[PATCH rdma-next 00/10] RDMA/bnxt_re: Add receive flow steering support

[PATCH rdma-next 00/10] RDMA/bnxt_re: Add receive flow steering support

[Kalesh AP]

Hi,

The RDMA stack allows for applications to create IB_QPT_RAW_PACKET
QPs, which receive plain Ethernet packets. This patch adds ib_create_flow()
and ib_destroy_flow() support in the bnxt_re driver. For now, only the
sniffer rule is supported to receive all port traffic. This is to support
tcpdump over the RDMA devices to capture the packets.

Patch#1 is Ethernet driver change to reserve more stats context to RDMA device.
Patch#2, #3 and #4 are code refactoring changes in preparation for subsequent patches.
Patch#5 adds support for unique GID.
Patch#6 adds support for mirror vnic.
Patch#7 adds support for flow create/destroy.
Patch#8 enables the feature by initializing FW with roce_mirror support.
Patch#9 is to improve the timeout value for the commands by using firmware provided message timeout value.
Patch#10 is another related cleanup patch to remove unnecessary checks.

This patch series is created on top of the below series posted on 08/14/2025:

[PATCH rdma-next 0/9] bnxt_re enhancements

Please review and apply.


Kalesh AP (3):
  RDMA/bnxt_re: Refactor hw context memory allocation
  RDMA/bnxt_re: Refactor stats context memory allocation
  RDMA/bnxt_re: Remove unnecessary condition checks

Saravanan Vajravel (7):
  bnxt_en: Enhance stats context reservation logic
  RDMA/bnxt_re: Add data structures for RoCE mirror support
  RDMA/bnxt_re: Add support for unique GID
  RDMA/bnxt_re: Add support for mirror vnic
  RDMA/bnxt_re: Add support for flow create/destroy
  RDMA/bnxt_re: Initialize fw with roce_mirror support
  RDMA/bnxt_re: Use firmware provided message timeout value

 drivers/infiniband/hw/bnxt_re/bnxt_re.h       |  13 ++
 drivers/infiniband/hw/bnxt_re/ib_verbs.c      | 146 +++++++++++-
 drivers/infiniband/hw/bnxt_re/ib_verbs.h      |  10 +
 drivers/infiniband/hw/bnxt_re/main.c          | 221 ++++++++++++++----
 drivers/infiniband/hw/bnxt_re/qplib_fp.c      |  12 +-
 drivers/infiniband/hw/bnxt_re/qplib_fp.h      |   1 +
 drivers/infiniband/hw/bnxt_re/qplib_rcfw.c    |   4 +
 drivers/infiniband/hw/bnxt_re/qplib_rcfw.h    |   1 +
 drivers/infiniband/hw/bnxt_re/qplib_res.c     |  38 +--
 drivers/infiniband/hw/bnxt_re/qplib_res.h     |  21 +-
 drivers/infiniband/hw/bnxt_re/qplib_sp.c      |  43 +++-
 drivers/infiniband/hw/bnxt_re/qplib_sp.h      |   5 +-
 drivers/infiniband/hw/bnxt_re/roce_hsi.h      |  41 +++-
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     |   8 +-
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |   3 +
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c |   6 +
 16 files changed, 486 insertions(+), 87 deletions(-)

-- 
2.43.5

[PATCH rdma-next 01/10] bnxt_en: Enhance stats context reservation logic

[Kalesh AP]

From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>

When the firmware advertises that the device is capable of supporting
port mirroring on RoCE device, reserve one additional stat_ctx.
To support port mirroring feature, RDMA driver allocates one stat_ctx
for exclusive use in RawEth QP.

Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 8 ++++++--
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     | 3 +++
 drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c | 6 ++++++
 3 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 5578ddcb465d..751840fff0dc 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -9601,10 +9601,10 @@ static int __bnxt_hwrm_ptp_qcfg(struct bnxt *bp)
 
 static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
 {
+	u32 flags, flags_ext, flags_ext2, flags_ext3;
+	struct bnxt_hw_resc *hw_resc = &bp->hw_resc;
 	struct hwrm_func_qcaps_output *resp;
 	struct hwrm_func_qcaps_input *req;
-	struct bnxt_hw_resc *hw_resc = &bp->hw_resc;
-	u32 flags, flags_ext, flags_ext2;
 	int rc;
 
 	rc = hwrm_req_init(bp, req, HWRM_FUNC_QCAPS);
@@ -9671,6 +9671,10 @@ static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
 	    (flags_ext2 & FUNC_QCAPS_RESP_FLAGS_EXT2_ROCE_VF_RESOURCE_MGMT_SUPPORTED))
 		bp->fw_cap |= BNXT_FW_CAP_ROCE_VF_RESC_MGMT_SUPPORTED;
 
+	flags_ext3 = le32_to_cpu(resp->flags_ext3);
+	if (flags_ext3 & FUNC_QCAPS_RESP_FLAGS_EXT3_MIRROR_ON_ROCE_SUPPORTED)
+		bp->fw_cap |= BNXT_FW_CAP_MIRROR_ON_ROCE;
+
 	bp->tx_push_thresh = 0;
 	if ((flags & FUNC_QCAPS_RESP_FLAGS_PUSH_MODE_SUPPORTED) &&
 	    BNXT_FW_MAJ(bp) > 217)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index fda0d3cc6227..fa2b39b55e68 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -2507,6 +2507,7 @@ struct bnxt {
 	#define BNXT_FW_CAP_VNIC_RE_FLUSH		BIT_ULL(40)
 	#define BNXT_FW_CAP_SW_MAX_RESOURCE_LIMITS	BIT_ULL(41)
 	#define BNXT_FW_CAP_NPAR_1_2			BIT_ULL(42)
+	#define BNXT_FW_CAP_MIRROR_ON_ROCE		BIT_ULL(43)
 
 	u32			fw_dbg_cap;
 
@@ -2528,6 +2529,8 @@ struct bnxt {
 	((bp)->fw_cap & BNXT_FW_CAP_ROCE_VF_RESC_MGMT_SUPPORTED)
 #define BNXT_SW_RES_LMT(bp)		\
 	((bp)->fw_cap & BNXT_FW_CAP_SW_MAX_RESOURCE_LIMITS)
+#define BNXT_MIRROR_ON_ROCE_CAP(bp)	\
+	((bp)->fw_cap & BNXT_FW_CAP_MIRROR_ON_ROCE)
 
 	u32			hwrm_spec_code;
 	u16			hwrm_cmd_seq;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
index 61cf201bb0dc..f8c2c72b382d 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ulp.c
@@ -100,6 +100,12 @@ void bnxt_set_dflt_ulp_stat_ctxs(struct bnxt *bp)
 		if (BNXT_PF(bp) && !bp->pf.port_id &&
 		    bp->port_count > 1)
 			bp->edev->ulp_num_ctxs++;
+
+		/* Reserve one additional stat_ctx when the device is capable
+		 * of supporting port mirroring on RDMA device.
+		 */
+		if (BNXT_MIRROR_ON_ROCE_CAP(bp))
+			bp->edev->ulp_num_ctxs++;
 	}
 }
 
-- 
2.43.5

[PATCH rdma-next 02/10] RDMA/bnxt_re: Add data structures for RoCE mirror support

[Kalesh AP]

From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>

Added data structures required for supporting mirroring on
RoCE device.

Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/bnxt_re.h    | 5 +++++
 drivers/infiniband/hw/bnxt_re/ib_verbs.h   | 5 +++++
 drivers/infiniband/hw/bnxt_re/qplib_fp.h   | 1 +
 drivers/infiniband/hw/bnxt_re/qplib_rcfw.h | 1 +
 drivers/infiniband/hw/bnxt_re/qplib_res.h  | 6 ++++++
 5 files changed, 18 insertions(+)

diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
index b5d0e38c7396..1cb57c8246cc 100644
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -172,6 +172,7 @@ struct bnxt_re_dev {
 	struct list_head		list;
 	unsigned long			flags;
 #define BNXT_RE_FLAG_NETDEV_REGISTERED		0
+#define BNXT_RE_FLAG_STATS_CTX3_ALLOC		1
 #define BNXT_RE_FLAG_HAVE_L2_REF		3
 #define BNXT_RE_FLAG_RCFW_CHANNEL_EN		4
 #define BNXT_RE_FLAG_QOS_WORK_REG		5
@@ -229,6 +230,10 @@ struct bnxt_re_dev {
 	struct bnxt_re_dbg_cc_config_params *cc_config_params;
 #define BNXT_VPD_FLD_LEN	32
 	char			board_partno[BNXT_VPD_FLD_LEN];
+	/* RoCE mirror */
+	u16			mirror_vnic_id;
+	union			ib_gid ugid;
+	u32			ugid_index;
 };
 
 #define to_bnxt_re_dev(ptr, member)	\
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.h b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
index fe00ab691a51..445a28b3cd96 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.h
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
@@ -164,6 +164,11 @@ struct bnxt_re_user_mmap_entry {
 	u8 mmap_flag;
 };
 
+struct bnxt_re_flow {
+	struct ib_flow		ib_flow;
+	struct bnxt_re_dev	*rdev;
+};
+
 static inline u16 bnxt_re_get_swqe_size(int nsge)
 {
 	return sizeof(struct sq_send_hdr) + nsge * sizeof(struct sq_sge);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.h b/drivers/infiniband/hw/bnxt_re/qplib_fp.h
index 074c539c69c1..3bd995ced9ca 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_fp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.h
@@ -345,6 +345,7 @@ struct bnxt_qplib_qp {
 	u32				msn_tbl_sz;
 	bool				is_host_msn_tbl;
 	u8				tos_dscp;
+	u32				ugid_index;
 };
 
 #define BNXT_RE_MAX_MSG_SIZE	0x80000000
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
index ff873c5f1b25..988c89b4232e 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
@@ -236,6 +236,7 @@ struct bnxt_qplib_rcfw {
 	atomic_t timeout_send;
 	/* cached from chip cctx for quick reference in slow path */
 	u16 max_timeout;
+	bool roce_mirror;
 };
 
 struct bnxt_qplib_cmdqmsg {
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.h b/drivers/infiniband/hw/bnxt_re/qplib_res.h
index 6a13927674b4..12e2fa23794a 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.h
@@ -65,6 +65,7 @@ struct bnxt_qplib_drv_modes {
 	bool db_push;
 	bool dbr_pacing;
 	u32 toggle_bits;
+	u8 roce_mirror;
 };
 
 enum bnxt_re_toggle_modes {
@@ -582,6 +583,11 @@ static inline u8 bnxt_qplib_dbr_pacing_en(struct bnxt_qplib_chip_ctx *cctx)
 	return cctx->modes.dbr_pacing;
 }
 
+static inline u8 bnxt_qplib_roce_mirror_supported(struct bnxt_qplib_chip_ctx *cctx)
+{
+	return cctx->modes.roce_mirror;
+}
+
 static inline bool _is_alloc_mr_unified(u16 dev_cap_flags)
 {
 	return dev_cap_flags & CREQ_QUERY_FUNC_RESP_SB_MR_REGISTER_ALLOC;
-- 
2.43.5

[PATCH rdma-next 03/10] RDMA/bnxt_re: Refactor hw context memory allocation

[Kalesh AP]

This patch is in preparation for other patches in this series.
There is no functional changes intended.

1. Rename bnxt_qplib_alloc_ctx() to bnxt_qplib_alloc_hwctx().
2. Rename bnxt_qplib_free_ctx() to bnxt_qplib_free_hwctx().
3. Reduce the number of arguments of bnxt_qplib_alloc_hwctx()
   by moving a check outside of it.

Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/main.c      | 18 ++++++++++--------
 drivers/infiniband/hw/bnxt_re/qplib_res.c | 17 ++++++-----------
 drivers/infiniband/hw/bnxt_re/qplib_res.h |  9 ++++-----
 3 files changed, 20 insertions(+), 24 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 3ae5f0d08f3a..9f752cb0c135 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -2011,7 +2011,7 @@ static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev, u8 op_type)
 			ibdev_warn(&rdev->ibdev,
 				   "Failed to deinitialize RCFW: %#x", rc);
 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
-		bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
+		bnxt_qplib_free_hwctx(&rdev->qplib_res, &rdev->qplib_ctx);
 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
@@ -2139,12 +2139,14 @@ static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 op_type)
 	bnxt_qplib_query_version(&rdev->rcfw);
 	bnxt_re_set_resource_limits(rdev);
 
-	rc = bnxt_qplib_alloc_ctx(&rdev->qplib_res, &rdev->qplib_ctx, 0,
-				  bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx));
-	if (rc) {
-		ibdev_err(&rdev->ibdev,
-			  "Failed to allocate QPLIB context: %#x\n", rc);
-		goto disable_rcfw;
+	if (!rdev->is_virtfn &&
+	    !bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx)) {
+		rc = bnxt_qplib_alloc_hwctx(&rdev->qplib_res, &rdev->qplib_ctx);
+		if (rc) {
+			ibdev_err(&rdev->ibdev,
+				  "Failed to allocate hw context: %#x\n", rc);
+			goto disable_rcfw;
+		}
 	}
 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
 					 rdev->qplib_ctx.stats.dma_map,
@@ -2206,7 +2208,7 @@ static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 op_type)
 free_sctx:
 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
 free_ctx:
-	bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
+	bnxt_qplib_free_hwctx(&rdev->qplib_res, &rdev->qplib_ctx);
 disable_rcfw:
 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
 free_ring:
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.c b/drivers/infiniband/hw/bnxt_re/qplib_res.c
index 6cd05207ffed..db2ee7246861 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.c
@@ -350,8 +350,8 @@ int bnxt_qplib_alloc_init_hwq(struct bnxt_qplib_hwq *hwq,
 }
 
 /* Context Tables */
-void bnxt_qplib_free_ctx(struct bnxt_qplib_res *res,
-			 struct bnxt_qplib_ctx *ctx)
+void bnxt_qplib_free_hwctx(struct bnxt_qplib_res *res,
+			   struct bnxt_qplib_ctx *ctx)
 {
 	int i;
 
@@ -464,7 +464,7 @@ static int bnxt_qplib_setup_tqm_rings(struct bnxt_qplib_res *res,
 }
 
 /*
- * Routine: bnxt_qplib_alloc_ctx
+ * Routine: bnxt_qplib_alloc_hwctx
  * Description:
  *     Context tables are memories which are used by the chip fw.
  *     The 6 tables defined are:
@@ -484,17 +484,13 @@ static int bnxt_qplib_setup_tqm_rings(struct bnxt_qplib_res *res,
  * Returns:
  *     0 if success, else -ERRORS
  */
-int bnxt_qplib_alloc_ctx(struct bnxt_qplib_res *res,
-			 struct bnxt_qplib_ctx *ctx,
-			 bool virt_fn, bool is_p5)
+int bnxt_qplib_alloc_hwctx(struct bnxt_qplib_res *res,
+			   struct bnxt_qplib_ctx *ctx)
 {
 	struct bnxt_qplib_hwq_attr hwq_attr = {};
 	struct bnxt_qplib_sg_info sginfo = {};
 	int rc;
 
-	if (virt_fn || is_p5)
-		goto stats_alloc;
-
 	/* QPC Tables */
 	sginfo.pgsize = PAGE_SIZE;
 	sginfo.pgshft = PAGE_SHIFT;
@@ -540,7 +536,6 @@ int bnxt_qplib_alloc_ctx(struct bnxt_qplib_res *res,
 	rc = bnxt_qplib_alloc_init_hwq(&ctx->tim_tbl, &hwq_attr);
 	if (rc)
 		goto fail;
-stats_alloc:
 	/* Stats */
 	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &ctx->stats);
 	if (rc)
@@ -549,7 +544,7 @@ int bnxt_qplib_alloc_ctx(struct bnxt_qplib_res *res,
 	return 0;
 
 fail:
-	bnxt_qplib_free_ctx(res, ctx);
+	bnxt_qplib_free_hwctx(res, ctx);
 	return rc;
 }
 
