[PATCH for-next 0/6] RDMA/hns: Add trace support

[PATCH for-next 0/6] RDMA/hns: Add trace support

[Junxian Huang]

Add trace support for hns. Set tracepoints for flushe CQE, WQE,
AEQE, MT/MTR and CMDQ.

Patch #5 fixes the dependency issue of hns_roce_hw_v2.h on hnae3.h,
otherwise there will be a compilation error when hns_roce_hw_v2.h
is included in hns_roce_trace.h in patch #6.

Junxian Huang (6):
  RDMA/hns: Add trace for flush CQE
  RDMA/hns: Add trace for WQE dumping
  RDMA/hns: Add trace for AEQE dumping
  RDMA/hns: Add trace for MR/MTR attribute dumping
  RDMA/hns: Include hnae3.h in hns_roce_hw_v2.h
  RDMA/hns: Add trace for CMDQ dumping

 drivers/infiniband/hw/hns/hns_roce_ah.c       |   1 -
 drivers/infiniband/hw/hns/hns_roce_device.h   |  20 ++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c    |  19 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h    |   1 +
 drivers/infiniband/hw/hns/hns_roce_main.c     |   1 -
 drivers/infiniband/hw/hns/hns_roce_mr.c       |   3 +
 drivers/infiniband/hw/hns/hns_roce_restrack.c |   1 -
 drivers/infiniband/hw/hns/hns_roce_trace.h    | 213 ++++++++++++++++++
 8 files changed, 255 insertions(+), 4 deletions(-)
 create mode 100644 drivers/infiniband/hw/hns/hns_roce_trace.h

--
2.33.0

[PATCH for-next 1/6] RDMA/hns: Add trace for flush CQE

[Junxian Huang]

Add trace to print the producer index of QP when triggering flush CQE.

Output example:
$ cat /sys/kernel/debug/tracing/trace
  tracer: nop

  entries-in-buffer/entries-written: 2/2   #P:128

                                 _-----=> irqs-off/BH-disabled
                                / _----=> need-resched
                               | / _---=> hardirq/softirq
                               || / _--=> preempt-depth
                               ||| / _-=> migrate-disable
                               |||| /     delay
            TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
               | |         |   |||||     |         |
      ib_send_bw-11474   [075] d..1.  2393.434738: sq_flush_cqe: SQ 0x2 flush head 0xb5c7.
      ib_send_bw-11474   [075] d..1.  2393.434739: rq_flush_cqe: RQ 0x2 flush head 0.

Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 drivers/infiniband/hw/hns/hns_roce_device.h | 17 +++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  |  5 +++
 drivers/infiniband/hw/hns/hns_roce_trace.h  | 50 +++++++++++++++++++++
 3 files changed, 72 insertions(+)
 create mode 100644 drivers/infiniband/hw/hns/hns_roce_trace.h

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 560a1d9de408..080bd049b0f8 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -1027,6 +1027,23 @@ struct hns_roce_dev {
 	atomic64_t *dfx_cnt;
 };
 
+enum hns_roce_trace_type {
+	TRACE_SQ,
+	TRACE_RQ,
+};
+
+static inline const char *trace_type_to_str(enum hns_roce_trace_type type)
+{
+	switch (type) {
+	case TRACE_SQ:
+		return "SQ";
+	case TRACE_RQ:
+		return "RQ";
+	default:
+		return "UNKNOWN";
+	}
+}
+
 static inline struct hns_roce_dev *to_hr_dev(struct ib_device *ib_dev)
 {
 	return container_of(ib_dev, struct hns_roce_dev, ib_dev);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index c6399490e3a5..f29265f28ba9 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -50,6 +50,9 @@
 #include "hns_roce_hem.h"
 #include "hns_roce_hw_v2.h"
 
+#define CREATE_TRACE_POINTS
+#include "hns_roce_trace.h"
+
 enum {
 	CMD_RST_PRC_OTHERS,
 	CMD_RST_PRC_SUCCESS,
@@ -5312,6 +5315,7 @@ static void v2_set_flushed_fields(struct ib_qp *ibqp,
 		return;
 
 	spin_lock_irqsave(&hr_qp->sq.lock, sq_flag);
+	trace_sq_flush_cqe(hr_qp->qpn, hr_qp->sq.head, TRACE_SQ);
 	hr_reg_write(context, QPC_SQ_PRODUCER_IDX, hr_qp->sq.head);
 	hr_reg_clear(qpc_mask, QPC_SQ_PRODUCER_IDX);
 	hr_qp->state = IB_QPS_ERR;
@@ -5321,6 +5325,7 @@ static void v2_set_flushed_fields(struct ib_qp *ibqp,
 		return;
 
