patches for DPDK stable branches
 help / color / mirror / Atom feed
* [PATCH 20.11] net/mlx5: destroy indirect actions on port stop
@ 2022-08-11  7:37 Dmitry Kozlyuk
  2022-08-12  7:21 ` Xueming(Steven) Li
  0 siblings, 1 reply; 2+ messages in thread
From: Dmitry Kozlyuk @ 2022-08-11  7:37 UTC (permalink / raw)
  To: stable; +Cc: Xueming Li

[ Upstream commit ec4e11d41d129ebc7c395b567827492e56fb08b7
  cannot be backported, because it avoids the issue
  by advertising and using a new capability
  of keeping indirect actions across restart.
  The behavior was unspecified before then,
  so it is correct for 20.11 with this patch
  to behave differently compared to the upstream. ]

Shared RSS actions were not destroyed on port stop,
preventing release of involved RX queue resources.
As a result, associated pool of such queues was exhaused
after a number of port stop and start operations.
Diagnostic message from rte_eth_dev_start():

    Rx queue allocation failed: Cannot allocate memory

When built with RTE_LIBRTE_MLX5_DEBUG,
the following assert was triggered:

    testpmd> port start 0
    PANIC in mlx5_rxq_start():
    line 169        assert "!rxq_ctrl->obj" failed

Destroy indirect actions on port stop to release the resources.

Fixes: d7cfcddded61 ("net/mlx5: translate shared action for RSS action")

Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
---
 drivers/net/mlx5/mlx5_trigger.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 6273122e40..c8dc0398ea 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -13,6 +13,7 @@
 #include <mlx5_malloc.h>
 
 #include "mlx5.h"
+#include "mlx5_flow.h"
 #include "mlx5_mr.h"
 #include "mlx5_rxtx.h"
 #include "mlx5_utils.h"
@@ -1192,6 +1193,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
 	mlx5_traffic_disable(dev);
 	/* All RX queue flags will be cleared in the flush interface. */
 	mlx5_flow_list_flush(dev, &priv->flows, true);
+	mlx5_shared_action_flush(dev);
 	mlx5_rx_intr_vec_disable(dev);
 	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
 	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
-- 
2.25.1


^ permalink raw reply	[flat|nested] 2+ messages in thread

* RE: [PATCH 20.11] net/mlx5: destroy indirect actions on port stop
  2022-08-11  7:37 [PATCH 20.11] net/mlx5: destroy indirect actions on port stop Dmitry Kozlyuk
@ 2022-08-12  7:21 ` Xueming(Steven) Li
  0 siblings, 0 replies; 2+ messages in thread
From: Xueming(Steven) Li @ 2022-08-12  7:21 UTC (permalink / raw)
  To: Dmitry Kozlyuk, stable

Thanks, applied to 20.11 LTS queue.

> -----Original Message-----
> From: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
> Sent: Thursday, August 11, 2022 3:37 PM
> To: stable@dpdk.org
> Cc: Xueming(Steven) Li <xuemingl@nvidia.com>
> Subject: [PATCH 20.11] net/mlx5: destroy indirect actions on port stop
> 
> [ Upstream commit ec4e11d41d129ebc7c395b567827492e56fb08b7
>   cannot be backported, because it avoids the issue
>   by advertising and using a new capability
>   of keeping indirect actions across restart.
>   The behavior was unspecified before then,
>   so it is correct for 20.11 with this patch
>   to behave differently compared to the upstream. ]
> 
> Shared RSS actions were not destroyed on port stop, preventing release of involved RX queue resources.
> As a result, associated pool of such queues was exhaused after a number of port stop and start operations.
> Diagnostic message from rte_eth_dev_start():
> 
>     Rx queue allocation failed: Cannot allocate memory
> 
> When built with RTE_LIBRTE_MLX5_DEBUG,
> the following assert was triggered:
> 
>     testpmd> port start 0
>     PANIC in mlx5_rxq_start():
>     line 169        assert "!rxq_ctrl->obj" failed
> 
> Destroy indirect actions on port stop to release the resources.
> 
> Fixes: d7cfcddded61 ("net/mlx5: translate shared action for RSS action")
> 
> Signed-off-by: Dmitry Kozlyuk <dkozlyuk@nvidia.com>
> ---
>  drivers/net/mlx5/mlx5_trigger.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c index 6273122e40..c8dc0398ea 100644
> --- a/drivers/net/mlx5/mlx5_trigger.c
> +++ b/drivers/net/mlx5/mlx5_trigger.c
> @@ -13,6 +13,7 @@
>  #include <mlx5_malloc.h>
> 
>  #include "mlx5.h"
> +#include "mlx5_flow.h"
>  #include "mlx5_mr.h"
>  #include "mlx5_rxtx.h"
>  #include "mlx5_utils.h"
> @@ -1192,6 +1193,7 @@ mlx5_dev_stop(struct rte_eth_dev *dev)
>  	mlx5_traffic_disable(dev);
>  	/* All RX queue flags will be cleared in the flush interface. */
>  	mlx5_flow_list_flush(dev, &priv->flows, true);
> +	mlx5_shared_action_flush(dev);
>  	mlx5_rx_intr_vec_disable(dev);
>  	priv->sh->port[priv->dev_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
>  	priv->sh->port[priv->dev_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
> --
> 2.25.1


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2022-08-12  7:21 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-08-11  7:37 [PATCH 20.11] net/mlx5: destroy indirect actions on port stop Dmitry Kozlyuk
2022-08-12  7:21 ` Xueming(Steven) Li

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).