net/smc/smc_tx.c

   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
  35 /***************************** sndbuf producer *******************************/
  36
  37 /* callback implementation for sk.sk_write_space()
  38  * to wakeup sndbuf producers that blocked with smc_tx_wait().
  39  * called under sk_socket lock.
  40  */
  41 static void smc_tx_write_space(struct sock *sk)
  42 {
  43         struct socket *sock = sk->sk_socket;
  44         struct smc_sock *smc = smc_sk(sk);
  45         struct socket_wq *wq;
  46
  47         /* similar to sk_stream_write_space */
  48         if (atomic_read(&smc->conn.sndbuf_space) && sock) {
  49                 if (test_bit(SOCK_NOSPACE, &sock->flags))
  50                         SMC_STAT_RMB_TX_FULL(smc, !smc->conn.lnk);
  51                 clear_bit(SOCK_NOSPACE, &sock->flags);
  52                 rcu_read_lock();
  53                 wq = rcu_dereference(sk->sk_wq);
  54                 if (skwq_has_sleeper(wq))
  55                         wake_up_interruptible_poll(&wq->wait,
  56                                                    EPOLLOUT | EPOLLWRNORM |
  57                                                    EPOLLWRBAND);
  58                 if (wq && wq->fasync_list && !(sk->sk_shutdown & SEND_SHUTDOWN))
  59                         sock_wake_async(wq, SOCK_WAKE_SPACE, POLL_OUT);
  60                 rcu_read_unlock();
  61         }
  62 }
  63
  64 /* Wakeup sndbuf producers that blocked with smc_tx_wait().
  65  * Cf. tcp_data_snd_check()=>tcp_check_space()=>tcp_new_space().
  66  */
  67 void smc_tx_sndbuf_nonfull(struct smc_sock *smc)
  68 {
  69         if (smc->sk.sk_socket &&
  70             test_bit(SOCK_NOSPACE, &smc->sk.sk_socket->flags))
  71                 smc->sk.sk_write_space(&smc->sk);
  72 }
  73
  74 /* blocks sndbuf producer until at least one byte of free space available
  75  * or urgent Byte was consumed
  76  */
  77 static int smc_tx_wait(struct smc_sock *smc, int flags)
  78 {
  79         DEFINE_WAIT_FUNC(wait, woken_wake_function);
  80         struct smc_connection *conn = &smc->conn;
  81         struct sock *sk = &smc->sk;
  82         long timeo;
  83         int rc = 0;
  84
  85         /* similar to sk_stream_wait_memory */
  86         timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
  87         add_wait_queue(sk_sleep(sk), &wait);
  88         while (1) {
  89                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
  90                 if (sk->sk_err ||
  91                     (sk->sk_shutdown & SEND_SHUTDOWN) ||
  92                     conn->killed ||
  93                     conn->local_tx_ctrl.conn_state_flags.peer_done_writing) {
  94                         rc = -EPIPE;
  95                         break;
  96                 }
  97                 if (smc_cdc_rxed_any_close(conn)) {
  98                         rc = -ECONNRESET;
  99                         break;
 100                 }
 101                 if (!timeo) {
 102                         /* ensure EPOLLOUT is subsequently generated */
 103                         set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 104                         rc = -EAGAIN;
 105                         break;
 106                 }
 107                 if (signal_pending(current)) {
 108                         rc = sock_intr_errno(timeo);
 109                         break;
 110                 }
 111                 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 112                 if (atomic_read(&conn->sndbuf_space) && !conn->urg_tx_pend)
 113                         break; /* at least 1 byte of free & no urgent data */
 114                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 115                 sk_wait_event(sk, &timeo,
 116                               READ_ONCE(sk->sk_err) ||
 117                               (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) ||
 118                               smc_cdc_rxed_any_close(conn) ||
 119                               (atomic_read(&conn->sndbuf_space) &&
 120                                !conn->urg_tx_pend),
 121                               &wait);
 122         }
 123         remove_wait_queue(sk_sleep(sk), &wait);
 124         return rc;
 125 }
 126
 127 static bool smc_tx_is_corked(struct smc_sock *smc)
 128 {
 129         struct tcp_sock *tp = tcp_sk(smc->clcsock->sk);
 130
 131         return (tp->nonagle & TCP_NAGLE_CORK) ? true : false;
 132 }
 133
 134 /* If we have pending CDC messages, do not send:
 135  * Because CQE of this CDC message will happen shortly, it gives
 136  * a chance to coalesce future sendmsg() payload in to one RDMA Write,
 137  * without need for a timer, and with no latency trade off.
