Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Connection Data Control (CDC)
6 * handles flow control
7 *
8 * Copyright IBM Corp. 2016
9 *
10 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
11 */
12
13#include <linux/spinlock.h>
14
15#include "smc.h"
16#include "smc_wr.h"
17#include "smc_cdc.h"
18#include "smc_tx.h"
19#include "smc_rx.h"
20#include "smc_close.h"
21
22/********************************** send *************************************/
23
24/* handler for send/transmission completion of a CDC msg */
25static void smc_cdc_tx_handler(struct smc_wr_tx_pend_priv *pnd_snd,
26 struct smc_link *link,
27 enum ib_wc_status wc_status)
28{
29 struct smc_cdc_tx_pend *cdcpend = (struct smc_cdc_tx_pend *)pnd_snd;
30 struct smc_connection *conn = cdcpend->conn;
31 struct smc_sock *smc;
32 int diff;
33
34 if (!conn)
35 /* already dismissed */
36 return;
37
38 smc = container_of(conn, struct smc_sock, conn);
39 bh_lock_sock(&smc->sk);
40 if (!wc_status) {
41 diff = smc_curs_diff(cdcpend->conn->sndbuf_desc->len,
42 &cdcpend->conn->tx_curs_fin,
43 &cdcpend->cursor);
44 /* sndbuf_space is decreased in smc_sendmsg */
45 smp_mb__before_atomic();
46 atomic_add(diff, &cdcpend->conn->sndbuf_space);
47 /* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
48 smp_mb__after_atomic();
49 smc_curs_copy(&conn->tx_curs_fin, &cdcpend->cursor, conn);
50 smc_curs_copy(&conn->local_tx_ctrl_fin, &cdcpend->p_cursor,
51 conn);
52 conn->tx_cdc_seq_fin = cdcpend->ctrl_seq;
53 }
54 smc_tx_sndbuf_nonfull(smc);
55 bh_unlock_sock(&smc->sk);
56}
57
58int smc_cdc_get_free_slot(struct smc_connection *conn,
59 struct smc_link *link,
60 struct smc_wr_buf **wr_buf,
61 struct smc_rdma_wr **wr_rdma_buf,
62 struct smc_cdc_tx_pend **pend)
63{
64 int rc;
65
66 rc = smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
67 wr_rdma_buf,
68 (struct smc_wr_tx_pend_priv **)pend);
69 if (conn->killed) {
70 /* abnormal termination */
71 if (!rc)
72 smc_wr_tx_put_slot(link,
73 (struct smc_wr_tx_pend_priv *)pend);
74 rc = -EPIPE;
75 }
76 return rc;
77}
78
79static inline void smc_cdc_add_pending_send(struct smc_connection *conn,
80 struct smc_cdc_tx_pend *pend)
81{
82 BUILD_BUG_ON_MSG(
83 sizeof(struct smc_cdc_msg) > SMC_WR_BUF_SIZE,
84 "must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_cdc_msg)");
85 BUILD_BUG_ON_MSG(
86 offsetofend(struct smc_cdc_msg, reserved) > SMC_WR_TX_SIZE,
87 "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()");
88 BUILD_BUG_ON_MSG(
89 sizeof(struct smc_cdc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
90 "must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_cdc_tx_pend)");
91 pend->conn = conn;
92 pend->cursor = conn->tx_curs_sent;
93 pend->p_cursor = conn->local_tx_ctrl.prod;
94 pend->ctrl_seq = conn->tx_cdc_seq;
95}
96
97int smc_cdc_msg_send(struct smc_connection *conn,
98 struct smc_wr_buf *wr_buf,
99 struct smc_cdc_tx_pend *pend)
100{
101 struct smc_link *link = conn->lnk;
102 union smc_host_cursor cfed;
103 int rc;
104
105 smc_cdc_add_pending_send(conn, pend);
106
107 conn->tx_cdc_seq++;
108 conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
109 smc_host_msg_to_cdc((struct smc_cdc_msg *)wr_buf, conn, &cfed);
110 rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
111 if (!rc) {
112 smc_curs_copy(&conn->rx_curs_confirmed, &cfed, conn);
113 conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
114 } else {
115 conn->tx_cdc_seq--;
116 conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
117 }
118
119 return rc;
120}
121
