From: Sagi Grimberg <sagi@grimberg.me>
To: Aurelien Aptel <aaptel@nvidia.com>,
linux-nvme@lists.infradead.org, netdev@vger.kernel.org,
hch@lst.de, kbusch@kernel.org, axboe@fb.com,
chaitanyak@nvidia.com, davem@davemloft.net, kuba@kernel.org
Cc: Yoray Zack <yorayz@nvidia.com>,
aurelien.aptel@gmail.com, smalin@nvidia.com, malin1024@gmail.com,
ogerlitz@nvidia.com, borisp@nvidia.com, galshalom@nvidia.com,
mgurtovoy@nvidia.com
Subject: Re: [PATCH v12 09/26] nvme-tcp: RX DDGST offload
Date: Wed, 9 Aug 2023 10:59:10 +0300 [thread overview]
Message-ID: <2a75b296-edff-3151-7c6e-22209f09a100@grimberg.me> (raw)
In-Reply-To: <20230712161513.134860-10-aaptel@nvidia.com>
On 7/12/23 19:14, Aurelien Aptel wrote:
> From: Yoray Zack <yorayz@nvidia.com>
>
> Enable rx side of DDGST offload when supported.
>
> At the end of the capsule, check if all the skb bits are on, and if not
> recalculate the DDGST in SW and check it.
>
> Signed-off-by: Yoray Zack <yorayz@nvidia.com>
> Signed-off-by: Boris Pismenny <borisp@nvidia.com>
> Signed-off-by: Ben Ben-Ishay <benishay@nvidia.com>
> Signed-off-by: Or Gerlitz <ogerlitz@nvidia.com>
> Signed-off-by: Shai Malin <smalin@nvidia.com>
> Signed-off-by: Aurelien Aptel <aaptel@nvidia.com>
> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
> ---
> drivers/nvme/host/tcp.c | 120 +++++++++++++++++++++++++++++++++++++---
> 1 file changed, 112 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
> index 6057cd424a19..df58668cbad6 100644
> --- a/drivers/nvme/host/tcp.c
> +++ b/drivers/nvme/host/tcp.c
> @@ -130,6 +130,7 @@ enum nvme_tcp_queue_flags {
> NVME_TCP_Q_LIVE = 1,
> NVME_TCP_Q_POLLING = 2,
> NVME_TCP_Q_OFF_DDP = 3,
> + NVME_TCP_Q_OFF_DDGST_RX = 4,
> };
>
> enum nvme_tcp_recv_state {
> @@ -168,6 +169,7 @@ struct nvme_tcp_queue {
> */
> atomic64_t resync_req;
> struct ulp_ddp_limits ddp_limits;
> + bool ddp_ddgst_valid;
> #endif
>
> /* send state */
> @@ -358,9 +360,29 @@ static inline bool is_netdev_ulp_offload_active(struct net_device *netdev,
> if (test_bit(ULP_DDP_C_NVME_TCP_BIT, netdev->ulp_ddp_caps.active))
> return true;
>
> + /*
> + * Otherwise, if netdev supports nvme-tcp ddgst offload and it
> + * was enabled, we can offload
> + */
> + if (queue->data_digest && test_bit(ULP_DDP_C_NVME_TCP_DDGST_RX_BIT,
> + netdev->ulp_ddp_caps.active))
> + return true;
> +
> return false;
> }
>
> +static inline bool nvme_tcp_ddp_ddgst_ok(struct nvme_tcp_queue *queue)
> +{
> + return queue->ddp_ddgst_valid;
> +}
> +
> +static inline void nvme_tcp_ddp_ddgst_update(struct nvme_tcp_queue *queue,
> + struct sk_buff *skb)
> +{
> + if (queue->ddp_ddgst_valid)
> + queue->ddp_ddgst_valid = skb_is_ulp_crc(skb);
> +}
> +
> static int nvme_tcp_req_map_ddp_sg(struct nvme_tcp_request *req, struct request *rq)
> {
> int ret;
> @@ -375,6 +397,38 @@ static int nvme_tcp_req_map_ddp_sg(struct nvme_tcp_request *req, struct request
> return 0;
> }
>
> +static void nvme_tcp_ddp_ddgst_recalc(struct ahash_request *hash,
> + struct request *rq,
> + __le32 *ddgst)
> +{
> + struct nvme_tcp_request *req;
> +
> + if (!rq)
> + return;
> +
> + req = blk_mq_rq_to_pdu(rq);
> +
> + if (!req->offloaded) {
> + /* if we have DDGST_RX offload without DDP the request
> + * wasn't mapped, so we need to map it here
> + */
> + if (nvme_tcp_req_map_ddp_sg(req, rq))
> + return;
grr.. wondering if this is something we want to support (crc without
ddp).
