mirror of
https://github.com/torvalds/linux.git
synced 2024-11-23 04:31:50 +00:00
6b88af839d
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>
492 lines
14 KiB
C
492 lines
14 KiB
C
// SPDX-License-Identifier: GPL-2.0
|
|
/*
|
|
* Shared Memory Communications over RDMA (SMC-R) and RoCE
|
|
*
|
|
* Connection Data Control (CDC)
|
|
* handles flow control
|
|
*
|
|
* Copyright IBM Corp. 2016
|
|
*
|
|
* Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
|
|
*/
|
|
|
|
#include <linux/spinlock.h>
|
|
|
|
#include "smc.h"
|
|
#include "smc_wr.h"
|
|
#include "smc_cdc.h"
|
|
#include "smc_tx.h"
|
|
#include "smc_rx.h"
|
|
#include "smc_close.h"
|
|
|
|
/********************************** send *************************************/
|
|
|
|
/* handler for send/transmission completion of a CDC msg */
|
|
static void smc_cdc_tx_handler(struct smc_wr_tx_pend_priv *pnd_snd,
|
|
struct smc_link *link,
|
|
enum ib_wc_status wc_status)
|
|
{
|
|
struct smc_cdc_tx_pend *cdcpend = (struct smc_cdc_tx_pend *)pnd_snd;
|
|
struct smc_connection *conn = cdcpend->conn;
|
|
struct smc_sock *smc;
|
|
int diff;
|
|
|
|
smc = container_of(conn, struct smc_sock, conn);
|
|
bh_lock_sock(&smc->sk);
|
|
if (!wc_status) {
|
|
diff = smc_curs_diff(cdcpend->conn->sndbuf_desc->len,
|
|
&cdcpend->conn->tx_curs_fin,
|
|
&cdcpend->cursor);
|
|
/* sndbuf_space is decreased in smc_sendmsg */
|
|
smp_mb__before_atomic();
|
|
atomic_add(diff, &cdcpend->conn->sndbuf_space);
|
|
/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
|
|
smp_mb__after_atomic();
|
|
smc_curs_copy(&conn->tx_curs_fin, &cdcpend->cursor, conn);
|
|
smc_curs_copy(&conn->local_tx_ctrl_fin, &cdcpend->p_cursor,
|
|
conn);
|
|
conn->tx_cdc_seq_fin = cdcpend->ctrl_seq;
|
|
}
|
|
|
|
if (atomic_dec_and_test(&conn->cdc_pend_tx_wr)) {
|
|
/* 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()
|
|
*/
|
|
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);
|
|
}
|
|
WARN_ON(atomic_read(&conn->cdc_pend_tx_wr) < 0);
|
|
|
|
smc_tx_sndbuf_nonfull(smc);
|
|
bh_unlock_sock(&smc->sk);
|
|
}
|
|
|
|
int smc_cdc_get_free_slot(struct smc_connection *conn,
|
|
struct smc_link *link,
|
|
struct smc_wr_buf **wr_buf,
|
|
struct smc_rdma_wr **wr_rdma_buf,
|
|
struct smc_cdc_tx_pend **pend)
|
|
{
|
|
int rc;
|
|
|
|
rc = smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
|
|
wr_rdma_buf,
|
|
(struct smc_wr_tx_pend_priv **)pend);
|
|
if (conn->killed) {
|
|
/* abnormal termination */
|
|
if (!rc)
|
|
smc_wr_tx_put_slot(link,
|
|
(struct smc_wr_tx_pend_priv *)pend);
|
|
rc = -EPIPE;
|
|
}
|
|
return rc;
|
|
}
|
|
|
|
static inline void smc_cdc_add_pending_send(struct smc_connection *conn,
|
|
struct smc_cdc_tx_pend *pend)
|
|
{
|
|
BUILD_BUG_ON_MSG(
|
|
sizeof(struct smc_cdc_msg) > SMC_WR_BUF_SIZE,
|
|
"must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_cdc_msg)");
|
|
BUILD_BUG_ON_MSG(
|
|
offsetofend(struct smc_cdc_msg, reserved) > SMC_WR_TX_SIZE,
|
|
"must adapt SMC_WR_TX_SIZE to sizeof(struct smc_cdc_msg); if not all smc_wr upper layer protocols use the same message size any more, must start to set link->wr_tx_sges[i].length on each individual smc_wr_tx_send()");
|
|
BUILD_BUG_ON_MSG(
|
|
sizeof(struct smc_cdc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
|
|
"must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_cdc_tx_pend)");
|
|
pend->conn = conn;
|
|
pend->cursor = conn->tx_curs_sent;
|
|
pend->p_cursor = conn->local_tx_ctrl.prod;
|
|
pend->ctrl_seq = conn->tx_cdc_seq;