122/* send a validation msg indicating the move of a conn to an other QP link */
123int smcr_cdc_msg_send_validation(struct smc_connection *conn,
124 struct smc_cdc_tx_pend *pend,
125 struct smc_wr_buf *wr_buf)
126{
127 struct smc_host_cdc_msg *local = &conn->local_tx_ctrl;
128 struct smc_link *link = conn->lnk;
129 struct smc_cdc_msg *peer;
130 int rc;
131
132 peer = (struct smc_cdc_msg *)wr_buf;
133 peer->common.type = local->common.type;
134 peer->len = local->len;
135 peer->seqno = htons(conn->tx_cdc_seq_fin); /* seqno last compl. tx */
136 peer->token = htonl(local->token);
137 peer->prod_flags.failover_validation = 1;
138
139 rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
140 return rc;
141}
142
143static int smcr_cdc_get_slot_and_msg_send(struct smc_connection *conn)
144{
145 struct smc_cdc_tx_pend *pend;
146 struct smc_wr_buf *wr_buf;
147 struct smc_link *link;
148 bool again = false;
149 int rc;
150
151again:
152 link = conn->lnk;
153 rc = smc_cdc_get_free_slot(conn, link, &wr_buf, NULL, &pend);
154 if (rc)
155 return rc;
156
157 spin_lock_bh(&conn->send_lock);
158 if (link != conn->lnk) {
159 /* link of connection changed, try again one time*/
160 spin_unlock_bh(&conn->send_lock);
161 smc_wr_tx_put_slot(link,
162 (struct smc_wr_tx_pend_priv *)pend);
163 if (again)
164 return -ENOLINK;
165 again = true;
166 goto again;
167 }
168 rc = smc_cdc_msg_send(conn, wr_buf, pend);
169 spin_unlock_bh(&conn->send_lock);
170 return rc;
171}
172
173int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn)
174{
175 int rc;
176
177 if (!conn->lgr || (conn->lgr->is_smcd && conn->lgr->peer_shutdown))
178 return -EPIPE;
179
180 if (conn->lgr->is_smcd) {
181 spin_lock_bh(&conn->send_lock);
182 rc = smcd_cdc_msg_send(conn);
183 spin_unlock_bh(&conn->send_lock);
184 } else {
185 rc = smcr_cdc_get_slot_and_msg_send(conn);
186 }
187
188 return rc;
189}
190
191static bool smc_cdc_tx_filter(struct smc_wr_tx_pend_priv *tx_pend,
192 unsigned long data)
193{
194 struct smc_connection *conn = (struct smc_connection *)data;
195 struct smc_cdc_tx_pend *cdc_pend =
196 (struct smc_cdc_tx_pend *)tx_pend;
197
198 return cdc_pend->conn == conn;
199}
200
201static void smc_cdc_tx_dismisser(struct smc_wr_tx_pend_priv *tx_pend)
202{
203 struct smc_cdc_tx_pend *cdc_pend =
204 (struct smc_cdc_tx_pend *)tx_pend;
205
206 cdc_pend->conn = NULL;
207}
208
209void smc_cdc_tx_dismiss_slots(struct smc_connection *conn)
210{
211 struct smc_link *link = conn->lnk;
212
213 smc_wr_tx_dismiss_slots(link, SMC_CDC_MSG_TYPE,
214 smc_cdc_tx_filter, smc_cdc_tx_dismisser,
215 (unsigned long)conn);
216}
217
218/* Send a SMC-D CDC header.
219 * This increments the free space available in our send buffer.
220 * Also update the confirmed receive buffer with what was sent to the peer.
221 */
222int smcd_cdc_msg_send(struct smc_connection *conn)
223{
224 struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
225 union smc_host_cursor curs;
226 struct smcd_cdc_msg cdc;
227 int rc, diff;
228
229 memset(&cdc, 0, sizeof(cdc));
230 cdc.common.type = SMC_CDC_MSG_TYPE;
231 curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.prod.acurs);
232 cdc.prod.wrap = curs.wrap;
233 cdc.prod.count = curs.count;
234 curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.cons.acurs);
235 cdc.cons.wrap = curs.wrap;
236 cdc.cons.count = curs.count;
237 cdc.cons.prod_flags = conn->local_tx_ctrl.prod_flags;
238 cdc.cons.conn_state_flags = conn->local_tx_ctrl.conn_state_flags;
239 rc = smcd_tx_ism_write(conn, &cdc, sizeof(cdc), 0, 1);
