Re: [PATCH net-next 5/5] net/mlx5e: XDP, Use page fragments for linear data in multibuf-mode
From: Jakub Kicinski <kuba@kernel.org>
Date: 2026-03-24 02:42:40
Also in:
bpf, linux-rdma, lkml
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.
quoted hunk ↗ jump to hunk
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
[ ... ]
quoted hunk ↗ jump to hunk
@@ -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,
[ ... ]
quoted hunk ↗ jump to hunk
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
[ ... ]
quoted hunk ↗ jump to hunk
@@ -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; }
[ ... ]
quoted hunk ↗ jump to hunk
@@ -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
[ ... ]
quoted hunk ↗ jump to hunk
@@ -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; +} +
[ ... ]
quoted hunk ↗ jump to hunk
@@ -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