[PATCH 00/18] DPAA PMD improvements

[PATCH 00/18] DPAA PMD improvements

[Hemant Agrawal]

This patch series add various improvement and performance related
optimizations for DPAA PMD

Ashish Jain (2):
  net/dpaa: fix the mbuf packet type if zero
  net/dpaa: set the correct frame size in device MTU

Hemant Agrawal (11):
  net/dpaa: fix coverity reported issues
  net/dpaa: fix FW version code
  bus/dpaa: update platform soc value register routines
  net/dpaa: add frame count based tail drop with CGR
  bus/dpaa: add support to create dynamic HW portal
  bus/dpaa: query queue frame count support
  net/dpaa: add Rx queue count support
  net/dpaa: add support for loopback API
  app/testpmd: add support for loopback config for dpaa
  bus/dpaa: add support for static queues
  net/dpaa: integrate the support of push mode in PMD

Nipun Gupta (5):
  bus/dpaa: optimize the qman HW stashing settings
  bus/dpaa: optimize the endianness conversions
  net/dpaa: change Tx HW budget to 7
  net/dpaa: optimize the Tx burst
  net/dpaa: optimize Rx path

 app/test-pmd/Makefile                     |   4 +
 app/test-pmd/cmdline.c                    |   7 +
 doc/guides/nics/dpaa.rst                  |  11 ++
 drivers/bus/dpaa/base/qbman/qman.c        | 172 ++++++++++++++++++--
 drivers/bus/dpaa/base/qbman/qman.h        |   4 +-
 drivers/bus/dpaa/base/qbman/qman_driver.c | 153 +++++++++++++++---
 drivers/bus/dpaa/base/qbman/qman_priv.h   |   6 +-
 drivers/bus/dpaa/dpaa_bus.c               |  43 ++++-
 drivers/bus/dpaa/include/fsl_qman.h       |  44 +++--
 drivers/bus/dpaa/include/fsl_usd.h        |   4 +
 drivers/bus/dpaa/include/process.h        |  11 +-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  18 +++
 drivers/bus/dpaa/rte_dpaa_bus.h           |  15 ++
 drivers/net/dpaa/Makefile                 |   3 +
 drivers/net/dpaa/dpaa_ethdev.c            | 259 ++++++++++++++++++++++++++----
 drivers/net/dpaa/dpaa_ethdev.h            |  21 ++-
 drivers/net/dpaa/dpaa_rxtx.c              | 163 +++++++++++++------
 drivers/net/dpaa/dpaa_rxtx.h              |   7 +-
 drivers/net/dpaa/rte_pmd_dpaa.h           |  37 +++++
 drivers/net/dpaa/rte_pmd_dpaa_version.map |   8 +
 20 files changed, 837 insertions(+), 153 deletions(-)
 create mode 100644 drivers/net/dpaa/rte_pmd_dpaa.h

-- 
2.7.4

[PATCH 01/18] net/dpaa: fix coverity reported issues

[Hemant Agrawal]

Fixes: 05ba55bc2b1a ("net/dpaa: add packet dump for debugging")
Fixes: 37f9b54bd3cf ("net/dpaa: support Tx and Rx queue setup")
Cc: stable@dpdk.org

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 6 +++---
 drivers/net/dpaa/dpaa_rxtx.c   | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index cf5a2ec..3023302 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -723,7 +723,7 @@ static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 static int dpaa_rx_queue_init(struct qman_fq *fq,
 			      uint32_t fqid)
 {
-	struct qm_mcc_initfq opts;
+	struct qm_mcc_initfq opts = {0};
 	int ret;
 
 	PMD_INIT_FUNC_TRACE();
@@ -769,7 +769,7 @@ static int dpaa_rx_queue_init(struct qman_fq *fq,
 static int dpaa_tx_queue_init(struct qman_fq *fq,
 			      struct fman_if *fman_intf)
 {
-	struct qm_mcc_initfq opts;
+	struct qm_mcc_initfq opts = {0};
 	int ret;
 
 	PMD_INIT_FUNC_TRACE();
@@ -800,7 +800,7 @@ static int dpaa_tx_queue_init(struct qman_fq *fq,
 /* Initialise a DEBUG FQ ([rt]x_error, rx_default). */
 static int dpaa_debug_queue_init(struct qman_fq *fq, uint32_t fqid)
 {
-	struct qm_mcc_initfq opts;
+	struct qm_mcc_initfq opts = {0};
 	int ret;
 
 	PMD_INIT_FUNC_TRACE();
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 41e57f2..771e141 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -665,7 +665,7 @@ tx_on_external_pool(struct qman_fq *txq, struct rte_mbuf *mbuf,
 		return 1;
 	}
 
-	DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, dpaa_intf->bp_info->bpid);
+	DPAA_MBUF_TO_CONTIG_FD(dmable_mbuf, fd_arr, dpaa_intf->bp_info->bpid);
 
 	return 0;
 }
-- 
2.7.4

[PATCH 02/18] net/dpaa: fix the mbuf packet type if zero

[Hemant Agrawal]

From: Ashish Jain <ashish.jain@nxp.com>

Populate the mbuf field packet_type which is required
for calculating checksum while transmitting frames

Fixes: 8cffdcbe85aa ("net/dpaa: support scattered Rx")
Cc: stable@dpdk.org

Signed-off-by: Ashish Jain <ashish.jain@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 771e141..c0cfec9 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -58,6 +58,7 @@
 #include <rte_ip.h>
 #include <rte_tcp.h>
 #include <rte_udp.h>
+#include <rte_net.h>
 
 #include "dpaa_ethdev.h"
 #include "dpaa_rxtx.h"
@@ -504,6 +505,15 @@ dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 	fd->opaque_addr = 0;
 
 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
+		if (!mbuf->packet_type) {
+			struct rte_net_hdr_lens hdr_lens;
+
+			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+					| RTE_PTYPE_L4_MASK);
+			mbuf->l2_len = hdr_lens.l2_len;
+			mbuf->l3_len = hdr_lens.l3_len;
+		}
 		if (temp->data_off < DEFAULT_TX_ICEOF
 			+ sizeof(struct dpaa_eth_parse_results_t))
 			temp->data_off = DEFAULT_TX_ICEOF
@@ -611,6 +621,15 @@ tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
 	}
 
 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
+		if (!mbuf->packet_type) {
+			struct rte_net_hdr_lens hdr_lens;
+
+			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+					| RTE_PTYPE_L4_MASK);
+			mbuf->l2_len = hdr_lens.l2_len;
+			mbuf->l3_len = hdr_lens.l3_len;
+		}
 		if (mbuf->data_off < (DEFAULT_TX_ICEOF +
 		    sizeof(struct dpaa_eth_parse_results_t))) {
 			DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
-- 
2.7.4

[PATCH 03/18] net/dpaa: fix FW version code

[Hemant Agrawal]

fix the soc id path and missing fclose

Fixes: cf0fab1d2ca5 ("net/dpaa: support firmware version get API")
Cc: stable@dpdk.org

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 14 +++++---------
 drivers/net/dpaa/dpaa_ethdev.h |  2 +-
 2 files changed, 6 insertions(+), 10 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 3023302..29678c5 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -212,19 +212,15 @@ dpaa_fw_version_get(struct rte_eth_dev *dev __rte_unused,
 		DPAA_PMD_ERR("Unable to open SoC device");
 		return -ENOTSUP; /* Not supported on this infra */
 	}
-
-	ret = fscanf(svr_file, "svr:%x", &svr_ver);
-	if (ret <= 0) {
+	if (fscanf(svr_file, "svr:%x", &svr_ver) <= 0)
 		DPAA_PMD_ERR("Unable to read SoC device");
-		return -ENOTSUP; /* Not supported on this infra */
-	}
 
-	ret = snprintf(fw_version, fw_size,
-		       "svr:%x-fman-v%x",
-		       svr_ver,
-		       fman_ip_rev);
+	fclose(svr_file);
 
+	ret = snprintf(fw_version, fw_size, "SVR:%x-fman-v%x",
+		       svr_ver, fman_ip_rev);
 	ret += 1; /* add the size of '\0' */
+
 	if (fw_size < (uint32_t)ret)
 		return ret;
 	else
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 5457d61..ec5ae13 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -46,7 +46,7 @@
 /* DPAA SoC identifier; If this is not available, it can be concluded
  * that board is non-DPAA. Single slot is currently supported.
  */
-#define DPAA_SOC_ID_FILE		"sys/devices/soc0/soc_id"
+#define DPAA_SOC_ID_FILE		"/sys/devices/soc0/soc_id"
 
 #define DPAA_MBUF_HW_ANNOTATION		64
 #define DPAA_FD_PTA_SIZE		64
-- 
2.7.4

[PATCH 04/18] bus/dpaa: update platform soc value register routines

[Hemant Agrawal]

This patch update the logic and expose the soc value
register, so that it can be used by other modules as well.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/dpaa_bus.c               | 12 ++++++++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  8 ++++++++
 drivers/bus/dpaa/rte_dpaa_bus.h           | 11 +++++++++++
 drivers/net/dpaa/dpaa_ethdev.c            |  4 +++-
 drivers/net/dpaa/dpaa_ethdev.h            |  5 -----
 5 files changed, 34 insertions(+), 6 deletions(-)

diff --git a/drivers/bus/dpaa/dpaa_bus.c b/drivers/bus/dpaa/dpaa_bus.c
index 1cc8c89..f1bc62a 100644
--- a/drivers/bus/dpaa/dpaa_bus.c
+++ b/drivers/bus/dpaa/dpaa_bus.c
@@ -77,6 +77,8 @@ struct netcfg_info *dpaa_netcfg;
 /* define a variable to hold the portal_key, once created.*/
 pthread_key_t dpaa_portal_key;
 
+unsigned int dpaa_svr_family;
+
 RTE_DEFINE_PER_LCORE(bool, _dpaa_io);
 
 static inline void
@@ -443,6 +445,8 @@ rte_dpaa_bus_probe(void)
 	int ret = -1;
 	struct rte_dpaa_device *dev;
 	struct rte_dpaa_driver *drv;
+	FILE *svr_file = NULL;
+	unsigned int svr_ver;
 
 	BUS_INIT_FUNC_TRACE();
 
@@ -462,6 +466,14 @@ rte_dpaa_bus_probe(void)
 			break;
 		}
 	}
+
+	svr_file = fopen(DPAA_SOC_ID_FILE, "r");
+	if (svr_file) {
+		if (fscanf(svr_file, "svr:%x", &svr_ver) > 0)
+			dpaa_svr_family = svr_ver & SVR_MASK;
+		fclose(svr_file);
+	}
+
 	return 0;
 }
 
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index fb9d532..eeeb458 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -64,3 +64,11 @@ DPDK_17.11 {
 
 	local: *;
 };
+
+DPDK_18.02 {
+	global:
+
+	dpaa_svr_family;
+
+	local: *;
+} DPDK_17.11;
diff --git a/drivers/bus/dpaa/rte_dpaa_bus.h b/drivers/bus/dpaa/rte_dpaa_bus.h
index eafc944..40caf72 100644
--- a/drivers/bus/dpaa/rte_dpaa_bus.h
+++ b/drivers/bus/dpaa/rte_dpaa_bus.h
@@ -46,6 +46,17 @@
 #define DEV_TO_DPAA_DEVICE(ptr)	\
 		container_of(ptr, struct rte_dpaa_device, device)
 
+/* DPAA SoC identifier; If this is not available, it can be concluded
+ * that board is non-DPAA. Single slot is currently supported.
+ */
+#define DPAA_SOC_ID_FILE	"/sys/devices/soc0/soc_id"
+
+#define SVR_LS1043A_FAMILY	0x87920000
+#define SVR_LS1046A_FAMILY	0x87070000
+#define SVR_MASK		0xffff0000
+
+extern unsigned int dpaa_svr_family;
+
 struct rte_dpaa_device;
 struct rte_dpaa_driver;
 
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 29678c5..4ad9afc 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -212,7 +212,9 @@ dpaa_fw_version_get(struct rte_eth_dev *dev __rte_unused,
 		DPAA_PMD_ERR("Unable to open SoC device");
 		return -ENOTSUP; /* Not supported on this infra */
 	}
-	if (fscanf(svr_file, "svr:%x", &svr_ver) <= 0)
+	if (fscanf(svr_file, "svr:%x", &svr_ver) > 0)
+		dpaa_svr_family = svr_ver & SVR_MASK;
+	else
 		DPAA_PMD_ERR("Unable to read SoC device");
 
 	fclose(svr_file);
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index ec5ae13..3f06d63 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -43,11 +43,6 @@
 #include <of.h>
 #include <netcfg.h>
 
-/* DPAA SoC identifier; If this is not available, it can be concluded
- * that board is non-DPAA. Single slot is currently supported.
- */
-#define DPAA_SOC_ID_FILE		"/sys/devices/soc0/soc_id"
-
 #define DPAA_MBUF_HW_ANNOTATION		64
 #define DPAA_FD_PTA_SIZE		64
 
-- 
2.7.4

[PATCH 05/18] net/dpaa: set the correct frame size in device MTU

[Hemant Agrawal]

From: Ashish Jain <ashish.jain@nxp.com>

Setting correct frame size in dpaa_dev_mtu_set
api call. Also setting correct max frame size in
hardware in dev_configure for jumbo frames

Signed-off-by: Ashish Jain <ashish.jain@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 20 +++++++++++++-------
 drivers/net/dpaa/dpaa_ethdev.h |  4 ++++
 2 files changed, 17 insertions(+), 7 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 4ad9afc..adcc219 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -111,19 +111,21 @@ static int
 dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	uint32_t frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN
+				+ VLAN_TAG_SIZE;
 
 	PMD_INIT_FUNC_TRACE();
 
-	if (mtu < ETHER_MIN_MTU)
+	if ((mtu < ETHER_MIN_MTU) || (frame_size > DPAA_MAX_RX_PKT_LEN))
 		return -EINVAL;
-	if (mtu > ETHER_MAX_LEN)
+	if (frame_size > ETHER_MAX_LEN)
 		dev->data->dev_conf.rxmode.jumbo_frame = 1;
 	else
 		dev->data->dev_conf.rxmode.jumbo_frame = 0;
 
-	dev->data->dev_conf.rxmode.max_rx_pkt_len = mtu;
+	dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
 
-	fman_if_set_maxfrm(dpaa_intf->fif, mtu);
+	fman_if_set_maxfrm(dpaa_intf->fif, frame_size);
 
 	return 0;
 }
@@ -131,15 +133,19 @@ dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 static int
 dpaa_eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 {
+	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+
 	PMD_INIT_FUNC_TRACE();
 
 	if (dev->data->dev_conf.rxmode.jumbo_frame == 1) {
 		if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
-		    DPAA_MAX_RX_PKT_LEN)
-			return dpaa_mtu_set(dev,
+		    DPAA_MAX_RX_PKT_LEN) {
+			fman_if_set_maxfrm(dpaa_intf->fif,
 				dev->data->dev_conf.rxmode.max_rx_pkt_len);
-		else
+			return 0;
+		} else {
 			return -1;
+		}
 	}
 	return 0;
 }
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 3f06d63..ef726d3 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -71,6 +71,10 @@
 /*Maximum number of slots available in TX ring*/
 #define MAX_TX_RING_SLOTS	8
 
+#ifndef VLAN_TAG_SIZE
+#define VLAN_TAG_SIZE   4 /** < Vlan Header Length */
+#endif
+
 /* PCD frame queues */
 #define DPAA_PCD_FQID_START		0x400
 #define DPAA_PCD_FQID_MULTIPLIER	0x100
-- 
2.7.4

[PATCH 06/18] net/dpaa: add frame count based tail drop with CGR

[Hemant Agrawal]

Replace the byte based tail queue congestion support
with frame count based congestion groups.
It can easily map to number of RX descriptors for a queue.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  5 ++
 drivers/net/dpaa/dpaa_ethdev.c            | 98 +++++++++++++++++++++++++++----
 drivers/net/dpaa/dpaa_ethdev.h            |  8 +--
 3 files changed, 97 insertions(+), 14 deletions(-)

diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index eeeb458..f412362 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -69,6 +69,11 @@ DPDK_18.02 {
 	global:
 
 	dpaa_svr_family;
+	qman_alloc_cgrid_range;
+	qman_create_cgr;
+	qman_delete_cgr;
+	qman_modify_cgr;
+	qman_release_cgrid_range;
 
 	local: *;
 } DPDK_17.11;
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index adcc219..6482998 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -73,6 +73,9 @@
 /* Keep track of whether QMAN and BMAN have been globally initialized */
 static int is_global_init;
 
+/* Per FQ Taildrop in frame count */
+static unsigned int td_threshold = CGR_RX_PERFQ_THRESH;
+
 struct rte_dpaa_xstats_name_off {
 	char name[RTE_ETH_XSTATS_NAME_SIZE];
 	uint32_t offset;
@@ -447,12 +450,13 @@ static void dpaa_eth_multicast_disable(struct rte_eth_dev *dev)
 
 static
 int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
-			    uint16_t nb_desc __rte_unused,
+			    uint16_t nb_desc,
 			    unsigned int socket_id __rte_unused,
 			    const struct rte_eth_rxconf *rx_conf __rte_unused,
 			    struct rte_mempool *mp)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	struct qman_fq *rxq = &dpaa_intf->rx_queues[queue_idx];
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -488,7 +492,23 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 			    dpaa_intf->name, fd_offset,
 			fman_if_get_fdoff(dpaa_intf->fif));
 	}
-	dev->data->rx_queues[queue_idx] = &dpaa_intf->rx_queues[queue_idx];
+
+	dev->data->rx_queues[queue_idx] = rxq;
+
+	/* configure the CGR size as per the desc size */
+	if (dpaa_intf->cgr_rx) {
+		struct qm_mcc_initcgr cgr_opts = {0};
+		int ret;
+
+		/* Enable tail drop with cgr on this queue */
+		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, nb_desc, 0);
+		ret = qman_modify_cgr(dpaa_intf->cgr_rx, 0, &cgr_opts);
+		if (ret) {
+			DPAA_PMD_WARN(
+				"rx taildrop modify fail on fqid %d (ret=%d)",
+				rxq->fqid, ret);
+		}
+	}
 
 	return 0;
 }
@@ -724,11 +744,21 @@ static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 }
 
 /* Initialise an Rx FQ */
-static int dpaa_rx_queue_init(struct qman_fq *fq,
+static int dpaa_rx_queue_init(struct qman_fq *fq, struct qman_cgr *cgr_rx,
 			      uint32_t fqid)
 {
 	struct qm_mcc_initfq opts = {0};
 	int ret;
+	u32 flags = 0;
+	struct qm_mcc_initcgr cgr_opts = {
+		.we_mask = QM_CGR_WE_CS_THRES |
+				QM_CGR_WE_CSTD_EN |
+				QM_CGR_WE_MODE,
+		.cgr = {
+			.cstd_en = QM_CGR_EN,
+			.mode = QMAN_CGR_MODE_FRAME
+		}
+	};
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -758,12 +788,24 @@ static int dpaa_rx_queue_init(struct qman_fq *fq,
 	opts.fqd.context_a.stashing.data_cl = DPAA_IF_RX_DATA_STASH;
 	opts.fqd.context_a.stashing.context_cl = DPAA_IF_RX_CONTEXT_STASH;
 
-	/*Enable tail drop */
-	opts.we_mask = opts.we_mask | QM_INITFQ_WE_TDTHRESH;
-	opts.fqd.fq_ctrl = opts.fqd.fq_ctrl | QM_FQCTRL_TDE;
-	qm_fqd_taildrop_set(&opts.fqd.td, CONG_THRESHOLD_RX_Q, 1);
-
-	ret = qman_init_fq(fq, 0, &opts);
+	if (cgr_rx) {
+		/* Enable tail drop with cgr on this queue */
+		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, td_threshold, 0);
+		cgr_rx->cb = NULL;
+		ret = qman_create_cgr(cgr_rx, QMAN_CGR_FLAG_USE_INIT,
+				      &cgr_opts);
+		if (ret) {
+			DPAA_PMD_WARN(
+				"rx taildrop init fail on rx fqid %d (ret=%d)",
+				fqid, ret);
+			goto without_cgr;
+		}
+		opts.we_mask |= QM_INITFQ_WE_CGID;
+		opts.fqd.cgid = cgr_rx->cgrid;
+		opts.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+	}
+without_cgr:
+	ret = qman_init_fq(fq, flags, &opts);
 	if (ret)
 		DPAA_PMD_ERR("init rx fqid %d failed with ret: %d", fqid, ret);
 	return ret;
@@ -845,6 +887,7 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 	struct fm_eth_port_cfg *cfg;
 	struct fman_if *fman_intf;
 	struct fman_if_bpool *bp, *tmp_bp;
+	uint32_t cgrid[DPAA_MAX_NUM_PCD_QUEUES];
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -881,10 +924,31 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 
 	dpaa_intf->rx_queues = rte_zmalloc(NULL,
 		sizeof(struct qman_fq) * num_rx_fqs, MAX_CACHELINE);
+
+	/* If congestion control is enabled globally*/
+	if (td_threshold) {
+		dpaa_intf->cgr_rx = rte_zmalloc(NULL,
+			sizeof(struct qman_cgr) * num_rx_fqs, MAX_CACHELINE);
+
+		ret = qman_alloc_cgrid_range(&cgrid[0], num_rx_fqs, 1, 0);
+		if (ret != num_rx_fqs) {
+			DPAA_PMD_WARN("insufficient CGRIDs available");
+			return -EINVAL;
+		}
+	} else {
+		dpaa_intf->cgr_rx = NULL;
+	}
+
 	for (loop = 0; loop < num_rx_fqs; loop++) {
 		fqid = DPAA_PCD_FQID_START + dpaa_intf->ifid *
 			DPAA_PCD_FQID_MULTIPLIER + loop;
-		ret = dpaa_rx_queue_init(&dpaa_intf->rx_queues[loop], fqid);
+
+		if (dpaa_intf->cgr_rx)
+			dpaa_intf->cgr_rx[loop].cgrid = cgrid[loop];
+
+		ret = dpaa_rx_queue_init(&dpaa_intf->rx_queues[loop],
+			dpaa_intf->cgr_rx ? &dpaa_intf->cgr_rx[loop] : NULL,
+			fqid);
 		if (ret)
 			return ret;
 		dpaa_intf->rx_queues[loop].dpaa_intf = dpaa_intf;
@@ -939,6 +1003,7 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 		DPAA_PMD_ERR("Failed to allocate %d bytes needed to "
 						"store MAC addresses",
 				ETHER_ADDR_LEN * DPAA_MAX_MAC_FILTER);
+		rte_free(dpaa_intf->cgr_rx);
 		rte_free(dpaa_intf->rx_queues);
 		rte_free(dpaa_intf->tx_queues);
 		dpaa_intf->rx_queues = NULL;
@@ -977,6 +1042,7 @@ static int
 dpaa_dev_uninit(struct rte_eth_dev *dev)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	int loop;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -994,6 +1060,18 @@ dpaa_dev_uninit(struct rte_eth_dev *dev)
 	if (dpaa_intf->fc_conf)
 		rte_free(dpaa_intf->fc_conf);
 
+	/* Release RX congestion Groups */
+	if (dpaa_intf->cgr_rx) {
+		for (loop = 0; loop < dpaa_intf->nb_rx_queues; loop++)
+			qman_delete_cgr(&dpaa_intf->cgr_rx[loop]);
+
+		qman_release_cgrid_range(dpaa_intf->cgr_rx[loop].cgrid,
+					 dpaa_intf->nb_rx_queues);
+	}
+
+	rte_free(dpaa_intf->cgr_rx);
+	dpaa_intf->cgr_rx = NULL;
+
 	rte_free(dpaa_intf->rx_queues);
 	dpaa_intf->rx_queues = NULL;
 
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index ef726d3..b26e411 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -60,10 +60,8 @@
 #define DPAA_MIN_RX_BUF_SIZE 512
 #define DPAA_MAX_RX_PKT_LEN  10240
 
-/* RX queue tail drop threshold
- * currently considering 32 KB packets.
- */
-#define CONG_THRESHOLD_RX_Q  (32 * 1024)
+/* RX queue tail drop threshold (CGR Based) in frame count */
+#define CGR_RX_PERFQ_THRESH 256
 
 /*max mac filter for memac(8) including primary mac addr*/
 #define DPAA_MAX_MAC_FILTER (MEMAC_NUM_OF_PADDRS + 1)
@@ -79,6 +77,7 @@
 #define DPAA_PCD_FQID_START		0x400
 #define DPAA_PCD_FQID_MULTIPLIER	0x100
 #define DPAA_DEFAULT_NUM_PCD_QUEUES	1
+#define DPAA_MAX_NUM_PCD_QUEUES		32
 
 #define DPAA_IF_TX_PRIORITY		3
 #define DPAA_IF_RX_PRIORITY		4
@@ -128,6 +127,7 @@ struct dpaa_if {
 	char *name;
 	const struct fm_eth_port_cfg *cfg;
 	struct qman_fq *rx_queues;
+	struct qman_cgr *cgr_rx;
 	struct qman_fq *tx_queues;
 	struct qman_fq debug_queues[2];
 	uint16_t nb_rx_queues;
-- 
2.7.4

[PATCH 07/18] bus/dpaa: optimize the qman HW stashing settings

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

The settings are tuned for performance.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 87fec60..77e4eeb 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -40,6 +40,7 @@
 
 #include "qman.h"
 #include <rte_branch_prediction.h>
+#include <rte_dpaa_bus.h>
 
 /* Compilation constants */
 #define DQRR_MAXFILL	15
@@ -532,7 +533,12 @@ struct qman_portal *qman_create_portal(
 
 	p = &portal->p;
 
-	portal->use_eqcr_ci_stashing = ((qman_ip_rev >= QMAN_REV30) ? 1 : 0);
+	if (dpaa_svr_family == SVR_LS1043A_FAMILY)
+		portal->use_eqcr_ci_stashing = 3;
+	else
+		portal->use_eqcr_ci_stashing =
+					((qman_ip_rev >= QMAN_REV30) ? 1 : 0);
+
 	/*
 	 * prep the low-level portal struct with the mapped addresses from the
 	 * config, everything that follows depends on it and "config" is more
@@ -545,7 +551,7 @@ struct qman_portal *qman_create_portal(
 	 * and stash with high-than-DQRR priority.
 	 */
 	if (qm_eqcr_init(p, qm_eqcr_pvb,
-			 portal->use_eqcr_ci_stashing ? 3 : 0, 1)) {
+			 portal->use_eqcr_ci_stashing, 1)) {
 		pr_err("Qman EQCR initialisation failed\n");
 		goto fail_eqcr;
 	}
-- 
2.7.4

[PATCH 08/18] bus/dpaa: optimize the endianness conversions

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c  | 7 ++++---
 drivers/bus/dpaa/include/fsl_qman.h | 2 ++
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 77e4eeb..400d920 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -935,7 +935,7 @@ static inline unsigned int __poll_portal_fast(struct qman_portal *p,
 	do {
 		qm_dqrr_pvb_update(&p->p);
 		dq = qm_dqrr_current(&p->p);
-		if (!dq)
+		if (unlikely(!dq))
 			break;
 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 	/* If running on an LE system the fields of the
@@ -1194,6 +1194,7 @@ int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq)
 	}
 	spin_lock_init(&fq->fqlock);
 	fq->fqid = fqid;
+	fq->fqid_le = cpu_to_be32(fqid);
 	fq->flags = flags;
 	fq->state = qman_fq_state_oos;
 	fq->cgr_groupid = 0;
@@ -1981,7 +1982,7 @@ int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags)
 
 int qman_enqueue_multi(struct qman_fq *fq,
 		       const struct qm_fd *fd,
-		int frames_to_send)
+		       int frames_to_send)
 {
 	struct qman_portal *p = get_affine_portal();
 	struct qm_portal *portal = &p->p;
@@ -2003,7 +2004,7 @@ int qman_enqueue_multi(struct qman_fq *fq,
 
 	/* try to send as many frames as possible */
 	while (eqcr->available && frames_to_send--) {
-		eq->fqid = cpu_to_be32(fq->fqid);
+		eq->fqid = fq->fqid_le;
 #ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
 		eq->tag = cpu_to_be32(fq->key);
 #else
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index eedfd7e..ebcfa43 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1230,6 +1230,8 @@ struct qman_fq {
 	 */
 	spinlock_t fqlock;
 	u32 fqid;
+	u32 fqid_le;
+
 	/* DPDK Interface */
 	void *dpaa_intf;
 
-- 
2.7.4

[PATCH 09/18] bus/dpaa: add support to create dynamic HW portal

[Hemant Agrawal]

HW portal is a processing context in DPAA. This patch allow
creation of a queue specific HW portal context.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        |  69 ++++++++++++--
 drivers/bus/dpaa/base/qbman/qman_driver.c | 153 +++++++++++++++++++++++++-----
 drivers/bus/dpaa/base/qbman/qman_priv.h   |   6 +-
 drivers/bus/dpaa/dpaa_bus.c               |  31 +++++-
 drivers/bus/dpaa/include/fsl_qman.h       |  25 ++---
 drivers/bus/dpaa/include/fsl_usd.h        |   4 +
 drivers/bus/dpaa/include/process.h        |  11 ++-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |   2 +
 drivers/bus/dpaa/rte_dpaa_bus.h           |   4 +
 9 files changed, 252 insertions(+), 53 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 400d920..6ae4bb3 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -650,11 +650,52 @@ struct qman_portal *qman_create_portal(
 	return NULL;
 }
 
+#define MAX_GLOBAL_PORTLAS 8
+static struct qman_portal global_portals[MAX_GLOBAL_PORTLAS];
+static int global_portals_used[MAX_GLOBAL_PORTLAS];
+
+static struct qman_portal *
+qman_alloc_global_portal(void)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_GLOBAL_PORTLAS; i++) {
+		if (global_portals_used[i] == 0) {
+			global_portals_used[i] = 1;
+			return &global_portals[i];
+		}
+	}
+	pr_err("No portal available (%x)\n", MAX_GLOBAL_PORTLAS);
+
+	return NULL;
+}
+
+static int
+qman_free_global_portal(struct qman_portal *portal)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_GLOBAL_PORTLAS; i++) {
+		if (&global_portals[i] == portal) {
+			global_portals_used[i] = 0;
+			return 0;
+		}
+	}
+	return -1;
+}
+
 struct qman_portal *qman_create_affine_portal(const struct qm_portal_config *c,
-					      const struct qman_cgrs *cgrs)
+					      const struct qman_cgrs *cgrs,
+					      int alloc)
 {
 	struct qman_portal *res;
-	struct qman_portal *portal = get_affine_portal();
+	struct qman_portal *portal;
+
+	if (alloc)
+		portal = qman_alloc_global_portal();
+	else
+		portal = get_affine_portal();
+
 	/* A criteria for calling this function (from qman_driver.c) is that
 	 * we're already affine to the cpu and won't schedule onto another cpu.
 	 */
@@ -704,13 +745,18 @@ void qman_destroy_portal(struct qman_portal *qm)
 	spin_lock_destroy(&qm->cgr_lock);
 }
 
-const struct qm_portal_config *qman_destroy_affine_portal(void)
+const struct qm_portal_config *
+qman_destroy_affine_portal(struct qman_portal *qp)
 {
 	/* We don't want to redirect if we're a slave, use "raw" */
-	struct qman_portal *qm = get_affine_portal();
+	struct qman_portal *qm;
 	const struct qm_portal_config *pcfg;
 	int cpu;
 
+	if (qp == NULL)
+		qm = get_affine_portal();
+	else
+		qm = qp;
 	pcfg = qm->config;
 	cpu = pcfg->cpu;
 
@@ -719,6 +765,9 @@ const struct qm_portal_config *qman_destroy_affine_portal(void)
 	spin_lock(&affine_mask_lock);
 	CPU_CLR(cpu, &affine_mask);
 	spin_unlock(&affine_mask_lock);
+
+	qman_free_global_portal(qm);
+
 	return pcfg;
 }
 
@@ -1125,27 +1174,27 @@ void qman_start_dequeues(void)
 		qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL);
 }
 
-void qman_static_dequeue_add(u32 pools)
+void qman_static_dequeue_add(u32 pools, struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 
 	pools &= p->config->pools;
 	p->sdqcr |= pools;
 	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
 }
 
-void qman_static_dequeue_del(u32 pools)
+void qman_static_dequeue_del(u32 pools, struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 
 	pools &= p->config->pools;
 	p->sdqcr &= ~pools;
 	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
 }
 
-u32 qman_static_dequeue_get(void)
+u32 qman_static_dequeue_get(struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 	return p->sdqcr;
 }
 
diff --git a/drivers/bus/dpaa/base/qbman/qman_driver.c b/drivers/bus/dpaa/base/qbman/qman_driver.c
index 7a68896..f5d4b37 100644
--- a/drivers/bus/dpaa/base/qbman/qman_driver.c
+++ b/drivers/bus/dpaa/base/qbman/qman_driver.c
@@ -57,8 +57,8 @@ void *qman_ccsr_map;
 /* The qman clock frequency */
 u32 qman_clk;
 
-static __thread int fd = -1;
-static __thread struct qm_portal_config pcfg;
+static __thread int qmfd = -1;
+static __thread struct qm_portal_config qpcfg;
 static __thread struct dpaa_ioctl_portal_map map = {
 	.type = dpaa_portal_qman
 };
@@ -77,16 +77,16 @@ static int fsl_qman_portal_init(uint32_t index, int is_shared)
 		error(0, ret, "pthread_getaffinity_np()");
 		return ret;
 	}
-	pcfg.cpu = -1;
+	qpcfg.cpu = -1;
 	for (loop = 0; loop < CPU_SETSIZE; loop++)
 		if (CPU_ISSET(loop, &cpuset)) {
-			if (pcfg.cpu != -1) {
+			if (qpcfg.cpu != -1) {
 				pr_err("Thread is not affine to 1 cpu\n");
 				return -EINVAL;
 			}
-			pcfg.cpu = loop;
+			qpcfg.cpu = loop;
 		}
-	if (pcfg.cpu == -1) {
+	if (qpcfg.cpu == -1) {
 		pr_err("Bug in getaffinity handling!\n");
 		return -EINVAL;
 	}
@@ -98,36 +98,36 @@ static int fsl_qman_portal_init(uint32_t index, int is_shared)
 		error(0, ret, "process_portal_map()");
 		return ret;
 	}
-	pcfg.channel = map.channel;
-	pcfg.pools = map.pools;
-	pcfg.index = map.index;
+	qpcfg.channel = map.channel;
+	qpcfg.pools = map.pools;
+	qpcfg.index = map.index;
 
 	/* Make the portal's cache-[enabled|inhibited] regions */
-	pcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena;
-	pcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh;
+	qpcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena;
+	qpcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh;
 
-	fd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
-	if (fd == -1) {
+	qmfd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
+	if (qmfd == -1) {
 		pr_err("QMan irq init failed\n");
 		process_portal_unmap(&map.addr);
 		return -EBUSY;
 	}
 
-	pcfg.is_shared = is_shared;
-	pcfg.node = NULL;
-	pcfg.irq = fd;
+	qpcfg.is_shared = is_shared;
+	qpcfg.node = NULL;
+	qpcfg.irq = qmfd;
 
-	portal = qman_create_affine_portal(&pcfg, NULL);
+	portal = qman_create_affine_portal(&qpcfg, NULL, 0);
 	if (!portal) {
 		pr_err("Qman portal initialisation failed (%d)\n",
-		       pcfg.cpu);
+		       qpcfg.cpu);
 		process_portal_unmap(&map.addr);
 		return -EBUSY;
 	}
 
 	irq_map.type = dpaa_portal_qman;
 	irq_map.portal_cinh = map.addr.cinh;
-	process_portal_irq_map(fd, &irq_map);
+	process_portal_irq_map(qmfd, &irq_map);
 	return 0;
 }
 
@@ -136,10 +136,10 @@ static int fsl_qman_portal_finish(void)
 	__maybe_unused const struct qm_portal_config *cfg;
 	int ret;
 
-	process_portal_irq_unmap(fd);
+	process_portal_irq_unmap(qmfd);
 
-	cfg = qman_destroy_affine_portal();
-	DPAA_BUG_ON(cfg != &pcfg);
+	cfg = qman_destroy_affine_portal(NULL);
+	DPAA_BUG_ON(cfg != &qpcfg);
 	ret = process_portal_unmap(&map.addr);
 	if (ret)
 		error(0, ret, "process_portal_unmap()");
@@ -161,14 +161,119 @@ int qman_thread_finish(void)
 
 void qman_thread_irq(void)
 {
-	qbman_invoke_irq(pcfg.irq);
+	qbman_invoke_irq(qpcfg.irq);
 
 	/* Now we need to uninhibit interrupts. This is the only code outside
 	 * the regular portal driver that manipulates any portal register, so
 	 * rather than breaking that encapsulation I am simply hard-coding the
 	 * offset to the inhibit register here.
 	 */
-	out_be32(pcfg.addr_virt[DPAA_PORTAL_CI] + 0xe0c, 0);
+	out_be32(qpcfg.addr_virt[DPAA_PORTAL_CI] + 0xe0c, 0);
+}
+
+struct qman_portal *fsl_qman_portal_create(void)
+{
+	cpu_set_t cpuset;
+	struct qman_portal *res;
+
+	struct qm_portal_config *q_pcfg;
+	int loop, ret;
+	struct dpaa_ioctl_irq_map irq_map;
+	struct dpaa_ioctl_portal_map q_map = {0};
+	int q_fd;
+
+	q_pcfg = kzalloc((sizeof(struct qm_portal_config)), 0);
+	if (!q_pcfg) {
+		error(0, -1, "q_pcfg kzalloc failed");
+		return NULL;
+	}
+
+	/* Verify the thread's cpu-affinity */
+	ret = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t),
+				     &cpuset);
+	if (ret) {
+		error(0, ret, "pthread_getaffinity_np()");
+		return NULL;
+	}
+
+	q_pcfg->cpu = -1;
+	for (loop = 0; loop < CPU_SETSIZE; loop++)
+		if (CPU_ISSET(loop, &cpuset)) {
+			if (q_pcfg->cpu != -1) {
+				pr_err("Thread is not affine to 1 cpu\n");
+				return NULL;
+			}
+			q_pcfg->cpu = loop;
+		}
+	if (q_pcfg->cpu == -1) {
+		pr_err("Bug in getaffinity handling!\n");
+		return NULL;
+	}
+
+	/* Allocate and map a qman portal */
+	q_map.type = dpaa_portal_qman;
+	q_map.index = QBMAN_ANY_PORTAL_IDX;
+	ret = process_portal_map(&q_map);
+	if (ret) {
+		error(0, ret, "process_portal_map()");
+		return NULL;
+	}
+	q_pcfg->channel = q_map.channel;
+	q_pcfg->pools = q_map.pools;
+	q_pcfg->index = q_map.index;
+
+	/* Make the portal's cache-[enabled|inhibited] regions */
+	q_pcfg->addr_virt[DPAA_PORTAL_CE] = q_map.addr.cena;
+	q_pcfg->addr_virt[DPAA_PORTAL_CI] = q_map.addr.cinh;
+
+	q_fd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
+	if (q_fd == -1) {
+		pr_err("QMan irq init failed\n");
+		goto err1;
+	}
+
+	q_pcfg->irq = q_fd;
+
+	res = qman_create_affine_portal(q_pcfg, NULL, true);
+	if (!res) {
+		pr_err("Qman portal initialisation failed (%d)\n",
+		       q_pcfg->cpu);
+		goto err2;
+	}
+
+	irq_map.type = dpaa_portal_qman;
+	irq_map.portal_cinh = q_map.addr.cinh;
+	process_portal_irq_map(q_fd, &irq_map);
+
+	return res;
+err2:
+	close(q_fd);
+err1:
+	process_portal_unmap(&q_map.addr);
+	return NULL;
+}
+
+int fsl_qman_portal_destroy(struct qman_portal *qp)
+{
+	const struct qm_portal_config *cfg;
+	struct dpaa_portal_map addr;
+	int ret;
+
+	cfg = qman_destroy_affine_portal(qp);
+	kfree(qp);
+
+	process_portal_irq_unmap(cfg->irq);
+
+	addr.cena = cfg->addr_virt[DPAA_PORTAL_CE];
+	addr.cinh = cfg->addr_virt[DPAA_PORTAL_CI];
+
+	ret = process_portal_unmap(&addr);
+	if (ret)
+		pr_err("process_portal_unmap() (%d)\n", ret);
+
+	kfree((void *)cfg);
+
+	return ret;
 }
 
 int qman_global_init(void)
diff --git a/drivers/bus/dpaa/base/qbman/qman_priv.h b/drivers/bus/dpaa/base/qbman/qman_priv.h
index 3e1d7f9..e78d90b 100644
--- a/drivers/bus/dpaa/base/qbman/qman_priv.h
+++ b/drivers/bus/dpaa/base/qbman/qman_priv.h
@@ -179,8 +179,10 @@ int qm_get_wpm(int *wpm);
 
 struct qman_portal *qman_create_affine_portal(
 			const struct qm_portal_config *config,
-			const struct qman_cgrs *cgrs);
-const struct qm_portal_config *qman_destroy_affine_portal(void);
+			const struct qman_cgrs *cgrs,
+			int alloc);
+const struct qm_portal_config *
+qman_destroy_affine_portal(struct qman_portal *q);
 
 struct qm_portal_config *qm_get_unused_portal(void);
 struct qm_portal_config *qm_get_unused_portal_idx(uint32_t idx);
diff --git a/drivers/bus/dpaa/dpaa_bus.c b/drivers/bus/dpaa/dpaa_bus.c
index f1bc62a..8d74643 100644
--- a/drivers/bus/dpaa/dpaa_bus.c
+++ b/drivers/bus/dpaa/dpaa_bus.c
@@ -290,8 +290,7 @@ _dpaa_portal_init(void *arg)
  * rte_dpaa_portal_init - Wrapper over _dpaa_portal_init with thread level check
  * XXX Complete this
  */
-int
-rte_dpaa_portal_init(void *arg)
+int rte_dpaa_portal_init(void *arg)
 {
 	if (unlikely(!RTE_PER_LCORE(_dpaa_io)))
 		return _dpaa_portal_init(arg);
@@ -299,6 +298,34 @@ rte_dpaa_portal_init(void *arg)
 	return 0;
 }
 
+int
+rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq)
+{
+	/* Affine above created portal with channel*/
+	u32 sdqcr;
+	struct qman_portal *qp;
+
+	if (unlikely(!RTE_PER_LCORE(_dpaa_io)))
+		_dpaa_portal_init(arg);
+
+	/* Initialise qman specific portals */
+	qp = fsl_qman_portal_create();
+	if (!qp) {
+		DPAA_BUS_LOG(ERR, "Unable to alloc fq portal");
+		return -1;
+	}
+	fq->qp = qp;
+	sdqcr = QM_SDQCR_CHANNELS_POOL_CONV(fq->ch_id);
+	qman_static_dequeue_add(sdqcr, qp);
+
+	return 0;
+}
+
+int rte_dpaa_portal_fq_close(struct qman_fq *fq)
+{
+	return fsl_qman_portal_destroy(fq->qp);
+}
+
 void
 dpaa_portal_finish(void *arg)
 {
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index ebcfa43..c5aef2d 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1223,21 +1223,24 @@ struct qman_fq_cb {
 struct qman_fq {
 	/* Caller of qman_create_fq() provides these demux callbacks */
 	struct qman_fq_cb cb;
-	/*
-	 * These are internal to the driver, don't touch. In particular, they
-	 * may change, be removed, or extended (so you shouldn't rely on
-	 * sizeof(qman_fq) being a constant).
-	 */
-	spinlock_t fqlock;
-	u32 fqid;
+
 	u32 fqid_le;
+	u16 ch_id;
+	u8 cgr_groupid;
+	u8 is_static;
 
 	/* DPDK Interface */
 	void *dpaa_intf;
 
+	/* affined portal in case of static queue */
+	struct qman_portal *qp;
+
 	volatile unsigned long flags;
+
 	enum qman_fq_state state;
-	int cgr_groupid;
+	u32 fqid;
+	spinlock_t fqlock;
+
 	struct rb_node node;
 #ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
 	u32 key;
@@ -1416,7 +1419,7 @@ void qman_start_dequeues(void);
  * (SDQCR). The requested pools are limited to those the portal has dequeue
  * access to.
  */
-void qman_static_dequeue_add(u32 pools);
+void qman_static_dequeue_add(u32 pools, struct qman_portal *qm);
 
 /**
  * qman_static_dequeue_del - Remove pool channels from the portal SDQCR
@@ -1426,7 +1429,7 @@ void qman_static_dequeue_add(u32 pools);
  * register (SDQCR). The requested pools are limited to those the portal has
  * dequeue access to.
  */
-void qman_static_dequeue_del(u32 pools);
+void qman_static_dequeue_del(u32 pools, struct qman_portal *qp);
 
 /**
  * qman_static_dequeue_get - return the portal's current SDQCR
@@ -1435,7 +1438,7 @@ void qman_static_dequeue_del(u32 pools);
  * entire register is returned, so if only the currently-enabled pool channels
  * are desired, mask the return value with QM_SDQCR_CHANNELS_POOL_MASK.
  */
-u32 qman_static_dequeue_get(void);
+u32 qman_static_dequeue_get(struct qman_portal *qp);
 
 /**
  * qman_dca - Perform a Discrete Consumption Acknowledgment
diff --git a/drivers/bus/dpaa/include/fsl_usd.h b/drivers/bus/dpaa/include/fsl_usd.h
index a3243af..038a89d 100644
--- a/drivers/bus/dpaa/include/fsl_usd.h
+++ b/drivers/bus/dpaa/include/fsl_usd.h
@@ -100,6 +100,10 @@ void bman_thread_irq(void);
 int qman_global_init(void);
 int bman_global_init(void);
 
+/* Direct portal create and destroy */
+struct qman_portal *fsl_qman_portal_create(void);
+int fsl_qman_portal_destroy(struct qman_portal *qp);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/drivers/bus/dpaa/include/process.h b/drivers/bus/dpaa/include/process.h
index 989ddcd..352e949 100644
--- a/drivers/bus/dpaa/include/process.h
+++ b/drivers/bus/dpaa/include/process.h
@@ -72,6 +72,11 @@ enum dpaa_portal_type {
 	dpaa_portal_bman,
 };
 
+struct dpaa_portal_map {
+	void *cinh;
+	void *cena;
+};
+
 struct dpaa_ioctl_portal_map {
 	/* Input parameter, is a qman or bman portal required. */
 	enum dpaa_portal_type type;
@@ -83,10 +88,8 @@ struct dpaa_ioctl_portal_map {
 	/* Return value if the map succeeds, this gives the mapped
 	 * cache-inhibited (cinh) and cache-enabled (cena) addresses.
 	 */
-	struct dpaa_portal_map {
-		void *cinh;
-		void *cena;
-	} addr;
+	struct dpaa_portal_map addr;
+
 	/* Qman-specific return values */
 	u16 channel;
 	uint32_t pools;
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index f412362..4e3afda 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -74,6 +74,8 @@ DPDK_18.02 {
 	qman_delete_cgr;
 	qman_modify_cgr;
 	qman_release_cgrid_range;
+	rte_dpaa_portal_fq_close;
+	rte_dpaa_portal_fq_init;
 
 	local: *;
 } DPDK_17.11;
diff --git a/drivers/bus/dpaa/rte_dpaa_bus.h b/drivers/bus/dpaa/rte_dpaa_bus.h
index 40caf72..b0f7d48 100644
--- a/drivers/bus/dpaa/rte_dpaa_bus.h
+++ b/drivers/bus/dpaa/rte_dpaa_bus.h
@@ -162,6 +162,10 @@ void rte_dpaa_driver_unregister(struct rte_dpaa_driver *driver);
  */
 int rte_dpaa_portal_init(void *arg);
 
+int rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq);
+
+int rte_dpaa_portal_fq_close(struct qman_fq *fq);
+
 /**
  * Cleanup a DPAA Portal
  */
-- 
2.7.4

[PATCH 10/18] net/dpaa: change Tx HW budget to 7

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

change the TX budget to 7 to sync best with the hw.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.h | 2 +-
 drivers/net/dpaa/dpaa_rxtx.c   | 5 +++--
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index b26e411..95d745e 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -67,7 +67,7 @@
 #define DPAA_MAX_MAC_FILTER (MEMAC_NUM_OF_PADDRS + 1)
 
 /*Maximum number of slots available in TX ring*/
-#define MAX_TX_RING_SLOTS	8
+#define DPAA_TX_BURST_SIZE	7
 
 #ifndef VLAN_TAG_SIZE
 #define VLAN_TAG_SIZE   4 /** < Vlan Header Length */
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index c0cfec9..1b0ca9a 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -695,7 +695,7 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	struct rte_mbuf *mbuf, *mi = NULL;
 	struct rte_mempool *mp;
 	struct dpaa_bp_info *bp_info;
-	struct qm_fd fd_arr[MAX_TX_RING_SLOTS];
+	struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
 	uint32_t frames_to_send, loop, i = 0;
 	uint16_t state;
 	int ret;
@@ -709,7 +709,8 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	DPAA_DP_LOG(DEBUG, "Transmitting %d buffers on queue: %p", nb_bufs, q);
 
 	while (nb_bufs) {
-		frames_to_send = (nb_bufs >> 3) ? MAX_TX_RING_SLOTS : nb_bufs;
+		frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
+				DPAA_TX_BURST_SIZE : nb_bufs;
 		for (loop = 0; loop < frames_to_send; loop++, i++) {
 			mbuf = bufs[i];
 			if (RTE_MBUF_DIRECT(mbuf)) {
-- 
2.7.4

[PATCH 11/18] net/dpaa: optimize the Tx burst

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Optimize it for best case. Create a function
for TX offloads to be used in multiple legs.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 73 ++++++++++++++++++++++++++++----------------
 1 file changed, 46 insertions(+), 27 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 1b0ca9a..33cc412 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -298,6 +298,30 @@ static inline void dpaa_checksum_offload(struct rte_mbuf *mbuf,
 	fd->cmd = DPAA_FD_CMD_RPD | DPAA_FD_CMD_DTC;
 }
 
+static inline void
+dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
+{
+	if (!mbuf->packet_type) {
+		struct rte_net_hdr_lens hdr_lens;
+
+		mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+				RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+				| RTE_PTYPE_L4_MASK);
+		mbuf->l2_len = hdr_lens.l2_len;
+		mbuf->l3_len = hdr_lens.l3_len;
+	}
+	if (mbuf->data_off < (DEFAULT_TX_ICEOF +
+	    sizeof(struct dpaa_eth_parse_results_t))) {
+		DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
+			"Not enough Headroom "
+			"space for correct Checksum offload."
+			"So Calculating checksum in Software.");
+		dpaa_checksum(mbuf);
+	} else {
+		dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
+	}
+}
+
 struct rte_mbuf *
 dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
 {
@@ -620,27 +644,8 @@ tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
 		rte_pktmbuf_free(mbuf);
 	}
 
-	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
-		if (!mbuf->packet_type) {
-			struct rte_net_hdr_lens hdr_lens;
-
-			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
-					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
-					| RTE_PTYPE_L4_MASK);
-			mbuf->l2_len = hdr_lens.l2_len;
-			mbuf->l3_len = hdr_lens.l3_len;
-		}
-		if (mbuf->data_off < (DEFAULT_TX_ICEOF +
-		    sizeof(struct dpaa_eth_parse_results_t))) {
-			DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
-				"Not enough Headroom "
-				"space for correct Checksum offload."
-				"So Calculating checksum in Software.");
-			dpaa_checksum(mbuf);
-		} else {
-			dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
-		}
-	}
+	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
+		dpaa_unsegmented_checksum(mbuf, fd_arr);
 }
 
 /* Handle all mbufs on dpaa BMAN managed pool */
@@ -696,7 +701,7 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	struct rte_mempool *mp;
 	struct dpaa_bp_info *bp_info;
 	struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
-	uint32_t frames_to_send, loop, i = 0;
+	uint32_t frames_to_send, loop, sent = 0;
 	uint16_t state;
 	int ret;
 
@@ -711,10 +716,23 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	while (nb_bufs) {
 		frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
 				DPAA_TX_BURST_SIZE : nb_bufs;
-		for (loop = 0; loop < frames_to_send; loop++, i++) {
-			mbuf = bufs[i];
-			if (RTE_MBUF_DIRECT(mbuf)) {
+		for (loop = 0; loop < frames_to_send; loop++) {
+			mbuf = *(bufs++);
+			if (likely(RTE_MBUF_DIRECT(mbuf))) {
 				mp = mbuf->pool;
+				bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
+				if (likely(mp->ops_index ==
+						bp_info->dpaa_ops_index &&
+					mbuf->nb_segs == 1 &&
+					rte_mbuf_refcnt_read(mbuf) == 1)) {
+					DPAA_MBUF_TO_CONTIG_FD(mbuf,
+						&fd_arr[loop], bp_info->bpid);
+					if (mbuf->ol_flags &
+						DPAA_TX_CKSUM_OFFLOAD_MASK)
+						dpaa_unsegmented_checksum(mbuf,
+							&fd_arr[loop]);
+					continue;
+				}
 			} else {
 				mi = rte_mbuf_from_indirect(mbuf);
 				mp = mi->pool;
@@ -755,11 +773,12 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 					frames_to_send - loop);
 		}
 		nb_bufs -= frames_to_send;
+		sent += frames_to_send;
 	}
 
-	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", i, q);
+	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", sent, q);
 
-	return i;
+	return sent;
 }
 
 uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
-- 
2.7.4

[PATCH 12/18] net/dpaa: optimize Rx path

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 48 ++++++++++++++++++++------------------------
 drivers/net/dpaa/dpaa_rxtx.h |  2 +-
 2 files changed, 23 insertions(+), 27 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 33cc412..2609953 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -123,12 +123,6 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 	DPAA_DP_LOG(DEBUG, " Parsing mbuf: %p with annotations: %p", m, annot);
 
 	switch (prs) {
-	case DPAA_PKT_TYPE_NONE:
-		m->packet_type = 0;
-		break;
-	case DPAA_PKT_TYPE_ETHER:
-		m->packet_type = RTE_PTYPE_L2_ETHER;
-		break;
 	case DPAA_PKT_TYPE_IPV4:
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV4;
@@ -137,6 +131,9 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV6;
 		break;
+	case DPAA_PKT_TYPE_ETHER:
+		m->packet_type = RTE_PTYPE_L2_ETHER;
+		break;
 	case DPAA_PKT_TYPE_IPV4_FRAG:
 	case DPAA_PKT_TYPE_IPV4_FRAG_UDP:
 	case DPAA_PKT_TYPE_IPV4_FRAG_TCP:
@@ -199,6 +196,9 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
 		break;
+	case DPAA_PKT_TYPE_NONE:
+		m->packet_type = 0;
+		break;
 	/* More switch cases can be added */
 	default:
 		dpaa_slow_parsing(m, prs);
@@ -209,12 +209,11 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 					<< DPAA_PKT_L3_LEN_SHIFT;
 
 	/* Set the hash values */
-	m->hash.rss = (uint32_t)(rte_be_to_cpu_64(annot->hash));
-	m->ol_flags = PKT_RX_RSS_HASH;
+	m->hash.rss = (uint32_t)(annot->hash);
 	/* All packets with Bad checksum are dropped by interface (and
 	 * corresponding notification issued to RX error queues).
 	 */
-	m->ol_flags |= PKT_RX_IP_CKSUM_GOOD;
+	m->ol_flags = PKT_RX_RSS_HASH | PKT_RX_IP_CKSUM_GOOD;
 
 	/* Check if Vlan is present */
 	if (prs & DPAA_PARSE_VLAN_MASK)
@@ -323,7 +322,7 @@ dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
 }
 
 struct rte_mbuf *
-dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
+dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 {
 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 	struct rte_mbuf *first_seg, *prev_seg, *cur_seg, *temp;
@@ -381,34 +380,31 @@ dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
 	return first_seg;
 }
 
-static inline struct rte_mbuf *dpaa_eth_fd_to_mbuf(struct qm_fd *fd,
-							uint32_t ifid)
+static inline struct rte_mbuf *
+dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 {
-	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 	struct rte_mbuf *mbuf;
-	void *ptr;
+	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
+	void *ptr = rte_dpaa_mem_ptov(qm_fd_addr(fd));
 	uint8_t format =
 		(fd->opaque & DPAA_FD_FORMAT_MASK) >> DPAA_FD_FORMAT_SHIFT;
-	uint16_t offset =
-		(fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
-	uint32_t length = fd->opaque & DPAA_FD_LENGTH_MASK;
+	uint16_t offset;
+	uint32_t length;
 
 	DPAA_DP_LOG(DEBUG, " FD--->MBUF");
 
 	if (unlikely(format == qm_fd_sg))
 		return dpaa_eth_sg_to_mbuf(fd, ifid);
 
+	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
+
+	offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
+	length = fd->opaque & DPAA_FD_LENGTH_MASK;
+
 	/* Ignoring case when format != qm_fd_contig */
 	dpaa_display_frame(fd);
-	ptr = rte_dpaa_mem_ptov(fd->addr);
-	/* Ignoring case when ptr would be NULL. That is only possible incase
-	 * of a corrupted packet
-	 */
 
 	mbuf = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
-	/* Prefetch the Parse results and packet data to L1 */
-	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
-	rte_prefetch0((void *)((uint8_t *)ptr + offset));
 
 	mbuf->data_off = offset;
 	mbuf->data_len = length;
@@ -488,11 +484,11 @@ static struct rte_mbuf *dpaa_get_dmable_mbuf(struct rte_mbuf *mbuf,
 	if (!dpaa_mbuf)
 		return NULL;
 
-	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + mbuf->data_off, (void *)
+	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + RTE_PKTMBUF_HEADROOM, (void *)
 		((uint8_t *)(mbuf->buf_addr) + mbuf->data_off), mbuf->pkt_len);
 
 	/* Copy only the required fields */
-	dpaa_mbuf->data_off = mbuf->data_off;
+	dpaa_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 	dpaa_mbuf->pkt_len = mbuf->pkt_len;
 	dpaa_mbuf->ol_flags = mbuf->ol_flags;
 	dpaa_mbuf->packet_type = mbuf->packet_type;
diff --git a/drivers/net/dpaa/dpaa_rxtx.h b/drivers/net/dpaa/dpaa_rxtx.h
index 2ffc4ff..b434b6d 100644
--- a/drivers/net/dpaa/dpaa_rxtx.h
+++ b/drivers/net/dpaa/dpaa_rxtx.h
@@ -288,7 +288,7 @@ uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
 			      struct rte_mbuf **bufs __rte_unused,
 			      uint16_t nb_bufs __rte_unused);
 
-struct rte_mbuf *dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid);
+struct rte_mbuf *dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid);
 
 int dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 			   struct qm_fd *fd,
-- 
2.7.4

[PATCH 13/18] bus/dpaa: query queue frame count support

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 22 ++++++++++++++++++++++
 drivers/bus/dpaa/include/fsl_qman.h       |  7 +++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  1 +
 3 files changed, 30 insertions(+)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 6ae4bb3..b2f82a3 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -1750,6 +1750,28 @@ int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np)
 	return 0;
 }
 
+int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq.fqid = cpu_to_be32(fq->fqid);
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP);
+
+	if (mcr->result == QM_MCR_RESULT_OK)
+		*frm_cnt = be24_to_cpu(mcr->queryfq_np.frm_cnt);
+	else if (mcr->result == QM_MCR_RESULT_ERR_FQID)
+		return -ERANGE;
+	else if (mcr->result != QM_MCR_RESULT_OK)
+		return -EIO;
+	return 0;
+}
+
 int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq)
 {
 	struct qm_mc_command *mcc;
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index c5aef2d..9090b63 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1649,6 +1649,13 @@ int qman_query_fq_has_pkts(struct qman_fq *fq);
 int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np);
 
 /**
+ * qman_query_fq_frmcnt - Queries fq frame count
+ * @fq: the frame queue object to be queried
+ * @frm_cnt: number of frames in the queue
+ */
+int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt);
+
+/**
  * qman_query_wq - Queries work queue lengths
  * @query_dedicated: If non-zero, query length of WQs in the channel dedicated
  *		to this software portal. Otherwise, query length of WQs in a
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index 4e3afda..212c75f 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -73,6 +73,7 @@ DPDK_18.02 {
 	qman_create_cgr;
 	qman_delete_cgr;
 	qman_modify_cgr;
+	qman_query_fq_frm_cnt;
 	qman_release_cgrid_range;
 	rte_dpaa_portal_fq_close;
 	rte_dpaa_portal_fq_init;
-- 
2.7.4

[PATCH 14/18] net/dpaa: add Rx queue count support

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 6482998..53b8c87 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -539,6 +539,22 @@ static void dpaa_eth_tx_queue_release(void *txq __rte_unused)
 	PMD_INIT_FUNC_TRACE();
 }
 
+static uint32_t
+dpaa_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	struct qman_fq *rxq = &dpaa_intf->rx_queues[rx_queue_id];
+	u32 frm_cnt = 0;
+
+	PMD_INIT_FUNC_TRACE();
+
+	if (qman_query_fq_frm_cnt(rxq, &frm_cnt) == 0) {
+		RTE_LOG(DEBUG, PMD, "RX frame count for q(%d) is %u\n",
+			rx_queue_id, frm_cnt);
+	}
+	return frm_cnt;
+}
+
 static int dpaa_link_down(struct rte_eth_dev *dev)
 {
 	PMD_INIT_FUNC_TRACE();
@@ -690,6 +706,7 @@ static struct eth_dev_ops dpaa_devops = {
 	.tx_queue_setup		  = dpaa_eth_tx_queue_setup,
 	.rx_queue_release	  = dpaa_eth_rx_queue_release,
 	.tx_queue_release	  = dpaa_eth_tx_queue_release,
+	.rx_queue_count		  = dpaa_dev_rx_queue_count,
 
 	.flow_ctrl_get		  = dpaa_flow_ctrl_get,
 	.flow_ctrl_set		  = dpaa_flow_ctrl_set,
-- 
2.7.4

[PATCH 15/18] net/dpaa: add support for loopback API

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/Makefile                 |  3 +++
 drivers/net/dpaa/dpaa_ethdev.c            | 42 +++++++++++++++++++++++++++++++
 drivers/net/dpaa/rte_pmd_dpaa.h           | 37 +++++++++++++++++++++++++++
 drivers/net/dpaa/rte_pmd_dpaa_version.map |  8 ++++++
 4 files changed, 90 insertions(+)
 create mode 100644 drivers/net/dpaa/rte_pmd_dpaa.h

diff --git a/drivers/net/dpaa/Makefile b/drivers/net/dpaa/Makefile
index 171686e..a99d1ee 100644
--- a/drivers/net/dpaa/Makefile
+++ b/drivers/net/dpaa/Makefile
@@ -60,4 +60,7 @@ LDLIBS += -lrte_mempool_dpaa
 LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring
 LDLIBS += -lrte_ethdev -lrte_net -lrte_kvargs
 
+# install this header file
+SYMLINK-$(CONFIG_RTE_LIBRTE_DPAA_PMD)-include := rte_pmd_dpaa.h
+
 include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 53b8c87..fcba929 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -64,6 +64,7 @@
 
 #include <dpaa_ethdev.h>
 #include <dpaa_rxtx.h>
+#include <rte_pmd_dpaa.h>
 
 #include <fsl_usd.h>
 #include <fsl_qman.h>
@@ -110,6 +111,8 @@ static const struct rte_dpaa_xstats_name_off dpaa_xstats_strings[] = {
 		offsetof(struct dpaa_if_stats, tund)},
 };
 
+static struct rte_dpaa_driver rte_dpaa_pmd;
+
 static int
 dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 {
@@ -733,6 +736,45 @@ static struct eth_dev_ops dpaa_devops = {
 	.fw_version_get		  = dpaa_fw_version_get,
 };
 
+static bool
+is_device_supported(struct rte_eth_dev *dev, struct rte_dpaa_driver *drv)
+{
+	if (strcmp(dev->device->driver->name,
+		   drv->driver.name))
+		return false;
+
+	return true;
+}
+
+static bool
+is_dpaa_supported(struct rte_eth_dev *dev)
+{
+	return is_device_supported(dev, &rte_dpaa_pmd);
+}
+
+int
+rte_pmd_dpaa_set_tx_loopback(uint8_t port, uint8_t on)
+{
+	struct rte_eth_dev *dev;
+	struct dpaa_if *dpaa_intf;
+
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(port, -ENODEV);
+
+	dev = &rte_eth_devices[port];
+
+	if (!is_dpaa_supported(dev))
+		return -ENOTSUP;
+
+	dpaa_intf = dev->data->dev_private;
+
+	if (on)
+		fman_if_loopback_enable(dpaa_intf->fif);
+	else
+		fman_if_loopback_disable(dpaa_intf->fif);
+
+	return 0;
+}
+
 static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 {
 	struct rte_eth_fc_conf *fc_conf;
diff --git a/drivers/net/dpaa/rte_pmd_dpaa.h b/drivers/net/dpaa/rte_pmd_dpaa.h
new file mode 100644
index 0000000..4464dd4
--- /dev/null
+++ b/drivers/net/dpaa/rte_pmd_dpaa.h
@@ -0,0 +1,37 @@
+/*-
+ *   Copyright 2017 NXP.
+ *
+ *   SPDX-License-Identifier:        BSD-3-Clause
+ */
+
+#ifndef _PMD_DPAA_H_
+#define _PMD_DPAA_H_
+
+/**
+ * @file rte_pmd_dpaa.h
+ *
+ * dpaa PMD specific functions.
+ *
+ * @b EXPERIMENTAL: this API may change, or be removed, without prior notice
+ *
+ */
+
+#include <rte_ethdev.h>
+
+/**
+ * Enable/Disable TX loopback
+ *
+ * @param port
+ *    The port identifier of the Ethernet device.
+ * @param on
+ *    1 - Enable TX loopback.
+ *    0 - Disable TX loopback.
+ * @return
+ *   - (0) if successful.
+ *   - (-ENODEV) if *port* invalid.
+ *   - (-EINVAL) if bad parameter.
+ */
+int rte_pmd_dpaa_set_tx_loopback(uint8_t port,
+				 uint8_t on);
+
+#endif /* _PMD_DPAA_H_ */
diff --git a/drivers/net/dpaa/rte_pmd_dpaa_version.map b/drivers/net/dpaa/rte_pmd_dpaa_version.map
index a70bd19..d76acbd 100644
--- a/drivers/net/dpaa/rte_pmd_dpaa_version.map
+++ b/drivers/net/dpaa/rte_pmd_dpaa_version.map
@@ -2,3 +2,11 @@ DPDK_17.11 {
 
 	local: *;
 };
+
+DPDK_18.02 {
+	global:
+
+	rte_pmd_dpaa_set_tx_loopback;
+
+	local: *;
+} DPDK_17.11;
-- 
2.7.4

[PATCH 16/18] app/testpmd: add support for loopback config for dpaa

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 app/test-pmd/Makefile  | 4 ++++
 app/test-pmd/cmdline.c | 7 +++++++
 2 files changed, 11 insertions(+)

diff --git a/app/test-pmd/Makefile b/app/test-pmd/Makefile
index d21308f..f60449b 100644
--- a/app/test-pmd/Makefile
+++ b/app/test-pmd/Makefile
@@ -71,6 +71,10 @@ ifeq ($(CONFIG_RTE_LIBRTE_PMD_BOND),y)
 LDLIBS += -lrte_pmd_bond
 endif
 
+ifeq ($(CONFIG_RTE_LIBRTE_DPAA_PMD),y)
+LDLIBS += -lrte_pmd_dpaa
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
 LDLIBS += -lrte_pmd_ixgbe
 endif
diff --git a/app/test-pmd/cmdline.c b/app/test-pmd/cmdline.c
index f71d963..32096aa 100644
--- a/app/test-pmd/cmdline.c
+++ b/app/test-pmd/cmdline.c
@@ -89,6 +89,9 @@
 #include <rte_eth_bond.h>
 #include <rte_eth_bond_8023ad.h>
 #endif
+#ifdef RTE_LIBRTE_DPAA_PMD
+#include <rte_pmd_dpaa.h>
+#endif
 #ifdef RTE_LIBRTE_IXGBE_PMD
 #include <rte_pmd_ixgbe.h>
 #endif
@@ -12620,6 +12623,10 @@ cmd_set_tx_loopback_parsed(
 	if (ret == -ENOTSUP)
 		ret = rte_pmd_bnxt_set_tx_loopback(res->port_id, is_on);
 #endif
+#ifdef RTE_LIBRTE_DPAA_PMD
+	if (ret == -ENOTSUP)
+		ret = rte_pmd_dpaa_set_tx_loopback(res->port_id, is_on);
+#endif
 
 	switch (ret) {
 	case 0:
-- 
2.7.4

[PATCH 17/18] bus/dpaa: add support for static queues

[Hemant Agrawal]

DPAA hardware support two kinds of queues:
1. Pull mode queue - where one needs to regularly pull the packets.
2. Push mode queue - where the hw pushes the packet to queue. These are
   high performance queues, but limitd in number.

This patch add the driver support for push m de queues.

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 64 +++++++++++++++++++++++++++++++
 drivers/bus/dpaa/base/qbman/qman.h        |  4 +-
 drivers/bus/dpaa/include/fsl_qman.h       | 10 +++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  2 +
 4 files changed, 78 insertions(+), 2 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index b2f82a3..42d509d 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -1080,6 +1080,70 @@ u16 qman_affine_channel(int cpu)
 	return affine_channels[cpu];
 }
 
+unsigned int qman_portal_poll_rx(unsigned int poll_limit,
+				 void **bufs,
+				 struct qman_portal *p)
+{
+	const struct qm_dqrr_entry *dq;
+	struct qman_fq *fq;
+	enum qman_cb_dqrr_result res;
+	unsigned int limit = 0;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	struct qm_dqrr_entry *shadow;
+#endif
+	unsigned int rx_number = 0;
+
+	do {
+		qm_dqrr_pvb_update(&p->p);
+		dq = qm_dqrr_current(&p->p);
+		if (unlikely(!dq))
+			break;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	/* If running on an LE system the fields of the
+	 * dequeue entry must be swapper.  Because the
+	 * QMan HW will ignore writes the DQRR entry is
+	 * copied and the index stored within the copy
+	 */
+		shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)];
+		*shadow = *dq;
+		dq = shadow;
+		shadow->fqid = be32_to_cpu(shadow->fqid);
+		shadow->contextB = be32_to_cpu(shadow->contextB);
+		shadow->seqnum = be16_to_cpu(shadow->seqnum);
+		hw_fd_to_cpu(&shadow->fd);
+#endif
+
+		/* SDQCR: context_b points to the FQ */
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+		fq = get_fq_table_entry(dq->contextB);
+#else
+		fq = (void *)(uintptr_t)dq->contextB;
+#endif
+		/* Now let the callback do its stuff */
+		res = fq->cb.dqrr_dpdk_cb(NULL, p, fq, dq, &bufs[rx_number]);
+		rx_number++;
+		/* Interpret 'dq' from a driver perspective. */
+		/*
+		 * Parking isn't possible unless HELDACTIVE was set. NB,
+		 * FORCEELIGIBLE implies HELDACTIVE, so we only need to
+		 * check for HELDACTIVE to cover both.
+		 */
+		DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) ||
+			    (res != qman_cb_dqrr_park));
+		qm_dqrr_cdc_consume_1ptr(&p->p, dq, res == qman_cb_dqrr_park);
+		/* Move forward */
+		qm_dqrr_next(&p->p);
+		/*
+		 * Entry processed and consumed, increment our counter.  The
+		 * callback can request that we exit after consuming the
+		 * entry, and we also exit if we reach our processing limit,
+		 * so loop back only if neither of these conditions is met.
+		 */
+	} while (likely(++limit < poll_limit));
+
+	return limit;
+}
+
 struct qm_dqrr_entry *qman_dequeue(struct qman_fq *fq)
 {
 	struct qman_portal *p = get_affine_portal();
diff --git a/drivers/bus/dpaa/base/qbman/qman.h b/drivers/bus/dpaa/base/qbman/qman.h
index 2c0f694..999e429 100644
--- a/drivers/bus/dpaa/base/qbman/qman.h
+++ b/drivers/bus/dpaa/base/qbman/qman.h
@@ -187,7 +187,7 @@ struct qm_eqcr {
 };
 
 struct qm_dqrr {
-	const struct qm_dqrr_entry *ring, *cursor;
+	struct qm_dqrr_entry *ring, *cursor;
 	u8 pi, ci, fill, ithresh, vbit;
 #ifdef RTE_LIBRTE_DPAA_HWDEBUG
 	enum qm_dqrr_dmode dmode;
@@ -460,7 +460,7 @@ static inline u8 DQRR_PTR2IDX(const struct qm_dqrr_entry *e)
 	return ((uintptr_t)e >> 6) & (QM_DQRR_SIZE - 1);
 }
 
-static inline const struct qm_dqrr_entry *DQRR_INC(
+static inline struct qm_dqrr_entry *DQRR_INC(
 						const struct qm_dqrr_entry *e)
 {
 	return DQRR_CARRYCLEAR(e + 1);
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index 9090b63..7ec07ee 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1157,6 +1157,12 @@ typedef enum qman_cb_dqrr_result (*qman_cb_dqrr)(struct qman_portal *qm,
 					struct qman_fq *fq,
 					const struct qm_dqrr_entry *dqrr);
 
+typedef enum qman_cb_dqrr_result (*qman_dpdk_cb_dqrr)(void *event,
+					struct qman_portal *qm,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dqrr,
+					void **bd);
+
 /*
  * This callback type is used when handling ERNs, FQRNs and FQRLs via MR. They
  * are always consumed after the callback returns.
@@ -1215,6 +1221,7 @@ enum qman_fq_state {
  */
 
 struct qman_fq_cb {
+	qman_dpdk_cb_dqrr dqrr_dpdk_cb;	/* for dequeued frames */
 	qman_cb_dqrr dqrr;	/* for dequeued frames */
 	qman_cb_mr ern;		/* for s/w ERNs */
 	qman_cb_mr fqs;		/* frame-queue state changes*/
@@ -1332,6 +1339,9 @@ int qman_get_portal_index(void);
  */
 u16 qman_affine_channel(int cpu);
 
+unsigned int qman_portal_poll_rx(unsigned int poll_limit,
+				 void **bufs, struct qman_portal *q);
+
 /**
  * qman_set_vdq - Issue a volatile dequeue command
  * @fq: Frame Queue on which the volatile dequeue command is issued
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index 212c75f..460cfbf 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -70,9 +70,11 @@ DPDK_18.02 {
 
 	dpaa_svr_family;
 	qman_alloc_cgrid_range;
+	qman_alloc_pool_range;
 	qman_create_cgr;
 	qman_delete_cgr;
 	qman_modify_cgr;
+	qman_portal_poll_rx;
 	qman_query_fq_frm_cnt;
 	qman_release_cgrid_range;
 	rte_dpaa_portal_fq_close;
-- 
2.7.4

[PATCH 18/18] net/dpaa: integrate the support of push mode in PMD

[Hemant Agrawal]

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
---
 doc/guides/nics/dpaa.rst       | 11 ++++++++
 drivers/net/dpaa/dpaa_ethdev.c | 62 ++++++++++++++++++++++++++++++++++++++----
 drivers/net/dpaa/dpaa_ethdev.h |  2 +-
 drivers/net/dpaa/dpaa_rxtx.c   | 34 +++++++++++++++++++++++
 drivers/net/dpaa/dpaa_rxtx.h   |  5 ++++
 5 files changed, 107 insertions(+), 7 deletions(-)

diff --git a/doc/guides/nics/dpaa.rst b/doc/guides/nics/dpaa.rst
index d331c05..7b6a0b1 100644
--- a/doc/guides/nics/dpaa.rst
+++ b/doc/guides/nics/dpaa.rst
@@ -315,6 +315,17 @@ state during application initialization:
   In case the application is configured to use lesser number of queues than
   configured above, it might result in packet loss (because of distribution).
 
+- ``DPAA_NUM_PUSH_QUEUES`` (default 4)
+
+  This defines the number of High performance queues to be used for ethdev Rx.
+  These queues use one private HW portal per queue configured, so they are
+  limited in the system. The first configured ethdev queues will be
+  automatically be assigned from the these high perf PUSH queues. Any queue
+  configuration beyond that will be standard Rx queues. The application can
+  choose to change their number if HW portals are limited.
+  The valid values are from '0' to '4'. The valuse shall be set to '0' if the
+  application want to use eventdev with DPAA device.
+
 
 Driver compilation and testing
 ------------------------------
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index fcba929..7798994 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -73,6 +73,14 @@
 
 /* Keep track of whether QMAN and BMAN have been globally initialized */
 static int is_global_init;
+/* At present we only allow up to 4 push mode queues - as each of this queue
+ * need dedicated portal and we are short of portals.
+ */
+#define DPAA_MAX_PUSH_MODE_QUEUE       4
+
+static int dpaa_push_mode_max_queue = DPAA_MAX_PUSH_MODE_QUEUE;
+static int dpaa_push_queue_idx; /* Queue index which are in push mode*/
+
 
 /* Per FQ Taildrop in frame count */
 static unsigned int td_threshold = CGR_RX_PERFQ_THRESH;
@@ -460,6 +468,9 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
 	struct qman_fq *rxq = &dpaa_intf->rx_queues[queue_idx];
+	struct qm_mcc_initfq opts = {0};
+	u32 flags = 0;
+	int ret;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -495,7 +506,41 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 			    dpaa_intf->name, fd_offset,
 			fman_if_get_fdoff(dpaa_intf->fif));
 	}
-
+	/* checking if push mode only, no error check for now */
+	if (dpaa_push_mode_max_queue > dpaa_push_queue_idx) {
+		dpaa_push_queue_idx++;
+		opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA;
+		opts.fqd.fq_ctrl = QM_FQCTRL_AVOIDBLOCK |
+				   QM_FQCTRL_CTXASTASHING |
+				   QM_FQCTRL_PREFERINCACHE;
+		opts.fqd.context_a.stashing.exclusive = 0;
+		opts.fqd.context_a.stashing.data_cl = DPAA_IF_RX_DATA_STASH;
+		opts.fqd.context_a.stashing.context_cl =
+						DPAA_IF_RX_CONTEXT_STASH;
+		if (dpaa_svr_family != SVR_LS1046A_FAMILY)
+			opts.fqd.context_a.stashing.annotation_cl =
+						DPAA_IF_RX_ANNOTATION_STASH;
+
+		/*Create a channel and associate given queue with the channel*/
+		qman_alloc_pool_range((u32 *)&rxq->ch_id, 1, 1, 0);
+		opts.we_mask = opts.we_mask | QM_INITFQ_WE_DESTWQ;
+		opts.fqd.dest.channel = rxq->ch_id;
+		opts.fqd.dest.wq = DPAA_IF_RX_PRIORITY;
+		flags = QMAN_INITFQ_FLAG_SCHED;
+
+		/* Configure tail drop */
+		if (dpaa_intf->cgr_rx) {
+			opts.we_mask |= QM_INITFQ_WE_CGID;
+			opts.fqd.cgid = dpaa_intf->cgr_rx[queue_idx].cgrid;
+			opts.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+		}
+		ret = qman_init_fq(rxq, flags, &opts);
+		if (ret)
+			DPAA_PMD_ERR("Channel/Queue association failed. fqid %d"
+				     " ret: %d", rxq->fqid, ret);
+		rxq->cb.dqrr_dpdk_cb = dpaa_rx_cb;
+		rxq->is_static = true;
+	}
 	dev->data->rx_queues[queue_idx] = rxq;
 
 	/* configure the CGR size as per the desc size */
@@ -835,11 +880,8 @@ static int dpaa_rx_queue_init(struct qman_fq *fq, struct qman_cgr *cgr_rx,
 			fqid, ret);
 		return ret;
 	}
-
-	opts.we_mask = QM_INITFQ_WE_DESTWQ | QM_INITFQ_WE_FQCTRL |
-		       QM_INITFQ_WE_CONTEXTA;
-
-	opts.fqd.dest.wq = DPAA_IF_RX_PRIORITY;
+	fq->is_static = false;
+	opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA;
 	opts.fqd.fq_ctrl = QM_FQCTRL_AVOIDBLOCK | QM_FQCTRL_CTXASTASHING |
 			   QM_FQCTRL_PREFERINCACHE;
 	opts.fqd.context_a.stashing.exclusive = 0;
@@ -973,6 +1015,14 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 	else
 		num_rx_fqs = DPAA_DEFAULT_NUM_PCD_QUEUES;
 
+	/* Number of High perf push mode queues. */
+	if (getenv("DPAA_NUM_PUSH_QUEUES")) {
+		dpaa_push_mode_max_queue =
+				atoi(getenv("DPAA_NUM_PUSH_QUEUES"));
+		if (dpaa_push_mode_max_queue > DPAA_MAX_PUSH_MODE_QUEUE)
+			dpaa_push_mode_max_queue = DPAA_MAX_PUSH_MODE_QUEUE;
+	}
+
 	/* Each device can not have more than DPAA_PCD_FQID_MULTIPLIER RX
 	 * queues.
 	 */
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 95d745e..c0a8430 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -80,7 +80,7 @@
 #define DPAA_MAX_NUM_PCD_QUEUES		32
 
 #define DPAA_IF_TX_PRIORITY		3
-#define DPAA_IF_RX_PRIORITY		4
+#define DPAA_IF_RX_PRIORITY		0
 #define DPAA_IF_DEBUG_PRIORITY		7
 
 #define DPAA_IF_RX_ANNOTATION_STASH	1
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 2609953..088fbe1 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -420,6 +420,37 @@ dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 	return mbuf;
 }
 
+enum qman_cb_dqrr_result dpaa_rx_cb(void *event __always_unused,
+				    struct qman_portal *qm __always_unused,
+				    struct qman_fq *fq,
+				    const struct qm_dqrr_entry *dqrr,
+				    void **bufs)
+{
+	const struct qm_fd *fd = &dqrr->fd;
+
+	*bufs = dpaa_eth_fd_to_mbuf(fd,
+			((struct dpaa_if *)fq->dpaa_intf)->ifid);
+	return qman_cb_dqrr_consume;
+}
+
+static uint16_t
+dpaa_eth_queue_portal_rx(struct qman_fq *fq,
+			 struct rte_mbuf **bufs,
+			 uint16_t nb_bufs)
+{
+	int ret;
+
+	if (unlikely(fq->qp == NULL)) {
+		ret = rte_dpaa_portal_fq_init((void *)0, fq);
+		if (ret) {
+			DPAA_PMD_ERR("Failure in affining portal %d", ret);
+			return 0;
+		}
+	}
+
+	return qman_portal_poll_rx(nb_bufs, (void **)bufs, fq->qp);
+}
+
 uint16_t dpaa_eth_queue_rx(void *q,
 			   struct rte_mbuf **bufs,
 			   uint16_t nb_bufs)
@@ -429,6 +460,9 @@ uint16_t dpaa_eth_queue_rx(void *q,
 	uint32_t num_rx = 0, ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
 	int ret;
 
+	if (likely(fq->is_static))
+		return dpaa_eth_queue_portal_rx(fq, bufs, nb_bufs);
+
 	ret = rte_dpaa_portal_init((void *)0);
 	if (ret) {
 		DPAA_PMD_ERR("Failure in affining portal");
diff --git a/drivers/net/dpaa/dpaa_rxtx.h b/drivers/net/dpaa/dpaa_rxtx.h
index b434b6d..de65ebc 100644
--- a/drivers/net/dpaa/dpaa_rxtx.h
+++ b/drivers/net/dpaa/dpaa_rxtx.h
@@ -294,4 +294,9 @@ int dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 			   struct qm_fd *fd,
 			   uint32_t bpid);
 
+enum qman_cb_dqrr_result dpaa_rx_cb(void *event,
+				    struct qman_portal *qm,
+				    struct qman_fq *fq,
+				    const struct qm_dqrr_entry *dqrr,
+				    void **bd);
 #endif
-- 
2.7.4

Re: [PATCH 15/18] net/dpaa: add support for loopback API

[Ferruh Yigit]

On 12/13/2017 12:05 PM, Hemant Agrawal wrote:
> Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

<...>

> @@ -0,0 +1,37 @@
> +/*-
> + *   Copyright 2017 NXP.
> + *
> + *   SPDX-License-Identifier:        BSD-3-Clause

I guess latest agreement was without break.

> + */
> +
> +#ifndef _PMD_DPAA_H_
> +#define _PMD_DPAA_H_
> +
> +/**
> + * @file rte_pmd_dpaa.h
> + *
> + * dpaa PMD specific functions.
> + *
> + * @b EXPERIMENTAL: this API may change, or be removed, without prior notice

What about adding a @warning for this note too, otherwise it is very easy to
miss note.

> + *
> + */
> +
> +#include <rte_ethdev.h>
> +
> +/**
> + * Enable/Disable TX loopback

I am for adding EXPERIMENTAL tag for API as well, otherwise it is easy to miss.
I suggest adding @warning as well to highlight is as done in rte_member.h

> + *
> + * @param port
> + *    The port identifier of the Ethernet device.
> + * @param on
> + *    1 - Enable TX loopback.
> + *    0 - Disable TX loopback.
> + * @return
> + *   - (0) if successful.
> + *   - (-ENODEV) if *port* invalid.
> + *   - (-EINVAL) if bad parameter.
> + */
> +int rte_pmd_dpaa_set_tx_loopback(uint8_t port,
> +				 uint8_t on);

PMD now has PMD specific API, this is public API, can you please update related
API documentations to document this API?

> +
> +#endif /* _PMD_DPAA_H_ */
> diff --git a/drivers/net/dpaa/rte_pmd_dpaa_version.map b/drivers/net/dpaa/rte_pmd_dpaa_version.map
> index a70bd19..d76acbd 100644
> --- a/drivers/net/dpaa/rte_pmd_dpaa_version.map
> +++ b/drivers/net/dpaa/rte_pmd_dpaa_version.map
> @@ -2,3 +2,11 @@ DPDK_17.11 {
>  
>  	local: *;
>  };
> +
> +DPDK_18.02 {

This API is EXPERIMENTAL (as far as I can see from above) so having an
experimental tag is better.

How to mark an API as experimental is not documented, we should indeed.

cc'ed Luca if I am missing steps related how to make an API experimental.

> +	global:
> +
> +	rte_pmd_dpaa_set_tx_loopback;
> +
> +	local: *;
> +} DPDK_17.11;
> 

Re: [PATCH 05/18] net/dpaa: set the correct frame size in device MTU

[Ferruh Yigit]

On 12/13/2017 12:05 PM, Hemant Agrawal wrote:
> From: Ashish Jain <ashish.jain@nxp.com>
> 
> Setting correct frame size in dpaa_dev_mtu_set
> api call. Also setting correct max frame size in
> hardware in dev_configure for jumbo frames
> 
> Signed-off-by: Ashish Jain <ashish.jain@nxp.com>
> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

<...>

> @@ -111,19 +111,21 @@ static int
>  dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
>  {
>  	struct dpaa_if *dpaa_intf = dev->data->dev_private;
> +	uint32_t frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN
> +				+ VLAN_TAG_SIZE;
>  
>  	PMD_INIT_FUNC_TRACE();
>  
> -	if (mtu < ETHER_MIN_MTU)
> +	if ((mtu < ETHER_MIN_MTU) || (frame_size > DPAA_MAX_RX_PKT_LEN))

checkpatch complains about extra parentheses:

CHECK:UNNECESSARY_PARENTHESES: Unnecessary parentheses around 'mtu < ETHER_MIN_MTU'
#42: FILE: drivers/net/dpaa/dpaa_ethdev.c:119:
+       if ((mtu < ETHER_MIN_MTU) || (frame_size > DPAA_MAX_RX_PKT_LEN))

CHECK:UNNECESSARY_PARENTHESES: Unnecessary parentheses around 'frame_size >
DPAA_MAX_RX_PKT_LEN'
#42: FILE: drivers/net/dpaa/dpaa_ethdev.c:119:
+       if ((mtu < ETHER_MIN_MTU) || (frame_size > DPAA_MAX_RX_PKT_LEN))

<...>

Re: [PATCH 17/18] bus/dpaa: add support for static queues

[Ferruh Yigit]

On 12/13/2017 12:05 PM, Hemant Agrawal wrote:
> DPAA hardware support two kinds of queues:
> 1. Pull mode queue - where one needs to regularly pull the packets.
> 2. Push mode queue - where the hw pushes the packet to queue. These are
>    high performance queues, but limitd in number.

limited

> 
> This patch add the driver support for push m de queues.

push mode

> 
> Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
> Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

<...>

Re: [PATCH 01/18] net/dpaa: fix coverity reported issues

[Ferruh Yigit]

On 12/13/2017 12:05 PM, Hemant Agrawal wrote:
> Fixes: 05ba55bc2b1a ("net/dpaa: add packet dump for debugging")
> Fixes: 37f9b54bd3cf ("net/dpaa: support Tx and Rx queue setup")
> Cc: stable@dpdk.org>
> Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>

Hi Hemant,

fix coverity issues is not very helpful as commit title, can you please document
what really fixed.

And there is a special format for coverity fixes:

"

    Coverity issue: ......
    Fixes: ............ ("...")
    Cc: stable@dpdk.org [if required]

    Signed-off-by: ....
"

There are samples in git history. It seems this format is not documented and
Marko will help to document it.

Re: [PATCH 00/18] DPAA PMD improvements

[Ferruh Yigit]

On 12/13/2017 12:05 PM, Hemant Agrawal wrote:
> This patch series add various improvement and performance related
> optimizations for DPAA PMD
> 
> Ashish Jain (2):
>   net/dpaa: fix the mbuf packet type if zero
>   net/dpaa: set the correct frame size in device MTU
> 
> Hemant Agrawal (11):
>   net/dpaa: fix coverity reported issues
>   net/dpaa: fix FW version code
>   bus/dpaa: update platform soc value register routines
>   net/dpaa: add frame count based tail drop with CGR
>   bus/dpaa: add support to create dynamic HW portal
>   bus/dpaa: query queue frame count support
>   net/dpaa: add Rx queue count support
>   net/dpaa: add support for loopback API
>   app/testpmd: add support for loopback config for dpaa
>   bus/dpaa: add support for static queues
>   net/dpaa: integrate the support of push mode in PMD
> 
> Nipun Gupta (5):
>   bus/dpaa: optimize the qman HW stashing settings
>   bus/dpaa: optimize the endianness conversions
>   net/dpaa: change Tx HW budget to 7
>   net/dpaa: optimize the Tx burst
>   net/dpaa: optimize Rx path

Hi Hemant,

There is new PMD specific API and it needs to be documented, I put more detailed
comments to the patch.

And since there will be a new version of the set commented on a few minor issues
too.

Thanks,
ferruh

[PATCH v2 00/18] DPAA PMD improvements

[Hemant Agrawal]

This patch series add various improvement and performance related
optimizations for DPAA PMD

v2:
 - fix the spelling of PORTALS
 - Add Akhil's patch wich is required for crypto
 - minor improvement in push mode patch
 
Akhil Goyal (1):
  bus/dpaa: support for enqueue frames of multiple queues

Ashish Jain (2):
  net/dpaa: fix the mbuf packet type if zero
  net/dpaa: set the correct frame size in device MTU

Hemant Agrawal (10):
  net/dpaa: fix FW version code
  bus/dpaa: update platform soc value register routines
  net/dpaa: add frame count based tail drop with CGR
  bus/dpaa: add support to create dynamic HW portal
  bus/dpaa: query queue frame count support
  net/dpaa: add Rx queue count support
  net/dpaa: add support for loopback API
  app/testpmd: add support for loopback config for dpaa
  bus/dpaa: add support for static queues
  net/dpaa: integrate the support of push mode in PMD

Nipun Gupta (5):
  bus/dpaa: optimize the qman HW stashing settings
  bus/dpaa: optimize the endianness conversions
  net/dpaa: change Tx HW budget to 7
  net/dpaa: optimize the Tx burst
  net/dpaa: optimize Rx path

 app/test-pmd/Makefile                     |   4 +
 app/test-pmd/cmdline.c                    |   7 +
 doc/guides/nics/dpaa.rst                  |  11 ++
 drivers/bus/dpaa/base/qbman/qman.c        | 238 ++++++++++++++++++++++++++--
 drivers/bus/dpaa/base/qbman/qman.h        |   4 +-
 drivers/bus/dpaa/base/qbman/qman_driver.c | 153 +++++++++++++++---
 drivers/bus/dpaa/base/qbman/qman_priv.h   |   6 +-
 drivers/bus/dpaa/dpaa_bus.c               |  43 ++++-
 drivers/bus/dpaa/include/fsl_qman.h       |  62 ++++++--
 drivers/bus/dpaa/include/fsl_usd.h        |   4 +
 drivers/bus/dpaa/include/process.h        |  11 +-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  21 +++
 drivers/bus/dpaa/rte_dpaa_bus.h           |  15 ++
 drivers/net/dpaa/Makefile                 |   3 +
 drivers/net/dpaa/dpaa_ethdev.c            | 253 ++++++++++++++++++++++++++----
 drivers/net/dpaa/dpaa_ethdev.h            |  21 ++-
 drivers/net/dpaa/dpaa_rxtx.c              | 161 +++++++++++++------
 drivers/net/dpaa/dpaa_rxtx.h              |   7 +-
 drivers/net/dpaa/rte_pmd_dpaa.h           |  35 +++++
 drivers/net/dpaa/rte_pmd_dpaa_version.map |   8 +
 20 files changed, 917 insertions(+), 150 deletions(-)
 create mode 100644 drivers/net/dpaa/rte_pmd_dpaa.h

-- 
2.7.4

[PATCH v2 01/18] net/dpaa: fix the mbuf packet type if zero

[Hemant Agrawal]

From: Ashish Jain <ashish.jain@nxp.com>

Populate the mbuf field packet_type which is required
for calculating checksum while transmitting frames

Fixes: 8cffdcbe85aa ("net/dpaa: support scattered Rx")
Cc: stable@dpdk.org

Signed-off-by: Ashish Jain <ashish.jain@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index c3a0920..630d7a5 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -32,6 +32,7 @@
 #include <rte_ip.h>
 #include <rte_tcp.h>
 #include <rte_udp.h>
+#include <rte_net.h>
 
 #include "dpaa_ethdev.h"
 #include "dpaa_rxtx.h"
@@ -478,6 +479,15 @@ dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 	fd->opaque_addr = 0;
 
 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
+		if (!mbuf->packet_type) {
+			struct rte_net_hdr_lens hdr_lens;
+
+			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+					| RTE_PTYPE_L4_MASK);
+			mbuf->l2_len = hdr_lens.l2_len;
+			mbuf->l3_len = hdr_lens.l3_len;
+		}
 		if (temp->data_off < DEFAULT_TX_ICEOF
 			+ sizeof(struct dpaa_eth_parse_results_t))
 			temp->data_off = DEFAULT_TX_ICEOF
@@ -585,6 +595,15 @@ tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
 	}
 
 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
+		if (!mbuf->packet_type) {
+			struct rte_net_hdr_lens hdr_lens;
+
+			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+					| RTE_PTYPE_L4_MASK);
+			mbuf->l2_len = hdr_lens.l2_len;
+			mbuf->l3_len = hdr_lens.l3_len;
+		}
 		if (mbuf->data_off < (DEFAULT_TX_ICEOF +
 		    sizeof(struct dpaa_eth_parse_results_t))) {
 			DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
-- 
2.7.4

[PATCH v2 02/18] net/dpaa: fix FW version code

[Hemant Agrawal]

fix the soc id path and missing fclose

Fixes: cf0fab1d2ca5 ("net/dpaa: support firmware version get API")
Cc: stable@dpdk.org

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 14 +++++---------
 drivers/net/dpaa/dpaa_ethdev.h |  2 +-
 2 files changed, 6 insertions(+), 10 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 7b4a6f1..db6574f 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -186,19 +186,15 @@ dpaa_fw_version_get(struct rte_eth_dev *dev __rte_unused,
 		DPAA_PMD_ERR("Unable to open SoC device");
 		return -ENOTSUP; /* Not supported on this infra */
 	}
-
-	ret = fscanf(svr_file, "svr:%x", &svr_ver);
-	if (ret <= 0) {
+	if (fscanf(svr_file, "svr:%x", &svr_ver) <= 0)
 		DPAA_PMD_ERR("Unable to read SoC device");
-		return -ENOTSUP; /* Not supported on this infra */
-	}
 
-	ret = snprintf(fw_version, fw_size,
-		       "svr:%x-fman-v%x",
-		       svr_ver,
-		       fman_ip_rev);
+	fclose(svr_file);
 
+	ret = snprintf(fw_version, fw_size, "SVR:%x-fman-v%x",
+		       svr_ver, fman_ip_rev);
 	ret += 1; /* add the size of '\0' */
+
 	if (fw_size < (uint32_t)ret)
 		return ret;
 	else
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index bd63ee0..254fca2 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -20,7 +20,7 @@
 /* DPAA SoC identifier; If this is not available, it can be concluded
  * that board is non-DPAA. Single slot is currently supported.
  */
-#define DPAA_SOC_ID_FILE		"sys/devices/soc0/soc_id"
+#define DPAA_SOC_ID_FILE		"/sys/devices/soc0/soc_id"
 
 #define DPAA_MBUF_HW_ANNOTATION		64
 #define DPAA_FD_PTA_SIZE		64
-- 
2.7.4

[PATCH v2 03/18] bus/dpaa: update platform soc value register routines

[Hemant Agrawal]

This patch update the logic and expose the soc value
register, so that it can be used by other modules as well.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/dpaa_bus.c               | 12 ++++++++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  8 ++++++++
 drivers/bus/dpaa/rte_dpaa_bus.h           | 11 +++++++++++
 drivers/net/dpaa/dpaa_ethdev.c            |  4 +++-
 drivers/net/dpaa/dpaa_ethdev.h            |  5 -----
 5 files changed, 34 insertions(+), 6 deletions(-)

diff --git a/drivers/bus/dpaa/dpaa_bus.c b/drivers/bus/dpaa/dpaa_bus.c
index 79f4858..a7c05b3 100644
--- a/drivers/bus/dpaa/dpaa_bus.c
+++ b/drivers/bus/dpaa/dpaa_bus.c
@@ -51,6 +51,8 @@ struct netcfg_info *dpaa_netcfg;
 /* define a variable to hold the portal_key, once created.*/
 pthread_key_t dpaa_portal_key;
 
+unsigned int dpaa_svr_family;
+
 RTE_DEFINE_PER_LCORE(bool, _dpaa_io);
 
 static inline void
@@ -417,6 +419,8 @@ rte_dpaa_bus_probe(void)
 	int ret = -1;
 	struct rte_dpaa_device *dev;
 	struct rte_dpaa_driver *drv;
+	FILE *svr_file = NULL;
+	unsigned int svr_ver;
 
 	BUS_INIT_FUNC_TRACE();
 
@@ -436,6 +440,14 @@ rte_dpaa_bus_probe(void)
 			break;
 		}
 	}
+
+	svr_file = fopen(DPAA_SOC_ID_FILE, "r");
+	if (svr_file) {
+		if (fscanf(svr_file, "svr:%x", &svr_ver) > 0)
+			dpaa_svr_family = svr_ver & SVR_MASK;
+		fclose(svr_file);
+	}
+
 	return 0;
 }
 
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index fb9d532..eeeb458 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -64,3 +64,11 @@ DPDK_17.11 {
 
 	local: *;
 };
+
+DPDK_18.02 {
+	global:
+
+	dpaa_svr_family;
+
+	local: *;
+} DPDK_17.11;
diff --git a/drivers/bus/dpaa/rte_dpaa_bus.h b/drivers/bus/dpaa/rte_dpaa_bus.h
index 5758274..d9e8c84 100644
--- a/drivers/bus/dpaa/rte_dpaa_bus.h
+++ b/drivers/bus/dpaa/rte_dpaa_bus.h
@@ -20,6 +20,17 @@
 #define DEV_TO_DPAA_DEVICE(ptr)	\
 		container_of(ptr, struct rte_dpaa_device, device)
 
+/* DPAA SoC identifier; If this is not available, it can be concluded
+ * that board is non-DPAA. Single slot is currently supported.
+ */
+#define DPAA_SOC_ID_FILE	"/sys/devices/soc0/soc_id"
+
+#define SVR_LS1043A_FAMILY	0x87920000
+#define SVR_LS1046A_FAMILY	0x87070000
+#define SVR_MASK		0xffff0000
+
+extern unsigned int dpaa_svr_family;
+
 struct rte_dpaa_device;
 struct rte_dpaa_driver;
 
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index db6574f..24943ef 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -186,7 +186,9 @@ dpaa_fw_version_get(struct rte_eth_dev *dev __rte_unused,
 		DPAA_PMD_ERR("Unable to open SoC device");
 		return -ENOTSUP; /* Not supported on this infra */
 	}
-	if (fscanf(svr_file, "svr:%x", &svr_ver) <= 0)
+	if (fscanf(svr_file, "svr:%x", &svr_ver) > 0)
+		dpaa_svr_family = svr_ver & SVR_MASK;
+	else
 		DPAA_PMD_ERR("Unable to read SoC device");
 
 	fclose(svr_file);
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 254fca2..9c3b42c 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -17,11 +17,6 @@
 #include <of.h>
 #include <netcfg.h>
 
-/* DPAA SoC identifier; If this is not available, it can be concluded
- * that board is non-DPAA. Single slot is currently supported.
- */
-#define DPAA_SOC_ID_FILE		"/sys/devices/soc0/soc_id"
-
 #define DPAA_MBUF_HW_ANNOTATION		64
 #define DPAA_FD_PTA_SIZE		64
 
-- 
2.7.4

[PATCH v2 04/18] net/dpaa: set the correct frame size in device MTU

[Hemant Agrawal]

From: Ashish Jain <ashish.jain@nxp.com>

Setting correct frame size in dpaa_dev_mtu_set
api call. Also setting correct max frame size in
hardware in dev_configure for jumbo frames

Signed-off-by: Ashish Jain <ashish.jain@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 20 +++++++++++++-------
 drivers/net/dpaa/dpaa_ethdev.h |  4 ++++
 2 files changed, 17 insertions(+), 7 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 24943ef..5a2ea4f 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -85,19 +85,21 @@ static int
 dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	uint32_t frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN
+				+ VLAN_TAG_SIZE;
 
 	PMD_INIT_FUNC_TRACE();
 
