* [PATCH] net/mlx5: fix check for orphan wait descriptor
@ 2022-08-11 5:50 Viacheslav Ovsiienko
2022-08-17 8:20 ` Raslan Darawsheh
0 siblings, 1 reply; 2+ messages in thread
From: Viacheslav Ovsiienko @ 2022-08-11 5:50 UTC (permalink / raw)
To: dev; +Cc: ferruh.yigit, matan, rasland, stable
The mlx5 PMD supports send scheduling feature, it allows
to send packets at specified moment of time, to do that
PMD pushes special wait descriptor (WQE) to the hardware
queue and then pushes descriptor for packet data as usual.
If queue is close to be full or there is no enough elts
buffers to store mbufs being sent the data descriptors might
be not pushed and the orphan wait WQE (not followed by the
data) might reside in queue on tx_burst routine exit.
To avoid orphan wait WQEs there was the check for enough
free space in the queue WQE buffer and enough amount of the
free elts in queue mbuf storage. This check was incomplete
and did not cover all the cases for Enhanced Multi-Packet
Write descriptors.
Fixes: 2f827f5ea6e1 ("net/mlx5: support scheduling on send routine template")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
drivers/net/mlx5/mlx5_tx.h | 74 +++++++++++++++++++++-----------------
1 file changed, 41 insertions(+), 33 deletions(-)
diff --git a/drivers/net/mlx5/mlx5_tx.h b/drivers/net/mlx5/mlx5_tx.h
index 20776919c2..f081921ffc 100644
--- a/drivers/net/mlx5/mlx5_tx.h
+++ b/drivers/net/mlx5/mlx5_tx.h
@@ -1642,6 +1642,9 @@ mlx5_tx_mseg_build(struct mlx5_txq_data *__rte_restrict txq,
* Pointer to TX queue structure.
* @param loc
* Pointer to burst routine local context.
+ * @param elts
+ * Number of free elements in elts buffer to be checked, for zero
+ * value the check is optimized out by compiler.
* @param olx
* Configured Tx offloads mask. It is fully defined at
* compile time and may be used for optimization.
@@ -1655,6 +1658,7 @@ mlx5_tx_mseg_build(struct mlx5_txq_data *__rte_restrict txq,
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_schedule_send(struct mlx5_txq_data *restrict txq,
struct mlx5_txq_local *restrict loc,
+ uint16_t elts,
unsigned int olx)
{
if (MLX5_TXOFF_CONFIG(TXPP) &&
@@ -1669,7 +1673,7 @@ mlx5_tx_schedule_send(struct mlx5_txq_data *restrict txq,
* to the queue and we won't get the orphan WAIT WQE.
*/
if (loc->wqe_free <= MLX5_WQE_SIZE_MAX / MLX5_WQE_SIZE ||
- loc->elts_free < NB_SEGS(loc->mbuf))
+ loc->elts_free < elts)
return MLX5_TXCMP_CODE_EXIT;
/* Convert the timestamp into completion to wait. */
ts = *RTE_MBUF_DYNFIELD(loc->mbuf, txq->ts_offset, uint64_t *);
@@ -1735,11 +1739,12 @@ mlx5_tx_packet_multi_tso(struct mlx5_txq_data *__rte_restrict txq,
struct mlx5_wqe *__rte_restrict wqe;
unsigned int ds, dlen, inlen, ntcp, vlan = 0;
+ MLX5_ASSERT(loc->elts_free >= NB_SEGS(loc->mbuf));
if (MLX5_TXOFF_CONFIG(TXPP)) {
enum mlx5_txcmp_code wret;
/* Generate WAIT for scheduling if requested. */
- wret = mlx5_tx_schedule_send(txq, loc, olx);
+ wret = mlx5_tx_schedule_send(txq, loc, 0, olx);
if (wret == MLX5_TXCMP_CODE_EXIT)
return MLX5_TXCMP_CODE_EXIT;
if (wret == MLX5_TXCMP_CODE_ERROR)
@@ -1833,11 +1838,12 @@ mlx5_tx_packet_multi_send(struct mlx5_txq_data *__rte_restrict txq,
unsigned int ds, nseg;
MLX5_ASSERT(NB_SEGS(loc->mbuf) > 1);
+ MLX5_ASSERT(loc->elts_free >= NB_SEGS(loc->mbuf));
if (MLX5_TXOFF_CONFIG(TXPP)) {
enum mlx5_txcmp_code wret;
/* Generate WAIT for scheduling if requested. */
- wret = mlx5_tx_schedule_send(txq, loc, olx);
+ wret = mlx5_tx_schedule_send(txq, loc, 0, olx);
if (wret == MLX5_TXCMP_CODE_EXIT)
return MLX5_TXCMP_CODE_EXIT;
if (wret == MLX5_TXCMP_CODE_ERROR)
@@ -1948,16 +1954,7 @@ mlx5_tx_packet_multi_inline(struct mlx5_txq_data *__rte_restrict txq,
MLX5_ASSERT(MLX5_TXOFF_CONFIG(INLINE));
MLX5_ASSERT(NB_SEGS(loc->mbuf) > 1);
- if (MLX5_TXOFF_CONFIG(TXPP)) {
- enum mlx5_txcmp_code wret;
-
- /* Generate WAIT for scheduling if requested. */
- wret = mlx5_tx_schedule_send(txq, loc, olx);
- if (wret == MLX5_TXCMP_CODE_EXIT)
- return MLX5_TXCMP_CODE_EXIT;
- if (wret == MLX5_TXCMP_CODE_ERROR)
- return MLX5_TXCMP_CODE_ERROR;
- }
+ MLX5_ASSERT(loc->elts_free >= NB_SEGS(loc->mbuf));
/*
* First calculate data length to be inlined
* to estimate the required space for WQE.
