xref: /openbmc/linux/net/smc/smc_tx.c (revision 1cc1e4c8)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Shared Memory Communications over RDMA (SMC-R) and RoCE
4  *
5  * Manage send buffer.
6  * Producer:
7  * Copy user space data into send buffer, if send buffer space available.
8  * Consumer:
9  * Trigger RDMA write into RMBE of peer and send CDC, if RMBE space available.
10  *
11  * Copyright IBM Corp. 2016
12  *
13  * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
14  */
15 
16 #include <linux/net.h>
17 #include <linux/rcupdate.h>
18 #include <linux/workqueue.h>
19 #include <linux/sched/signal.h>
20 
21 #include <net/sock.h>
22 #include <net/tcp.h>
23 
24 #include "smc.h"
25 #include "smc_wr.h"
26 #include "smc_cdc.h"
27 #include "smc_close.h"
28 #include "smc_ism.h"
29 #include "smc_tx.h"
30 #include "smc_stats.h"
31 #include "smc_tracepoint.h"
32 
33 #define SMC_TX_WORK_DELAY	0
34 #define SMC_TX_CORK_DELAY	(HZ >> 2)	/* 250 ms */
35 
36 /***************************** sndbuf producer *******************************/
37 
38 /* callback implementation for sk.sk_write_space()
39  * to wakeup sndbuf producers that blocked with smc_tx_wait().
40  * called under sk_socket lock.
41  */
42 static void smc_tx_write_space(struct sock *sk)
43 {
44 	struct socket *sock = sk->sk_socket;
45 	struct smc_sock *smc = smc_sk(sk);
46 	struct socket_wq *wq;
47 
48 	/* similar to sk_stream_write_space */
49 	if (atomic_read(&smc->conn.sndbuf_space) && sock) {
50 		if (test_bit(SOCK_NOSPACE, &sock->flags))
51 			SMC_STAT_RMB_TX_FULL(smc, !smc->conn.lnk);
52 		clear_bit(SOCK_NOSPACE, &sock->flags);
53 		rcu_read_lock();
54 		wq = rcu_dereference(sk->sk_wq);
55 		if (skwq_has_sleeper(wq))
56 			wake_up_interruptible_poll(&wq->wait,
57 						   EPOLLOUT | EPOLLWRNORM |
58 						   EPOLLWRBAND);
59 		if (wq && wq->fasync_list && !(sk->sk_shutdown & SEND_SHUTDOWN))
60 			sock_wake_async(wq, SOCK_WAKE_SPACE, POLL_OUT);
61 		rcu_read_unlock();
62 	}
63 }
64 
65 /* Wakeup sndbuf producers that blocked with smc_tx_wait().
66  * Cf. tcp_data_snd_check()=>tcp_check_space()=>tcp_new_space().
67  */
68 void smc_tx_sndbuf_nonfull(struct smc_sock *smc)
69 {
70 	if (smc->sk.sk_socket &&
71 	    test_bit(SOCK_NOSPACE, &smc->sk.sk_socket->flags))
72 		smc->sk.sk_write_space(&smc->sk);
73 }
74 
75 /* blocks sndbuf producer until at least one byte of free space available
76  * or urgent Byte was consumed
77  */
78 static int smc_tx_wait(struct smc_sock *smc, int flags)
79 {
80 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
81 	struct smc_connection *conn = &smc->conn;
82 	struct sock *sk = &smc->sk;
83 	long timeo;
84 	int rc = 0;
85 
86 	/* similar to sk_stream_wait_memory */
87 	timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
88 	add_wait_queue(sk_sleep(sk), &wait);
89 	while (1) {
90 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
91 		if (sk->sk_err ||
92 		    (sk->sk_shutdown & SEND_SHUTDOWN) ||
93 		    conn->killed ||
94 		    conn->local_tx_ctrl.conn_state_flags.peer_done_writing) {
95 			rc = -EPIPE;
96 			break;
97 		}
98 		if (smc_cdc_rxed_any_close(conn)) {
99 			rc = -ECONNRESET;
100 			break;
