[mirror_ubuntu-bionic-kernel.git] / net / smc / smc_tx.c

/*
 * Shared Memory Communications over RDMA (SMC-R) and RoCE
 *
 * Manage send buffer.
 * Producer:
 * Copy user space data into send buffer, if send buffer space available.
 * Consumer:
 * Trigger RDMA write into RMBE of peer and send CDC, if RMBE space available.
 *
 * Copyright IBM Corp. 2016
 *
 * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
 */

#include <linux/net.h>
#include <linux/rcupdate.h>
#include <linux/workqueue.h>
#include <linux/sched/signal.h>

#include <net/sock.h>

#include "smc.h"
#include "smc_wr.h"
#include "smc_cdc.h"
#include "smc_tx.h"

/***************************** sndbuf producer *******************************/

/* callback implementation for sk.sk_write_space()
 * to wakeup sndbuf producers that blocked with smc_tx_wait_memory().
 * called under sk_socket lock.
 */
static void smc_tx_write_space(struct sock *sk)
{
	struct socket *sock = sk->sk_socket;
	struct smc_sock *smc = smc_sk(sk);
	struct socket_wq *wq;

	/* similar to sk_stream_write_space */
	if (atomic_read(&smc->conn.sndbuf_space) && sock) {
		clear_bit(SOCK_NOSPACE, &sock->flags);
		rcu_read_lock();
		wq = rcu_dereference(sk->sk_wq);
		if (skwq_has_sleeper(wq))
			wake_up_interruptible_poll(&wq->wait,
						   POLLOUT | POLLWRNORM |
						   POLLWRBAND);
		if (wq && wq->fasync_list && !(sk->sk_shutdown & SEND_SHUTDOWN))
			sock_wake_async(wq, SOCK_WAKE_SPACE, POLL_OUT);
		rcu_read_unlock();
	}
}

/* Wakeup sndbuf producers that blocked with smc_tx_wait_memory().
 * Cf. tcp_data_snd_check()=>tcp_check_space()=>tcp_new_space().
 */
void smc_tx_sndbuf_nonfull(struct smc_sock *smc)
{
	if (smc->sk.sk_socket &&
	    test_bit(SOCK_NOSPACE, &smc->sk.sk_socket->flags))
		smc->sk.sk_write_space(&smc->sk);
}

/* blocks sndbuf producer until at least one byte of free space available */
static int smc_tx_wait_memory(struct smc_sock *smc, int flags)
{
	DEFINE_WAIT_FUNC(wait, woken_wake_function);
	struct smc_connection *conn = &smc->conn;
	struct sock *sk = &smc->sk;
	bool noblock;
	long timeo;
	int rc = 0;

	/* similar to sk_stream_wait_memory */
	timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
	noblock = timeo ? false : true;
	add_wait_queue(sk_sleep(sk), &wait);
	while (1) {
		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
		if (sk->sk_err ||
		    (sk->sk_shutdown & SEND_SHUTDOWN) ||
		    conn->local_tx_ctrl.conn_state_flags.peer_done_writing) {
			rc = -EPIPE;
			break;
		}
		if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
			rc = -ECONNRESET;
			break;
		}
		if (!timeo) {
			if (noblock)
				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
			rc = -EAGAIN;
			break;
		}
		if (signal_pending(current)) {
			rc = sock_intr_errno(timeo);
			break;
		}
		sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
		if (atomic_read(&conn->sndbuf_space))
			break; /* at least 1 byte of free space available */
		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
		sk->sk_write_pending++;
		sk_wait_event(sk, &timeo,
			      sk->sk_err ||
			      (sk->sk_shutdown & SEND_SHUTDOWN) ||
			      smc_cdc_rxed_any_close_or_senddone(conn) ||
			      atomic_read(&conn->sndbuf_space),
			      &wait);
		sk->sk_write_pending--;
	}
	remove_wait_queue(sk_sleep(sk), &wait);
	return rc;
}

/* sndbuf producer: main API called by socket layer.
 * called under sock lock.
 */
int smc_tx_sendmsg(struct smc_sock *smc, struct msghdr *msg, size_t len)
{
	size_t copylen, send_done = 0, send_remaining = len;
	size_t chunk_len, chunk_off, chunk_len_sum;
	struct smc_connection *conn = &smc->conn;
	union smc_host_cursor prep;
	struct sock *sk = &smc->sk;
	char *sndbuf_base;
	int tx_cnt_prep;
	int writespace;
	int rc, chunk;

	/* This should be in poll */
	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);

	if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
		rc = -EPIPE;
		goto out_err;
	}

	while (msg_data_left(msg)) {
		if (sk->sk_state == SMC_INIT)
			return -ENOTCONN;
		if (smc->sk.sk_shutdown & SEND_SHUTDOWN ||
		    (smc->sk.sk_err == ECONNABORTED) ||
		    conn->local_tx_ctrl.conn_state_flags.peer_conn_abort)
			return -EPIPE;
		if (smc_cdc_rxed_any_close(conn))
			return send_done ?: -ECONNRESET;

		if (!atomic_read(&conn->sndbuf_space)) {
			rc = smc_tx_wait_memory(smc, msg->msg_flags);
			if (rc) {
				if (send_done)
					return send_done;
				goto out_err;
			}
			continue;
		}