> + }
> +
> + req->ddp.sg_table.sgl = req->ddp.first_sgl;
Why is this assignment needed? why not pass req->ddp.first_sgl ?
> + ahash_request_set_crypt(hash, req->ddp.sg_table.sgl, (u8 *)ddgst,
> + req->data_len);
> + crypto_ahash_digest(hash);
> +
> + if (!req->offloaded) {
> + /* without DDP, ddp_teardown() won't be called, so
> + * free the table here
> + */
> + sg_free_table_chained(&req->ddp.sg_table, SG_CHUNK_SIZE);
> + }
> +}
> +
> static bool nvme_tcp_resync_request(struct sock *sk, u32 seq, u32 flags);
> static void nvme_tcp_ddp_teardown_done(void *ddp_ctx);
> static const struct ulp_ddp_ulp_ops nvme_tcp_ddp_ulp_ops = {
> @@ -434,6 +488,10 @@ static int nvme_tcp_offload_socket(struct nvme_tcp_queue *queue)
> {
> struct net_device *netdev = queue->ctrl->offloading_netdev;
> struct ulp_ddp_config config = {.type = ULP_DDP_NVME};
> + bool offload_ddp = test_bit(ULP_DDP_C_NVME_TCP_BIT,
> + netdev->ulp_ddp_caps.active);
> + bool offload_ddgst_rx = test_bit(ULP_DDP_C_NVME_TCP_DDGST_RX_BIT,
> + netdev->ulp_ddp_caps.active);
> int ret;
>
> config.nvmeotcp.pfv = NVME_TCP_PFV_1_0;
> @@ -460,7 +518,10 @@ static int nvme_tcp_offload_socket(struct nvme_tcp_queue *queue)
> }
>
> inet_csk(queue->sock->sk)->icsk_ulp_ddp_ops = &nvme_tcp_ddp_ulp_ops;
> - set_bit(NVME_TCP_Q_OFF_DDP, &queue->flags);
> + if (offload_ddp)
> + set_bit(NVME_TCP_Q_OFF_DDP, &queue->flags);
> + if (queue->data_digest && offload_ddgst_rx)
> + set_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags);
> return 0;
> }
>
> @@ -474,6 +535,7 @@ static void nvme_tcp_unoffload_socket(struct nvme_tcp_queue *queue)
> }
>
> clear_bit(NVME_TCP_Q_OFF_DDP, &queue->flags);
> + clear_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags);
>
> netdev->netdev_ops->ulp_ddp_ops->sk_del(netdev, queue->sock->sk);
>
> @@ -562,6 +624,20 @@ static inline bool is_netdev_ulp_offload_active(struct net_device *netdev,
> return false;
> }
>
> +static inline bool nvme_tcp_ddp_ddgst_ok(struct nvme_tcp_queue *queue)
> +{
> + return false;
> +}
> +
> +static inline void nvme_tcp_ddp_ddgst_update(struct nvme_tcp_queue *queue,
> + struct sk_buff *skb)
> +{}
> +
> +static void nvme_tcp_ddp_ddgst_recalc(struct ahash_request *hash,
> + struct request *rq,
> + __le32 *ddgst)
> +{}
> +
> static int nvme_tcp_setup_ddp(struct nvme_tcp_queue *queue, u16 command_id,
> struct request *rq)
> {
> @@ -844,6 +920,9 @@ static void nvme_tcp_init_recv_ctx(struct nvme_tcp_queue *queue)
> queue->pdu_offset = 0;
> queue->data_remaining = -1;
> queue->ddgst_remaining = 0;
> +#ifdef CONFIG_ULP_DDP
> + queue->ddp_ddgst_valid = true;
> +#endif
> }
>
> static void nvme_tcp_error_recovery(struct nvme_ctrl *ctrl)
> @@ -1047,7 +1126,8 @@ static int nvme_tcp_recv_pdu(struct nvme_tcp_queue *queue, struct sk_buff *skb,
> size_t rcv_len = min_t(size_t, *len, queue->pdu_remaining);
> int ret;
>
> - if (test_bit(NVME_TCP_Q_OFF_DDP, &queue->flags))
> + if (test_bit(NVME_TCP_Q_OFF_DDP, &queue->flags) ||
> + test_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags))
This now becomes two atomic bitops to check for each capability, where
its more likely that neighther are on...