240 if (rc)
241 return rc;
242 smc_curs_copy(&conn->rx_curs_confirmed, &curs, conn);
243 conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
244 /* Calculate transmitted data and increment free send buffer space */
245 diff = smc_curs_diff(conn->sndbuf_desc->len, &conn->tx_curs_fin,
246 &conn->tx_curs_sent);
247 /* increased by confirmed number of bytes */
248 smp_mb__before_atomic();
249 atomic_add(diff, &conn->sndbuf_space);
250 /* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
251 smp_mb__after_atomic();
252 smc_curs_copy(&conn->tx_curs_fin, &conn->tx_curs_sent, conn);
253
254 smc_tx_sndbuf_nonfull(smc);
255 return rc;
256}
257
258/********************************* receive ***********************************/
259
260static inline bool smc_cdc_before(u16 seq1, u16 seq2)
261{
262 return (s16)(seq1 - seq2) < 0;
263}
264
265static void smc_cdc_handle_urg_data_arrival(struct smc_sock *smc,
266 int *diff_prod)
267{
268 struct smc_connection *conn = &smc->conn;
269 char *base;
270
271 /* new data included urgent business */
272 smc_curs_copy(&conn->urg_curs, &conn->local_rx_ctrl.prod, conn);
273 conn->urg_state = SMC_URG_VALID;
274 if (!sock_flag(&smc->sk, SOCK_URGINLINE))
275 /* we'll skip the urgent byte, so don't account for it */
276 (*diff_prod)--;
277 base = (char *)conn->rmb_desc->cpu_addr + conn->rx_off;
278 if (conn->urg_curs.count)
279 conn->urg_rx_byte = *(base + conn->urg_curs.count - 1);
280 else
281 conn->urg_rx_byte = *(base + conn->rmb_desc->len - 1);
282 sk_send_sigurg(&smc->sk);
283}
284
285static void smc_cdc_msg_validate(struct smc_sock *smc, struct smc_cdc_msg *cdc,
286 struct smc_link *link)
287{
288 struct smc_connection *conn = &smc->conn;
289 u16 recv_seq = ntohs(cdc->seqno);
290 s16 diff;
291
292 /* check that seqnum was seen before */
293 diff = conn->local_rx_ctrl.seqno - recv_seq;
294 if (diff < 0) { /* diff larger than 0x7fff */
295 /* drop connection */
296 conn->out_of_sync = 1; /* prevent any further receives */
297 spin_lock_bh(&conn->send_lock);
298 conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
299 conn->lnk = link;
300 spin_unlock_bh(&conn->send_lock);
301 sock_hold(&smc->sk); /* sock_put in abort_work */
302 if (!schedule_work(&conn->abort_work))
303 sock_put(&smc->sk);
304 }
305}
306
307static void smc_cdc_msg_recv_action(struct smc_sock *smc,
308 struct smc_cdc_msg *cdc)
309{
310 union smc_host_cursor cons_old, prod_old;
311 struct smc_connection *conn = &smc->conn;
312 int diff_cons, diff_prod;
313
314 smc_curs_copy(&prod_old, &conn->local_rx_ctrl.prod, conn);
315 smc_curs_copy(&cons_old, &conn->local_rx_ctrl.cons, conn);
316 smc_cdc_msg_to_host(&conn->local_rx_ctrl, cdc, conn);
317
318 diff_cons = smc_curs_diff(conn->peer_rmbe_size, &cons_old,
319 &conn->local_rx_ctrl.cons);
320 if (diff_cons) {
321 /* peer_rmbe_space is decreased during data transfer with RDMA
322 * write
323 */
324 smp_mb__before_atomic();
325 atomic_add(diff_cons, &conn->peer_rmbe_space);
326 /* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
327 smp_mb__after_atomic();
328 }
329
330 diff_prod = smc_curs_diff(conn->rmb_desc->len, &prod_old,
331 &conn->local_rx_ctrl.prod);
332 if (diff_prod) {
333 if (conn->local_rx_ctrl.prod_flags.urg_data_present)
334 smc_cdc_handle_urg_data_arrival(smc, &diff_prod);
335 /* bytes_to_rcv is decreased in smc_recvmsg */
336 smp_mb__before_atomic();
337 atomic_add(diff_prod, &conn->bytes_to_rcv);
338 /* guarantee 0 <= bytes_to_rcv <= rmb_desc->len */
339 smp_mb__after_atomic();