		/* initialize variables for 1st iteration of subsequent loop */
		/* could be just 1 byte, even after smc_tx_wait_memory above */
		writespace = atomic_read(&conn->sndbuf_space);
		/* not more than what user space asked for */
		copylen = min_t(size_t, send_remaining, writespace);
		/* determine start of sndbuf */
		sndbuf_base = conn->sndbuf_desc->cpu_addr;
		smc_curs_write(&prep,
			       smc_curs_read(&conn->tx_curs_prep, conn),
			       conn);
		tx_cnt_prep = prep.count;
		/* determine chunks where to write into sndbuf */
		/* either unwrapped case, or 1st chunk of wrapped case */
		chunk_len = min_t(size_t,
				  copylen, conn->sndbuf_size - tx_cnt_prep);
		chunk_len_sum = chunk_len;
		chunk_off = tx_cnt_prep;
		smc_sndbuf_sync_sg_for_cpu(conn);
		for (chunk = 0; chunk < 2; chunk++) {
			rc = memcpy_from_msg(sndbuf_base + chunk_off,
					     msg, chunk_len);
			if (rc) {
				smc_sndbuf_sync_sg_for_device(conn);
				if (send_done)
					return send_done;
				goto out_err;
			}
			send_done += chunk_len;
			send_remaining -= chunk_len;

			if (chunk_len_sum == copylen)
				break; /* either on 1st or 2nd iteration */
			/* prepare next (== 2nd) iteration */
			chunk_len = copylen - chunk_len; /* remainder */
			chunk_len_sum += chunk_len;
			chunk_off = 0; /* modulo offset in send ring buffer */
		}
		smc_sndbuf_sync_sg_for_device(conn);
		/* update cursors */
		smc_curs_add(conn->sndbuf_size, &prep, copylen);
		smc_curs_write(&conn->tx_curs_prep,
			       smc_curs_read(&prep, conn),
			       conn);
		/* increased in send tasklet smc_cdc_tx_handler() */
		smp_mb__before_atomic();
		atomic_sub(copylen, &conn->sndbuf_space);
		/* guarantee 0 <= sndbuf_space <= sndbuf_size */
		smp_mb__after_atomic();
		/* since we just produced more new data into sndbuf,
		 * trigger sndbuf consumer: RDMA write into peer RMBE and CDC
		 */
		smc_tx_sndbuf_nonempty(conn);
	} /* while (msg_data_left(msg)) */

	return send_done;

out_err:
	rc = sk_stream_error(sk, msg->msg_flags, rc);
	/* make sure we wake any epoll edge trigger waiter */
	if (unlikely(rc == -EAGAIN))
		sk->sk_write_space(sk);
	return rc;
}

/***************************** sndbuf consumer *******************************/

/* sndbuf consumer: actual data transfer of one target chunk with RDMA write */
static int smc_tx_rdma_write(struct smc_connection *conn, int peer_rmbe_offset,
			     int num_sges, struct ib_sge sges[])
{
	struct smc_link_group *lgr = conn->lgr;
	struct ib_send_wr *failed_wr = NULL;
	struct ib_rdma_wr rdma_wr;
	struct smc_link *link;
	int rc;

	memset(&rdma_wr, 0, sizeof(rdma_wr));
	link = &lgr->lnk[SMC_SINGLE_LINK];
	rdma_wr.wr.wr_id = smc_wr_tx_get_next_wr_id(link);
	rdma_wr.wr.sg_list = sges;
	rdma_wr.wr.num_sge = num_sges;
	rdma_wr.wr.opcode = IB_WR_RDMA_WRITE;
	rdma_wr.remote_addr =
		lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].dma_addr +
		/* RMBE within RMB */
		((conn->peer_conn_idx - 1) * conn->peer_rmbe_size) +
		/* offset within RMBE */
		peer_rmbe_offset;
	rdma_wr.rkey = lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].rkey;
	rc = ib_post_send(link->roce_qp, &rdma_wr.wr, &failed_wr);
	if (rc)
		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
	return rc;
}

/* sndbuf consumer */
static inline void smc_tx_advance_cursors(struct smc_connection *conn,
					  union smc_host_cursor *prod,
					  union smc_host_cursor *sent,
					  size_t len)
{
	smc_curs_add(conn->peer_rmbe_size, prod, len);
	/* increased in recv tasklet smc_cdc_msg_rcv() */
	smp_mb__before_atomic();
	/* data in flight reduces usable snd_wnd */
	atomic_sub(len, &conn->peer_rmbe_space);
	/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
	smp_mb__after_atomic();
	smc_curs_add(conn->sndbuf_size, sent, len);
}