|
|
}
|
|
|
|
int smc_cdc_msg_send(struct smc_connection *conn,
|
|
struct smc_wr_buf *wr_buf,
|
|
struct smc_cdc_tx_pend *pend)
|
|
{
|
|
struct smc_link *link = conn->lnk;
|
|
union smc_host_cursor cfed;
|
|
int rc;
|
|
|
|
smc_cdc_add_pending_send(conn, pend);
|
|
|
|
conn->tx_cdc_seq++;
|
|
conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
|
|
smc_host_msg_to_cdc((struct smc_cdc_msg *)wr_buf, conn, &cfed);
|
|
|
|
atomic_inc(&conn->cdc_pend_tx_wr);
|
|
smp_mb__after_atomic(); /* Make sure cdc_pend_tx_wr added before post */
|
|
|
|
rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
|
|
if (!rc) {
|
|
smc_curs_copy(&conn->rx_curs_confirmed, &cfed, conn);
|
|
conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
|
|
} else {
|
|
conn->tx_cdc_seq--;
|
|
conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
|
|
atomic_dec(&conn->cdc_pend_tx_wr);
|
|
}
|
|
|
|
return rc;
|
|
}
|
|
|
|
/* send a validation msg indicating the move of a conn to an other QP link */
|
|
int smcr_cdc_msg_send_validation(struct smc_connection *conn,
|
|
struct smc_cdc_tx_pend *pend,
|
|
struct smc_wr_buf *wr_buf)
|
|
{
|
|
struct smc_host_cdc_msg *local = &conn->local_tx_ctrl;
|
|
struct smc_link *link = conn->lnk;
|
|
struct smc_cdc_msg *peer;
|
|
int rc;
|
|
|
|
peer = (struct smc_cdc_msg *)wr_buf;
|
|
peer->common.type = local->common.type;
|
|
peer->len = local->len;
|
|
peer->seqno = htons(conn->tx_cdc_seq_fin); /* seqno last compl. tx */
|
|
peer->token = htonl(local->token);
|
|
peer->prod_flags.failover_validation = 1;
|
|
|
|
/* We need to set pend->conn here to make sure smc_cdc_tx_handler()
|
|
* can handle properly
|
|
*/
|
|
smc_cdc_add_pending_send(conn, pend);
|
|
|
|
atomic_inc(&conn->cdc_pend_tx_wr);
|
|
smp_mb__after_atomic(); /* Make sure cdc_pend_tx_wr added before post */
|
|
|
|
rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
|
|
if (unlikely(rc))
|
|
atomic_dec(&conn->cdc_pend_tx_wr);
|
|
|
|
return rc;
|
|
}
|
|
|
|
static int smcr_cdc_get_slot_and_msg_send(struct smc_connection *conn)
|
|
{
|
|
struct smc_cdc_tx_pend *pend;
|
|
struct smc_wr_buf *wr_buf;
|
|
struct smc_link *link;
|
|
bool again = false;
|
|
int rc;
|
|
|
|
again:
|
|
link = conn->lnk;
|
|
if (!smc_wr_tx_link_hold(link))
|
|
return -ENOLINK;
|
|
rc = smc_cdc_get_free_slot(conn, link, &wr_buf, NULL, &pend);
|
|
if (rc)
|
|
goto put_out;
|
|
|
|
spin_lock_bh(&conn->send_lock);
|
|
if (link != conn->lnk) {
|
|
/* link of connection changed, try again one time*/
|
|
spin_unlock_bh(&conn->send_lock);
|
|
smc_wr_tx_put_slot(link,
|
|
(struct smc_wr_tx_pend_priv *)pend);
|
|
smc_wr_tx_link_put(link);
|
|
if (again)
|
|
return -ENOLINK;
|
|
again = true;
|
|
goto again;
|
|
}
|
|
rc = smc_cdc_msg_send(conn, wr_buf, pend);
|
|
spin_unlock_bh(&conn->send_lock);
|
|
put_out:
|
|
smc_wr_tx_link_put(link);
|
|
return rc;
|
|
}
|
|
|
|
int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn)
|
|
{
|
|
int rc;
|
|
|
|
if (!smc_conn_lgr_valid(conn) ||
|
|
(conn->lgr->is_smcd && conn->lgr->peer_shutdown))
|
|
return -EPIPE;
|
|
|
|
if (conn->lgr->is_smcd) {
|
|
spin_lock_bh(&conn->send_lock);
|
|
rc = smcd_cdc_msg_send(conn);
|
|
spin_unlock_bh(&conn->send_lock);
|
|
} else {
|
|
rc = smcr_cdc_get_slot_and_msg_send(conn);
|
|
}
|
|
|
|
return rc;
|
|
}
|
|
|
|
void smc_cdc_wait_pend_tx_wr(struct smc_connection *conn)
|
|
{
|
|
wait_event(conn->cdc_pend_tx_wq, !atomic_read(&conn->cdc_pend_tx_wr));
|
|
}
|
|
|
|
/* Send a SMC-D CDC header.
