DPDK patches and discussions
 help / color / mirror / Atom feed
From: Raslan Darawsheh <rasland@mellanox.com>
To: Alexander Kozyrev <akozyrev@mellanox.com>, "dev@dpdk.org" <dev@dpdk.org>
Cc: Matan Azrad <matan@mellanox.com>,
	Slava Ovsiienko <viacheslavo@mellanox.com>,
	"stable@dpdk.org" <stable@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH] net/mlx5: reduce txq completion index memory loads
Date: Tue, 17 Mar 2020 13:10:18 +0000	[thread overview]
Message-ID: <AM0PR05MB67077465D09295189A5F935BC2F60@AM0PR05MB6707.eurprd05.prod.outlook.com> (raw)
In-Reply-To: <1584372899-26732-1-git-send-email-akozyrev@mellanox.com>

Hi,

> -----Original Message-----
> From: Alexander Kozyrev <akozyrev@mellanox.com>
> Sent: Monday, March 16, 2020 5:35 PM
> To: dev@dpdk.org
> Cc: Raslan Darawsheh <rasland@mellanox.com>; Matan Azrad
> <matan@mellanox.com>; Slava Ovsiienko <viacheslavo@mellanox.com>;
> stable@dpdk.org
> Subject: [PATCH] net/mlx5: reduce txq completion index memory loads
> 
> There is a non-optimal check if doorbel is needed present in the
> mlx5_tx_handle_completion() function. Advancing a copy of the txq
> consumer index and checking this copy with initial value causes
> unnecessary memory loads and hurts the performance. It is better to
> have a simple small boolean variable for this purpose. That allows
> to eliminate all the excessive memory operations with the txq consumer
> index and restore the performance of the tx completions.
> 
> Fixes: 1fd9af0 ("net/mlx5: update Tx error handling routine")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Alexander Kozyrev <akozyrev@mellanox.com>
> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> ---
>  drivers/net/mlx5/mlx5_rxtx.c | 29 +++++++++++++----------------
>  1 file changed, 13 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
> index 5ac63da..f3bf763 100644
> --- a/drivers/net/mlx5/mlx5_rxtx.c
> +++ b/drivers/net/mlx5/mlx5_rxtx.c
> @@ -2160,7 +2160,7 @@ enum mlx5_txcmp_code {
>  {
>  	unsigned int count = MLX5_TX_COMP_MAX_CQE;
>  	volatile struct mlx5_cqe *last_cqe = NULL;
> -	uint16_t ci = txq->cq_ci;
> +	bool ring_doorbell = false;
>  	int ret;
> 
>  	static_assert(MLX5_CQE_STATUS_HW_OWN < 0, "Must be negative
> value");
> @@ -2168,8 +2168,8 @@ enum mlx5_txcmp_code {
>  	do {
>  		volatile struct mlx5_cqe *cqe;
> 
> -		cqe = &txq->cqes[ci & txq->cqe_m];
> -		ret = check_cqe(cqe, txq->cqe_s, ci);
> +		cqe = &txq->cqes[txq->cq_ci & txq->cqe_m];
> +		ret = check_cqe(cqe, txq->cqe_s, txq->cq_ci);
>  		if (unlikely(ret != MLX5_CQE_STATUS_SW_OWN)) {
>  			if (likely(ret != MLX5_CQE_STATUS_ERR)) {
>  				/* No new CQEs in completion queue. */
> @@ -2183,7 +2183,6 @@ enum mlx5_txcmp_code {
>  			 * here, before we might perform SQ reset.
>  			 */
>  			rte_wmb();
> -			txq->cq_ci = ci;
>  			ret = mlx5_tx_error_cqe_handle
>  				(txq, (volatile struct mlx5_err_cqe *)cqe);
>  			if (unlikely(ret < 0)) {
> @@ -2199,16 +2198,18 @@ enum mlx5_txcmp_code {
>  			 * MLX5_CQE_SYNDROME_WR_FLUSH_ERR status.
>  			 * The send queue is supposed to be empty.
>  			 */
> -			++ci;
> -			txq->cq_pi = ci;
> +			ring_doorbell = true;
> +			++txq->cq_ci;
> +			txq->cq_pi = txq->cq_ci;
>  			last_cqe = NULL;
>  			continue;
>  		}
>  		/* Normal transmit completion. */
> -		MLX5_ASSERT(ci != txq->cq_pi);
> -		MLX5_ASSERT((txq->fcqs[ci & txq->cqe_m] >> 16) ==
> +		MLX5_ASSERT(txq->cq_ci != txq->cq_pi);
> +		MLX5_ASSERT((txq->fcqs[txq->cq_ci & txq->cqe_m] >> 16)
> ==
>  			    cqe->wqe_counter);
> -		++ci;
> +		ring_doorbell = true;
> +		++txq->cq_ci;
>  		last_cqe = cqe;
>  		/*
>  		 * We have to restrict the amount of processed CQEs
> @@ -2221,14 +2222,10 @@ enum mlx5_txcmp_code {
>  		if (likely(--count == 0))
>  			break;
>  	} while (true);
> -	if (likely(ci != txq->cq_ci)) {
> -		/*
> -		 * Update completion queue consuming index
> -		 * and ring doorbell to notify hardware.
> -		 */
> +	if (likely(ring_doorbell)) {
> +		/* Ring doorbell to notify hardware. */
>  		rte_compiler_barrier();
> -		txq->cq_ci = ci;
> -		*txq->cq_db = rte_cpu_to_be_32(ci);
> +		*txq->cq_db = rte_cpu_to_be_32(txq->cq_ci);
>  		mlx5_tx_comp_flush(txq, last_cqe, olx);
>  	}
>  }
> --
> 1.8.3.1


Patch applied to next-net-mlx,

Kindest regards,
Raslan Darawsheh

      reply	other threads:[~2020-03-17 13:10 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-16 15:34 Alexander Kozyrev
2020-03-17 13:10 ` Raslan Darawsheh [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=AM0PR05MB67077465D09295189A5F935BC2F60@AM0PR05MB6707.eurprd05.prod.outlook.com \
    --to=rasland@mellanox.com \
    --cc=akozyrev@mellanox.com \
    --cc=dev@dpdk.org \
    --cc=matan@mellanox.com \
    --cc=stable@dpdk.org \
    --cc=viacheslavo@mellanox.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).