/* sndbuf consumer: prepare all necessary (src&dst) chunks of data transmit;
 * usable snd_wnd as max transmit
 */
static int smc_tx_rdma_writes(struct smc_connection *conn)
{
	size_t src_off, src_len, dst_off, dst_len; /* current chunk values */
	size_t len, dst_len_sum, src_len_sum, dstchunk, srcchunk;
	union smc_host_cursor sent, prep, prod, cons;
	struct ib_sge sges[SMC_IB_MAX_SEND_SGE];
	struct smc_link_group *lgr = conn->lgr;
	int to_send, rmbespace;
	struct smc_link *link;
	dma_addr_t dma_addr;
	int num_sges;
	int rc;

	/* source: sndbuf */
	smc_curs_write(&sent, smc_curs_read(&conn->tx_curs_sent, conn), conn);
	smc_curs_write(&prep, smc_curs_read(&conn->tx_curs_prep, conn), conn);
	/* cf. wmem_alloc - (snd_max - snd_una) */
	to_send = smc_curs_diff(conn->sndbuf_size, &sent, &prep);
	if (to_send <= 0)
		return 0;

	/* destination: RMBE */
	/* cf. snd_wnd */
	rmbespace = atomic_read(&conn->peer_rmbe_space);
	if (rmbespace <= 0)
		return 0;
	smc_curs_write(&prod,
		       smc_curs_read(&conn->local_tx_ctrl.prod, conn),
		       conn);
	smc_curs_write(&cons,
		       smc_curs_read(&conn->local_rx_ctrl.cons, conn),
		       conn);

	/* if usable snd_wnd closes ask peer to advertise once it opens again */
	conn->local_tx_ctrl.prod_flags.write_blocked = (to_send >= rmbespace);
	/* cf. usable snd_wnd */
	len = min(to_send, rmbespace);

	/* initialize variables for first iteration of subsequent nested loop */
	link = &lgr->lnk[SMC_SINGLE_LINK];
	dst_off = prod.count;
	if (prod.wrap == cons.wrap) {
		/* the filled destination area is unwrapped,
		 * hence the available free destination space is wrapped
		 * and we need 2 destination chunks of sum len; start with 1st
		 * which is limited by what's available in sndbuf
		 */
		dst_len = min_t(size_t,
				conn->peer_rmbe_size - prod.count, len);
	} else {
		/* the filled destination area is wrapped,
		 * hence the available free destination space is unwrapped
		 * and we need a single destination chunk of entire len
		 */
		dst_len = len;
	}
	dst_len_sum = dst_len;
	src_off = sent.count;
	/* dst_len determines the maximum src_len */
	if (sent.count + dst_len <= conn->sndbuf_size) {
		/* unwrapped src case: single chunk of entire dst_len */
		src_len = dst_len;
	} else {
		/* wrapped src case: 2 chunks of sum dst_len; start with 1st: */
		src_len = conn->sndbuf_size - sent.count;
	}
	src_len_sum = src_len;
	dma_addr = sg_dma_address(conn->sndbuf_desc->sgt[SMC_SINGLE_LINK].sgl);
	for (dstchunk = 0; dstchunk < 2; dstchunk++) {
		num_sges = 0;
		for (srcchunk = 0; srcchunk < 2; srcchunk++) {
			sges[srcchunk].addr = dma_addr + src_off;
			sges[srcchunk].length = src_len;
			sges[srcchunk].lkey = link->roce_pd->local_dma_lkey;
			num_sges++;
			src_off += src_len;
			if (src_off >= conn->sndbuf_size)
				src_off -= conn->sndbuf_size;
						/* modulo in send ring */
			if (src_len_sum == dst_len)
				break; /* either on 1st or 2nd iteration */
			/* prepare next (== 2nd) iteration */
			src_len = dst_len - src_len; /* remainder */
			src_len_sum += src_len;
		}
		rc = smc_tx_rdma_write(conn, dst_off, num_sges, sges);
		if (rc)
			return rc;
		if (dst_len_sum == len)
			break; /* either on 1st or 2nd iteration */
		/* prepare next (== 2nd) iteration */
		dst_off = 0; /* modulo offset in RMBE ring buffer */
		dst_len = len - dst_len; /* remainder */
		dst_len_sum += dst_len;
		src_len = min_t(int,
				dst_len, conn->sndbuf_size - sent.count);
		src_len_sum = src_len;
	}

	smc_tx_advance_cursors(conn, &prod, &sent, len);
	/* update connection's cursors with advanced local cursors */
	smc_curs_write(&conn->local_tx_ctrl.prod,
		       smc_curs_read(&prod, conn),
		       conn);
							/* dst: peer RMBE */
	smc_curs_write(&conn->tx_curs_sent,
		       smc_curs_read(&sent, conn),
		       conn);
							/* src: local sndbuf */

	return 0;
}

/* Wakeup sndbuf consumers from any context (IRQ or process)
 * since there is more data to transmit; usable snd_wnd as max transmit
 */
int smc_tx_sndbuf_nonempty(struct smc_connection *conn)
{
	struct smc_cdc_tx_pend *pend;
	struct smc_wr_buf *wr_buf;
	int rc;