|
|
* This increments the free space available in our send buffer.
|
|
* Also update the confirmed receive buffer with what was sent to the peer.
|
|
*/
|
|
int smcd_cdc_msg_send(struct smc_connection *conn)
|
|
{
|
|
struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
|
|
union smc_host_cursor curs;
|
|
struct smcd_cdc_msg cdc;
|
|
int rc, diff;
|
|
|
|
memset(&cdc, 0, sizeof(cdc));
|
|
cdc.common.type = SMC_CDC_MSG_TYPE;
|
|
curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.prod.acurs);
|
|
cdc.prod.wrap = curs.wrap;
|
|
cdc.prod.count = curs.count;
|
|
curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.cons.acurs);
|
|
cdc.cons.wrap = curs.wrap;
|
|
cdc.cons.count = curs.count;
|
|
cdc.cons.prod_flags = conn->local_tx_ctrl.prod_flags;
|
|
cdc.cons.conn_state_flags = conn->local_tx_ctrl.conn_state_flags;
|
|
rc = smcd_tx_ism_write(conn, &cdc, sizeof(cdc), 0, 1);
|
|
if (rc)
|
|
return rc;
|
|
smc_curs_copy(&conn->rx_curs_confirmed, &curs, conn);
|
|
conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
|
|
/* Calculate transmitted data and increment free send buffer space */
|
|
diff = smc_curs_diff(conn->sndbuf_desc->len, &conn->tx_curs_fin,
|
|
&conn->tx_curs_sent);
|
|
/* increased by confirmed number of bytes */
|
|
smp_mb__before_atomic();
|
|
atomic_add(diff, &conn->sndbuf_space);
|
|
/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
|
|
smp_mb__after_atomic();
|
|
smc_curs_copy(&conn->tx_curs_fin, &conn->tx_curs_sent, conn);
|
|
|
|
smc_tx_sndbuf_nonfull(smc);
|
|
return rc;
|
|
}
|
|
|
|
/********************************* receive ***********************************/
|
|
|
|
static inline bool smc_cdc_before(u16 seq1, u16 seq2)
|
|
{
|
|
return (s16)(seq1 - seq2) < 0;
|
|
}
|
|
|
|
static void smc_cdc_handle_urg_data_arrival(struct smc_sock *smc,
|
|
int *diff_prod)
|
|
{
|
|
struct smc_connection *conn = &smc->conn;
|
|
char *base;
|
|
|
|
/* new data included urgent business */
|
|
smc_curs_copy(&conn->urg_curs, &conn->local_rx_ctrl.prod, conn);
|
|
conn->urg_state = SMC_URG_VALID;
|
|
if (!sock_flag(&smc->sk, SOCK_URGINLINE))
|
|
/* we'll skip the urgent byte, so don't account for it */
|
|
(*diff_prod)--;
|
|
base = (char *)conn->rmb_desc->cpu_addr + conn->rx_off;
|
|
if (conn->urg_curs.count)
|
|
conn->urg_rx_byte = *(base + conn->urg_curs.count - 1);
|
|
else
|
|
conn->urg_rx_byte = *(base + conn->rmb_desc->len - 1);
|
|
sk_send_sigurg(&smc->sk);
|
|
}
|
|
|
|
static void smc_cdc_msg_validate(struct smc_sock *smc, struct smc_cdc_msg *cdc,
|
|
struct smc_link *link)
|
|
{
|
|
struct smc_connection *conn = &smc->conn;
|
|
u16 recv_seq = ntohs(cdc->seqno);
|
|
s16 diff;
|
|
|
|
/* check that seqnum was seen before */
|
|
diff = conn->local_rx_ctrl.seqno - recv_seq;
|
|
if (diff < 0) { /* diff larger than 0x7fff */
|
|
/* drop connection */
|
|
conn->out_of_sync = 1; /* prevent any further receives */
|
|
spin_lock_bh(&conn->send_lock);
|
|
conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
|
|
conn->lnk = link;
|
|
spin_unlock_bh(&conn->send_lock);
|
|
sock_hold(&smc->sk); /* sock_put in abort_work */
|
|
if (!queue_work(smc_close_wq, &conn->abort_work))
|
|
sock_put(&smc->sk);
|
|
}
|
|
}
|
|
|
|
static void smc_cdc_msg_recv_action(struct smc_sock *smc,
|
|
struct smc_cdc_msg *cdc)
|
|
{
|
|
union smc_host_cursor cons_old, prod_old;
|
|
struct smc_connection *conn = &smc->conn;
|
|
int diff_cons, diff_prod;
|
|
|
|
smc_curs_copy(&prod_old, &conn->local_rx_ctrl.prod, conn);
|
|
smc_curs_copy(&cons_old, &conn->local_rx_ctrl.cons, conn);
|
|
smc_cdc_msg_to_host(&conn->local_rx_ctrl, cdc, conn);
|
|
|
|
diff_cons = smc_curs_diff(conn->peer_rmbe_size, &cons_old,
|
|
&conn->local_rx_ctrl.cons);
|
|
if (diff_cons) {
|
|
/* peer_rmbe_space is decreased during data transfer with RDMA
|
|
* write
|
|
*/
|
|
smp_mb__before_atomic();
|
|
atomic_add(diff_cons, &conn->peer_rmbe_space);
|
|
/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
|
|
smp_mb__after_atomic();
|
|
}
|
|
|
|
diff_prod = smc_curs_diff(conn->rmb_desc->len, &prod_old,
|
|
&conn->local_rx_ctrl.prod);
|
|
if (diff_prod) {
|
|
if (conn->local_rx_ctrl.prod_flags.urg_data_present)
|
|
smc_cdc_handle_urg_data_arrival(smc, &diff_prod);
|
|
/* bytes_to_rcv is decreased in smc_recvmsg */
|
|
smp_mb__before_atomic();
|
|
atomic_add(diff_prod, &conn->bytes_to_rcv);
|
|
/* guarantee 0 <= bytes_to_rcv <= rmb_desc->len */
|
|
smp_mb__after_atomic();
|
|
smc->sk.sk_data_ready(&smc->sk);
|
|
} else {
|
|
if (conn->local_rx_ctrl.prod_flags.write_blocked)
|
|
smc->sk.sk_data_ready(&smc->sk);
|
|
if (conn->local_rx_ctrl.prod_flags.urg_data_pending)
|
|
conn->urg_state = SMC_URG_NOTYET;
|
|
}
|
|
|
|
/* 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) {
|
|
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) {
|
|
/* urg data confirmed by peer, indicate we're ready for more */
|
|
conn->urg_tx_pend = false;
|
|
smc->sk.sk_write_space(&smc->sk);
|
|
}
|
|
|
|
if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
|
|
smc->sk.sk_err = ECONNRESET;
|
|
conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
|
|
}
|
|
if (smc_cdc_rxed_any_close_or_senddone(conn)) {
|
|
smc->sk.sk_shutdown |= RCV_SHUTDOWN;
|
|
if (smc->clcsock && smc->clcsock->sk)
|
|
smc->clcsock->sk->sk_shutdown |= RCV_SHUTDOWN;
|
|
sock_set_flag(&smc->sk, SOCK_DONE);
|
|
sock_hold(&smc->sk); /* sock_put in close_work */
|
|
if (!queue_work(smc_close_wq, &conn->close_work))
|
|
sock_put(&smc->sk);
|
|
}
|
|
}
|
|
|
|
/* called under tasklet context */
|
|
static void smc_cdc_msg_recv(struct smc_sock *smc, struct smc_cdc_msg *cdc)
|
|
{
|
|
sock_hold(&smc->sk);
|
|
bh_lock_sock(&smc->sk);
|
|
smc_cdc_msg_recv_action(smc, cdc);
|
|
bh_unlock_sock(&smc->sk);
|
|
sock_put(&smc->sk); /* no free sk in softirq-context */
|
|
}
|
|
|
|
/* Schedule a tasklet for this connection. Triggered from the ISM device IRQ
|
|
* handler to indicate update in the DMBE.
