netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Aurelien Aptel <aaptel@nvidia.com>
To: linux-nvme@lists.infradead.org, netdev@vger.kernel.org,
	sagi@grimberg.me, hch@lst.de, kbusch@kernel.org, axboe@fb.com,
	chaitanyak@nvidia.com, davem@davemloft.net, kuba@kernel.org
Cc: aaptel@nvidia.com, aurelien.aptel@gmail.com, smalin@nvidia.com,
	malin1024@gmail.com, ogerlitz@nvidia.com, yorayz@nvidia.com,
	borisp@nvidia.com, galshalom@nvidia.com, mgurtovoy@nvidia.com,
	tariqt@nvidia.com, gus@collabora.com
Subject: [PATCH v28 18/20] net/mlx5e: NVMEoTCP, async ddp invalidation
Date: Wed, 30 Apr 2025 08:57:39 +0000	[thread overview]
Message-ID: <20250430085741.5108-19-aaptel@nvidia.com> (raw)
In-Reply-To: <20250430085741.5108-1-aaptel@nvidia.com>

From: Ben Ben-Ishay <benishay@nvidia.com>

After the ULP consumed the buffers of the offloaded request, it calls the
ddp_teardown op to release the NIC mapping for them and allow the NIC to
reuse the HW contexts associated with offloading this IO. We do a
fast/async un-mapping via UMR WQE. In this case, the ULP does holds off
with completing the request towards the upper/application layers until the
HW unmapping is done.

When the corresponding CQE is received, a notification is done via the
the teardown_done ddp callback advertised by the ULP in the ddp context.

Signed-off-by: Ben Ben-Ishay <benishay@nvidia.com>
Signed-off-by: Boris Pismenny <borisp@nvidia.com>
Signed-off-by: Or Gerlitz <ogerlitz@nvidia.com>
Signed-off-by: Yoray Zack <yorayz@nvidia.com>
Signed-off-by: Aurelien Aptel <aaptel@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/en/txrx.h |  4 ++
 .../mellanox/mlx5/core/en_accel/nvmeotcp.c    | 58 +++++++++++++++++--
 .../mellanox/mlx5/core/en_accel/nvmeotcp.h    |  1 +
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   |  6 ++
 4 files changed, 63 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
index 528e5c97f5a4..b9264a7fe587 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
@@ -73,6 +73,7 @@ enum mlx5e_icosq_wqe_type {
 #endif
 #ifdef CONFIG_MLX5_EN_NVMEOTCP
 	MLX5E_ICOSQ_WQE_UMR_NVMEOTCP,
+	MLX5E_ICOSQ_WQE_UMR_NVMEOTCP_INVALIDATE,
 	MLX5E_ICOSQ_WQE_SET_PSV_NVMEOTCP,
 #endif
 };
@@ -264,6 +265,9 @@ struct mlx5e_icosq_wqe_info {
 		struct {
 			struct mlx5e_nvmeotcp_queue *queue;
 		} nvmeotcp_q;
+		struct {
+			struct mlx5e_nvmeotcp_queue_entry *entry;
+		} nvmeotcp_qe;
 #endif
 	};
 };
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.c b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.c
index 48dd242af2bb..639a9187d88c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.c
@@ -173,6 +173,13 @@ build_nvmeotcp_klm_umr(struct mlx5e_nvmeotcp_queue *queue,
 	cseg->qpn_ds = cpu_to_be32((sqn << MLX5_WQE_CTRL_QPN_SHIFT) | ds_cnt);
 	cseg->general_id = cpu_to_be32(id);
 
+	if (!klm_entries) { /* this is invalidate */
+		ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
+		ucseg->flags = MLX5_UMR_INLINE;
+		mkc->status = MLX5_MKEY_STATUS_FREE;
+		return;
+	}
+
 	if (klm_type == KLM_UMR && !klm_offset) {
 		ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_XLT_OCT_SIZE |
 					       MLX5_MKEY_MASK_LEN |
@@ -285,8 +292,8 @@ build_nvmeotcp_static_params(struct mlx5e_nvmeotcp_queue *queue,
 
 static void
 mlx5e_nvmeotcp_fill_wi(struct mlx5e_nvmeotcp_queue *nvmeotcp_queue,
-		       struct mlx5e_icosq *sq, u32 wqebbs, u16 pi,
-		       enum wqe_type type)
+		       struct mlx5e_icosq *sq, u32 wqebbs,
+		       u16 pi, u16 ccid, enum wqe_type type)
 {
 	struct mlx5e_icosq_wqe_info *wi = &sq->db.wqe_info[pi];
 