	spin_lock_bh(&conn->send_lock);
	rc = smc_cdc_get_free_slot(&conn->lgr->lnk[SMC_SINGLE_LINK], &wr_buf,
				   &pend);
	if (rc < 0) {
		if (rc == -EBUSY) {
			struct smc_sock *smc =
				container_of(conn, struct smc_sock, conn);

			if (smc->sk.sk_err == ECONNABORTED) {
				rc = sock_error(&smc->sk);
				goto out_unlock;
			}
			rc = 0;
			schedule_work(&conn->tx_work);
		}
		goto out_unlock;
	}

	rc = smc_tx_rdma_writes(conn);
	if (rc) {
		smc_wr_tx_put_slot(&conn->lgr->lnk[SMC_SINGLE_LINK],
				   (struct smc_wr_tx_pend_priv *)pend);
		goto out_unlock;
	}

	rc = smc_cdc_msg_send(conn, wr_buf, pend);

out_unlock:
	spin_unlock_bh(&conn->send_lock);
	return rc;
}

/* Wakeup sndbuf consumers from process context
 * since there is more data to transmit
 */
static void smc_tx_work(struct work_struct *work)
{
	struct smc_connection *conn = container_of(work,
						   struct smc_connection,
						   tx_work);
	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
	int rc;

	lock_sock(&smc->sk);
	rc = smc_tx_sndbuf_nonempty(conn);
	if (!rc && conn->local_rx_ctrl.prod_flags.write_blocked &&
	    !atomic_read(&conn->bytes_to_rcv))
		conn->local_rx_ctrl.prod_flags.write_blocked = 0;
	release_sock(&smc->sk);
}

void smc_tx_consumer_update(struct smc_connection *conn)
{
	union smc_host_cursor cfed, cons;
	struct smc_cdc_tx_pend *pend;
	struct smc_wr_buf *wr_buf;
	int to_confirm, rc;

	smc_curs_write(&cons,
		       smc_curs_read(&conn->local_tx_ctrl.cons, conn),
		       conn);
	smc_curs_write(&cfed,
		       smc_curs_read(&conn->rx_curs_confirmed, conn),
		       conn);
	to_confirm = smc_curs_diff(conn->rmbe_size, &cfed, &cons);

	if (conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||
	    ((to_confirm > conn->rmbe_update_limit) &&
	     ((to_confirm > (conn->rmbe_size / 2)) ||
	      conn->local_rx_ctrl.prod_flags.write_blocked))) {
		rc = smc_cdc_get_free_slot(&conn->lgr->lnk[SMC_SINGLE_LINK],
					   &wr_buf, &pend);
		if (!rc)
			rc = smc_cdc_msg_send(conn, wr_buf, pend);
		if (rc < 0) {
			schedule_work(&conn->tx_work);
			return;
		}
		smc_curs_write(&conn->rx_curs_confirmed,
			       smc_curs_read(&conn->local_tx_ctrl.cons, conn),
			       conn);
		conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
	}
	if (conn->local_rx_ctrl.prod_flags.write_blocked &&
	    !atomic_read(&conn->bytes_to_rcv))
		conn->local_rx_ctrl.prod_flags.write_blocked = 0;
}