@@ -2063,6 +2060,16 @@ mlx5_tx_packet_multi_inline(struct mlx5_txq_data *__rte_restrict txq,
* supposing no any mbufs is being freed during inlining.
*/
do_build:
+ if (MLX5_TXOFF_CONFIG(TXPP)) {
+ enum mlx5_txcmp_code wret;
+
+ /* Generate WAIT for scheduling if requested. */
+ wret = mlx5_tx_schedule_send(txq, loc, 0, olx);
+ if (wret == MLX5_TXCMP_CODE_EXIT)
+ return MLX5_TXCMP_CODE_EXIT;
+ if (wret == MLX5_TXCMP_CODE_ERROR)
+ return MLX5_TXCMP_CODE_ERROR;
+ }
MLX5_ASSERT(inlen <= txq->inlen_send);
ds = NB_SEGS(loc->mbuf) + 2 + (inlen -
MLX5_ESEG_MIN_INLINE_SIZE +
@@ -2223,7 +2230,7 @@ mlx5_tx_burst_tso(struct mlx5_txq_data *__rte_restrict txq,
enum mlx5_txcmp_code wret;
/* Generate WAIT for scheduling if requested. */
- wret = mlx5_tx_schedule_send(txq, loc, olx);
+ wret = mlx5_tx_schedule_send(txq, loc, 1, olx);
if (wret == MLX5_TXCMP_CODE_EXIT)
return MLX5_TXCMP_CODE_EXIT;
if (wret == MLX5_TXCMP_CODE_ERROR)
@@ -2601,16 +2608,6 @@ mlx5_tx_burst_empw_simple(struct mlx5_txq_data *__rte_restrict txq,
next_empw:
MLX5_ASSERT(NB_SEGS(loc->mbuf) == 1);
- if (MLX5_TXOFF_CONFIG(TXPP)) {
- enum mlx5_txcmp_code wret;
-
- /* Generate WAIT for scheduling if requested. */
- wret = mlx5_tx_schedule_send(txq, loc, olx);
- if (wret == MLX5_TXCMP_CODE_EXIT)
- return MLX5_TXCMP_CODE_EXIT;
- if (wret == MLX5_TXCMP_CODE_ERROR)
- return MLX5_TXCMP_CODE_ERROR;
- }
part = RTE_MIN(pkts_n, MLX5_TXOFF_CONFIG(MPW) ?
MLX5_MPW_MAX_PACKETS :
MLX5_EMPW_MAX_PACKETS);
@@ -2621,6 +2618,16 @@ mlx5_tx_burst_empw_simple(struct mlx5_txq_data *__rte_restrict txq,
/* But we still able to send at least minimal eMPW. */
part = loc->elts_free;
}
+ if (MLX5_TXOFF_CONFIG(TXPP)) {
+ enum mlx5_txcmp_code wret;
+
+ /* Generate WAIT for scheduling if requested. */
+ wret = mlx5_tx_schedule_send(txq, loc, 0, olx);
+ if (wret == MLX5_TXCMP_CODE_EXIT)
+ return MLX5_TXCMP_CODE_EXIT;
+ if (wret == MLX5_TXCMP_CODE_ERROR)
+ return MLX5_TXCMP_CODE_ERROR;
+ }
/* Check whether we have enough WQEs */
if (unlikely(loc->wqe_free < ((2 + part + 3) / 4))) {
if (unlikely(loc->wqe_free <
@@ -2775,23 +2782,23 @@ mlx5_tx_burst_empw_inline(struct mlx5_txq_data *__rte_restrict txq,
unsigned int slen = 0;
MLX5_ASSERT(NB_SEGS(loc->mbuf) == 1);
+ /*
+ * Limits the amount of packets in one WQE
+ * to improve CQE latency generation.