101 		}
102 		if (!timeo) {
103 			/* ensure EPOLLOUT is subsequently generated */
104 			set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
105 			rc = -EAGAIN;
106 			break;
107 		}
108 		if (signal_pending(current)) {
109 			rc = sock_intr_errno(timeo);
110 			break;
111 		}
112 		sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
113 		if (atomic_read(&conn->sndbuf_space) && !conn->urg_tx_pend)
114 			break; /* at least 1 byte of free & no urgent data */
115 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
116 		sk_wait_event(sk, &timeo,
117 			      sk->sk_err ||
118 			      (sk->sk_shutdown & SEND_SHUTDOWN) ||
119 			      smc_cdc_rxed_any_close(conn) ||
120 			      (atomic_read(&conn->sndbuf_space) &&
121 			       !conn->urg_tx_pend),
122 			      &wait);
123 	}
124 	remove_wait_queue(sk_sleep(sk), &wait);
125 	return rc;
126 }
127 
128 static bool smc_tx_is_corked(struct smc_sock *smc)
129 {
130 	struct tcp_sock *tp = tcp_sk(smc->clcsock->sk);
131 
132 	return (tp->nonagle & TCP_NAGLE_CORK) ? true : false;
133 }
134 
135 /* sndbuf producer: main API called by socket layer.
136  * called under sock lock.
137  */
138 int smc_tx_sendmsg(struct smc_sock *smc, struct msghdr *msg, size_t len)
139 {
140 	size_t copylen, send_done = 0, send_remaining = len;
141 	size_t chunk_len, chunk_off, chunk_len_sum;
142 	struct smc_connection *conn = &smc->conn;
143 	union smc_host_cursor prep;
144 	struct sock *sk = &smc->sk;
145 	char *sndbuf_base;
146 	int tx_cnt_prep;
147 	int writespace;
148 	int rc, chunk;
149 
150 	/* This should be in poll */
151 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
152 
153 	if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
154 		rc = -EPIPE;
155 		goto out_err;
156 	}
157 
158 	if (sk->sk_state == SMC_INIT)
159 		return -ENOTCONN;
160 
161 	if (len > conn->sndbuf_desc->len)
162 		SMC_STAT_RMB_TX_SIZE_SMALL(smc, !conn->lnk);
163 
164 	if (len > conn->peer_rmbe_size)
165 		SMC_STAT_RMB_TX_PEER_SIZE_SMALL(smc, !conn->lnk);
166 
167 	if (msg->msg_flags & MSG_OOB)
168 		SMC_STAT_INC(smc, urg_data_cnt);
169 
170 	while (msg_data_left(msg)) {
171 		if (smc->sk.sk_shutdown & SEND_SHUTDOWN ||
172 		    (smc->sk.sk_err == ECONNABORTED) ||
173 		    conn->killed)
174 			return -EPIPE;
175 		if (smc_cdc_rxed_any_close(conn))
176 			return send_done ?: -ECONNRESET;
177 
178 		if (msg->msg_flags & MSG_OOB)
179 			conn->local_tx_ctrl.prod_flags.urg_data_pending = 1;
180 
181 		if (!atomic_read(&conn->sndbuf_space) || conn->urg_tx_pend) {
182 			if (send_done)
183 				return send_done;
184 			rc = smc_tx_wait(smc, msg->msg_flags);
185 			if (rc)
186 				goto out_err;
187 			continue;
188 		}
189 
190 		/* initialize variables for 1st iteration of subsequent loop */
191 		/* could be just 1 byte, even after smc_tx_wait above */
192 		writespace = atomic_read(&conn->sndbuf_space);
193 		/* not more than what user space asked for */
194 		copylen = min_t(size_t, send_remaining, writespace);
195 		/* determine start of sndbuf */
196 		sndbuf_base = conn->sndbuf_desc->cpu_addr;
197 		smc_curs_copy(&prep, &conn->tx_curs_prep, conn);
198 		tx_cnt_prep = prep.count;