-	if (mtu < ETHER_MIN_MTU)
+	if (mtu < ETHER_MIN_MTU || frame_size > DPAA_MAX_RX_PKT_LEN)
 		return -EINVAL;
-	if (mtu > ETHER_MAX_LEN)
+	if (frame_size > ETHER_MAX_LEN)
 		dev->data->dev_conf.rxmode.jumbo_frame = 1;
 	else
 		dev->data->dev_conf.rxmode.jumbo_frame = 0;
 
-	dev->data->dev_conf.rxmode.max_rx_pkt_len = mtu;
+	dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
 
-	fman_if_set_maxfrm(dpaa_intf->fif, mtu);
+	fman_if_set_maxfrm(dpaa_intf->fif, frame_size);
 
 	return 0;
 }
@@ -105,15 +107,19 @@ dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 static int
 dpaa_eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 {
+	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+
 	PMD_INIT_FUNC_TRACE();
 
 	if (dev->data->dev_conf.rxmode.jumbo_frame == 1) {
 		if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
-		    DPAA_MAX_RX_PKT_LEN)
-			return dpaa_mtu_set(dev,
+		    DPAA_MAX_RX_PKT_LEN) {
+			fman_if_set_maxfrm(dpaa_intf->fif,
 				dev->data->dev_conf.rxmode.max_rx_pkt_len);
-		else
+			return 0;
+		} else {
 			return -1;
+		}
 	}
 	return 0;
 }
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 9c3b42c..548ccff 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -45,6 +45,10 @@
 /*Maximum number of slots available in TX ring*/
 #define MAX_TX_RING_SLOTS	8
 
+#ifndef VLAN_TAG_SIZE
+#define VLAN_TAG_SIZE   4 /** < Vlan Header Length */
+#endif
+
 /* PCD frame queues */
 #define DPAA_PCD_FQID_START		0x400
 #define DPAA_PCD_FQID_MULTIPLIER	0x100
-- 
2.7.4

[PATCH v2 05/18] net/dpaa: add frame count based tail drop with CGR

[Hemant Agrawal]

Replace the byte based tail queue congestion support
with frame count based congestion groups.
It can easily map to number of RX descriptors for a queue.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  5 ++
 drivers/net/dpaa/dpaa_ethdev.c            | 98 +++++++++++++++++++++++++++----
 drivers/net/dpaa/dpaa_ethdev.h            |  8 +--
 3 files changed, 97 insertions(+), 14 deletions(-)

diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index eeeb458..f412362 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -69,6 +69,11 @@ DPDK_18.02 {
 	global:
 
 	dpaa_svr_family;
+	qman_alloc_cgrid_range;
+	qman_create_cgr;
+	qman_delete_cgr;
+	qman_modify_cgr;
+	qman_release_cgrid_range;
 
 	local: *;
 } DPDK_17.11;
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 5a2ea4f..5d94af5 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -47,6 +47,9 @@
 /* Keep track of whether QMAN and BMAN have been globally initialized */
 static int is_global_init;
 
+/* Per FQ Taildrop in frame count */
+static unsigned int td_threshold = CGR_RX_PERFQ_THRESH;
+
 struct rte_dpaa_xstats_name_off {
 	char name[RTE_ETH_XSTATS_NAME_SIZE];
 	uint32_t offset;
@@ -421,12 +424,13 @@ static void dpaa_eth_multicast_disable(struct rte_eth_dev *dev)
 
 static
 int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
-			    uint16_t nb_desc __rte_unused,
+			    uint16_t nb_desc,
 			    unsigned int socket_id __rte_unused,
 			    const struct rte_eth_rxconf *rx_conf __rte_unused,
 			    struct rte_mempool *mp)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	struct qman_fq *rxq = &dpaa_intf->rx_queues[queue_idx];
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -462,7 +466,23 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 			    dpaa_intf->name, fd_offset,
 			fman_if_get_fdoff(dpaa_intf->fif));
 	}
-	dev->data->rx_queues[queue_idx] = &dpaa_intf->rx_queues[queue_idx];
+
+	dev->data->rx_queues[queue_idx] = rxq;
+
+	/* configure the CGR size as per the desc size */
+	if (dpaa_intf->cgr_rx) {
+		struct qm_mcc_initcgr cgr_opts = {0};
+		int ret;
+
+		/* Enable tail drop with cgr on this queue */
+		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, nb_desc, 0);
+		ret = qman_modify_cgr(dpaa_intf->cgr_rx, 0, &cgr_opts);
+		if (ret) {
+			DPAA_PMD_WARN(
+				"rx taildrop modify fail on fqid %d (ret=%d)",
+				rxq->fqid, ret);
+		}
+	}
 
 	return 0;
 }
@@ -698,11 +718,21 @@ static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 }
 
 /* Initialise an Rx FQ */
-static int dpaa_rx_queue_init(struct qman_fq *fq,
+static int dpaa_rx_queue_init(struct qman_fq *fq, struct qman_cgr *cgr_rx,
 			      uint32_t fqid)
 {
 	struct qm_mcc_initfq opts = {0};
 	int ret;
+	u32 flags = 0;
+	struct qm_mcc_initcgr cgr_opts = {
+		.we_mask = QM_CGR_WE_CS_THRES |
+				QM_CGR_WE_CSTD_EN |
+				QM_CGR_WE_MODE,
+		.cgr = {
+			.cstd_en = QM_CGR_EN,
+			.mode = QMAN_CGR_MODE_FRAME
+		}
+	};
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -732,12 +762,24 @@ static int dpaa_rx_queue_init(struct qman_fq *fq,
 	opts.fqd.context_a.stashing.data_cl = DPAA_IF_RX_DATA_STASH;
 	opts.fqd.context_a.stashing.context_cl = DPAA_IF_RX_CONTEXT_STASH;
 
-	/*Enable tail drop */
-	opts.we_mask = opts.we_mask | QM_INITFQ_WE_TDTHRESH;
-	opts.fqd.fq_ctrl = opts.fqd.fq_ctrl | QM_FQCTRL_TDE;
-	qm_fqd_taildrop_set(&opts.fqd.td, CONG_THRESHOLD_RX_Q, 1);
-
-	ret = qman_init_fq(fq, 0, &opts);
+	if (cgr_rx) {
+		/* Enable tail drop with cgr on this queue */
+		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, td_threshold, 0);
+		cgr_rx->cb = NULL;
+		ret = qman_create_cgr(cgr_rx, QMAN_CGR_FLAG_USE_INIT,
+				      &cgr_opts);
+		if (ret) {
+			DPAA_PMD_WARN(
+				"rx taildrop init fail on rx fqid %d (ret=%d)",
+				fqid, ret);
+			goto without_cgr;
+		}
+		opts.we_mask |= QM_INITFQ_WE_CGID;
+		opts.fqd.cgid = cgr_rx->cgrid;
+		opts.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+	}
+without_cgr:
+	ret = qman_init_fq(fq, flags, &opts);
 	if (ret)
 		DPAA_PMD_ERR("init rx fqid %d failed with ret: %d", fqid, ret);
 	return ret;
@@ -819,6 +861,7 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 	struct fm_eth_port_cfg *cfg;
 	struct fman_if *fman_intf;
 	struct fman_if_bpool *bp, *tmp_bp;
+	uint32_t cgrid[DPAA_MAX_NUM_PCD_QUEUES];
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -855,10 +898,31 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 
 	dpaa_intf->rx_queues = rte_zmalloc(NULL,
 		sizeof(struct qman_fq) * num_rx_fqs, MAX_CACHELINE);
+
+	/* If congestion control is enabled globally*/
+	if (td_threshold) {
+		dpaa_intf->cgr_rx = rte_zmalloc(NULL,
+			sizeof(struct qman_cgr) * num_rx_fqs, MAX_CACHELINE);
+
+		ret = qman_alloc_cgrid_range(&cgrid[0], num_rx_fqs, 1, 0);
+		if (ret != num_rx_fqs) {
+			DPAA_PMD_WARN("insufficient CGRIDs available");
+			return -EINVAL;
+		}
+	} else {
+		dpaa_intf->cgr_rx = NULL;
+	}
+
 	for (loop = 0; loop < num_rx_fqs; loop++) {
 		fqid = DPAA_PCD_FQID_START + dpaa_intf->ifid *
 			DPAA_PCD_FQID_MULTIPLIER + loop;
-		ret = dpaa_rx_queue_init(&dpaa_intf->rx_queues[loop], fqid);
+
+		if (dpaa_intf->cgr_rx)
+			dpaa_intf->cgr_rx[loop].cgrid = cgrid[loop];
+
+		ret = dpaa_rx_queue_init(&dpaa_intf->rx_queues[loop],
+			dpaa_intf->cgr_rx ? &dpaa_intf->cgr_rx[loop] : NULL,
+			fqid);
 		if (ret)
 			return ret;
 		dpaa_intf->rx_queues[loop].dpaa_intf = dpaa_intf;
@@ -913,6 +977,7 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 		DPAA_PMD_ERR("Failed to allocate %d bytes needed to "
 						"store MAC addresses",
 				ETHER_ADDR_LEN * DPAA_MAX_MAC_FILTER);
+		rte_free(dpaa_intf->cgr_rx);
 		rte_free(dpaa_intf->rx_queues);
 		rte_free(dpaa_intf->tx_queues);
 		dpaa_intf->rx_queues = NULL;
@@ -951,6 +1016,7 @@ static int
 dpaa_dev_uninit(struct rte_eth_dev *dev)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	int loop;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -968,6 +1034,18 @@ dpaa_dev_uninit(struct rte_eth_dev *dev)
 	if (dpaa_intf->fc_conf)
 		rte_free(dpaa_intf->fc_conf);
 
+	/* Release RX congestion Groups */
+	if (dpaa_intf->cgr_rx) {
+		for (loop = 0; loop < dpaa_intf->nb_rx_queues; loop++)
+			qman_delete_cgr(&dpaa_intf->cgr_rx[loop]);
+
+		qman_release_cgrid_range(dpaa_intf->cgr_rx[loop].cgrid,
+					 dpaa_intf->nb_rx_queues);
+	}
+
+	rte_free(dpaa_intf->cgr_rx);
+	dpaa_intf->cgr_rx = NULL;
+
 	rte_free(dpaa_intf->rx_queues);
 	dpaa_intf->rx_queues = NULL;
 
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 548ccff..f00a77a 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -34,10 +34,8 @@
 #define DPAA_MIN_RX_BUF_SIZE 512
 #define DPAA_MAX_RX_PKT_LEN  10240
 
-/* RX queue tail drop threshold
- * currently considering 32 KB packets.
- */
-#define CONG_THRESHOLD_RX_Q  (32 * 1024)
+/* RX queue tail drop threshold (CGR Based) in frame count */
+#define CGR_RX_PERFQ_THRESH 256
 
 /*max mac filter for memac(8) including primary mac addr*/
 #define DPAA_MAX_MAC_FILTER (MEMAC_NUM_OF_PADDRS + 1)
@@ -53,6 +51,7 @@
 #define DPAA_PCD_FQID_START		0x400
 #define DPAA_PCD_FQID_MULTIPLIER	0x100
 #define DPAA_DEFAULT_NUM_PCD_QUEUES	1
+#define DPAA_MAX_NUM_PCD_QUEUES		32
 
 #define DPAA_IF_TX_PRIORITY		3
 #define DPAA_IF_RX_PRIORITY		4
@@ -102,6 +101,7 @@ struct dpaa_if {
 	char *name;
 	const struct fm_eth_port_cfg *cfg;
 	struct qman_fq *rx_queues;
+	struct qman_cgr *cgr_rx;
 	struct qman_fq *tx_queues;
 	struct qman_fq debug_queues[2];
 	uint16_t nb_rx_queues;
-- 
2.7.4

[PATCH v2 06/18] bus/dpaa: optimize the qman HW stashing settings

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

The settings are tuned for performance.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index a53459f..49bc317 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -7,6 +7,7 @@
 
 #include "qman.h"
 #include <rte_branch_prediction.h>
+#include <rte_dpaa_bus.h>
 
 /* Compilation constants */
 #define DQRR_MAXFILL	15
@@ -503,7 +504,12 @@ struct qman_portal *qman_create_portal(
 
 	p = &portal->p;
 
-	portal->use_eqcr_ci_stashing = ((qman_ip_rev >= QMAN_REV30) ? 1 : 0);
+	if (dpaa_svr_family == SVR_LS1043A_FAMILY)
+		portal->use_eqcr_ci_stashing = 3;
+	else
+		portal->use_eqcr_ci_stashing =
+					((qman_ip_rev >= QMAN_REV30) ? 1 : 0);
+
 	/*
 	 * prep the low-level portal struct with the mapped addresses from the
 	 * config, everything that follows depends on it and "config" is more
@@ -516,7 +522,7 @@ struct qman_portal *qman_create_portal(
 	 * and stash with high-than-DQRR priority.
 	 */
 	if (qm_eqcr_init(p, qm_eqcr_pvb,
-			 portal->use_eqcr_ci_stashing ? 3 : 0, 1)) {
+			 portal->use_eqcr_ci_stashing, 1)) {
 		pr_err("Qman EQCR initialisation failed\n");
 		goto fail_eqcr;
 	}
-- 
2.7.4

[PATCH v2 07/18] bus/dpaa: optimize the endianness conversions

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c  | 7 ++++---
 drivers/bus/dpaa/include/fsl_qman.h | 2 ++
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 49bc317..b6fd40b 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -906,7 +906,7 @@ static inline unsigned int __poll_portal_fast(struct qman_portal *p,
 	do {
 		qm_dqrr_pvb_update(&p->p);
 		dq = qm_dqrr_current(&p->p);
-		if (!dq)
+		if (unlikely(!dq))
 			break;
 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 	/* If running on an LE system the fields of the
@@ -1165,6 +1165,7 @@ int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq)
 	}
 	spin_lock_init(&fq->fqlock);
 	fq->fqid = fqid;
+	fq->fqid_le = cpu_to_be32(fqid);
 	fq->flags = flags;
 	fq->state = qman_fq_state_oos;
 	fq->cgr_groupid = 0;
@@ -1953,7 +1954,7 @@ int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags)
 
 int qman_enqueue_multi(struct qman_fq *fq,
 		       const struct qm_fd *fd,
-		int frames_to_send)
+		       int frames_to_send)
 {
 	struct qman_portal *p = get_affine_portal();
 	struct qm_portal *portal = &p->p;
@@ -1975,7 +1976,7 @@ int qman_enqueue_multi(struct qman_fq *fq,
 
 	/* try to send as many frames as possible */
 	while (eqcr->available && frames_to_send--) {
-		eq->fqid = cpu_to_be32(fq->fqid);
+		eq->fqid = fq->fqid_le;
 #ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
 		eq->tag = cpu_to_be32(fq->key);
 #else
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index 5830ad5..5027230 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1197,6 +1197,8 @@ struct qman_fq {
 	 */
 	spinlock_t fqlock;
 	u32 fqid;
+	u32 fqid_le;
+
 	/* DPDK Interface */
 	void *dpaa_intf;
 
-- 
2.7.4

[PATCH v2 08/18] bus/dpaa: add support to create dynamic HW portal

[Hemant Agrawal]

HW portal is a processing context in DPAA. This patch allow
creation of a queue specific HW portal context.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        |  69 ++++++++++++--
 drivers/bus/dpaa/base/qbman/qman_driver.c | 153 +++++++++++++++++++++++++-----
 drivers/bus/dpaa/base/qbman/qman_priv.h   |   6 +-
 drivers/bus/dpaa/dpaa_bus.c               |  31 +++++-
 drivers/bus/dpaa/include/fsl_qman.h       |  25 ++---
 drivers/bus/dpaa/include/fsl_usd.h        |   4 +
 drivers/bus/dpaa/include/process.h        |  11 ++-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |   2 +
 drivers/bus/dpaa/rte_dpaa_bus.h           |   4 +
 9 files changed, 252 insertions(+), 53 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index b6fd40b..d8fb25a 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -621,11 +621,52 @@ struct qman_portal *qman_create_portal(
 	return NULL;
 }
 
+#define MAX_GLOBAL_PORTALS 8
+static struct qman_portal global_portals[MAX_GLOBAL_PORTALS];
+static int global_portals_used[MAX_GLOBAL_PORTALS];
+
+static struct qman_portal *
+qman_alloc_global_portal(void)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_GLOBAL_PORTALS; i++) {
+		if (global_portals_used[i] == 0) {
+			global_portals_used[i] = 1;
+			return &global_portals[i];
+		}
+	}
+	pr_err("No portal available (%x)\n", MAX_GLOBAL_PORTALS);
+
+	return NULL;
+}
+
+static int
+qman_free_global_portal(struct qman_portal *portal)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_GLOBAL_PORTALS; i++) {
+		if (&global_portals[i] == portal) {
+			global_portals_used[i] = 0;
+			return 0;
+		}
+	}
+	return -1;
+}
+
 struct qman_portal *qman_create_affine_portal(const struct qm_portal_config *c,
-					      const struct qman_cgrs *cgrs)
+					      const struct qman_cgrs *cgrs,
+					      int alloc)
 {
 	struct qman_portal *res;
-	struct qman_portal *portal = get_affine_portal();
+	struct qman_portal *portal;
+
+	if (alloc)
+		portal = qman_alloc_global_portal();
+	else
+		portal = get_affine_portal();
+
 	/* A criteria for calling this function (from qman_driver.c) is that
 	 * we're already affine to the cpu and won't schedule onto another cpu.
 	 */
@@ -675,13 +716,18 @@ void qman_destroy_portal(struct qman_portal *qm)
 	spin_lock_destroy(&qm->cgr_lock);
 }
 
-const struct qm_portal_config *qman_destroy_affine_portal(void)
+const struct qm_portal_config *
+qman_destroy_affine_portal(struct qman_portal *qp)
 {
 	/* We don't want to redirect if we're a slave, use "raw" */
-	struct qman_portal *qm = get_affine_portal();
+	struct qman_portal *qm;
 	const struct qm_portal_config *pcfg;
 	int cpu;
 
+	if (qp == NULL)
+		qm = get_affine_portal();
+	else
+		qm = qp;
 	pcfg = qm->config;
 	cpu = pcfg->cpu;
 
@@ -690,6 +736,9 @@ const struct qm_portal_config *qman_destroy_affine_portal(void)
 	spin_lock(&affine_mask_lock);
 	CPU_CLR(cpu, &affine_mask);
 	spin_unlock(&affine_mask_lock);
+
+	qman_free_global_portal(qm);
+
 	return pcfg;
 }
 
@@ -1096,27 +1145,27 @@ void qman_start_dequeues(void)
 		qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL);
 }
 
-void qman_static_dequeue_add(u32 pools)
+void qman_static_dequeue_add(u32 pools, struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 
 	pools &= p->config->pools;
 	p->sdqcr |= pools;
 	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
 }
 
-void qman_static_dequeue_del(u32 pools)
+void qman_static_dequeue_del(u32 pools, struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 
 	pools &= p->config->pools;
 	p->sdqcr &= ~pools;
 	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
 }
 
-u32 qman_static_dequeue_get(void)
+u32 qman_static_dequeue_get(struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 	return p->sdqcr;
 }
 
diff --git a/drivers/bus/dpaa/base/qbman/qman_driver.c b/drivers/bus/dpaa/base/qbman/qman_driver.c
index c17d15f..7cfa8ee 100644
--- a/drivers/bus/dpaa/base/qbman/qman_driver.c
+++ b/drivers/bus/dpaa/base/qbman/qman_driver.c
@@ -24,8 +24,8 @@ void *qman_ccsr_map;
 /* The qman clock frequency */
 u32 qman_clk;
 
-static __thread int fd = -1;
-static __thread struct qm_portal_config pcfg;
+static __thread int qmfd = -1;
+static __thread struct qm_portal_config qpcfg;
 static __thread struct dpaa_ioctl_portal_map map = {
 	.type = dpaa_portal_qman
 };
@@ -44,16 +44,16 @@ static int fsl_qman_portal_init(uint32_t index, int is_shared)
 		error(0, ret, "pthread_getaffinity_np()");
 		return ret;
 	}
-	pcfg.cpu = -1;
+	qpcfg.cpu = -1;
 	for (loop = 0; loop < CPU_SETSIZE; loop++)
 		if (CPU_ISSET(loop, &cpuset)) {
-			if (pcfg.cpu != -1) {
+			if (qpcfg.cpu != -1) {
 				pr_err("Thread is not affine to 1 cpu\n");
 				return -EINVAL;
 			}
-			pcfg.cpu = loop;
+			qpcfg.cpu = loop;
 		}
-	if (pcfg.cpu == -1) {
+	if (qpcfg.cpu == -1) {
 		pr_err("Bug in getaffinity handling!\n");
 		return -EINVAL;
 	}
@@ -65,36 +65,36 @@ static int fsl_qman_portal_init(uint32_t index, int is_shared)
 		error(0, ret, "process_portal_map()");
 		return ret;
 	}
-	pcfg.channel = map.channel;
-	pcfg.pools = map.pools;
-	pcfg.index = map.index;
+	qpcfg.channel = map.channel;
+	qpcfg.pools = map.pools;
+	qpcfg.index = map.index;
 
 	/* Make the portal's cache-[enabled|inhibited] regions */
-	pcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena;
-	pcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh;
+	qpcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena;
+	qpcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh;
 
-	fd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
-	if (fd == -1) {
+	qmfd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
+	if (qmfd == -1) {
 		pr_err("QMan irq init failed\n");
 		process_portal_unmap(&map.addr);
 		return -EBUSY;
 	}
 
-	pcfg.is_shared = is_shared;
-	pcfg.node = NULL;
-	pcfg.irq = fd;
+	qpcfg.is_shared = is_shared;
+	qpcfg.node = NULL;
+	qpcfg.irq = qmfd;
 
-	portal = qman_create_affine_portal(&pcfg, NULL);
+	portal = qman_create_affine_portal(&qpcfg, NULL, 0);
 	if (!portal) {
 		pr_err("Qman portal initialisation failed (%d)\n",
-		       pcfg.cpu);
+		       qpcfg.cpu);
 		process_portal_unmap(&map.addr);
 		return -EBUSY;
 	}
 
 	irq_map.type = dpaa_portal_qman;
 	irq_map.portal_cinh = map.addr.cinh;
-	process_portal_irq_map(fd, &irq_map);
+	process_portal_irq_map(qmfd, &irq_map);
 	return 0;
 }
 
@@ -103,10 +103,10 @@ static int fsl_qman_portal_finish(void)
 	__maybe_unused const struct qm_portal_config *cfg;
 	int ret;
 
-	process_portal_irq_unmap(fd);
+	process_portal_irq_unmap(qmfd);
 
-	cfg = qman_destroy_affine_portal();
-	DPAA_BUG_ON(cfg != &pcfg);
+	cfg = qman_destroy_affine_portal(NULL);
+	DPAA_BUG_ON(cfg != &qpcfg);
 	ret = process_portal_unmap(&map.addr);
 	if (ret)
 		error(0, ret, "process_portal_unmap()");
@@ -128,14 +128,119 @@ int qman_thread_finish(void)
 
 void qman_thread_irq(void)
 {
-	qbman_invoke_irq(pcfg.irq);
+	qbman_invoke_irq(qpcfg.irq);
 
 	/* Now we need to uninhibit interrupts. This is the only code outside
 	 * the regular portal driver that manipulates any portal register, so
 	 * rather than breaking that encapsulation I am simply hard-coding the
 	 * offset to the inhibit register here.
 	 */
-	out_be32(pcfg.addr_virt[DPAA_PORTAL_CI] + 0xe0c, 0);
+	out_be32(qpcfg.addr_virt[DPAA_PORTAL_CI] + 0xe0c, 0);
+}
+
+struct qman_portal *fsl_qman_portal_create(void)
+{
+	cpu_set_t cpuset;
+	struct qman_portal *res;
+
+	struct qm_portal_config *q_pcfg;
+	int loop, ret;
+	struct dpaa_ioctl_irq_map irq_map;
+	struct dpaa_ioctl_portal_map q_map = {0};
+	int q_fd;
+
+	q_pcfg = kzalloc((sizeof(struct qm_portal_config)), 0);
+	if (!q_pcfg) {
+		error(0, -1, "q_pcfg kzalloc failed");
+		return NULL;
+	}
+
+	/* Verify the thread's cpu-affinity */
+	ret = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t),
+				     &cpuset);
+	if (ret) {
+		error(0, ret, "pthread_getaffinity_np()");
+		return NULL;
+	}
+
+	q_pcfg->cpu = -1;
+	for (loop = 0; loop < CPU_SETSIZE; loop++)
+		if (CPU_ISSET(loop, &cpuset)) {
+			if (q_pcfg->cpu != -1) {
+				pr_err("Thread is not affine to 1 cpu\n");
+				return NULL;
+			}
+			q_pcfg->cpu = loop;
+		}
+	if (q_pcfg->cpu == -1) {
+		pr_err("Bug in getaffinity handling!\n");
+		return NULL;
+	}
+
+	/* Allocate and map a qman portal */
+	q_map.type = dpaa_portal_qman;
+	q_map.index = QBMAN_ANY_PORTAL_IDX;
+	ret = process_portal_map(&q_map);
+	if (ret) {
+		error(0, ret, "process_portal_map()");
+		return NULL;
+	}
+	q_pcfg->channel = q_map.channel;
+	q_pcfg->pools = q_map.pools;
+	q_pcfg->index = q_map.index;
+
+	/* Make the portal's cache-[enabled|inhibited] regions */
+	q_pcfg->addr_virt[DPAA_PORTAL_CE] = q_map.addr.cena;
+	q_pcfg->addr_virt[DPAA_PORTAL_CI] = q_map.addr.cinh;
+
+	q_fd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
+	if (q_fd == -1) {
+		pr_err("QMan irq init failed\n");
+		goto err1;
+	}
+
+	q_pcfg->irq = q_fd;
+
+	res = qman_create_affine_portal(q_pcfg, NULL, true);
+	if (!res) {
+		pr_err("Qman portal initialisation failed (%d)\n",
+		       q_pcfg->cpu);
+		goto err2;
+	}
+
+	irq_map.type = dpaa_portal_qman;
+	irq_map.portal_cinh = q_map.addr.cinh;
+	process_portal_irq_map(q_fd, &irq_map);
+
+	return res;
+err2:
+	close(q_fd);
+err1:
+	process_portal_unmap(&q_map.addr);
+	return NULL;
+}
+
+int fsl_qman_portal_destroy(struct qman_portal *qp)
+{
+	const struct qm_portal_config *cfg;
+	struct dpaa_portal_map addr;
+	int ret;
+
+	cfg = qman_destroy_affine_portal(qp);
+	kfree(qp);
+
+	process_portal_irq_unmap(cfg->irq);
+
+	addr.cena = cfg->addr_virt[DPAA_PORTAL_CE];
+	addr.cinh = cfg->addr_virt[DPAA_PORTAL_CI];
+
+	ret = process_portal_unmap(&addr);
+	if (ret)
+		pr_err("process_portal_unmap() (%d)\n", ret);
+
+	kfree((void *)cfg);
+
+	return ret;
 }
 
 int qman_global_init(void)
diff --git a/drivers/bus/dpaa/base/qbman/qman_priv.h b/drivers/bus/dpaa/base/qbman/qman_priv.h
index db0b310..9e4471e 100644
--- a/drivers/bus/dpaa/base/qbman/qman_priv.h
+++ b/drivers/bus/dpaa/base/qbman/qman_priv.h
@@ -146,8 +146,10 @@ int qm_get_wpm(int *wpm);
 
 struct qman_portal *qman_create_affine_portal(
 			const struct qm_portal_config *config,
-			const struct qman_cgrs *cgrs);
-const struct qm_portal_config *qman_destroy_affine_portal(void);
+			const struct qman_cgrs *cgrs,
+			int alloc);
+const struct qm_portal_config *
+qman_destroy_affine_portal(struct qman_portal *q);
 
 struct qm_portal_config *qm_get_unused_portal(void);
 struct qm_portal_config *qm_get_unused_portal_idx(uint32_t idx);
diff --git a/drivers/bus/dpaa/dpaa_bus.c b/drivers/bus/dpaa/dpaa_bus.c
index a7c05b3..329a125 100644
--- a/drivers/bus/dpaa/dpaa_bus.c
+++ b/drivers/bus/dpaa/dpaa_bus.c
@@ -264,8 +264,7 @@ _dpaa_portal_init(void *arg)
  * rte_dpaa_portal_init - Wrapper over _dpaa_portal_init with thread level check
  * XXX Complete this
  */
-int
-rte_dpaa_portal_init(void *arg)
+int rte_dpaa_portal_init(void *arg)
 {
 	if (unlikely(!RTE_PER_LCORE(_dpaa_io)))
 		return _dpaa_portal_init(arg);
@@ -273,6 +272,34 @@ rte_dpaa_portal_init(void *arg)
 	return 0;
 }
 
+int
+rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq)
+{
+	/* Affine above created portal with channel*/
+	u32 sdqcr;
+	struct qman_portal *qp;
+
+	if (unlikely(!RTE_PER_LCORE(_dpaa_io)))
+		_dpaa_portal_init(arg);
+
+	/* Initialise qman specific portals */
+	qp = fsl_qman_portal_create();
+	if (!qp) {
+		DPAA_BUS_LOG(ERR, "Unable to alloc fq portal");
+		return -1;
+	}
+	fq->qp = qp;
+	sdqcr = QM_SDQCR_CHANNELS_POOL_CONV(fq->ch_id);
+	qman_static_dequeue_add(sdqcr, qp);
+
+	return 0;
+}
+
+int rte_dpaa_portal_fq_close(struct qman_fq *fq)
+{
+	return fsl_qman_portal_destroy(fq->qp);
+}
+
 void
 dpaa_portal_finish(void *arg)
 {
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index 5027230..fc00d8d 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1190,21 +1190,24 @@ struct qman_fq_cb {
 struct qman_fq {
 	/* Caller of qman_create_fq() provides these demux callbacks */
 	struct qman_fq_cb cb;
-	/*
-	 * These are internal to the driver, don't touch. In particular, they
-	 * may change, be removed, or extended (so you shouldn't rely on
-	 * sizeof(qman_fq) being a constant).
-	 */
-	spinlock_t fqlock;
-	u32 fqid;
+
 	u32 fqid_le;
+	u16 ch_id;
+	u8 cgr_groupid;
+	u8 is_static;
 
 	/* DPDK Interface */
 	void *dpaa_intf;
 
+	/* affined portal in case of static queue */
+	struct qman_portal *qp;
+
 	volatile unsigned long flags;
+
 	enum qman_fq_state state;
-	int cgr_groupid;
+	u32 fqid;
+	spinlock_t fqlock;
+
 	struct rb_node node;
 #ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
 	u32 key;
@@ -1383,7 +1386,7 @@ void qman_start_dequeues(void);
  * (SDQCR). The requested pools are limited to those the portal has dequeue
  * access to.
  */
-void qman_static_dequeue_add(u32 pools);
+void qman_static_dequeue_add(u32 pools, struct qman_portal *qm);
 
 /**
  * qman_static_dequeue_del - Remove pool channels from the portal SDQCR
@@ -1393,7 +1396,7 @@ void qman_static_dequeue_add(u32 pools);
  * register (SDQCR). The requested pools are limited to those the portal has
  * dequeue access to.
  */
-void qman_static_dequeue_del(u32 pools);
+void qman_static_dequeue_del(u32 pools, struct qman_portal *qp);
 
 /**
  * qman_static_dequeue_get - return the portal's current SDQCR
@@ -1402,7 +1405,7 @@ void qman_static_dequeue_del(u32 pools);
  * entire register is returned, so if only the currently-enabled pool channels
  * are desired, mask the return value with QM_SDQCR_CHANNELS_POOL_MASK.
  */
-u32 qman_static_dequeue_get(void);
+u32 qman_static_dequeue_get(struct qman_portal *qp);
 
 /**
  * qman_dca - Perform a Discrete Consumption Acknowledgment
diff --git a/drivers/bus/dpaa/include/fsl_usd.h b/drivers/bus/dpaa/include/fsl_usd.h
index ada92f2..e183617 100644
--- a/drivers/bus/dpaa/include/fsl_usd.h
+++ b/drivers/bus/dpaa/include/fsl_usd.h
@@ -67,6 +67,10 @@ void bman_thread_irq(void);
 int qman_global_init(void);
 int bman_global_init(void);
 
+/* Direct portal create and destroy */
+struct qman_portal *fsl_qman_portal_create(void);
+int fsl_qman_portal_destroy(struct qman_portal *qp);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/drivers/bus/dpaa/include/process.h b/drivers/bus/dpaa/include/process.h
index 079849b..d9ec94e 100644
--- a/drivers/bus/dpaa/include/process.h
+++ b/drivers/bus/dpaa/include/process.h
@@ -39,6 +39,11 @@ enum dpaa_portal_type {
 	dpaa_portal_bman,
 };
 
+struct dpaa_portal_map {
+	void *cinh;
+	void *cena;
+};
+
 struct dpaa_ioctl_portal_map {
 	/* Input parameter, is a qman or bman portal required. */
 	enum dpaa_portal_type type;
@@ -50,10 +55,8 @@ struct dpaa_ioctl_portal_map {
 	/* Return value if the map succeeds, this gives the mapped
 	 * cache-inhibited (cinh) and cache-enabled (cena) addresses.
 	 */
-	struct dpaa_portal_map {
-		void *cinh;
-		void *cena;
-	} addr;
+	struct dpaa_portal_map addr;
+
 	/* Qman-specific return values */
 	u16 channel;
 	uint32_t pools;
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index f412362..4e3afda 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -74,6 +74,8 @@ DPDK_18.02 {
 	qman_delete_cgr;
 	qman_modify_cgr;
 	qman_release_cgrid_range;
+	rte_dpaa_portal_fq_close;
+	rte_dpaa_portal_fq_init;
 
 	local: *;
 } DPDK_17.11;
diff --git a/drivers/bus/dpaa/rte_dpaa_bus.h b/drivers/bus/dpaa/rte_dpaa_bus.h
index d9e8c84..f626066 100644
--- a/drivers/bus/dpaa/rte_dpaa_bus.h
+++ b/drivers/bus/dpaa/rte_dpaa_bus.h
@@ -136,6 +136,10 @@ void rte_dpaa_driver_unregister(struct rte_dpaa_driver *driver);
  */
 int rte_dpaa_portal_init(void *arg);
 
+int rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq);
+
+int rte_dpaa_portal_fq_close(struct qman_fq *fq);
+
 /**
  * Cleanup a DPAA Portal
  */
-- 
2.7.4