Is this really racing with anything? maybe just check with bitwise AND?
or a local variable (or struct member)
I don't think that we should add any more overhead for the normal path
than we already have.
> nvme_tcp_resync_response(queue, skb, *offset);
>
> ret = skb_copy_bits(skb, *offset,
> @@ -1111,6 +1191,10 @@ static int nvme_tcp_recv_data(struct nvme_tcp_queue *queue, struct sk_buff *skb,
> nvme_cid_to_rq(nvme_tcp_tagset(queue), pdu->command_id);
> struct nvme_tcp_request *req = blk_mq_rq_to_pdu(rq);
>
> + if (queue->data_digest &&
> + test_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags))
And a third atomic bitop..
> + nvme_tcp_ddp_ddgst_update(queue, skb);
> +
> while (true) {
> int recv_len, ret;
>
> @@ -1139,7 +1223,8 @@ static int nvme_tcp_recv_data(struct nvme_tcp_queue *queue, struct sk_buff *skb,
> recv_len = min_t(size_t, recv_len,
> iov_iter_count(&req->iter));
>
> - if (queue->data_digest)
> + if (queue->data_digest &&
> + !test_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags))
and a 4'th atomic bitop...
> ret = skb_copy_and_hash_datagram_iter(skb, *offset,
> &req->iter, recv_len, queue->rcv_hash);
> else
> @@ -1181,8 +1266,11 @@ static int nvme_tcp_recv_ddgst(struct nvme_tcp_queue *queue,
> char *ddgst = (char *)&queue->recv_ddgst;
> size_t recv_len = min_t(size_t, *len, queue->ddgst_remaining);
> off_t off = NVME_TCP_DIGEST_LENGTH - queue->ddgst_remaining;
> + struct request *rq;
> int ret;
>
> + if (test_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags))
> + nvme_tcp_ddp_ddgst_update(queue, skb);
and a 5'th atomic bitop...
> ret = skb_copy_bits(skb, *offset, &ddgst[off], recv_len);
> if (unlikely(ret))
> return ret;
> @@ -1193,9 +1281,25 @@ static int nvme_tcp_recv_ddgst(struct nvme_tcp_queue *queue,
> if (queue->ddgst_remaining)
> return 0;
>
> + rq = nvme_cid_to_rq(nvme_tcp_tagset(queue),
> + pdu->command_id);
> +
> + if (test_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags)) {
and a 6'th... ok this is just spraying atomic bitops on the data
path. Please find a better solution to this.
> + /*
> + * If HW successfully offloaded the digest
> + * verification, we can skip it
> + */
> + if (nvme_tcp_ddp_ddgst_ok(queue))
> + goto out;
> + /*
> + * Otherwise we have to recalculate and verify the
> + * digest with the software-fallback
> + */
> + nvme_tcp_ddp_ddgst_recalc(queue->rcv_hash, rq,
> + &queue->exp_ddgst);
> + }
> +
> if (queue->recv_ddgst != queue->exp_ddgst) {
> - struct request *rq = nvme_cid_to_rq(nvme_tcp_tagset(queue),
> - pdu->command_id);
> struct nvme_tcp_request *req = blk_mq_rq_to_pdu(rq);
>
> req->status = cpu_to_le16(NVME_SC_DATA_XFER_ERROR);
> @@ -1206,9 +1310,8 @@ static int nvme_tcp_recv_ddgst(struct nvme_tcp_queue *queue,
> le32_to_cpu(queue->exp_ddgst));
> }
>
> +out:
> if (pdu->hdr.flags & NVME_TCP_F_DATA_SUCCESS) {
> - struct request *rq = nvme_cid_to_rq(nvme_tcp_tagset(queue),
> - pdu->command_id);
> struct nvme_tcp_request *req = blk_mq_rq_to_pdu(rq);
>
> nvme_tcp_end_request(rq, le16_to_cpu(req->status));
> @@ -2125,7 +2228,8 @@ static void __nvme_tcp_stop_queue(struct nvme_tcp_queue *queue)
> kernel_sock_shutdown(queue->sock, SHUT_RDWR);
> nvme_tcp_restore_sock_ops(queue);
> cancel_work_sync(&queue->io_work);
> - if (test_bit(NVME_TCP_Q_OFF_DDP, &queue->flags))
> + if (test_bit(NVME_TCP_Q_OFF_DDP, &queue->flags) ||
> + test_bit(NVME_TCP_Q_OFF_DDGST_RX, &queue->flags))
> nvme_tcp_unoffload_socket(queue);
> }
>
next prev parent reply other threads:[~2023-08-09 7:59 UTC|newest]
Thread overview: 62+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-07-12 16:14 [PATCH v12 00/26] nvme-tcp receive offloads Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 01/26] net: Introduce direct data placement tcp offload Aurelien Aptel
2023-08-09 7:15 ` Sagi Grimberg
2023-08-10 14:46 ` Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 02/26] net/ethtool: add new stringset ETH_SS_ULP_DDP_{CAPS,STATS} Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 03/26] net/ethtool: add ULP_DDP_{GET,SET} operations for caps and stats Aurelien Aptel