 138  * Algorithm here:
 139  *  1. First message should never cork
 140  *  2. If we have pending Tx CDC messages, wait for the first CDC
 141  *     message's completion
 142  *  3. Don't cork to much data in a single RDMA Write to prevent burst
 143  *     traffic, total corked message should not exceed sendbuf/2
 144  */
 145 static bool smc_should_autocork(struct smc_sock *smc)
 146 {
 147         struct smc_connection *conn = &smc->conn;
 148         int corking_size;
 149
 150         corking_size = min_t(unsigned int, conn->sndbuf_desc->len >> 1,
 151                              sock_net(&smc->sk)->smc.sysctl_autocorking_size);
 152
 153         if (atomic_read(&conn->cdc_pend_tx_wr) == 0 ||
 154             smc_tx_prepared_sends(conn) > corking_size)
 155                 return false;
 156         return true;
 157 }
 158
 159 static bool smc_tx_should_cork(struct smc_sock *smc, struct msghdr *msg)
 160 {
 161         struct smc_connection *conn = &smc->conn;
 162
 163         if (smc_should_autocork(smc))
 164                 return true;
 165
 166         /* for a corked socket defer the RDMA writes if
 167          * sndbuf_space is still available. The applications
 168          * should known how/when to uncork it.
 169          */
 170         if ((msg->msg_flags & MSG_MORE ||
 171              smc_tx_is_corked(smc)) &&
 172             atomic_read(&conn->sndbuf_space))
 173                 return true;
 174
 175         return false;
 176 }
 177
 178 /* sndbuf producer: main API called by socket layer.
 179  * called under sock lock.
 180  */
 181 int smc_tx_sendmsg(struct smc_sock *smc, struct msghdr *msg, size_t len)
 182 {
 183         size_t copylen, send_done = 0, send_remaining = len;
 184         size_t chunk_len, chunk_off, chunk_len_sum;
 185         struct smc_connection *conn = &smc->conn;
 186         union smc_host_cursor prep;
 187         struct sock *sk = &smc->sk;
 188         char *sndbuf_base;
 189         int tx_cnt_prep;
 190         int writespace;
 191         int rc, chunk;
 192
 193         /* This should be in poll */
 194         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 195
 196         if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
 197                 rc = -EPIPE;
 198                 goto out_err;
 199         }
 200
 201         if (sk->sk_state == SMC_INIT)
 202                 return -ENOTCONN;
 203
 204         if (len > conn->sndbuf_desc->len)
 205                 SMC_STAT_RMB_TX_SIZE_SMALL(smc, !conn->lnk);
 206
 207         if (len > conn->peer_rmbe_size)
 208                 SMC_STAT_RMB_TX_PEER_SIZE_SMALL(smc, !conn->lnk);
 209
 210         if (msg->msg_flags & MSG_OOB)
 211                 SMC_STAT_INC(smc, urg_data_cnt);
 212
 213         while (msg_data_left(msg)) {
 214                 if (smc->sk.sk_shutdown & SEND_SHUTDOWN ||
 215                     (smc->sk.sk_err == ECONNABORTED) ||
 216                     conn->killed)
 217                         return -EPIPE;
 218                 if (smc_cdc_rxed_any_close(conn))
 219                         return send_done ?: -ECONNRESET;
 220
 221                 if (msg->msg_flags & MSG_OOB)
 222                         conn->local_tx_ctrl.prod_flags.urg_data_pending = 1;
 223
 224                 if (!atomic_read(&conn->sndbuf_space) || conn->urg_tx_pend) {
 225                         if (send_done)
 226                                 return send_done;
 227                         rc = smc_tx_wait(smc, msg->msg_flags);
 228                         if (rc)
 229                                 goto out_err;
 230                         continue;
 231                 }
 232
 233                 /* initialize variables for 1st iteration of subsequent loop */