/***************************** send initialize *******************************/

/* Initialize send properties on connection establishment. NB: not __init! */
void smc_tx_init(struct smc_sock *smc)
{
	smc->sk.sk_write_space = smc_tx_write_space;
	INIT_WORK(&smc->conn.tx_work, smc_tx_work);
	spin_lock_init(&smc->conn.send_lock);
}
Commit	Line	Data
e6727f39 UB	1	/*
	2	* Shared Memory Communications over RDMA (SMC-R) and RoCE
	3	*
	4	* Manage send buffer.
	5	* Producer:
	6	* Copy user space data into send buffer, if send buffer space available.
	7	* Consumer:
	8	* Trigger RDMA write into RMBE of peer and send CDC, if RMBE space available.
	9	*
	10	* Copyright IBM Corp. 2016
	11	*
	12	* Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
	13	*/
	14
	15	#include <linux/net.h>
	16	#include <linux/rcupdate.h>
	17	#include <linux/workqueue.h>
c3edc401 IM	18	#include <linux/sched/signal.h>
c3edc401 IM	19
e6727f39 UB	20	#include <net/sock.h>
	21
	22	#include "smc.h"
	23	#include "smc_wr.h"
	24	#include "smc_cdc.h"
	25	#include "smc_tx.h"
	26
	27	/*************************** sndbuf producer *****************************/
	28
	29	/* callback implementation for sk.sk_write_space()
	30	* to wakeup sndbuf producers that blocked with smc_tx_wait_memory().
	31	* called under sk_socket lock.
	32	*/
	33	static void smc_tx_write_space(struct sock *sk)
	34	{
	35	struct socket *sock = sk->sk_socket;
	36	struct smc_sock *smc = smc_sk(sk);
	37	struct socket_wq *wq;
	38
	39	/* similar to sk_stream_write_space */
	40	if (atomic_read(&smc->conn.sndbuf_space) && sock) {
	41	clear_bit(SOCK_NOSPACE, &sock->flags);
	42	rcu_read_lock();
	43	wq = rcu_dereference(sk->sk_wq);
	44	if (skwq_has_sleeper(wq))
	45	wake_up_interruptible_poll(&wq->wait,
	46	POLLOUT \| POLLWRNORM \|
	47	POLLWRBAND);
	48	if (wq && wq->fasync_list && !(sk->sk_shutdown & SEND_SHUTDOWN))
	49	sock_wake_async(wq, SOCK_WAKE_SPACE, POLL_OUT);
	50	rcu_read_unlock();
	51	}
	52	}
	53
	54	/* Wakeup sndbuf producers that blocked with smc_tx_wait_memory().
	55	* Cf. tcp_data_snd_check()=>tcp_check_space()=>tcp_new_space().
	56	*/
	57	void smc_tx_sndbuf_nonfull(struct smc_sock *smc)
	58	{
	59	if (smc->sk.sk_socket &&
	60	test_bit(SOCK_NOSPACE, &smc->sk.sk_socket->flags))
	61	smc->sk.sk_write_space(&smc->sk);
	62	}
	63
	64	/* blocks sndbuf producer until at least one byte of free space available */
	65	static int smc_tx_wait_memory(struct smc_sock *smc, int flags)
	66	{
	67	DEFINE_WAIT_FUNC(wait, woken_wake_function);
	68	struct smc_connection *conn = &smc->conn;
	69	struct sock *sk = &smc->sk;
	70	bool noblock;
	71	long timeo;
	72	int rc = 0;
	73
	74	/* similar to sk_stream_wait_memory */
	75	timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
	76	noblock = timeo ? false : true;
	77	add_wait_queue(sk_sleep(sk), &wait);
	78	while (1) {
	79	sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
	80	if (sk->sk_err \|\|
	81	(sk->sk_shutdown & SEND_SHUTDOWN) \|\|
	82	conn->local_tx_ctrl.conn_state_flags.peer_done_writing) {
	83	rc = -EPIPE;
84	break;
85	}
86	if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
87	rc = -ECONNRESET;
88	break;
89	}
90	if (!timeo) {
91	if (noblock)
92	set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
93	rc = -EAGAIN;
94	break;
95	}
96	if (signal_pending(current)) {
97	rc = sock_intr_errno(timeo);
98	break;
99	}
100	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
101	if (atomic_read(&conn->sndbuf_space))
102	break; /* at least 1 byte of free space available */
103	set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
104	sk->sk_write_pending++;
105	sk_wait_event(sk, &timeo,
106	sk->sk_err \|\|
107	(sk->sk_shutdown & SEND_SHUTDOWN) \|\|
108	smc_cdc_rxed_any_close_or_senddone(conn) \|\|
109	atomic_read(&conn->sndbuf_space),
110	&wait);
111	sk->sk_write_pending--;
112	}
113	remove_wait_queue(sk_sleep(sk), &wait);
114	return rc;
115	}
116
117	/* sndbuf producer: main API called by socket layer.
118	* called under sock lock.
119	*/
120	int smc_tx_sendmsg(struct smc_sock smc, struct msghdr msg, size_t len)
121	{
122	size_t copylen, send_done = 0, send_remaining = len;
123	size_t chunk_len, chunk_off, chunk_len_sum;
124	struct smc_connection *conn = &smc->conn;
