1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Shared Memory Communications over RDMA (SMC-R) and RoCE 4 * 5 * Basic Transport Functions exploiting Infiniband API 6 * 7 * Copyright IBM Corp. 2016 8 * 9 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> 10 */ 11 12 #include <linux/socket.h> 13 #include <linux/if_vlan.h> 14 #include <linux/random.h> 15 #include <linux/workqueue.h> 16 #include <linux/wait.h> 17 #include <linux/reboot.h> 18 #include <net/tcp.h> 19 #include <net/sock.h> 20 #include <rdma/ib_verbs.h> 21 #include <rdma/ib_cache.h> 22 23 #include "smc.h" 24 #include "smc_clc.h" 25 #include "smc_core.h" 26 #include "smc_ib.h" 27 #include "smc_wr.h" 28 #include "smc_llc.h" 29 #include "smc_cdc.h" 30 #include "smc_close.h" 31 #include "smc_ism.h" 32 33 #define SMC_LGR_NUM_INCR 256 34 #define SMC_LGR_FREE_DELAY_SERV (600 * HZ) 35 #define SMC_LGR_FREE_DELAY_CLNT (SMC_LGR_FREE_DELAY_SERV + 10 * HZ) 36 #define SMC_LGR_FREE_DELAY_FAST (8 * HZ) 37 38 static struct smc_lgr_list smc_lgr_list = { /* established link groups */ 39 .lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock), 40 .list = LIST_HEAD_INIT(smc_lgr_list.list), 41 .num = 0, 42 }; 43 44 static atomic_t lgr_cnt = ATOMIC_INIT(0); /* number of existing link groups */ 45 static DECLARE_WAIT_QUEUE_HEAD(lgrs_deleted); 46 47 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb, 48 struct smc_buf_desc *buf_desc); 49 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft); 50 51 /* return head of link group list and its lock for a given link group */ 52 static inline struct list_head *smc_lgr_list_head(struct smc_link_group *lgr, 53 spinlock_t **lgr_lock) 54 { 55 if (lgr->is_smcd) { 56 *lgr_lock = &lgr->smcd->lgr_lock; 57 return &lgr->smcd->lgr_list; 58 } 59 60 *lgr_lock = &smc_lgr_list.lock; 61 return &smc_lgr_list.list; 62 } 63 64 static void smc_lgr_schedule_free_work(struct smc_link_group *lgr) 65 { 66 /* client link group creation always follows the server link group 67 * creation. For client use a somewhat higher removal delay time, 68 * otherwise there is a risk of out-of-sync link groups. 69 */ 70 if (!lgr->freeing && !lgr->freefast) { 71 mod_delayed_work(system_wq, &lgr->free_work, 72 (!lgr->is_smcd && lgr->role == SMC_CLNT) ? 73 SMC_LGR_FREE_DELAY_CLNT : 74 SMC_LGR_FREE_DELAY_SERV); 75 } 76 } 77 78 void smc_lgr_schedule_free_work_fast(struct smc_link_group *lgr) 79 { 80 if (!lgr->freeing && !lgr->freefast) { 81 lgr->freefast = 1; 82 mod_delayed_work(system_wq, &lgr->free_work, 83 SMC_LGR_FREE_DELAY_FAST); 84 } 85 } 86 87 /* Register connection's alert token in our lookup structure. 88 * To use rbtrees we have to implement our own insert core. 89 * Requires @conns_lock 90 * @smc connection to register 91 * Returns 0 on success, != otherwise. 92 */ 93 static void smc_lgr_add_alert_token(struct smc_connection *conn) 94 { 95 struct rb_node **link, *parent = NULL; 96 u32 token = conn->alert_token_local; 97 98 link = &conn->lgr->conns_all.rb_node; 99 while (*link) { 100 struct smc_connection *cur = rb_entry(*link, 101 struct smc_connection, alert_node); 102 103 parent = *link; 104 if (cur->alert_token_local > token) 105 link = &parent->rb_left; 106 else 107 link = &parent->rb_right; 108 } 109 /* Put the new node there */ 110 rb_link_node(&conn->alert_node, parent, link); 111 rb_insert_color(&conn->alert_node, &conn->lgr->conns_all); 112 } 113 114 /* Register connection in link group by assigning an alert token 115 * registered in a search tree. 116 * Requires @conns_lock 117 * Note that '0' is a reserved value and not assigned. 118 */ 119 static void smc_lgr_register_conn(struct smc_connection *conn) 120 { 121 struct smc_sock *smc = container_of(conn, struct smc_sock, conn); 122 static atomic_t nexttoken = ATOMIC_INIT(0); 123 124 /* find a new alert_token_local value not yet used by some connection 125 * in this link group 126 */ 127 sock_hold(&smc->sk); /* sock_put in smc_lgr_unregister_conn() */ 128 while (!conn->alert_token_local) { 129 conn->alert_token_local = atomic_inc_return(&nexttoken); 130 if (smc_lgr_find_conn(conn->alert_token_local, conn->lgr)) 131 conn->alert_token_local = 0; 132 } 133 smc_lgr_add_alert_token(conn); 134 conn->lgr->conns_num++; 135 } 136 137 /* Unregister connection and reset the alert token of the given connection< 138 */ 139 static void __smc_lgr_unregister_conn(struct smc_connection *conn) 140 { 141 struct smc_sock *smc = container_of(conn, struct smc_sock, conn); 142 struct smc_link_group *lgr = conn->lgr; 143 144 rb_erase(&conn->alert_node, &lgr->conns_all); 145 lgr->conns_num--; 146 conn->alert_token_local = 0; 147 sock_put(&smc->sk); /* sock_hold in smc_lgr_register_conn() */ 148 } 149 150 /* Unregister connection from lgr 151 */ 152 static void smc_lgr_unregister_conn(struct smc_connection *conn) 153 { 154 struct smc_link_group *lgr = conn->lgr; 155 156 if (!lgr) 157 return; 158 write_lock_bh(&lgr->conns_lock); 159 if (conn->alert_token_local) { 160 __smc_lgr_unregister_conn(conn); 161 } 162 write_unlock_bh(&lgr->conns_lock); 163 conn->lgr = NULL; 164 } 165 166 /* Send delete link, either as client to request the initiation 167 * of the DELETE LINK sequence from server; or as server to 168 * initiate the delete processing. See smc_llc_rx_delete_link(). 169 */ 170 static int smc_link_send_delete(struct smc_link *lnk, bool orderly) 171 { 172 if (lnk->state == SMC_LNK_ACTIVE && 173 !smc_llc_send_delete_link(lnk, SMC_LLC_REQ, orderly)) { 174 smc_llc_link_deleting(lnk); 175 return 0; 176 } 177 return -ENOTCONN; 178 } 179 180 static void smc_lgr_free(struct smc_link_group *lgr); 181 182 static void smc_lgr_free_work(struct work_struct *work) 183 { 184 struct smc_link_group *lgr = container_of(to_delayed_work(work), 185 struct smc_link_group, 186 free_work); 187 spinlock_t *lgr_lock; 188 struct smc_link *lnk; 189 bool conns; 190 191 smc_lgr_list_head(lgr, &lgr_lock); 192 spin_lock_bh(lgr_lock); 193 if (lgr->freeing) { 194 spin_unlock_bh(lgr_lock); 195 return; 196 } 197 read_lock_bh(&lgr->conns_lock); 198 conns = RB_EMPTY_ROOT(&lgr->conns_all); 199 read_unlock_bh(&lgr->conns_lock); 200 if (!conns) { /* number of lgr connections is no longer zero */ 201 spin_unlock_bh(lgr_lock); 202 return; 203 } 204 list_del_init(&lgr->list); /* remove from smc_lgr_list */ 205 206 lnk = &lgr->lnk[SMC_SINGLE_LINK]; 207 if (!lgr->is_smcd && !lgr->terminating) { 208 /* try to send del link msg, on error free lgr immediately */ 209 if (lnk->state == SMC_LNK_ACTIVE && 210 !smc_link_send_delete(lnk, true)) { 211 /* reschedule in case we never receive a response */ 212 smc_lgr_schedule_free_work(lgr); 213 spin_unlock_bh(lgr_lock); 214 return; 215 } 216 } 217 lgr->freeing = 1; /* this instance does the freeing, no new schedule */ 218 spin_unlock_bh(lgr_lock); 219 cancel_delayed_work(&lgr->free_work); 220 221 if (!lgr->is_smcd && lnk->state != SMC_LNK_INACTIVE) 222 smc_llc_link_inactive(lnk); 223 if (lgr->is_smcd && !lgr->terminating) 224 smc_ism_signal_shutdown(lgr); 225 smc_lgr_free(lgr); 226 } 227 228 static void smc_lgr_terminate_work(struct work_struct *work) 229 { 230 struct smc_link_group *lgr = container_of(work, struct smc_link_group, 231 terminate_work); 232 233 __smc_lgr_terminate(lgr, true); 234 } 235 236 /* create a new SMC link group */ 237 static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini) 238 { 239 struct smc_link_group *lgr; 240 struct list_head *lgr_list; 241 struct smc_link *lnk; 242 spinlock_t *lgr_lock; 243 u8 rndvec[3]; 244 int rc = 0; 245 int i; 246 247 if (ini->is_smcd && ini->vlan_id) { 248 if (smc_ism_get_vlan(ini->ism_dev, ini->vlan_id)) { 249 rc = SMC_CLC_DECL_ISMVLANERR; 250 goto out; 251 } 252 } 253 254 lgr = kzalloc(sizeof(*lgr), GFP_KERNEL); 255 if (!lgr) { 256 rc = SMC_CLC_DECL_MEM; 257 goto ism_put_vlan; 258 } 259 lgr->is_smcd = ini->is_smcd; 260 lgr->sync_err = 0; 261 lgr->terminating = 0; 262 lgr->freefast = 0; 263 lgr->freeing = 0; 264 lgr->vlan_id = ini->vlan_id; 265 rwlock_init(&lgr->sndbufs_lock); 266 rwlock_init(&lgr->rmbs_lock); 267 rwlock_init(&lgr->conns_lock); 268 for (i = 0; i < SMC_RMBE_SIZES; i++) { 269 INIT_LIST_HEAD(&lgr->sndbufs[i]); 270 INIT_LIST_HEAD(&lgr->rmbs[i]); 271 } 272 smc_lgr_list.num += SMC_LGR_NUM_INCR; 273 memcpy(&lgr->id, (u8 *)&smc_lgr_list.num, SMC_LGR_ID_SIZE); 274 INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work); 275 INIT_WORK(&lgr->terminate_work, smc_lgr_terminate_work); 276 lgr->conns_all = RB_ROOT; 277 if (ini->is_smcd) { 278 /* SMC-D specific settings */ 279 get_device(&ini->ism_dev->dev); 280 lgr->peer_gid = ini->ism_gid; 281 lgr->smcd = ini->ism_dev; 282 lgr_list = &ini->ism_dev->lgr_list; 283 lgr_lock = &lgr->smcd->lgr_lock; 284 lgr->peer_shutdown = 0; 285 atomic_inc(&ini->ism_dev->lgr_cnt); 286 } else { 287 /* SMC-R specific settings */ 288 get_device(&ini->ib_dev->ibdev->dev); 289 lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT; 290 memcpy(lgr->peer_systemid, ini->ib_lcl->id_for_peer, 291 SMC_SYSTEMID_LEN); 292 293 lnk = &lgr->lnk[SMC_SINGLE_LINK]; 294 /* initialize link */ 295 lnk->state = SMC_LNK_ACTIVATING; 296 lnk->link_id = SMC_SINGLE_LINK; 297 lnk->smcibdev = ini->ib_dev; 298 lnk->ibport = ini->ib_port; 299 lgr_list = &smc_lgr_list.list; 300 lgr_lock = &smc_lgr_list.lock; 301 lnk->path_mtu = 302 ini->ib_dev->pattr[ini->ib_port - 1].active_mtu; 303 if (!ini->ib_dev->initialized) 304 smc_ib_setup_per_ibdev(ini->ib_dev); 305 get_random_bytes(rndvec, sizeof(rndvec)); 306 lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) + 307 (rndvec[2] << 16); 308 rc = smc_ib_determine_gid(lnk->smcibdev, lnk->ibport, 309 ini->vlan_id, lnk->gid, 310 &lnk->sgid_index); 311 if (rc) 312 goto free_lgr; 313 rc = smc_llc_link_init(lnk); 314 if (rc) 315 goto free_lgr; 316 rc = smc_wr_alloc_link_mem(lnk); 317 if (rc) 318 goto clear_llc_lnk; 319 rc = smc_ib_create_protection_domain(lnk); 320 if (rc) 321 goto free_link_mem; 322 rc = smc_ib_create_queue_pair(lnk); 323 if (rc) 324 goto dealloc_pd; 325 rc = smc_wr_create_link(lnk); 326 if (rc) 327 goto destroy_qp; 328 atomic_inc(&lgr_cnt); 329 atomic_inc(&ini->ib_dev->lnk_cnt); 330 } 331 smc->conn.lgr = lgr; 332 spin_lock_bh(lgr_lock); 333 list_add(&lgr->list, lgr_list); 334 spin_unlock_bh(lgr_lock); 335 return 0; 336 337 destroy_qp: 338 smc_ib_destroy_queue_pair(lnk); 339 dealloc_pd: 340 smc_ib_dealloc_protection_domain(lnk); 341 free_link_mem: 342 smc_wr_free_link_mem(lnk); 343 clear_llc_lnk: 344 smc_llc_link_clear(lnk); 345 free_lgr: 346 kfree(lgr); 347 ism_put_vlan: 348 if (ini->is_smcd && ini->vlan_id) 349 smc_ism_put_vlan(ini->ism_dev, ini->vlan_id); 350 out: 351 if (rc < 0) { 352 if (rc == -ENOMEM) 353 rc = SMC_CLC_DECL_MEM; 354 else 355 rc = SMC_CLC_DECL_INTERR; 356 } 357 return rc; 358 } 359 360 static void smc_buf_unuse(struct smc_connection *conn, 361 struct smc_link_group *lgr) 362 { 363 if (conn->sndbuf_desc) 364 conn->sndbuf_desc->used = 0; 365 if (conn->rmb_desc) { 366 if (!conn->rmb_desc->regerr) { 367 if (!lgr->is_smcd && !list_empty(&lgr->list)) { 368 /* unregister rmb with peer */ 369 smc_llc_do_delete_rkey( 370 &lgr->lnk[SMC_SINGLE_LINK], 371 conn->rmb_desc); 372 } 373 conn->rmb_desc->used = 0; 374 } else { 375 /* buf registration failed, reuse not possible */ 376 write_lock_bh(&lgr->rmbs_lock); 377 list_del(&conn->rmb_desc->list); 378 write_unlock_bh(&lgr->rmbs_lock); 379 380 smc_buf_free(lgr, true, conn->rmb_desc); 381 } 382 } 383 } 384 385 /* remove a finished connection from its link group */ 386 void smc_conn_free(struct smc_connection *conn) 387 { 388 struct smc_link_group *lgr = conn->lgr; 389 390 if (!lgr) 391 return; 392 if (lgr->is_smcd) { 393 if (!list_empty(&lgr->list)) 394 smc_ism_unset_conn(conn); 395 tasklet_kill(&conn->rx_tsklet); 396 } else { 397 smc_cdc_tx_dismiss_slots(conn); 398 } 399 if (!list_empty(&lgr->list)) { 400 smc_lgr_unregister_conn(conn); 401 smc_buf_unuse(conn, lgr); /* allow buffer reuse */ 402 } 403 404 if (!lgr->conns_num) 405 smc_lgr_schedule_free_work(lgr); 406 } 407 408 static void smc_link_clear(struct smc_link *lnk) 409 { 410 lnk->peer_qpn = 0; 411 smc_llc_link_clear(lnk); 412 smc_ib_modify_qp_reset(lnk); 413 smc_wr_free_link(lnk); 414 smc_ib_destroy_queue_pair(lnk); 415 smc_ib_dealloc_protection_domain(lnk); 416 smc_wr_free_link_mem(lnk); 417 if (!atomic_dec_return(&lnk->smcibdev->lnk_cnt)) 418 wake_up(&lnk->smcibdev->lnks_deleted); 419 } 420 421 static void smcr_buf_free(struct smc_link_group *lgr, bool is_rmb, 422 struct smc_buf_desc *buf_desc) 423 { 424 struct smc_link *lnk = &lgr->lnk[SMC_SINGLE_LINK]; 425 426 if (is_rmb) { 427 if (buf_desc->mr_rx[SMC_SINGLE_LINK]) 428 smc_ib_put_memory_region( 429 buf_desc->mr_rx[SMC_SINGLE_LINK]); 430 smc_ib_buf_unmap_sg(lnk->smcibdev, buf_desc, 431 DMA_FROM_DEVICE); 432 } else { 433 smc_ib_buf_unmap_sg(lnk->smcibdev, buf_desc, 434 DMA_TO_DEVICE); 435 } 436 sg_free_table(&buf_desc->sgt[SMC_SINGLE_LINK]); 437 if (buf_desc->pages) 438 __free_pages(buf_desc->pages, buf_desc->order); 439 kfree(buf_desc); 440 } 441 442 static void smcd_buf_free(struct smc_link_group *lgr, bool is_dmb, 443 struct smc_buf_desc *buf_desc) 444 { 445 if (is_dmb) { 446 /* restore original buf len */ 447 buf_desc->len += sizeof(struct smcd_cdc_msg); 448 smc_ism_unregister_dmb(lgr->smcd, buf_desc); 449 } else { 450 kfree(buf_desc->cpu_addr); 451 } 452 kfree(buf_desc); 453 } 454 455 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb, 456 struct smc_buf_desc *buf_desc) 457 { 458 if (lgr->is_smcd) 459 smcd_buf_free(lgr, is_rmb, buf_desc); 460 else 461 smcr_buf_free(lgr, is_rmb, buf_desc); 462 } 463 464 static void __smc_lgr_free_bufs(struct smc_link_group *lgr, bool is_rmb) 465 { 466 struct smc_buf_desc *buf_desc, *bf_desc; 467 struct list_head *buf_list; 468 int i; 469 470 for (i = 0; i < SMC_RMBE_SIZES; i++) { 471 if (is_rmb) 472 buf_list = &lgr->rmbs[i]; 473 else 474 buf_list = &lgr->sndbufs[i]; 475 list_for_each_entry_safe(buf_desc, bf_desc, buf_list, 476 list) { 477 list_del(&buf_desc->list); 478 smc_buf_free(lgr, is_rmb, buf_desc); 479 } 480 } 481 } 482 483 static void smc_lgr_free_bufs(struct smc_link_group *lgr) 484 { 485 /* free send buffers */ 486 __smc_lgr_free_bufs(lgr, false); 487 /* free rmbs */ 488 __smc_lgr_free_bufs(lgr, true); 489 } 490 491 /* remove a link group */ 492 static void smc_lgr_free(struct smc_link_group *lgr) 493 { 494 smc_lgr_free_bufs(lgr); 495 if (lgr->is_smcd) { 496 if (!lgr->terminating) { 497 smc_ism_put_vlan(lgr->smcd, lgr->vlan_id); 498 put_device(&lgr->smcd->dev); 499 } 500 if (!atomic_dec_return(&lgr->smcd->lgr_cnt)) 501 wake_up(&lgr->smcd->lgrs_deleted); 502 } else { 503 smc_link_clear(&lgr->lnk[SMC_SINGLE_LINK]); 504 put_device(&lgr->lnk[SMC_SINGLE_LINK].smcibdev->ibdev->dev); 505 if (!atomic_dec_return(&lgr_cnt)) 506 wake_up(&lgrs_deleted); 507 } 508 kfree(lgr); 509 } 510 511 void smc_lgr_forget(struct smc_link_group *lgr) 512 { 513 struct list_head *lgr_list; 514 spinlock_t *lgr_lock; 515 516 lgr_list = smc_lgr_list_head(lgr, &lgr_lock); 517 spin_lock_bh(lgr_lock); 518 /* do not use this link group for new connections */ 519 if (!list_empty(lgr_list)) 520 list_del_init(lgr_list); 521 spin_unlock_bh(lgr_lock); 522 } 523 524 static void smcd_unregister_all_dmbs(struct smc_link_group *lgr) 525 { 526 int i; 527 528 for (i = 0; i < SMC_RMBE_SIZES; i++) { 529 struct smc_buf_desc *buf_desc; 530 531 list_for_each_entry(buf_desc, &lgr->rmbs[i], list) { 532 buf_desc->len += sizeof(struct smcd_cdc_msg); 533 smc_ism_unregister_dmb(lgr->smcd, buf_desc); 534 } 535 } 536 } 537 538 static void smc_sk_wake_ups(struct smc_sock *smc) 539 { 540 smc->sk.sk_write_space(&smc->sk); 541 smc->sk.sk_data_ready(&smc->sk); 542 smc->sk.sk_state_change(&smc->sk); 543 } 544 545 /* kill a connection */ 546 static void smc_conn_kill(struct smc_connection *conn, bool soft) 547 { 548 struct smc_sock *smc = container_of(conn, struct smc_sock, conn); 549 550 if (conn->lgr->is_smcd && conn->lgr->peer_shutdown) 551 conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1; 552 else 553 smc_close_abort(conn); 554 conn->killed = 1; 555 smc->sk.sk_err = ECONNABORTED; 556 smc_sk_wake_ups(smc); 557 if (conn->lgr->is_smcd) { 558 smc_ism_unset_conn(conn); 559 if (soft) 560 tasklet_kill(&conn->rx_tsklet); 561 else 562 tasklet_unlock_wait(&conn->rx_tsklet); 563 } else { 564 smc_cdc_tx_dismiss_slots(conn); 565 } 566 smc_lgr_unregister_conn(conn); 567 smc_close_active_abort(smc); 568 } 569 570 static void smc_lgr_cleanup(struct smc_link_group *lgr) 571 { 572 if (lgr->is_smcd) { 573 smc_ism_signal_shutdown(lgr); 574 smcd_unregister_all_dmbs(lgr); 575 smc_ism_put_vlan(lgr->smcd, lgr->vlan_id); 576 put_device(&lgr->smcd->dev); 577 } else { 578 struct smc_link *lnk = &lgr->lnk[SMC_SINGLE_LINK]; 579 580 if (lnk->state != SMC_LNK_INACTIVE) 581 smc_llc_link_inactive(lnk); 582 } 583 } 584 585 /* terminate link group 586 * @soft: true if link group shutdown can take its time 587 * false if immediate link group shutdown is required 588 */ 589 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft) 590 { 591 struct smc_connection *conn; 592 struct smc_sock *smc; 593 struct rb_node *node; 594 595 if (lgr->terminating) 596 return; /* lgr already terminating */ 597 if (!soft) 598 cancel_delayed_work_sync(&lgr->free_work); 599 lgr->terminating = 1; 600 if (!lgr->is_smcd) 601 smc_llc_link_inactive(&lgr->lnk[SMC_SINGLE_LINK]); 602 603 /* kill remaining link group connections */ 604 read_lock_bh(&lgr->conns_lock); 605 node = rb_first(&lgr->conns_all); 606 while (node) { 607 read_unlock_bh(&lgr->conns_lock); 608 conn = rb_entry(node, struct smc_connection, alert_node); 609 smc = container_of(conn, struct smc_sock, conn); 610 sock_hold(&smc->sk); /* sock_put below */ 611 lock_sock(&smc->sk); 612 smc_conn_kill(conn, soft); 613 release_sock(&smc->sk); 614 sock_put(&smc->sk); /* sock_hold above */ 615 read_lock_bh(&lgr->conns_lock); 616 node = rb_first(&lgr->conns_all); 617 } 618 read_unlock_bh(&lgr->conns_lock); 619 smc_lgr_cleanup(lgr); 620 if (soft) 621 smc_lgr_schedule_free_work_fast(lgr); 622 else 623 smc_lgr_free(lgr); 624 } 625 626 /* unlink link group and schedule termination */ 627 void smc_lgr_terminate_sched(struct smc_link_group *lgr) 628 { 629 spinlock_t *lgr_lock; 630 631 smc_lgr_list_head(lgr, &lgr_lock); 632 spin_lock_bh(lgr_lock); 633 if (list_empty(&lgr->list) || lgr->terminating || lgr->freeing) { 634 spin_unlock_bh(lgr_lock); 635 return; /* lgr already terminating */ 636 } 637 list_del_init(&lgr->list); 638 spin_unlock_bh(lgr_lock); 639 schedule_work(&lgr->terminate_work); 640 } 641 642 /* Called when IB port is terminated */ 643 void smc_port_terminate(struct smc_ib_device *smcibdev, u8 ibport) 644 { 645 struct smc_link_group *lgr, *l; 646 LIST_HEAD(lgr_free_list); 647 648 spin_lock_bh(&smc_lgr_list.lock); 649 list_for_each_entry_safe(lgr, l, &smc_lgr_list.list, list) { 650 if (!lgr->is_smcd && 651 lgr->lnk[SMC_SINGLE_LINK].smcibdev == smcibdev && 652 lgr->lnk[SMC_SINGLE_LINK].ibport == ibport) { 653 list_move(&lgr->list, &lgr_free_list); 654 lgr->freeing = 1; 655 } 656 } 657 spin_unlock_bh(&smc_lgr_list.lock); 658 659 list_for_each_entry_safe(lgr, l, &lgr_free_list, list) { 660 list_del_init(&lgr->list); 661 __smc_lgr_terminate(lgr, false); 662 } 663 } 664 665 /* Called when peer lgr shutdown (regularly or abnormally) is received */ 666 void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid, unsigned short vlan) 667 { 668 struct smc_link_group *lgr, *l; 669 LIST_HEAD(lgr_free_list); 670 671 /* run common cleanup function and build free list */ 672 spin_lock_bh(&dev->lgr_lock); 673 list_for_each_entry_safe(lgr, l, &dev->lgr_list, list) { 674 if ((!peer_gid || lgr->peer_gid == peer_gid) && 675 (vlan == VLAN_VID_MASK || lgr->vlan_id == vlan)) { 676 if (peer_gid) /* peer triggered termination */ 677 lgr->peer_shutdown = 1; 678 list_move(&lgr->list, &lgr_free_list); 679 } 680 } 681 spin_unlock_bh(&dev->lgr_lock); 682 683 /* cancel the regular free workers and actually free lgrs */ 684 list_for_each_entry_safe(lgr, l, &lgr_free_list, list) { 685 list_del_init(&lgr->list); 686 schedule_work(&lgr->terminate_work); 687 } 688 } 689 690 /* Called when an SMCD device is removed or the smc module is unloaded */ 691 void smc_smcd_terminate_all(struct smcd_dev *smcd) 692 { 693 struct smc_link_group *lgr, *lg; 694 LIST_HEAD(lgr_free_list); 695 696 spin_lock_bh(&smcd->lgr_lock); 697 list_splice_init(&smcd->lgr_list, &lgr_free_list); 698 list_for_each_entry(lgr, &lgr_free_list, list) 699 lgr->freeing = 1; 700 spin_unlock_bh(&smcd->lgr_lock); 701 702 list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) { 703 list_del_init(&lgr->list); 704 __smc_lgr_terminate(lgr, false); 705 } 706 707 if (atomic_read(&smcd->lgr_cnt)) 708 wait_event(smcd->lgrs_deleted, !atomic_read(&smcd->lgr_cnt)); 709 } 710 711 /* Called when an SMCR device is removed or the smc module is unloaded. 712 * If smcibdev is given, all SMCR link groups using this device are terminated. 713 * If smcibdev is NULL, all SMCR link groups are terminated. 714 */ 715 void smc_smcr_terminate_all(struct smc_ib_device *smcibdev) 716 { 717 struct smc_link_group *lgr, *lg; 718 LIST_HEAD(lgr_free_list); 719 720 spin_lock_bh(&smc_lgr_list.lock); 721 if (!smcibdev) { 722 list_splice_init(&smc_lgr_list.list, &lgr_free_list); 723 list_for_each_entry(lgr, &lgr_free_list, list) 724 lgr->freeing = 1; 725 } else { 726 list_for_each_entry_safe(lgr, lg, &smc_lgr_list.list, list) { 727 if (lgr->lnk[SMC_SINGLE_LINK].smcibdev == smcibdev) { 728 list_move(&lgr->list, &lgr_free_list); 729 lgr->freeing = 1; 730 } 731 } 732 } 733 spin_unlock_bh(&smc_lgr_list.lock); 734 735 list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) { 736 list_del_init(&lgr->list); 737 __smc_lgr_terminate(lgr, false); 738 } 739 740 if (smcibdev) { 741 if (atomic_read(&smcibdev->lnk_cnt)) 742 wait_event(smcibdev->lnks_deleted, 743 !atomic_read(&smcibdev->lnk_cnt)); 744 } else { 745 if (atomic_read(&lgr_cnt)) 746 wait_event(lgrs_deleted, !atomic_read(&lgr_cnt)); 747 } 748 } 749 750 /* Determine vlan of internal TCP socket. 751 * @vlan_id: address to store the determined vlan id into 752 */ 753 int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini) 754 { 755 struct dst_entry *dst = sk_dst_get(clcsock->sk); 756 struct net_device *ndev; 757 int i, nest_lvl, rc = 0; 758 759 ini->vlan_id = 0; 760 if (!dst) { 761 rc = -ENOTCONN; 762 goto out; 763 } 764 if (!dst->dev) { 765 rc = -ENODEV; 766 goto out_rel; 767 } 768 769 ndev = dst->dev; 770 if (is_vlan_dev(ndev)) { 771 ini->vlan_id = vlan_dev_vlan_id(ndev); 772 goto out_rel; 773 } 774 775 rtnl_lock(); 776 nest_lvl = ndev->lower_level; 777 for (i = 0; i < nest_lvl; i++) { 778 struct list_head *lower = &ndev->adj_list.lower; 779 780 if (list_empty(lower)) 781 break; 782 lower = lower->next; 783 ndev = (struct net_device *)netdev_lower_get_next(ndev, &lower); 784 if (is_vlan_dev(ndev)) { 785 ini->vlan_id = vlan_dev_vlan_id(ndev); 786 break; 787 } 788 } 789 rtnl_unlock(); 790 791 out_rel: 792 dst_release(dst); 793 out: 794 return rc; 795 } 796 797 static bool smcr_lgr_match(struct smc_link_group *lgr, 798 struct smc_clc_msg_local *lcl, 799 enum smc_lgr_role role, u32 clcqpn) 800 { 801 return !memcmp(lgr->peer_systemid, lcl->id_for_peer, 802 SMC_SYSTEMID_LEN) && 803 !memcmp(lgr->lnk[SMC_SINGLE_LINK].peer_gid, &lcl->gid, 804 SMC_GID_SIZE) && 805 !memcmp(lgr->lnk[SMC_SINGLE_LINK].peer_mac, lcl->mac, 806 sizeof(lcl->mac)) && 807 lgr->role == role && 808 (lgr->role == SMC_SERV || 809 lgr->lnk[SMC_SINGLE_LINK].peer_qpn == clcqpn); 810 } 811 812 static bool smcd_lgr_match(struct smc_link_group *lgr, 813 struct smcd_dev *smcismdev, u64 peer_gid) 814 { 815 return lgr->peer_gid == peer_gid && lgr->smcd == smcismdev; 816 } 817 818 /* create a new SMC connection (and a new link group if necessary) */ 819 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini) 820 { 