 234                 /* could be just 1 byte, even after smc_tx_wait above */
 235                 writespace = atomic_read(&conn->sndbuf_space);
 236                 /* not more than what user space asked for */
 237                 copylen = min_t(size_t, send_remaining, writespace);
 238                 /* determine start of sndbuf */
 239                 sndbuf_base = conn->sndbuf_desc->cpu_addr;
 240                 smc_curs_copy(&prep, &conn->tx_curs_prep, conn);
 241                 tx_cnt_prep = prep.count;
 242                 /* determine chunks where to write into sndbuf */
 243                 /* either unwrapped case, or 1st chunk of wrapped case */
 244                 chunk_len = min_t(size_t, copylen, conn->sndbuf_desc->len -
 245                                   tx_cnt_prep);
 246                 chunk_len_sum = chunk_len;
 247                 chunk_off = tx_cnt_prep;
 248                 for (chunk = 0; chunk < 2; chunk++) {
 249                         rc = memcpy_from_msg(sndbuf_base + chunk_off,
 250                                              msg, chunk_len);
 251                         if (rc) {
 252                                 smc_sndbuf_sync_sg_for_device(conn);
 253                                 if (send_done)
 254                                         return send_done;
 255                                 goto out_err;
 256                         }
 257                         send_done += chunk_len;
 258                         send_remaining -= chunk_len;
 259
 260                         if (chunk_len_sum == copylen)
 261                                 break; /* either on 1st or 2nd iteration */
 262                         /* prepare next (== 2nd) iteration */
 263                         chunk_len = copylen - chunk_len; /* remainder */
 264                         chunk_len_sum += chunk_len;
 265                         chunk_off = 0; /* modulo offset in send ring buffer */
 266                 }
 267                 smc_sndbuf_sync_sg_for_device(conn);
 268                 /* update cursors */
 269                 smc_curs_add(conn->sndbuf_desc->len, &prep, copylen);
 270                 smc_curs_copy(&conn->tx_curs_prep, &prep, conn);
 271                 /* increased in send tasklet smc_cdc_tx_handler() */
 272                 smp_mb__before_atomic();
 273                 atomic_sub(copylen, &conn->sndbuf_space);
 274                 /* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
 275                 smp_mb__after_atomic();
 276                 /* since we just produced more new data into sndbuf,
 277                  * trigger sndbuf consumer: RDMA write into peer RMBE and CDC
 278                  */
 279                 if ((msg->msg_flags & MSG_OOB) && !send_remaining)
 280                         conn->urg_tx_pend = true;
 281                 /* If we need to cork, do nothing and wait for the next
 282                  * sendmsg() call or push on tx completion
 283                  */
 284                 if (!smc_tx_should_cork(smc, msg))
 285                         smc_tx_sndbuf_nonempty(conn);
 286
 287                 trace_smc_tx_sendmsg(smc, copylen);
 288         } /* while (msg_data_left(msg)) */
 289
 290         return send_done;
 291
 292 out_err:
 293         rc = sk_stream_error(sk, msg->msg_flags, rc);
 294         /* make sure we wake any epoll edge trigger waiter */
 295         if (unlikely(rc == -EAGAIN))
 296                 sk->sk_write_space(sk);
 297         return rc;
 298 }
 299
 300 /***************************** sndbuf consumer *******************************/
 301
 302 /* sndbuf consumer: actual data transfer of one target chunk with ISM write */
 303 int smcd_tx_ism_write(struct smc_connection *conn, void *data, size_t len,