125	union smc_host_cursor prep;
126	struct sock *sk = &smc->sk;
127	char *sndbuf_base;
128	int tx_cnt_prep;
129	int writespace;
130	int rc, chunk;
131
132	/* This should be in poll */
133	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
134
135	if (sk->sk_err \|\| (sk->sk_shutdown & SEND_SHUTDOWN)) {
136	rc = -EPIPE;
137	goto out_err;
138	}
139
140	while (msg_data_left(msg)) {
141	if (sk->sk_state == SMC_INIT)
142	return -ENOTCONN;
143	if (smc->sk.sk_shutdown & SEND_SHUTDOWN \|\|
b38d7324	144	(smc->sk.sk_err == ECONNABORTED) \|\|
e6727f39 UB	145	conn->local_tx_ctrl.conn_state_flags.peer_conn_abort)
	146	return -EPIPE;
	147	if (smc_cdc_rxed_any_close(conn))
	148	return send_done ?: -ECONNRESET;
	149
	150	if (!atomic_read(&conn->sndbuf_space)) {
	151	rc = smc_tx_wait_memory(smc, msg->msg_flags);
	152	if (rc) {
	153	if (send_done)
	154	return send_done;
	155	goto out_err;
	156	}
	157	continue;
	158	}
	159
	160	/* initialize variables for 1st iteration of subsequent loop */
	161	/* could be just 1 byte, even after smc_tx_wait_memory above */
	162	writespace = atomic_read(&conn->sndbuf_space);
	163	/* not more than what user space asked for */
	164	copylen = min_t(size_t, send_remaining, writespace);
	165	/* determine start of sndbuf */
	166	sndbuf_base = conn->sndbuf_desc->cpu_addr;
	167	smc_curs_write(&prep,
	168	smc_curs_read(&conn->tx_curs_prep, conn),
	169	conn);
	170	tx_cnt_prep = prep.count;
	171	/* determine chunks where to write into sndbuf */
	172	/* either unwrapped case, or 1st chunk of wrapped case */
	173	chunk_len = min_t(size_t,
	174	copylen, conn->sndbuf_size - tx_cnt_prep);
	175	chunk_len_sum = chunk_len;
	176	chunk_off = tx_cnt_prep;
10428dd8	177	smc_sndbuf_sync_sg_for_cpu(conn);
e6727f39 UB	178	for (chunk = 0; chunk < 2; chunk++) {
	179	rc = memcpy_from_msg(sndbuf_base + chunk_off,
	180	msg, chunk_len);
	181	if (rc) {
10428dd8	182	smc_sndbuf_sync_sg_for_device(conn);
e6727f39 UB	183	if (send_done)
	184	return send_done;
	185	goto out_err;
	186	}
	187	send_done += chunk_len;
	188	send_remaining -= chunk_len;
	189
	190	if (chunk_len_sum == copylen)
	191	break; /* either on 1st or 2nd iteration */
	192	/* prepare next (== 2nd) iteration */
	193	chunk_len = copylen - chunk_len; /* remainder */
	194	chunk_len_sum += chunk_len;
	195	chunk_off = 0; /* modulo offset in send ring buffer */
	196	}
10428dd8	197	smc_sndbuf_sync_sg_for_device(conn);
e6727f39 UB	198	/* update cursors */
	199	smc_curs_add(conn->sndbuf_size, &prep, copylen);
	200	smc_curs_write(&conn->tx_curs_prep,
	201	smc_curs_read(&prep, conn),
	202	conn);
	203	/* increased in send tasklet smc_cdc_tx_handler() */
	204	smp_mb__before_atomic();
	205	atomic_sub(copylen, &conn->sndbuf_space);
	206	/* guarantee 0 <= sndbuf_space <= sndbuf_size */
	207	smp_mb__after_atomic();
	208	/* since we just produced more new data into sndbuf,
	209	* trigger sndbuf consumer: RDMA write into peer RMBE and CDC
	210	*/
	211	smc_tx_sndbuf_nonempty(conn);
	212	} /* while (msg_data_left(msg)) */
	213
	214	return send_done;
	215
	216	out_err:
	217	rc = sk_stream_error(sk, msg->msg_flags, rc);
	218	/* make sure we wake any epoll edge trigger waiter */
	219	if (unlikely(rc == -EAGAIN))
	220	sk->sk_write_space(sk);
	221	return rc;
	222	}
	223
	224	/*************************** sndbuf consumer *****************************/
	225
	226	/* sndbuf consumer: actual data transfer of one target chunk with RDMA write */
	227	static int smc_tx_rdma_write(struct smc_connection *conn, int peer_rmbe_offset,
	228	int num_sges, struct ib_sge sges[])
	229	{
	230	struct smc_link_group *lgr = conn->lgr;
	231	struct ib_send_wr *failed_wr = NULL;
	232	struct ib_rdma_wr rdma_wr;
	233	struct smc_link *link;
	234	int rc;
	235
	236	memset(&rdma_wr, 0, sizeof(rdma_wr));
	237	link = &lgr->lnk[SMC_SINGLE_LINK];
	238	rdma_wr.wr.wr_id = smc_wr_tx_get_next_wr_id(link);
	239	rdma_wr.wr.sg_list = sges;
	240	rdma_wr.wr.num_sge = num_sges;
	241	rdma_wr.wr.opcode = IB_WR_RDMA_WRITE;
	242	rdma_wr.remote_addr =
	243	lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].dma_addr +
	244	/* RMBE within RMB */
	245	((conn->peer_conn_idx - 1) * conn->peer_rmbe_size) +
	246	/* offset within RMBE */
	247	peer_rmbe_offset;