 	spin_lock_irqsave(&hr_qp->rq.lock, rq_flag);
+	trace_rq_flush_cqe(hr_qp->qpn, hr_qp->rq.head, TRACE_RQ);
 	hr_reg_write(context, QPC_RQ_PRODUCER_IDX, hr_qp->rq.head);
 	hr_reg_clear(qpc_mask, QPC_RQ_PRODUCER_IDX);
 	spin_unlock_irqrestore(&hr_qp->rq.lock, rq_flag);
diff --git a/drivers/infiniband/hw/hns/hns_roce_trace.h b/drivers/infiniband/hw/hns/hns_roce_trace.h
new file mode 100644
index 000000000000..104a4abf9177
--- /dev/null
+++ b/drivers/infiniband/hw/hns/hns_roce_trace.h
@@ -0,0 +1,50 @@
+/* SPDX-License-Identifier: GPL-2.0+ */
+/*
+ * Copyright (c) 2025 Hisilicon Limited.
+ */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM hns_roce
+
+#if !defined(__HNS_ROCE_TRACE_H) || defined(TRACE_HEADER_MULTI_READ)
+#define __HNS_ROCE_TRACE_H
+
+#include <linux/tracepoint.h>
+#include "hns_roce_device.h"
+
+DECLARE_EVENT_CLASS(flush_head_template,
+		    TP_PROTO(unsigned long qpn, u32 pi,
+			     enum hns_roce_trace_type type),
+		    TP_ARGS(qpn, pi, type),
+
+		    TP_STRUCT__entry(__field(unsigned long, qpn)
+				     __field(u32, pi)
+				     __field(enum hns_roce_trace_type, type)
+		    ),
+
+		    TP_fast_assign(__entry->qpn = qpn;
+				   __entry->pi = pi;
+				   __entry->type = type;
+		    ),
+
+		    TP_printk("%s 0x%lx flush head 0x%x.",
+			      trace_type_to_str(__entry->type),
+			      __entry->qpn, __entry->pi)
+);
+
+DEFINE_EVENT(flush_head_template, sq_flush_cqe,
+	     TP_PROTO(unsigned long qpn, u32 pi,
+		      enum hns_roce_trace_type type),
+	     TP_ARGS(qpn, pi, type));
+DEFINE_EVENT(flush_head_template, rq_flush_cqe,
+	     TP_PROTO(unsigned long qpn, u32 pi,
+		      enum hns_roce_trace_type type),
+	     TP_ARGS(qpn, pi, type));
+
+#endif /* __HNS_ROCE_TRACE_H */
+
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_FILE hns_roce_trace
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#include <trace/define_trace.h>
-- 
2.33.0

[PATCH for-next 2/6] RDMA/hns: Add trace for WQE dumping

[Junxian Huang]

Add trace for WQE dumping, including SQ, RQ and SRQ.

Output example:
$ cat /sys/kernel/debug/tracing/trace
  tracer: nop

  entries-in-buffer/entries-written: 2/2   #P:128

                                 _-----=> irqs-off/BH-disabled
                               / _----=> need-resched
                               | / _---=> hardirq/softirq
			       || / _--=> preempt-depth
                               ||| / _-=> migrate-disable
                               |||| /     delay
            TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
               | |         |   |||||     |         |
  roce_test_main-22730   [074] d..1. 16133.898282: sq_wqe: SQ 0xc wqe
(0x0/0xffff0820a6076060): {0x180,0x639c,0x0,0x1000000,0x0,0x0,0x0,0x0,
0x639c,0x300,0xf7e38000,0x0,0x0,0x0,0x0,0x0}

Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 drivers/infiniband/hw/hns/hns_roce_device.h |  3 ++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  |  8 ++++
 drivers/infiniband/hw/hns/hns_roce_trace.h  | 44 +++++++++++++++++++++
 3 files changed, 55 insertions(+)

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 080bd049b0f8..1dcc9cbb4678 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -1030,6 +1030,7 @@ struct hns_roce_dev {
 enum hns_roce_trace_type {
 	TRACE_SQ,
 	TRACE_RQ,
+	TRACE_SRQ,
 };
 
 static inline const char *trace_type_to_str(enum hns_roce_trace_type type)
@@ -1039,6 +1040,8 @@ static inline const char *trace_type_to_str(enum hns_roce_trace_type type)
 		return "SQ";
 	case TRACE_RQ:
 		return "RQ";
+	case TRACE_SRQ:
+		return "SRQ";
 	default:
 		return "UNKNOWN";
 	}
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index f29265f28ba9..a86884cd1b25 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -741,6 +741,8 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
 		else
 			ret = set_ud_wqe(qp, wr, wqe, &sge_idx, owner_bit);
 