 304                       u32 offset, int signal)
 305 {
 306         int rc;
 307
 308         rc = smc_ism_write(conn->lgr->smcd, conn->peer_token,
 309                            conn->peer_rmbe_idx, signal, conn->tx_off + offset,
 310                            data, len);
 311         if (rc)
 312                 conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
 313         return rc;
 314 }
 315
 316 /* sndbuf consumer: actual data transfer of one target chunk with RDMA write */
 317 static int smc_tx_rdma_write(struct smc_connection *conn, int peer_rmbe_offset,
 318                              int num_sges, struct ib_rdma_wr *rdma_wr)
 319 {
 320         struct smc_link_group *lgr = conn->lgr;
 321         struct smc_link *link = conn->lnk;
 322         int rc;
 323
 324         rdma_wr->wr.wr_id = smc_wr_tx_get_next_wr_id(link);
 325         rdma_wr->wr.num_sge = num_sges;
 326         rdma_wr->remote_addr =
 327                 lgr->rtokens[conn->rtoken_idx][link->link_idx].dma_addr +
 328                 /* RMBE within RMB */
 329                 conn->tx_off +
 330                 /* offset within RMBE */
 331                 peer_rmbe_offset;
 332         rdma_wr->rkey = lgr->rtokens[conn->rtoken_idx][link->link_idx].rkey;
 333         rc = ib_post_send(link->roce_qp, &rdma_wr->wr, NULL);
 334         if (rc)
 335                 smcr_link_down_cond_sched(link);
 336         return rc;
 337 }
 338
 339 /* sndbuf consumer */
 340 static inline void smc_tx_advance_cursors(struct smc_connection *conn,
 341                                           union smc_host_cursor *prod,
 342                                           union smc_host_cursor *sent,
 343                                           size_t len)
 344 {
 345         smc_curs_add(conn->peer_rmbe_size, prod, len);
 346         /* increased in recv tasklet smc_cdc_msg_rcv() */
 347         smp_mb__before_atomic();
 348         /* data in flight reduces usable snd_wnd */
 349         atomic_sub(len, &conn->peer_rmbe_space);
 350         /* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
 351         smp_mb__after_atomic();
 352         smc_curs_add(conn->sndbuf_desc->len, sent, len);
 353 }
 354
 355 /* SMC-R helper for smc_tx_rdma_writes() */
 356 static int smcr_tx_rdma_writes(struct smc_connection *conn, size_t len,
 357                                size_t src_off, size_t src_len,
 358                                size_t dst_off, size_t dst_len,
 359                                struct smc_rdma_wr *wr_rdma_buf)
 360 {
 361         struct smc_link *link = conn->lnk;
 362
 363         dma_addr_t dma_addr =
 364                 sg_dma_address(conn->sndbuf_desc->sgt[link->link_idx].sgl);
 365         u64 virt_addr = (uintptr_t)conn->sndbuf_desc->cpu_addr;
 366         int src_len_sum = src_len, dst_len_sum = dst_len;
 367         int sent_count = src_off;
 368         int srcchunk, dstchunk;
 369         int num_sges;
 370         int rc;
 371
 372         for (dstchunk = 0; dstchunk < 2; dstchunk++) {
 373                 struct ib_rdma_wr *wr = &wr_rdma_buf->wr_tx_rdma[dstchunk];
 374                 struct ib_sge *sge = wr->wr.sg_list;
 375                 u64 base_addr = dma_addr;
 376
 377                 if (dst_len < link->qp_attr.cap.max_inline_data) {
 378                         base_addr = virt_addr;
 379                         wr->wr.send_flags |= IB_SEND_INLINE;
 380                 } else {
 381                         wr->wr.send_flags &= ~IB_SEND_INLINE;
 382                 }
 383
 384                 num_sges = 0;
 385                 for (srcchunk = 0; srcchunk < 2; srcchunk++) {
 386                         sge[srcchunk].addr = conn->sndbuf_desc->is_vm ?
 387                                 (virt_addr + src_off) : (base_addr + src_off);
 388                         sge[srcchunk].length = src_len;
 389                         if (conn->sndbuf_desc->is_vm)
 390                                 sge[srcchunk].lkey =
 391                                         conn->sndbuf_desc->mr[link->link_idx]->lkey;
 392                         num_sges++;
 393
 394                         src_off += src_len;
 395                         if (src_off >= conn->sndbuf_desc->len)
 396                                 src_off -= conn->sndbuf_desc->len;
 397                                                 /* modulo in send ring */
 398                         if (src_len_sum == dst_len)
 399                                 break; /* either on 1st or 2nd iteration */
 400                         /* prepare next (== 2nd) iteration */
 401                         src_len = dst_len - src_len; /* remainder */
 402                         src_len_sum += src_len;
 403                 }
 404                 rc = smc_tx_rdma_write(conn, dst_off, num_sges, wr);
 405                 if (rc)
 406                         return rc;
 407                 if (dst_len_sum == len)
 408                         break; /* either on 1st or 2nd iteration */
 409                 /* prepare next (== 2nd) iteration */
 410                 dst_off = 0; /* modulo offset in RMBE ring buffer */
 411                 dst_len = len - dst_len; /* remainder */
 412                 dst_len_sum += dst_len;
 413                 src_len = min_t(int, dst_len, conn->sndbuf_desc->len -
 414                                 sent_count);
 415                 src_len_sum = src_len;
 416         }
 417         return 0;
 418 }
 419
 420 /* SMC-D helper for smc_tx_rdma_writes() */
 421 static int smcd_tx_rdma_writes(struct smc_connection *conn, size_t len,
 422                                size_t src_off, size_t src_len,
 423                                size_t dst_off, size_t dst_len)
 424 {
 425         int src_len_sum = src_len, dst_len_sum = dst_len;
 426         int srcchunk, dstchunk;
 427         int rc;
 428
 429         for (dstchunk = 0; dstchunk < 2; dstchunk++) {
 430                 for (srcchunk = 0; srcchunk < 2; srcchunk++) {
 431                         void *data = conn->sndbuf_desc->cpu_addr + src_off;
 432
 433                         rc = smcd_tx_ism_write(conn, data, src_len, dst_off +
 434                                                sizeof(struct smcd_cdc_msg), 0);
 435                         if (rc)
 436                                 return rc;
 437                         dst_off += src_len;
 438                         src_off += src_len;
 439                         if (src_off >= conn->sndbuf_desc->len)
 440                                 src_off -= conn->sndbuf_desc->len;
 441                                                 /* modulo in send ring */
 442                         if (src_len_sum == dst_len)
 443                                 break; /* either on 1st or 2nd iteration */
 444                         /* prepare next (== 2nd) iteration */
 445                         src_len = dst_len - src_len; /* remainder */
 446                         src_len_sum += src_len;
 447                 }
 448                 if (dst_len_sum == len)
 449                         break; /* either on 1st or 2nd iteration */
 450                 /* prepare next (== 2nd) iteration */
 451                 dst_off = 0; /* modulo offset in RMBE ring buffer */
 452                 dst_len = len - dst_len; /* remainder */
 453                 dst_len_sum += dst_len;
 454                 src_len = min_t(int, dst_len, conn->sndbuf_desc->len - src_off);
 455                 src_len_sum = src_len;
 456         }
 457         return 0;
 458 }
 459
 460 /* sndbuf consumer: prepare all necessary (src&dst) chunks of data transmit;
 461  * usable snd_wnd as max transmit
 462  */
 463 static int smc_tx_rdma_writes(struct smc_connection *conn,
 464                               struct smc_rdma_wr *wr_rdma_buf)
 465 {
 466         size_t len, src_len, dst_off, dst_len; /* current chunk values */
