is_solicited(ohdr));
break;
case OP(RDMA_WRITE_FIRST):
diff --git a/drivers/infiniband/hw/hfi1/ud.c b/drivers/infiniband/hw/hfi1/ud.c
index f1ba510..066afbc 100644
--- a/drivers/infiniband/hw/hfi1/ud.c
+++ b/drivers/infiniband/hw/hfi1/ud.c
@@ -1050,8 +1050,7 @@ void hfi1_ud_rcv(struct hfi1_packet *packet)
wc.port_num = qp->port_num;
/* Signal completion event if the solicited bit is set. */
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
- (ohdr->bth[0] &
- cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+ ib_bth_is_solicited(ohdr));
return;
drop:
diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
index e4a9ba1..61d3bc6 100644
--- a/drivers/infiniband/hw/qib/qib_rc.c
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -1916,8 +1916,7 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct ib_header *hdr,
wc.port_num = 0;
/* Signal completion event if the solicited bit is set. */
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
- (ohdr->bth[0] &
- cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+ ib_bth_is_solicited(ohdr));
break;
case OP(RDMA_WRITE_FIRST):
diff --git a/drivers/infiniband/hw/qib/qib_uc.c b/drivers/infiniband/hw/qib/qib_uc.c
index bddcc37..d311db6 100644
--- a/drivers/infiniband/hw/qib/qib_uc.c
+++ b/drivers/infiniband/hw/qib/qib_uc.c
@@ -403,8 +403,7 @@ void qib_uc_rcv(struct qib_ibport *ibp, struct ib_header *hdr,
wc.port_num = 0;
/* Signal completion event if the solicited bit is set. */
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
- (ohdr->bth[0] &
- cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+ ib_bth_is_solicited(ohdr));
break;
case OP(RDMA_WRITE_FIRST):
diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c
index 15962ed..59a4f03 100644
--- a/drivers/infiniband/hw/qib/qib_ud.c
+++ b/drivers/infiniband/hw/qib/qib_ud.c
@@ -581,8 +581,7 @@ void qib_ud_rcv(struct qib_ibport *ibp, struct ib_header *hdr,
wc.port_num = qp->port_num;
/* Signal completion event if the solicited bit is set. */
rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
- (ohdr->bth[0] &
- cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+ ib_bth_is_solicited(ohdr));
return;
drop:
diff --git a/include/rdma/ib_hdrs.h b/include/rdma/ib_hdrs.h
index c124d51..6a86d14 100644
--- a/include/rdma/ib_hdrs.h
+++ b/include/rdma/ib_hdrs.h
@@ -331,4 +331,8 @@ static inline u8 ib_bth_get_tver(struct ib_other_headers *ohdr)
IB_BTH_TVER_MASK);
}
+static inline bool ib_bth_is_solicited(struct ib_other_headers *ohdr)
+{
+ return !!(ohdr->bth[0] & cpu_to_be32(IB_BTH_SOLICITED));
+}
#endif /* IB_HDRS_H */
--
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] 27+ messages in thread* [PATCH for-next 10/16] IB/hfi1: Optimize packet type comparison using 9B and bypass code paths
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (8 preceding siblings ...)
2018-01-26 14:32 ` [PATCH for-next 09/16] IB/hfi1: Compute BTH only for RDMA_WRITE_LAST/SEND_LAST packet Dennis Dalessandro
@ 2018-01-26 14:32 ` Dennis Dalessandro
2018-01-26 14:32 ` [PATCH for-next 11/16] IB/hfi1: Look up ibport using a pointer in receive path Dennis Dalessandro
` (5 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:32 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Don Hiatt, Mike Marciniszyn,
Sebastian Sanchez
From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
The packet type comparison used to find out if a packet is a bypass
packet in the hot path is an expensive operation as seen in a profile.
Determine packet's pkey and migration bit through the bypass and 9B
code paths instead.
Reviewed-by: Don Hiatt <don.hiatt-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/driver.c | 4 ++++
drivers/infiniband/hw/hfi1/hfi.h | 2 ++
drivers/infiniband/hw/hfi1/ruc.c | 15 ++-------------
drivers/infiniband/hw/hfi1/verbs.h | 5 +++++
include/rdma/ib_hdrs.h | 5 +++++
5 files changed, 18 insertions(+), 13 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index e6a318c..7ef8634 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -1440,6 +1440,8 @@ static int hfi1_setup_9B_packet(struct hfi1_packet *packet)
packet->extra_byte = 0;
packet->fecn = ib_bth_get_fecn(packet->ohdr);
packet->becn = ib_bth_get_becn(packet->ohdr);
+ packet->pkey = ib_bth_get_pkey(packet->ohdr);
+ packet->migrated = ib_bth_is_migration(packet->ohdr);
return 0;
drop:
@@ -1508,6 +1510,8 @@ static int hfi1_setup_bypass_packet(struct hfi1_packet *packet)
packet->extra_byte = SIZE_OF_LT;
packet->fecn = hfi1_16B_get_fecn(packet->hdr);
packet->becn = hfi1_16B_get_becn(packet->hdr);
+ packet->pkey = hfi1_16B_get_pkey(packet->hdr);
+ packet->migrated = opa_bth_is_migration(packet->ohdr);
if (hfi1_bypass_ingress_pkt_check(packet))
goto drop;
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index 8aea811..e0ed2b4 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -341,6 +341,7 @@ struct hfi1_packet {
u32 slid;
u16 tlen;
s16 etail;
+ u16 pkey;
u8 hlen;
u8 numpkt;
u8 rsize;
@@ -353,6 +354,7 @@ struct hfi1_packet {
u8 opcode;
bool becn;
bool fecn;
+ bool migrated;
};
/* Packet types */
diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c