+		trace_sq_wqe(qp->qpn, wqe_idx, wqe, 1 << qp->sq.wqe_shift,
+			     wr->wr_id, TRACE_SQ);
 		if (unlikely(ret)) {
 			*bad_wr = wr;
 			goto out;
@@ -810,6 +812,9 @@ static void fill_rq_wqe(struct hns_roce_qp *hr_qp, const struct ib_recv_wr *wr,
 
 	wqe = hns_roce_get_recv_wqe(hr_qp, wqe_idx);
 	fill_recv_sge_to_wqe(wr, wqe, max_sge, hr_qp->rq.rsv_sge);
+
+	trace_rq_wqe(hr_qp->qpn, wqe_idx, wqe, 1 << hr_qp->rq.wqe_shift,
+		     wr->wr_id, TRACE_RQ);
 }
 
 static int hns_roce_v2_post_recv(struct ib_qp *ibqp,
@@ -987,6 +992,9 @@ static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq,
 		fill_recv_sge_to_wqe(wr, wqe, max_sge, srq->rsv_sge);
 		fill_wqe_idx(srq, wqe_idx);
 		srq->wrid[wqe_idx] = wr->wr_id;
+
+		trace_srq_wqe(srq->srqn, wqe_idx, wqe, 1 << srq->wqe_shift,
+			      wr->wr_id, TRACE_SRQ);
 	}
 
 	if (likely(nreq)) {
diff --git a/drivers/infiniband/hw/hns/hns_roce_trace.h b/drivers/infiniband/hw/hns/hns_roce_trace.h
index 104a4abf9177..71da01b19916 100644
--- a/drivers/infiniband/hw/hns/hns_roce_trace.h
+++ b/drivers/infiniband/hw/hns/hns_roce_trace.h
@@ -41,6 +41,50 @@ DEFINE_EVENT(flush_head_template, rq_flush_cqe,
 		      enum hns_roce_trace_type type),
 	     TP_ARGS(qpn, pi, type));
 
+#define MAX_SGE_PER_WQE 64
+#define MAX_WQE_SIZE (MAX_SGE_PER_WQE * HNS_ROCE_SGE_SIZE)
+DECLARE_EVENT_CLASS(wqe_template,
+		    TP_PROTO(unsigned long qpn, u32 idx, void *wqe, u32 len,
+			     u64 id, enum hns_roce_trace_type type),
+		    TP_ARGS(qpn, idx, wqe, len, id, type),
+
+		    TP_STRUCT__entry(__field(unsigned long, qpn)
+				     __field(u32, idx)
+				     __array(__le32, wqe,
+					     MAX_WQE_SIZE / sizeof(__le32))
+				     __field(u32, len)
+				     __field(u64, id)
+				     __field(enum hns_roce_trace_type, type)
+				     ),
+
+		    TP_fast_assign(__entry->qpn = qpn;
+				   __entry->idx = idx;
+				   __entry->id = id;
+				   memcpy(__entry->wqe, wqe, len);
+				   __entry->len = len / sizeof(__le32);
+				   __entry->type = type;
+				   ),
+
+		    TP_printk("%s 0x%lx wqe(0x%x/0x%llx): %s",
+			      trace_type_to_str(__entry->type),
+			      __entry->qpn, __entry->idx, __entry->id,
+			      __print_array(__entry->wqe, __entry->len,
+					    sizeof(__le32)))
+);
+
+DEFINE_EVENT(wqe_template, sq_wqe,
+	     TP_PROTO(unsigned long qpn, u32 idx, void *wqe, u32 len, u64 id,
+		      enum hns_roce_trace_type type),
+	     TP_ARGS(qpn, idx, wqe, len, id, type));
+DEFINE_EVENT(wqe_template, rq_wqe,
+	     TP_PROTO(unsigned long qpn, u32 idx, void *wqe, u32 len, u64 id,
+		      enum hns_roce_trace_type type),
+	     TP_ARGS(qpn, idx, wqe, len, id, type));
+DEFINE_EVENT(wqe_template, srq_wqe,
+	     TP_PROTO(unsigned long qpn, u32 idx, void *wqe, u32 len, u64 id,
+		      enum hns_roce_trace_type type),
+	     TP_ARGS(qpn, idx, wqe, len, id, type));
+
 #endif /* __HNS_ROCE_TRACE_H */
 
 #undef TRACE_INCLUDE_FILE
-- 
2.33.0

[PATCH for-next 3/6] RDMA/hns: Add trace for AEQE dumping

[Junxian Huang]

Add trace for AEQE dumping.

