[PATCH] net/mlx5: fix error CQE dumping for vectorized Rx burst

[PATCH] net/mlx5: fix error CQE dumping for vectorized Rx burst

[Alexander Kozyrev]

There is a dump file with debug information created
for an error CQE to help with troubleshooting later.
It starts with the last CQE, which, presumably is the error CQE.
But this is only true for the scalar Rx burst routing since
we handle CQEs there one by one and detect the error immediately.
For vectorized Rx bursts, we may already move to another CQE
when we detect the error since we handle CQEs in batches there.
Go back to the error CQE in this case to dump proper CQE.

Fixes: 88c0733535 ("net/mlx5: extend Rx completion with error handling")
Cc: stable@dpdk.org

Signed-off-by: Alexander Kozyrev <akozyrev@nvidia.com>
---
 drivers/net/mlx5/mlx5_rx.c       | 16 +++++++++++-----
 drivers/net/mlx5/mlx5_rx.h       |  3 ++-
 drivers/net/mlx5/mlx5_rxtx_vec.c | 12 +++++++-----
 3 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rx.c b/drivers/net/mlx5/mlx5_rx.c
index 917c517b83..7612d15f01 100644
--- a/drivers/net/mlx5/mlx5_rx.c
+++ b/drivers/net/mlx5/mlx5_rx.c
@@ -425,12 +425,14 @@ mlx5_rxq_initialize(struct mlx5_rxq_data *rxq)
  * @param[in] vec
  *   1 when called from vectorized Rx burst, need to prepare mbufs for the RQ.
  *   0 when called from non-vectorized Rx burst.
+ * @param[in] err_n
+ *   Number of CQEs to check for an error.
  *
  * @return
  *   MLX5_RECOVERY_ERROR_RET in case of recovery error, otherwise the CQE status.
  */
 int
-mlx5_rx_err_handle(struct mlx5_rxq_data *rxq, uint8_t vec)
+mlx5_rx_err_handle(struct mlx5_rxq_data *rxq, uint8_t vec, uint16_t err_n)
 {
 	const uint16_t cqe_n = 1 << rxq->cqe_n;
 	const uint16_t cqe_mask = cqe_n - 1;
@@ -442,13 +444,18 @@ mlx5_rx_err_handle(struct mlx5_rxq_data *rxq, uint8_t vec)
 		volatile struct mlx5_cqe *cqe;
 		volatile struct mlx5_err_cqe *err_cqe;
 	} u = {
-		.cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_mask],
+		.cqe = &(*rxq->cqes)[(rxq->cq_ci - vec) & cqe_mask],
 	};
 	struct mlx5_mp_arg_queue_state_modify sm;
-	int ret;
+	int ret, i;
 
 	switch (rxq->err_state) {
 	case MLX5_RXQ_ERR_STATE_NO_ERROR:
+		for (i = 0; i < (int)err_n; i++) {
+			u.cqe = &(*rxq->cqes)[(rxq->cq_ci - vec - i) & cqe_mask];
+			if (MLX5_CQE_OPCODE(u.cqe->op_own) == MLX5_CQE_RESP_ERR)
+				break;
+		}
 		rxq->err_state = MLX5_RXQ_ERR_STATE_NEED_RESET;
 		/* Fall-through */
 	case MLX5_RXQ_ERR_STATE_NEED_RESET:
@@ -507,7 +514,6 @@ mlx5_rx_err_handle(struct mlx5_rxq_data *rxq, uint8_t vec)
 					rxq->elts_ci : rxq->rq_ci;
 				uint32_t elt_idx;
 				struct rte_mbuf **elt;
-				int i;
 				unsigned int n = elts_n - (elts_ci -
 							  rxq->rq_pi);
 
@@ -628,7 +634,7 @@ mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
 			if (unlikely(ret != MLX5_CQE_STATUS_SW_OWN)) {
 				if (unlikely(ret == MLX5_CQE_STATUS_ERR ||
 					     rxq->err_state)) {
-					ret = mlx5_rx_err_handle(rxq, 0);
+					ret = mlx5_rx_err_handle(rxq, 0, 1);
 					if (ret == MLX5_CQE_STATUS_HW_OWN ||
 					    ret == MLX5_RECOVERY_ERROR_RET)
 						return MLX5_ERROR_CQE_RET;
diff --git a/drivers/net/mlx5/mlx5_rx.h b/drivers/net/mlx5/mlx5_rx.h
index e078aaf3dc..4ba53ebc48 100644
--- a/drivers/net/mlx5/mlx5_rx.h
+++ b/drivers/net/mlx5/mlx5_rx.h
@@ -286,7 +286,8 @@ int mlx5_hrxq_modify(struct rte_eth_dev *dev, uint32_t hxrq_idx,
 