index 1033068..e7d4e04 100644
--- a/drivers/infiniband/hw/hfi1/ruc.c
+++ b/drivers/infiniband/hw/hfi1/ruc.c
@@ -225,19 +225,8 @@ int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct hfi1_packet *packet)
u32 dlid = packet->dlid;
u32 slid = packet->slid;
u32 sl = packet->sl;
- int migrated;
- u32 bth0, bth1;
- u16 pkey;
-
- bth0 = be32_to_cpu(packet->ohdr->bth[0]);
- bth1 = be32_to_cpu(packet->ohdr->bth[1]);
- if (packet->etype == RHF_RCV_TYPE_BYPASS) {
- pkey = hfi1_16B_get_pkey(packet->hdr);
- migrated = bth1 & OPA_BTH_MIG_REQ;
- } else {
- pkey = ib_bth_get_pkey(packet->ohdr);
- migrated = bth0 & IB_BTH_MIG_REQ;
- }
+ bool migrated = packet->migrated;
+ u16 pkey = packet->pkey;
if (qp->s_mig_state == IB_MIG_ARMED && migrated) {
if (!packet->grh) {
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index 26c1ed6..90a170e 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -410,6 +410,11 @@ static inline void cacheless_memcpy(void *dst, void *src, size_t n)
__copy_user_nocache(dst, (void __user *)src, n, 0);
}
+static inline bool opa_bth_is_migration(struct ib_other_headers *ohdr)
+{
+ return !!(ohdr->bth[1] & cpu_to_be32(OPA_BTH_MIG_REQ));
+}
+
extern const enum ib_wc_opcode ib_hfi1_wc_opcode[];
extern const u8 hdr_len_by_opcode[];
diff --git a/include/rdma/ib_hdrs.h b/include/rdma/ib_hdrs.h
index 6a86d14..2aa19ec 100644
--- a/include/rdma/ib_hdrs.h
+++ b/include/rdma/ib_hdrs.h
@@ -335,4 +335,9 @@ static inline bool ib_bth_is_solicited(struct ib_other_headers *ohdr)
{
return !!(ohdr->bth[0] & cpu_to_be32(IB_BTH_SOLICITED));
}
+
+static inline bool ib_bth_is_migration(struct ib_other_headers *ohdr)
+{
+ return !!(ohdr->bth[0] & cpu_to_be32(IB_BTH_MIG_REQ));
+}
#endif /* IB_HDRS_H */
--
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] 27+ messages in thread* [PATCH for-next 11/16] IB/hfi1: Look up ibport using a pointer in receive path
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (9 preceding siblings ...)
2018-01-26 14:32 ` [PATCH for-next 10/16] IB/hfi1: Optimize packet type comparison using 9B and bypass code paths Dennis Dalessandro
@ 2018-01-26 14:32 ` Dennis Dalessandro
2018-01-26 14:32 ` [PATCH for-next 12/16] IB/hfi1: Remove unnecessary fecn and becn fields Dennis Dalessandro
` (4 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:32 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Mike Marciniszyn, Sebastian Sanchez
From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
In the receive path, hfi1_ibport is looked up by indexing into an
array. A profile shows this to be expensive. The receive context
data has a pointer to the ibport data, use that pointer instead.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/driver.c | 14 ++++++++------
drivers/infiniband/hw/hfi1/rc.c | 36 ++++++++++++++++++++---------------
drivers/infiniband/hw/hfi1/verbs.h | 3 +--
3 files changed, 30 insertions(+), 23 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index 7ef8634..bdae9fb 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -634,9 +634,10 @@ static void __prescan_rxq(struct hfi1_packet *packet)
}
}
-static void process_rcv_qp_work(struct hfi1_ctxtdata *rcd)
+static void process_rcv_qp_work(struct hfi1_packet *packet)
{
struct rvt_qp *qp, *nqp;
+ struct hfi1_ctxtdata *rcd = packet->rcd;
/*
* Iterate over all QPs waiting to respond.
@@ -646,7 +647,8 @@ static void process_rcv_qp_work(struct hfi1_ctxtdata *rcd)
list_del_init(&qp->rspwait);
if (qp->r_flags & RVT_R_RSP_NAK) {
qp->r_flags &= ~RVT_R_RSP_NAK;
- hfi1_send_rc_ack(rcd, qp, 0);
+ packet->qp = qp;
+ hfi1_send_rc_ack(packet, 0);
}
if (qp->r_flags & RVT_R_RSP_SEND) {
unsigned long flags;
@@ -667,7 +669,7 @@ static noinline int max_packet_exceeded(struct hfi1_packet *packet, int thread)
if (thread) {
if ((packet->numpkt & (MAX_PKT_RECV_THREAD - 1)) == 0)
/* allow defered processing */
- process_rcv_qp_work(packet->rcd);
+ process_rcv_qp_work(packet);
cond_resched();
return RCV_PKT_OK;
} else {
@@ -809,7 +811,7 @@ int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread)
last = RCV_PKT_DONE;
process_rcv_update(last, &packet);
}
- process_rcv_qp_work(rcd);
+ process_rcv_qp_work(&packet);
rcd->head = packet.rhqoff;
bail:
finish_packet(&packet);
@@ -838,7 +840,7 @@ int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread)
last = RCV_PKT_DONE;
process_rcv_update(last, &packet);
}
- process_rcv_qp_work(rcd);
+ process_rcv_qp_work(&packet);
rcd->head = packet.rhqoff;
bail:
finish_packet(&packet);
@@ -1068,7 +1070,7 @@ int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)
process_rcv_update(last, &packet);
}
- process_rcv_qp_work(rcd);
+ process_rcv_qp_work(&packet);
rcd->head = packet.rhqoff;
bail:
diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c
index 70aa82b..daf50cc 100644
--- a/drivers/infiniband/hw/hfi1/rc.c
+++ b/drivers/infiniband/hw/hfi1/rc.c
@@ -730,14 +730,16 @@ static inline void hfi1_make_bth_aeth(struct rvt_qp *qp,
ohdr->bth[2] = cpu_to_be32(mask_psn(qp->r_ack_psn));
}
-static inline void hfi1_queue_rc_ack(struct rvt_qp *qp, bool is_fecn)