Output example:
$ cat /sys/kernel/debug/tracing/trace
  tracer: nop

  entries-in-buffer/entries-written: 2/2   #P:128

                                 _-----=> irqs-off/BH-disabled
                               / _----=> need-resched
                               | / _---=> hardirq/softirq
                               || / _--=> preempt-depth
                               ||| / _-=> migrate-disable
                               |||| /     delay
            TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
               | |         |   |||||     |         |
	  <idle>-0       [120] d.h1.  7995.835587: ae_info: event 19 aeqe:
{0x80006013,0x0,0x0,0x10d2c,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0}

Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c |  1 +
 drivers/infiniband/hw/hns/hns_roce_trace.h | 19 +++++++++++++++++++
 2 files changed, 20 insertions(+)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index a86884cd1b25..ae8c790d4211 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -6260,6 +6260,7 @@ static irqreturn_t hns_roce_v2_aeq_int(struct hns_roce_dev *hr_dev,
 		eq->sub_type = sub_type;
 		++eq->cons_index;
 		aeqe_found = IRQ_HANDLED;
+		trace_ae_info(event_type, aeqe, eq->eqe_size);
 
 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_AEQE_CNT]);
 
diff --git a/drivers/infiniband/hw/hns/hns_roce_trace.h b/drivers/infiniband/hw/hns/hns_roce_trace.h
index 71da01b19916..11b00564bfba 100644
--- a/drivers/infiniband/hw/hns/hns_roce_trace.h
+++ b/drivers/infiniband/hw/hns/hns_roce_trace.h
@@ -85,6 +85,25 @@ DEFINE_EVENT(wqe_template, srq_wqe,
 		      enum hns_roce_trace_type type),
 	     TP_ARGS(qpn, idx, wqe, len, id, type));
 
+TRACE_EVENT(ae_info,
+	    TP_PROTO(int event_type, void *aeqe, unsigned int len),
+	    TP_ARGS(event_type, aeqe, len),
+
+	    TP_STRUCT__entry(__field(int, event_type)
+			     __array(__le32, aeqe,
+				     HNS_ROCE_V3_EQE_SIZE / sizeof(__le32))
+			     __field(u32, len)
+	    ),
+
+	    TP_fast_assign(__entry->event_type = event_type;
+			   __entry->len = len / sizeof(__le32);
+			   memcpy(__entry->aeqe, aeqe, len);
+	    ),
+
+	    TP_printk("event %2d aeqe: %s", __entry->event_type,
+		      __print_array(__entry->aeqe, __entry->len, sizeof(__le32)))
+);
+
 #endif /* __HNS_ROCE_TRACE_H */
 
 #undef TRACE_INCLUDE_FILE
-- 
2.33.0

[PATCH for-next 4/6] RDMA/hns: Add trace for MR/MTR attribute dumping

[Junxian Huang]

Add trace for MR/MTR attribute dumping.

Output example:
$ cat /sys/kernel/debug/tracing/trace
  tracer: nop

  entries-in-buffer/entries-written: 2/2   #P:128

                                 _-----=> irqs-off/BH-disabled
                               / _----=> need-resched
                               | / _---=> hardirq/softirq
                               || / _--=> preempt-depth
                               ||| / _-=> migrate-disable
                               |||| /     delay
            TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
               | |         |   |||||     |         |
      ib_send_bw-14751   [111] .....  8763.823038: buf_attr: rg cnt:1,
pg_sft:0xc, mtt_only:no, rg 0 (sz:131072, hop:2), rg 1 (sz:0, hop:0),
rg 2 (sz:0, hop:0)

      ib_send_bw-14751   [111] .....  8763.823118: drv_mr:
iova:0xffffb2968000, size:131072, key:512, pd:1, pbl_hop:1, npages:4,
type:0, status:0

Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 drivers/infiniband/hw/hns/hns_roce_mr.c    |  3 +
 drivers/infiniband/hw/hns/hns_roce_trace.h | 65 ++++++++++++++++++++++
 2 files changed, 68 insertions(+)

diff --git a/drivers/infiniband/hw/hns/hns_roce_mr.c b/drivers/infiniband/hw/hns/hns_roce_mr.c
index 09da3496843b..a462a557b818 100644
--- a/drivers/infiniband/hw/hns/hns_roce_mr.c
+++ b/drivers/infiniband/hw/hns/hns_roce_mr.c
@@ -38,6 +38,7 @@
 #include "hns_roce_device.h"
 #include "hns_roce_cmd.h"
 #include "hns_roce_hem.h"
+#include "hns_roce_trace.h"
 
 static u32 hw_index_to_key(int ind)
 {
@@ -159,6 +160,7 @@ static int hns_roce_mr_enable(struct hns_roce_dev *hr_dev,
 	if (IS_ERR(mailbox))
 		return PTR_ERR(mailbox);
 
+	trace_drv_mr(mr);
 	if (mr->type != MR_TYPE_FRMR)
 		ret = hr_dev->hw->write_mtpt(hr_dev, mailbox->buf, mr);
 	else
@@ -1146,6 +1148,7 @@ int hns_roce_mtr_create(struct hns_roce_dev *hr_dev, struct hns_roce_mtr *mtr,
 	struct ib_device *ibdev = &hr_dev->ib_dev;
 	int ret;
 
