From: Dust Li <dust.li@linux.alibaba.com>

Date: Tue, 1 Mar 2022 17:44:02 +0800

Subject: net/smc: don't send in the BH context if sock_owned_by_user

Patch-mainline: v5.18-rc1

Git-commit: 6b88af839d204c9283ae09357555e5c4f56c6da5

References: jsc#PED-612

Send data all the way down to the RDMA device is a time

consuming operation(get a new slot, maybe do RDMA Write

and send a CDC, etc). Moving those operations from BH

to user context is good for performance.

If the sock_lock is hold by user, we don't try to send

data out in the BH context, but just mark we should

send. Since the user will release the sock_lock soon, we

can do the sending there.

Add smc_release_cb() which will be called in release_sock()

and try send in the callback if needed.

This patch moves the sending part out from BH if sock lock

is hold by user. In my testing environment, this saves about

20% softirq in the qperf 4K tcp_bw test in the sender side

with no noticeable throughput drop.

Signed-off-by: Dust Li <dust.li@linux.alibaba.com>

Signed-off-by: David S. Miller <davem@davemloft.net>

Acked-by: Lee, Chun-Yi <jlee@suse.com>

---

 net/smc/af_smc.c  |   16 ++++++++++++++++

 net/smc/smc.h     |    4 ++++

 net/smc/smc_cdc.c |   19 ++++++++++++++-----

 3 files changed, 34 insertions(+), 5 deletions(-)

--- a/net/smc/af_smc.c

+++ b/net/smc/af_smc.c

@@ -193,12 +193,27 @@ void smc_unhash_sk(struct sock *sk)

 }

 EXPORT_SYMBOL_GPL(smc_unhash_sk);

+/* This will be called before user really release sock_lock. So do the

+ * work which we didn't do because of user hold the sock_lock in the

+ * BH context

+ */

+static void smc_release_cb(struct sock *sk)

+{

+	struct smc_sock *smc = smc_sk(sk);

+

+	if (smc->conn.tx_in_release_sock) {

+		smc_tx_pending(&smc->conn);

+		smc->conn.tx_in_release_sock = false;

+	}

+}

+

 struct proto smc_proto = {

 	.name		= "SMC",

 	.owner		= THIS_MODULE,

 	.keepalive	= smc_set_keepalive,

 	.hash		= smc_hash_sk,

 	.unhash		= smc_unhash_sk,

+	.release_cb	= smc_release_cb,

 	.obj_size	= sizeof(struct smc_sock),

 	.h.smc_hash	= &smc_v4_hashinfo,

 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,

@@ -211,6 +226,7 @@ struct proto smc_proto6 = {

 	.keepalive	= smc_set_keepalive,

 	.hash		= smc_hash_sk,

 	.unhash		= smc_unhash_sk,

+	.release_cb	= smc_release_cb,

 	.obj_size	= sizeof(struct smc_sock),

 	.h.smc_hash	= &smc_v6_hashinfo,

 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,

--- a/net/smc/smc.h

+++ b/net/smc/smc.h

@@ -213,6 +213,10 @@ struct smc_connection {

 						 * data still pending

 						 */

 	char			urg_rx_byte;	/* urgent byte */

+	bool			tx_in_release_sock;

+						/* flush pending tx data in

+						 * sock release_cb()

+						 */

 	atomic_t		bytes_to_rcv;	/* arrived data,

 						 * not yet received

 						 */

--- a/net/smc/smc_cdc.c

+++ b/net/smc/smc_cdc.c

@@ -49,10 +49,15 @@ static void smc_cdc_tx_handler(struct sm

 	}

 	if (atomic_dec_and_test(&conn->cdc_pend_tx_wr)) {

-		/* If this is the last pending WR complete, we must push to

-		 * prevent hang when autocork enabled.

+		/* If user owns the sock_lock, mark the connection need sending.

+		 * User context will later try to send when it release sock_lock

+		 * in smc_release_cb()

 		 */

-		smc_tx_sndbuf_nonempty(conn);

+		if (sock_owned_by_user(&smc->sk))

+			conn->tx_in_release_sock = true;

+		else

+			smc_tx_pending(conn);

+

 		if (unlikely(wq_has_sleeper(&conn->cdc_pend_tx_wq)))

 			wake_up(&conn->cdc_pend_tx_wq);

 	}

@@ -355,8 +360,12 @@ static void smc_cdc_msg_recv_action(stru

 	/* trigger sndbuf consumer: RDMA write into peer RMBE and CDC */

 	if ((diff_cons && smc_tx_prepared_sends(conn)) ||

 	    conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||

-	    conn->local_rx_ctrl.prod_flags.urg_data_pending)

-		smc_tx_sndbuf_nonempty(conn);

+	    conn->local_rx_ctrl.prod_flags.urg_data_pending) {

+		if (!sock_owned_by_user(&smc->sk))

+			smc_tx_pending(conn);

+		else

+			conn->tx_in_release_sock = true;

+	}

 	if (diff_cons && conn->urg_tx_pend &&

 	    atomic_read(&conn->peer_rmbe_space) == conn->peer_rmbe_size) {