199 		/* determine chunks where to write into sndbuf */
200 		/* either unwrapped case, or 1st chunk of wrapped case */
201 		chunk_len = min_t(size_t, copylen, conn->sndbuf_desc->len -
202 				  tx_cnt_prep);
203 		chunk_len_sum = chunk_len;
204 		chunk_off = tx_cnt_prep;
205 		smc_sndbuf_sync_sg_for_cpu(conn);
206 		for (chunk = 0; chunk < 2; chunk++) {
207 			rc = memcpy_from_msg(sndbuf_base + chunk_off,
208 					     msg, chunk_len);
209 			if (rc) {
210 				smc_sndbuf_sync_sg_for_device(conn);
211 				if (send_done)
212 					return send_done;
213 				goto out_err;
214 			}
215 			send_done += chunk_len;
216 			send_remaining -= chunk_len;
217 
218 			if (chunk_len_sum == copylen)
219 				break; /* either on 1st or 2nd iteration */
220 			/* prepare next (== 2nd) iteration */
221 			chunk_len = copylen - chunk_len; /* remainder */
222 			chunk_len_sum += chunk_len;
223 			chunk_off = 0; /* modulo offset in send ring buffer */
224 		}
225 		smc_sndbuf_sync_sg_for_device(conn);
226 		/* update cursors */
227 		smc_curs_add(conn->sndbuf_desc->len, &prep, copylen);
228 		smc_curs_copy(&conn->tx_curs_prep, &prep, conn);
229 		/* increased in send tasklet smc_cdc_tx_handler() */
230 		smp_mb__before_atomic();
231 		atomic_sub(copylen, &conn->sndbuf_space);
232 		/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
233 		smp_mb__after_atomic();
234 		/* since we just produced more new data into sndbuf,
235 		 * trigger sndbuf consumer: RDMA write into peer RMBE and CDC
236 		 */
237 		if ((msg->msg_flags & MSG_OOB) && !send_remaining)
238 			conn->urg_tx_pend = true;
239 		if ((msg->msg_flags & MSG_MORE || smc_tx_is_corked(smc)) &&
240 		    (atomic_read(&conn->sndbuf_space) >
241 						(conn->sndbuf_desc->len >> 1)))
242 			/* for a corked socket defer the RDMA writes if there
243 			 * is still sufficient sndbuf_space available
244 			 */
245 			queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work,
246 					   SMC_TX_CORK_DELAY);
247 		else
248 			smc_tx_sndbuf_nonempty(conn);
249 
250 		trace_smc_tx_sendmsg(smc, copylen);
251 	} /* while (msg_data_left(msg)) */
252 
253 	return send_done;
254 
255 out_err:
256 	rc = sk_stream_error(sk, msg->msg_flags, rc);
257 	/* make sure we wake any epoll edge trigger waiter */
258 	if (unlikely(rc == -EAGAIN))
259 		sk->sk_write_space(sk);
260 	return rc;
261 }
262 
263 /***************************** sndbuf consumer *******************************/
264 
265 /* sndbuf consumer: actual data transfer of one target chunk with ISM write */
266 int smcd_tx_ism_write(struct smc_connection *conn, void *data, size_t len,
267 		      u32 offset, int signal)
268 {
269 	struct smc_ism_position pos;
270 	int rc;
271 
272 	memset(&pos, 0, sizeof(pos));
273 	pos.token = conn->peer_token;
274 	pos.index = conn->peer_rmbe_idx;
275 	pos.offset = conn->tx_off + offset;
276 	pos.signal = signal;
277 	rc = smc_ism_write(conn->lgr->smcd, &pos, data, len);
278 	if (rc)
279 		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
280 	return rc;
281 }
282 
283 /* sndbuf consumer: actual data transfer of one target chunk with RDMA write */
284 static int smc_tx_rdma_write(struct smc_connection *conn, int peer_rmbe_offset,