+	trace_buf_attr(buf_attr);
 	/* The caller has its own buffer list and invokes the hns_roce_mtr_map()
 	 * to finish the MTT configuration.
 	 */
diff --git a/drivers/infiniband/hw/hns/hns_roce_trace.h b/drivers/infiniband/hw/hns/hns_roce_trace.h
index 11b00564bfba..2e60ab5943af 100644
--- a/drivers/infiniband/hw/hns/hns_roce_trace.h
+++ b/drivers/infiniband/hw/hns/hns_roce_trace.h
@@ -10,6 +10,7 @@
 #define __HNS_ROCE_TRACE_H
 
 #include <linux/tracepoint.h>
+#include <linux/string_choices.h>
 #include "hns_roce_device.h"
 
 DECLARE_EVENT_CLASS(flush_head_template,
@@ -104,6 +105,70 @@ TRACE_EVENT(ae_info,
 		      __print_array(__entry->aeqe, __entry->len, sizeof(__le32)))
 );
 
+TRACE_EVENT(drv_mr,
+	    TP_PROTO(struct hns_roce_mr *mr),
+	    TP_ARGS(mr),
+
+	    TP_STRUCT__entry(__field(u64, iova)
+			     __field(u64, size)
+			     __field(u32, key)
+			     __field(u32, pd)
+			     __field(u32, pbl_hop_num)
+			     __field(u32, npages)
+			     __field(int, type)
+			     __field(int, enabled)
+	    ),
+
+	    TP_fast_assign(__entry->iova = mr->iova;
+			   __entry->size = mr->size;
+			   __entry->key = mr->key;
+			   __entry->pd = mr->pd;
+			   __entry->pbl_hop_num = mr->pbl_hop_num;
+			   __entry->npages = mr->npages;
+			   __entry->type = mr->type;
+			   __entry->enabled = mr->enabled;
+	    ),
+
+	    TP_printk("iova:0x%llx, size:%llu, key:%u, pd:%u, pbl_hop:%u, npages:%u, type:%d, status:%d",
+		      __entry->iova, __entry->size, __entry->key,
+		      __entry->pd, __entry->pbl_hop_num, __entry->npages,
+		      __entry->type, __entry->enabled)
+);
+
+TRACE_EVENT(buf_attr,
+	    TP_PROTO(struct hns_roce_buf_attr *attr),
+	    TP_ARGS(attr),
+
+	    TP_STRUCT__entry(__field(unsigned int, region_count)
+			     __field(unsigned int, region0_size)
+			     __field(int, region0_hopnum)
+			     __field(unsigned int, region1_size)
+			     __field(int, region1_hopnum)
+			     __field(unsigned int, region2_size)
+			     __field(int, region2_hopnum)
+			     __field(unsigned int, page_shift)
+			     __field(bool, mtt_only)
+	    ),
+
+	    TP_fast_assign(__entry->region_count = attr->region_count;
+			   __entry->region0_size = attr->region[0].size;
+			   __entry->region0_hopnum = attr->region[0].hopnum;
+			   __entry->region1_size = attr->region[1].size;
+			   __entry->region1_hopnum = attr->region[1].hopnum;
+			   __entry->region2_size = attr->region[2].size;
+			   __entry->region2_hopnum = attr->region[2].hopnum;
+			   __entry->page_shift = attr->page_shift;
+			   __entry->mtt_only = attr->mtt_only;
+	    ),
+
+	    TP_printk("rg cnt:%u, pg_sft:0x%x, mtt_only:%s, rg 0 (sz:%u, hop:%u), rg 1 (sz:%u, hop:%u), rg 2 (sz:%u, hop:%u)\n",
+		      __entry->region_count, __entry->page_shift,
+		      str_yes_no(__entry->mtt_only),
+		      __entry->region0_size, __entry->region0_hopnum,
+		      __entry->region1_size, __entry->region1_hopnum,
+		      __entry->region2_size, __entry->region2_hopnum)
+);
+
 #endif /* __HNS_ROCE_TRACE_H */
 
 #undef TRACE_INCLUDE_FILE
-- 
2.33.0

[PATCH for-next 5/6] RDMA/hns: Include hnae3.h in hns_roce_hw_v2.h

[Junxian Huang]

hns_roce_hw_v2.h has a direct dependency on hnae3.h due to the
inline function hns_roce_write64(), but it doesn't include this
header currently. This leads to that files including
hns_roce_hw_v2.h must also include hnae3.h to avoid compilation
errors, even if they themselves don't really rely on hnae3.h.
This doesn't make sense, hns_roce_hw_v2.h should include hnae3.h
directly.