@@ -298,6 +305,10 @@ mlx5e_nvmeotcp_fill_wi(struct mlx5e_nvmeotcp_queue *nvmeotcp_queue,
 		wi->wqe_type = MLX5E_ICOSQ_WQE_SET_PSV_NVMEOTCP;
 		wi->nvmeotcp_q.queue = nvmeotcp_queue;
 		break;
+	case KLM_INV_UMR:
+		wi->wqe_type = MLX5E_ICOSQ_WQE_UMR_NVMEOTCP_INVALIDATE;
+		wi->nvmeotcp_qe.entry = &nvmeotcp_queue->ccid_table[ccid];
+		break;
 	default:
 		/* cases where no further action is required upon
 		 * completion, such as ddp setup
@@ -319,7 +330,7 @@ mlx5e_nvmeotcp_rx_post_static_params_wqe(struct mlx5e_nvmeotcp_queue *queue,
 	wqebbs = MLX5E_TRANSPORT_SET_STATIC_PARAMS_WQEBBS;
 	pi = mlx5e_icosq_get_next_pi(sq, wqebbs);
 	wqe = MLX5E_TRANSPORT_FETCH_SET_STATIC_PARAMS_WQE(sq, pi);
-	mlx5e_nvmeotcp_fill_wi(NULL, sq, wqebbs, pi, BSF_UMR);
+	mlx5e_nvmeotcp_fill_wi(NULL, sq, wqebbs, pi, 0, BSF_UMR);
 	build_nvmeotcp_static_params(queue, wqe, resync_seq, queue->crc_rx);
 	sq->pc += wqebbs;
 	mlx5e_notify_hw(&sq->wq, sq->pc, sq->uar_map, &wqe->ctrl);
@@ -337,7 +348,7 @@ mlx5e_nvmeotcp_rx_post_progress_params_wqe(struct mlx5e_nvmeotcp_queue *queue,
 	wqebbs = MLX5E_NVMEOTCP_PROGRESS_PARAMS_WQEBBS;
 	pi = mlx5e_icosq_get_next_pi(sq, wqebbs);
 	wqe = MLX5E_NVMEOTCP_FETCH_PROGRESS_PARAMS_WQE(sq, pi);
-	mlx5e_nvmeotcp_fill_wi(queue, sq, wqebbs, pi, SET_PSV_UMR);
+	mlx5e_nvmeotcp_fill_wi(queue, sq, wqebbs, pi, 0, SET_PSV_UMR);
 	build_nvmeotcp_progress_params(queue, wqe, seq);
 	sq->pc += wqebbs;
 	mlx5e_notify_hw(&sq->wq, sq->pc, sq->uar_map, &wqe->ctrl);
@@ -363,7 +374,7 @@ post_klm_wqe(struct mlx5e_nvmeotcp_queue *queue,
 	wqebbs = DIV_ROUND_UP(wqe_sz, MLX5_SEND_WQE_BB);
 	pi = mlx5e_icosq_get_next_pi(sq, wqebbs);
 	wqe = MLX5E_NVMEOTCP_FETCH_KLM_WQE(sq, pi);
-	mlx5e_nvmeotcp_fill_wi(queue, sq, wqebbs, pi, wqe_type);
+	mlx5e_nvmeotcp_fill_wi(queue, sq, wqebbs, pi, ccid, wqe_type);
 	build_nvmeotcp_klm_umr(queue, wqe, ccid, cur_klm_entries, klm_offset,
 			       klm_length, wqe_type);
 	sq->pc += wqebbs;
@@ -378,7 +389,10 @@ mlx5e_nvmeotcp_post_klm_wqe(struct mlx5e_nvmeotcp_queue *queue,
 	struct mlx5e_icosq *sq = &queue->sq;
 	u32 klm_offset = 0, wqes, i;
 