+static inline void hfi1_queue_rc_ack(struct hfi1_packet *packet, bool is_fecn)
{
- struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+ struct rvt_qp *qp = packet->qp;
+ struct hfi1_ibport *ibp;
unsigned long flags;
spin_lock_irqsave(&qp->s_lock, flags);
if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK))
goto unlock;
+ ibp = rcd_to_iport(packet->rcd);
this_cpu_inc(*ibp->rvp.rc_qacks);
qp->s_flags |= RVT_S_ACK_PENDING | RVT_S_RESP_PENDING;
qp->s_nak_state = qp->r_nak_state;
@@ -751,13 +753,14 @@ static inline void hfi1_queue_rc_ack(struct rvt_qp *qp, bool is_fecn)
spin_unlock_irqrestore(&qp->s_lock, flags);
}
-static inline void hfi1_make_rc_ack_9B(struct rvt_qp *qp,
+static inline void hfi1_make_rc_ack_9B(struct hfi1_packet *packet,
struct hfi1_opa_header *opa_hdr,
u8 sc5, bool is_fecn,
u64 *pbc_flags, u32 *hwords,
u32 *nwords)
{
- struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+ struct rvt_qp *qp = packet->qp;
+ struct hfi1_ibport *ibp = rcd_to_iport(packet->rcd);
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
struct ib_header *hdr = &opa_hdr->ibh;
struct ib_other_headers *ohdr;
@@ -798,13 +801,14 @@ static inline void hfi1_make_rc_ack_9B(struct rvt_qp *qp,
hfi1_make_bth_aeth(qp, ohdr, bth0, bth1);
}
-static inline void hfi1_make_rc_ack_16B(struct rvt_qp *qp,
+static inline void hfi1_make_rc_ack_16B(struct hfi1_packet *packet,
struct hfi1_opa_header *opa_hdr,
u8 sc5, bool is_fecn,
u64 *pbc_flags, u32 *hwords,
u32 *nwords)
{
- struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+ struct rvt_qp *qp = packet->qp;
+ struct hfi1_ibport *ibp = rcd_to_iport(packet->rcd);
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
struct hfi1_16b_header *hdr = &opa_hdr->opah;
struct ib_other_headers *ohdr;
@@ -850,7 +854,7 @@ static inline void hfi1_make_rc_ack_16B(struct rvt_qp *qp,
hfi1_make_bth_aeth(qp, ohdr, bth0, bth1);
}
-typedef void (*hfi1_make_rc_ack)(struct rvt_qp *qp,
+typedef void (*hfi1_make_rc_ack)(struct hfi1_packet *packet,
struct hfi1_opa_header *opa_hdr,
u8 sc5, bool is_fecn,
u64 *pbc_flags, u32 *hwords,
@@ -870,9 +874,10 @@ typedef void (*hfi1_make_rc_ack)(struct rvt_qp *qp,
* Note that RDMA reads and atomics are handled in the
* send side QP state and send engine.
*/
-void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd,
- struct rvt_qp *qp, bool is_fecn)
+void hfi1_send_rc_ack(struct hfi1_packet *packet, bool is_fecn)
{
+ struct hfi1_ctxtdata *rcd = packet->rcd;
+ struct rvt_qp *qp = packet->qp;
struct hfi1_ibport *ibp = rcd_to_iport(rcd);
struct hfi1_qp_priv *priv = qp->priv;
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
@@ -889,14 +894,14 @@ void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd,
/* Don't send ACK or NAK if a RDMA read or atomic is pending. */
if (qp->s_flags & RVT_S_RESP_PENDING) {
- hfi1_queue_rc_ack(qp, is_fecn);
+ hfi1_queue_rc_ack(packet, is_fecn);
return;
}
/* Ensure s_rdma_ack_cnt changes are committed */
smp_read_barrier_depends();
if (qp->s_rdma_ack_cnt) {
- hfi1_queue_rc_ack(qp, is_fecn);
+ hfi1_queue_rc_ack(packet, is_fecn);
return;
}
@@ -905,7 +910,7 @@ void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd,
return;
/* Make the appropriate header */
- hfi1_make_rc_ack_tbl[priv->hdr_type](qp, &opa_hdr, sc5, is_fecn,
+ hfi1_make_rc_ack_tbl[priv->hdr_type](packet, &opa_hdr, sc5, is_fecn,
&pbc_flags, &hwords, &nwords);
plen = 2 /* PBC */ + hwords + nwords;
@@ -919,7 +924,7 @@ void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd,
* so that when enough buffer space becomes available,
* the ACK is sent ahead of other outgoing packets.
*/
- hfi1_queue_rc_ack(qp, is_fecn);
+ hfi1_queue_rc_ack(packet, is_fecn);
return;
}
trace_ack_output_ibhdr(dd_from_ibdev(qp->ibqp.device),
@@ -1537,7 +1542,7 @@ static void rc_rcv_resp(struct hfi1_packet *packet)
void *data = packet->payload;
u32 tlen = packet->tlen;
struct rvt_qp *qp = packet->qp;
- struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+ struct hfi1_ibport *ibp;
struct ib_other_headers *ohdr = packet->ohdr;
struct rvt_swqe *wqe;
enum ib_wc_status status;
@@ -1695,6 +1700,7 @@ static void rc_rcv_resp(struct hfi1_packet *packet)
goto ack_err;
ack_seq_err:
+ ibp = rcd_to_iport(rcd);
rdma_seq_err(qp, ibp, psn, rcd);
goto ack_done;
@@ -2476,7 +2482,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
qp->r_ack_psn = qp->r_psn;
send_ack:
- hfi1_send_rc_ack(rcd, qp, is_fecn);
+ hfi1_send_rc_ack(packet, is_fecn);
}
void hfi1_rc_hdrerr(
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index 90a170e..ec26ed2 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -363,8 +363,7 @@ void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
enum ib_wc_status status);
-void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp,
- bool is_fecn);
+void hfi1_send_rc_ack(struct hfi1_packet *packet, bool is_fecn);
int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
--
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] 27+ messages in thread* [PATCH for-next 12/16] IB/hfi1: Remove unnecessary fecn and becn fields
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (10 preceding siblings ...)