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.h b/drivers/infiniband/hw/bnxt_re/qplib_res.h
index 12e2fa23794a..9d866cfdebab 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.h
@@ -433,11 +433,10 @@ void bnxt_qplib_cleanup_res(struct bnxt_qplib_res *res);
 int bnxt_qplib_init_res(struct bnxt_qplib_res *res);
 void bnxt_qplib_free_res(struct bnxt_qplib_res *res);
 int bnxt_qplib_alloc_res(struct bnxt_qplib_res *res, struct net_device *netdev);
-void bnxt_qplib_free_ctx(struct bnxt_qplib_res *res,
-			 struct bnxt_qplib_ctx *ctx);
-int bnxt_qplib_alloc_ctx(struct bnxt_qplib_res *res,
-			 struct bnxt_qplib_ctx *ctx,
-			 bool virt_fn, bool is_p5);
+void bnxt_qplib_free_hwctx(struct bnxt_qplib_res *res,
+			   struct bnxt_qplib_ctx *ctx);
+int bnxt_qplib_alloc_hwctx(struct bnxt_qplib_res *res,
+			   struct bnxt_qplib_ctx *ctx);
 int bnxt_qplib_map_db_bar(struct bnxt_qplib_res *res);
 void bnxt_qplib_unmap_db_bar(struct bnxt_qplib_res *res);
 
-- 
2.43.5

[PATCH rdma-next 04/10] RDMA/bnxt_re: Refactor stats context memory allocation

[Kalesh AP]

Moved the stats context allocation logic to a new function.
The stats context memory allocation code has been moved from
bnxt_qplib_alloc_hwctx() to the newly added bnxt_re_get_stats_ctx()
function. Also, the code to send the firmware command has been moved.

This patch is in preparation for other patches in this series.
There is no functional changes intended.

Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/main.c      | 46 ++++++++++++++++++-----
 drivers/infiniband/hw/bnxt_re/qplib_res.c | 21 +++--------
 drivers/infiniband/hw/bnxt_re/qplib_res.h |  5 +++
 3 files changed, 47 insertions(+), 25 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 9f752cb0c135..c25eb2525a8f 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -935,8 +935,7 @@ static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
 }
 
 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
-				       dma_addr_t dma_map,
-				       u32 *fw_stats_ctx_id)
+				       struct bnxt_qplib_stats *stats)
 {
 	struct bnxt_qplib_chip_ctx *chip_ctx = rdev->chip_ctx;
 	struct hwrm_stat_ctx_alloc_output resp = {};
@@ -945,21 +944,21 @@ static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
 	struct bnxt_fw_msg fw_msg = {};
 	int rc = -EINVAL;
 
-	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
+	stats->fw_id = INVALID_STATS_CTX_ID;
 
 	if (!en_dev)
 		return rc;
 
 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_ALLOC);
 	req.update_period_ms = cpu_to_le32(1000);
-	req.stats_dma_addr = cpu_to_le64(dma_map);
+	req.stats_dma_addr = cpu_to_le64(stats->dma_map);
 	req.stats_dma_length = cpu_to_le16(chip_ctx->hw_stats_size);
 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (!rc)
-		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
+		stats->fw_id = le32_to_cpu(resp.stat_ctx_id);
 
 	return rc;
 }
@@ -1986,6 +1985,36 @@ static void bnxt_re_free_nqr_mem(struct bnxt_re_dev *rdev)
 	rdev->nqr = NULL;
 }
 
+static int bnxt_re_get_stats_ctx(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
+	struct bnxt_qplib_res *res = &rdev->qplib_res;
+	int rc;
+
+	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &hctx->stats);
+	if (rc)
+		return rc;
+
+	rc = bnxt_re_net_stats_ctx_alloc(rdev, &hctx->stats);
+	if (rc)
+		goto free_stat_mem;
+
+	return 0;
+free_stat_mem:
+	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats);
+
+	return rc;
+}
+
+static void bnxt_re_put_stats_ctx(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
+	struct bnxt_qplib_res *res = &rdev->qplib_res;
+
+	bnxt_re_net_stats_ctx_free(rdev, hctx->stats.fw_id);
+	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats);
+}
+
 static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev, u8 op_type)
 {
 	u8 type;
@@ -2010,7 +2039,7 @@ static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev, u8 op_type)
 		if (rc)
 			ibdev_warn(&rdev->ibdev,
 				   "Failed to deinitialize RCFW: %#x", rc);
-		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
+		bnxt_re_put_stats_ctx(rdev);
 		bnxt_qplib_free_hwctx(&rdev->qplib_res, &rdev->qplib_ctx);
 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
@@ -2148,9 +2177,8 @@ static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 op_type)
 			goto disable_rcfw;
 		}
 	}
-	rc = bnxt_re_net_stats_ctx_alloc(rdev,
-					 rdev->qplib_ctx.stats.dma_map,
-					 &rdev->qplib_ctx.stats.fw_id);
+
+	rc = bnxt_re_get_stats_ctx(rdev);
 	if (rc) {
 		ibdev_err(&rdev->ibdev,
 			  "Failed to allocate stats context: %#x\n", rc);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.c b/drivers/infiniband/hw/bnxt_re/qplib_res.c
index db2ee7246861..8d04f98b2606 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.c
@@ -53,12 +53,6 @@
 #include "qplib_sp.h"
 #include "qplib_rcfw.h"
 
-static void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
-				      struct bnxt_qplib_stats *stats);
-static int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
-				      struct bnxt_qplib_chip_ctx *cctx,
-				      struct bnxt_qplib_stats *stats);
-
 /* PBL */
 static void __free_pbl(struct bnxt_qplib_res *res, struct bnxt_qplib_pbl *pbl,
 		       bool is_umem)
@@ -365,7 +359,6 @@ void bnxt_qplib_free_hwctx(struct bnxt_qplib_res *res,
 	/* restore original pde level before destroy */
 	ctx->tqm_ctx.pde.level = ctx->tqm_ctx.pde_level;
 	bnxt_qplib_free_hwq(res, &ctx->tqm_ctx.pde);
-	bnxt_qplib_free_stats_ctx(res->pdev, &ctx->stats);
 }
 
 static int bnxt_qplib_alloc_tqm_rings(struct bnxt_qplib_res *res,
@@ -534,10 +527,6 @@ int bnxt_qplib_alloc_hwctx(struct bnxt_qplib_res *res,
 	hwq_attr.depth = ctx->qpc_count * 16;
 	hwq_attr.stride = 1;
 	rc = bnxt_qplib_alloc_init_hwq(&ctx->tim_tbl, &hwq_attr);
-	if (rc)
-		goto fail;
-	/* Stats */
-	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &ctx->stats);
 	if (rc)
 		goto fail;
 
@@ -825,8 +814,8 @@ static int bnxt_qplib_alloc_dpi_tbl(struct bnxt_qplib_res *res,
 }
 
 /* Stats */
-static void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
-				      struct bnxt_qplib_stats *stats)
+void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
+			       struct bnxt_qplib_stats *stats)
 {
 	if (stats->dma) {
 		dma_free_coherent(&pdev->dev, stats->size,
@@ -836,9 +825,9 @@ static void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
 	stats->fw_id = -1;
 }
 
-static int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
-				      struct bnxt_qplib_chip_ctx *cctx,
-				      struct bnxt_qplib_stats *stats)
+int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
+			       struct bnxt_qplib_chip_ctx *cctx,
+			       struct bnxt_qplib_stats *stats)
 {
 	memset(stats, 0, sizeof(*stats));
 	stats->fw_id = -1;
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.h b/drivers/infiniband/hw/bnxt_re/qplib_res.h
index 9d866cfdebab..ed1be06c2c60 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.h
@@ -441,6 +441,11 @@ int bnxt_qplib_map_db_bar(struct bnxt_qplib_res *res);
 void bnxt_qplib_unmap_db_bar(struct bnxt_qplib_res *res);
 
 int bnxt_qplib_determine_atomics(struct pci_dev *dev);
+int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
+			       struct bnxt_qplib_chip_ctx *cctx,
+			       struct bnxt_qplib_stats *stats);
+void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
+			       struct bnxt_qplib_stats *stats);
 
 static inline void bnxt_qplib_hwq_incr_prod(struct bnxt_qplib_db_info *dbinfo,
 					    struct bnxt_qplib_hwq *hwq, u32 cnt)
-- 
2.43.5

[PATCH rdma-next 05/10] RDMA/bnxt_re: Add support for unique GID

[Kalesh AP]

From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>

- RawEth QP requires unique GID so that per function stats_ctx
  is not polluted by packets mirrored to RoCE vnic.
- Added support to add unique GID when RawEth type QP is created.
- Added support to destroy unique GID when RawEth type QP is
  destroyed.
- Allocated exclusive stats_ctx to use for RawEth type QP.

Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/ib_verbs.c  | 57 ++++++++++++++++++++++-
 drivers/infiniband/hw/bnxt_re/main.c      | 42 +++++++++++++++++
 drivers/infiniband/hw/bnxt_re/qplib_res.h |  1 +
 drivers/infiniband/hw/bnxt_re/qplib_sp.c  |  6 ++-
 drivers/infiniband/hw/bnxt_re/qplib_sp.h  |  3 +-
 5 files changed, 105 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index 195a9ba6f65d..c83809c72f5b 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -288,7 +288,9 @@ int bnxt_re_query_port(struct ib_device *ibdev, u32 port_num,
 	}
 	port_attr->max_mtu = IB_MTU_4096;
 	port_attr->active_mtu = iboe_get_mtu(rdev->netdev->mtu);
-	port_attr->gid_tbl_len = dev_attr->max_sgid;
+	/* One GID is reserved for RawEth QP. Report one less */
+	port_attr->gid_tbl_len = (rdev->rcfw.roce_mirror ? (dev_attr->max_sgid - 1) :
+				  dev_attr->max_sgid);
 	port_attr->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
 				    IB_PORT_DEVICE_MGMT_SUP |
 				    IB_PORT_VENDOR_CLASS_SUP;
@@ -429,7 +431,7 @@ int bnxt_re_add_gid(const struct ib_gid_attr *attr, void **context)
 
 	rc = bnxt_qplib_add_sgid(sgid_tbl, (struct bnxt_qplib_gid *)&attr->gid,
 				 rdev->qplib_res.netdev->dev_addr,
-				 vlan_id, true, &tbl_idx);
+				 vlan_id, true, &tbl_idx, false, 0);
 	if (rc == -EALREADY) {
 		ctx_tbl = sgid_tbl->ctx;
 		ctx_tbl[tbl_idx]->refcnt++;
@@ -955,6 +957,20 @@ static int bnxt_re_destroy_gsi_sqp(struct bnxt_re_qp *qp)
 	return rc;
 }
 
+static void bnxt_re_del_unique_gid(struct bnxt_re_dev *rdev)
+{
+	int rc;
+
+	if (!rdev->rcfw.roce_mirror)
+		return;
+
+	rc = bnxt_qplib_del_sgid(&rdev->qplib_res.sgid_tbl,
+				 (struct bnxt_qplib_gid *)&rdev->ugid,
+				 0xFFFF, true);
+	if (rc)
+		dev_err(rdev_to_dev(rdev), "Failed to delete unique GID, rc: %d\n", rc);
+}
+
 /* Queue Pairs */
 int bnxt_re_destroy_qp(struct ib_qp *ib_qp, struct ib_udata *udata)
 {
@@ -994,6 +1010,9 @@ int bnxt_re_destroy_qp(struct ib_qp *ib_qp, struct ib_udata *udata)
 	else if (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD)
 		atomic_dec(&rdev->stats.res.ud_qp_count);
 
+	if (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_RAW_ETHERTYPE)
+		bnxt_re_del_unique_gid(rdev);
+
 	ib_umem_release(qp->rumem);
 	ib_umem_release(qp->sumem);
 
@@ -1595,6 +1614,29 @@ static bool bnxt_re_test_qp_limits(struct bnxt_re_dev *rdev,
 	return rc;
 }
 
+static int bnxt_re_add_unique_gid(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
+	struct bnxt_qplib_res *res = &rdev->qplib_res;
+	int rc;
+
+	if (!rdev->rcfw.roce_mirror)
+		return 0;
+
+	rdev->ugid.global.subnet_prefix = cpu_to_be64(0xfe8000000000abcdLL);
+	addrconf_ifid_eui48(&rdev->ugid.raw[8], rdev->netdev);
+
+	rc = bnxt_qplib_add_sgid(&res->sgid_tbl,
+				 (struct bnxt_qplib_gid *)&rdev->ugid,
+				 rdev->qplib_res.netdev->dev_addr,
+				 0xFFFF, true, &rdev->ugid_index, true,
+				 hctx->stats3.fw_id);
+	if (rc)
+		dev_err(rdev_to_dev(rdev), "Failed to add unique GID. rc = %d\n", rc);
+
+	return rc;
+}
+
 int bnxt_re_create_qp(struct ib_qp *ib_qp, struct ib_qp_init_attr *qp_init_attr,
 		      struct ib_udata *udata)
 {
@@ -1656,6 +1698,17 @@ int bnxt_re_create_qp(struct ib_qp *ib_qp, struct ib_qp_init_attr *qp_init_attr,
 		}
 	}
 
+	/* Support for RawEth QP is added to capture TCP pkt dump.
+	 * So unique SGID is used to avoid incorrect statistics on per
+	 * function stats_ctx
+	 */
+	if (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_RAW_ETHERTYPE) {
+		rc = bnxt_re_add_unique_gid(rdev);
+		if (rc)
+			goto qp_destroy;
+		qp->qplib_qp.ugid_index = rdev->ugid_index;
+	}
+
 	qp->ib_qp.qp_num = qp->qplib_qp.id;
 	if (qp_init_attr->qp_type == IB_QPT_GSI)
 		rdev->gsi_ctx.gsi_qp = qp;
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index c25eb2525a8f..479c2a390885 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -2006,6 +2006,42 @@ static int bnxt_re_get_stats_ctx(struct bnxt_re_dev *rdev)
 	return rc;
 }
 
+static int bnxt_re_get_stats3_ctx(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
+	struct bnxt_qplib_res *res = &rdev->qplib_res;
+	int rc;
+
+	if (!rdev->rcfw.roce_mirror)
+		return 0;
+
+	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &hctx->stats3);
+	if (rc)
+		return rc;
+
+	rc = bnxt_re_net_stats_ctx_alloc(rdev, &hctx->stats3);
+	if (rc)
+		goto free_stat_mem;
+
+	return 0;
+free_stat_mem:
+	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats3);
+
+	return rc;
+}
+
+static void bnxt_re_put_stats3_ctx(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
+	struct bnxt_qplib_res *res = &rdev->qplib_res;
+
+	if (!rdev->rcfw.roce_mirror)
+		return;
+
+	bnxt_re_net_stats_ctx_free(rdev, hctx->stats3.fw_id);
+	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats3);
+}
+
 static void bnxt_re_put_stats_ctx(struct bnxt_re_dev *rdev)
 {
 	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
@@ -2028,6 +2064,8 @@ static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev, u8 op_type)
 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
 		cancel_delayed_work_sync(&rdev->worker);
 
+	bnxt_re_put_stats3_ctx(rdev);
+
 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
 			       &rdev->flags))
 		bnxt_re_cleanup_res(rdev);
@@ -2232,6 +2270,10 @@ static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 op_type)
 	if (!rdev->is_virtfn)
 		bnxt_re_read_vpd_info(rdev);
 
+	rc = bnxt_re_get_stats3_ctx(rdev);
+	if (rc)
+		goto fail;
+
 	return 0;
 free_sctx:
 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.h b/drivers/infiniband/hw/bnxt_re/qplib_res.h
index ed1be06c2c60..2ea3b7f232a3 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_res.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_res.h
@@ -304,6 +304,7 @@ struct bnxt_qplib_ctx {
 	struct bnxt_qplib_hwq		tim_tbl;
 	struct bnxt_qplib_tqm_ctx	tqm_ctx;
 	struct bnxt_qplib_stats		stats;
+	struct bnxt_qplib_stats		stats3;
 	struct bnxt_qplib_vf_res	vf_res;
 };
 
