Bug ID 1200
Summary mlx5_tx_handle_completion: bad timestamp
Product DPDK
Version 22.03
Hardware x86
OS Linux
Status UNCONFIRMED
Severity normal
Priority Normal
Component core
Assignee dev@dpdk.org
Reporter 7532yahoo@gmail.com
Target Milestone ---

In the file `dpdk/drivers/net/mlx5/mlx5_tx.c` method
`mlx5_tx_handle_completion` I inspect the timestamp field of a cqe:

@@ -192,6 +196,9 @@ mlx5_tx_handle_completion(struct mlx5_txq_data
*__rte_restrict txq,

                cqe = &txq->cqes[txq->cq_ci & txq->cqe_m];
                ret = check_cqe(cqe, txq->cqe_s, txq->cq_ci);
+    if (ret==MLX5_CQE_STATUS_HW_OWN) {
+      printf("line:%d cqe timestamp: %lu\n", __LINE__, cqe->timestamp);
+    }

Consider transmitting several thousand packets. This print to stdout with
regard to the timestamp,

* prints 0 for timestamp for the first ~2000 packet completions
* prints non-zero numbers thereafter, however, not monotonically increasing
e.g. completion event timestamp for occurrence  N is 10x smaller than N-1
          


You are receiving this mail because: