net/smc/smc_tx.c

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