285 			     int num_sges, struct ib_rdma_wr *rdma_wr)
286 {
287 	struct smc_link_group *lgr = conn->lgr;
288 	struct smc_link *link = conn->lnk;
289 	int rc;
290 
291 	rdma_wr->wr.wr_id = smc_wr_tx_get_next_wr_id(link);
292 	rdma_wr->wr.num_sge = num_sges;
293 	rdma_wr->remote_addr =
294 		lgr->rtokens[conn->rtoken_idx][link->link_idx].dma_addr +
295 		/* RMBE within RMB */
296 		conn->tx_off +
297 		/* offset within RMBE */
298 		peer_rmbe_offset;
299 	rdma_wr->rkey = lgr->rtokens[conn->rtoken_idx][link->link_idx].rkey;
300 	rc = ib_post_send(link->roce_qp, &rdma_wr->wr, NULL);
301 	if (rc)
302 		smcr_link_down_cond_sched(link);
303 	return rc;
304 }
305 
306 /* sndbuf consumer */
307 static inline void smc_tx_advance_cursors(struct smc_connection *conn,
308 					  union smc_host_cursor *prod,
309 					  union smc_host_cursor *sent,
310 					  size_t len)
311 {
312 	smc_curs_add(conn->peer_rmbe_size, prod, len);
313 	/* increased in recv tasklet smc_cdc_msg_rcv() */
314 	smp_mb__before_atomic();
315 	/* data in flight reduces usable snd_wnd */
316 	atomic_sub(len, &conn->peer_rmbe_space);
317 	/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
318 	smp_mb__after_atomic();
319 	smc_curs_add(conn->sndbuf_desc->len, sent, len);
320 }
321 
322 /* SMC-R helper for smc_tx_rdma_writes() */
323 static int smcr_tx_rdma_writes(struct smc_connection *conn, size_t len,
324 			       size_t src_off, size_t src_len,
325 			       size_t dst_off, size_t dst_len,
326 			       struct smc_rdma_wr *wr_rdma_buf)
327 {
328 	struct smc_link *link = conn->lnk;
329 
330 	dma_addr_t dma_addr =
331 		sg_dma_address(conn->sndbuf_desc->sgt[link->link_idx].sgl);
332 	int src_len_sum = src_len, dst_len_sum = dst_len;
333 	int sent_count = src_off;
334 	int srcchunk, dstchunk;
335 	int num_sges;
336 	int rc;
337 
338 	for (dstchunk = 0; dstchunk < 2; dstchunk++) {
339 		struct ib_sge *sge =
340 			wr_rdma_buf->wr_tx_rdma[dstchunk].wr.sg_list;
341 
342 		num_sges = 0;
343 		for (srcchunk = 0; srcchunk < 2; srcchunk++) {
344 			sge[srcchunk].addr = dma_addr + src_off;
345 			sge[srcchunk].length = src_len;
346 			num_sges++;
347 
348 			src_off += src_len;
349 			if (src_off >= conn->sndbuf_desc->len)
350 				src_off -= conn->sndbuf_desc->len;
351 						/* modulo in send ring */
352 			if (src_len_sum == dst_len)
353 				break; /* either on 1st or 2nd iteration */
354 			/* prepare next (== 2nd) iteration */
355 			src_len = dst_len - src_len; /* remainder */
356 			src_len_sum += src_len;
357 		}
358 		rc = smc_tx_rdma_write(conn, dst_off, num_sges,
359 				       &wr_rdma_buf->wr_tx_rdma[dstchunk]);
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, dst_len, conn->sndbuf_desc->len -
369 				sent_count);
370 		src_len_sum = src_len;
371 	}
372 	return 0;
373 }
374 
375 /* SMC-D helper for smc_tx_rdma_writes() */
376 static int smcd_tx_rdma_writes(struct smc_connection *conn, size_t len,
377 			       size_t src_off, size_t src_len,
378 			       size_t dst_off, size_t dst_len)
379 {
380 	int src_len_sum = src_len, dst_len_sum = dst_len;
381 	int srcchunk, dstchunk;
382 	int rc;
383 
384 	for (dstchunk = 0; dstchunk < 2; dstchunk++) {
385 		for (srcchunk = 0; srcchunk < 2; srcchunk++) {
386 			void *data = conn->sndbuf_desc->cpu_addr + src_off;
387 
388 			rc = smcd_tx_ism_write(conn, data, src_len, dst_off +
389 					       sizeof(struct smcd_cdc_msg), 0);
390 			if (rc)
391 				return rc;
392 			dst_off += src_len;
393 			src_off += src_len;
394 			if (src_off >= conn->sndbuf_desc->len)
395 				src_off -= conn->sndbuf_desc->len;
396 						/* modulo in send ring */
397 			if (src_len_sum == dst_len)
398 				break; /* either on 1st or 2nd iteration */
399 			/* prepare next (== 2nd) iteration */
400 			src_len = dst_len - src_len; /* remainder */
401 			src_len_sum += src_len;
402 		}
403 		if (dst_len_sum == len)
404 			break; /* either on 1st or 2nd iteration */
405 		/* prepare next (== 2nd) iteration */
406 		dst_off = 0; /* modulo offset in RMBE ring buffer */
407 		dst_len = len - dst_len; /* remainder */
408 		dst_len_sum += dst_len;
409 		src_len = min_t(int, dst_len, conn->sndbuf_desc->len - src_off);
410 		src_len_sum = src_len;
411 	}
412 	return 0;
413 }
414 
415 /* sndbuf consumer: prepare all necessary (src&dst) chunks of data transmit;
416  * usable snd_wnd as max transmit
417  */
418 static int smc_tx_rdma_writes(struct smc_connection *conn,
419 			      struct smc_rdma_wr *wr_rdma_buf)
420 {
421 	size_t len, src_len, dst_off, dst_len; /* current chunk values */
422 	union smc_host_cursor sent, prep, prod, cons;
423 	struct smc_cdc_producer_flags *pflags;
424 	int to_send, rmbespace;
425 	int rc;
426 
427 	/* source: sndbuf */
428 	smc_curs_copy(&sent, &conn->tx_curs_sent, conn);
429 	smc_curs_copy(&prep, &conn->tx_curs_prep, conn);
430 	/* cf. wmem_alloc - (snd_max - snd_una) */
431 	to_send = smc_curs_diff(conn->sndbuf_desc->len, &sent, &prep);
432 	if (to_send <= 0)
433 		return 0;
434 
435 	/* destination: RMBE */
436 	/* cf. snd_wnd */
437 	rmbespace = atomic_read(&conn->peer_rmbe_space);
438 	if (rmbespace <= 0) {
439 		struct smc_sock *smc = container_of(conn, struct smc_sock,
440 						    conn);
441 		SMC_STAT_RMB_TX_PEER_FULL(smc, !conn->lnk);
442 		return 0;
443 	}
444 	smc_curs_copy(&prod, &conn->local_tx_ctrl.prod, conn);
445 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
446 
447 	/* if usable snd_wnd closes ask peer to advertise once it opens again */
448 	pflags = &conn->local_tx_ctrl.prod_flags;
449 	pflags->write_blocked = (to_send >= rmbespace);
450 	/* cf. usable snd_wnd */
451 	len = min(to_send, rmbespace);
452 
453 	/* initialize variables for first iteration of subsequent nested loop */
454 	dst_off = prod.count;
455 	if (prod.wrap == cons.wrap) {
456 		/* the filled destination area is unwrapped,
457 		 * hence the available free destination space is wrapped
458 		 * and we need 2 destination chunks of sum len; start with 1st
459 		 * which is limited by what's available in sndbuf
460 		 */
461 		dst_len = min_t(size_t,
462 				conn->peer_rmbe_size - prod.count, len);
463 	} else {
464 		/* the filled destination area is wrapped,
465 		 * hence the available free destination space is unwrapped
466 		 * and we need a single destination chunk of entire len
467 		 */