+ */
+ nlim = RTE_MIN(pkts_n, MLX5_TXOFF_CONFIG(MPW) ?
+ MLX5_MPW_INLINE_MAX_PACKETS :
+ MLX5_EMPW_MAX_PACKETS);
if (MLX5_TXOFF_CONFIG(TXPP)) {
enum mlx5_txcmp_code wret;
/* Generate WAIT for scheduling if requested. */
- wret = mlx5_tx_schedule_send(txq, loc, olx);
+ wret = mlx5_tx_schedule_send(txq, loc, nlim, olx);
if (wret == MLX5_TXCMP_CODE_EXIT)
return MLX5_TXCMP_CODE_EXIT;
if (wret == MLX5_TXCMP_CODE_ERROR)
return MLX5_TXCMP_CODE_ERROR;
}
- /*
- * Limits the amount of packets in one WQE
- * to improve CQE latency generation.
- */
- nlim = RTE_MIN(pkts_n, MLX5_TXOFF_CONFIG(MPW) ?
- MLX5_MPW_INLINE_MAX_PACKETS :
- MLX5_EMPW_MAX_PACKETS);
/* Check whether we have minimal amount WQEs */
if (unlikely(loc->wqe_free <
((2 + MLX5_EMPW_MIN_PACKETS + 3) / 4)))
@@ -3074,11 +3081,12 @@ mlx5_tx_burst_single_send(struct mlx5_txq_data *__rte_restrict txq,
enum mlx5_txcmp_code ret;
MLX5_ASSERT(NB_SEGS(loc->mbuf) == 1);
+ MLX5_ASSERT(loc->elts_free);
if (MLX5_TXOFF_CONFIG(TXPP)) {
enum mlx5_txcmp_code wret;
/* Generate WAIT for scheduling if requested. */
- wret = mlx5_tx_schedule_send(txq, loc, olx);
+ wret = mlx5_tx_schedule_send(txq, loc, 0, olx);
if (wret == MLX5_TXCMP_CODE_EXIT)
return MLX5_TXCMP_CODE_EXIT;
if (wret == MLX5_TXCMP_CODE_ERROR)
--
2.18.1
^ permalink raw reply [flat|nested] 2+ messages in thread* RE: [PATCH] net/mlx5: fix check for orphan wait descriptor
2022-08-11 5:50 [PATCH] net/mlx5: fix check for orphan wait descriptor Viacheslav Ovsiienko
@ 2022-08-17 8:20 ` Raslan Darawsheh
0 siblings, 0 replies; 2+ messages in thread
From: Raslan Darawsheh @ 2022-08-17 8:20 UTC (permalink / raw)
To: Slava Ovsiienko, dev; +Cc: ferruh.yigit, Matan Azrad, stable
Hi,
> -----Original Message-----
> From: Slava Ovsiienko <viacheslavo@nvidia.com>
> Sent: Thursday, August 11, 2022 8:51 AM
> To: dev@dpdk.org
> Cc: ferruh.yigit@intel.com; Matan Azrad <matan@nvidia.com>; Raslan
> Darawsheh <rasland@nvidia.com>; stable@dpdk.org
> Subject: [PATCH] net/mlx5: fix check for orphan wait descriptor
>
> The mlx5 PMD supports send scheduling feature, it allows
> to send packets at specified moment of time, to do that
> PMD pushes special wait descriptor (WQE) to the hardware
> queue and then pushes descriptor for packet data as usual.
> If queue is close to be full or there is no enough elts
> buffers to store mbufs being sent the data descriptors might
> be not pushed and the orphan wait WQE (not followed by the
> data) might reside in queue on tx_burst routine exit.
>
> To avoid orphan wait WQEs there was the check for enough
> free space in the queue WQE buffer and enough amount of the
> free elts in queue mbuf storage. This check was incomplete
> and did not cover all the cases for Enhanced Multi-Packet
> Write descriptors.
>
> Fixes: 2f827f5ea6e1 ("net/mlx5: support scheduling on send routine
> template")
> Cc: stable@dpdk.org
>
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
Patch applied to next-net-mlx,
Kindest regards,
Raslan Darawsheh
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2022-08-17 8:20 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-08-11 5:50 [PATCH] net/mlx5: fix check for orphan wait descriptor Viacheslav Ovsiienko
2022-08-17 8:20 ` Raslan Darawsheh
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).