340 smc->sk.sk_data_ready(&smc->sk);
341 } else {
342 if (conn->local_rx_ctrl.prod_flags.write_blocked)
343 smc->sk.sk_data_ready(&smc->sk);
344 if (conn->local_rx_ctrl.prod_flags.urg_data_pending)
345 conn->urg_state = SMC_URG_NOTYET;
346 }
347
348 /* trigger sndbuf consumer: RDMA write into peer RMBE and CDC */
349 if ((diff_cons && smc_tx_prepared_sends(conn)) ||
350 conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||
351 conn->local_rx_ctrl.prod_flags.urg_data_pending)
352 smc_tx_sndbuf_nonempty(conn);
353
354 if (diff_cons && conn->urg_tx_pend &&
355 atomic_read(&conn->peer_rmbe_space) == conn->peer_rmbe_size) {
356 /* urg data confirmed by peer, indicate we're ready for more */
357 conn->urg_tx_pend = false;
358 smc->sk.sk_write_space(&smc->sk);
359 }
360
361 if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
362 smc->sk.sk_err = ECONNRESET;
363 conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
364 }
365 if (smc_cdc_rxed_any_close_or_senddone(conn)) {
366 smc->sk.sk_shutdown |= RCV_SHUTDOWN;
367 if (smc->clcsock && smc->clcsock->sk)
368 smc->clcsock->sk->sk_shutdown |= RCV_SHUTDOWN;
369 sock_set_flag(&smc->sk, SOCK_DONE);
370 sock_hold(&smc->sk); /* sock_put in close_work */
371 if (!schedule_work(&conn->close_work))
372 sock_put(&smc->sk);
373 }
374}
375
376/* called under tasklet context */
377static void smc_cdc_msg_recv(struct smc_sock *smc, struct smc_cdc_msg *cdc)
378{
379 sock_hold(&smc->sk);
380 bh_lock_sock(&smc->sk);
381 smc_cdc_msg_recv_action(smc, cdc);
382 bh_unlock_sock(&smc->sk);
383 sock_put(&smc->sk); /* no free sk in softirq-context */
384}
385
386/* Schedule a tasklet for this connection. Triggered from the ISM device IRQ
387 * handler to indicate update in the DMBE.
388 *
389 * Context:
390 * - tasklet context
391 */
392static void smcd_cdc_rx_tsklet(unsigned long data)
393{
394 struct smc_connection *conn = (struct smc_connection *)data;
395 struct smcd_cdc_msg *data_cdc;
396 struct smcd_cdc_msg cdc;
397 struct smc_sock *smc;
398
399 if (!conn || conn->killed)
400 return;
401
402 data_cdc = (struct smcd_cdc_msg *)conn->rmb_desc->cpu_addr;
403 smcd_curs_copy(&cdc.prod, &data_cdc->prod, conn);
404 smcd_curs_copy(&cdc.cons, &data_cdc->cons, conn);
405 smc = container_of(conn, struct smc_sock, conn);
406 smc_cdc_msg_recv(smc, (struct smc_cdc_msg *)&cdc);
407}
408
409/* Initialize receive tasklet. Called from ISM device IRQ handler to start
410 * receiver side.
411 */
412void smcd_cdc_rx_init(struct smc_connection *conn)
413{
414 tasklet_init(&conn->rx_tsklet, smcd_cdc_rx_tsklet, (unsigned long)conn);
415}
416
417/***************************** init, exit, misc ******************************/
418
419static void smc_cdc_rx_handler(struct ib_wc *wc, void *buf)
420{
421 struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
422 struct smc_cdc_msg *cdc = buf;
423 struct smc_connection *conn;
424 struct smc_link_group *lgr;
425 struct smc_sock *smc;
426
427 if (wc->byte_len < offsetof(struct smc_cdc_msg, reserved))
428 return; /* short message */
429 if (cdc->len != SMC_WR_TX_SIZE)
430 return; /* invalid message */
431
432 /* lookup connection */
433 lgr = smc_get_lgr(link);
434 read_lock_bh(&lgr->conns_lock);
435 conn = smc_lgr_find_conn(ntohl(cdc->token), lgr);
436 read_unlock_bh(&lgr->conns_lock);
437 if (!conn || conn->out_of_sync)
438 return;
439 smc = container_of(conn, struct smc_sock, conn);
440
441 if (cdc->prod_flags.failover_validation) {
442 smc_cdc_msg_validate(smc, cdc, link);
443 return;
444 }
445 if (smc_cdc_before(ntohs(cdc->seqno),