2018-01-26 14:32 ` [PATCH for-next 11/16] IB/hfi1: Look up ibport using a pointer in receive path Dennis Dalessandro
@ 2018-01-26 14:32 ` Dennis Dalessandro
2018-01-26 14:33 ` [PATCH for-next 13/16] IB/hfi1: Optimize process_receive_ib() Dennis Dalessandro
` (3 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:32 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Mike Marciniszyn, Sebastian Sanchez
From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
packet->fecn and packet->becn are calculated in the hot path
and are never used. Remove these fields as they show to be
costly in a profile. Also, remove initialization for
becn and fecn in process_ecn() as they're unconditionally
assigned in the function and ensure fecn and becn variables
use a boolean type.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/driver.c | 4 ----
drivers/infiniband/hw/hfi1/hfi.h | 16 +++++-----------
drivers/infiniband/hw/hfi1/rc.c | 2 +-
drivers/infiniband/hw/hfi1/ruc.c | 4 ++--
drivers/infiniband/hw/hfi1/trace.c | 8 ++++----
drivers/infiniband/hw/hfi1/trace_ibhdrs.h | 16 ++++++++--------
include/rdma/ib_hdrs.h | 10 ++++------
7 files changed, 24 insertions(+), 36 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index bdae9fb..ea09c56 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -1440,8 +1440,6 @@ static int hfi1_setup_9B_packet(struct hfi1_packet *packet)
packet->sc = hfi1_9B_get_sc5(hdr, packet->rhf);
packet->pad = ib_bth_get_pad(packet->ohdr);
packet->extra_byte = 0;
- packet->fecn = ib_bth_get_fecn(packet->ohdr);
- packet->becn = ib_bth_get_becn(packet->ohdr);
packet->pkey = ib_bth_get_pkey(packet->ohdr);
packet->migrated = ib_bth_is_migration(packet->ohdr);
@@ -1510,8 +1508,6 @@ static int hfi1_setup_bypass_packet(struct hfi1_packet *packet)
packet->sl = ibp->sc_to_sl[packet->sc];
packet->pad = hfi1_16B_bth_get_pad(packet->ohdr);
packet->extra_byte = SIZE_OF_LT;
- packet->fecn = hfi1_16B_get_fecn(packet->hdr);
- packet->becn = hfi1_16B_get_becn(packet->hdr);
packet->pkey = hfi1_16B_get_pkey(packet->hdr);
packet->migrated = opa_bth_is_migration(packet->ohdr);
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index e0ed2b4..d451299 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -352,8 +352,6 @@ struct hfi1_packet {
u8 sc;
u8 sl;
u8 opcode;
- bool becn;
- bool fecn;
bool migrated;
};
@@ -1788,19 +1786,15 @@ void hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
static inline bool process_ecn(struct rvt_qp *qp, struct hfi1_packet *pkt,
bool do_cnp)
{
- struct ib_other_headers *ohdr = pkt->ohdr;
-
- u32 bth1;
- bool becn = false;
- bool fecn = false;
+ bool becn;
+ bool fecn;
if (pkt->etype == RHF_RCV_TYPE_BYPASS) {
fecn = hfi1_16B_get_fecn(pkt->hdr);
becn = hfi1_16B_get_becn(pkt->hdr);
} else {
- bth1 = be32_to_cpu(ohdr->bth[1]);
- fecn = bth1 & IB_FECN_SMASK;
- becn = bth1 & IB_BECN_SMASK;
+ fecn = ib_bth_get_fecn(pkt->ohdr);
+ becn = ib_bth_get_becn(pkt->ohdr);
}
if (unlikely(fecn || becn)) {
hfi1_process_ecn_slowpath(qp, pkt, do_cnp);
@@ -2425,7 +2419,7 @@ static inline void hfi1_make_ib_hdr(struct ib_header *hdr,
static inline void hfi1_make_16b_hdr(struct hfi1_16b_header *hdr,
u32 slid, u32 dlid,
u16 len, u16 pkey,
- u8 becn, u8 fecn, u8 l4,
+ bool becn, bool fecn, u8 l4,
u8 sc)
{
u32 lrh0 = 0;
diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c
index daf50cc..93ea03c 100644
--- a/drivers/infiniband/hw/hfi1/rc.c
+++ b/drivers/infiniband/hw/hfi1/rc.c
@@ -814,7 +814,7 @@ static inline void hfi1_make_rc_ack_16B(struct hfi1_packet *packet,
struct ib_other_headers *ohdr;
u32 bth0, bth1 = 0;
u16 len, pkey;
- u8 becn = !!is_fecn;
+ bool becn = is_fecn;
u8 l4 = OPA_16B_L4_IB_LOCAL;
u8 extra_bytes;
diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c
index e7d4e04..edef320 100644
--- a/drivers/infiniband/hw/hfi1/ruc.c
+++ b/drivers/infiniband/hw/hfi1/ruc.c
@@ -751,7 +751,7 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
ps->s_txreq->s_cur_size);
u32 nwords = SIZE_OF_CRC + ((ps->s_txreq->s_cur_size +
extra_bytes + SIZE_OF_LT) >> 2);
- u8 becn = 0;
+ bool becn = false;
if (unlikely(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH) &&
hfi1_check_mcast(rdma_ah_get_dlid(&qp->remote_ah_attr))) {
@@ -788,7 +788,7 @@ static inline void hfi1_make_ruc_header_16B(struct rvt_qp *qp,
if (qp->s_flags & RVT_S_ECN) {
qp->s_flags &= ~RVT_S_ECN;
/* we recently received a FECN, so return a BECN */
- becn = 1;
+ becn = true;
}