 467         union smc_host_cursor sent, prep, prod, cons;
 468         struct smc_cdc_producer_flags *pflags;
 469         int to_send, rmbespace;
 470         int rc;
 471
 472         /* source: sndbuf */
 473         smc_curs_copy(&sent, &conn->tx_curs_sent, conn);
 474         smc_curs_copy(&prep, &conn->tx_curs_prep, conn);
 475         /* cf. wmem_alloc - (snd_max - snd_una) */
 476         to_send = smc_curs_diff(conn->sndbuf_desc->len, &sent, &prep);
 477         if (to_send <= 0)
 478                 return 0;
 479
 480         /* destination: RMBE */
 481         /* cf. snd_wnd */
 482         rmbespace = atomic_read(&conn->peer_rmbe_space);
 483         if (rmbespace <= 0) {
 484                 struct smc_sock *smc = container_of(conn, struct smc_sock,
 485                                                     conn);
 486                 SMC_STAT_RMB_TX_PEER_FULL(smc, !conn->lnk);
 487                 return 0;
 488         }
 489         smc_curs_copy(&prod, &conn->local_tx_ctrl.prod, conn);
 490         smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
 491
 492         /* if usable snd_wnd closes ask peer to advertise once it opens again */
 493         pflags = &conn->local_tx_ctrl.prod_flags;
 494         pflags->write_blocked = (to_send >= rmbespace);
 495         /* cf. usable snd_wnd */
 496         len = min(to_send, rmbespace);
 497
 498         /* initialize variables for first iteration of subsequent nested loop */
 499         dst_off = prod.count;
 500         if (prod.wrap == cons.wrap) {
 501                 /* the filled destination area is unwrapped,
 502                  * hence the available free destination space is wrapped
 503                  * and we need 2 destination chunks of sum len; start with 1st
 504                  * which is limited by what's available in sndbuf
 505                  */
 506                 dst_len = min_t(size_t,
 507                                 conn->peer_rmbe_size - prod.count, len);
 508         } else {
 509                 /* the filled destination area is wrapped,
 510                  * hence the available free destination space is unwrapped
 511                  * and we need a single destination chunk of entire len
 512                  */
 513                 dst_len = len;
 514         }
 515         /* dst_len determines the maximum src_len */
 516         if (sent.count + dst_len <= conn->sndbuf_desc->len) {
 517                 /* unwrapped src case: single chunk of entire dst_len */
 518                 src_len = dst_len;
 519         } else {
 520                 /* wrapped src case: 2 chunks of sum dst_len; start with 1st: */
 521                 src_len = conn->sndbuf_desc->len - sent.count;
 522         }
 523
 524         if (conn->lgr->is_smcd)
 525                 rc = smcd_tx_rdma_writes(conn, len, sent.count, src_len,
 526                                          dst_off, dst_len);
 527         else
 528                 rc = smcr_tx_rdma_writes(conn, len, sent.count, src_len,
 529                                          dst_off, dst_len, wr_rdma_buf);
 530         if (rc)
 531                 return rc;
 532
 533         if (conn->urg_tx_pend && len == to_send)
 534                 pflags->urg_data_present = 1;
 535         smc_tx_advance_cursors(conn, &prod, &sent, len);
 536         /* update connection's cursors with advanced local cursors */
 537         smc_curs_copy(&conn->local_tx_ctrl.prod, &prod, conn);
 538                                                         /* dst: peer RMBE */
 539         smc_curs_copy(&conn->tx_curs_sent, &sent, conn);/* src: local sndbuf */
 540
 541         return 0;
 542 }
 543
 544 /* Wakeup sndbuf consumers from any context (IRQ or process)
 545  * since there is more data to transmit; usable snd_wnd as max transmit
 546  */
 547 static int smcr_tx_sndbuf_nonempty(struct smc_connection *conn)
 548 {
 549         struct smc_cdc_producer_flags *pflags = &conn->local_tx_ctrl.prod_flags;
 550         struct smc_link *link = conn->lnk;
 551         struct smc_rdma_wr *wr_rdma_buf;
 552         struct smc_cdc_tx_pend *pend;