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
index 79edff6bda95..d10741151543 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -308,7 +308,8 @@ int bnxt_qplib_del_sgid(struct bnxt_qplib_sgid_tbl *sgid_tbl,
 
 int bnxt_qplib_add_sgid(struct bnxt_qplib_sgid_tbl *sgid_tbl,
 			struct bnxt_qplib_gid *gid, const u8 *smac,
-			u16 vlan_id, bool update, u32 *index)
+			u16 vlan_id, bool update, u32 *index,
+			bool is_ugid, u32 stats_ctx_id)
 {
 	struct bnxt_qplib_res *res = to_bnxt_qplib(sgid_tbl,
 						   struct bnxt_qplib_res,
@@ -373,6 +374,9 @@ int bnxt_qplib_add_sgid(struct bnxt_qplib_sgid_tbl *sgid_tbl,
 		req.src_mac[1] = cpu_to_be16(((u16 *)smac)[1]);
 		req.src_mac[2] = cpu_to_be16(((u16 *)smac)[2]);
 
+		req.stats_ctx = cpu_to_le16(CMDQ_ADD_GID_STATS_CTX_STATS_CTX_VALID |
+					    (u16)stats_ctx_id);
+
 		bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
 					sizeof(resp), 0);
 		rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.h b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
index e9834e7fc383..58f90f3e57f7 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
@@ -323,7 +323,8 @@ int bnxt_qplib_del_sgid(struct bnxt_qplib_sgid_tbl *sgid_tbl,
 			struct bnxt_qplib_gid *gid, u16 vlan_id, bool update);
 int bnxt_qplib_add_sgid(struct bnxt_qplib_sgid_tbl *sgid_tbl,
 			struct bnxt_qplib_gid *gid, const u8 *mac, u16 vlan_id,
-			bool update, u32 *index);
+			bool update, u32 *index,
+			bool is_ugid, u32 stats_ctx_id);
 int bnxt_qplib_update_sgid(struct bnxt_qplib_sgid_tbl *sgid_tbl,
 			   struct bnxt_qplib_gid *gid, u16 gid_idx,
 			   const u8 *smac);
-- 
2.43.5

[PATCH rdma-next 06/10] RDMA/bnxt_re: Add support for mirror vnic

[Kalesh AP]

From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>

Added below support:
- Querying the pre-reserved mirror_vnic_id
- Allocating/freeing mirror_vnic
- Configuring mirror vnic to associate it with raw qp

These functions will be used in the subsequent patch in this series.

Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/bnxt_re.h |  4 ++
 drivers/infiniband/hw/bnxt_re/main.c    | 67 +++++++++++++++++++++++++
 2 files changed, 71 insertions(+)

diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
index 1cb57c8246cc..3f7f6cefe771 100644
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -250,6 +250,10 @@ int bnxt_re_assign_pma_port_counters(struct bnxt_re_dev *rdev, struct ib_mad *ou
 int bnxt_re_assign_pma_port_ext_counters(struct bnxt_re_dev *rdev,
 					 struct ib_mad *out_mad);
 
+void bnxt_re_hwrm_free_vnic(struct bnxt_re_dev *rdev);
+int bnxt_re_hwrm_alloc_vnic(struct bnxt_re_dev *rdev);
+int bnxt_re_hwrm_cfg_vnic(struct bnxt_re_dev *rdev, u32 qp_id);
+
 static inline struct device *rdev_to_dev(struct bnxt_re_dev *rdev)
 {
 	if (rdev)
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 479c2a390885..07d25deffabe 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -540,6 +540,72 @@ static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
 	fw_msg->timeout = timeout;
 }
 
+void bnxt_re_hwrm_free_vnic(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_en_dev *en_dev = rdev->en_dev;
+	struct hwrm_vnic_free_input req = {};
+	struct bnxt_fw_msg fw_msg = {};
+	int rc;
+
+	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VNIC_FREE);
+
+	req.vnic_id = cpu_to_le32(rdev->mirror_vnic_id);
+	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), NULL,
+			    0, DFLT_HWRM_CMD_TIMEOUT);
+	rc = bnxt_send_msg(en_dev, &fw_msg);
+	if (rc)
+		ibdev_dbg(&rdev->ibdev,
+			  "Failed to free vnic, rc = %d\n", rc);
+}
+
+int bnxt_re_hwrm_alloc_vnic(struct bnxt_re_dev *rdev)
+{
+	struct bnxt_en_dev *en_dev = rdev->en_dev;
+	struct hwrm_vnic_alloc_output resp = {};
+	struct hwrm_vnic_alloc_input req = {};
+	struct bnxt_fw_msg fw_msg = {};
+	int rc;
+
+	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VNIC_ALLOC);
+
+	req.vnic_id = cpu_to_le16(rdev->mirror_vnic_id);
+	req.flags = cpu_to_le32(VNIC_ALLOC_REQ_FLAGS_VNIC_ID_VALID);
+	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
+			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+	rc = bnxt_send_msg(en_dev, &fw_msg);
+	if (rc)
+		ibdev_dbg(&rdev->ibdev,
+			  "Failed to alloc vnic, rc = %d\n", rc);
+
+	return rc;
+}
+
+int bnxt_re_hwrm_cfg_vnic(struct bnxt_re_dev *rdev, u32 qp_id)
+{
+	struct bnxt_en_dev *en_dev = rdev->en_dev;
+	struct hwrm_vnic_cfg_input req = {};
+	struct bnxt_fw_msg fw_msg = {};
+	int rc;
+
+	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VNIC_CFG);
+
+	req.flags = cpu_to_le32(VNIC_CFG_REQ_FLAGS_ROCE_ONLY_VNIC_MODE);
+	req.enables = cpu_to_le32(VNIC_CFG_REQ_ENABLES_RAW_QP_ID |
+				  VNIC_CFG_REQ_ENABLES_MRU);
+	req.vnic_id = cpu_to_le16(rdev->mirror_vnic_id);
+	req.raw_qp_id = cpu_to_le32(qp_id);
+	req.mru = cpu_to_le16(rdev->netdev->mtu + VLAN_ETH_HLEN);
+
+	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), NULL,
+			    0, DFLT_HWRM_CMD_TIMEOUT);
+	rc = bnxt_send_msg(en_dev, &fw_msg);
+	if (rc)
+		ibdev_dbg(&rdev->ibdev,
+			  "Failed to cfg vnic, rc = %d\n", rc);
+
+	return rc;
+}
+
 /* Query device config using common hwrm */
 static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
 			     u32 *offset)
@@ -558,6 +624,7 @@ static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
 	if (!rc) {
 		*db_len = PAGE_ALIGN(le16_to_cpu(resp.l2_doorbell_bar_size_kb) * 1024);
 		*offset = PAGE_ALIGN(le16_to_cpu(resp.legacy_l2_db_size_kb) * 1024);
+		rdev->mirror_vnic_id = le16_to_cpu(resp.mirror_vnic_id);
 	}
 	return rc;
 }
-- 
2.43.5

[PATCH rdma-next 07/10] RDMA/bnxt_re: Add support for flow create/destroy

[Kalesh AP]

From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>

- Added support for create_flow and destroy_flow verbs. These
  verbs are used on RawEth QP to add a specific flow action.
- To support TCP dump on RoCE, added IB_FLOW_ATTR_SNIFFER
  attribute.
- In create_flow verb, driver allocates mirror_vnic and configure it
  with RawEth QP. Once this is done, driver will enable mirroring.
- In destroy_flow, driver will disable mirroring and free the mirror
  vnic.

Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/bnxt_re.h  |  1 +
 drivers/infiniband/hw/bnxt_re/ib_verbs.c | 87 ++++++++++++++++++++++++
 drivers/infiniband/hw/bnxt_re/ib_verbs.h |  5 ++
 drivers/infiniband/hw/bnxt_re/main.c     |  2 +
 drivers/infiniband/hw/bnxt_re/qplib_sp.c | 37 ++++++++++
 drivers/infiniband/hw/bnxt_re/qplib_sp.h |  2 +
 drivers/infiniband/hw/bnxt_re/roce_hsi.h | 40 ++++++++++-
 7 files changed, 173 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
index 3f7f6cefe771..3a219d67746c 100644
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -234,6 +234,7 @@ struct bnxt_re_dev {
 	u16			mirror_vnic_id;
 	union			ib_gid ugid;
 	u32			ugid_index;
+	u8			sniffer_flow_created:1;
 };
 
 #define to_bnxt_re_dev(ptr, member)	\
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index c83809c72f5b..90c23d0ee262 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -4449,6 +4449,93 @@ void bnxt_re_dealloc_ucontext(struct ib_ucontext *ib_uctx)
 	}
 }
 
+static int bnxt_re_setup_vnic(struct bnxt_re_dev *rdev, struct bnxt_re_qp *qp)
+{
+	int rc;
+
+	rc = bnxt_re_hwrm_alloc_vnic(rdev);
+	if (rc)
+		return rc;
+
+	rc = bnxt_re_hwrm_cfg_vnic(rdev, qp->qplib_qp.id);
+	if (rc)
+		goto out_free_vnic;
+
+	return 0;
+out_free_vnic:
+	bnxt_re_hwrm_free_vnic(rdev);
+	return rc;
+}
+
+struct ib_flow *bnxt_re_create_flow(struct ib_qp *ib_qp,
+				    struct ib_flow_attr *attr,
+				    struct ib_udata *udata)
+{
+	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
+	struct bnxt_re_dev *rdev = qp->rdev;
+	struct bnxt_re_flow *flow;
+	int rc;
+
+	if (attr->type != IB_FLOW_ATTR_SNIFFER ||
+	    !rdev->rcfw.roce_mirror)
+		return ERR_PTR(-EOPNOTSUPP);
+
+	mutex_lock(&rdev->qp_lock);
+	if (rdev->sniffer_flow_created) {
+		ibdev_err(&rdev->ibdev, "RoCE Mirroring is already Configured\n");
+		mutex_unlock(&rdev->qp_lock);
+		return ERR_PTR(-EBUSY);
+	}
+
+	flow = kzalloc(sizeof(*flow), GFP_KERNEL);
+	if (!flow) {
+		mutex_unlock(&rdev->qp_lock);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	flow->rdev = rdev;
+
+	rc = bnxt_re_setup_vnic(rdev, qp);
+	if (rc)
+		goto out_free_flow;
+
+	rc = bnxt_qplib_create_flow(&rdev->qplib_res);
+	if (rc)
+		goto out_free_vnic;
+
+	rdev->sniffer_flow_created = 1;
+	mutex_unlock(&rdev->qp_lock);
+
+	return &flow->ib_flow;
+
+out_free_vnic:
+	bnxt_re_hwrm_free_vnic(rdev);
+out_free_flow:
+	mutex_unlock(&rdev->qp_lock);
+	kfree(flow);
+	return ERR_PTR(rc);
+}
+
+int bnxt_re_destroy_flow(struct ib_flow *flow_id)
+{
+	struct bnxt_re_flow *flow =
+		container_of(flow_id, struct bnxt_re_flow, ib_flow);
+	struct bnxt_re_dev *rdev = flow->rdev;
+	int rc;
+
+	mutex_lock(&rdev->qp_lock);
+	rc = bnxt_qplib_destroy_flow(&rdev->qplib_res);
+	if (rc)
+		ibdev_dbg(&rdev->ibdev, "failed to destroy_flow rc = %d\n", rc);
+	rdev->sniffer_flow_created = 0;
+
+	bnxt_re_hwrm_free_vnic(rdev);
+	mutex_unlock(&rdev->qp_lock);
+	kfree(flow);
+
+	return rc;
+}
+
 static struct bnxt_re_cq *bnxt_re_search_for_cq(struct bnxt_re_dev *rdev, u32 cq_id)
 {
 	struct bnxt_re_cq *cq = NULL, *tmp_cq;
diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.h b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
index 445a28b3cd96..76ba9ab04d5c 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.h
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.h
@@ -272,6 +272,11 @@ struct ib_mr *bnxt_re_reg_user_mr_dmabuf(struct ib_pd *ib_pd, u64 start,
 					 struct uverbs_attr_bundle *attrs);
 int bnxt_re_alloc_ucontext(struct ib_ucontext *ctx, struct ib_udata *udata);
 void bnxt_re_dealloc_ucontext(struct ib_ucontext *context);
+struct ib_flow *bnxt_re_create_flow(struct ib_qp *ib_qp,
+				    struct ib_flow_attr *attr,
+				    struct ib_udata *udata);
+int bnxt_re_destroy_flow(struct ib_flow *flow_id);
+
 int bnxt_re_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
 void bnxt_re_mmap_free(struct rdma_user_mmap_entry *rdma_entry);
 
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 07d25deffabe..fe1be036f8f2 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -1383,6 +1383,8 @@ static const struct ib_device_ops bnxt_re_dev_ops = {
 	.reg_user_mr_dmabuf = bnxt_re_reg_user_mr_dmabuf,
 	.req_notify_cq = bnxt_re_req_notify_cq,
 	.resize_cq = bnxt_re_resize_cq,
+	.create_flow = bnxt_re_create_flow,
+	.destroy_flow = bnxt_re_destroy_flow,
 	INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
 	INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
index d10741151543..9ef581ed785c 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c
@@ -1106,3 +1106,40 @@ int bnxt_qplib_query_cc_param(struct bnxt_qplib_res *res,
 	dma_free_coherent(&rcfw->pdev->dev, sbuf.size, sbuf.sb, sbuf.dma_addr);
 	return rc;
 }
+
+int bnxt_qplib_create_flow(struct bnxt_qplib_res *res)
+{
+	struct creq_roce_mirror_cfg_resp resp = {};
+	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
+	struct cmdq_roce_mirror_cfg req = {};
+	struct bnxt_qplib_cmdqmsg msg = {};
+
+	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
+				 CMDQ_BASE_OPCODE_ROCE_MIRROR_CFG,
+				 sizeof(req));
+
+	req.mirror_flags = (u8)CMDQ_ROCE_MIRROR_CFG_MIRROR_ENABLE;
+
+	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
+				sizeof(resp), 0);
+	return bnxt_qplib_rcfw_send_message(rcfw, &msg);
+}
+
+int bnxt_qplib_destroy_flow(struct bnxt_qplib_res *res)
+{
+	struct creq_roce_mirror_cfg_resp resp = {};
+	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
+	struct cmdq_roce_mirror_cfg req = {};
+	struct bnxt_qplib_cmdqmsg msg = {};
+
+	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
+				 CMDQ_BASE_OPCODE_ROCE_MIRROR_CFG,
+				 sizeof(req));
+
+	req.mirror_flags &= ~((u8)CMDQ_ROCE_MIRROR_CFG_MIRROR_ENABLE);
+
+	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
+				sizeof(resp), 0);
+
+	return bnxt_qplib_rcfw_send_message(rcfw, &msg);
+}
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.h b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
index 58f90f3e57f7..147b5d9c0313 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_sp.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.h
@@ -360,6 +360,8 @@ int bnxt_qplib_read_context(struct bnxt_qplib_rcfw *rcfw, u8 type, u32 xid,
 int bnxt_qplib_query_cc_param(struct bnxt_qplib_res *res,
 			      struct bnxt_qplib_cc_param *cc_param);
 void bnxt_qplib_query_version(struct bnxt_qplib_rcfw *rcfw);
+int bnxt_qplib_create_flow(struct bnxt_qplib_res *res);
+int bnxt_qplib_destroy_flow(struct bnxt_qplib_res *res);
 
 #define BNXT_VAR_MAX_WQE       4352
 #define BNXT_VAR_MAX_SLOT_ALIGN 256
diff --git a/drivers/infiniband/hw/bnxt_re/roce_hsi.h b/drivers/infiniband/hw/bnxt_re/roce_hsi.h
index f9ac37335a1d..cfdf69a3fe9a 100644
--- a/drivers/infiniband/hw/bnxt_re/roce_hsi.h
+++ b/drivers/infiniband/hw/bnxt_re/roce_hsi.h
@@ -144,7 +144,8 @@ struct cmdq_base {
 	#define CMDQ_BASE_OPCODE_MODIFY_CQ              0x90UL
 	#define CMDQ_BASE_OPCODE_QUERY_QP_EXTEND        0x91UL
 	#define CMDQ_BASE_OPCODE_QUERY_ROCE_STATS_EXT   0x92UL
-	#define CMDQ_BASE_OPCODE_LAST                  CMDQ_BASE_OPCODE_QUERY_ROCE_STATS_EXT
+	#define CMDQ_BASE_OPCODE_ROCE_MIRROR_CFG	0x99UL
+	#define CMDQ_BASE_OPCODE_LAST                   CMDQ_BASE_OPCODE_ROCE_MIRROR_CFG
 	u8	cmd_size;
 	__le16	flags;
 	__le16	cookie;
@@ -2109,6 +2110,43 @@ struct creq_query_roce_stats_ext_resp_sb {
 	__le64	dup_req;
 };
 