[PATCH v2 09/18] net/dpaa: change Tx HW budget to 7

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

change the TX budget to 7 to sync best with the hw.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.h | 2 +-
 drivers/net/dpaa/dpaa_rxtx.c   | 5 +++--
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index f00a77a..1b36567 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -41,7 +41,7 @@
 #define DPAA_MAX_MAC_FILTER (MEMAC_NUM_OF_PADDRS + 1)
 
 /*Maximum number of slots available in TX ring*/
-#define MAX_TX_RING_SLOTS	8
+#define DPAA_TX_BURST_SIZE	7
 
 #ifndef VLAN_TAG_SIZE
 #define VLAN_TAG_SIZE   4 /** < Vlan Header Length */
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 630d7a5..565ca50 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -669,7 +669,7 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	struct rte_mbuf *mbuf, *mi = NULL;
 	struct rte_mempool *mp;
 	struct dpaa_bp_info *bp_info;
-	struct qm_fd fd_arr[MAX_TX_RING_SLOTS];
+	struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
 	uint32_t frames_to_send, loop, i = 0;
 	uint16_t state;
 	int ret;
@@ -683,7 +683,8 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	DPAA_DP_LOG(DEBUG, "Transmitting %d buffers on queue: %p", nb_bufs, q);
 
 	while (nb_bufs) {
-		frames_to_send = (nb_bufs >> 3) ? MAX_TX_RING_SLOTS : nb_bufs;
+		frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
+				DPAA_TX_BURST_SIZE : nb_bufs;
 		for (loop = 0; loop < frames_to_send; loop++, i++) {
 			mbuf = bufs[i];
 			if (RTE_MBUF_DIRECT(mbuf)) {
-- 
2.7.4

[PATCH v2 10/18] net/dpaa: optimize the Tx burst

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Optimize it for best case. Create a function
for TX offloads to be used in multiple legs.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 73 ++++++++++++++++++++++++++++----------------
 1 file changed, 46 insertions(+), 27 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 565ca50..148f265 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -272,6 +272,30 @@ static inline void dpaa_checksum_offload(struct rte_mbuf *mbuf,
 	fd->cmd = DPAA_FD_CMD_RPD | DPAA_FD_CMD_DTC;
 }
 
+static inline void
+dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
+{
+	if (!mbuf->packet_type) {
+		struct rte_net_hdr_lens hdr_lens;
+
+		mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+				RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+				| RTE_PTYPE_L4_MASK);
+		mbuf->l2_len = hdr_lens.l2_len;
+		mbuf->l3_len = hdr_lens.l3_len;
+	}
+	if (mbuf->data_off < (DEFAULT_TX_ICEOF +
+	    sizeof(struct dpaa_eth_parse_results_t))) {
+		DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
+			"Not enough Headroom "
+			"space for correct Checksum offload."
+			"So Calculating checksum in Software.");
+		dpaa_checksum(mbuf);
+	} else {
+		dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
+	}
+}
+
 struct rte_mbuf *
 dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
 {
@@ -594,27 +618,8 @@ tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
 		rte_pktmbuf_free(mbuf);
 	}
 
-	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
-		if (!mbuf->packet_type) {
-			struct rte_net_hdr_lens hdr_lens;
-
-			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
-					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
-					| RTE_PTYPE_L4_MASK);
-			mbuf->l2_len = hdr_lens.l2_len;
-			mbuf->l3_len = hdr_lens.l3_len;
-		}
-		if (mbuf->data_off < (DEFAULT_TX_ICEOF +
-		    sizeof(struct dpaa_eth_parse_results_t))) {
-			DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
-				"Not enough Headroom "
-				"space for correct Checksum offload."
-				"So Calculating checksum in Software.");
-			dpaa_checksum(mbuf);
-		} else {
-			dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
-		}
-	}
+	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
+		dpaa_unsegmented_checksum(mbuf, fd_arr);
 }
 
 /* Handle all mbufs on dpaa BMAN managed pool */
@@ -670,7 +675,7 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	struct rte_mempool *mp;
 	struct dpaa_bp_info *bp_info;
 	struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
-	uint32_t frames_to_send, loop, i = 0;
+	uint32_t frames_to_send, loop, sent = 0;
 	uint16_t state;
 	int ret;
 
@@ -685,10 +690,23 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	while (nb_bufs) {
 		frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
 				DPAA_TX_BURST_SIZE : nb_bufs;
-		for (loop = 0; loop < frames_to_send; loop++, i++) {
-			mbuf = bufs[i];
-			if (RTE_MBUF_DIRECT(mbuf)) {
+		for (loop = 0; loop < frames_to_send; loop++) {
+			mbuf = *(bufs++);
+			if (likely(RTE_MBUF_DIRECT(mbuf))) {
 				mp = mbuf->pool;
+				bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
+				if (likely(mp->ops_index ==
+						bp_info->dpaa_ops_index &&
+					mbuf->nb_segs == 1 &&
+					rte_mbuf_refcnt_read(mbuf) == 1)) {
+					DPAA_MBUF_TO_CONTIG_FD(mbuf,
+						&fd_arr[loop], bp_info->bpid);
+					if (mbuf->ol_flags &
+						DPAA_TX_CKSUM_OFFLOAD_MASK)
+						dpaa_unsegmented_checksum(mbuf,
+							&fd_arr[loop]);
+					continue;
+				}
 			} else {
 				mi = rte_mbuf_from_indirect(mbuf);
 				mp = mi->pool;
@@ -729,11 +747,12 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 					frames_to_send - loop);
 		}
 		nb_bufs -= frames_to_send;
+		sent += frames_to_send;
 	}
 
-	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", i, q);
+	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", sent, q);
 
-	return i;
+	return sent;
 }
 
 uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
-- 
2.7.4

[PATCH v2 11/18] net/dpaa: optimize Rx path

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 48 ++++++++++++++++++++------------------------
 drivers/net/dpaa/dpaa_rxtx.h |  2 +-
 2 files changed, 23 insertions(+), 27 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 148f265..98671fa 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -97,12 +97,6 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 	DPAA_DP_LOG(DEBUG, " Parsing mbuf: %p with annotations: %p", m, annot);
 
 	switch (prs) {
-	case DPAA_PKT_TYPE_NONE:
-		m->packet_type = 0;
-		break;
-	case DPAA_PKT_TYPE_ETHER:
-		m->packet_type = RTE_PTYPE_L2_ETHER;
-		break;
 	case DPAA_PKT_TYPE_IPV4:
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV4;
@@ -111,6 +105,9 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV6;
 		break;
+	case DPAA_PKT_TYPE_ETHER:
+		m->packet_type = RTE_PTYPE_L2_ETHER;
+		break;
 	case DPAA_PKT_TYPE_IPV4_FRAG:
 	case DPAA_PKT_TYPE_IPV4_FRAG_UDP:
 	case DPAA_PKT_TYPE_IPV4_FRAG_TCP:
@@ -173,6 +170,9 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
 		break;
+	case DPAA_PKT_TYPE_NONE:
+		m->packet_type = 0;
+		break;
 	/* More switch cases can be added */
 	default:
 		dpaa_slow_parsing(m, prs);
@@ -183,12 +183,11 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 					<< DPAA_PKT_L3_LEN_SHIFT;
 
 	/* Set the hash values */
-	m->hash.rss = (uint32_t)(rte_be_to_cpu_64(annot->hash));
-	m->ol_flags = PKT_RX_RSS_HASH;
+	m->hash.rss = (uint32_t)(annot->hash);
 	/* All packets with Bad checksum are dropped by interface (and
 	 * corresponding notification issued to RX error queues).
 	 */
-	m->ol_flags |= PKT_RX_IP_CKSUM_GOOD;
+	m->ol_flags = PKT_RX_RSS_HASH | PKT_RX_IP_CKSUM_GOOD;
 
 	/* Check if Vlan is present */
 	if (prs & DPAA_PARSE_VLAN_MASK)
@@ -297,7 +296,7 @@ dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
 }
 
 struct rte_mbuf *
-dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
+dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 {
 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 	struct rte_mbuf *first_seg, *prev_seg, *cur_seg, *temp;
@@ -355,34 +354,31 @@ dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
 	return first_seg;
 }
 
-static inline struct rte_mbuf *dpaa_eth_fd_to_mbuf(struct qm_fd *fd,
-							uint32_t ifid)
+static inline struct rte_mbuf *
+dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 {
-	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 	struct rte_mbuf *mbuf;
-	void *ptr;
+	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
+	void *ptr = rte_dpaa_mem_ptov(qm_fd_addr(fd));
 	uint8_t format =
 		(fd->opaque & DPAA_FD_FORMAT_MASK) >> DPAA_FD_FORMAT_SHIFT;
-	uint16_t offset =
-		(fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
-	uint32_t length = fd->opaque & DPAA_FD_LENGTH_MASK;
+	uint16_t offset;
+	uint32_t length;
 
 	DPAA_DP_LOG(DEBUG, " FD--->MBUF");
 
 	if (unlikely(format == qm_fd_sg))
 		return dpaa_eth_sg_to_mbuf(fd, ifid);
 
+	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
+
+	offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
+	length = fd->opaque & DPAA_FD_LENGTH_MASK;
+
 	/* Ignoring case when format != qm_fd_contig */
 	dpaa_display_frame(fd);
-	ptr = rte_dpaa_mem_ptov(fd->addr);
-	/* Ignoring case when ptr would be NULL. That is only possible incase
-	 * of a corrupted packet
-	 */
 
 	mbuf = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
-	/* Prefetch the Parse results and packet data to L1 */
-	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
-	rte_prefetch0((void *)((uint8_t *)ptr + offset));
 
 	mbuf->data_off = offset;
 	mbuf->data_len = length;
@@ -462,11 +458,11 @@ static struct rte_mbuf *dpaa_get_dmable_mbuf(struct rte_mbuf *mbuf,
 	if (!dpaa_mbuf)
 		return NULL;
 
-	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + mbuf->data_off, (void *)
+	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + RTE_PKTMBUF_HEADROOM, (void *)
 		((uint8_t *)(mbuf->buf_addr) + mbuf->data_off), mbuf->pkt_len);
 
 	/* Copy only the required fields */
-	dpaa_mbuf->data_off = mbuf->data_off;
+	dpaa_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 	dpaa_mbuf->pkt_len = mbuf->pkt_len;
 	dpaa_mbuf->ol_flags = mbuf->ol_flags;
 	dpaa_mbuf->packet_type = mbuf->packet_type;
diff --git a/drivers/net/dpaa/dpaa_rxtx.h b/drivers/net/dpaa/dpaa_rxtx.h
index 9308b3a..78e804f 100644
--- a/drivers/net/dpaa/dpaa_rxtx.h
+++ b/drivers/net/dpaa/dpaa_rxtx.h
@@ -262,7 +262,7 @@ uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
 			      struct rte_mbuf **bufs __rte_unused,
 			      uint16_t nb_bufs __rte_unused);
 
-struct rte_mbuf *dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid);
+struct rte_mbuf *dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid);
 
 int dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 			   struct qm_fd *fd,
-- 
2.7.4

[PATCH v2 12/18] bus/dpaa: query queue frame count support

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 22 ++++++++++++++++++++++
 drivers/bus/dpaa/include/fsl_qman.h       |  7 +++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  1 +
 3 files changed, 30 insertions(+)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index d8fb25a..ffb008e 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -1722,6 +1722,28 @@ int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np)
 	return 0;
 }
 
+int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq.fqid = cpu_to_be32(fq->fqid);
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP);
+
+	if (mcr->result == QM_MCR_RESULT_OK)
+		*frm_cnt = be24_to_cpu(mcr->queryfq_np.frm_cnt);
+	else if (mcr->result == QM_MCR_RESULT_ERR_FQID)
+		return -ERANGE;
+	else if (mcr->result != QM_MCR_RESULT_OK)
+		return -EIO;
+	return 0;
+}
+
 int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq)
 {
 	struct qm_mc_command *mcc;
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index fc00d8d..d769d50 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1616,6 +1616,13 @@ int qman_query_fq_has_pkts(struct qman_fq *fq);
 int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np);
 
 /**
+ * qman_query_fq_frmcnt - Queries fq frame count
+ * @fq: the frame queue object to be queried
+ * @frm_cnt: number of frames in the queue
+ */
+int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt);
+
+/**
  * qman_query_wq - Queries work queue lengths
  * @query_dedicated: If non-zero, query length of WQs in the channel dedicated
  *		to this software portal. Otherwise, query length of WQs in a
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index 4e3afda..212c75f 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -73,6 +73,7 @@ DPDK_18.02 {
 	qman_create_cgr;
 	qman_delete_cgr;
 	qman_modify_cgr;
+	qman_query_fq_frm_cnt;
 	qman_release_cgrid_range;
 	rte_dpaa_portal_fq_close;
 	rte_dpaa_portal_fq_init;
-- 
2.7.4

[PATCH v2 13/18] net/dpaa: add Rx queue count support

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 5d94af5..de016ab 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -513,6 +513,22 @@ static void dpaa_eth_tx_queue_release(void *txq __rte_unused)
 	PMD_INIT_FUNC_TRACE();
 }
 
+static uint32_t
+dpaa_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	struct qman_fq *rxq = &dpaa_intf->rx_queues[rx_queue_id];
+	u32 frm_cnt = 0;
+
+	PMD_INIT_FUNC_TRACE();
+
+	if (qman_query_fq_frm_cnt(rxq, &frm_cnt) == 0) {
+		RTE_LOG(DEBUG, PMD, "RX frame count for q(%d) is %u\n",
+			rx_queue_id, frm_cnt);
+	}
+	return frm_cnt;
+}
+
 static int dpaa_link_down(struct rte_eth_dev *dev)
 {
 	PMD_INIT_FUNC_TRACE();
@@ -664,6 +680,7 @@ static struct eth_dev_ops dpaa_devops = {
 	.tx_queue_setup		  = dpaa_eth_tx_queue_setup,
 	.rx_queue_release	  = dpaa_eth_rx_queue_release,
 	.tx_queue_release	  = dpaa_eth_tx_queue_release,
+	.rx_queue_count		  = dpaa_dev_rx_queue_count,
 
 	.flow_ctrl_get		  = dpaa_flow_ctrl_get,
 	.flow_ctrl_set		  = dpaa_flow_ctrl_set,
-- 
2.7.4

[PATCH v2 14/18] net/dpaa: add support for loopback API

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/Makefile                 |  3 +++
 drivers/net/dpaa/dpaa_ethdev.c            | 42 +++++++++++++++++++++++++++++++
 drivers/net/dpaa/rte_pmd_dpaa.h           | 35 ++++++++++++++++++++++++++
 drivers/net/dpaa/rte_pmd_dpaa_version.map |  8 ++++++
 4 files changed, 88 insertions(+)
 create mode 100644 drivers/net/dpaa/rte_pmd_dpaa.h

diff --git a/drivers/net/dpaa/Makefile b/drivers/net/dpaa/Makefile
index e5f662f..b1fc5a0 100644
--- a/drivers/net/dpaa/Makefile
+++ b/drivers/net/dpaa/Makefile
@@ -34,4 +34,7 @@ LDLIBS += -lrte_mempool_dpaa
 LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring
 LDLIBS += -lrte_ethdev -lrte_net -lrte_kvargs
 
+# install this header file
+SYMLINK-$(CONFIG_RTE_LIBRTE_DPAA_PMD)-include := rte_pmd_dpaa.h
+
 include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index de016ab..85ccea4 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -38,6 +38,7 @@
 
 #include <dpaa_ethdev.h>
 #include <dpaa_rxtx.h>
+#include <rte_pmd_dpaa.h>
 
 #include <fsl_usd.h>
 #include <fsl_qman.h>
@@ -84,6 +85,8 @@ static const struct rte_dpaa_xstats_name_off dpaa_xstats_strings[] = {
 		offsetof(struct dpaa_if_stats, tund)},
 };
 
+static struct rte_dpaa_driver rte_dpaa_pmd;
+
 static int
 dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 {
@@ -707,6 +710,45 @@ static struct eth_dev_ops dpaa_devops = {
 	.fw_version_get		  = dpaa_fw_version_get,
 };
 
+static bool
+is_device_supported(struct rte_eth_dev *dev, struct rte_dpaa_driver *drv)
+{
+	if (strcmp(dev->device->driver->name,
+		   drv->driver.name))
+		return false;
+
+	return true;
+}
+
+static bool
+is_dpaa_supported(struct rte_eth_dev *dev)
+{
+	return is_device_supported(dev, &rte_dpaa_pmd);
+}
+
+int
+rte_pmd_dpaa_set_tx_loopback(uint8_t port, uint8_t on)
+{
+	struct rte_eth_dev *dev;
+	struct dpaa_if *dpaa_intf;
+
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(port, -ENODEV);
+
+	dev = &rte_eth_devices[port];
+
+	if (!is_dpaa_supported(dev))
+		return -ENOTSUP;
+
+	dpaa_intf = dev->data->dev_private;
+
+	if (on)
+		fman_if_loopback_enable(dpaa_intf->fif);
+	else
+		fman_if_loopback_disable(dpaa_intf->fif);
+
+	return 0;
+}
+
 static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 {
 	struct rte_eth_fc_conf *fc_conf;
diff --git a/drivers/net/dpaa/rte_pmd_dpaa.h b/drivers/net/dpaa/rte_pmd_dpaa.h
new file mode 100644
index 0000000..04955a4
--- /dev/null
+++ b/drivers/net/dpaa/rte_pmd_dpaa.h
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2017 NXP.
+ */
+
+#ifndef _PMD_DPAA_H_
+#define _PMD_DPAA_H_
+
+/**
+ * @file rte_pmd_dpaa.h
+ *
+ * dpaa PMD specific functions.
+ *
+ * @b EXPERIMENTAL: this API may change, or be removed, without prior notice
+ *
+ */
+
+#include <rte_ethdev.h>
+
+/**
+ * Enable/Disable TX loopback
+ *
+ * @param port
+ *    The port identifier of the Ethernet device.
+ * @param on
+ *    1 - Enable TX loopback.
+ *    0 - Disable TX loopback.
+ * @return
+ *   - (0) if successful.
+ *   - (-ENODEV) if *port* invalid.
+ *   - (-EINVAL) if bad parameter.
+ */
+int rte_pmd_dpaa_set_tx_loopback(uint8_t port,
+				 uint8_t on);
+
+#endif /* _PMD_DPAA_H_ */
diff --git a/drivers/net/dpaa/rte_pmd_dpaa_version.map b/drivers/net/dpaa/rte_pmd_dpaa_version.map
index a70bd19..d76acbd 100644
--- a/drivers/net/dpaa/rte_pmd_dpaa_version.map
+++ b/drivers/net/dpaa/rte_pmd_dpaa_version.map
@@ -2,3 +2,11 @@ DPDK_17.11 {
 
 	local: *;
 };
+
+DPDK_18.02 {
+	global:
+
+	rte_pmd_dpaa_set_tx_loopback;
+
+	local: *;
+} DPDK_17.11;
-- 
2.7.4

[PATCH v2 15/18] app/testpmd: add support for loopback config for dpaa

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 app/test-pmd/Makefile  | 4 ++++
 app/test-pmd/cmdline.c | 7 +++++++
 2 files changed, 11 insertions(+)

diff --git a/app/test-pmd/Makefile b/app/test-pmd/Makefile
index 82b3481..34125e5 100644
--- a/app/test-pmd/Makefile
+++ b/app/test-pmd/Makefile
@@ -43,6 +43,10 @@ ifeq ($(CONFIG_RTE_LIBRTE_PMD_BOND),y)
 LDLIBS += -lrte_pmd_bond
 endif
 
+ifeq ($(CONFIG_RTE_LIBRTE_DPAA_PMD),y)
+LDLIBS += -lrte_pmd_dpaa
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
 LDLIBS += -lrte_pmd_ixgbe
 endif
diff --git a/app/test-pmd/cmdline.c b/app/test-pmd/cmdline.c
index f71d963..32096aa 100644
--- a/app/test-pmd/cmdline.c
+++ b/app/test-pmd/cmdline.c
@@ -89,6 +89,9 @@
 #include <rte_eth_bond.h>
 #include <rte_eth_bond_8023ad.h>
 #endif
+#ifdef RTE_LIBRTE_DPAA_PMD
+#include <rte_pmd_dpaa.h>
+#endif
 #ifdef RTE_LIBRTE_IXGBE_PMD
 #include <rte_pmd_ixgbe.h>
 #endif
@@ -12620,6 +12623,10 @@ cmd_set_tx_loopback_parsed(
 	if (ret == -ENOTSUP)
 		ret = rte_pmd_bnxt_set_tx_loopback(res->port_id, is_on);
 #endif
+#ifdef RTE_LIBRTE_DPAA_PMD
+	if (ret == -ENOTSUP)
+		ret = rte_pmd_dpaa_set_tx_loopback(res->port_id, is_on);
+#endif
 
 	switch (ret) {
 	case 0:
-- 
2.7.4

[PATCH v2 16/18] bus/dpaa: add support for static queues

[Hemant Agrawal]

DPAA hardware support two kinds of queues:
1. Pull mode queue - where one needs to regularly pull the packets.
2. Push mode queue - where the hw pushes the packet to queue. These are
   high performance queues, but limitd in number.

This patch add the driver support for push m de queues.

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 64 +++++++++++++++++++++++++++++++
 drivers/bus/dpaa/base/qbman/qman.h        |  4 +-
 drivers/bus/dpaa/include/fsl_qman.h       | 14 ++++++-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  4 ++
 4 files changed, 83 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index ffb008e..7e285a5 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -1051,6 +1051,70 @@ u16 qman_affine_channel(int cpu)
 	return affine_channels[cpu];
 }
 
+unsigned int qman_portal_poll_rx(unsigned int poll_limit,
+				 void **bufs,
+				 struct qman_portal *p)
+{
+	const struct qm_dqrr_entry *dq;
+	struct qman_fq *fq;
+	enum qman_cb_dqrr_result res;
+	unsigned int limit = 0;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	struct qm_dqrr_entry *shadow;
+#endif
+	unsigned int rx_number = 0;
+
+	do {
+		qm_dqrr_pvb_update(&p->p);
+		dq = qm_dqrr_current(&p->p);
+		if (unlikely(!dq))
+			break;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	/* If running on an LE system the fields of the
+	 * dequeue entry must be swapper.  Because the
+	 * QMan HW will ignore writes the DQRR entry is
+	 * copied and the index stored within the copy
+	 */
+		shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)];
+		*shadow = *dq;
+		dq = shadow;
+		shadow->fqid = be32_to_cpu(shadow->fqid);
+		shadow->contextB = be32_to_cpu(shadow->contextB);
+		shadow->seqnum = be16_to_cpu(shadow->seqnum);
+		hw_fd_to_cpu(&shadow->fd);
+#endif
+
+		/* SDQCR: context_b points to the FQ */
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+		fq = get_fq_table_entry(dq->contextB);
+#else
+		fq = (void *)(uintptr_t)dq->contextB;
+#endif
+		/* Now let the callback do its stuff */
+		res = fq->cb.dqrr_dpdk_cb(NULL, p, fq, dq, &bufs[rx_number]);
+		rx_number++;
+		/* Interpret 'dq' from a driver perspective. */
+		/*
+		 * Parking isn't possible unless HELDACTIVE was set. NB,
+		 * FORCEELIGIBLE implies HELDACTIVE, so we only need to
+		 * check for HELDACTIVE to cover both.
+		 */
+		DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) ||
+			    (res != qman_cb_dqrr_park));
+		qm_dqrr_cdc_consume_1ptr(&p->p, dq, res == qman_cb_dqrr_park);
+		/* Move forward */
+		qm_dqrr_next(&p->p);
+		/*
+		 * Entry processed and consumed, increment our counter.  The
+		 * callback can request that we exit after consuming the
+		 * entry, and we also exit if we reach our processing limit,
+		 * so loop back only if neither of these conditions is met.
+		 */
+	} while (likely(++limit < poll_limit));
+
+	return limit;
+}
+
 struct qm_dqrr_entry *qman_dequeue(struct qman_fq *fq)
 {
 	struct qman_portal *p = get_affine_portal();
diff --git a/drivers/bus/dpaa/base/qbman/qman.h b/drivers/bus/dpaa/base/qbman/qman.h
index a433369..4346d86 100644
--- a/drivers/bus/dpaa/base/qbman/qman.h
+++ b/drivers/bus/dpaa/base/qbman/qman.h
@@ -154,7 +154,7 @@ struct qm_eqcr {
 };
 
 struct qm_dqrr {
-	const struct qm_dqrr_entry *ring, *cursor;
+	struct qm_dqrr_entry *ring, *cursor;
 	u8 pi, ci, fill, ithresh, vbit;
 #ifdef RTE_LIBRTE_DPAA_HWDEBUG
 	enum qm_dqrr_dmode dmode;
@@ -441,7 +441,7 @@ static inline u8 DQRR_PTR2IDX(const struct qm_dqrr_entry *e)
 	return ((uintptr_t)e >> 6) & (QM_DQRR_SIZE - 1);
 }
 
-static inline const struct qm_dqrr_entry *DQRR_INC(
+static inline struct qm_dqrr_entry *DQRR_INC(
 						const struct qm_dqrr_entry *e)
 {
 	return DQRR_CARRYCLEAR(e + 1);
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index d769d50..ad40d80 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1124,6 +1124,12 @@ typedef enum qman_cb_dqrr_result (*qman_cb_dqrr)(struct qman_portal *qm,
 					struct qman_fq *fq,
 					const struct qm_dqrr_entry *dqrr);
 
+typedef enum qman_cb_dqrr_result (*qman_dpdk_cb_dqrr)(void *event,
+					struct qman_portal *qm,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dqrr,
+					void **bd);
+
 /*
  * This callback type is used when handling ERNs, FQRNs and FQRLs via MR. They
  * are always consumed after the callback returns.
@@ -1182,7 +1188,10 @@ enum qman_fq_state {
  */
 
 struct qman_fq_cb {
-	qman_cb_dqrr dqrr;	/* for dequeued frames */
+	union { /* for dequeued frames */
+		qman_dpdk_cb_dqrr dqrr_dpdk_cb;
+		qman_cb_dqrr dqrr;
+	};
 	qman_cb_mr ern;		/* for s/w ERNs */
 	qman_cb_mr fqs;		/* frame-queue state changes*/
 };
@@ -1299,6 +1308,9 @@ int qman_get_portal_index(void);
  */
 u16 qman_affine_channel(int cpu);
 
+unsigned int qman_portal_poll_rx(unsigned int poll_limit,
+				 void **bufs, struct qman_portal *q);
+
 /**
  * qman_set_vdq - Issue a volatile dequeue command
  * @fq: Frame Queue on which the volatile dequeue command is issued
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index 212c75f..ac455cd 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -70,11 +70,15 @@ DPDK_18.02 {
 
 	dpaa_svr_family;
 	qman_alloc_cgrid_range;
+	qman_alloc_pool_range;
 	qman_create_cgr;
 	qman_delete_cgr;
 	qman_modify_cgr;
+	qman_oos_fq;
+	qman_portal_poll_rx;
 	qman_query_fq_frm_cnt;
 	qman_release_cgrid_range;
+	qman_retire_fq;
 	rte_dpaa_portal_fq_close;
 	rte_dpaa_portal_fq_init;
 
-- 
2.7.4

[PATCH v2 17/18] net/dpaa: integrate the support of push mode in PMD

[Hemant Agrawal]

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
---
 doc/guides/nics/dpaa.rst       | 11 ++++++++
 drivers/net/dpaa/dpaa_ethdev.c | 64 +++++++++++++++++++++++++++++++++++++-----
 drivers/net/dpaa/dpaa_ethdev.h |  2 +-
 drivers/net/dpaa/dpaa_rxtx.c   | 34 ++++++++++++++++++++++
 drivers/net/dpaa/dpaa_rxtx.h   |  5 ++++
 5 files changed, 108 insertions(+), 8 deletions(-)

diff --git a/doc/guides/nics/dpaa.rst b/doc/guides/nics/dpaa.rst
index a62f128..0a13996 100644
--- a/doc/guides/nics/dpaa.rst
+++ b/doc/guides/nics/dpaa.rst
@@ -290,6 +290,17 @@ state during application initialization:
   In case the application is configured to use lesser number of queues than
   configured above, it might result in packet loss (because of distribution).
 
+- ``DPAA_PUSH_QUEUES_NUMBER`` (default 4)
+
+  This defines the number of High performance queues to be used for ethdev Rx.
+  These queues use one private HW portal per queue configured, so they are
+  limited in the system. The first configured ethdev queues will be
+  automatically be assigned from the these high perf PUSH queues. Any queue
+  configuration beyond that will be standard Rx queues. The application can
+  choose to change their number if HW portals are limited.
+  The valid values are from '0' to '4'. The valuse shall be set to '0' if the
+  application want to use eventdev with DPAA device.
+
 
 Driver compilation and testing
 ------------------------------
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 85ccea4..444c122 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -47,6 +47,14 @@
 
 /* Keep track of whether QMAN and BMAN have been globally initialized */
 static int is_global_init;
+/* At present we only allow up to 4 push mode queues - as each of this queue
+ * need dedicated portal and we are short of portals.
+ */
+#define DPAA_MAX_PUSH_MODE_QUEUE       4
+
+static int dpaa_push_mode_max_queue = DPAA_MAX_PUSH_MODE_QUEUE;
+static int dpaa_push_queue_idx; /* Queue index which are in push mode*/
+
 
 /* Per FQ Taildrop in frame count */
 static unsigned int td_threshold = CGR_RX_PERFQ_THRESH;
@@ -434,6 +442,9 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
 	struct qman_fq *rxq = &dpaa_intf->rx_queues[queue_idx];
+	struct qm_mcc_initfq opts = {0};
+	u32 flags = 0;
+	int ret;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -469,13 +480,45 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 			    dpaa_intf->name, fd_offset,
 			fman_if_get_fdoff(dpaa_intf->fif));
 	}
-
+	/* checking if push mode only, no error check for now */
+	if (dpaa_push_mode_max_queue > dpaa_push_queue_idx) {
+		dpaa_push_queue_idx++;
+		opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA;
+		opts.fqd.fq_ctrl = QM_FQCTRL_AVOIDBLOCK |
+				   QM_FQCTRL_CTXASTASHING |
+				   QM_FQCTRL_PREFERINCACHE;
+		opts.fqd.context_a.stashing.exclusive = 0;
+		opts.fqd.context_a.stashing.annotation_cl =
+						DPAA_IF_RX_ANNOTATION_STASH;
+		opts.fqd.context_a.stashing.data_cl = DPAA_IF_RX_DATA_STASH;
+		opts.fqd.context_a.stashing.context_cl =
+						DPAA_IF_RX_CONTEXT_STASH;
+
+		/*Create a channel and associate given queue with the channel*/
+		qman_alloc_pool_range((u32 *)&rxq->ch_id, 1, 1, 0);
+		opts.we_mask = opts.we_mask | QM_INITFQ_WE_DESTWQ;
+		opts.fqd.dest.channel = rxq->ch_id;
+		opts.fqd.dest.wq = DPAA_IF_RX_PRIORITY;
+		flags = QMAN_INITFQ_FLAG_SCHED;
+
+		/* Configure tail drop */
+		if (dpaa_intf->cgr_rx) {
+			opts.we_mask |= QM_INITFQ_WE_CGID;
+			opts.fqd.cgid = dpaa_intf->cgr_rx[queue_idx].cgrid;
+			opts.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+		}
+		ret = qman_init_fq(rxq, flags, &opts);
+		if (ret)
+			DPAA_PMD_ERR("Channel/Queue association failed. fqid %d"
+				     " ret: %d", rxq->fqid, ret);
+		rxq->cb.dqrr_dpdk_cb = dpaa_rx_cb;
+		rxq->is_static = true;
+	}
 	dev->data->rx_queues[queue_idx] = rxq;
 
 	/* configure the CGR size as per the desc size */
 	if (dpaa_intf->cgr_rx) {
 		struct qm_mcc_initcgr cgr_opts = {0};
-		int ret;
 
 		/* Enable tail drop with cgr on this queue */
 		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, nb_desc, 0);
@@ -809,11 +852,8 @@ static int dpaa_rx_queue_init(struct qman_fq *fq, struct qman_cgr *cgr_rx,
 			fqid, ret);
 		return ret;
 	}
-
-	opts.we_mask = QM_INITFQ_WE_DESTWQ | QM_INITFQ_WE_FQCTRL |
-		       QM_INITFQ_WE_CONTEXTA;
-
-	opts.fqd.dest.wq = DPAA_IF_RX_PRIORITY;
+	fq->is_static = false;
+	opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA;
 	opts.fqd.fq_ctrl = QM_FQCTRL_AVOIDBLOCK | QM_FQCTRL_CTXASTASHING |
 			   QM_FQCTRL_PREFERINCACHE;
 	opts.fqd.context_a.stashing.exclusive = 0;
@@ -947,6 +987,16 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 	else
 		num_rx_fqs = DPAA_DEFAULT_NUM_PCD_QUEUES;
 
+	/* if push mode queues to be enabled. Currenly we are allowing only
+	 * one queue per thread.
+	 */
+	if (getenv("DPAA_PUSH_QUEUES_NUMBER")) {
+		dpaa_push_mode_max_queue =
+				atoi(getenv("DPAA_PUSH_QUEUES_NUMBER"));
+		if (dpaa_push_mode_max_queue > DPAA_MAX_PUSH_MODE_QUEUE)
+			dpaa_push_mode_max_queue = DPAA_MAX_PUSH_MODE_QUEUE;
+	}
+
 	/* Each device can not have more than DPAA_PCD_FQID_MULTIPLIER RX
 	 * queues.
 	 */
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 1b36567..1fa6caf 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -54,7 +54,7 @@
 #define DPAA_MAX_NUM_PCD_QUEUES		32
 
 #define DPAA_IF_TX_PRIORITY		3
-#define DPAA_IF_RX_PRIORITY		4
+#define DPAA_IF_RX_PRIORITY		0
 #define DPAA_IF_DEBUG_PRIORITY		7
 
 #define DPAA_IF_RX_ANNOTATION_STASH	1
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 98671fa..0413932 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -394,6 +394,37 @@ dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 	return mbuf;
 }
 