 553         struct smc_wr_buf *wr_buf;
 554         int rc;
 555
 556         if (!link || !smc_wr_tx_link_hold(link))
 557                 return -ENOLINK;
 558         rc = smc_cdc_get_free_slot(conn, link, &wr_buf, &wr_rdma_buf, &pend);
 559         if (rc < 0) {
 560                 smc_wr_tx_link_put(link);
 561                 if (rc == -EBUSY) {
 562                         struct smc_sock *smc =
 563                                 container_of(conn, struct smc_sock, conn);
 564
 565                         if (smc->sk.sk_err == ECONNABORTED)
 566                                 return sock_error(&smc->sk);
 567                         if (conn->killed)
 568                                 return -EPIPE;
 569                         rc = 0;
 570                         mod_delayed_work(conn->lgr->tx_wq, &conn->tx_work,
 571                                          SMC_TX_WORK_DELAY);
 572                 }
 573                 return rc;
 574         }
 575
 576         spin_lock_bh(&conn->send_lock);
 577         if (link != conn->lnk) {
 578                 /* link of connection changed, tx_work will restart */
 579                 smc_wr_tx_put_slot(link,
 580                                    (struct smc_wr_tx_pend_priv *)pend);
 581                 rc = -ENOLINK;
 582                 goto out_unlock;
 583         }
 584         if (!pflags->urg_data_present) {
 585                 rc = smc_tx_rdma_writes(conn, wr_rdma_buf);
 586                 if (rc) {
 587                         smc_wr_tx_put_slot(link,
 588                                            (struct smc_wr_tx_pend_priv *)pend);
 589                         goto out_unlock;
 590                 }
 591         }
 592
 593         rc = smc_cdc_msg_send(conn, wr_buf, pend);
 594         if (!rc && pflags->urg_data_present) {
 595                 pflags->urg_data_pending = 0;
 596                 pflags->urg_data_present = 0;
 597         }
 598
 599 out_unlock:
 600         spin_unlock_bh(&conn->send_lock);
 601         smc_wr_tx_link_put(link);
 602         return rc;
 603 }
 604
 605 static int smcd_tx_sndbuf_nonempty(struct smc_connection *conn)
 606 {
 607         struct smc_cdc_producer_flags *pflags = &conn->local_tx_ctrl.prod_flags;
 608         int rc = 0;
 609
 610         spin_lock_bh(&conn->send_lock);
 611         if (!pflags->urg_data_present)
 612                 rc = smc_tx_rdma_writes(conn, NULL);
 613         if (!rc)
 614                 rc = smcd_cdc_msg_send(conn);
 615
 616         if (!rc && pflags->urg_data_present) {
 617                 pflags->urg_data_pending = 0;
 618                 pflags->urg_data_present = 0;
 619         }
 620         spin_unlock_bh(&conn->send_lock);
 621         return rc;
 622 }
 623
 624 static int __smc_tx_sndbuf_nonempty(struct smc_connection *conn)
 625 {
 626         struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
 627         int rc = 0;
 628
 629         /* No data in the send queue */
 630         if (unlikely(smc_tx_prepared_sends(conn) <= 0))
 631                 goto out;
 632
 633         /* Peer don't have RMBE space */
 634         if (unlikely(atomic_read(&conn->peer_rmbe_space) <= 0)) {
 635                 SMC_STAT_RMB_TX_PEER_FULL(smc, !conn->lnk);
 636                 goto out;
 637         }
 638
 639         if (conn->killed ||
 640             conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
 641                 rc = -EPIPE;    /* connection being aborted */
 642                 goto out;
 643         }
 644         if (conn->lgr->is_smcd)
 645                 rc = smcd_tx_sndbuf_nonempty(conn);
 646         else
 647                 rc = smcr_tx_sndbuf_nonempty(conn);
 648
 649         if (!rc) {
 650                 /* trigger socket release if connection is closing */
 651                 smc_close_wake_tx_prepared(smc);
 652         }
 653
 654 out:
 655         return rc;
 656 }
 657
 658 int smc_tx_sndbuf_nonempty(struct smc_connection *conn)
 659 {
 660         int rc;
 661
 662         /* This make sure only one can send simultaneously to prevent wasting
 663          * of CPU and CDC slot.