	248	rdma_wr.rkey = lgr->rtokens[conn->rtoken_idx][SMC_SINGLE_LINK].rkey;
	249	rc = ib_post_send(link->roce_qp, &rdma_wr.wr, &failed_wr);
	250	if (rc)
	251	conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
	252	return rc;
	253	}
	254
	255	/* sndbuf consumer */
	256	static inline void smc_tx_advance_cursors(struct smc_connection *conn,
	257	union smc_host_cursor *prod,
	258	union smc_host_cursor *sent,
	259	size_t len)
	260	{
	261	smc_curs_add(conn->peer_rmbe_size, prod, len);
262	/* increased in recv tasklet smc_cdc_msg_rcv() */
263	smp_mb__before_atomic();
264	/* data in flight reduces usable snd_wnd */
265	atomic_sub(len, &conn->peer_rmbe_space);
266	/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
267	smp_mb__after_atomic();
268	smc_curs_add(conn->sndbuf_size, sent, len);
269	}
270
271	/* sndbuf consumer: prepare all necessary (src&dst) chunks of data transmit;
272	* usable snd_wnd as max transmit
273	*/
274	static int smc_tx_rdma_writes(struct smc_connection *conn)
275	{
276	size_t src_off, src_len, dst_off, dst_len; /* current chunk values */
277	size_t len, dst_len_sum, src_len_sum, dstchunk, srcchunk;
278	union smc_host_cursor sent, prep, prod, cons;
279	struct ib_sge sges[SMC_IB_MAX_SEND_SGE];
280	struct smc_link_group *lgr = conn->lgr;
281	int to_send, rmbespace;
282	struct smc_link *link;
9d8fb617	283	dma_addr_t dma_addr;
e6727f39 UB	284	int num_sges;
	285	int rc;
	286
	287	/* source: sndbuf */
	288	smc_curs_write(&sent, smc_curs_read(&conn->tx_curs_sent, conn), conn);
	289	smc_curs_write(&prep, smc_curs_read(&conn->tx_curs_prep, conn), conn);
	290	/* cf. wmem_alloc - (snd_max - snd_una) */
	291	to_send = smc_curs_diff(conn->sndbuf_size, &sent, &prep);
	292	if (to_send <= 0)
	293	return 0;
	294
	295	/* destination: RMBE */
	296	/* cf. snd_wnd */
	297	rmbespace = atomic_read(&conn->peer_rmbe_space);
	298	if (rmbespace <= 0)
	299	return 0;
	300	smc_curs_write(&prod,
	301	smc_curs_read(&conn->local_tx_ctrl.prod, conn),
	302	conn);
	303	smc_curs_write(&cons,
	304	smc_curs_read(&conn->local_rx_ctrl.cons, conn),
	305	conn);
	306
	307	/* if usable snd_wnd closes ask peer to advertise once it opens again */
	308	conn->local_tx_ctrl.prod_flags.write_blocked = (to_send >= rmbespace);
	309	/* cf. usable snd_wnd */
	310	len = min(to_send, rmbespace);
	311
	312	/* initialize variables for first iteration of subsequent nested loop */
	313	link = &lgr->lnk[SMC_SINGLE_LINK];
	314	dst_off = prod.count;
	315	if (prod.wrap == cons.wrap) {
	316	/* the filled destination area is unwrapped,
	317	* hence the available free destination space is wrapped
	318	* and we need 2 destination chunks of sum len; start with 1st
	319	* which is limited by what's available in sndbuf
	320	*/
	321	dst_len = min_t(size_t,
	322	conn->peer_rmbe_size - prod.count, len);
	323	} else {
	324	/* the filled destination area is wrapped,
	325	* hence the available free destination space is unwrapped
	326	* and we need a single destination chunk of entire len
	327	*/
	328	dst_len = len;
	329	}
	330	dst_len_sum = dst_len;
	331	src_off = sent.count;
	332	/* dst_len determines the maximum src_len */
	333	if (sent.count + dst_len <= conn->sndbuf_size) {
	334	/* unwrapped src case: single chunk of entire dst_len */
	335	src_len = dst_len;
	336	} else {
	337	/* wrapped src case: 2 chunks of sum dst_len; start with 1st: */
	338	src_len = conn->sndbuf_size - sent.count;
	339	}
	340	src_len_sum = src_len;
9d8fb617	341	dma_addr = sg_dma_address(conn->sndbuf_desc->sgt[SMC_SINGLE_LINK].sgl);
e6727f39 UB	342	for (dstchunk = 0; dstchunk < 2; dstchunk++) {
	343	num_sges = 0;
	344	for (srcchunk = 0; srcchunk < 2; srcchunk++) {
9d8fb617	345	sges[srcchunk].addr = dma_addr + src_off;
e6727f39 UB	346	sges[srcchunk].length = src_len;
	347	sges[srcchunk].lkey = link->roce_pd->local_dma_lkey;
	348	num_sges++;
	349	src_off += src_len;
	350	if (src_off >= conn->sndbuf_size)
	351	src_off -= conn->sndbuf_size;
	352	/* modulo in send ring */
	353	if (src_len_sum == dst_len)
	354	break; /* either on 1st or 2nd iteration */
	355	/* prepare next (== 2nd) iteration */
	356	src_len = dst_len - src_len; /* remainder */
	357	src_len_sum += src_len;
	358	}