446 conn->local_rx_ctrl.seqno))
447 /* received seqno is old */
448 return;
449
450 smc_cdc_msg_recv(smc, cdc);
451}
452
453static struct smc_wr_rx_handler smc_cdc_rx_handlers[] = {
454 {
455 .handler = smc_cdc_rx_handler,
456 .type = SMC_CDC_MSG_TYPE
457 },
458 {
459 .handler = NULL,
460 }
461};
462
463int __init smc_cdc_init(void)
464{
465 struct smc_wr_rx_handler *handler;
466 int rc = 0;
467
468 for (handler = smc_cdc_rx_handlers; handler->handler; handler++) {
469 INIT_HLIST_NODE(&handler->list);
470 rc = smc_wr_rx_register_handler(handler);
471 if (rc)
472 break;
473 }
474 return rc;
475}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Connection Data Control (CDC)
6 * handles flow control
7 *
8 * Copyright IBM Corp. 2016
9 *
10 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
11 */
12
13#include <linux/spinlock.h>
14
15#include "smc.h"
16#include "smc_wr.h"
17#include "smc_cdc.h"
18#include "smc_tx.h"
19#include "smc_rx.h"
20#include "smc_close.h"
21
22/********************************** send *************************************/
23
24/* handler for send/transmission completion of a CDC msg */
25static void smc_cdc_tx_handler(struct smc_wr_tx_pend_priv *pnd_snd,
26 struct smc_link *link,
27 enum ib_wc_status wc_status)
28{
29 struct smc_cdc_tx_pend *cdcpend = (struct smc_cdc_tx_pend *)pnd_snd;
30 struct smc_connection *conn = cdcpend->conn;
31 struct smc_sock *smc;
32 int diff;
33
34 if (!conn)
35 /* already dismissed */
36 return;
37
38 smc = container_of(conn, struct smc_sock, conn);
39 bh_lock_sock(&smc->sk);
40 if (!wc_status) {
41 diff = smc_curs_diff(cdcpend->conn->sndbuf_desc->len,
42 &cdcpend->conn->tx_curs_fin,
43 &cdcpend->cursor);
44 /* sndbuf_space is decreased in smc_sendmsg */
45 smp_mb__before_atomic();
46 atomic_add(diff, &cdcpend->conn->sndbuf_space);
47 /* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
48 smp_mb__after_atomic();
49 smc_curs_copy(&conn->tx_curs_fin, &cdcpend->cursor, conn);
50 smc_curs_copy(&conn->local_tx_ctrl_fin, &cdcpend->p_cursor,
51 conn);
52 conn->tx_cdc_seq_fin = cdcpend->ctrl_seq;
53 }
54 smc_tx_sndbuf_nonfull(smc);
55 bh_unlock_sock(&smc->sk);
56}
57
58int smc_cdc_get_free_slot(struct smc_connection *conn,
59 struct smc_link *link,
60 struct smc_wr_buf **wr_buf,
61 struct smc_rdma_wr **wr_rdma_buf,
62 struct smc_cdc_tx_pend **pend)
63{
64 int rc;
65
66 rc = smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
67 wr_rdma_buf,
68 (struct smc_wr_tx_pend_priv **)pend);
69 if (conn->killed) {
70 /* abnormal termination */
71 if (!rc)
72 smc_wr_tx_put_slot(link,
73 (struct smc_wr_tx_pend_priv *)pend);
74 rc = -EPIPE;
75 }
76 return rc;
77}
78
79static inline void smc_cdc_add_pending_send(struct smc_connection *conn,
80 struct smc_cdc_tx_pend *pend)
81{
82 BUILD_BUG_ON_MSG(
83 sizeof(struct smc_cdc_msg) > SMC_WR_BUF_SIZE,
84 "must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_cdc_msg)");
85 BUILD_BUG_ON_MSG(
86 offsetofend(struct smc_cdc_msg, reserved) > SMC_WR_TX_SIZE,
87 "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()");
88 BUILD_BUG_ON_MSG(
89 sizeof(struct smc_cdc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
90 "must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_cdc_tx_pend)");
91 pend->conn = conn;
92 pend->cursor = conn->tx_curs_sent;
93 pend->p_cursor = conn->local_tx_ctrl.prod;
94 pend->ctrl_seq = conn->tx_cdc_seq;
95}
96
97int smc_cdc_msg_send(struct smc_connection *conn,
98 struct smc_wr_buf *wr_buf,
99 struct smc_cdc_tx_pend *pend)
100{