+enum qman_cb_dqrr_result dpaa_rx_cb(void *event __always_unused,
+				    struct qman_portal *qm __always_unused,
+				    struct qman_fq *fq,
+				    const struct qm_dqrr_entry *dqrr,
+				    void **bufs)
+{
+	const struct qm_fd *fd = &dqrr->fd;
+
+	*bufs = dpaa_eth_fd_to_mbuf(fd,
+			((struct dpaa_if *)fq->dpaa_intf)->ifid);
+	return qman_cb_dqrr_consume;
+}
+
+static uint16_t
+dpaa_eth_queue_portal_rx(struct qman_fq *fq,
+			 struct rte_mbuf **bufs,
+			 uint16_t nb_bufs)
+{
+	int ret;
+
+	if (unlikely(fq->qp == NULL)) {
+		ret = rte_dpaa_portal_fq_init((void *)0, fq);
+		if (ret) {
+			DPAA_PMD_ERR("Failure in affining portal %d", ret);
+			return 0;
+		}
+	}
+
+	return qman_portal_poll_rx(nb_bufs, (void **)bufs, fq->qp);
+}
+
 uint16_t dpaa_eth_queue_rx(void *q,
 			   struct rte_mbuf **bufs,
 			   uint16_t nb_bufs)
@@ -403,6 +434,9 @@ uint16_t dpaa_eth_queue_rx(void *q,
 	uint32_t num_rx = 0, ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
 	int ret;
 
+	if (likely(fq->is_static))
+		return dpaa_eth_queue_portal_rx(fq, bufs, nb_bufs);
+
 	ret = rte_dpaa_portal_init((void *)0);
 	if (ret) {
 		DPAA_PMD_ERR("Failure in affining portal");
diff --git a/drivers/net/dpaa/dpaa_rxtx.h b/drivers/net/dpaa/dpaa_rxtx.h
index 78e804f..29d8f95 100644
--- a/drivers/net/dpaa/dpaa_rxtx.h
+++ b/drivers/net/dpaa/dpaa_rxtx.h
@@ -268,4 +268,9 @@ int dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 			   struct qm_fd *fd,
 			   uint32_t bpid);
 
+enum qman_cb_dqrr_result dpaa_rx_cb(void *event,
+				    struct qman_portal *qm,
+				    struct qman_fq *fq,
+				    const struct qm_dqrr_entry *dqrr,
+				    void **bd);
 #endif
-- 
2.7.4

[PATCH v2 18/18] bus/dpaa: support for enqueue frames of multiple queues

[Hemant Agrawal]

From: Akhil Goyal <akhil.goyal@nxp.com>

Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com>
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 66 +++++++++++++++++++++++++++++++
 drivers/bus/dpaa/include/fsl_qman.h       | 14 +++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  1 +
 3 files changed, 81 insertions(+)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 7e285a5..e171356 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -2158,6 +2158,72 @@ int qman_enqueue_multi(struct qman_fq *fq,
 	return sent;
 }
 
+int
+qman_enqueue_multi_fq(struct qman_fq *fq[], const struct qm_fd *fd,
+		      int frames_to_send)
+{
+	struct qman_portal *p = get_affine_portal();
+	struct qm_portal *portal = &p->p;
+
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	struct qm_eqcr_entry *eq = eqcr->cursor, *prev_eq;
+
+	u8 i, diff, old_ci, sent = 0;
+
+	/* Update the available entries if no entry is free */
+	if (!eqcr->available) {
+		old_ci = eqcr->ci;
+		eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1);
+		diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci);
+		eqcr->available += diff;
+		if (!diff)
+			return 0;
+	}
+
+	/* try to send as many frames as possible */
+	while (eqcr->available && frames_to_send--) {
+		eq->fqid = fq[sent]->fqid_le;
+		eq->fd.opaque_addr = fd->opaque_addr;
+		eq->fd.addr = cpu_to_be40(fd->addr);
+		eq->fd.status = cpu_to_be32(fd->status);
+		eq->fd.opaque = cpu_to_be32(fd->opaque);
+
+		eq = (void *)((unsigned long)(eq + 1) &
+			(~(unsigned long)(QM_EQCR_SIZE << 6)));
+		eqcr->available--;
+		sent++;
+		fd++;
+	}
+	lwsync();
+
+	/* In order for flushes to complete faster, all lines are recorded in
+	 * 32 bit word.
+	 */
+	eq = eqcr->cursor;
+	for (i = 0; i < sent; i++) {
+		eq->__dont_write_directly__verb =
+			QM_EQCR_VERB_CMD_ENQUEUE | eqcr->vbit;
+		prev_eq = eq;
+		eq = (void *)((unsigned long)(eq + 1) &
+			(~(unsigned long)(QM_EQCR_SIZE << 6)));
+		if (unlikely((prev_eq + 1) != eq))
+			eqcr->vbit ^= QM_EQCR_VERB_VBIT;
+	}
+
+	/* We need  to flush all the lines but without load/store operations
+	 * between them
+	 */
+	eq = eqcr->cursor;
+	for (i = 0; i < sent; i++) {
+		dcbf(eq);
+		eq = (void *)((unsigned long)(eq + 1) &
+			(~(unsigned long)(QM_EQCR_SIZE << 6)));
+	}
+	/* Update cursor for the next call */
+	eqcr->cursor = eq;
+	return sent;
+}
+
 int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags,
 		     struct qman_fq *orp, u16 orp_seqnum)
 {
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index ad40d80..0e3e4fe 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1703,6 +1703,20 @@ int qman_enqueue_multi(struct qman_fq *fq,
 		       const struct qm_fd *fd,
 		int frames_to_send);
 
+/**
+ * qman_enqueue_multi_fq - Enqueue multiple frames to their respective frame
+ * queues.
+ * @fq[]: Array of frame queue objects to enqueue to
+ * @fd: pointer to first descriptor of frame to be enqueued
+ * @frames_to_send: number of frames to be sent.
+ *
+ * This API is similar to qman_enqueue_multi(), but it takes fd which needs
+ * to be processed by different frame queues.
+ */
+int
+qman_enqueue_multi_fq(struct qman_fq *fq[], const struct qm_fd *fd,
+		      int frames_to_send);
+
 typedef int (*qman_cb_precommit) (void *arg);
 
 /**
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index ac455cd..64068de 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -73,6 +73,7 @@ DPDK_18.02 {
 	qman_alloc_pool_range;
 	qman_create_cgr;
 	qman_delete_cgr;
+	qman_enqueue_multi_fq;
 	qman_modify_cgr;
 	qman_oos_fq;
 	qman_portal_poll_rx;
-- 
2.7.4

Re: [PATCH 01/18] net/dpaa: fix coverity reported issues

[Hemant Agrawal]

On 1/9/2018 4:16 PM, Ferruh Yigit wrote:
> On 12/13/2017 12:05 PM, Hemant Agrawal wrote:
>> Fixes: 05ba55bc2b1a ("net/dpaa: add packet dump for debugging")
>> Fixes: 37f9b54bd3cf ("net/dpaa: support Tx and Rx queue setup")
>> Cc: stable@dpdk.org>
>> Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
>
> Hi Hemant,
>
> fix coverity issues is not very helpful as commit title, can you please document
> what really fixed.
>
> And there is a special format for coverity fixes:
>
> "
>
>     Coverity issue: ......
>     Fixes: ............ ("...")
>     Cc: stable@dpdk.org [if required]
>
>     Signed-off-by: ....
> "
>
> There are samples in git history. It seems this format is not documented and
> Marko will help to document it.
>
Hi Ferruh,
	thanks for the review. Please ignore v2, I will fix it in v3.

Regards,
Hemant

[PATCH v3 00/19 ]DPAA PMD improvements

[Hemant Agrawal]

This patch series add various improvement and performance related
optimizations for DPAA PMD

v3:
 - handling review comments from Ferruh
 - update the API doc for new PMD specific API

v2:
 - fix the spelling of PORTALS
 - Add Akhil's patch wich is required for crypto
 - minor improvement in push mode patch

Akhil Goyal (1):
  bus/dpaa: support for enqueue frames of multiple queues

Ashish Jain (2):
  net/dpaa: fix the mbuf packet type if zero
  net/dpaa: set the correct frame size in device MTU

Hemant Agrawal (11):
  net/dpaa: fix uninitialized and unused variables
  net/dpaa: fix FW version code
  bus/dpaa: update platform soc value register routines
  net/dpaa: add frame count based tail drop with CGR
  bus/dpaa: add support to create dynamic HW portal
  bus/dpaa: query queue frame count support
  net/dpaa: add Rx queue count support
  net/dpaa: add support for loopback API
  app/testpmd: add support for loopback config for dpaa
  bus/dpaa: add support for static queues
  net/dpaa: integrate the support of push mode in PMD

Nipun Gupta (5):
  bus/dpaa: optimize the qman HW stashing settings
  bus/dpaa: optimize the endianness conversions
  net/dpaa: change Tx HW budget to 7
  net/dpaa: optimize the Tx burst
  net/dpaa: optimize Rx path

 app/test-pmd/Makefile                     |   4 +
 app/test-pmd/cmdline.c                    |   7 +
 doc/api/doxy-api-index.md                 |   1 +
 doc/api/doxy-api.conf                     |   1 +
 doc/guides/nics/dpaa.rst                  |  11 ++
 drivers/bus/dpaa/base/qbman/qman.c        | 238 +++++++++++++++++++++++++--
 drivers/bus/dpaa/base/qbman/qman.h        |   4 +-
 drivers/bus/dpaa/base/qbman/qman_driver.c | 153 +++++++++++++++---
 drivers/bus/dpaa/base/qbman/qman_priv.h   |   6 +-
 drivers/bus/dpaa/dpaa_bus.c               |  43 ++++-
 drivers/bus/dpaa/include/fsl_qman.h       |  62 +++++--
 drivers/bus/dpaa/include/fsl_usd.h        |   4 +
 drivers/bus/dpaa/include/process.h        |  11 +-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  21 +++
 drivers/bus/dpaa/rte_dpaa_bus.h           |  15 ++
 drivers/net/dpaa/Makefile                 |   3 +
 drivers/net/dpaa/dpaa_ethdev.c            | 259 ++++++++++++++++++++++++++----
 drivers/net/dpaa/dpaa_ethdev.h            |  21 ++-
 drivers/net/dpaa/dpaa_rxtx.c              | 163 +++++++++++++------
 drivers/net/dpaa/dpaa_rxtx.h              |   7 +-
 drivers/net/dpaa/rte_pmd_dpaa.h           |  39 +++++
 drivers/net/dpaa/rte_pmd_dpaa_version.map |   8 +
 22 files changed, 927 insertions(+), 154 deletions(-)
 create mode 100644 drivers/net/dpaa/rte_pmd_dpaa.h

-- 
2.7.4

[PATCH v3 01/19] net/dpaa: fix uninitialized and unused variables

[Hemant Agrawal]

This patch fixes the issues reported by NXP's internal
coverity build.

Fixes: 05ba55bc2b1a ("net/dpaa: add packet dump for debugging")
Fixes: 37f9b54bd3cf ("net/dpaa: support Tx and Rx queue setup")
Cc: stable@dpdk.org

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 6 +++---
 drivers/net/dpaa/dpaa_rxtx.c   | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 2eb00b5..7b4a6f1 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -697,7 +697,7 @@ static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 static int dpaa_rx_queue_init(struct qman_fq *fq,
 			      uint32_t fqid)
 {
-	struct qm_mcc_initfq opts;
+	struct qm_mcc_initfq opts = {0};
 	int ret;
 
 	PMD_INIT_FUNC_TRACE();
@@ -743,7 +743,7 @@ static int dpaa_rx_queue_init(struct qman_fq *fq,
 static int dpaa_tx_queue_init(struct qman_fq *fq,
 			      struct fman_if *fman_intf)
 {
-	struct qm_mcc_initfq opts;
+	struct qm_mcc_initfq opts = {0};
 	int ret;
 
 	PMD_INIT_FUNC_TRACE();
@@ -774,7 +774,7 @@ static int dpaa_tx_queue_init(struct qman_fq *fq,
 /* Initialise a DEBUG FQ ([rt]x_error, rx_default). */
 static int dpaa_debug_queue_init(struct qman_fq *fq, uint32_t fqid)
 {
-	struct qm_mcc_initfq opts;
+	struct qm_mcc_initfq opts = {0};
 	int ret;
 
 	PMD_INIT_FUNC_TRACE();
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 905ecc0..c3a0920 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -639,7 +639,7 @@ tx_on_external_pool(struct qman_fq *txq, struct rte_mbuf *mbuf,
 		return 1;
 	}
 
-	DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, dpaa_intf->bp_info->bpid);
+	DPAA_MBUF_TO_CONTIG_FD(dmable_mbuf, fd_arr, dpaa_intf->bp_info->bpid);
 
 	return 0;
 }
-- 
2.7.4

[PATCH v3 02/19] net/dpaa: fix the mbuf packet type if zero

[Hemant Agrawal]

From: Ashish Jain <ashish.jain@nxp.com>

Populate the mbuf field packet_type which is required
for calculating checksum while transmitting frames

Fixes: 8cffdcbe85aa ("net/dpaa: support scattered Rx")
Cc: stable@dpdk.org

Signed-off-by: Ashish Jain <ashish.jain@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index c3a0920..630d7a5 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -32,6 +32,7 @@
 #include <rte_ip.h>
 #include <rte_tcp.h>
 #include <rte_udp.h>
+#include <rte_net.h>
 
 #include "dpaa_ethdev.h"
 #include "dpaa_rxtx.h"
@@ -478,6 +479,15 @@ dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 	fd->opaque_addr = 0;
 
 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
+		if (!mbuf->packet_type) {
+			struct rte_net_hdr_lens hdr_lens;
+
+			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+					| RTE_PTYPE_L4_MASK);
+			mbuf->l2_len = hdr_lens.l2_len;
+			mbuf->l3_len = hdr_lens.l3_len;
+		}
 		if (temp->data_off < DEFAULT_TX_ICEOF
 			+ sizeof(struct dpaa_eth_parse_results_t))
 			temp->data_off = DEFAULT_TX_ICEOF
@@ -585,6 +595,15 @@ tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
 	}
 
 	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
+		if (!mbuf->packet_type) {
+			struct rte_net_hdr_lens hdr_lens;
+
+			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+					| RTE_PTYPE_L4_MASK);
+			mbuf->l2_len = hdr_lens.l2_len;
+			mbuf->l3_len = hdr_lens.l3_len;
+		}
 		if (mbuf->data_off < (DEFAULT_TX_ICEOF +
 		    sizeof(struct dpaa_eth_parse_results_t))) {
 			DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
-- 
2.7.4

[PATCH v3 03/19] net/dpaa: fix FW version code

[Hemant Agrawal]

fix the soc id path and missing fclose

Fixes: cf0fab1d2ca5 ("net/dpaa: support firmware version get API")
Cc: stable@dpdk.org

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 14 +++++---------
 drivers/net/dpaa/dpaa_ethdev.h |  2 +-
 2 files changed, 6 insertions(+), 10 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 7b4a6f1..db6574f 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -186,19 +186,15 @@ dpaa_fw_version_get(struct rte_eth_dev *dev __rte_unused,
 		DPAA_PMD_ERR("Unable to open SoC device");
 		return -ENOTSUP; /* Not supported on this infra */
 	}
-
-	ret = fscanf(svr_file, "svr:%x", &svr_ver);
-	if (ret <= 0) {
+	if (fscanf(svr_file, "svr:%x", &svr_ver) <= 0)
 		DPAA_PMD_ERR("Unable to read SoC device");
-		return -ENOTSUP; /* Not supported on this infra */
-	}
 
-	ret = snprintf(fw_version, fw_size,
-		       "svr:%x-fman-v%x",
-		       svr_ver,
-		       fman_ip_rev);
+	fclose(svr_file);
 
+	ret = snprintf(fw_version, fw_size, "SVR:%x-fman-v%x",
+		       svr_ver, fman_ip_rev);
 	ret += 1; /* add the size of '\0' */
+
 	if (fw_size < (uint32_t)ret)
 		return ret;
 	else
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index bd63ee0..254fca2 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -20,7 +20,7 @@
 /* DPAA SoC identifier; If this is not available, it can be concluded
  * that board is non-DPAA. Single slot is currently supported.
  */
-#define DPAA_SOC_ID_FILE		"sys/devices/soc0/soc_id"
+#define DPAA_SOC_ID_FILE		"/sys/devices/soc0/soc_id"
 
 #define DPAA_MBUF_HW_ANNOTATION		64
 #define DPAA_FD_PTA_SIZE		64
-- 
2.7.4

[PATCH v3 04/19] bus/dpaa: update platform soc value register routines

[Hemant Agrawal]

This patch update the logic and expose the soc value
register, so that it can be used by other modules as well.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/dpaa_bus.c               | 12 ++++++++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  8 ++++++++
 drivers/bus/dpaa/rte_dpaa_bus.h           | 11 +++++++++++
 drivers/net/dpaa/dpaa_ethdev.c            |  4 +++-
 drivers/net/dpaa/dpaa_ethdev.h            |  5 -----
 5 files changed, 34 insertions(+), 6 deletions(-)

diff --git a/drivers/bus/dpaa/dpaa_bus.c b/drivers/bus/dpaa/dpaa_bus.c
index 79f4858..a7c05b3 100644
--- a/drivers/bus/dpaa/dpaa_bus.c
+++ b/drivers/bus/dpaa/dpaa_bus.c
@@ -51,6 +51,8 @@ struct netcfg_info *dpaa_netcfg;
 /* define a variable to hold the portal_key, once created.*/
 pthread_key_t dpaa_portal_key;
 
+unsigned int dpaa_svr_family;
+
 RTE_DEFINE_PER_LCORE(bool, _dpaa_io);
 
 static inline void
@@ -417,6 +419,8 @@ rte_dpaa_bus_probe(void)
 	int ret = -1;
 	struct rte_dpaa_device *dev;
 	struct rte_dpaa_driver *drv;
+	FILE *svr_file = NULL;
+	unsigned int svr_ver;
 
 	BUS_INIT_FUNC_TRACE();
 
@@ -436,6 +440,14 @@ rte_dpaa_bus_probe(void)
 			break;
 		}
 	}
+
+	svr_file = fopen(DPAA_SOC_ID_FILE, "r");
+	if (svr_file) {
+		if (fscanf(svr_file, "svr:%x", &svr_ver) > 0)
+			dpaa_svr_family = svr_ver & SVR_MASK;
+		fclose(svr_file);
+	}
+
 	return 0;
 }
 
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index fb9d532..eeeb458 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -64,3 +64,11 @@ DPDK_17.11 {
 
 	local: *;
 };
+
+DPDK_18.02 {
+	global:
+
+	dpaa_svr_family;
+
+	local: *;
+} DPDK_17.11;
diff --git a/drivers/bus/dpaa/rte_dpaa_bus.h b/drivers/bus/dpaa/rte_dpaa_bus.h
index 5758274..d9e8c84 100644
--- a/drivers/bus/dpaa/rte_dpaa_bus.h
+++ b/drivers/bus/dpaa/rte_dpaa_bus.h
@@ -20,6 +20,17 @@
 #define DEV_TO_DPAA_DEVICE(ptr)	\
 		container_of(ptr, struct rte_dpaa_device, device)
 
+/* DPAA SoC identifier; If this is not available, it can be concluded
+ * that board is non-DPAA. Single slot is currently supported.
+ */
+#define DPAA_SOC_ID_FILE	"/sys/devices/soc0/soc_id"
+
+#define SVR_LS1043A_FAMILY	0x87920000
+#define SVR_LS1046A_FAMILY	0x87070000
+#define SVR_MASK		0xffff0000
+
+extern unsigned int dpaa_svr_family;
+
 struct rte_dpaa_device;
 struct rte_dpaa_driver;
 
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index db6574f..24943ef 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -186,7 +186,9 @@ dpaa_fw_version_get(struct rte_eth_dev *dev __rte_unused,
 		DPAA_PMD_ERR("Unable to open SoC device");
 		return -ENOTSUP; /* Not supported on this infra */
 	}
-	if (fscanf(svr_file, "svr:%x", &svr_ver) <= 0)
+	if (fscanf(svr_file, "svr:%x", &svr_ver) > 0)
+		dpaa_svr_family = svr_ver & SVR_MASK;
+	else
 		DPAA_PMD_ERR("Unable to read SoC device");
 
 	fclose(svr_file);
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 254fca2..9c3b42c 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -17,11 +17,6 @@
 #include <of.h>
 #include <netcfg.h>
 
-/* DPAA SoC identifier; If this is not available, it can be concluded
- * that board is non-DPAA. Single slot is currently supported.
- */
-#define DPAA_SOC_ID_FILE		"/sys/devices/soc0/soc_id"
-
 #define DPAA_MBUF_HW_ANNOTATION		64
 #define DPAA_FD_PTA_SIZE		64
 
-- 
2.7.4

[PATCH v3 05/19] net/dpaa: set the correct frame size in device MTU

[Hemant Agrawal]

From: Ashish Jain <ashish.jain@nxp.com>

Setting correct frame size in dpaa_dev_mtu_set
api call. Also setting correct max frame size in
hardware in dev_configure for jumbo frames

Signed-off-by: Ashish Jain <ashish.jain@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 20 +++++++++++++-------
 drivers/net/dpaa/dpaa_ethdev.h |  4 ++++
 2 files changed, 17 insertions(+), 7 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 24943ef..5a2ea4f 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -85,19 +85,21 @@ static int
 dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	uint32_t frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN
+				+ VLAN_TAG_SIZE;
 
 	PMD_INIT_FUNC_TRACE();
 
-	if (mtu < ETHER_MIN_MTU)
+	if (mtu < ETHER_MIN_MTU || frame_size > DPAA_MAX_RX_PKT_LEN)
 		return -EINVAL;
-	if (mtu > ETHER_MAX_LEN)
+	if (frame_size > ETHER_MAX_LEN)
 		dev->data->dev_conf.rxmode.jumbo_frame = 1;
 	else
 		dev->data->dev_conf.rxmode.jumbo_frame = 0;
 
-	dev->data->dev_conf.rxmode.max_rx_pkt_len = mtu;
+	dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
 
-	fman_if_set_maxfrm(dpaa_intf->fif, mtu);
+	fman_if_set_maxfrm(dpaa_intf->fif, frame_size);
 
 	return 0;
 }
@@ -105,15 +107,19 @@ dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 static int
 dpaa_eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 {
+	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+
 	PMD_INIT_FUNC_TRACE();
 
 	if (dev->data->dev_conf.rxmode.jumbo_frame == 1) {
 		if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
-		    DPAA_MAX_RX_PKT_LEN)
-			return dpaa_mtu_set(dev,
+		    DPAA_MAX_RX_PKT_LEN) {
+			fman_if_set_maxfrm(dpaa_intf->fif,
 				dev->data->dev_conf.rxmode.max_rx_pkt_len);
-		else
+			return 0;
+		} else {
 			return -1;
+		}
 	}
 	return 0;
 }
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 9c3b42c..548ccff 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -45,6 +45,10 @@
 /*Maximum number of slots available in TX ring*/
 #define MAX_TX_RING_SLOTS	8
 
+#ifndef VLAN_TAG_SIZE
+#define VLAN_TAG_SIZE   4 /** < Vlan Header Length */
+#endif
+
 /* PCD frame queues */
 #define DPAA_PCD_FQID_START		0x400
 #define DPAA_PCD_FQID_MULTIPLIER	0x100
-- 
2.7.4

[PATCH v3 06/19] net/dpaa: add frame count based tail drop with CGR

[Hemant Agrawal]

Replace the byte based tail queue congestion support
with frame count based congestion groups.
It can easily map to number of RX descriptors for a queue.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  5 ++
 drivers/net/dpaa/dpaa_ethdev.c            | 98 +++++++++++++++++++++++++++----
 drivers/net/dpaa/dpaa_ethdev.h            |  8 +--
 3 files changed, 97 insertions(+), 14 deletions(-)

diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index eeeb458..f412362 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -69,6 +69,11 @@ DPDK_18.02 {
 	global:
 
 	dpaa_svr_family;
+	qman_alloc_cgrid_range;
+	qman_create_cgr;
+	qman_delete_cgr;
+	qman_modify_cgr;
+	qman_release_cgrid_range;
 
 	local: *;
 } DPDK_17.11;
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 5a2ea4f..5d94af5 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -47,6 +47,9 @@
 /* Keep track of whether QMAN and BMAN have been globally initialized */
 static int is_global_init;
 
+/* Per FQ Taildrop in frame count */
+static unsigned int td_threshold = CGR_RX_PERFQ_THRESH;
+
 struct rte_dpaa_xstats_name_off {
 	char name[RTE_ETH_XSTATS_NAME_SIZE];
 	uint32_t offset;
@@ -421,12 +424,13 @@ static void dpaa_eth_multicast_disable(struct rte_eth_dev *dev)
 
 static
 int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
-			    uint16_t nb_desc __rte_unused,
+			    uint16_t nb_desc,
 			    unsigned int socket_id __rte_unused,
 			    const struct rte_eth_rxconf *rx_conf __rte_unused,
 			    struct rte_mempool *mp)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	struct qman_fq *rxq = &dpaa_intf->rx_queues[queue_idx];
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -462,7 +466,23 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 			    dpaa_intf->name, fd_offset,
 			fman_if_get_fdoff(dpaa_intf->fif));
 	}
-	dev->data->rx_queues[queue_idx] = &dpaa_intf->rx_queues[queue_idx];
+
+	dev->data->rx_queues[queue_idx] = rxq;
+
+	/* configure the CGR size as per the desc size */
+	if (dpaa_intf->cgr_rx) {
+		struct qm_mcc_initcgr cgr_opts = {0};
+		int ret;
+
+		/* Enable tail drop with cgr on this queue */
+		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, nb_desc, 0);
+		ret = qman_modify_cgr(dpaa_intf->cgr_rx, 0, &cgr_opts);
+		if (ret) {
+			DPAA_PMD_WARN(
+				"rx taildrop modify fail on fqid %d (ret=%d)",
+				rxq->fqid, ret);
+		}
+	}
 
 	return 0;
 }
@@ -698,11 +718,21 @@ static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 }
 
 /* Initialise an Rx FQ */
-static int dpaa_rx_queue_init(struct qman_fq *fq,
+static int dpaa_rx_queue_init(struct qman_fq *fq, struct qman_cgr *cgr_rx,
 			      uint32_t fqid)
 {
 	struct qm_mcc_initfq opts = {0};
 	int ret;
+	u32 flags = 0;
+	struct qm_mcc_initcgr cgr_opts = {
+		.we_mask = QM_CGR_WE_CS_THRES |
+				QM_CGR_WE_CSTD_EN |
+				QM_CGR_WE_MODE,
+		.cgr = {
+			.cstd_en = QM_CGR_EN,
+			.mode = QMAN_CGR_MODE_FRAME
+		}
+	};
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -732,12 +762,24 @@ static int dpaa_rx_queue_init(struct qman_fq *fq,
 	opts.fqd.context_a.stashing.data_cl = DPAA_IF_RX_DATA_STASH;
 	opts.fqd.context_a.stashing.context_cl = DPAA_IF_RX_CONTEXT_STASH;
 
-	/*Enable tail drop */
-	opts.we_mask = opts.we_mask | QM_INITFQ_WE_TDTHRESH;
-	opts.fqd.fq_ctrl = opts.fqd.fq_ctrl | QM_FQCTRL_TDE;
-	qm_fqd_taildrop_set(&opts.fqd.td, CONG_THRESHOLD_RX_Q, 1);
-
-	ret = qman_init_fq(fq, 0, &opts);
+	if (cgr_rx) {
+		/* Enable tail drop with cgr on this queue */
+		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, td_threshold, 0);
+		cgr_rx->cb = NULL;
+		ret = qman_create_cgr(cgr_rx, QMAN_CGR_FLAG_USE_INIT,
+				      &cgr_opts);
+		if (ret) {
+			DPAA_PMD_WARN(
+				"rx taildrop init fail on rx fqid %d (ret=%d)",
+				fqid, ret);
+			goto without_cgr;
+		}
+		opts.we_mask |= QM_INITFQ_WE_CGID;
+		opts.fqd.cgid = cgr_rx->cgrid;
+		opts.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+	}
+without_cgr:
+	ret = qman_init_fq(fq, flags, &opts);
 	if (ret)
 		DPAA_PMD_ERR("init rx fqid %d failed with ret: %d", fqid, ret);
 	return ret;
@@ -819,6 +861,7 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 	struct fm_eth_port_cfg *cfg;
 	struct fman_if *fman_intf;
 	struct fman_if_bpool *bp, *tmp_bp;
+	uint32_t cgrid[DPAA_MAX_NUM_PCD_QUEUES];
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -855,10 +898,31 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 
 	dpaa_intf->rx_queues = rte_zmalloc(NULL,
 		sizeof(struct qman_fq) * num_rx_fqs, MAX_CACHELINE);
+
+	/* If congestion control is enabled globally*/
+	if (td_threshold) {
+		dpaa_intf->cgr_rx = rte_zmalloc(NULL,
+			sizeof(struct qman_cgr) * num_rx_fqs, MAX_CACHELINE);
+
+		ret = qman_alloc_cgrid_range(&cgrid[0], num_rx_fqs, 1, 0);
+		if (ret != num_rx_fqs) {
+			DPAA_PMD_WARN("insufficient CGRIDs available");
+			return -EINVAL;
+		}
+	} else {
+		dpaa_intf->cgr_rx = NULL;
+	}
+
 	for (loop = 0; loop < num_rx_fqs; loop++) {
 		fqid = DPAA_PCD_FQID_START + dpaa_intf->ifid *
 			DPAA_PCD_FQID_MULTIPLIER + loop;
-		ret = dpaa_rx_queue_init(&dpaa_intf->rx_queues[loop], fqid);
+
+		if (dpaa_intf->cgr_rx)
+			dpaa_intf->cgr_rx[loop].cgrid = cgrid[loop];
+
+		ret = dpaa_rx_queue_init(&dpaa_intf->rx_queues[loop],
+			dpaa_intf->cgr_rx ? &dpaa_intf->cgr_rx[loop] : NULL,
+			fqid);
 		if (ret)
 			return ret;
 		dpaa_intf->rx_queues[loop].dpaa_intf = dpaa_intf;
@@ -913,6 +977,7 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 		DPAA_PMD_ERR("Failed to allocate %d bytes needed to "
 						"store MAC addresses",
 				ETHER_ADDR_LEN * DPAA_MAX_MAC_FILTER);
+		rte_free(dpaa_intf->cgr_rx);
 		rte_free(dpaa_intf->rx_queues);
 		rte_free(dpaa_intf->tx_queues);
 		dpaa_intf->rx_queues = NULL;
@@ -951,6 +1016,7 @@ static int
 dpaa_dev_uninit(struct rte_eth_dev *dev)
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	int loop;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -968,6 +1034,18 @@ dpaa_dev_uninit(struct rte_eth_dev *dev)
 	if (dpaa_intf->fc_conf)
 		rte_free(dpaa_intf->fc_conf);
 
+	/* Release RX congestion Groups */
+	if (dpaa_intf->cgr_rx) {
+		for (loop = 0; loop < dpaa_intf->nb_rx_queues; loop++)
+			qman_delete_cgr(&dpaa_intf->cgr_rx[loop]);
+
+		qman_release_cgrid_range(dpaa_intf->cgr_rx[loop].cgrid,
+					 dpaa_intf->nb_rx_queues);
+	}
+
+	rte_free(dpaa_intf->cgr_rx);
+	dpaa_intf->cgr_rx = NULL;
+
 	rte_free(dpaa_intf->rx_queues);
 	dpaa_intf->rx_queues = NULL;
 
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 548ccff..f00a77a 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -34,10 +34,8 @@
 #define DPAA_MIN_RX_BUF_SIZE 512
 #define DPAA_MAX_RX_PKT_LEN  10240
 
-/* RX queue tail drop threshold
- * currently considering 32 KB packets.
- */
-#define CONG_THRESHOLD_RX_Q  (32 * 1024)
+/* RX queue tail drop threshold (CGR Based) in frame count */
+#define CGR_RX_PERFQ_THRESH 256
 
 /*max mac filter for memac(8) including primary mac addr*/
 #define DPAA_MAX_MAC_FILTER (MEMAC_NUM_OF_PADDRS + 1)
@@ -53,6 +51,7 @@
 #define DPAA_PCD_FQID_START		0x400
 #define DPAA_PCD_FQID_MULTIPLIER	0x100
 #define DPAA_DEFAULT_NUM_PCD_QUEUES	1
+#define DPAA_MAX_NUM_PCD_QUEUES		32
 
 #define DPAA_IF_TX_PRIORITY		3
 #define DPAA_IF_RX_PRIORITY		4
@@ -102,6 +101,7 @@ struct dpaa_if {
 	char *name;
 	const struct fm_eth_port_cfg *cfg;
 	struct qman_fq *rx_queues;
+	struct qman_cgr *cgr_rx;
 	struct qman_fq *tx_queues;
 	struct qman_fq debug_queues[2];
 	uint16_t nb_rx_queues;
-- 
2.7.4

[PATCH v3 07/19] bus/dpaa: optimize the qman HW stashing settings

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

The settings are tuned for performance.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index a53459f..49bc317 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -7,6 +7,7 @@
 
 #include "qman.h"
 #include <rte_branch_prediction.h>
+#include <rte_dpaa_bus.h>
 
 /* Compilation constants */
 #define DQRR_MAXFILL	15
@@ -503,7 +504,12 @@ struct qman_portal *qman_create_portal(
 
 	p = &portal->p;
 
-	portal->use_eqcr_ci_stashing = ((qman_ip_rev >= QMAN_REV30) ? 1 : 0);
+	if (dpaa_svr_family == SVR_LS1043A_FAMILY)
+		portal->use_eqcr_ci_stashing = 3;
+	else
+		portal->use_eqcr_ci_stashing =
+					((qman_ip_rev >= QMAN_REV30) ? 1 : 0);
+
 	/*
 	 * prep the low-level portal struct with the mapped addresses from the
 	 * config, everything that follows depends on it and "config" is more
@@ -516,7 +522,7 @@ struct qman_portal *qman_create_portal(
 	 * and stash with high-than-DQRR priority.
 	 */
 	if (qm_eqcr_init(p, qm_eqcr_pvb,
-			 portal->use_eqcr_ci_stashing ? 3 : 0, 1)) {
+			 portal->use_eqcr_ci_stashing, 1)) {
 		pr_err("Qman EQCR initialisation failed\n");
 		goto fail_eqcr;
 	}
-- 
2.7.4