+/* cmdq_roce_mirror_cfg (size:192b/24B) */
+struct cmdq_roce_mirror_cfg {
+	u8      opcode;
+	#define CMDQ_ROCE_MIRROR_CFG_OPCODE_ROCE_MIRROR_CFG	0x99UL
+	#define CMDQ_ROCE_MIRROR_CFG_OPCODE_LAST		\
+				CMDQ_ROCE_MIRROR_CFG_OPCODE_ROCE_MIRROR_CFG
+	u8      cmd_size;
+	__le16  flags;
+	__le16  cookie;
+	u8      resp_size;
+	u8      reserved8;
+	__le64  resp_addr;
+	u8      mirror_flags;
+	#define CMDQ_ROCE_MIRROR_CFG_MIRROR_ENABLE		0x1UL
+	u8      rsvd[7];
+};
+
+/* creq_roce_mirror_cfg_resp (size:128b/16B) */
+struct creq_roce_mirror_cfg_resp {
+	u8      type;
+	#define CREQ_ROCE_MIRROR_CFG_RESP_TYPE_MASK	0x3fUL
+	#define CREQ_ROCE_MIRROR_CFG_RESP_TYPE_SFT	0
+	#define CREQ_ROCE_MIRROR_CFG_RESP_TYPE_QP_EVENT	0x38UL
+	#define CREQ_ROCE_MIRROR_CFG_RESP_TYPE_LAST	\
+			CREQ_ROCE_MIRROR_CFG_RESP_TYPE_QP_EVENT
+	u8      status;
+	__le16  cookie;
+	__le32  reserved32;
+	u8      v;
+	#define CREQ_ROCE_MIRROR_CFG_RESP_V		0x1UL
+	u8      event;
+	#define CREQ_ROCE_MIRROR_CFG_RESP_EVENT_ROCE_MIRROR_CFG	0x99UL
+	#define CREQ_ROCE_MIRROR_CFG_RESP_EVENT_LAST	\
+			CREQ_ROCE_MIRROR_CFG_RESP_EVENT_ROCE_MIRROR_CFG
+	u8      reserved48[6];
+};
+
 /* cmdq_query_func (size:128b/16B) */
 struct cmdq_query_func {
 	u8	opcode;
-- 
2.43.5

[PATCH rdma-next 08/10] RDMA/bnxt_re: Initialize fw with roce_mirror support

[Kalesh AP]

From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>

- Check FW capability for roce_mirror support.
- Initialize FW with roce_mirror support.
- When modifying QP, use unique GID for sgid incase of RawEth QP.

Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Anantha Prabhu <anantha.prabhu@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/ib_verbs.c   |  2 ++
 drivers/infiniband/hw/bnxt_re/main.c       |  2 ++
 drivers/infiniband/hw/bnxt_re/qplib_fp.c   | 12 +++++++++---
 drivers/infiniband/hw/bnxt_re/qplib_rcfw.c |  4 ++++
 drivers/infiniband/hw/bnxt_re/roce_hsi.h   |  1 +
 5 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
index 90c23d0ee262..f12d6cd3ae93 100644
--- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c
+++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
@@ -1037,6 +1037,8 @@ static u8 __from_ib_qp_type(enum ib_qp_type type)
 		return CMDQ_CREATE_QP_TYPE_RC;
 	case IB_QPT_UD:
 		return CMDQ_CREATE_QP_TYPE_UD;
+	case IB_QPT_RAW_PACKET:
+		return CMDQ_CREATE_QP_TYPE_RAW_ETHERTYPE;
 	default:
 		return IB_QPT_MAX;
 	}
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index fe1be036f8f2..059a4963963a 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -654,6 +654,8 @@ int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev)
 	flags_ext2 = le32_to_cpu(resp.flags_ext2);
 	cctx->modes.dbr_pacing = flags_ext2 & FUNC_QCAPS_RESP_FLAGS_EXT2_DBR_PACING_EXT_SUPPORTED ||
 				 flags_ext2 & FUNC_QCAPS_RESP_FLAGS_EXT2_DBR_PACING_V0_SUPPORTED;
+	cctx->modes.roce_mirror = !!(le32_to_cpu(resp.flags_ext3) &
+				     FUNC_QCAPS_RESP_FLAGS_EXT3_MIRROR_ON_ROCE_SUPPORTED);
 	return 0;
 }
 
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
index 092310571dcc..43a4ef76272d 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
@@ -1335,6 +1335,7 @@ static bool is_optimized_state_transition(struct bnxt_qplib_qp *qp)
 
 int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
 {
+	struct bnxt_qplib_sgid_tbl *sgid_tbl = &res->sgid_tbl;
 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
 	struct creq_modify_qp_resp resp = {};
 	struct bnxt_qplib_cmdqmsg msg = {};
@@ -1386,9 +1387,14 @@ int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL)
 		req.flow_label = cpu_to_le32(qp->ah.flow_label);
 
-	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)
-		req.sgid_index = cpu_to_le16(res->sgid_tbl.hw_id
-					     [qp->ah.sgid_index]);
+	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX) {
+		if (qp->type == CMDQ_CREATE_QP_TYPE_RAW_ETHERTYPE)
+			req.sgid_index =
+				cpu_to_le16(sgid_tbl->hw_id[qp->ugid_index]);
+		else
+			req.sgid_index =
+				cpu_to_le16(sgid_tbl->hw_id[qp->ah.sgid_index]);
+	}
 
 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT)
 		req.hop_limit = qp->ah.hop_limit;
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
index b97e75404139..5e34395472c5 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
@@ -905,6 +905,10 @@ int bnxt_qplib_init_rcfw(struct bnxt_qplib_rcfw *rcfw,
 		flags |= CMDQ_INITIALIZE_FW_FLAGS_OPTIMIZE_MODIFY_QP_SUPPORTED;
 	if (rcfw->res->en_dev->flags & BNXT_EN_FLAG_ROCE_VF_RES_MGMT)
 		flags |= CMDQ_INITIALIZE_FW_FLAGS_L2_VF_RESOURCE_MGMT;
+	if (bnxt_qplib_roce_mirror_supported(rcfw->res->cctx)) {
+		flags |= CMDQ_INITIALIZE_FW_FLAGS_MIRROR_ON_ROCE_SUPPORTED;
+		rcfw->roce_mirror = true;
+	}
 	req.flags |= cpu_to_le16(flags);
 	req.stat_ctx_id = cpu_to_le32(ctx->stats.fw_id);
 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
diff --git a/drivers/infiniband/hw/bnxt_re/roce_hsi.h b/drivers/infiniband/hw/bnxt_re/roce_hsi.h
index cfdf69a3fe9a..99ecd72e72e2 100644
--- a/drivers/infiniband/hw/bnxt_re/roce_hsi.h
+++ b/drivers/infiniband/hw/bnxt_re/roce_hsi.h
@@ -219,6 +219,7 @@ struct cmdq_initialize_fw {
 	#define CMDQ_INITIALIZE_FW_FLAGS_HW_REQUESTER_RETX_SUPPORTED     0x2UL
 	#define CMDQ_INITIALIZE_FW_FLAGS_OPTIMIZE_MODIFY_QP_SUPPORTED    0x8UL
 	#define CMDQ_INITIALIZE_FW_FLAGS_L2_VF_RESOURCE_MGMT		 0x10UL
+	#define CMDQ_INITIALIZE_FW_FLAGS_MIRROR_ON_ROCE_SUPPORTED        0x80UL
 	__le16	cookie;
 	u8	resp_size;
 	u8	reserved8;
-- 
2.43.5

[PATCH rdma-next 09/10] RDMA/bnxt_re: Use firmware provided message timeout value

[Kalesh AP]

From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>

Before this patch, we used a hardcoded value of 500 msec as the default
value for L2 firmware message response timeout. With this commit,
the driver is using the firmware timeout value from the firmware.