Fixes: d3743fa94ccd ("RDMA/hns: Fix the chip hanging caused by sending doorbell during reset")
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 drivers/infiniband/hw/hns/hns_roce_ah.c       | 1 -
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c    | 1 -
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h    | 1 +
 drivers/infiniband/hw/hns/hns_roce_main.c     | 1 -
 drivers/infiniband/hw/hns/hns_roce_restrack.c | 1 -
 5 files changed, 1 insertion(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_ah.c b/drivers/infiniband/hw/hns/hns_roce_ah.c
index 4fc5b9d5fea8..307c35888b30 100644
--- a/drivers/infiniband/hw/hns/hns_roce_ah.c
+++ b/drivers/infiniband/hw/hns/hns_roce_ah.c
@@ -33,7 +33,6 @@
 #include <linux/pci.h>
 #include <rdma/ib_addr.h>
 #include <rdma/ib_cache.h>
-#include "hnae3.h"
 #include "hns_roce_device.h"
 #include "hns_roce_hw_v2.h"
 
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index ae8c790d4211..78d39afb2aa0 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -43,7 +43,6 @@
 #include <rdma/ib_umem.h>
 #include <rdma/uverbs_ioctl.h>
 
-#include "hnae3.h"
 #include "hns_roce_common.h"
 #include "hns_roce_device.h"
 #include "hns_roce_cmd.h"
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 91a5665465ff..bc7466830eaf 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -34,6 +34,7 @@
 #define _HNS_ROCE_HW_V2_H
 
 #include <linux/bitops.h>
+#include "hnae3.h"
 
 #define HNS_ROCE_V2_MAX_RC_INL_INN_SZ		32
 #define HNS_ROCE_V2_MTT_ENTRY_SZ		64
diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
index cf89a8db4f64..02f7acb30c5d 100644
--- a/drivers/infiniband/hw/hns/hns_roce_main.c
+++ b/drivers/infiniband/hw/hns/hns_roce_main.c
@@ -37,7 +37,6 @@
 #include <rdma/ib_smi.h>
 #include <rdma/ib_user_verbs.h>
 #include <rdma/ib_cache.h>
-#include "hnae3.h"
 #include "hns_roce_common.h"
 #include "hns_roce_device.h"
 #include "hns_roce_hem.h"
diff --git a/drivers/infiniband/hw/hns/hns_roce_restrack.c b/drivers/infiniband/hw/hns/hns_roce_restrack.c
index 356d98816949..f637b73b946e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_restrack.c
+++ b/drivers/infiniband/hw/hns/hns_roce_restrack.c
@@ -4,7 +4,6 @@
 #include <rdma/rdma_cm.h>
 #include <rdma/restrack.h>
 #include <uapi/rdma/rdma_netlink.h>
-#include "hnae3.h"
 #include "hns_roce_common.h"
 #include "hns_roce_device.h"
 #include "hns_roce_hw_v2.h"
-- 
2.33.0

[PATCH for-next 6/6] RDMA/hns: Add trace for CMDQ dumping

[Junxian Huang]

Add trace for CMDQ dumping.

Output example:
$ cat /sys/kernel/debug/tracing/trace
  tracer: nop

  entries-in-buffer/entries-written: 2/2   #P:128

                                 _-----=> irqs-off/BH-disabled
                               / _----=> need-resched
                               | / _---=> hardirq/softirq
                               || / _--=> preempt-depth
                               ||| / _-=> migrate-disable
                               |||| /     delay
            TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
               | |         |   |||||     |         |
  kworker/u512:1-14003   [089] b..1. 50737.238304: cmdq_req: 0000:bd:00.0 cmdq opcode:0x8500, flag:0x1, retval:0x0, data:{0x2,0x0,0x0,0xffff0000,0x32323232,0x0}

  kworker/u512:1-14003   [089] b..1. 50737.238316: cmdq_resp: 0000:bd:00.0 cmdq opcode:0x8500, flag:0x2, retval:0x0, data:{0x2,0x0,0x0,0xffff0000,0x32323232,0x0}

Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c |  4 +++
 drivers/infiniband/hw/hns/hns_roce_trace.h | 35 ++++++++++++++++++++++
 2 files changed, 39 insertions(+)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 78d39afb2aa0..db33b4d329d1 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -1321,6 +1321,8 @@ static int __hns_roce_cmq_send_one(struct hns_roce_dev *hr_dev,
 	tail = csq->head;
 