[PATCH v3 08/19] bus/dpaa: optimize the endianness conversions

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c  | 7 ++++---
 drivers/bus/dpaa/include/fsl_qman.h | 2 ++
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 49bc317..b6fd40b 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -906,7 +906,7 @@ static inline unsigned int __poll_portal_fast(struct qman_portal *p,
 	do {
 		qm_dqrr_pvb_update(&p->p);
 		dq = qm_dqrr_current(&p->p);
-		if (!dq)
+		if (unlikely(!dq))
 			break;
 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
 	/* If running on an LE system the fields of the
@@ -1165,6 +1165,7 @@ int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq)
 	}
 	spin_lock_init(&fq->fqlock);
 	fq->fqid = fqid;
+	fq->fqid_le = cpu_to_be32(fqid);
 	fq->flags = flags;
 	fq->state = qman_fq_state_oos;
 	fq->cgr_groupid = 0;
@@ -1953,7 +1954,7 @@ int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags)
 
 int qman_enqueue_multi(struct qman_fq *fq,
 		       const struct qm_fd *fd,
-		int frames_to_send)
+		       int frames_to_send)
 {
 	struct qman_portal *p = get_affine_portal();
 	struct qm_portal *portal = &p->p;
@@ -1975,7 +1976,7 @@ int qman_enqueue_multi(struct qman_fq *fq,
 
 	/* try to send as many frames as possible */
 	while (eqcr->available && frames_to_send--) {
-		eq->fqid = cpu_to_be32(fq->fqid);
+		eq->fqid = fq->fqid_le;
 #ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
 		eq->tag = cpu_to_be32(fq->key);
 #else
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index 5830ad5..5027230 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1197,6 +1197,8 @@ struct qman_fq {
 	 */
 	spinlock_t fqlock;
 	u32 fqid;
+	u32 fqid_le;
+
 	/* DPDK Interface */
 	void *dpaa_intf;
 
-- 
2.7.4

[PATCH v3 09/19] bus/dpaa: add support to create dynamic HW portal

[Hemant Agrawal]

HW portal is a processing context in DPAA. This patch allow
creation of a queue specific HW portal context.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        |  69 ++++++++++++--
 drivers/bus/dpaa/base/qbman/qman_driver.c | 153 +++++++++++++++++++++++++-----
 drivers/bus/dpaa/base/qbman/qman_priv.h   |   6 +-
 drivers/bus/dpaa/dpaa_bus.c               |  31 +++++-
 drivers/bus/dpaa/include/fsl_qman.h       |  25 ++---
 drivers/bus/dpaa/include/fsl_usd.h        |   4 +
 drivers/bus/dpaa/include/process.h        |  11 ++-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |   2 +
 drivers/bus/dpaa/rte_dpaa_bus.h           |   4 +
 9 files changed, 252 insertions(+), 53 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index b6fd40b..d8fb25a 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -621,11 +621,52 @@ struct qman_portal *qman_create_portal(
 	return NULL;
 }
 
+#define MAX_GLOBAL_PORTALS 8
+static struct qman_portal global_portals[MAX_GLOBAL_PORTALS];
+static int global_portals_used[MAX_GLOBAL_PORTALS];
+
+static struct qman_portal *
+qman_alloc_global_portal(void)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_GLOBAL_PORTALS; i++) {
+		if (global_portals_used[i] == 0) {
+			global_portals_used[i] = 1;
+			return &global_portals[i];
+		}
+	}
+	pr_err("No portal available (%x)\n", MAX_GLOBAL_PORTALS);
+
+	return NULL;
+}
+
+static int
+qman_free_global_portal(struct qman_portal *portal)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_GLOBAL_PORTALS; i++) {
+		if (&global_portals[i] == portal) {
+			global_portals_used[i] = 0;
+			return 0;
+		}
+	}
+	return -1;
+}
+
 struct qman_portal *qman_create_affine_portal(const struct qm_portal_config *c,
-					      const struct qman_cgrs *cgrs)
+					      const struct qman_cgrs *cgrs,
+					      int alloc)
 {
 	struct qman_portal *res;
-	struct qman_portal *portal = get_affine_portal();
+	struct qman_portal *portal;
+
+	if (alloc)
+		portal = qman_alloc_global_portal();
+	else
+		portal = get_affine_portal();
+
 	/* A criteria for calling this function (from qman_driver.c) is that
 	 * we're already affine to the cpu and won't schedule onto another cpu.
 	 */
@@ -675,13 +716,18 @@ void qman_destroy_portal(struct qman_portal *qm)
 	spin_lock_destroy(&qm->cgr_lock);
 }
 
-const struct qm_portal_config *qman_destroy_affine_portal(void)
+const struct qm_portal_config *
+qman_destroy_affine_portal(struct qman_portal *qp)
 {
 	/* We don't want to redirect if we're a slave, use "raw" */
-	struct qman_portal *qm = get_affine_portal();
+	struct qman_portal *qm;
 	const struct qm_portal_config *pcfg;
 	int cpu;
 
+	if (qp == NULL)
+		qm = get_affine_portal();
+	else
+		qm = qp;
 	pcfg = qm->config;
 	cpu = pcfg->cpu;
 
@@ -690,6 +736,9 @@ const struct qm_portal_config *qman_destroy_affine_portal(void)
 	spin_lock(&affine_mask_lock);
 	CPU_CLR(cpu, &affine_mask);
 	spin_unlock(&affine_mask_lock);
+
+	qman_free_global_portal(qm);
+
 	return pcfg;
 }
 
@@ -1096,27 +1145,27 @@ void qman_start_dequeues(void)
 		qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL);
 }
 
-void qman_static_dequeue_add(u32 pools)
+void qman_static_dequeue_add(u32 pools, struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 
 	pools &= p->config->pools;
 	p->sdqcr |= pools;
 	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
 }
 
-void qman_static_dequeue_del(u32 pools)
+void qman_static_dequeue_del(u32 pools, struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 
 	pools &= p->config->pools;
 	p->sdqcr &= ~pools;
 	qm_dqrr_sdqcr_set(&p->p, p->sdqcr);
 }
 
-u32 qman_static_dequeue_get(void)
+u32 qman_static_dequeue_get(struct qman_portal *qp)
 {
-	struct qman_portal *p = get_affine_portal();
+	struct qman_portal *p = qp ? qp : get_affine_portal();
 	return p->sdqcr;
 }
 
diff --git a/drivers/bus/dpaa/base/qbman/qman_driver.c b/drivers/bus/dpaa/base/qbman/qman_driver.c
index c17d15f..7cfa8ee 100644
--- a/drivers/bus/dpaa/base/qbman/qman_driver.c
+++ b/drivers/bus/dpaa/base/qbman/qman_driver.c
@@ -24,8 +24,8 @@ void *qman_ccsr_map;
 /* The qman clock frequency */
 u32 qman_clk;
 
-static __thread int fd = -1;
-static __thread struct qm_portal_config pcfg;
+static __thread int qmfd = -1;
+static __thread struct qm_portal_config qpcfg;
 static __thread struct dpaa_ioctl_portal_map map = {
 	.type = dpaa_portal_qman
 };
@@ -44,16 +44,16 @@ static int fsl_qman_portal_init(uint32_t index, int is_shared)
 		error(0, ret, "pthread_getaffinity_np()");
 		return ret;
 	}
-	pcfg.cpu = -1;
+	qpcfg.cpu = -1;
 	for (loop = 0; loop < CPU_SETSIZE; loop++)
 		if (CPU_ISSET(loop, &cpuset)) {
-			if (pcfg.cpu != -1) {
+			if (qpcfg.cpu != -1) {
 				pr_err("Thread is not affine to 1 cpu\n");
 				return -EINVAL;
 			}
-			pcfg.cpu = loop;
+			qpcfg.cpu = loop;
 		}
-	if (pcfg.cpu == -1) {
+	if (qpcfg.cpu == -1) {
 		pr_err("Bug in getaffinity handling!\n");
 		return -EINVAL;
 	}
@@ -65,36 +65,36 @@ static int fsl_qman_portal_init(uint32_t index, int is_shared)
 		error(0, ret, "process_portal_map()");
 		return ret;
 	}
-	pcfg.channel = map.channel;
-	pcfg.pools = map.pools;
-	pcfg.index = map.index;
+	qpcfg.channel = map.channel;
+	qpcfg.pools = map.pools;
+	qpcfg.index = map.index;
 
 	/* Make the portal's cache-[enabled|inhibited] regions */
-	pcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena;
-	pcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh;
+	qpcfg.addr_virt[DPAA_PORTAL_CE] = map.addr.cena;
+	qpcfg.addr_virt[DPAA_PORTAL_CI] = map.addr.cinh;
 
-	fd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
-	if (fd == -1) {
+	qmfd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
+	if (qmfd == -1) {
 		pr_err("QMan irq init failed\n");
 		process_portal_unmap(&map.addr);
 		return -EBUSY;
 	}
 
-	pcfg.is_shared = is_shared;
-	pcfg.node = NULL;
-	pcfg.irq = fd;
+	qpcfg.is_shared = is_shared;
+	qpcfg.node = NULL;
+	qpcfg.irq = qmfd;
 
-	portal = qman_create_affine_portal(&pcfg, NULL);
+	portal = qman_create_affine_portal(&qpcfg, NULL, 0);
 	if (!portal) {
 		pr_err("Qman portal initialisation failed (%d)\n",
-		       pcfg.cpu);
+		       qpcfg.cpu);
 		process_portal_unmap(&map.addr);
 		return -EBUSY;
 	}
 
 	irq_map.type = dpaa_portal_qman;
 	irq_map.portal_cinh = map.addr.cinh;
-	process_portal_irq_map(fd, &irq_map);
+	process_portal_irq_map(qmfd, &irq_map);
 	return 0;
 }
 
@@ -103,10 +103,10 @@ static int fsl_qman_portal_finish(void)
 	__maybe_unused const struct qm_portal_config *cfg;
 	int ret;
 
-	process_portal_irq_unmap(fd);
+	process_portal_irq_unmap(qmfd);
 
-	cfg = qman_destroy_affine_portal();
-	DPAA_BUG_ON(cfg != &pcfg);
+	cfg = qman_destroy_affine_portal(NULL);
+	DPAA_BUG_ON(cfg != &qpcfg);
 	ret = process_portal_unmap(&map.addr);
 	if (ret)
 		error(0, ret, "process_portal_unmap()");
@@ -128,14 +128,119 @@ int qman_thread_finish(void)
 
 void qman_thread_irq(void)
 {
-	qbman_invoke_irq(pcfg.irq);
+	qbman_invoke_irq(qpcfg.irq);
 
 	/* Now we need to uninhibit interrupts. This is the only code outside
 	 * the regular portal driver that manipulates any portal register, so
 	 * rather than breaking that encapsulation I am simply hard-coding the
 	 * offset to the inhibit register here.
 	 */
-	out_be32(pcfg.addr_virt[DPAA_PORTAL_CI] + 0xe0c, 0);
+	out_be32(qpcfg.addr_virt[DPAA_PORTAL_CI] + 0xe0c, 0);
+}
+
+struct qman_portal *fsl_qman_portal_create(void)
+{
+	cpu_set_t cpuset;
+	struct qman_portal *res;
+
+	struct qm_portal_config *q_pcfg;
+	int loop, ret;
+	struct dpaa_ioctl_irq_map irq_map;
+	struct dpaa_ioctl_portal_map q_map = {0};
+	int q_fd;
+
+	q_pcfg = kzalloc((sizeof(struct qm_portal_config)), 0);
+	if (!q_pcfg) {
+		error(0, -1, "q_pcfg kzalloc failed");
+		return NULL;
+	}
+
+	/* Verify the thread's cpu-affinity */
+	ret = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t),
+				     &cpuset);
+	if (ret) {
+		error(0, ret, "pthread_getaffinity_np()");
+		return NULL;
+	}
+
+	q_pcfg->cpu = -1;
+	for (loop = 0; loop < CPU_SETSIZE; loop++)
+		if (CPU_ISSET(loop, &cpuset)) {
+			if (q_pcfg->cpu != -1) {
+				pr_err("Thread is not affine to 1 cpu\n");
+				return NULL;
+			}
+			q_pcfg->cpu = loop;
+		}
+	if (q_pcfg->cpu == -1) {
+		pr_err("Bug in getaffinity handling!\n");
+		return NULL;
+	}
+
+	/* Allocate and map a qman portal */
+	q_map.type = dpaa_portal_qman;
+	q_map.index = QBMAN_ANY_PORTAL_IDX;
+	ret = process_portal_map(&q_map);
+	if (ret) {
+		error(0, ret, "process_portal_map()");
+		return NULL;
+	}
+	q_pcfg->channel = q_map.channel;
+	q_pcfg->pools = q_map.pools;
+	q_pcfg->index = q_map.index;
+
+	/* Make the portal's cache-[enabled|inhibited] regions */
+	q_pcfg->addr_virt[DPAA_PORTAL_CE] = q_map.addr.cena;
+	q_pcfg->addr_virt[DPAA_PORTAL_CI] = q_map.addr.cinh;
+
+	q_fd = open(QMAN_PORTAL_IRQ_PATH, O_RDONLY);
+	if (q_fd == -1) {
+		pr_err("QMan irq init failed\n");
+		goto err1;
+	}
+
+	q_pcfg->irq = q_fd;
+
+	res = qman_create_affine_portal(q_pcfg, NULL, true);
+	if (!res) {
+		pr_err("Qman portal initialisation failed (%d)\n",
+		       q_pcfg->cpu);
+		goto err2;
+	}
+
+	irq_map.type = dpaa_portal_qman;
+	irq_map.portal_cinh = q_map.addr.cinh;
+	process_portal_irq_map(q_fd, &irq_map);
+
+	return res;
+err2:
+	close(q_fd);
+err1:
+	process_portal_unmap(&q_map.addr);
+	return NULL;
+}
+
+int fsl_qman_portal_destroy(struct qman_portal *qp)
+{
+	const struct qm_portal_config *cfg;
+	struct dpaa_portal_map addr;
+	int ret;
+
+	cfg = qman_destroy_affine_portal(qp);
+	kfree(qp);
+
+	process_portal_irq_unmap(cfg->irq);
+
+	addr.cena = cfg->addr_virt[DPAA_PORTAL_CE];
+	addr.cinh = cfg->addr_virt[DPAA_PORTAL_CI];
+
+	ret = process_portal_unmap(&addr);
+	if (ret)
+		pr_err("process_portal_unmap() (%d)\n", ret);
+
+	kfree((void *)cfg);
+
+	return ret;
 }
 
 int qman_global_init(void)
diff --git a/drivers/bus/dpaa/base/qbman/qman_priv.h b/drivers/bus/dpaa/base/qbman/qman_priv.h
index db0b310..9e4471e 100644
--- a/drivers/bus/dpaa/base/qbman/qman_priv.h
+++ b/drivers/bus/dpaa/base/qbman/qman_priv.h
@@ -146,8 +146,10 @@ int qm_get_wpm(int *wpm);
 
 struct qman_portal *qman_create_affine_portal(
 			const struct qm_portal_config *config,
-			const struct qman_cgrs *cgrs);
-const struct qm_portal_config *qman_destroy_affine_portal(void);
+			const struct qman_cgrs *cgrs,
+			int alloc);
+const struct qm_portal_config *
+qman_destroy_affine_portal(struct qman_portal *q);
 
 struct qm_portal_config *qm_get_unused_portal(void);
 struct qm_portal_config *qm_get_unused_portal_idx(uint32_t idx);
diff --git a/drivers/bus/dpaa/dpaa_bus.c b/drivers/bus/dpaa/dpaa_bus.c
index a7c05b3..329a125 100644
--- a/drivers/bus/dpaa/dpaa_bus.c
+++ b/drivers/bus/dpaa/dpaa_bus.c
@@ -264,8 +264,7 @@ _dpaa_portal_init(void *arg)
  * rte_dpaa_portal_init - Wrapper over _dpaa_portal_init with thread level check
  * XXX Complete this
  */
-int
-rte_dpaa_portal_init(void *arg)
+int rte_dpaa_portal_init(void *arg)
 {
 	if (unlikely(!RTE_PER_LCORE(_dpaa_io)))
 		return _dpaa_portal_init(arg);
@@ -273,6 +272,34 @@ rte_dpaa_portal_init(void *arg)
 	return 0;
 }
 
+int
+rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq)
+{
+	/* Affine above created portal with channel*/
+	u32 sdqcr;
+	struct qman_portal *qp;
+
+	if (unlikely(!RTE_PER_LCORE(_dpaa_io)))
+		_dpaa_portal_init(arg);
+
+	/* Initialise qman specific portals */
+	qp = fsl_qman_portal_create();
+	if (!qp) {
+		DPAA_BUS_LOG(ERR, "Unable to alloc fq portal");
+		return -1;
+	}
+	fq->qp = qp;
+	sdqcr = QM_SDQCR_CHANNELS_POOL_CONV(fq->ch_id);
+	qman_static_dequeue_add(sdqcr, qp);
+
+	return 0;
+}
+
+int rte_dpaa_portal_fq_close(struct qman_fq *fq)
+{
+	return fsl_qman_portal_destroy(fq->qp);
+}
+
 void
 dpaa_portal_finish(void *arg)
 {
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index 5027230..fc00d8d 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1190,21 +1190,24 @@ struct qman_fq_cb {
 struct qman_fq {
 	/* Caller of qman_create_fq() provides these demux callbacks */
 	struct qman_fq_cb cb;
-	/*
-	 * These are internal to the driver, don't touch. In particular, they
-	 * may change, be removed, or extended (so you shouldn't rely on
-	 * sizeof(qman_fq) being a constant).
-	 */
-	spinlock_t fqlock;
-	u32 fqid;
+
 	u32 fqid_le;
+	u16 ch_id;
+	u8 cgr_groupid;
+	u8 is_static;
 
 	/* DPDK Interface */
 	void *dpaa_intf;
 
+	/* affined portal in case of static queue */
+	struct qman_portal *qp;
+
 	volatile unsigned long flags;
+
 	enum qman_fq_state state;
-	int cgr_groupid;
+	u32 fqid;
+	spinlock_t fqlock;
+
 	struct rb_node node;
 #ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
 	u32 key;
@@ -1383,7 +1386,7 @@ void qman_start_dequeues(void);
  * (SDQCR). The requested pools are limited to those the portal has dequeue
  * access to.
  */
-void qman_static_dequeue_add(u32 pools);
+void qman_static_dequeue_add(u32 pools, struct qman_portal *qm);
 
 /**
  * qman_static_dequeue_del - Remove pool channels from the portal SDQCR
@@ -1393,7 +1396,7 @@ void qman_static_dequeue_add(u32 pools);
  * register (SDQCR). The requested pools are limited to those the portal has
  * dequeue access to.
  */
-void qman_static_dequeue_del(u32 pools);
+void qman_static_dequeue_del(u32 pools, struct qman_portal *qp);
 
 /**
  * qman_static_dequeue_get - return the portal's current SDQCR
@@ -1402,7 +1405,7 @@ void qman_static_dequeue_del(u32 pools);
  * entire register is returned, so if only the currently-enabled pool channels
  * are desired, mask the return value with QM_SDQCR_CHANNELS_POOL_MASK.
  */
-u32 qman_static_dequeue_get(void);
+u32 qman_static_dequeue_get(struct qman_portal *qp);
 
 /**
  * qman_dca - Perform a Discrete Consumption Acknowledgment
diff --git a/drivers/bus/dpaa/include/fsl_usd.h b/drivers/bus/dpaa/include/fsl_usd.h
index ada92f2..e183617 100644
--- a/drivers/bus/dpaa/include/fsl_usd.h
+++ b/drivers/bus/dpaa/include/fsl_usd.h
@@ -67,6 +67,10 @@ void bman_thread_irq(void);
 int qman_global_init(void);
 int bman_global_init(void);
 
+/* Direct portal create and destroy */
+struct qman_portal *fsl_qman_portal_create(void);
+int fsl_qman_portal_destroy(struct qman_portal *qp);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/drivers/bus/dpaa/include/process.h b/drivers/bus/dpaa/include/process.h
index 079849b..d9ec94e 100644
--- a/drivers/bus/dpaa/include/process.h
+++ b/drivers/bus/dpaa/include/process.h
@@ -39,6 +39,11 @@ enum dpaa_portal_type {
 	dpaa_portal_bman,
 };
 
+struct dpaa_portal_map {
+	void *cinh;
+	void *cena;
+};
+
 struct dpaa_ioctl_portal_map {
 	/* Input parameter, is a qman or bman portal required. */
 	enum dpaa_portal_type type;
@@ -50,10 +55,8 @@ struct dpaa_ioctl_portal_map {
 	/* Return value if the map succeeds, this gives the mapped
 	 * cache-inhibited (cinh) and cache-enabled (cena) addresses.
 	 */
-	struct dpaa_portal_map {
-		void *cinh;
-		void *cena;
-	} addr;
+	struct dpaa_portal_map addr;
+
 	/* Qman-specific return values */
 	u16 channel;
 	uint32_t pools;
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index f412362..4e3afda 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -74,6 +74,8 @@ DPDK_18.02 {
 	qman_delete_cgr;
 	qman_modify_cgr;
 	qman_release_cgrid_range;
+	rte_dpaa_portal_fq_close;
+	rte_dpaa_portal_fq_init;
 
 	local: *;
 } DPDK_17.11;
diff --git a/drivers/bus/dpaa/rte_dpaa_bus.h b/drivers/bus/dpaa/rte_dpaa_bus.h
index d9e8c84..f626066 100644
--- a/drivers/bus/dpaa/rte_dpaa_bus.h
+++ b/drivers/bus/dpaa/rte_dpaa_bus.h
@@ -136,6 +136,10 @@ void rte_dpaa_driver_unregister(struct rte_dpaa_driver *driver);
  */
 int rte_dpaa_portal_init(void *arg);
 
+int rte_dpaa_portal_fq_init(void *arg, struct qman_fq *fq);
+
+int rte_dpaa_portal_fq_close(struct qman_fq *fq);
+
 /**
  * Cleanup a DPAA Portal
  */
-- 
2.7.4

[PATCH v3 10/19] net/dpaa: change Tx HW budget to 7

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

change the TX budget to 7 to sync best with the hw.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.h | 2 +-
 drivers/net/dpaa/dpaa_rxtx.c   | 5 +++--
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index f00a77a..1b36567 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -41,7 +41,7 @@
 #define DPAA_MAX_MAC_FILTER (MEMAC_NUM_OF_PADDRS + 1)
 
 /*Maximum number of slots available in TX ring*/
-#define MAX_TX_RING_SLOTS	8
+#define DPAA_TX_BURST_SIZE	7
 
 #ifndef VLAN_TAG_SIZE
 #define VLAN_TAG_SIZE   4 /** < Vlan Header Length */
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 630d7a5..565ca50 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -669,7 +669,7 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	struct rte_mbuf *mbuf, *mi = NULL;
 	struct rte_mempool *mp;
 	struct dpaa_bp_info *bp_info;
-	struct qm_fd fd_arr[MAX_TX_RING_SLOTS];
+	struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
 	uint32_t frames_to_send, loop, i = 0;
 	uint16_t state;
 	int ret;
@@ -683,7 +683,8 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	DPAA_DP_LOG(DEBUG, "Transmitting %d buffers on queue: %p", nb_bufs, q);
 
 	while (nb_bufs) {
-		frames_to_send = (nb_bufs >> 3) ? MAX_TX_RING_SLOTS : nb_bufs;
+		frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
+				DPAA_TX_BURST_SIZE : nb_bufs;
 		for (loop = 0; loop < frames_to_send; loop++, i++) {
 			mbuf = bufs[i];
 			if (RTE_MBUF_DIRECT(mbuf)) {
-- 
2.7.4

[PATCH v3 11/19] net/dpaa: optimize the Tx burst

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Optimize it for best case. Create a function
for TX offloads to be used in multiple legs.

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 73 ++++++++++++++++++++++++++++----------------
 1 file changed, 46 insertions(+), 27 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 565ca50..148f265 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -272,6 +272,30 @@ static inline void dpaa_checksum_offload(struct rte_mbuf *mbuf,
 	fd->cmd = DPAA_FD_CMD_RPD | DPAA_FD_CMD_DTC;
 }
 
+static inline void
+dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
+{
+	if (!mbuf->packet_type) {
+		struct rte_net_hdr_lens hdr_lens;
+
+		mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
+				RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
+				| RTE_PTYPE_L4_MASK);
+		mbuf->l2_len = hdr_lens.l2_len;
+		mbuf->l3_len = hdr_lens.l3_len;
+	}
+	if (mbuf->data_off < (DEFAULT_TX_ICEOF +
+	    sizeof(struct dpaa_eth_parse_results_t))) {
+		DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
+			"Not enough Headroom "
+			"space for correct Checksum offload."
+			"So Calculating checksum in Software.");
+		dpaa_checksum(mbuf);
+	} else {
+		dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
+	}
+}
+
 struct rte_mbuf *
 dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
 {
@@ -594,27 +618,8 @@ tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
 		rte_pktmbuf_free(mbuf);
 	}
 
-	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
-		if (!mbuf->packet_type) {
-			struct rte_net_hdr_lens hdr_lens;
-
-			mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
-					RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
-					| RTE_PTYPE_L4_MASK);
-			mbuf->l2_len = hdr_lens.l2_len;
-			mbuf->l3_len = hdr_lens.l3_len;
-		}
-		if (mbuf->data_off < (DEFAULT_TX_ICEOF +
-		    sizeof(struct dpaa_eth_parse_results_t))) {
-			DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
-				"Not enough Headroom "
-				"space for correct Checksum offload."
-				"So Calculating checksum in Software.");
-			dpaa_checksum(mbuf);
-		} else {
-			dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
-		}
-	}
+	if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
+		dpaa_unsegmented_checksum(mbuf, fd_arr);
 }
 
 /* Handle all mbufs on dpaa BMAN managed pool */
@@ -670,7 +675,7 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	struct rte_mempool *mp;
 	struct dpaa_bp_info *bp_info;
 	struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
-	uint32_t frames_to_send, loop, i = 0;
+	uint32_t frames_to_send, loop, sent = 0;
 	uint16_t state;
 	int ret;
 
@@ -685,10 +690,23 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 	while (nb_bufs) {
 		frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
 				DPAA_TX_BURST_SIZE : nb_bufs;
-		for (loop = 0; loop < frames_to_send; loop++, i++) {
-			mbuf = bufs[i];
-			if (RTE_MBUF_DIRECT(mbuf)) {
+		for (loop = 0; loop < frames_to_send; loop++) {
+			mbuf = *(bufs++);
+			if (likely(RTE_MBUF_DIRECT(mbuf))) {
 				mp = mbuf->pool;
+				bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
+				if (likely(mp->ops_index ==
+						bp_info->dpaa_ops_index &&
+					mbuf->nb_segs == 1 &&
+					rte_mbuf_refcnt_read(mbuf) == 1)) {
+					DPAA_MBUF_TO_CONTIG_FD(mbuf,
+						&fd_arr[loop], bp_info->bpid);
+					if (mbuf->ol_flags &
+						DPAA_TX_CKSUM_OFFLOAD_MASK)
+						dpaa_unsegmented_checksum(mbuf,
+							&fd_arr[loop]);
+					continue;
+				}
 			} else {
 				mi = rte_mbuf_from_indirect(mbuf);
 				mp = mi->pool;
@@ -729,11 +747,12 @@ dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 					frames_to_send - loop);
 		}
 		nb_bufs -= frames_to_send;
+		sent += frames_to_send;
 	}
 
-	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", i, q);
+	DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", sent, q);
 
-	return i;
+	return sent;
 }
 
 uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
-- 
2.7.4

[PATCH v3 12/19] net/dpaa: optimize Rx path

[Hemant Agrawal]

From: Nipun Gupta <nipun.gupta@nxp.com>

Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_rxtx.c | 48 ++++++++++++++++++++------------------------
 drivers/net/dpaa/dpaa_rxtx.h |  2 +-
 2 files changed, 23 insertions(+), 27 deletions(-)

diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 148f265..98671fa 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -97,12 +97,6 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 	DPAA_DP_LOG(DEBUG, " Parsing mbuf: %p with annotations: %p", m, annot);
 
 	switch (prs) {
-	case DPAA_PKT_TYPE_NONE:
-		m->packet_type = 0;
-		break;
-	case DPAA_PKT_TYPE_ETHER:
-		m->packet_type = RTE_PTYPE_L2_ETHER;
-		break;
 	case DPAA_PKT_TYPE_IPV4:
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV4;
@@ -111,6 +105,9 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV6;
 		break;
+	case DPAA_PKT_TYPE_ETHER:
+		m->packet_type = RTE_PTYPE_L2_ETHER;
+		break;
 	case DPAA_PKT_TYPE_IPV4_FRAG:
 	case DPAA_PKT_TYPE_IPV4_FRAG_UDP:
 	case DPAA_PKT_TYPE_IPV4_FRAG_TCP:
@@ -173,6 +170,9 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 		m->packet_type = RTE_PTYPE_L2_ETHER |
 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
 		break;
+	case DPAA_PKT_TYPE_NONE:
+		m->packet_type = 0;
+		break;
 	/* More switch cases can be added */
 	default:
 		dpaa_slow_parsing(m, prs);
@@ -183,12 +183,11 @@ static inline void dpaa_eth_packet_info(struct rte_mbuf *m,
 					<< DPAA_PKT_L3_LEN_SHIFT;
 
 	/* Set the hash values */
-	m->hash.rss = (uint32_t)(rte_be_to_cpu_64(annot->hash));
-	m->ol_flags = PKT_RX_RSS_HASH;
+	m->hash.rss = (uint32_t)(annot->hash);
 	/* All packets with Bad checksum are dropped by interface (and
 	 * corresponding notification issued to RX error queues).
 	 */
-	m->ol_flags |= PKT_RX_IP_CKSUM_GOOD;
+	m->ol_flags = PKT_RX_RSS_HASH | PKT_RX_IP_CKSUM_GOOD;
 
 	/* Check if Vlan is present */
 	if (prs & DPAA_PARSE_VLAN_MASK)
@@ -297,7 +296,7 @@ dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
 }
 
 struct rte_mbuf *
-dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
+dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 {
 	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 	struct rte_mbuf *first_seg, *prev_seg, *cur_seg, *temp;
@@ -355,34 +354,31 @@ dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid)
 	return first_seg;
 }
 
-static inline struct rte_mbuf *dpaa_eth_fd_to_mbuf(struct qm_fd *fd,
-							uint32_t ifid)
+static inline struct rte_mbuf *
+dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 {
-	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 	struct rte_mbuf *mbuf;
-	void *ptr;
+	struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
+	void *ptr = rte_dpaa_mem_ptov(qm_fd_addr(fd));
 	uint8_t format =
 		(fd->opaque & DPAA_FD_FORMAT_MASK) >> DPAA_FD_FORMAT_SHIFT;
-	uint16_t offset =
-		(fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
-	uint32_t length = fd->opaque & DPAA_FD_LENGTH_MASK;
+	uint16_t offset;
+	uint32_t length;
 
 	DPAA_DP_LOG(DEBUG, " FD--->MBUF");
 
 	if (unlikely(format == qm_fd_sg))
 		return dpaa_eth_sg_to_mbuf(fd, ifid);
 
+	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
+
+	offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
+	length = fd->opaque & DPAA_FD_LENGTH_MASK;
+
 	/* Ignoring case when format != qm_fd_contig */
 	dpaa_display_frame(fd);
-	ptr = rte_dpaa_mem_ptov(fd->addr);
-	/* Ignoring case when ptr would be NULL. That is only possible incase
-	 * of a corrupted packet
-	 */
 
 	mbuf = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
-	/* Prefetch the Parse results and packet data to L1 */
-	rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
-	rte_prefetch0((void *)((uint8_t *)ptr + offset));
 
 	mbuf->data_off = offset;
 	mbuf->data_len = length;
@@ -462,11 +458,11 @@ static struct rte_mbuf *dpaa_get_dmable_mbuf(struct rte_mbuf *mbuf,
 	if (!dpaa_mbuf)
 		return NULL;
 
-	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + mbuf->data_off, (void *)
+	memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + RTE_PKTMBUF_HEADROOM, (void *)
 		((uint8_t *)(mbuf->buf_addr) + mbuf->data_off), mbuf->pkt_len);
 
 	/* Copy only the required fields */
-	dpaa_mbuf->data_off = mbuf->data_off;
+	dpaa_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 	dpaa_mbuf->pkt_len = mbuf->pkt_len;
 	dpaa_mbuf->ol_flags = mbuf->ol_flags;
 	dpaa_mbuf->packet_type = mbuf->packet_type;
diff --git a/drivers/net/dpaa/dpaa_rxtx.h b/drivers/net/dpaa/dpaa_rxtx.h
index 9308b3a..78e804f 100644
--- a/drivers/net/dpaa/dpaa_rxtx.h
+++ b/drivers/net/dpaa/dpaa_rxtx.h
@@ -262,7 +262,7 @@ uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
 			      struct rte_mbuf **bufs __rte_unused,
 			      uint16_t nb_bufs __rte_unused);
 
-struct rte_mbuf *dpaa_eth_sg_to_mbuf(struct qm_fd *fd, uint32_t ifid);
+struct rte_mbuf *dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid);
 
 int dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 			   struct qm_fd *fd,
-- 
2.7.4

[PATCH v3 13/19] bus/dpaa: query queue frame count support

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 22 ++++++++++++++++++++++
 drivers/bus/dpaa/include/fsl_qman.h       |  7 +++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  1 +
 3 files changed, 30 insertions(+)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index d8fb25a..ffb008e 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -1722,6 +1722,28 @@ int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np)
 	return 0;
 }
 
+int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt)
+{
+	struct qm_mc_command *mcc;
+	struct qm_mc_result *mcr;
+	struct qman_portal *p = get_affine_portal();
+
+	mcc = qm_mc_start(&p->p);
+	mcc->queryfq.fqid = cpu_to_be32(fq->fqid);
+	qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP);
+	while (!(mcr = qm_mc_result(&p->p)))
+		cpu_relax();
+	DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP);
+
+	if (mcr->result == QM_MCR_RESULT_OK)
+		*frm_cnt = be24_to_cpu(mcr->queryfq_np.frm_cnt);
+	else if (mcr->result == QM_MCR_RESULT_ERR_FQID)
+		return -ERANGE;
+	else if (mcr->result != QM_MCR_RESULT_OK)
+		return -EIO;
+	return 0;
+}
+
 int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq)
 {
 	struct qm_mc_command *mcc;
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index fc00d8d..d769d50 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1616,6 +1616,13 @@ int qman_query_fq_has_pkts(struct qman_fq *fq);
 int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np);
 