468 		dst_len = len;
469 	}
470 	/* dst_len determines the maximum src_len */
471 	if (sent.count + dst_len <= conn->sndbuf_desc->len) {
472 		/* unwrapped src case: single chunk of entire dst_len */
473 		src_len = dst_len;
474 	} else {
475 		/* wrapped src case: 2 chunks of sum dst_len; start with 1st: */
476 		src_len = conn->sndbuf_desc->len - sent.count;
477 	}
478 
479 	if (conn->lgr->is_smcd)
480 		rc = smcd_tx_rdma_writes(conn, len, sent.count, src_len,
481 					 dst_off, dst_len);
482 	else
483 		rc = smcr_tx_rdma_writes(conn, len, sent.count, src_len,
484 					 dst_off, dst_len, wr_rdma_buf);
485 	if (rc)
486 		return rc;
487 
488 	if (conn->urg_tx_pend && len == to_send)
489 		pflags->urg_data_present = 1;
490 	smc_tx_advance_cursors(conn, &prod, &sent, len);
491 	/* update connection's cursors with advanced local cursors */
492 	smc_curs_copy(&conn->local_tx_ctrl.prod, &prod, conn);
493 							/* dst: peer RMBE */
494 	smc_curs_copy(&conn->tx_curs_sent, &sent, conn);/* src: local sndbuf */
495 
496 	return 0;
497 }
498 
499 /* Wakeup sndbuf consumers from any context (IRQ or process)
500  * since there is more data to transmit; usable snd_wnd as max transmit
501  */
502 static int smcr_tx_sndbuf_nonempty(struct smc_connection *conn)
503 {
504 	struct smc_cdc_producer_flags *pflags = &conn->local_tx_ctrl.prod_flags;
505 	struct smc_link *link = conn->lnk;
506 	struct smc_rdma_wr *wr_rdma_buf;
507 	struct smc_cdc_tx_pend *pend;
508 	struct smc_wr_buf *wr_buf;
509 	int rc;
510 
511 	if (!link || !smc_wr_tx_link_hold(link))
512 		return -ENOLINK;
513 	rc = smc_cdc_get_free_slot(conn, link, &wr_buf, &wr_rdma_buf, &pend);
514 	if (rc < 0) {
515 		smc_wr_tx_link_put(link);
516 		if (rc == -EBUSY) {
517 			struct smc_sock *smc =
518 				container_of(conn, struct smc_sock, conn);
519 
520 			if (smc->sk.sk_err == ECONNABORTED)
521 				return sock_error(&smc->sk);
522 			if (conn->killed)
523 				return -EPIPE;
524 			rc = 0;
525 			mod_delayed_work(conn->lgr->tx_wq, &conn->tx_work,
526 					 SMC_TX_WORK_DELAY);
527 		}
528 		return rc;
529 	}
530 
531 	spin_lock_bh(&conn->send_lock);
532 	if (link != conn->lnk) {
533 		/* link of connection changed, tx_work will restart */
534 		smc_wr_tx_put_slot(link,
535 				   (struct smc_wr_tx_pend_priv *)pend);
536 		rc = -ENOLINK;
537 		goto out_unlock;
538 	}
539 	if (!pflags->urg_data_present) {
540 		rc = smc_tx_rdma_writes(conn, wr_rdma_buf);
541 		if (rc) {
542 			smc_wr_tx_put_slot(link,
543 					   (struct smc_wr_tx_pend_priv *)pend);
544 			goto out_unlock;
545 		}
546 	}
547 
548 	rc = smc_cdc_msg_send(conn, wr_buf, pend);
549 	if (!rc && pflags->urg_data_present) {
550 		pflags->urg_data_pending = 0;
551 		pflags->urg_data_present = 0;
552 	}
553 
554 out_unlock:
555 	spin_unlock_bh(&conn->send_lock);
556 	smc_wr_tx_link_put(link);
557 	return rc;
558 }
559 
560 static int smcd_tx_sndbuf_nonempty(struct smc_connection *conn)
561 {
562 	struct smc_cdc_producer_flags *pflags = &conn->local_tx_ctrl.prod_flags;
563 	int rc = 0;
564 
565 	spin_lock_bh(&conn->send_lock);
566 	if (!pflags->urg_data_present)
567 		rc = smc_tx_rdma_writes(conn, NULL);
568 	if (!rc)