 	for (i = 0; i < num; i++) {
+		trace_cmdq_req(hr_dev, &desc[i]);
+
 		csq->desc[csq->head++] = desc[i];
 		if (csq->head == csq->desc_num)
 			csq->head = 0;
@@ -1335,6 +1337,8 @@ static int __hns_roce_cmq_send_one(struct hns_roce_dev *hr_dev,
 	if (hns_roce_cmq_csq_done(hr_dev)) {
 		ret = 0;
 		for (i = 0; i < num; i++) {
+			trace_cmdq_resp(hr_dev, &csq->desc[tail]);
+
 			/* check the result of hardware write back */
 			desc_ret = le16_to_cpu(csq->desc[tail++].retval);
 			if (tail == csq->desc_num)
diff --git a/drivers/infiniband/hw/hns/hns_roce_trace.h b/drivers/infiniband/hw/hns/hns_roce_trace.h
index 2e60ab5943af..adc3d66ce06c 100644
--- a/drivers/infiniband/hw/hns/hns_roce_trace.h
+++ b/drivers/infiniband/hw/hns/hns_roce_trace.h
@@ -12,6 +12,7 @@
 #include <linux/tracepoint.h>
 #include <linux/string_choices.h>
 #include "hns_roce_device.h"
+#include "hns_roce_hw_v2.h"
 
 DECLARE_EVENT_CLASS(flush_head_template,
 		    TP_PROTO(unsigned long qpn, u32 pi,
@@ -169,6 +170,40 @@ TRACE_EVENT(buf_attr,
 		      __entry->region2_size, __entry->region2_hopnum)
 );
 
+DECLARE_EVENT_CLASS(cmdq,
+		    TP_PROTO(struct hns_roce_dev *hr_dev,
+			     struct hns_roce_cmq_desc *desc),
+		    TP_ARGS(hr_dev, desc),
+
+		    TP_STRUCT__entry(__string(dev_name, dev_name(hr_dev->dev))
+				     __field(u16, opcode)
+				     __field(u16, flag)
+				     __field(u16, retval)
+				     __array(__le32, data, 6)
+		    ),
+
+		    TP_fast_assign(__assign_str(dev_name);
+				   __entry->opcode = le16_to_cpu(desc->opcode);
+				   __entry->flag = le16_to_cpu(desc->flag);
+				   __entry->retval = le16_to_cpu(desc->retval);
+				   memcpy(__entry->data, desc->data, 6 * sizeof(__le32));
+		    ),
+
+		    TP_printk("%s cmdq opcode:0x%x, flag:0x%x, retval:0x%x, data:%s\n",
+			      __get_str(dev_name), __entry->opcode,
+			      __entry->flag, __entry->retval,
+			      __print_array(__entry->data, 6, sizeof(__le32)))
+);
+
+DEFINE_EVENT(cmdq, cmdq_req,
+	     TP_PROTO(struct hns_roce_dev *hr_dev,
+		      struct hns_roce_cmq_desc *desc),
+	     TP_ARGS(hr_dev, desc));
+DEFINE_EVENT(cmdq, cmdq_resp,
+	     TP_PROTO(struct hns_roce_dev *hr_dev,
+		      struct hns_roce_cmq_desc *desc),
+	     TP_ARGS(hr_dev, desc));
+
 #endif /* __HNS_ROCE_TRACE_H */
 
 #undef TRACE_INCLUDE_FILE
-- 
2.33.0

Re: [PATCH for-next 0/6] RDMA/hns: Add trace support

[Leon Romanovsky]

On Fri, 18 Apr 2025 16:56:41 +0800, Junxian Huang wrote:
> Add trace support for hns. Set tracepoints for flushe CQE, WQE,
> AEQE, MT/MTR and CMDQ.
> 
> Patch #5 fixes the dependency issue of hns_roce_hw_v2.h on hnae3.h,
> otherwise there will be a compilation error when hns_roce_hw_v2.h
> is included in hns_roce_trace.h in patch #6.
> 
> [...]

Applied, thanks!

[1/6] RDMA/hns: Add trace for flush CQE
      https://git.kernel.org/rdma/rdma/c/96cb704b04419c
[2/6] RDMA/hns: Add trace for WQE dumping
      https://git.kernel.org/rdma/rdma/c/cc1131a3004ed6
[3/6] RDMA/hns: Add trace for AEQE dumping
      https://git.kernel.org/rdma/rdma/c/39091301472a12
[4/6] RDMA/hns: Add trace for MR/MTR attribute dumping
      https://git.kernel.org/rdma/rdma/c/0f38b1359190fb
[5/6] RDMA/hns: Include hnae3.h in hns_roce_hw_v2.h
      https://git.kernel.org/rdma/rdma/c/07474abfb48070
[6/6] RDMA/hns: Add trace for CMDQ dumping
      https://git.kernel.org/rdma/rdma/c/6ce9c93102b98b

Best regards,
-- 
Leon Romanovsky <leon@kernel.org>

Re: [PATCH for-next 0/6] RDMA/hns: Add trace support

[Leon Romanovsky]