-	wqes = DIV_ROUND_UP(klm_length, queue->max_klms_per_wqe);
+	if (wqe_type == KLM_INV_UMR)
+		wqes = 1;
+	else
+		wqes = DIV_ROUND_UP(klm_length, queue->max_klms_per_wqe);
 
 	spin_lock_bh(&queue->sq_lock);
 
@@ -905,12 +919,44 @@ void mlx5e_nvmeotcp_ctx_complete(struct mlx5e_icosq_wqe_info *wi)
 	complete(&queue->static_params_done);
 }
 
+void mlx5e_nvmeotcp_ddp_inv_done(struct mlx5e_icosq_wqe_info *wi)
+{
+	struct mlx5e_nvmeotcp_queue_entry *q_entry = wi->nvmeotcp_qe.entry;
+	struct mlx5e_nvmeotcp_queue *queue = q_entry->queue;
+	struct mlx5_core_dev *mdev = queue->priv->mdev;
+	struct ulp_ddp_io *ddp = q_entry->ddp;
+	const struct ulp_ddp_ulp_ops *ulp_ops;
+
+	dma_unmap_sg(mdev->device, ddp->sg_table.sgl,
+		     ddp->nents, DMA_FROM_DEVICE);
+
+	q_entry->sgl_length = 0;
+
+	ulp_ops = inet_csk(queue->sk)->icsk_ulp_ddp_ops;
+	if (ulp_ops && ulp_ops->ddp_teardown_done)
+		ulp_ops->ddp_teardown_done(q_entry->ddp_ctx);
+}
+
 static void
 mlx5e_nvmeotcp_ddp_teardown(struct net_device *netdev,
 			    struct sock *sk,
 			    struct ulp_ddp_io *ddp,
 			    void *ddp_ctx)
 {
+	struct mlx5e_nvmeotcp_queue_entry *q_entry;
+	struct mlx5e_nvmeotcp_queue *queue;
+
+	queue = container_of(ulp_ddp_get_ctx(sk), struct mlx5e_nvmeotcp_queue,
+			     ulp_ddp_ctx);
+	q_entry  = &queue->ccid_table[ddp->command_id];
+	WARN_ONCE(q_entry->sgl_length == 0,
+		  "Invalidation of empty sgl (CID 0x%x, queue 0x%x)\n",
+		  ddp->command_id, queue->id);
+
+	q_entry->ddp_ctx = ddp_ctx;
+	q_entry->queue = queue;
+
+	mlx5e_nvmeotcp_post_klm_wqe(queue, KLM_INV_UMR, ddp->command_id, 0);
 }
 
 static void
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.h b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.h
index 4850c19e18c7..67805adc6fdf 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/nvmeotcp.h
@@ -113,6 +113,7 @@ void mlx5e_nvmeotcp_cleanup(struct mlx5e_priv *priv);
 struct mlx5e_nvmeotcp_queue *
 mlx5e_nvmeotcp_get_queue(struct mlx5e_nvmeotcp *nvmeotcp, int id);
 void mlx5e_nvmeotcp_put_queue(struct mlx5e_nvmeotcp_queue *queue);
+void mlx5e_nvmeotcp_ddp_inv_done(struct mlx5e_icosq_wqe_info *wi);
 void mlx5e_nvmeotcp_ctx_complete(struct mlx5e_icosq_wqe_info *wi);
 static inline void mlx5e_nvmeotcp_init_rx(struct mlx5e_priv *priv) {}
 void mlx5e_nvmeotcp_cleanup_rx(struct mlx5e_priv *priv);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 38c8825d8678..8af51d1886bd 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -958,6 +958,9 @@ void mlx5e_free_icosq_descs(struct mlx5e_icosq *sq)
 			break;
 #endif
 #ifdef CONFIG_MLX5_EN_NVMEOTCP
+		case MLX5E_ICOSQ_WQE_UMR_NVMEOTCP_INVALIDATE:
+			mlx5e_nvmeotcp_ddp_inv_done(wi);
+			break;
 		case MLX5E_ICOSQ_WQE_SET_PSV_NVMEOTCP:
 			mlx5e_nvmeotcp_ctx_complete(wi);
 			break;
@@ -1068,6 +1071,9 @@ int mlx5e_poll_ico_cq(struct mlx5e_cq *cq, int budget)
 #ifdef CONFIG_MLX5_EN_NVMEOTCP
 			case MLX5E_ICOSQ_WQE_UMR_NVMEOTCP:
 				break;
+			case MLX5E_ICOSQ_WQE_UMR_NVMEOTCP_INVALIDATE:
+				mlx5e_nvmeotcp_ddp_inv_done(wi);
+				break;
 			case MLX5E_ICOSQ_WQE_SET_PSV_NVMEOTCP:
 				mlx5e_nvmeotcp_ctx_complete(wi);
 				break;
-- 
2.34.1