2023-07-15 10:14 ` Simon Horman
2023-07-17 9:45 ` Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 04/26] Documentation: document netlink ULP_DDP_GET/SET messages Aurelien Aptel
2023-07-15 10:17 ` Simon Horman
2023-07-17 9:47 ` Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 05/26] iov_iter: skip copy if src == dst for direct data placement Aurelien Aptel
2023-08-16 0:24 ` Max Gurtovoy
2023-07-12 16:14 ` [PATCH v12 06/26] net/tls,core: export get_netdev_for_sock Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 07/26] nvme-tcp: Add DDP offload control path Aurelien Aptel
2023-08-01 2:25 ` Chaitanya Kulkarni
2023-08-09 7:39 ` Sagi Grimberg
2023-08-11 5:28 ` Chaitanya Kulkarni
2023-08-16 0:50 ` Max Gurtovoy
2023-08-09 7:13 ` Sagi Grimberg
2023-08-14 16:11 ` Aurelien Aptel
2023-08-14 18:54 ` Sagi Grimberg
2023-08-16 12:30 ` Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 08/26] nvme-tcp: Add DDP data-path Aurelien Aptel
2023-08-09 7:35 ` Sagi Grimberg
2023-08-14 16:12 ` Aurelien Aptel
2023-08-14 19:01 ` Sagi Grimberg
2023-08-17 13:28 ` Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 09/26] nvme-tcp: RX DDGST offload Aurelien Aptel
2023-08-09 7:59 ` Sagi Grimberg [this message]
2023-08-10 14:48 ` Aurelien Aptel
2023-08-13 13:49 ` Sagi Grimberg
2023-07-12 16:14 ` [PATCH v12 10/26] nvme-tcp: Deal with netdevice DOWN events Aurelien Aptel
2023-08-09 8:00 ` Sagi Grimberg
2023-08-16 13:03 ` Aurelien Aptel
2023-08-16 14:10 ` Sagi Grimberg
2023-08-17 14:09 ` Aurelien Aptel
2023-08-20 10:50 ` Sagi Grimberg
2023-08-21 12:33 ` Aurelien Aptel
2023-07-12 16:14 ` [PATCH v12 11/26] nvme-tcp: Add modparam to control the ULP offload enablement Aurelien Aptel
2023-08-09 8:03 ` Sagi Grimberg
2023-08-10 14:50 ` Aurelien Aptel
2023-08-16 1:05 ` Max Gurtovoy
2023-07-12 16:14 ` [PATCH v12 12/26] nvme-tcp: Only enable offload with TLS if the driver supports it Aurelien Aptel
2023-08-09 8:05 ` Sagi Grimberg
2023-08-10 14:52 ` Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 13/26] Documentation: add ULP DDP offload documentation Aurelien Aptel
2023-07-15 10:32 ` Simon Horman
2023-07-17 9:48 ` Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 14/26] net/mlx5e: Rename from tls to transport static params Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 15/26] net/mlx5e: Refactor ico sq polling to get budget Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 16/26] net/mlx5e: Have mdev pointer directly on the icosq structure Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 17/26] net/mlx5e: Refactor doorbell function to allow avoiding a completion Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 18/26] net/mlx5: Add NVMEoTCP caps, HW bits, 128B CQE and enumerations Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 19/26] net/mlx5e: NVMEoTCP, offload initialization Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 20/26] net/mlx5e: TCP flow steering for nvme-tcp acceleration Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 21/26] net/mlx5e: NVMEoTCP, use KLM UMRs for buffer registration Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 22/26] net/mlx5e: NVMEoTCP, queue init/teardown Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 23/26] net/mlx5e: NVMEoTCP, ddp setup and resync Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 24/26] net/mlx5e: NVMEoTCP, async ddp invalidation Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 25/26] net/mlx5e: NVMEoTCP, data-path for DDP+DDGST offload Aurelien Aptel
2023-07-12 16:15 ` [PATCH v12 26/26] net/mlx5e: NVMEoTCP, statistics Aurelien Aptel
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=2a75b296-edff-3151-7c6e-22209f09a100@grimberg.me \
--to=sagi@grimberg.me \
--cc=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=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=smalin@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).