569 		rc = smcd_cdc_msg_send(conn);
570 
571 	if (!rc && pflags->urg_data_present) {
572 		pflags->urg_data_pending = 0;
573 		pflags->urg_data_present = 0;
574 	}
575 	spin_unlock_bh(&conn->send_lock);
576 	return rc;
577 }
578 
579 int smc_tx_sndbuf_nonempty(struct smc_connection *conn)
580 {
581 	int rc;
582 
583 	if (conn->killed ||
584 	    conn->local_rx_ctrl.conn_state_flags.peer_conn_abort)
585 		return -EPIPE;	/* connection being aborted */
586 	if (conn->lgr->is_smcd)
587 		rc = smcd_tx_sndbuf_nonempty(conn);
588 	else
589 		rc = smcr_tx_sndbuf_nonempty(conn);
590 
591 	if (!rc) {
592 		/* trigger socket release if connection is closing */
593 		struct smc_sock *smc = container_of(conn, struct smc_sock,
594 						    conn);
595 		smc_close_wake_tx_prepared(smc);
596 	}
597 	return rc;
598 }
599 
600 /* Wakeup sndbuf consumers from process context
601  * since there is more data to transmit
602  */
603 void smc_tx_work(struct work_struct *work)
604 {
605 	struct smc_connection *conn = container_of(to_delayed_work(work),
606 						   struct smc_connection,
607 						   tx_work);
608 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
609 	int rc;
610 
611 	lock_sock(&smc->sk);
612 	if (smc->sk.sk_err)
613 		goto out;
614 
615 	rc = smc_tx_sndbuf_nonempty(conn);
616 	if (!rc && conn->local_rx_ctrl.prod_flags.write_blocked &&
617 	    !atomic_read(&conn->bytes_to_rcv))
618 		conn->local_rx_ctrl.prod_flags.write_blocked = 0;
619 
620 out:
621 	release_sock(&smc->sk);
622 }
623 
624 void smc_tx_consumer_update(struct smc_connection *conn, bool force)
625 {
626 	union smc_host_cursor cfed, cons, prod;
627 	int sender_free = conn->rmb_desc->len;
628 	int to_confirm;
629 
630 	smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
631 	smc_curs_copy(&cfed, &conn->rx_curs_confirmed, conn);
632 	to_confirm = smc_curs_diff(conn->rmb_desc->len, &cfed, &cons);
633 	if (to_confirm > conn->rmbe_update_limit) {
634 		smc_curs_copy(&prod, &conn->local_rx_ctrl.prod, conn);
635 		sender_free = conn->rmb_desc->len -
636 			      smc_curs_diff_large(conn->rmb_desc->len,
637 						  &cfed, &prod);
638 	}
639 
640 	if (conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||
641 	    force ||
642 	    ((to_confirm > conn->rmbe_update_limit) &&
643 	     ((sender_free <= (conn->rmb_desc->len / 2)) ||
644 	      conn->local_rx_ctrl.prod_flags.write_blocked))) {
645 		if (conn->killed ||
646 		    conn->local_rx_ctrl.conn_state_flags.peer_conn_abort)
647 			return;
648 		if ((smc_cdc_get_slot_and_msg_send(conn) < 0) &&
649 		    !conn->killed) {
650 			queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work,
651 					   SMC_TX_WORK_DELAY);
652 			return;
653 		}
654 	}
655 	if (conn->local_rx_ctrl.prod_flags.write_blocked &&
656 	    !atomic_read(&conn->bytes_to_rcv))
657 		conn->local_rx_ctrl.prod_flags.write_blocked = 0;
658 }
659 
660 /***************************** send initialize *******************************/
661 
662 /* Initialize send properties on connection establishment. NB: not __init! */
663 void smc_tx_init(struct smc_sock *smc)
664 {
665 	smc->sk.sk_write_space = smc_tx_write_space;
666 }
667