	359	rc = smc_tx_rdma_write(conn, dst_off, num_sges, sges);
	360	if (rc)
	361	return rc;
	362	if (dst_len_sum == len)
	363	break; /* either on 1st or 2nd iteration */
	364	/* prepare next (== 2nd) iteration */
	365	dst_off = 0; /* modulo offset in RMBE ring buffer */
	366	dst_len = len - dst_len; /* remainder */
	367	dst_len_sum += dst_len;
	368	src_len = min_t(int,
	369	dst_len, conn->sndbuf_size - sent.count);
	370	src_len_sum = src_len;
	371	}
	372
	373	smc_tx_advance_cursors(conn, &prod, &sent, len);
	374	/* update connection's cursors with advanced local cursors */
	375	smc_curs_write(&conn->local_tx_ctrl.prod,
	376	smc_curs_read(&prod, conn),
	377	conn);
	378	/* dst: peer RMBE */
	379	smc_curs_write(&conn->tx_curs_sent,
	380	smc_curs_read(&sent, conn),
	381	conn);
	382	/* src: local sndbuf */
	383
	384	return 0;
	385	}
	386
	387	/* Wakeup sndbuf consumers from any context (IRQ or process)
	388	* since there is more data to transmit; usable snd_wnd as max transmit
	389	*/
	390	int smc_tx_sndbuf_nonempty(struct smc_connection *conn)
	391	{
	392	struct smc_cdc_tx_pend *pend;
	393	struct smc_wr_buf *wr_buf;
	394	int rc;
	395
	396	spin_lock_bh(&conn->send_lock);
	397	rc = smc_cdc_get_free_slot(&conn->lgr->lnk[SMC_SINGLE_LINK], &wr_buf,
	398	&pend);
	399	if (rc < 0) {
	400	if (rc == -EBUSY) {
b38d7324 UB	401	struct smc_sock *smc =
	402	container_of(conn, struct smc_sock, conn);
	403
	404	if (smc->sk.sk_err == ECONNABORTED) {
	405	rc = sock_error(&smc->sk);
	406	goto out_unlock;
	407	}
e6727f39 UB	408	rc = 0;
	409	schedule_work(&conn->tx_work);
	410	}
	411	goto out_unlock;
	412	}
	413
	414	rc = smc_tx_rdma_writes(conn);
	415	if (rc) {
	416	smc_wr_tx_put_slot(&conn->lgr->lnk[SMC_SINGLE_LINK],
	417	(struct smc_wr_tx_pend_priv *)pend);
	418	goto out_unlock;
	419	}
	420
	421	rc = smc_cdc_msg_send(conn, wr_buf, pend);
	422
	423	out_unlock:
	424	spin_unlock_bh(&conn->send_lock);
	425	return rc;
	426	}
	427
	428	/* Wakeup sndbuf consumers from process context
	429	* since there is more data to transmit
	430	*/
	431	static void smc_tx_work(struct work_struct *work)
	432	{
	433	struct smc_connection *conn = container_of(work,
	434	struct smc_connection,
	435	tx_work);
	436	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
90cacb2e	437	int rc;
e6727f39 UB	438
e6727f39 UB	439	lock_sock(&smc->sk);
90cacb2e UB	440	rc = smc_tx_sndbuf_nonempty(conn);
	441	if (!rc && conn->local_rx_ctrl.prod_flags.write_blocked &&
	442	!atomic_read(&conn->bytes_to_rcv))
	443	conn->local_rx_ctrl.prod_flags.write_blocked = 0;
e6727f39 UB	444	release_sock(&smc->sk);
	445	}
	446
952310cc UB	447	void smc_tx_consumer_update(struct smc_connection *conn)
	448	{
	449	union smc_host_cursor cfed, cons;
	450	struct smc_cdc_tx_pend *pend;
	451	struct smc_wr_buf *wr_buf;
	452	int to_confirm, rc;
	453
	454	smc_curs_write(&cons,
	455	smc_curs_read(&conn->local_tx_ctrl.cons, conn),
	456	conn);
	457	smc_curs_write(&cfed,
	458	smc_curs_read(&conn->rx_curs_confirmed, conn),
	459	conn);
	460	to_confirm = smc_curs_diff(conn->rmbe_size, &cfed, &cons);
	461
	462	if (conn->local_rx_ctrl.prod_flags.cons_curs_upd_req \|\|
	463	((to_confirm > conn->rmbe_update_limit) &&
	464	((to_confirm > (conn->rmbe_size / 2)) \|\|
	465	conn->local_rx_ctrl.prod_flags.write_blocked))) {
	466	rc = smc_cdc_get_free_slot(&conn->lgr->lnk[SMC_SINGLE_LINK],
	467	&wr_buf, &pend);
	468	if (!rc)
	469	rc = smc_cdc_msg_send(conn, wr_buf, pend);
	470	if (rc < 0) {
	471	schedule_work(&conn->tx_work);
	472	return;
	473	}
	474	smc_curs_write(&conn->rx_curs_confirmed,
	475	smc_curs_read(&conn->local_tx_ctrl.cons, conn),
	476	conn);
	477	conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
	478	}
	479	if (conn->local_rx_ctrl.prod_flags.write_blocked &&
	480	!atomic_read(&conn->bytes_to_rcv))
	481	conn->local_rx_ctrl.prod_flags.write_blocked = 0;
	482	}
	483
e6727f39 UB	484	/*************************** send initialize *****************************/
	485
	486	/* Initialize send properties on connection establishment. NB: not __init! */
	487	void smc_tx_init(struct smc_sock *smc)
	488	{
	489	smc->sk.sk_write_space = smc_tx_write_space;
	490	INIT_WORK(&smc->conn.tx_work, smc_tx_work);
	491	spin_lock_init(&smc->conn.send_lock);
	492	}