hfi1_make_ruc_bth(qp, ohdr, bth0, bth1, bth2);
diff --git a/drivers/infiniband/hw/hfi1/trace.c b/drivers/infiniband/hw/hfi1/trace.c
index 959a804..89bd985 100644
--- a/drivers/infiniband/hw/hfi1/trace.c
+++ b/drivers/infiniband/hw/hfi1/trace.c
@@ -138,7 +138,7 @@ u8 hfi1_trace_opa_hdr_len(struct hfi1_opa_header *opa_hdr)
}
void hfi1_trace_parse_9b_bth(struct ib_other_headers *ohdr,
- u8 *ack, u8 *becn, u8 *fecn, u8 *mig,
+ u8 *ack, bool *becn, bool *fecn, u8 *mig,
u8 *se, u8 *pad, u8 *opcode, u8 *tver,
u16 *pkey, u32 *psn, u32 *qpn)
{
@@ -184,7 +184,7 @@ void hfi1_trace_parse_9b_hdr(struct ib_header *hdr, bool sc5,
}
void hfi1_trace_parse_16b_hdr(struct hfi1_16b_header *hdr,
- u8 *age, u8 *becn, u8 *fecn,
+ u8 *age, bool *becn, bool *fecn,
u8 *l4, u8 *rc, u8 *sc,
u16 *entropy, u16 *len, u16 *pkey,
u32 *dlid, u32 *slid)
@@ -207,7 +207,7 @@ void hfi1_trace_parse_16b_hdr(struct hfi1_16b_header *hdr,
#define LRH_16B_PRN "age:%d becn:%d fecn:%d l4:%d " \
"rc:%d sc:%d pkey:0x%.4x entropy:0x%.4x"
const char *hfi1_trace_fmt_lrh(struct trace_seq *p, bool bypass,
- u8 age, u8 becn, u8 fecn, u8 l4,
+ u8 age, bool becn, bool fecn, u8 l4,
u8 lnh, const char *lnh_name, u8 lver,
u8 rc, u8 sc, u8 sl, u16 entropy,
u16 len, u16 pkey, u32 dlid, u32 slid)
@@ -235,7 +235,7 @@ void hfi1_trace_parse_16b_hdr(struct hfi1_16b_header *hdr,
"op:0x%.2x,%s se:%d m:%d pad:%d tver:%d " \
"qpn:0x%.6x a:%d psn:0x%.8x"
const char *hfi1_trace_fmt_bth(struct trace_seq *p, bool bypass,
- u8 ack, u8 becn, u8 fecn, u8 mig,
+ u8 ack, bool becn, bool fecn, u8 mig,
u8 se, u8 pad, u8 opcode, const char *opname,
u8 tver, u16 pkey, u32 psn, u32 qpn)
{
diff --git a/drivers/infiniband/hw/hfi1/trace_ibhdrs.h b/drivers/infiniband/hw/hfi1/trace_ibhdrs.h
index fb63127..2847626 100644
--- a/drivers/infiniband/hw/hfi1/trace_ibhdrs.h
+++ b/drivers/infiniband/hw/hfi1/trace_ibhdrs.h
@@ -101,7 +101,7 @@
u8 hfi1_trace_packet_hdr_len(struct hfi1_packet *packet);
const char *hfi1_trace_get_packet_l4_str(u8 l4);
void hfi1_trace_parse_9b_bth(struct ib_other_headers *ohdr,
- u8 *ack, u8 *becn, u8 *fecn, u8 *mig,
+ u8 *ack, bool *becn, bool *fecn, u8 *mig,
u8 *se, u8 *pad, u8 *opcode, u8 *tver,
u16 *pkey, u32 *psn, u32 *qpn);
void hfi1_trace_parse_9b_hdr(struct ib_header *hdr, bool sc5,
@@ -112,19 +112,19 @@ void hfi1_trace_parse_16b_bth(struct ib_other_headers *ohdr,
u8 *pad, u8 *se, u8 *tver,
u32 *psn, u32 *qpn);
void hfi1_trace_parse_16b_hdr(struct hfi1_16b_header *hdr,
- u8 *age, u8 *becn, u8 *fecn,
+ u8 *age, bool *becn, bool *fecn,
u8 *l4, u8 *rc, u8 *sc,
u16 *entropy, u16 *len, u16 *pkey,
u32 *dlid, u32 *slid);
const char *hfi1_trace_fmt_lrh(struct trace_seq *p, bool bypass,
- u8 age, u8 becn, u8 fecn, u8 l4,
+ u8 age, bool becn, bool fecn, u8 l4,
u8 lnh, const char *lnh_name, u8 lver,
u8 rc, u8 sc, u8 sl, u16 entropy,
u16 len, u16 pkey, u32 dlid, u32 slid);
const char *hfi1_trace_fmt_bth(struct trace_seq *p, bool bypass,
- u8 ack, u8 becn, u8 fecn, u8 mig,
+ u8 ack, bool becn, bool fecn, u8 mig,
u8 se, u8 pad, u8 opcode, const char *opname,
u8 tver, u16 pkey, u32 psn, u32 qpn);
@@ -148,8 +148,8 @@ void hfi1_trace_parse_16b_hdr(struct hfi1_16b_header *hdr,
__field(u8, etype)
__field(u8, ack)
__field(u8, age)
- __field(u8, becn)
- __field(u8, fecn)
+ __field(bool, becn)
+ __field(bool, fecn)
__field(u8, l2)
__field(u8, l4)
__field(u8, lnh)
@@ -290,8 +290,8 @@ void hfi1_trace_parse_16b_hdr(struct hfi1_16b_header *hdr,
__field(u8, hdr_type)
__field(u8, ack)
__field(u8, age)
- __field(u8, becn)
- __field(u8, fecn)
+ __field(bool, becn)
+ __field(bool, fecn)
__field(u8, l4)
__field(u8, lnh)
__field(u8, lver)
diff --git a/include/rdma/ib_hdrs.h b/include/rdma/ib_hdrs.h
index 2aa19ec..710c8be 100644
--- a/include/rdma/ib_hdrs.h
+++ b/include/rdma/ib_hdrs.h
@@ -313,16 +313,14 @@ static inline u32 ib_bth_get_qpn(struct ib_other_headers *ohdr)
return (u32)((be32_to_cpu(ohdr->bth[1])) & IB_QPN_MASK);
}
-static inline u8 ib_bth_get_becn(struct ib_other_headers *ohdr)
+static inline bool ib_bth_get_becn(struct ib_other_headers *ohdr)
{
- return (u8)((be32_to_cpu(ohdr->bth[1]) >> IB_BECN_SHIFT) &
- IB_BECN_MASK);
+ return !!((ohdr->bth[1]) & cpu_to_be32(IB_BECN_SMASK));
}
-static inline u8 ib_bth_get_fecn(struct ib_other_headers *ohdr)
+static inline bool ib_bth_get_fecn(struct ib_other_headers *ohdr)
{
- return (u8)((be32_to_cpu(ohdr->bth[1]) >> IB_FECN_SHIFT) &
- IB_FECN_MASK);
+ return !!((ohdr->bth[1]) & cpu_to_be32(IB_FECN_SMASK));
}