 uint16_t mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n);
 void mlx5_rxq_initialize(struct mlx5_rxq_data *rxq);
-__rte_noinline int mlx5_rx_err_handle(struct mlx5_rxq_data *rxq, uint8_t vec);
+__rte_noinline int mlx5_rx_err_handle(struct mlx5_rxq_data *rxq,
+				      uint8_t vec, uint16_t err_n);
 void mlx5_mprq_buf_free(struct mlx5_mprq_buf *buf);
 uint16_t mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts,
 			    uint16_t pkts_n);
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec.c b/drivers/net/mlx5/mlx5_rxtx_vec.c
index 0e2eab068a..c6be2be763 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec.c
+++ b/drivers/net/mlx5/mlx5_rxtx_vec.c
@@ -74,7 +74,7 @@ rxq_handle_pending_error(struct mlx5_rxq_data *rxq, struct rte_mbuf **pkts,
 	rxq->stats.ipackets -= (pkts_n - n);
 	rxq->stats.ibytes -= err_bytes;
 #endif
-	mlx5_rx_err_handle(rxq, 1);
+	mlx5_rx_err_handle(rxq, 1, pkts_n);
 	return n;
 }
 
@@ -253,8 +253,6 @@ rxq_copy_mprq_mbuf_v(struct mlx5_rxq_data *rxq,
 	}
 	rxq->rq_pi += i;
 	rxq->cq_ci += i;
-	rte_io_wmb();
-	*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
 	if (rq_ci != rxq->rq_ci) {
 		rxq->rq_ci = rq_ci;
 		rte_io_wmb();
@@ -361,8 +359,6 @@ rxq_burst_v(struct mlx5_rxq_data *rxq, struct rte_mbuf **pkts,
 			rxq->decompressed -= n;
 		}
 	}
-	rte_io_wmb();
-	*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
 	*no_cq = !rcvd_pkt;
 	return rcvd_pkt;
 }
@@ -390,6 +386,7 @@ mlx5_rx_burst_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	bool no_cq = false;
 
 	do {
+		err = 0;
 		nb_rx = rxq_burst_v(rxq, pkts + tn, pkts_n - tn,
 				    &err, &no_cq);
 		if (unlikely(err | rxq->err_state))
@@ -397,6 +394,8 @@ mlx5_rx_burst_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		tn += nb_rx;
 		if (unlikely(no_cq))
 			break;
+		rte_io_wmb();
+		*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
 	} while (tn != pkts_n);
 	return tn;
 }
@@ -524,6 +523,7 @@ mlx5_rx_burst_mprq_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	bool no_cq = false;
 
 	do {
+		err = 0;
 		nb_rx = rxq_burst_mprq_v(rxq, pkts + tn, pkts_n - tn,
 					 &err, &no_cq);
 		if (unlikely(err | rxq->err_state))
@@ -531,6 +531,8 @@ mlx5_rx_burst_mprq_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		tn += nb_rx;
 		if (unlikely(no_cq))
 			break;
+		rte_io_wmb();
+		*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
 	} while (tn != pkts_n);
 	return tn;
 }
-- 
2.18.2

RE: [PATCH] net/mlx5: fix error CQE dumping for vectorized Rx burst

[Matan Azrad]

From: Alexander Kozyrev
> There is a dump file with debug information created for an error CQE to help
> with troubleshooting later.
> It starts with the last CQE, which, presumably is the error CQE.
> But this is only true for the scalar Rx burst routing since we handle CQEs there
> one by one and detect the error immediately.
> For vectorized Rx bursts, we may already move to another CQE when we
> detect the error since we handle CQEs in batches there.
> Go back to the error CQE in this case to dump proper CQE.
> 
> Fixes: 88c0733535 ("net/mlx5: extend Rx completion with error handling")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Alexander Kozyrev <akozyrev@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>

RE: [PATCH] net/mlx5: fix error CQE dumping for vectorized Rx burst

[Raslan Darawsheh]

Hi,

> -----Original Message-----
> From: Alexander Kozyrev <akozyrev@nvidia.com>
> Sent: Friday, January 27, 2023 5:22 AM
> To: dev@dpdk.org
> Cc: stable@dpdk.org; Raslan Darawsheh <rasland@nvidia.com>; Matan
> Azrad <matan@nvidia.com>; Slava Ovsiienko <viacheslavo@nvidia.com>
> Subject: [PATCH] net/mlx5: fix error CQE dumping for vectorized Rx burst
> 
> There is a dump file with debug information created
> for an error CQE to help with troubleshooting later.
> It starts with the last CQE, which, presumably is the error CQE.
> But this is only true for the scalar Rx burst routing since
> we handle CQEs there one by one and detect the error immediately.
> For vectorized Rx bursts, we may already move to another CQE
> when we detect the error since we handle CQEs in batches there.
> Go back to the error CQE in this case to dump proper CQE.
> 
> Fixes: 88c0733535 ("net/mlx5: extend Rx completion with error handling")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Alexander Kozyrev <akozyrev@nvidia.com>

Patch applied to next-net-mlx,

Kindest regards,
Raslan Darawsheh