  parent reply	other threads:[~2025-04-30  8:59 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-04-30  8:57 [PATCH v28 00/20] nvme-tcp receive offloads Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 01/20] net: Introduce direct data placement tcp offload Aurelien Aptel
2025-05-14  7:12   ` Eric Dumazet
2025-05-16 14:47     ` Aurelien Aptel
2025-05-16 22:31       ` Jakub Kicinski
2025-05-17  7:38       ` Eric Dumazet
2025-05-22 15:01         ` Aurelien Aptel
2025-06-04 12:33           ` Aurelien Aptel
2025-06-04 12:55             ` Eric Dumazet
2025-06-05 11:54               ` Aurelien Aptel
2025-06-05 12:44                 ` Eric Dumazet
2025-04-30  8:57 ` [PATCH v28 02/20] netlink: add new family to manage ULP_DDP enablement and stats Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 03/20] iov_iter: skip copy if src == dst for direct data placement Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 04/20] net/tls,core: export get_netdev_for_sock Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 05/20] nvme-tcp: Add DDP offload control path Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 06/20] nvme-tcp: Add DDP data-path Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 07/20] nvme-tcp: RX DDGST offload Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 08/20] nvme-tcp: Deal with netdevice DOWN events Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 09/20] Documentation: add ULP DDP offload documentation Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 10/20] net/mlx5e: Rename from tls to transport static params Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 11/20] net/mlx5e: Refactor ico sq polling to get budget Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 12/20] net/mlx5: Add NVMEoTCP caps, HW bits, 128B CQE and enumerations Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 13/20] net/mlx5e: NVMEoTCP, offload initialization Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 14/20] net/mlx5e: TCP flow steering for nvme-tcp acceleration Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 15/20] net/mlx5e: NVMEoTCP, use KLM UMRs for buffer registration Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 16/20] net/mlx5e: NVMEoTCP, queue init/teardown Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 17/20] net/mlx5e: NVMEoTCP, ddp setup and resync Aurelien Aptel
2025-04-30  8:57 ` Aurelien Aptel [this message]
2025-04-30  8:57 ` [PATCH v28 19/20] net/mlx5e: NVMEoTCP, data-path for DDP+DDGST offload Aurelien Aptel
2025-04-30  8:57 ` [PATCH v28 20/20] net/mlx5e: NVMEoTCP, statistics Aurelien Aptel
2025-04-30 12:52 ` [PATCH v28 00/20] nvme-tcp receive offloads Gustavo Padovan
2025-05-05 20:43   ` Jakub Kicinski
2025-05-05 21:51     ` Keith Busch
2025-05-05 22:51       ` Jakub Kicinski
2025-05-13 12:56         ` Aurelien Aptel
2025-05-13 14:36           ` Jakub Kicinski
2025-05-06 13:34       ` Sagi Grimberg

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=20250430085741.5108-19-aaptel@nvidia.com \
    --to=aaptel@nvidia.com \
    --cc=aurelien.aptel@gmail.com \
    --cc=axboe@fb.com \
    --cc=borisp@nvidia.com \
    --cc=chaitanyak@nvidia.com \
    --cc=davem@davemloft.net \
    --cc=galshalom@nvidia.com \
    --cc=gus@collabora.com \
    --cc=hch@lst.de \
    --cc=kbusch@kernel.org \
    --cc=kuba@kernel.org \
    --cc=linux-nvme@lists.infradead.org \
    --cc=malin1024@gmail.com \
    --cc=mgurtovoy@nvidia.com \
    --cc=netdev@vger.kernel.org \
    --cc=ogerlitz@nvidia.com \
    --cc=sagi@grimberg.me \
    --cc=smalin@nvidia.com \
    --cc=tariqt@nvidia.com \
    --cc=yorayz@nvidia.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).