|
|
*
|
|
* Context:
|
|
* - tasklet context
|
|
*/
|
|
static void smcd_cdc_rx_tsklet(struct tasklet_struct *t)
|
|
{
|
|
struct smc_connection *conn = from_tasklet(conn, t, rx_tsklet);
|
|
struct smcd_cdc_msg *data_cdc;
|
|
struct smcd_cdc_msg cdc;
|
|
struct smc_sock *smc;
|
|
|
|
if (!conn || conn->killed)
|
|
return;
|
|
|
|
data_cdc = (struct smcd_cdc_msg *)conn->rmb_desc->cpu_addr;
|
|
smcd_curs_copy(&cdc.prod, &data_cdc->prod, conn);
|
|
smcd_curs_copy(&cdc.cons, &data_cdc->cons, conn);
|
|
smc = container_of(conn, struct smc_sock, conn);
|
|
smc_cdc_msg_recv(smc, (struct smc_cdc_msg *)&cdc);
|
|
}
|
|
|
|
/* Initialize receive tasklet. Called from ISM device IRQ handler to start
|
|
* receiver side.
|
|
*/
|
|
void smcd_cdc_rx_init(struct smc_connection *conn)
|
|
{
|
|
tasklet_setup(&conn->rx_tsklet, smcd_cdc_rx_tsklet);
|
|
}
|
|
|
|
/***************************** init, exit, misc ******************************/
|
|
|
|
static void smc_cdc_rx_handler(struct ib_wc *wc, void *buf)
|
|
{
|
|
struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
|
|
struct smc_cdc_msg *cdc = buf;
|
|
struct smc_connection *conn;
|
|
struct smc_link_group *lgr;
|
|
struct smc_sock *smc;
|
|
|
|
if (wc->byte_len < offsetof(struct smc_cdc_msg, reserved))
|
|
return; /* short message */
|
|
if (cdc->len != SMC_WR_TX_SIZE)
|
|
return; /* invalid message */
|
|
|
|
/* lookup connection */
|
|
lgr = smc_get_lgr(link);
|
|
read_lock_bh(&lgr->conns_lock);
|
|
conn = smc_lgr_find_conn(ntohl(cdc->token), lgr);
|
|
read_unlock_bh(&lgr->conns_lock);
|
|
if (!conn || conn->out_of_sync)
|
|
return;
|
|
smc = container_of(conn, struct smc_sock, conn);
|
|
|
|
if (cdc->prod_flags.failover_validation) {
|
|
smc_cdc_msg_validate(smc, cdc, link);
|
|
return;
|
|
}
|
|
if (smc_cdc_before(ntohs(cdc->seqno),
|
|
conn->local_rx_ctrl.seqno))
|
|
/* received seqno is old */
|
|
return;
|
|
|
|
smc_cdc_msg_recv(smc, cdc);
|
|
}
|
|
|
|
static struct smc_wr_rx_handler smc_cdc_rx_handlers[] = {
|
|
{
|
|
.handler = smc_cdc_rx_handler,
|
|
.type = SMC_CDC_MSG_TYPE
|
|
},
|
|
{
|
|
.handler = NULL,
|
|
}
|
|
};
|
|
|
|
int __init smc_cdc_init(void)
|
|
{
|
|
struct smc_wr_rx_handler *handler;
|
|
int rc = 0;
|
|
|
|
for (handler = smc_cdc_rx_handlers; handler->handler; handler++) {
|
|
INIT_HLIST_NODE(&handler->list);
|
|
rc = smc_wr_rx_register_handler(handler);
|
|
if (rc)
|
|
break;
|
|
}
|
|
return rc;
|
|
}
|