101 struct smc_link *link = conn->lnk;
102 union smc_host_cursor cfed;
103 int rc;
104
105 smc_cdc_add_pending_send(conn, pend);
106
107 conn->tx_cdc_seq++;
108 conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
109 smc_host_msg_to_cdc((struct smc_cdc_msg *)wr_buf, conn, &cfed);
110 rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
111 if (!rc) {
112 smc_curs_copy(&conn->rx_curs_confirmed, &cfed, conn);
113 conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
114 } else {
115 conn->tx_cdc_seq--;
116 conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
117 }
118
119 return rc;
120}
121
122/* send a validation msg indicating the move of a conn to an other QP link */
123int smcr_cdc_msg_send_validation(struct smc_connection *conn,
124 struct smc_cdc_tx_pend *pend,
125 struct smc_wr_buf *wr_buf)
126{
127 struct smc_host_cdc_msg *local = &conn->local_tx_ctrl;
128 struct smc_link *link = conn->lnk;
129 struct smc_cdc_msg *peer;
130 int rc;
131
132 peer = (struct smc_cdc_msg *)wr_buf;
133 peer->common.type = local->common.type;
134 peer->len = local->len;
135 peer->seqno = htons(conn->tx_cdc_seq_fin); /* seqno last compl. tx */
136 peer->token = htonl(local->token);
137 peer->prod_flags.failover_validation = 1;
138
139 rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
140 return rc;
141}
142
143static int smcr_cdc_get_slot_and_msg_send(struct smc_connection *conn)
144{
145 struct smc_cdc_tx_pend *pend;
146 struct smc_wr_buf *wr_buf;
147 struct smc_link *link;
148 bool again = false;
149 int rc;
150
151again:
152 link = conn->lnk;
153 if (!smc_wr_tx_link_hold(link))
154 return -ENOLINK;
155 rc = smc_cdc_get_free_slot(conn, link, &wr_buf, NULL, &pend);
156 if (rc)
157 goto put_out;
158
159 spin_lock_bh(&conn->send_lock);
160 if (link != conn->lnk) {
161 /* link of connection changed, try again one time*/
162 spin_unlock_bh(&conn->send_lock);
163 smc_wr_tx_put_slot(link,
164 (struct smc_wr_tx_pend_priv *)pend);
165 smc_wr_tx_link_put(link);
166 if (again)
167 return -ENOLINK;
168 again = true;
169 goto again;
170 }
171 rc = smc_cdc_msg_send(conn, wr_buf, pend);
172 spin_unlock_bh(&conn->send_lock);
173put_out:
174 smc_wr_tx_link_put(link);
175 return rc;
176}
177
178int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn)
179{
180 int rc;
181
182 if (!conn->lgr || (conn->lgr->is_smcd && conn->lgr->peer_shutdown))
183 return -EPIPE;
184
185 if (conn->lgr->is_smcd) {
186 spin_lock_bh(&conn->send_lock);
187 rc = smcd_cdc_msg_send(conn);
188 spin_unlock_bh(&conn->send_lock);
189 } else {
190 rc = smcr_cdc_get_slot_and_msg_send(conn);
191 }
192
193 return rc;
194}
195
196static bool smc_cdc_tx_filter(struct smc_wr_tx_pend_priv *tx_pend,
197 unsigned long data)
198{
199 struct smc_connection *conn = (struct smc_connection *)data;
200 struct smc_cdc_tx_pend *cdc_pend =
201 (struct smc_cdc_tx_pend *)tx_pend;
202
203 return cdc_pend->conn == conn;
204}
205
206static void smc_cdc_tx_dismisser(struct smc_wr_tx_pend_priv *tx_pend)
207{
208 struct smc_cdc_tx_pend *cdc_pend =
209 (struct smc_cdc_tx_pend *)tx_pend;
210
211 cdc_pend->conn = NULL;
212}
213
214void smc_cdc_tx_dismiss_slots(struct smc_connection *conn)
215{
216 struct smc_link *link = conn->lnk;
217
218 smc_wr_tx_dismiss_slots(link, SMC_CDC_MSG_TYPE,
219 smc_cdc_tx_filter, smc_cdc_tx_dismisser,
220 (unsigned long)conn);
221}
222
223/* Send a SMC-D CDC header.
224 * This increments the free space available in our send buffer.
225 * Also update the confirmed receive buffer with what was sent to the peer.