On Fri, Apr 18, 2025 at 04:56:41PM +0800, Junxian Huang wrote:
> Add trace support for hns. Set tracepoints for flushe CQE, WQE,
> AEQE, MT/MTR and CMDQ.
> 
> Patch #5 fixes the dependency issue of hns_roce_hw_v2.h on hnae3.h,
> otherwise there will be a compilation error when hns_roce_hw_v2.h
> is included in hns_roce_trace.h in patch #6.
> 
> Junxian Huang (6):
>   RDMA/hns: Add trace for flush CQE
>   RDMA/hns: Add trace for WQE dumping
>   RDMA/hns: Add trace for AEQE dumping
>   RDMA/hns: Add trace for MR/MTR attribute dumping
>   RDMA/hns: Include hnae3.h in hns_roce_hw_v2.h
>   RDMA/hns: Add trace for CMDQ dumping
> 
>  drivers/infiniband/hw/hns/hns_roce_ah.c       |   1 -
>  drivers/infiniband/hw/hns/hns_roce_device.h   |  20 ++
>  drivers/infiniband/hw/hns/hns_roce_hw_v2.c    |  19 +-
>  drivers/infiniband/hw/hns/hns_roce_hw_v2.h    |   1 +
>  drivers/infiniband/hw/hns/hns_roce_main.c     |   1 -
>  drivers/infiniband/hw/hns/hns_roce_mr.c       |   3 +
>  drivers/infiniband/hw/hns/hns_roce_restrack.c |   1 -
>  drivers/infiniband/hw/hns/hns_roce_trace.h    | 213 ++++++++++++++++++
>  8 files changed, 255 insertions(+), 4 deletions(-)
>  create mode 100644 drivers/infiniband/hw/hns/hns_roce_trace.h

Please change trace_drv_* calls to have a name of your driver, e.g.
trace_drv_mr() -> trace_hns_mr().

Thanks

> 
> --
> 2.33.0
> 
> 

Re: [PATCH for-next 0/6] RDMA/hns: Add trace support

[Leon Romanovsky]

On Sun, Apr 20, 2025 at 08:21:13AM -0400, Leon Romanovsky wrote:
> 
> On Fri, 18 Apr 2025 16:56:41 +0800, Junxian Huang wrote:
> > Add trace support for hns. Set tracepoints for flushe CQE, WQE,
> > AEQE, MT/MTR and CMDQ.
> > 
> > Patch #5 fixes the dependency issue of hns_roce_hw_v2.h on hnae3.h,
> > otherwise there will be a compilation error when hns_roce_hw_v2.h
> > is included in hns_roce_trace.h in patch #6.
> > 
> > [...]
> 
> Applied, thanks!

I dropped this.

Thanks

Re: [PATCH for-next 0/6] RDMA/hns: Add trace support

[Junxian Huang]

On 2025/4/20 23:11, Leon Romanovsky wrote:
> On Fri, Apr 18, 2025 at 04:56:41PM +0800, Junxian Huang wrote:
>> Add trace support for hns. Set tracepoints for flushe CQE, WQE,
>> AEQE, MT/MTR and CMDQ.
>>
>> Patch #5 fixes the dependency issue of hns_roce_hw_v2.h on hnae3.h,
>> otherwise there will be a compilation error when hns_roce_hw_v2.h
>> is included in hns_roce_trace.h in patch #6.
>>
>> Junxian Huang (6):
>>   RDMA/hns: Add trace for flush CQE
>>   RDMA/hns: Add trace for WQE dumping
>>   RDMA/hns: Add trace for AEQE dumping
>>   RDMA/hns: Add trace for MR/MTR attribute dumping
>>   RDMA/hns: Include hnae3.h in hns_roce_hw_v2.h
>>   RDMA/hns: Add trace for CMDQ dumping
>>
>>  drivers/infiniband/hw/hns/hns_roce_ah.c       |   1 -
>>  drivers/infiniband/hw/hns/hns_roce_device.h   |  20 ++
>>  drivers/infiniband/hw/hns/hns_roce_hw_v2.c    |  19 +-
>>  drivers/infiniband/hw/hns/hns_roce_hw_v2.h    |   1 +
>>  drivers/infiniband/hw/hns/hns_roce_main.c     |   1 -
>>  drivers/infiniband/hw/hns/hns_roce_mr.c       |   3 +
>>  drivers/infiniband/hw/hns/hns_roce_restrack.c |   1 -
>>  drivers/infiniband/hw/hns/hns_roce_trace.h    | 213 ++++++++++++++++++
>>  8 files changed, 255 insertions(+), 4 deletions(-)
>>  create mode 100644 drivers/infiniband/hw/hns/hns_roce_trace.h
> 
> Please change trace_drv_* calls to have a name of your driver, e.g.
> trace_drv_mr() -> trace_hns_mr().
> 

Sure

> Thanks
> 
>>
>> --
>> 2.33.0
>>
>>