821 struct smc_connection *conn = &smc->conn; 822 struct list_head *lgr_list; 823 struct smc_link_group *lgr; 824 enum smc_lgr_role role; 825 spinlock_t *lgr_lock; 826 int rc = 0; 827 828 lgr_list = ini->is_smcd ? &ini->ism_dev->lgr_list : &smc_lgr_list.list; 829 lgr_lock = ini->is_smcd ? &ini->ism_dev->lgr_lock : &smc_lgr_list.lock; 830 ini->cln_first_contact = SMC_FIRST_CONTACT; 831 role = smc->listen_smc ? SMC_SERV : SMC_CLNT; 832 if (role == SMC_CLNT && ini->srv_first_contact) 833 /* create new link group as well */ 834 goto create; 835 836 /* determine if an existing link group can be reused */ 837 spin_lock_bh(lgr_lock); 838 list_for_each_entry(lgr, lgr_list, list) { 839 write_lock_bh(&lgr->conns_lock); 840 if ((ini->is_smcd ? 841 smcd_lgr_match(lgr, ini->ism_dev, ini->ism_gid) : 842 smcr_lgr_match(lgr, ini->ib_lcl, role, ini->ib_clcqpn)) && 843 !lgr->sync_err && 844 lgr->vlan_id == ini->vlan_id && 845 (role == SMC_CLNT || 846 lgr->conns_num < SMC_RMBS_PER_LGR_MAX)) { 847 /* link group found */ 848 ini->cln_first_contact = SMC_REUSE_CONTACT; 849 conn->lgr = lgr; 850 smc_lgr_register_conn(conn); /* add smc conn to lgr */ 851 if (delayed_work_pending(&lgr->free_work)) 852 cancel_delayed_work(&lgr->free_work); 853 write_unlock_bh(&lgr->conns_lock); 854 break; 855 } 856 write_unlock_bh(&lgr->conns_lock); 857 } 858 spin_unlock_bh(lgr_lock); 859 860 if (role == SMC_CLNT && !ini->srv_first_contact && 861 ini->cln_first_contact == SMC_FIRST_CONTACT) { 862 /* Server reuses a link group, but Client wants to start 863 * a new one 864 * send out_of_sync decline, reason synchr. error 865 */ 866 return SMC_CLC_DECL_SYNCERR; 867 } 868 869 create: 870 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 871 rc = smc_lgr_create(smc, ini); 872 if (rc) 873 goto out; 874 lgr = conn->lgr; 875 write_lock_bh(&lgr->conns_lock); 876 smc_lgr_register_conn(conn); /* add smc conn to lgr */ 877 write_unlock_bh(&lgr->conns_lock); 878 } 879 conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE; 880 conn->local_tx_ctrl.len = SMC_WR_TX_SIZE; 881 conn->urg_state = SMC_URG_READ; 882 if (ini->is_smcd) { 883 conn->rx_off = sizeof(struct smcd_cdc_msg); 884 smcd_cdc_rx_init(conn); /* init tasklet for this conn */ 885 } 886 #ifndef KERNEL_HAS_ATOMIC64 887 spin_lock_init(&conn->acurs_lock); 888 #endif 889 890 out: 891 return rc; 892 } 893 894 /* convert the RMB size into the compressed notation - minimum 16K. 895 * In contrast to plain ilog2, this rounds towards the next power of 2, 896 * so the socket application gets at least its desired sndbuf / rcvbuf size. 897 */ 898 static u8 smc_compress_bufsize(int size) 899 { 900 u8 compressed; 901 902 if (size <= SMC_BUF_MIN_SIZE) 903 return 0; 904 905 size = (size - 1) >> 14; 906 compressed = ilog2(size) + 1; 907 if (compressed >= SMC_RMBE_SIZES) 908 compressed = SMC_RMBE_SIZES - 1; 909 return compressed; 910 } 911 912 /* convert the RMB size from compressed notation into integer */ 913 int smc_uncompress_bufsize(u8 compressed) 914 { 915 u32 size; 916 917 size = 0x00000001 << (((int)compressed) + 14); 918 return (int)size; 919 } 920 921 /* try to reuse a sndbuf or rmb description slot for a certain 922 * buffer size; if not available, return NULL 923 */ 924 static struct smc_buf_desc *smc_buf_get_slot(int compressed_bufsize, 925 rwlock_t *lock, 926 struct list_head *buf_list) 927 { 928 struct smc_buf_desc *buf_slot; 929 930 read_lock_bh(lock); 931 list_for_each_entry(buf_slot, buf_list, list) { 932 if (cmpxchg(&buf_slot->used, 0, 1) == 0) { 933 read_unlock_bh(lock); 934 return buf_slot; 935 } 936 } 937 read_unlock_bh(lock); 938 return NULL; 939 } 940 941 /* one of the conditions for announcing a receiver's current window size is 942 * that it "results in a minimum increase in the window size of 10% of the 943 * receive buffer space" [RFC7609] 944 */ 945 static inline int smc_rmb_wnd_update_limit(int rmbe_size) 946 { 947 return min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2); 948 } 949 950 static struct smc_buf_desc *smcr_new_buf_create(struct smc_link_group *lgr, 951 bool is_rmb, int bufsize) 952 { 953 struct smc_buf_desc *buf_desc; 954 struct smc_link *lnk; 955 int rc; 956 957 /* try to alloc a new buffer */ 958 buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL); 959 if (!buf_desc) 960 return ERR_PTR(-ENOMEM); 961 962 buf_desc->order = get_order(bufsize); 963 buf_desc->pages = alloc_pages(GFP_KERNEL | __GFP_NOWARN | 964 __GFP_NOMEMALLOC | __GFP_COMP | 965 __GFP_NORETRY | __GFP_ZERO, 966 buf_desc->order); 967 if (!buf_desc->pages) { 968 kfree(buf_desc); 969 return ERR_PTR(-EAGAIN); 970 } 971 buf_desc->cpu_addr = (void *)page_address(buf_desc->pages); 972 973 /* build the sg table from the pages */ 974 lnk = &lgr->lnk[SMC_SINGLE_LINK]; 975 rc = sg_alloc_table(&buf_desc->sgt[SMC_SINGLE_LINK], 1, 976 GFP_KERNEL); 977 if (rc) { 978 smc_buf_free(lgr, is_rmb, buf_desc); 979 return ERR_PTR(rc); 980 } 981 sg_set_buf(buf_desc->sgt[SMC_SINGLE_LINK].sgl, 982 buf_desc->cpu_addr, bufsize); 983 984 /* map sg table to DMA address */ 985 rc = smc_ib_buf_map_sg(lnk->smcibdev, buf_desc, 986 is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE); 987 /* SMC protocol depends on mapping to one DMA address only */ 988 if (rc != 1) { 989 smc_buf_free(lgr, is_rmb, buf_desc); 990 return ERR_PTR(-EAGAIN); 991 } 992 993 /* create a new memory region for the RMB */ 994 if (is_rmb) { 995 rc = smc_ib_get_memory_region(lnk->roce_pd, 996 IB_ACCESS_REMOTE_WRITE | 997 IB_ACCESS_LOCAL_WRITE, 998 buf_desc); 999 if (rc) { 1000 smc_buf_free(lgr, is_rmb, buf_desc); 1001 return ERR_PTR(rc); 1002 } 1003 } 1004 1005 buf_desc->len = bufsize; 1006 return buf_desc; 1007 } 1008 1009 #define SMCD_DMBE_SIZES 7 /* 0 -> 16KB, 1 -> 32KB, .. 6 -> 1MB */ 1010 1011 static struct smc_buf_desc *smcd_new_buf_create(struct smc_link_group *lgr, 1012 bool is_dmb, int bufsize) 1013 { 1014 struct smc_buf_desc *buf_desc; 1015 int rc; 1016 1017 if (smc_compress_bufsize(bufsize) > SMCD_DMBE_SIZES) 1018 return ERR_PTR(-EAGAIN); 1019 1020 /* try to alloc a new DMB */ 1021 buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL); 1022 if (!buf_desc) 1023 return ERR_PTR(-ENOMEM); 1024 if (is_dmb) { 1025 rc = smc_ism_register_dmb(lgr, bufsize, buf_desc); 1026 if (rc) { 1027 kfree(buf_desc); 1028 return ERR_PTR(-EAGAIN); 1029 } 1030 buf_desc->pages = virt_to_page(buf_desc->cpu_addr); 1031 /* CDC header stored in buf. So, pretend it was smaller */ 1032 buf_desc->len = bufsize - sizeof(struct smcd_cdc_msg); 1033 } else { 1034 buf_desc->cpu_addr = kzalloc(bufsize, GFP_KERNEL | 1035 __GFP_NOWARN | __GFP_NORETRY | 1036 __GFP_NOMEMALLOC); 1037 if (!buf_desc->cpu_addr) { 1038 kfree(buf_desc); 1039 return ERR_PTR(-EAGAIN); 1040 } 1041 buf_desc->len = bufsize; 1042 } 1043 return buf_desc; 1044 } 1045 1046 static int __smc_buf_create(struct smc_sock *smc, bool is_smcd, bool is_rmb) 1047 { 1048 struct smc_buf_desc *buf_desc = ERR_PTR(-ENOMEM); 1049 struct smc_connection *conn = &smc->conn; 1050 struct smc_link_group *lgr = conn->lgr; 1051 struct list_head *buf_list; 1052 int bufsize, bufsize_short; 1053 int sk_buf_size; 1054 rwlock_t *lock; 1055 1056 if (is_rmb) 1057 /* use socket recv buffer size (w/o overhead) as start value */ 1058 sk_buf_size = smc->sk.sk_rcvbuf / 2; 1059 else 1060 /* use socket send buffer size (w/o overhead) as start value */ 1061 sk_buf_size = smc->sk.sk_sndbuf / 2; 1062 1063 for (bufsize_short = smc_compress_bufsize(sk_buf_size); 1064 bufsize_short >= 0; bufsize_short--) { 1065 1066 if (is_rmb) { 1067 lock = &lgr->rmbs_lock; 1068 buf_list = &lgr->rmbs[bufsize_short]; 1069 } else { 1070 lock = &lgr->sndbufs_lock; 1071 buf_list = &lgr->sndbufs[bufsize_short]; 1072 } 1073 bufsize = smc_uncompress_bufsize(bufsize_short); 1074 if ((1 << get_order(bufsize)) > SG_MAX_SINGLE_ALLOC) 1075 continue; 1076 1077 /* check for reusable slot in the link group */ 1078 buf_desc = smc_buf_get_slot(bufsize_short, lock, buf_list); 1079 if (buf_desc) { 1080 memset(buf_desc->cpu_addr, 0, bufsize); 1081 break; /* found reusable slot */ 1082 } 1083 1084 if (is_smcd) 1085 buf_desc = smcd_new_buf_create(lgr, is_rmb, bufsize); 1086 else 1087 buf_desc = smcr_new_buf_create(lgr, is_rmb, bufsize); 1088 1089 if (PTR_ERR(buf_desc) == -ENOMEM) 1090 break; 1091 if (IS_ERR(buf_desc)) 1092 continue; 1093 1094 buf_desc->used = 1; 1095 write_lock_bh(lock); 1096 list_add(&buf_desc->list, buf_list); 1097 write_unlock_bh(lock); 1098 break; /* found */ 1099 } 1100 1101 if (IS_ERR(buf_desc)) 1102 return -ENOMEM; 1103 1104 if (is_rmb) { 1105 conn->rmb_desc = buf_desc; 1106 conn->rmbe_size_short = bufsize_short; 1107 smc->sk.sk_rcvbuf = bufsize * 2; 1108 atomic_set(&conn->bytes_to_rcv, 0); 1109 conn->rmbe_update_limit = 1110 smc_rmb_wnd_update_limit(buf_desc->len); 1111 if (is_smcd) 1112 smc_ism_set_conn(conn); /* map RMB/smcd_dev to conn */ 1113 } else { 1114 conn->sndbuf_desc = buf_desc; 1115 smc->sk.sk_sndbuf = bufsize * 2; 1116 atomic_set(&conn->sndbuf_space, bufsize); 1117 } 1118 return 0; 1119 } 1120 1121 void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn) 1122 { 1123 struct smc_link_group *lgr = conn->lgr; 1124 1125 if (!conn->lgr || conn->lgr->is_smcd) 1126 return; 1127 smc_ib_sync_sg_for_cpu(lgr->lnk[SMC_SINGLE_LINK].smcibdev, 1128 conn->sndbuf_desc, DMA_TO_DEVICE); 1129 } 1130 1131 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn) 1132 { 1133 struct smc_link_group *lgr = conn->lgr; 1134 1135 if (!conn->lgr || conn->lgr->is_smcd) 1136 return; 1137 smc_ib_sync_sg_for_device(lgr->lnk[SMC_SINGLE_LINK].smcibdev, 