226 */
227int smcd_cdc_msg_send(struct smc_connection *conn)
228{
229 struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
230 union smc_host_cursor curs;
231 struct smcd_cdc_msg cdc;
232 int rc, diff;
233
234 memset(&cdc, 0, sizeof(cdc));
235 cdc.common.type = SMC_CDC_MSG_TYPE;
236 curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.prod.acurs);
237 cdc.prod.wrap = curs.wrap;
238 cdc.prod.count = curs.count;
239 curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.cons.acurs);
240 cdc.cons.wrap = curs.wrap;
241 cdc.cons.count = curs.count;
242 cdc.cons.prod_flags = conn->local_tx_ctrl.prod_flags;
243 cdc.cons.conn_state_flags = conn->local_tx_ctrl.conn_state_flags;
244 rc = smcd_tx_ism_write(conn, &cdc, sizeof(cdc), 0, 1);
245 if (rc)
246 return rc;
247 smc_curs_copy(&conn->rx_curs_confirmed, &curs, conn);
248 conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
249 /* Calculate transmitted data and increment free send buffer space */
250 diff = smc_curs_diff(conn->sndbuf_desc->len, &conn->tx_curs_fin,
251 &conn->tx_curs_sent);
252 /* increased by confirmed number of bytes */
253 smp_mb__before_atomic();
254 atomic_add(diff, &conn->sndbuf_space);
255 /* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
256 smp_mb__after_atomic();
257 smc_curs_copy(&conn->tx_curs_fin, &conn->tx_curs_sent, conn);
258
259 smc_tx_sndbuf_nonfull(smc);
260 return rc;
261}
262
263/********************************* receive ***********************************/
264
265static inline bool smc_cdc_before(u16 seq1, u16 seq2)
266{
267 return (s16)(seq1 - seq2) < 0;
268}
269
270static void smc_cdc_handle_urg_data_arrival(struct smc_sock *smc,
271 int *diff_prod)
272{
273 struct smc_connection *conn = &smc->conn;
274 char *base;
275
276 /* new data included urgent business */
277 smc_curs_copy(&conn->urg_curs, &conn->local_rx_ctrl.prod, conn);
278 conn->urg_state = SMC_URG_VALID;
279 if (!sock_flag(&smc->sk, SOCK_URGINLINE))
280 /* we'll skip the urgent byte, so don't account for it */
281 (*diff_prod)--;
282 base = (char *)conn->rmb_desc->cpu_addr + conn->rx_off;
283 if (conn->urg_curs.count)
284 conn->urg_rx_byte = *(base + conn->urg_curs.count - 1);
285 else
286 conn->urg_rx_byte = *(base + conn->rmb_desc->len - 1);
287 sk_send_sigurg(&smc->sk);
288}
289
290static void smc_cdc_msg_validate(struct smc_sock *smc, struct smc_cdc_msg *cdc,
291 struct smc_link *link)
292{
293 struct smc_connection *conn = &smc->conn;
294 u16 recv_seq = ntohs(cdc->seqno);
295 s16 diff;
296
297 /* check that seqnum was seen before */
298 diff = conn->local_rx_ctrl.seqno - recv_seq;
299 if (diff < 0) { /* diff larger than 0x7fff */
300 /* drop connection */
301 conn->out_of_sync = 1; /* prevent any further receives */
302 spin_lock_bh(&conn->send_lock);
303 conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
304 conn->lnk = link;
305 spin_unlock_bh(&conn->send_lock);
306 sock_hold(&smc->sk); /* sock_put in abort_work */
307 if (!queue_work(smc_close_wq, &conn->abort_work))
308 sock_put(&smc->sk);
309 }
310}
311
312static void smc_cdc_msg_recv_action(struct smc_sock *smc,
313 struct smc_cdc_msg *cdc)
314{
315 union smc_host_cursor cons_old, prod_old;
316 struct smc_connection *conn = &smc->conn;
317 int diff_cons, diff_prod;
318
319 smc_curs_copy(&prod_old, &conn->local_rx_ctrl.prod, conn);
320 smc_curs_copy(&cons_old, &conn->local_rx_ctrl.cons, conn);
321 smc_cdc_msg_to_host(&conn->local_rx_ctrl, cdc, conn);
322
323 diff_cons = smc_curs_diff(conn->peer_rmbe_size, &cons_old,
324 &conn->local_rx_ctrl.cons);
325 if (diff_cons) {
326 /* peer_rmbe_space is decreased during data transfer with RDMA
327 * write
328 */
329 smp_mb__before_atomic();
330 atomic_add(diff_cons, &conn->peer_rmbe_space);
331 /* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
332 smp_mb__after_atomic();
333 }
334
335 diff_prod = smc_curs_diff(conn->rmb_desc->len, &prod_old,
336 &conn->local_rx_ctrl.prod);
337 if (diff_prod) {
338 if (conn->local_rx_ctrl.prod_flags.urg_data_present)
339 smc_cdc_handle_urg_data_arrival(smc, &diff_prod);
340 /* bytes_to_rcv is decreased in smc_recvmsg */
341 smp_mb__before_atomic();
342 atomic_add(diff_prod, &conn->bytes_to_rcv);