static inline u8 ib_bth_get_tver(struct ib_other_headers *ohdr)
--
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] 27+ messages in thread* [PATCH for-next 13/16] IB/hfi1: Optimize process_receive_ib()
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (11 preceding siblings ...)
2018-01-26 14:32 ` [PATCH for-next 12/16] IB/hfi1: Remove unnecessary fecn and becn fields Dennis Dalessandro
@ 2018-01-26 14:33 ` Dennis Dalessandro
2018-01-26 14:33 ` [PATCH for-next 14/16] IB/hfi1: Re-order IRQ cleanup to address driver cleanup race Dennis Dalessandro
` (2 subsequent siblings)
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:33 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Don Hiatt, Mike Marciniszyn,
Sebastian Sanchez
From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
The arguments for trace_hfi1_rcvhdr() get computed every
time in the hot path regardless of the whether the trace
is on or off. This is seen to be costly with a profile.
The handling of fault inject isolates the verbs device for
all packets regardless of the presence of a RHF_DC_ERR error.
Fix the first by computing trace_hfi1_rcvhdr() arguments within
the trace itself, so that when the trace is off, the argument
data isn't computed. Fix the second by moving the error check to
handle_eflags() when an RHF error occurs and by testing for
RHF_DC_ERR before executing the reset of handle_eflags().
Reviewed-by: Don Hiatt <don.hiatt-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/driver.c | 21 +++++++--------------
drivers/infiniband/hw/hfi1/trace_rx.h | 28 ++++++++++------------------
2 files changed, 17 insertions(+), 32 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index ea09c56..c7e6b67 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -256,7 +256,12 @@ static void rcv_hdrerr(struct hfi1_ctxtdata *rcd, struct hfi1_pportdata *ppd,
u32 mlid_base;
struct hfi1_ibport *ibp = rcd_to_iport(rcd);
struct hfi1_devdata *dd = ppd->dd;
- struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
+ struct hfi1_ibdev *verbs_dev = &dd->verbs_dev;
+ struct rvt_dev_info *rdi = &verbs_dev->rdi;
+
+ if ((packet->rhf & RHF_DC_ERR) &&
+ hfi1_dbg_fault_suppress_err(verbs_dev))
+ return;
if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR))
return;
@@ -1554,19 +1559,7 @@ int process_receive_ib(struct hfi1_packet *packet)
if (hfi1_setup_9B_packet(packet))
return RHF_RCV_CONTINUE;
- trace_hfi1_rcvhdr(packet->rcd->ppd->dd,
- packet->rcd->ctxt,
- rhf_err_flags(packet->rhf),
- RHF_RCV_TYPE_IB,
- packet->hlen,
- packet->tlen,
- packet->updegr,
- rhf_egr_index(packet->rhf));
-
- if (unlikely(
- (hfi1_dbg_fault_suppress_err(&packet->rcd->dd->verbs_dev) &&
- (packet->rhf & RHF_DC_ERR))))
- return RHF_RCV_CONTINUE;
+ trace_hfi1_rcvhdr(packet, RHF_RCV_TYPE_IB);
if (unlikely(rhf_err_flags(packet->rhf))) {
handle_eflags(packet);
diff --git a/drivers/infiniband/hw/hfi1/trace_rx.h b/drivers/infiniband/hw/hfi1/trace_rx.h
index 4d487fe..f768415 100644
--- a/drivers/infiniband/hw/hfi1/trace_rx.h
+++ b/drivers/infiniband/hw/hfi1/trace_rx.h
@@ -63,17 +63,9 @@
#define TRACE_SYSTEM hfi1_rx
TRACE_EVENT(hfi1_rcvhdr,
- TP_PROTO(struct hfi1_devdata *dd,
- u32 ctxt,
- u64 eflags,
- u32 etype,
- u32 hlen,
- u32 tlen,
- u32 updegr,
- u32 etail
- ),
- TP_ARGS(dd, ctxt, eflags, etype, hlen, tlen, updegr, etail),
- TP_STRUCT__entry(DD_DEV_ENTRY(dd)
+ TP_PROTO(struct hfi1_packet *packet, u32 etype),
+ TP_ARGS(packet, etype),
+ TP_STRUCT__entry(DD_DEV_ENTRY(packet->rcd->dd)
__field(u64, eflags)
__field(u32, ctxt)
__field(u32, etype)
@@ -82,14 +74,14 @@
__field(u32, updegr)
__field(u32, etail)
),
- TP_fast_assign(DD_DEV_ASSIGN(dd);
- __entry->eflags = eflags;
- __entry->ctxt = ctxt;
+ TP_fast_assign(DD_DEV_ASSIGN(packet->rcd->dd);
+ __entry->eflags = rhf_err_flags(packet->rhf);
+ __entry->ctxt = packet->rcd->ctxt;
__entry->etype = etype;
- __entry->hlen = hlen;
- __entry->tlen = tlen;
- __entry->updegr = updegr;
- __entry->etail = etail;
+ __entry->hlen = packet->hlen;
+ __entry->tlen = packet->tlen;
+ __entry->updegr = packet->updegr;
+ __entry->etail = rhf_egr_index(packet->rhf);
),
TP_printk(
"[%s] ctxt %d eflags 0x%llx etype %d,%s hlen %d tlen %d updegr %d etail %d",
--
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] 27+ messages in thread* [PATCH for-next 14/16] IB/hfi1: Re-order IRQ cleanup to address driver cleanup race
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (12 preceding siblings ...)