As part of this change moved bnxt_re_query_hwrm_intf_version() to
bnxt_re_setup_chip_ctx() so that timeout value is queries before
sending first command.

Signed-off-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Selvin Xavier <selvin.xavier@broadcom.com>
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Reviewed-by: Bhargava Chenna Marreddy <bhargava.marreddy@broadcom.com>
Co-developed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/bnxt_re.h |  3 +++
 drivers/infiniband/hw/bnxt_re/main.c    | 33 ++++++++++++++-----------
 2 files changed, 22 insertions(+), 14 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
index 3a219d67746c..4ac6a312e053 100644
--- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h
+++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h
@@ -288,4 +288,7 @@ static inline int bnxt_re_read_context_allowed(struct bnxt_re_dev *rdev)
 #define BNXT_RE_CONTEXT_TYPE_MRW_SIZE_P7	192
 #define BNXT_RE_CONTEXT_TYPE_SRQ_SIZE_P7	192
 
+#define BNXT_RE_HWRM_CMD_TIMEOUT(rdev)		\
+		((rdev)->chip_ctx->hwrm_cmd_max_timeout * 1000)
+
 #endif
diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 059a4963963a..3e1161721738 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -80,6 +80,7 @@ MODULE_LICENSE("Dual BSD/GPL");
 static DEFINE_MUTEX(bnxt_re_mutex);
 
 static int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev);
+static int bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev);
 
 static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
 			     u32 *offset);
@@ -188,6 +189,10 @@ static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev)
 	rdev->qplib_res.is_vf = BNXT_EN_VF(en_dev);
 	rdev->qplib_res.en_dev = en_dev;
 
+	rc = bnxt_re_query_hwrm_intf_version(rdev);
+	if (rc)
+		goto free_dev_attr;
+
 	bnxt_re_set_drv_mode(rdev);
 
 	bnxt_re_set_db_offset(rdev);
@@ -551,7 +556,7 @@ void bnxt_re_hwrm_free_vnic(struct bnxt_re_dev *rdev)
 
 	req.vnic_id = cpu_to_le32(rdev->mirror_vnic_id);
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), NULL,
-			    0, DFLT_HWRM_CMD_TIMEOUT);
+			    0, BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (rc)
 		ibdev_dbg(&rdev->ibdev,
@@ -571,7 +576,7 @@ int bnxt_re_hwrm_alloc_vnic(struct bnxt_re_dev *rdev)
 	req.vnic_id = cpu_to_le16(rdev->mirror_vnic_id);
 	req.flags = cpu_to_le32(VNIC_ALLOC_REQ_FLAGS_VNIC_ID_VALID);
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (rc)
 		ibdev_dbg(&rdev->ibdev,
@@ -597,7 +602,7 @@ int bnxt_re_hwrm_cfg_vnic(struct bnxt_re_dev *rdev, u32 qp_id)
 	req.mru = cpu_to_le16(rdev->netdev->mtu + VLAN_ETH_HLEN);
 
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), NULL,
-			    0, DFLT_HWRM_CMD_TIMEOUT);
+			    0, BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (rc)
 		ibdev_dbg(&rdev->ibdev,
@@ -619,7 +624,7 @@ static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_QCFG);
 	req.fid = cpu_to_le16(0xffff);
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (!rc) {
 		*db_len = PAGE_ALIGN(le16_to_cpu(resp.l2_doorbell_bar_size_kb) * 1024);
@@ -644,7 +649,7 @@ int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev)
 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_QCAPS);
 	req.fid = cpu_to_le16(0xffff);
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (rc)
@@ -672,7 +677,7 @@ static int bnxt_re_hwrm_dbr_pacing_qcfg(struct bnxt_re_dev *rdev)
 	cctx = rdev->chip_ctx;
 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_DBR_PACING_QCFG);
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (rc)
 		return rc;
@@ -932,7 +937,7 @@ static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
 	req.ring_type = type;
 	req.ring_id = cpu_to_le16(fw_ring_id);
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (rc)
 		ibdev_err(&rdev->ibdev, "Failed to free HW ring:%d :%#x",
@@ -968,7 +973,7 @@ static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev,
 	req.ring_type = ring_attr->type;
 	req.int_mode = ring_attr->mode;
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (!rc)
 		*fw_ring_id = le16_to_cpu(resp.ring_id);
@@ -994,7 +999,7 @@ static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_FREE);
 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (rc)
 		ibdev_err(&rdev->ibdev, "Failed to free HW stats context %#x",
@@ -1024,7 +1029,7 @@ static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
 	req.stats_dma_length = cpu_to_le16(chip_ctx->hw_stats_size);
 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
-			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
+			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
 	rc = bnxt_send_msg(en_dev, &fw_msg);
 	if (!rc)
 		stats->fw_id = le32_to_cpu(resp.stat_ctx_id);
@@ -1984,7 +1989,7 @@ static void bnxt_re_read_vpd_info(struct bnxt_re_dev *rdev)
 	kfree(vpd_data);
 }
 
-static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
+static int bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
 {
 	struct bnxt_en_dev *en_dev = rdev->en_dev;
 	struct hwrm_ver_get_output resp = {};
@@ -2003,7 +2008,7 @@ static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
 	if (rc) {
 		ibdev_err(&rdev->ibdev, "Failed to query HW version, rc = 0x%x",
 			  rc);
-		return;
+		return rc;
 	}
 
 	cctx = rdev->chip_ctx;
@@ -2017,6 +2022,8 @@ static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
 
 	if (!cctx->hwrm_cmd_max_timeout)
 		cctx->hwrm_cmd_max_timeout = RCFW_FW_STALL_MAX_TIMEOUT;
+
+	return 0;
 }
 
 static int bnxt_re_ib_init(struct bnxt_re_dev *rdev)
@@ -2223,8 +2230,6 @@ static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 op_type)
 	/* Check whether VF or PF */
 	bnxt_re_get_sriov_func_type(rdev);
 
-	bnxt_re_query_hwrm_intf_version(rdev);
-
 	/* Establish RCFW Communication Channel to initialize the context
 	 * memory for the function and all child VFs
 	 */
-- 
2.43.5

[PATCH rdma-next 10/10] RDMA/bnxt_re: Remove unnecessary condition checks

[Kalesh AP]

The check for "rdev" and "en_dev" pointer validity always
return false.

Remove them.

Reviewed-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/main.c | 17 -----------------
 1 file changed, 17 deletions(-)

diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 3e1161721738..43af0dba0749 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -922,14 +922,6 @@ static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
 	struct bnxt_fw_msg fw_msg = {};
 	int rc = -EINVAL;
 
-	if (!rdev)
-		return rc;
-
-	en_dev = rdev->en_dev;
-
-	if (!en_dev)
-		return rc;
-
 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
 		return 0;
 
@@ -955,9 +947,6 @@ static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev,
 	struct bnxt_fw_msg fw_msg = {};
 	int rc = -EINVAL;
 
-	if (!en_dev)
-		return rc;
-
 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_RING_ALLOC);
 	req.enables = 0;
 	req.page_tbl_addr =  cpu_to_le64(ring_attr->dma_arr[0]);
@@ -990,9 +979,6 @@ static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
 	struct bnxt_fw_msg fw_msg = {};
 	int rc = -EINVAL;
 
-	if (!en_dev)
-		return rc;
-
 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
 		return 0;
 
@@ -1020,9 +1006,6 @@ static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
 
 	stats->fw_id = INVALID_STATS_CTX_ID;
 
-	if (!en_dev)
-		return rc;
-
 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_ALLOC);
 	req.update_period_ms = cpu_to_le32(1000);
 	req.stats_dma_addr = cpu_to_le64(stats->dma_map);
-- 
2.43.5

Re: [PATCH rdma-next 01/10] bnxt_en: Enhance stats context reservation logic

[Jakub Kicinski]

On Fri, 22 Aug 2025 09:37:52 +0530 Kalesh AP wrote:
> From: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
> 
> When the firmware advertises that the device is capable of supporting
> port mirroring on RoCE device, reserve one additional stat_ctx.
> To support port mirroring feature, RDMA driver allocates one stat_ctx
> for exclusive use in RawEth QP.

Acked-by: Jakub Kicinski <kuba@kernel.org>
-- 
pw-bot: nap