Re: [PATCH net-next 5/5] net/mlx5e: XDP, Use page fragments for linear data in multibuf-mode

[Jakub Kicinski]

This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.

For local reproduction steps see https://netdev-ai.bots.linux.dev/ai-local.html
---
net/mlx5e: XDP, Use page fragments for linear data in multibuf-mode

This commit optimizes XDP multi-buffer mode by splitting pages into fixed-
size fragments instead of allocating whole pages for the linear part of XDP
buffers. This reduces allocations significantly, especially on systems with
larger page sizes, improving performance by 15.80% on 4K pages and 31.62%
on 64K pages for XDP_DROP tests.

> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
> index 592234780f2b..2270e2e550dd 100644
> --- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h

[ ... ]

> @@ -1081,6 +1085,8 @@ bool mlx5e_reset_rx_moderation(struct dim_cq_moder *cq_moder, u8 cq_period_mode,
>  bool mlx5e_reset_rx_channels_moderation(struct mlx5e_channels *chs, u8 cq_period_mode,
>  					bool dim_enabled, bool keep_dim_state);
>
> +void mlx5e_mpwqe_dealloc_linear_page(struct mlx5e_rq *rq);
> +
>  struct mlx5e_sq_param;
>  int mlx5e_open_xdpsq(struct mlx5e_channel *c, struct mlx5e_params *params,
>  		     struct mlx5e_sq_param *param, struct xsk_buff_pool *xsk_pool,

[ ... ]

> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
> index 8b3c82f6f038..b376abc561fd 100644
> --- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c

[ ... ]

> @@ -385,10 +385,22 @@ static int mlx5e_rq_alloc_mpwqe_linear_info(struct mlx5e_rq *rq, int node,
>  	if (!li)
>  		return -ENOMEM;
>
> +	linear_frag_count =
> +		BIT(rq->mpwqe.page_shift - MLX5E_XDP_LOG_MAX_LINEAR_SZ);
> +	if (linear_frag_count > U16_MAX) {
> +		netdev_warn(rq->netdev,
> +			    "rq %d: linear_frag_count (%u) larger than expected (%u), page_shift: %u, log_max_linear_sz: %u\n",
> +			    rq->ix, linear_frag_count, U16_MAX,
> +			    rq->mpwqe.page_shift, MLX5E_XDP_LOG_MAX_LINEAR_SZ);
> +		kvfree(li);
> +		return -EINVAL;
> +	}
> +
> +	li->max_frags = linear_frag_count;
>  	rq->mpwqe.linear_info = li;
>
> -	/* additional page per packet for the linear part */
> -	*pool_size *= 2;
> +	/* Set to max to force allocation on first run. */
> +	li->frag_page.frags = li->max_frags;
>
>  	return 0;
>  }

[ ... ]

> @@ -1347,6 +1358,8 @@ void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
>  			mlx5_wq_ll_pop(wq, wqe_ix_be,
>  				       &wqe->next.next_wqe_index);
>  		}
> +
> +		mlx5e_mpwqe_dealloc_linear_page(rq);
>  	} else {
>  		struct mlx5_wq_cyc *wq = &rq->wqe.wq;
>  		u16 missing = mlx5_wq_cyc_missing(wq);
> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> index feb042d84b8e..2ac38536afe9 100644
> --- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c

[ ... ]

> @@ -300,6 +300,35 @@ static void mlx5e_page_release_fragmented(struct page_pool *pp,
>  		page_pool_put_unrefed_netmem(pp, netmem, -1, true);
>  }
>
> +static int mlx5e_mpwqe_linear_page_refill(struct mlx5e_rq *rq)
> +{
> +	struct mlx5e_mpw_linear_info *li = rq->mpwqe.linear_info;
> +
> +	if (likely(li->frag_page.frags < li->max_frags))
> +		return 0;
> +
> +	if (likely(li->frag_page.netmem)) {
> +		mlx5e_page_release_fragmented(rq->page_pool, &li->frag_page);
> +		li->frag_page.netmem = 0;
> +	}
> +
> +	return mlx5e_page_alloc_fragmented(rq->page_pool, &li->frag_page);
> +}
> +
> +static void *mlx5e_mpwqe_get_linear_page_frag(struct mlx5e_rq *rq)
> +{
> +	struct mlx5e_mpw_linear_info *li = rq->mpwqe.linear_info;
> +	u32 frag_offset;
> +
> +	if (unlikely(mlx5e_mpwqe_linear_page_refill(rq)))
> +		return NULL;
> +
> +	frag_offset = li->frag_page.frags << MLX5E_XDP_LOG_MAX_LINEAR_SZ;
> +	WARN_ON(frag_offset >= BIT(rq->mpwqe.page_shift));
> +
> +	return netmem_address(li->frag_page.netmem) + frag_offset;
> +}
> +

[ ... ]

> @@ -702,6 +731,16 @@ static void mlx5e_dealloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
>  	bitmap_fill(wi->skip_release_bitmap, rq->mpwqe.pages_per_wqe);
>  }
>
> +void mlx5e_mpwqe_dealloc_linear_page(struct mlx5e_rq *rq)
> +{
> +	struct mlx5e_mpw_linear_info *li = rq->mpwqe.linear_info;
> +
> +	if (!li || !li->frag_page.netmem)
> +		return;
> +
> +	mlx5e_page_release_fragmented(rq->page_pool, &li->frag_page);
> +}

Does this function need to reset li->frag_page.netmem to 0 and
li->frag_page.frags to li->max_frags after releasing the page?

When an RQ is flushed and then reactivated (for example, in error recovery
via mlx5e_rx_reporter_err_rq_cqe_recover()), the following sequence occurs:

mlx5e_rx_reporter_err_rq_cqe_recover()
  mlx5e_flush_rq()
    mlx5e_free_rx_descs()
      mlx5e_mpwqe_dealloc_linear_page()
        mlx5e_page_release_fragmented()  [releases the page]
        [but doesn't reset netmem or frags]
  mlx5e_activate_rq()
    [RQ becomes active again]

When the next XDP multi-buffer packet arrives, mlx5e_mpwqe_get_linear_page_frag()
is called:

mlx5e_skb_from_cqe_mpwrq_nonlinear()
  mlx5e_mpwqe_get_linear_page_frag()
    mlx5e_mpwqe_linear_page_refill()
      if (likely(li->frag_page.frags < li->max_frags))
        return 0;  [skips allocation if frags < max_frags]
    netmem_address(li->frag_page.netmem) + frag_offset  [UAF]

If li->frag_page.frags was not reset to li->max_frags in
mlx5e_mpwqe_dealloc_linear_page(), the refill function will see frags <
max_frags and skip the allocation, then compute the address using the stale
netmem pointer that was already returned to the page pool.

Shouldn't this match the initialization in mlx5e_rq_alloc_mpwqe_linear_info()
which sets li->frag_page.frags = li->max_frags?
-- 
pw-bot: cr