 664          * Record whether someone has tried to push while we are pushing.
 665          */
 666         if (atomic_inc_return(&conn->tx_pushing) > 1)
 667                 return 0;
 668
 669 again:
 670         atomic_set(&conn->tx_pushing, 1);
 671         smp_wmb(); /* Make sure tx_pushing is 1 before real send */
 672         rc = __smc_tx_sndbuf_nonempty(conn);
 673
 674         /* We need to check whether someone else have added some data into
 675          * the send queue and tried to push but failed after the atomic_set()
 676          * when we are pushing.
 677          * If so, we need to push again to prevent those data hang in the send
 678          * queue.
 679          */
 680         if (unlikely(!atomic_dec_and_test(&conn->tx_pushing)))
 681                 goto again;
 682
 683         return rc;
 684 }
 685
 686 /* Wakeup sndbuf consumers from process context
 687  * since there is more data to transmit. The caller
 688  * must hold sock lock.
 689  */
 690 void smc_tx_pending(struct smc_connection *conn)
 691 {
 692         struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
 693         int rc;
 694
 695         if (smc->sk.sk_err)
 696                 return;
 697
 698         rc = smc_tx_sndbuf_nonempty(conn);
 699         if (!rc && conn->local_rx_ctrl.prod_flags.write_blocked &&
 700             !atomic_read(&conn->bytes_to_rcv))
 701                 conn->local_rx_ctrl.prod_flags.write_blocked = 0;
 702 }
 703
 704 /* Wakeup sndbuf consumers from process context
 705  * since there is more data to transmit in locked
 706  * sock.
 707  */
 708 void smc_tx_work(struct work_struct *work)
 709 {
 710         struct smc_connection *conn = container_of(to_delayed_work(work),
 711                                                    struct smc_connection,
 712                                                    tx_work);
 713         struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
 714
 715         lock_sock(&smc->sk);
 716         smc_tx_pending(conn);
 717         release_sock(&smc->sk);
 718 }
 719
 720 void smc_tx_consumer_update(struct smc_connection *conn, bool force)
 721 {
 722         union smc_host_cursor cfed, cons, prod;
 723         int sender_free = conn->rmb_desc->len;
 724         int to_confirm;
 725
 726         smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
 727         smc_curs_copy(&cfed, &conn->rx_curs_confirmed, conn);
 728         to_confirm = smc_curs_diff(conn->rmb_desc->len, &cfed, &cons);
 729         if (to_confirm > conn->rmbe_update_limit) {
 730                 smc_curs_copy(&prod, &conn->local_rx_ctrl.prod, conn);
 731                 sender_free = conn->rmb_desc->len -
 732                               smc_curs_diff_large(conn->rmb_desc->len,
 733                                                   &cfed, &prod);
 734         }
 735
 736         if (conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||
 737             force ||
 738             ((to_confirm > conn->rmbe_update_limit) &&
 739              ((sender_free <= (conn->rmb_desc->len / 2)) ||
 740               conn->local_rx_ctrl.prod_flags.write_blocked))) {
 741                 if (conn->killed ||
 742                     conn->local_rx_ctrl.conn_state_flags.peer_conn_abort)
 743                         return;
 744                 if ((smc_cdc_get_slot_and_msg_send(conn) < 0) &&
 745                     !conn->killed) {
 746                         queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work,
 747                                            SMC_TX_WORK_DELAY);
 748                         return;
 749                 }
 750         }
 751         if (conn->local_rx_ctrl.prod_flags.write_blocked &&
 752             !atomic_read(&conn->bytes_to_rcv))
 753                 conn->local_rx_ctrl.prod_flags.write_blocked = 0;
 754 }
 755
 756 /***************************** send initialize *******************************/
 757
 758 /* Initialize send properties on connection establishment. NB: not __init! */
 759 void smc_tx_init(struct smc_sock *smc)
 760 {
 761         smc->sk.sk_write_space = smc_tx_write_space;
 762 }