2018-01-26 14:33 ` [PATCH for-next 13/16] IB/hfi1: Optimize process_receive_ib() Dennis Dalessandro
@ 2018-01-26 14:33 ` Dennis Dalessandro
2018-01-26 14:33 ` [PATCH for-next 15/16] IB/core: Map iWarp AH type to undefined in rdma_ah_find_type Dennis Dalessandro
2018-01-26 14:33 ` [PATCH for-next 16/16] IB/ipoib: Fix for potential no-carrier state Dennis Dalessandro
15 siblings, 0 replies; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:33 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Michael J. Ruhl,
Mike Marciniszyn, Patel Jay P, Alex Estrin
From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
The pci_request_irq() interfaces always adds the IRQF_SHARED bit to
all IRQ requests.
When the kernel is built with CONFIG_DEBUG_SHIRQ config flag, if the
IRQF_SHARED bit is set, a call to the IRQ handler is made from the
__free_irq() function. This is testing a race condition between the
IRQ cleanup and an IRQ racing the cleanup. The HFI driver should be
able to handle this race, but does not.
This race can cause traces that start with this footprint:
BUG: unable to handle kernel NULL pointer dereference at (null)
Call Trace:
<hfi1 irq handler>
...
__free_irq+0x1b3/0x2d0
free_irq+0x35/0x70
pci_free_irq+0x1c/0x30
clean_up_interrupts+0x53/0xf0 [hfi1]
hfi1_start_cleanup+0x122/0x190 [hfi1]
postinit_cleanup+0x1d/0x280 [hfi1]
remove_one+0x233/0x250 [hfi1]
pci_device_remove+0x39/0xc0
Export IRQ cleanup function so it can be called from other modules.
Using the exported cleanup function:
Re-order the driver cleanup code to clean up IRQ resources before
other resources, eliminating the race.
Re-order error path for init so that the race does not occur.
Reduce severity on spurious error message for SDMA IRQs to info.
Reviewed-by: Alex Estrin <alex.estrin-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Patel Jay P <jay.p.patel-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/hw/hfi1/chip.c | 18 ++++++++++++------
drivers/infiniband/hw/hfi1/hfi.h | 1 +
drivers/infiniband/hw/hfi1/init.c | 4 +++-
3 files changed, 16 insertions(+), 7 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index 793e8b7..cb9095d 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -8319,8 +8319,8 @@ static irqreturn_t sdma_interrupt(int irq, void *data)
/* handle the interrupt(s) */
sdma_engine_interrupt(sde, status);
} else {
- dd_dev_err_ratelimited(dd, "SDMA engine %u interrupt, but no status bits set\n",
- sde->this_idx);
+ dd_dev_info_ratelimited(dd, "SDMA engine %u interrupt, but no status bits set\n",
+ sde->this_idx);
}
return IRQ_HANDLED;
}
@@ -13031,7 +13031,14 @@ static void disable_intx(struct pci_dev *pdev)
pci_intx(pdev, 0);
}
-static void clean_up_interrupts(struct hfi1_devdata *dd)
+/**
+ * hfi1_clean_up_interrupts() - Free all IRQ resources
+ * @dd: valid device data data structure
+ *
+ * Free the MSI or INTx IRQs and assoicated PCI resources,
+ * if they have been allocated.