343 /* guarantee 0 <= bytes_to_rcv <= rmb_desc->len */
344 smp_mb__after_atomic();
345 smc->sk.sk_data_ready(&smc->sk);
346 } else {
347 if (conn->local_rx_ctrl.prod_flags.write_blocked)
348 smc->sk.sk_data_ready(&smc->sk);
349 if (conn->local_rx_ctrl.prod_flags.urg_data_pending)
350 conn->urg_state = SMC_URG_NOTYET;
351 }
352
353 /* trigger sndbuf consumer: RDMA write into peer RMBE and CDC */
354 if ((diff_cons && smc_tx_prepared_sends(conn)) ||
355 conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||
356 conn->local_rx_ctrl.prod_flags.urg_data_pending)
357 smc_tx_sndbuf_nonempty(conn);
358
359 if (diff_cons && conn->urg_tx_pend &&
360 atomic_read(&conn->peer_rmbe_space) == conn->peer_rmbe_size) {
361 /* urg data confirmed by peer, indicate we're ready for more */
362 conn->urg_tx_pend = false;
363 smc->sk.sk_write_space(&smc->sk);
364 }
365
366 if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
367 smc->sk.sk_err = ECONNRESET;
368 conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
369 }
370 if (smc_cdc_rxed_any_close_or_senddone(conn)) {
371 smc->sk.sk_shutdown |= RCV_SHUTDOWN;
372 if (smc->clcsock && smc->clcsock->sk)
373 smc->clcsock->sk->sk_shutdown |= RCV_SHUTDOWN;
374 sock_set_flag(&smc->sk, SOCK_DONE);
375 sock_hold(&smc->sk); /* sock_put in close_work */
376 if (!queue_work(smc_close_wq, &conn->close_work))
377 sock_put(&smc->sk);
378 }
379}
380
381/* called under tasklet context */
382static void smc_cdc_msg_recv(struct smc_sock *smc, struct smc_cdc_msg *cdc)
383{
384 sock_hold(&smc->sk);
385 bh_lock_sock(&smc->sk);
386 smc_cdc_msg_recv_action(smc, cdc);
387 bh_unlock_sock(&smc->sk);
388 sock_put(&smc->sk); /* no free sk in softirq-context */
389}
390
391/* Schedule a tasklet for this connection. Triggered from the ISM device IRQ
392 * handler to indicate update in the DMBE.
393 *
394 * Context:
395 * - tasklet context
396 */
397static void smcd_cdc_rx_tsklet(struct tasklet_struct *t)
398{
399 struct smc_connection *conn = from_tasklet(conn, t, rx_tsklet);
400 struct smcd_cdc_msg *data_cdc;
401 struct smcd_cdc_msg cdc;
402 struct smc_sock *smc;
403
404 if (!conn || conn->killed)
405 return;
406
407 data_cdc = (struct smcd_cdc_msg *)conn->rmb_desc->cpu_addr;
408 smcd_curs_copy(&cdc.prod, &data_cdc->prod, conn);
409 smcd_curs_copy(&cdc.cons, &data_cdc->cons, conn);
410 smc = container_of(conn, struct smc_sock, conn);
411 smc_cdc_msg_recv(smc, (struct smc_cdc_msg *)&cdc);
412}
413
414/* Initialize receive tasklet. Called from ISM device IRQ handler to start
415 * receiver side.
416 */
417void smcd_cdc_rx_init(struct smc_connection *conn)
418{
419 tasklet_setup(&conn->rx_tsklet, smcd_cdc_rx_tsklet);
420}
421
422/***************************** init, exit, misc ******************************/
423
424static void smc_cdc_rx_handler(struct ib_wc *wc, void *buf)
425{
426 struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
427 struct smc_cdc_msg *cdc = buf;
428 struct smc_connection *conn;
429 struct smc_link_group *lgr;
430 struct smc_sock *smc;
431
432 if (wc->byte_len < offsetof(struct smc_cdc_msg, reserved))
433 return; /* short message */
434 if (cdc->len != SMC_WR_TX_SIZE)
435 return; /* invalid message */
436
437 /* lookup connection */
438 lgr = smc_get_lgr(link);
439 read_lock_bh(&lgr->conns_lock);
440 conn = smc_lgr_find_conn(ntohl(cdc->token), lgr);
441 read_unlock_bh(&lgr->conns_lock);
442 if (!conn || conn->out_of_sync)
443 return;
444 smc = container_of(conn, struct smc_sock, conn);
445
446 if (cdc->prod_flags.failover_validation) {
447 smc_cdc_msg_validate(smc, cdc, link);
448 return;
449 }
450 if (smc_cdc_before(ntohs(cdc->seqno),
451 conn->local_rx_ctrl.seqno))
452 /* received seqno is old */
453 return;
454
455 smc_cdc_msg_recv(smc, cdc);
456}
457
458static struct smc_wr_rx_handler smc_cdc_rx_handlers[] = {
459 {
460 .handler = smc_cdc_rx_handler,
461 .type = SMC_CDC_MSG_TYPE
462 },
463 {
464 .handler = NULL,
465 }
466};
467
468int __init smc_cdc_init(void)
469{
470 struct smc_wr_rx_handler *handler;
471 int rc = 0;
472
473 for (handler = smc_cdc_rx_handlers; handler->handler; handler++) {
474 INIT_HLIST_NODE(&handler->list);
475 rc = smc_wr_rx_register_handler(handler);
476 if (rc)
477 break;
478 }
479 return rc;
480}