 /**
+ * qman_query_fq_frmcnt - Queries fq frame count
+ * @fq: the frame queue object to be queried
+ * @frm_cnt: number of frames in the queue
+ */
+int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt);
+
+/**
  * qman_query_wq - Queries work queue lengths
  * @query_dedicated: If non-zero, query length of WQs in the channel dedicated
  *		to this software portal. Otherwise, query length of WQs in a
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index 4e3afda..212c75f 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -73,6 +73,7 @@ DPDK_18.02 {
 	qman_create_cgr;
 	qman_delete_cgr;
 	qman_modify_cgr;
+	qman_query_fq_frm_cnt;
 	qman_release_cgrid_range;
 	rte_dpaa_portal_fq_close;
 	rte_dpaa_portal_fq_init;
-- 
2.7.4

[PATCH v3 14/19] net/dpaa: add Rx queue count support

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/net/dpaa/dpaa_ethdev.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 5d94af5..de016ab 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -513,6 +513,22 @@ static void dpaa_eth_tx_queue_release(void *txq __rte_unused)
 	PMD_INIT_FUNC_TRACE();
 }
 
+static uint32_t
+dpaa_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+	struct dpaa_if *dpaa_intf = dev->data->dev_private;
+	struct qman_fq *rxq = &dpaa_intf->rx_queues[rx_queue_id];
+	u32 frm_cnt = 0;
+
+	PMD_INIT_FUNC_TRACE();
+
+	if (qman_query_fq_frm_cnt(rxq, &frm_cnt) == 0) {
+		RTE_LOG(DEBUG, PMD, "RX frame count for q(%d) is %u\n",
+			rx_queue_id, frm_cnt);
+	}
+	return frm_cnt;
+}
+
 static int dpaa_link_down(struct rte_eth_dev *dev)
 {
 	PMD_INIT_FUNC_TRACE();
@@ -664,6 +680,7 @@ static struct eth_dev_ops dpaa_devops = {
 	.tx_queue_setup		  = dpaa_eth_tx_queue_setup,
 	.rx_queue_release	  = dpaa_eth_rx_queue_release,
 	.tx_queue_release	  = dpaa_eth_tx_queue_release,
+	.rx_queue_count		  = dpaa_dev_rx_queue_count,
 
 	.flow_ctrl_get		  = dpaa_flow_ctrl_get,
 	.flow_ctrl_set		  = dpaa_flow_ctrl_set,
-- 
2.7.4

[PATCH v3 15/19] net/dpaa: add support for loopback API

[Hemant Agrawal]

PMD specific API is being added as an EXPERIMENTAL API

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 doc/api/doxy-api-index.md                 |  1 +
 doc/api/doxy-api.conf                     |  1 +
 drivers/net/dpaa/Makefile                 |  3 +++
 drivers/net/dpaa/dpaa_ethdev.c            | 42 +++++++++++++++++++++++++++++++
 drivers/net/dpaa/rte_pmd_dpaa.h           | 39 ++++++++++++++++++++++++++++
 drivers/net/dpaa/rte_pmd_dpaa_version.map |  8 ++++++
 6 files changed, 94 insertions(+)
 create mode 100644 drivers/net/dpaa/rte_pmd_dpaa.h

diff --git a/doc/api/doxy-api-index.md b/doc/api/doxy-api-index.md
index 3492702..38314af 100644
--- a/doc/api/doxy-api-index.md
+++ b/doc/api/doxy-api-index.md
@@ -60,6 +60,7 @@ The public API headers are grouped by topics:
   [ixgbe]              (@ref rte_pmd_ixgbe.h),
   [i40e]               (@ref rte_pmd_i40e.h),
   [bnxt]               (@ref rte_pmd_bnxt.h),
+  [dpaa]               (@ref rte_pmd_dpaa.h),
   [crypto_scheduler]   (@ref rte_cryptodev_scheduler.h)
 
 - **memory**:
diff --git a/doc/api/doxy-api.conf b/doc/api/doxy-api.conf
index b2cbe94..09e3232 100644
--- a/doc/api/doxy-api.conf
+++ b/doc/api/doxy-api.conf
@@ -33,6 +33,7 @@ INPUT                   = doc/api/doxy-api-index.md \
                           drivers/crypto/scheduler \
                           drivers/net/bnxt \
                           drivers/net/bonding \
+                          drivers/net/dpaa \
                           drivers/net/i40e \
                           drivers/net/ixgbe \
                           drivers/net/softnic \
diff --git a/drivers/net/dpaa/Makefile b/drivers/net/dpaa/Makefile
index e5f662f..b1fc5a0 100644
--- a/drivers/net/dpaa/Makefile
+++ b/drivers/net/dpaa/Makefile
@@ -34,4 +34,7 @@ LDLIBS += -lrte_mempool_dpaa
 LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring
 LDLIBS += -lrte_ethdev -lrte_net -lrte_kvargs
 
+# install this header file
+SYMLINK-$(CONFIG_RTE_LIBRTE_DPAA_PMD)-include := rte_pmd_dpaa.h
+
 include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index de016ab..85ccea4 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -38,6 +38,7 @@
 
 #include <dpaa_ethdev.h>
 #include <dpaa_rxtx.h>
+#include <rte_pmd_dpaa.h>
 
 #include <fsl_usd.h>
 #include <fsl_qman.h>
@@ -84,6 +85,8 @@ static const struct rte_dpaa_xstats_name_off dpaa_xstats_strings[] = {
 		offsetof(struct dpaa_if_stats, tund)},
 };
 
+static struct rte_dpaa_driver rte_dpaa_pmd;
+
 static int
 dpaa_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 {
@@ -707,6 +710,45 @@ static struct eth_dev_ops dpaa_devops = {
 	.fw_version_get		  = dpaa_fw_version_get,
 };
 
+static bool
+is_device_supported(struct rte_eth_dev *dev, struct rte_dpaa_driver *drv)
+{
+	if (strcmp(dev->device->driver->name,
+		   drv->driver.name))
+		return false;
+
+	return true;
+}
+
+static bool
+is_dpaa_supported(struct rte_eth_dev *dev)
+{
+	return is_device_supported(dev, &rte_dpaa_pmd);
+}
+
+int
+rte_pmd_dpaa_set_tx_loopback(uint8_t port, uint8_t on)
+{
+	struct rte_eth_dev *dev;
+	struct dpaa_if *dpaa_intf;
+
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(port, -ENODEV);
+
+	dev = &rte_eth_devices[port];
+
+	if (!is_dpaa_supported(dev))
+		return -ENOTSUP;
+
+	dpaa_intf = dev->data->dev_private;
+
+	if (on)
+		fman_if_loopback_enable(dpaa_intf->fif);
+	else
+		fman_if_loopback_disable(dpaa_intf->fif);
+
+	return 0;
+}
+
 static int dpaa_fc_set_default(struct dpaa_if *dpaa_intf)
 {
 	struct rte_eth_fc_conf *fc_conf;
diff --git a/drivers/net/dpaa/rte_pmd_dpaa.h b/drivers/net/dpaa/rte_pmd_dpaa.h
new file mode 100644
index 0000000..9614be8
--- /dev/null
+++ b/drivers/net/dpaa/rte_pmd_dpaa.h
@@ -0,0 +1,39 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2018 NXP
+ */
+
+#ifndef _PMD_DPAA_H_
+#define _PMD_DPAA_H_
+
+/**
+ * @file rte_pmd_dpaa.h
+ *
+ * NXP dpaa PMD specific functions.
+ *
+ * @warning
+ * @b EXPERIMENTAL: this API may change, or be removed, without prior notice
+ *
+ */
+
+#include <rte_ethdev.h>
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change, or be removed, without prior notice
+ *
+ * Enable/Disable TX loopback
+ *
+ * @param port
+ *    The port identifier of the Ethernet device.
+ * @param on
+ *    1 - Enable TX loopback.
+ *    0 - Disable TX loopback.
+ * @return
+ *   - (0) if successful.
+ *   - (-ENODEV) if *port* invalid.
+ *   - (-EINVAL) if bad parameter.
+ */
+int
+rte_pmd_dpaa_set_tx_loopback(uint8_t port, uint8_t on);
+
+#endif /* _PMD_DPAA_H_ */
diff --git a/drivers/net/dpaa/rte_pmd_dpaa_version.map b/drivers/net/dpaa/rte_pmd_dpaa_version.map
index a70bd19..d1f3ea4 100644
--- a/drivers/net/dpaa/rte_pmd_dpaa_version.map
+++ b/drivers/net/dpaa/rte_pmd_dpaa_version.map
@@ -2,3 +2,11 @@ DPDK_17.11 {
 
 	local: *;
 };
+
+EXPERIMENTAL {
+	global:
+
+	rte_pmd_dpaa_set_tx_loopback;
+
+	local: *;
+} DPDK_17.11;
-- 
2.7.4

[PATCH v3 16/19] app/testpmd: add support for loopback config for dpaa

[Hemant Agrawal]

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 app/test-pmd/Makefile  | 4 ++++
 app/test-pmd/cmdline.c | 7 +++++++
 2 files changed, 11 insertions(+)

diff --git a/app/test-pmd/Makefile b/app/test-pmd/Makefile
index 82b3481..34125e5 100644
--- a/app/test-pmd/Makefile
+++ b/app/test-pmd/Makefile
@@ -43,6 +43,10 @@ ifeq ($(CONFIG_RTE_LIBRTE_PMD_BOND),y)
 LDLIBS += -lrte_pmd_bond
 endif
 
+ifeq ($(CONFIG_RTE_LIBRTE_DPAA_PMD),y)
+LDLIBS += -lrte_pmd_dpaa
+endif
+
 ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
 LDLIBS += -lrte_pmd_ixgbe
 endif
diff --git a/app/test-pmd/cmdline.c b/app/test-pmd/cmdline.c
index f71d963..32096aa 100644
--- a/app/test-pmd/cmdline.c
+++ b/app/test-pmd/cmdline.c
@@ -89,6 +89,9 @@
 #include <rte_eth_bond.h>
 #include <rte_eth_bond_8023ad.h>
 #endif
+#ifdef RTE_LIBRTE_DPAA_PMD
+#include <rte_pmd_dpaa.h>
+#endif
 #ifdef RTE_LIBRTE_IXGBE_PMD
 #include <rte_pmd_ixgbe.h>
 #endif
@@ -12620,6 +12623,10 @@ cmd_set_tx_loopback_parsed(
 	if (ret == -ENOTSUP)
 		ret = rte_pmd_bnxt_set_tx_loopback(res->port_id, is_on);
 #endif
+#ifdef RTE_LIBRTE_DPAA_PMD
+	if (ret == -ENOTSUP)
+		ret = rte_pmd_dpaa_set_tx_loopback(res->port_id, is_on);
+#endif
 
 	switch (ret) {
 	case 0:
-- 
2.7.4

[PATCH v3 17/19] bus/dpaa: add support for static queues

[Hemant Agrawal]

DPAA hardware support two kinds of queues:
1. Pull mode queue - where one needs to regularly pull the packets.
2. Push mode queue - where the hw pushes the packet to queue. These are
   high performance queues, but limited in number.

This patch add the driver support for push mode queues.

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 64 +++++++++++++++++++++++++++++++
 drivers/bus/dpaa/base/qbman/qman.h        |  4 +-
 drivers/bus/dpaa/include/fsl_qman.h       | 14 ++++++-
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  4 ++
 4 files changed, 83 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index ffb008e..7e285a5 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -1051,6 +1051,70 @@ u16 qman_affine_channel(int cpu)
 	return affine_channels[cpu];
 }
 
+unsigned int qman_portal_poll_rx(unsigned int poll_limit,
+				 void **bufs,
+				 struct qman_portal *p)
+{
+	const struct qm_dqrr_entry *dq;
+	struct qman_fq *fq;
+	enum qman_cb_dqrr_result res;
+	unsigned int limit = 0;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	struct qm_dqrr_entry *shadow;
+#endif
+	unsigned int rx_number = 0;
+
+	do {
+		qm_dqrr_pvb_update(&p->p);
+		dq = qm_dqrr_current(&p->p);
+		if (unlikely(!dq))
+			break;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	/* If running on an LE system the fields of the
+	 * dequeue entry must be swapper.  Because the
+	 * QMan HW will ignore writes the DQRR entry is
+	 * copied and the index stored within the copy
+	 */
+		shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)];
+		*shadow = *dq;
+		dq = shadow;
+		shadow->fqid = be32_to_cpu(shadow->fqid);
+		shadow->contextB = be32_to_cpu(shadow->contextB);
+		shadow->seqnum = be16_to_cpu(shadow->seqnum);
+		hw_fd_to_cpu(&shadow->fd);
+#endif
+
+		/* SDQCR: context_b points to the FQ */
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+		fq = get_fq_table_entry(dq->contextB);
+#else
+		fq = (void *)(uintptr_t)dq->contextB;
+#endif
+		/* Now let the callback do its stuff */
+		res = fq->cb.dqrr_dpdk_cb(NULL, p, fq, dq, &bufs[rx_number]);
+		rx_number++;
+		/* Interpret 'dq' from a driver perspective. */
+		/*
+		 * Parking isn't possible unless HELDACTIVE was set. NB,
+		 * FORCEELIGIBLE implies HELDACTIVE, so we only need to
+		 * check for HELDACTIVE to cover both.
+		 */
+		DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) ||
+			    (res != qman_cb_dqrr_park));
+		qm_dqrr_cdc_consume_1ptr(&p->p, dq, res == qman_cb_dqrr_park);
+		/* Move forward */
+		qm_dqrr_next(&p->p);
+		/*
+		 * Entry processed and consumed, increment our counter.  The
+		 * callback can request that we exit after consuming the
+		 * entry, and we also exit if we reach our processing limit,
+		 * so loop back only if neither of these conditions is met.
+		 */
+	} while (likely(++limit < poll_limit));
+
+	return limit;
+}
+
 struct qm_dqrr_entry *qman_dequeue(struct qman_fq *fq)
 {
 	struct qman_portal *p = get_affine_portal();
diff --git a/drivers/bus/dpaa/base/qbman/qman.h b/drivers/bus/dpaa/base/qbman/qman.h
index a433369..4346d86 100644
--- a/drivers/bus/dpaa/base/qbman/qman.h
+++ b/drivers/bus/dpaa/base/qbman/qman.h
@@ -154,7 +154,7 @@ struct qm_eqcr {
 };
 
 struct qm_dqrr {
-	const struct qm_dqrr_entry *ring, *cursor;
+	struct qm_dqrr_entry *ring, *cursor;
 	u8 pi, ci, fill, ithresh, vbit;
 #ifdef RTE_LIBRTE_DPAA_HWDEBUG
 	enum qm_dqrr_dmode dmode;
@@ -441,7 +441,7 @@ static inline u8 DQRR_PTR2IDX(const struct qm_dqrr_entry *e)
 	return ((uintptr_t)e >> 6) & (QM_DQRR_SIZE - 1);
 }
 
-static inline const struct qm_dqrr_entry *DQRR_INC(
+static inline struct qm_dqrr_entry *DQRR_INC(
 						const struct qm_dqrr_entry *e)
 {
 	return DQRR_CARRYCLEAR(e + 1);
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index d769d50..ad40d80 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1124,6 +1124,12 @@ typedef enum qman_cb_dqrr_result (*qman_cb_dqrr)(struct qman_portal *qm,
 					struct qman_fq *fq,
 					const struct qm_dqrr_entry *dqrr);
 
+typedef enum qman_cb_dqrr_result (*qman_dpdk_cb_dqrr)(void *event,
+					struct qman_portal *qm,
+					struct qman_fq *fq,
+					const struct qm_dqrr_entry *dqrr,
+					void **bd);
+
 /*
  * This callback type is used when handling ERNs, FQRNs and FQRLs via MR. They
  * are always consumed after the callback returns.
@@ -1182,7 +1188,10 @@ enum qman_fq_state {
  */
 
 struct qman_fq_cb {
-	qman_cb_dqrr dqrr;	/* for dequeued frames */
+	union { /* for dequeued frames */
+		qman_dpdk_cb_dqrr dqrr_dpdk_cb;
+		qman_cb_dqrr dqrr;
+	};
 	qman_cb_mr ern;		/* for s/w ERNs */
 	qman_cb_mr fqs;		/* frame-queue state changes*/
 };
@@ -1299,6 +1308,9 @@ int qman_get_portal_index(void);
  */
 u16 qman_affine_channel(int cpu);
 
+unsigned int qman_portal_poll_rx(unsigned int poll_limit,
+				 void **bufs, struct qman_portal *q);
+
 /**
  * qman_set_vdq - Issue a volatile dequeue command
  * @fq: Frame Queue on which the volatile dequeue command is issued
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index 212c75f..ac455cd 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -70,11 +70,15 @@ DPDK_18.02 {
 
 	dpaa_svr_family;
 	qman_alloc_cgrid_range;
+	qman_alloc_pool_range;
 	qman_create_cgr;
 	qman_delete_cgr;
 	qman_modify_cgr;
+	qman_oos_fq;
+	qman_portal_poll_rx;
 	qman_query_fq_frm_cnt;
 	qman_release_cgrid_range;
+	qman_retire_fq;
 	rte_dpaa_portal_fq_close;
 	rte_dpaa_portal_fq_init;
 
-- 
2.7.4

[PATCH v3 18/19] net/dpaa: integrate the support of push mode in PMD

[Hemant Agrawal]

Signed-off-by: Sunil Kumar Kori <sunil.kori@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
---
 doc/guides/nics/dpaa.rst       | 11 ++++++++
 drivers/net/dpaa/dpaa_ethdev.c | 64 +++++++++++++++++++++++++++++++++++++-----
 drivers/net/dpaa/dpaa_ethdev.h |  2 +-
 drivers/net/dpaa/dpaa_rxtx.c   | 34 ++++++++++++++++++++++
 drivers/net/dpaa/dpaa_rxtx.h   |  5 ++++
 5 files changed, 108 insertions(+), 8 deletions(-)

diff --git a/doc/guides/nics/dpaa.rst b/doc/guides/nics/dpaa.rst
index a62f128..0a13996 100644
--- a/doc/guides/nics/dpaa.rst
+++ b/doc/guides/nics/dpaa.rst
@@ -290,6 +290,17 @@ state during application initialization:
   In case the application is configured to use lesser number of queues than
   configured above, it might result in packet loss (because of distribution).
 
+- ``DPAA_PUSH_QUEUES_NUMBER`` (default 4)
+
+  This defines the number of High performance queues to be used for ethdev Rx.
+  These queues use one private HW portal per queue configured, so they are
+  limited in the system. The first configured ethdev queues will be
+  automatically be assigned from the these high perf PUSH queues. Any queue
+  configuration beyond that will be standard Rx queues. The application can
+  choose to change their number if HW portals are limited.
+  The valid values are from '0' to '4'. The valuse shall be set to '0' if the
+  application want to use eventdev with DPAA device.
+
 
 Driver compilation and testing
 ------------------------------
diff --git a/drivers/net/dpaa/dpaa_ethdev.c b/drivers/net/dpaa/dpaa_ethdev.c
index 85ccea4..444c122 100644
--- a/drivers/net/dpaa/dpaa_ethdev.c
+++ b/drivers/net/dpaa/dpaa_ethdev.c
@@ -47,6 +47,14 @@
 
 /* Keep track of whether QMAN and BMAN have been globally initialized */
 static int is_global_init;
+/* At present we only allow up to 4 push mode queues - as each of this queue
+ * need dedicated portal and we are short of portals.
+ */
+#define DPAA_MAX_PUSH_MODE_QUEUE       4
+
+static int dpaa_push_mode_max_queue = DPAA_MAX_PUSH_MODE_QUEUE;
+static int dpaa_push_queue_idx; /* Queue index which are in push mode*/
+
 
 /* Per FQ Taildrop in frame count */
 static unsigned int td_threshold = CGR_RX_PERFQ_THRESH;
@@ -434,6 +442,9 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 {
 	struct dpaa_if *dpaa_intf = dev->data->dev_private;
 	struct qman_fq *rxq = &dpaa_intf->rx_queues[queue_idx];
+	struct qm_mcc_initfq opts = {0};
+	u32 flags = 0;
+	int ret;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -469,13 +480,45 @@ int dpaa_eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 			    dpaa_intf->name, fd_offset,
 			fman_if_get_fdoff(dpaa_intf->fif));
 	}
-
+	/* checking if push mode only, no error check for now */
+	if (dpaa_push_mode_max_queue > dpaa_push_queue_idx) {
+		dpaa_push_queue_idx++;
+		opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA;
+		opts.fqd.fq_ctrl = QM_FQCTRL_AVOIDBLOCK |
+				   QM_FQCTRL_CTXASTASHING |
+				   QM_FQCTRL_PREFERINCACHE;
+		opts.fqd.context_a.stashing.exclusive = 0;
+		opts.fqd.context_a.stashing.annotation_cl =
+						DPAA_IF_RX_ANNOTATION_STASH;
+		opts.fqd.context_a.stashing.data_cl = DPAA_IF_RX_DATA_STASH;
+		opts.fqd.context_a.stashing.context_cl =
+						DPAA_IF_RX_CONTEXT_STASH;
+
+		/*Create a channel and associate given queue with the channel*/
+		qman_alloc_pool_range((u32 *)&rxq->ch_id, 1, 1, 0);
+		opts.we_mask = opts.we_mask | QM_INITFQ_WE_DESTWQ;
+		opts.fqd.dest.channel = rxq->ch_id;
+		opts.fqd.dest.wq = DPAA_IF_RX_PRIORITY;
+		flags = QMAN_INITFQ_FLAG_SCHED;
+
+		/* Configure tail drop */
+		if (dpaa_intf->cgr_rx) {
+			opts.we_mask |= QM_INITFQ_WE_CGID;
+			opts.fqd.cgid = dpaa_intf->cgr_rx[queue_idx].cgrid;
+			opts.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+		}
+		ret = qman_init_fq(rxq, flags, &opts);
+		if (ret)
+			DPAA_PMD_ERR("Channel/Queue association failed. fqid %d"
+				     " ret: %d", rxq->fqid, ret);
+		rxq->cb.dqrr_dpdk_cb = dpaa_rx_cb;
+		rxq->is_static = true;
+	}
 	dev->data->rx_queues[queue_idx] = rxq;
 
 	/* configure the CGR size as per the desc size */
 	if (dpaa_intf->cgr_rx) {
 		struct qm_mcc_initcgr cgr_opts = {0};
-		int ret;
 
 		/* Enable tail drop with cgr on this queue */
 		qm_cgr_cs_thres_set64(&cgr_opts.cgr.cs_thres, nb_desc, 0);
@@ -809,11 +852,8 @@ static int dpaa_rx_queue_init(struct qman_fq *fq, struct qman_cgr *cgr_rx,
 			fqid, ret);
 		return ret;
 	}
-
-	opts.we_mask = QM_INITFQ_WE_DESTWQ | QM_INITFQ_WE_FQCTRL |
-		       QM_INITFQ_WE_CONTEXTA;
-
-	opts.fqd.dest.wq = DPAA_IF_RX_PRIORITY;
+	fq->is_static = false;
+	opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA;
 	opts.fqd.fq_ctrl = QM_FQCTRL_AVOIDBLOCK | QM_FQCTRL_CTXASTASHING |
 			   QM_FQCTRL_PREFERINCACHE;
 	opts.fqd.context_a.stashing.exclusive = 0;
@@ -947,6 +987,16 @@ dpaa_dev_init(struct rte_eth_dev *eth_dev)
 	else
 		num_rx_fqs = DPAA_DEFAULT_NUM_PCD_QUEUES;
 
+	/* if push mode queues to be enabled. Currenly we are allowing only
+	 * one queue per thread.
+	 */
+	if (getenv("DPAA_PUSH_QUEUES_NUMBER")) {
+		dpaa_push_mode_max_queue =
+				atoi(getenv("DPAA_PUSH_QUEUES_NUMBER"));
+		if (dpaa_push_mode_max_queue > DPAA_MAX_PUSH_MODE_QUEUE)
+			dpaa_push_mode_max_queue = DPAA_MAX_PUSH_MODE_QUEUE;
+	}
+
 	/* Each device can not have more than DPAA_PCD_FQID_MULTIPLIER RX
 	 * queues.
 	 */
diff --git a/drivers/net/dpaa/dpaa_ethdev.h b/drivers/net/dpaa/dpaa_ethdev.h
index 1b36567..1fa6caf 100644
--- a/drivers/net/dpaa/dpaa_ethdev.h
+++ b/drivers/net/dpaa/dpaa_ethdev.h
@@ -54,7 +54,7 @@
 #define DPAA_MAX_NUM_PCD_QUEUES		32
 
 #define DPAA_IF_TX_PRIORITY		3
-#define DPAA_IF_RX_PRIORITY		4
+#define DPAA_IF_RX_PRIORITY		0
 #define DPAA_IF_DEBUG_PRIORITY		7
 
 #define DPAA_IF_RX_ANNOTATION_STASH	1
diff --git a/drivers/net/dpaa/dpaa_rxtx.c b/drivers/net/dpaa/dpaa_rxtx.c
index 98671fa..0413932 100644
--- a/drivers/net/dpaa/dpaa_rxtx.c
+++ b/drivers/net/dpaa/dpaa_rxtx.c
@@ -394,6 +394,37 @@ dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 	return mbuf;
 }
 
+enum qman_cb_dqrr_result dpaa_rx_cb(void *event __always_unused,
+				    struct qman_portal *qm __always_unused,
+				    struct qman_fq *fq,
+				    const struct qm_dqrr_entry *dqrr,
+				    void **bufs)
+{
+	const struct qm_fd *fd = &dqrr->fd;
+
+	*bufs = dpaa_eth_fd_to_mbuf(fd,
+			((struct dpaa_if *)fq->dpaa_intf)->ifid);
+	return qman_cb_dqrr_consume;
+}
+
+static uint16_t
+dpaa_eth_queue_portal_rx(struct qman_fq *fq,
+			 struct rte_mbuf **bufs,
+			 uint16_t nb_bufs)
+{
+	int ret;
+
+	if (unlikely(fq->qp == NULL)) {
+		ret = rte_dpaa_portal_fq_init((void *)0, fq);
+		if (ret) {
+			DPAA_PMD_ERR("Failure in affining portal %d", ret);
+			return 0;
+		}
+	}
+
+	return qman_portal_poll_rx(nb_bufs, (void **)bufs, fq->qp);
+}
+
 uint16_t dpaa_eth_queue_rx(void *q,
 			   struct rte_mbuf **bufs,
 			   uint16_t nb_bufs)
@@ -403,6 +434,9 @@ uint16_t dpaa_eth_queue_rx(void *q,
 	uint32_t num_rx = 0, ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
 	int ret;
 
+	if (likely(fq->is_static))
+		return dpaa_eth_queue_portal_rx(fq, bufs, nb_bufs);
+
 	ret = rte_dpaa_portal_init((void *)0);
 	if (ret) {
 		DPAA_PMD_ERR("Failure in affining portal");
diff --git a/drivers/net/dpaa/dpaa_rxtx.h b/drivers/net/dpaa/dpaa_rxtx.h
index 78e804f..29d8f95 100644
--- a/drivers/net/dpaa/dpaa_rxtx.h
+++ b/drivers/net/dpaa/dpaa_rxtx.h
@@ -268,4 +268,9 @@ int dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 			   struct qm_fd *fd,
 			   uint32_t bpid);
 
+enum qman_cb_dqrr_result dpaa_rx_cb(void *event,
+				    struct qman_portal *qm,
+				    struct qman_fq *fq,
+				    const struct qm_dqrr_entry *dqrr,
+				    void **bd);
 #endif
-- 
2.7.4

[PATCH v3 19/19] bus/dpaa: support for enqueue frames of multiple queues

[Hemant Agrawal]

From: Akhil Goyal <akhil.goyal@nxp.com>

Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com>
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/dpaa/base/qbman/qman.c        | 66 +++++++++++++++++++++++++++++++
 drivers/bus/dpaa/include/fsl_qman.h       | 14 +++++++
 drivers/bus/dpaa/rte_bus_dpaa_version.map |  1 +
 3 files changed, 81 insertions(+)

diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 7e285a5..e171356 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -2158,6 +2158,72 @@ int qman_enqueue_multi(struct qman_fq *fq,
 	return sent;
 }
 
+int
+qman_enqueue_multi_fq(struct qman_fq *fq[], const struct qm_fd *fd,
+		      int frames_to_send)
+{
+	struct qman_portal *p = get_affine_portal();
+	struct qm_portal *portal = &p->p;
+
+	register struct qm_eqcr *eqcr = &portal->eqcr;
+	struct qm_eqcr_entry *eq = eqcr->cursor, *prev_eq;
+
+	u8 i, diff, old_ci, sent = 0;
+
+	/* Update the available entries if no entry is free */
+	if (!eqcr->available) {
+		old_ci = eqcr->ci;
+		eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1);
+		diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci);
+		eqcr->available += diff;
+		if (!diff)
+			return 0;
+	}
+
+	/* try to send as many frames as possible */
+	while (eqcr->available && frames_to_send--) {
+		eq->fqid = fq[sent]->fqid_le;
+		eq->fd.opaque_addr = fd->opaque_addr;
+		eq->fd.addr = cpu_to_be40(fd->addr);
+		eq->fd.status = cpu_to_be32(fd->status);
+		eq->fd.opaque = cpu_to_be32(fd->opaque);
+
+		eq = (void *)((unsigned long)(eq + 1) &
+			(~(unsigned long)(QM_EQCR_SIZE << 6)));
+		eqcr->available--;
+		sent++;
+		fd++;
+	}
+	lwsync();
+
+	/* In order for flushes to complete faster, all lines are recorded in
+	 * 32 bit word.
+	 */
+	eq = eqcr->cursor;
+	for (i = 0; i < sent; i++) {
+		eq->__dont_write_directly__verb =
+			QM_EQCR_VERB_CMD_ENQUEUE | eqcr->vbit;
+		prev_eq = eq;
+		eq = (void *)((unsigned long)(eq + 1) &
+			(~(unsigned long)(QM_EQCR_SIZE << 6)));
+		if (unlikely((prev_eq + 1) != eq))
+			eqcr->vbit ^= QM_EQCR_VERB_VBIT;
+	}
+
+	/* We need  to flush all the lines but without load/store operations
+	 * between them
+	 */
+	eq = eqcr->cursor;
+	for (i = 0; i < sent; i++) {
+		dcbf(eq);
+		eq = (void *)((unsigned long)(eq + 1) &
+			(~(unsigned long)(QM_EQCR_SIZE << 6)));
+	}
+	/* Update cursor for the next call */
+	eqcr->cursor = eq;
+	return sent;
+}
+
 int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags,
 		     struct qman_fq *orp, u16 orp_seqnum)
 {
diff --git a/drivers/bus/dpaa/include/fsl_qman.h b/drivers/bus/dpaa/include/fsl_qman.h
index ad40d80..0e3e4fe 100644
--- a/drivers/bus/dpaa/include/fsl_qman.h
+++ b/drivers/bus/dpaa/include/fsl_qman.h
@@ -1703,6 +1703,20 @@ int qman_enqueue_multi(struct qman_fq *fq,
 		       const struct qm_fd *fd,
 		int frames_to_send);
 
+/**
+ * qman_enqueue_multi_fq - Enqueue multiple frames to their respective frame
+ * queues.
+ * @fq[]: Array of frame queue objects to enqueue to
+ * @fd: pointer to first descriptor of frame to be enqueued
+ * @frames_to_send: number of frames to be sent.
+ *
+ * This API is similar to qman_enqueue_multi(), but it takes fd which needs
+ * to be processed by different frame queues.
+ */
+int
+qman_enqueue_multi_fq(struct qman_fq *fq[], const struct qm_fd *fd,
+		      int frames_to_send);
+
 typedef int (*qman_cb_precommit) (void *arg);
 
 /**
diff --git a/drivers/bus/dpaa/rte_bus_dpaa_version.map b/drivers/bus/dpaa/rte_bus_dpaa_version.map
index ac455cd..64068de 100644
--- a/drivers/bus/dpaa/rte_bus_dpaa_version.map
+++ b/drivers/bus/dpaa/rte_bus_dpaa_version.map
@@ -73,6 +73,7 @@ DPDK_18.02 {
 	qman_alloc_pool_range;
 	qman_create_cgr;
 	qman_delete_cgr;
+	qman_enqueue_multi_fq;
 	qman_modify_cgr;
 	qman_oos_fq;
 	qman_portal_poll_rx;
-- 
2.7.4

Re: [PATCH v3 00/19 ]DPAA PMD improvements

[Ferruh Yigit]

On 1/10/2018 10:46 AM, Hemant Agrawal wrote:
> This patch series add various improvement and performance related
> optimizations for DPAA PMD
> 
> v3:
>  - handling review comments from Ferruh
>  - update the API doc for new PMD specific API
> 
> v2:
>  - fix the spelling of PORTALS
>  - Add Akhil's patch wich is required for crypto
>  - minor improvement in push mode patch
> 
> Akhil Goyal (1):
>   bus/dpaa: support for enqueue frames of multiple queues
> 
> Ashish Jain (2):
>   net/dpaa: fix the mbuf packet type if zero
>   net/dpaa: set the correct frame size in device MTU
> 
> Hemant Agrawal (11):
>   net/dpaa: fix uninitialized and unused variables
>   net/dpaa: fix FW version code
>   bus/dpaa: update platform soc value register routines
>   net/dpaa: add frame count based tail drop with CGR
>   bus/dpaa: add support to create dynamic HW portal
>   bus/dpaa: query queue frame count support
>   net/dpaa: add Rx queue count support
>   net/dpaa: add support for loopback API
>   app/testpmd: add support for loopback config for dpaa
>   bus/dpaa: add support for static queues
>   net/dpaa: integrate the support of push mode in PMD
> 
> Nipun Gupta (5):
>   bus/dpaa: optimize the qman HW stashing settings
>   bus/dpaa: optimize the endianness conversions
>   net/dpaa: change Tx HW budget to 7
>   net/dpaa: optimize the Tx burst
>   net/dpaa: optimize Rx path

Series applied to dpdk-next-net/master, thanks.