+ */
+void hfi1_clean_up_interrupts(struct hfi1_devdata *dd)
{
int i;
@@ -13392,7 +13399,7 @@ static int set_up_interrupts(struct hfi1_devdata *dd)
return 0;
fail:
- clean_up_interrupts(dd);
+ hfi1_clean_up_interrupts(dd);
return ret;
}
@@ -14819,7 +14826,6 @@ void hfi1_start_cleanup(struct hfi1_devdata *dd)
aspm_exit(dd);
free_cntrs(dd);
free_rcverr(dd);
- clean_up_interrupts(dd);
finish_chip_resources(dd);
}
@@ -15275,7 +15281,7 @@ struct hfi1_devdata *hfi1_init_dd(struct pci_dev *pdev,
bail_free_cntrs:
free_cntrs(dd);
bail_clear_intr:
- clean_up_interrupts(dd);
+ hfi1_clean_up_interrupts(dd);
bail_cleanup:
hfi1_pcie_ddcleanup(dd);
bail_free:
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index d451299..90bc8c7 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -1960,6 +1960,7 @@ int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num,
int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
int hfi1_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent);
+void hfi1_clean_up_interrupts(struct hfi1_devdata *dd);
void hfi1_pcie_cleanup(struct pci_dev *pdev);
int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev);
void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index d89cd86..8c4f040 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -1067,8 +1067,9 @@ static void shutdown_device(struct hfi1_devdata *dd)
}
dd->flags &= ~HFI1_INITTED;
- /* mask interrupts, but not errors */
+ /* mask and clean up interrupts, but not errors */
set_intr_state(dd, 0);
+ hfi1_clean_up_interrupts(dd);
for (pidx = 0; pidx < dd->num_pports; ++pidx) {
ppd = dd->pport + pidx;
@@ -1714,6 +1715,7 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
dd_dev_err(dd, "Failed to create /dev devices: %d\n", -j);
if (initfail || ret) {
+ hfi1_clean_up_interrupts(dd);
stop_timers(dd);
flush_workqueue(ib_wq);
for (pidx = 0; pidx < dd->num_pports; ++pidx) {
--
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] 27+ messages in thread* [PATCH for-next 15/16] IB/core: Map iWarp AH type to undefined in rdma_ah_find_type
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (13 preceding siblings ...)
2018-01-26 14:33 ` [PATCH for-next 14/16] IB/hfi1: Re-order IRQ cleanup to address driver cleanup race Dennis Dalessandro
@ 2018-01-26 14:33 ` Dennis Dalessandro
[not found] ` <20180126143315.6868.67354.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2018-01-26 14:33 ` [PATCH for-next 16/16] IB/ipoib: Fix for potential no-carrier state Dennis Dalessandro
15 siblings, 1 reply; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:33 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Don Hiatt, Sean Hefty,
Ira Weiny, Parav Pandit
From: Don Hiatt <don.hiatt-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
iWarp devices do not support the creation of address handles
so return AH_ATTR_TYPE_UNDEFINED for all iWarp devices.
While we are here reduce the size of port_num to u8 and add
a comment.
Fixes: 44c58487d51a ("IB/core: Define 'ib' and 'roce' rdma_ah_attr types")
Reported-by: Parav Pandit <parav-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
CC: Sean Hefty <sean.hefty-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Ira Weiny <ira.weiny-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Don Hiatt <don.hiatt-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
include/rdma/ib_verbs.h | 15 +++++++++++----
1 files changed, 11 insertions(+), 4 deletions(-)
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index f25c036..7d2a7ea 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -881,6 +881,7 @@ enum rdma_ah_attr_type {
RDMA_AH_ATTR_TYPE_IB,
RDMA_AH_ATTR_TYPE_ROCE,
RDMA_AH_ATTR_TYPE_OPA,
+ RDMA_AH_ATTR_TYPE_UNDEFINED,
};
struct ib_ah_attr {
@@ -3789,13 +3790,19 @@ static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
grh->traffic_class = traffic_class;
}
-/*Get AH type */
+/**
+ * rdma_ah_find_type - Return address handle type.
+ *
+ * @dev: Device to be checked
+ * @port_num: Port number
+ */
static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
- u32 port_num)
+ u8 port_num)
{
- if ((rdma_protocol_roce(dev, port_num)) ||
- (rdma_protocol_iwarp(dev, port_num)))
+ if (rdma_protocol_roce(dev, port_num))
return RDMA_AH_ATTR_TYPE_ROCE;
+ else if (rdma_protocol_iwarp(dev, port_num))
+ return RDMA_AH_ATTR_TYPE_UNDEFINED;
else if ((rdma_protocol_ib(dev, port_num)) &&
(rdma_cap_opa_ah(dev, port_num)))
return RDMA_AH_ATTR_TYPE_OPA;
--
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] 27+ messages in thread* [PATCH for-next 16/16] IB/ipoib: Fix for potential no-carrier state
[not found] ` <20180126142640.6868.12402.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
` (14 preceding siblings ...)
2018-01-26 14:33 ` [PATCH for-next 15/16] IB/core: Map iWarp AH type to undefined in rdma_ah_find_type Dennis Dalessandro
@ 2018-01-26 14:33 ` Dennis Dalessandro
[not found] ` <20180126143323.6868.89162.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
15 siblings, 1 reply; 27+ messages in thread
From: Dennis Dalessandro @ 2018-01-26 14:33 UTC (permalink / raw)
To: jgg-uk2M96/98Pc, dledford-H+wXaHxf7aLQT0dZR+AlfA
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA, Mike Marciniszyn, Ira Weiny,
Alex Estrin
From: Alex Estrin <alex.estrin-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
On reboot SM can program port pkey table before ipoib registered its
event handler, which could result in missing pkey event and leave root
interface with initial pkey value from index 0.
Since OPA port starts with invalid pkey in index 0, root interface will
fail to initialize and stay down with no-carrier flag.
For IB ipoib interface may end up with pkey different from value
opensm put in pkey table idx 0, resulting in connectivity issues
(different mcast groups, for example).
Close the window by calling event handler after registration
to make sure ipoib pkey is in sync with port pkey table.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Ira Weiny <ira.weiny-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Alex Estrin <alex.estrin-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/ulp/ipoib/ipoib_main.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 5930c7d..161ba8c 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -2306,6 +2306,9 @@ void ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
priv->ca, ipoib_event);
ib_register_event_handler(&priv->event_handler);
+ /* call event handler to ensure pkey in sync */
+ queue_work(ipoib_workqueue, &priv->flush_heavy);
+
result = register_netdev(priv->dev);
if (result) {
pr_warn("%s: couldn't register ipoib port %d; error %d\n",
--
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] 27+ messages in thread