1138 conn->sndbuf_desc, DMA_TO_DEVICE); 1139 } 1140 1141 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn) 1142 { 1143 struct smc_link_group *lgr = conn->lgr; 1144 1145 if (!conn->lgr || conn->lgr->is_smcd) 1146 return; 1147 smc_ib_sync_sg_for_cpu(lgr->lnk[SMC_SINGLE_LINK].smcibdev, 1148 conn->rmb_desc, DMA_FROM_DEVICE); 1149 } 1150 1151 void smc_rmb_sync_sg_for_device(struct smc_connection *conn) 1152 { 1153 struct smc_link_group *lgr = conn->lgr; 1154 1155 if (!conn->lgr || conn->lgr->is_smcd) 1156 return; 1157 smc_ib_sync_sg_for_device(lgr->lnk[SMC_SINGLE_LINK].smcibdev, 1158 conn->rmb_desc, DMA_FROM_DEVICE); 1159 } 1160 1161 /* create the send and receive buffer for an SMC socket; 1162 * receive buffers are called RMBs; 1163 * (even though the SMC protocol allows more than one RMB-element per RMB, 1164 * the Linux implementation uses just one RMB-element per RMB, i.e. uses an 1165 * extra RMB for every connection in a link group 1166 */ 1167 int smc_buf_create(struct smc_sock *smc, bool is_smcd) 1168 { 1169 int rc; 1170 1171 /* create send buffer */ 1172 rc = __smc_buf_create(smc, is_smcd, false); 1173 if (rc) 1174 return rc; 1175 /* create rmb */ 1176 rc = __smc_buf_create(smc, is_smcd, true); 1177 if (rc) 1178 smc_buf_free(smc->conn.lgr, false, smc->conn.sndbuf_desc); 1179 return rc; 1180 } 1181 1182 static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr) 1183 { 1184 int i; 1185 1186 for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) { 1187 if (!test_and_set_bit(i, lgr->rtokens_used_mask)) 1188 return i; 1189 } 1190 return -ENOSPC; 1191 } 1192 1193 /* add a new rtoken from peer */ 1194 int smc_rtoken_add(struct smc_link_group *lgr, __be64 nw_vaddr, __be32 nw_rkey) 1195 { 1196 u64 dma_addr = be64_to_cpu(nw_vaddr); 1197 u32 rkey = ntohl(nw_rkey); 1198 int i; 1199 1200 for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) { 1201 if ((lgr->rtokens[i][SMC_SINGLE_LINK].rkey == rkey) && 1202 (lgr->rtokens[i][SMC_SINGLE_LINK].dma_addr == dma_addr) && 1203 test_bit(i, lgr->rtokens_used_mask)) { 1204 /* already in list */ 1205 return i; 1206 } 1207 } 1208 i = smc_rmb_reserve_rtoken_idx(lgr); 1209 if (i < 0) 1210 return i; 1211 lgr->rtokens[i][SMC_SINGLE_LINK].rkey = rkey; 1212 lgr->rtokens[i][SMC_SINGLE_LINK].dma_addr = dma_addr; 1213 return i; 1214 } 1215 1216 /* delete an rtoken */ 1217 int smc_rtoken_delete(struct smc_link_group *lgr, __be32 nw_rkey) 1218 { 1219 u32 rkey = ntohl(nw_rkey); 1220 int i; 1221 1222 for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) { 1223 if (lgr->rtokens[i][SMC_SINGLE_LINK].rkey == rkey && 1224 test_bit(i, lgr->rtokens_used_mask)) { 1225 lgr->rtokens[i][SMC_SINGLE_LINK].rkey = 0; 1226 lgr->rtokens[i][SMC_SINGLE_LINK].dma_addr = 0; 1227 1228 clear_bit(i, lgr->rtokens_used_mask); 1229 return 0; 1230 } 1231 } 1232 return -ENOENT; 1233 } 1234 1235 /* save rkey and dma_addr received from peer during clc handshake */ 1236 int smc_rmb_rtoken_handling(struct smc_connection *conn, 1237 struct smc_clc_msg_accept_confirm *clc) 1238 { 1239 conn->rtoken_idx = smc_rtoken_add(conn->lgr, clc->rmb_dma_addr, 1240 clc->rmb_rkey); 1241 if (conn->rtoken_idx < 0) 1242 return conn->rtoken_idx; 1243 return 0; 1244 } 1245 1246 static void smc_core_going_away(void) 1247 { 1248 struct smc_ib_device *smcibdev; 1249 struct smcd_dev *smcd; 1250 1251 spin_lock(&smc_ib_devices.lock); 1252 list_for_each_entry(smcibdev, &smc_ib_devices.list, list) { 1253 int i; 1254 1255 for (i = 0; i < SMC_MAX_PORTS; i++) 1256 set_bit(i, smcibdev->ports_going_away); 1257 } 1258 spin_unlock(&smc_ib_devices.lock); 1259 1260 spin_lock(&smcd_dev_list.lock); 1261 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1262 smcd->going_away = 1; 1263 } 1264 spin_unlock(&smcd_dev_list.lock); 1265 } 1266 1267 /* Clean up all SMC link groups */ 1268 static void smc_lgrs_shutdown(void) 1269 { 1270 struct smcd_dev *smcd; 1271 1272 smc_core_going_away(); 1273 1274 smc_smcr_terminate_all(NULL); 1275 1276 spin_lock(&smcd_dev_list.lock); 1277 list_for_each_entry(smcd, &smcd_dev_list.list, list) 1278 smc_smcd_terminate_all(smcd); 1279 spin_unlock(&smcd_dev_list.lock); 1280 } 1281 1282 static int smc_core_reboot_event(struct notifier_block *this, 1283 unsigned long event, void *ptr) 1284 { 1285 smc_lgrs_shutdown(); 1286 smc_ib_unregister_client(); 1287 return 0; 1288 } 1289 1290 static struct notifier_block smc_reboot_notifier = { 1291 .notifier_call = smc_core_reboot_event, 1292 }; 1293 1294 int __init smc_core_init(void) 1295 { 1296 return register_reboot_notifier(&smc_reboot_notifier); 1297 } 1298 1299 /* Called (from smc_exit) when module is removed */ 1300 void smc_core_exit(void) 1301 { 1302 unregister_reboot_notifier(&smc_reboot_notifier); 1303 smc_lgrs_shutdown(); 1304 } 1305