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 <linux/mutex.h> 19 #include <linux/list.h> 20 #include <linux/smc.h> 21 #include <net/tcp.h> 22 #include <net/sock.h> 23 #include <rdma/ib_verbs.h> 24 #include <rdma/ib_cache.h> 25 26 #include "smc.h" 27 #include "smc_clc.h" 28 #include "smc_core.h" 29 #include "smc_ib.h" 30 #include "smc_wr.h" 31 #include "smc_llc.h" 32 #include "smc_cdc.h" 33 #include "smc_close.h" 34 #include "smc_ism.h" 35 #include "smc_netlink.h" 36 #include "smc_stats.h" 37 #include "smc_tracepoint.h" 38 39 #define SMC_LGR_NUM_INCR 256 40 #define SMC_LGR_FREE_DELAY_SERV (600 * HZ) 41 #define SMC_LGR_FREE_DELAY_CLNT (SMC_LGR_FREE_DELAY_SERV + 10 * HZ) 42 43 struct smc_lgr_list smc_lgr_list = { /* established link groups */ 44 .lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock), 45 .list = LIST_HEAD_INIT(smc_lgr_list.list), 46 .num = 0, 47 }; 48 49 static atomic_t lgr_cnt = ATOMIC_INIT(0); /* number of existing link groups */ 50 static DECLARE_WAIT_QUEUE_HEAD(lgrs_deleted); 51 52 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb, 53 struct smc_buf_desc *buf_desc); 54 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft); 55 56 static void smc_link_down_work(struct work_struct *work); 57 58 /* return head of link group list and its lock for a given link group */ 59 static inline struct list_head *smc_lgr_list_head(struct smc_link_group *lgr, 60 spinlock_t **lgr_lock) 61 { 62 if (lgr->is_smcd) { 63 *lgr_lock = &lgr->smcd->lgr_lock; 64 return &lgr->smcd->lgr_list; 65 } 66 67 *lgr_lock = &smc_lgr_list.lock; 68 return &smc_lgr_list.list; 69 } 70 71 static void smc_ibdev_cnt_inc(struct smc_link *lnk) 72 { 73 atomic_inc(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]); 74 } 75 76 static void smc_ibdev_cnt_dec(struct smc_link *lnk) 77 { 78 atomic_dec(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]); 79 } 80 81 static void smc_lgr_schedule_free_work(struct smc_link_group *lgr) 82 { 83 /* client link group creation always follows the server link group 84 * creation. For client use a somewhat higher removal delay time, 85 * otherwise there is a risk of out-of-sync link groups. 86 */ 87 if (!lgr->freeing) { 88 mod_delayed_work(system_wq, &lgr->free_work, 89 (!lgr->is_smcd && lgr->role == SMC_CLNT) ? 90 SMC_LGR_FREE_DELAY_CLNT : 91 SMC_LGR_FREE_DELAY_SERV); 92 } 93 } 94 95 /* Register connection's alert token in our lookup structure. 96 * To use rbtrees we have to implement our own insert core. 97 * Requires @conns_lock 98 * @smc connection to register 99 * Returns 0 on success, != otherwise. 100 */ 101 static void smc_lgr_add_alert_token(struct smc_connection *conn) 102 { 103 struct rb_node **link, *parent = NULL; 104 u32 token = conn->alert_token_local; 105 106 link = &conn->lgr->conns_all.rb_node; 107 while (*link) { 108 struct smc_connection *cur = rb_entry(*link, 109 struct smc_connection, alert_node); 110 111 parent = *link; 112 if (cur->alert_token_local > token) 113 link = &parent->rb_left; 114 else 115 link = &parent->rb_right; 116 } 117 /* Put the new node there */ 118 rb_link_node(&conn->alert_node, parent, link); 119 rb_insert_color(&conn->alert_node, &conn->lgr->conns_all); 120 } 121 122 /* assign an SMC-R link to the connection */ 123 static int smcr_lgr_conn_assign_link(struct smc_connection *conn, bool first) 124 { 125 enum smc_link_state expected = first ? SMC_LNK_ACTIVATING : 126 SMC_LNK_ACTIVE; 127 int i, j; 128 129 /* do link balancing */ 130 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 131 struct smc_link *lnk = &conn->lgr->lnk[i]; 132 133 if (lnk->state != expected || lnk->link_is_asym) 134 continue; 135 if (conn->lgr->role == SMC_CLNT) { 136 conn->lnk = lnk; /* temporary, SMC server assigns link*/ 137 break; 138 } 139 if (conn->lgr->conns_num % 2) { 140 for (j = i + 1; j < SMC_LINKS_PER_LGR_MAX; j++) { 141 struct smc_link *lnk2; 142 143 lnk2 = &conn->lgr->lnk[j]; 144 if (lnk2->state == expected && 145 !lnk2->link_is_asym) { 146 conn->lnk = lnk2; 147 break; 148 } 149 } 150 } 151 if (!conn->lnk) 152 conn->lnk = lnk; 153 break; 154 } 155 if (!conn->lnk) 156 return SMC_CLC_DECL_NOACTLINK; 157 atomic_inc(&conn->lnk->conn_cnt); 158 return 0; 159 } 160 161 /* Register connection in link group by assigning an alert token 162 * registered in a search tree. 163 * Requires @conns_lock 164 * Note that '0' is a reserved value and not assigned. 165 */ 166 static int smc_lgr_register_conn(struct smc_connection *conn, bool first) 167 { 168 struct smc_sock *smc = container_of(conn, struct smc_sock, conn); 169 static atomic_t nexttoken = ATOMIC_INIT(0); 170 int rc; 171 172 if (!conn->lgr->is_smcd) { 173 rc = smcr_lgr_conn_assign_link(conn, first); 174 if (rc) { 175 conn->lgr = NULL; 176 return rc; 177 } 178 } 179 /* find a new alert_token_local value not yet used by some connection 180 * in this link group 181 */ 182 sock_hold(&smc->sk); /* sock_put in smc_lgr_unregister_conn() */ 183 while (!conn->alert_token_local) { 184 conn->alert_token_local = atomic_inc_return(&nexttoken); 185 if (smc_lgr_find_conn(conn->alert_token_local, conn->lgr)) 186 conn->alert_token_local = 0; 187 } 188 smc_lgr_add_alert_token(conn); 189 conn->lgr->conns_num++; 190 return 0; 191 } 192 193 /* Unregister connection and reset the alert token of the given connection< 194 */ 195 static void __smc_lgr_unregister_conn(struct smc_connection *conn) 196 { 197 struct smc_sock *smc = container_of(conn, struct smc_sock, conn); 198 struct smc_link_group *lgr = conn->lgr; 199 200 rb_erase(&conn->alert_node, &lgr->conns_all); 201 if (conn->lnk) 202 atomic_dec(&conn->lnk->conn_cnt); 203 lgr->conns_num--; 204 conn->alert_token_local = 0; 205 sock_put(&smc->sk); /* sock_hold in smc_lgr_register_conn() */ 206 } 207 208 /* Unregister connection from lgr 209 */ 210 static void smc_lgr_unregister_conn(struct smc_connection *conn) 211 { 212 struct smc_link_group *lgr = conn->lgr; 213 214 if (!lgr) 215 return; 216 write_lock_bh(&lgr->conns_lock); 217 if (conn->alert_token_local) { 218 __smc_lgr_unregister_conn(conn); 219 } 220 write_unlock_bh(&lgr->conns_lock); 221 conn->lgr = NULL; 222 } 223 224 int smc_nl_get_sys_info(struct sk_buff *skb, struct netlink_callback *cb) 225 { 226 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 227 char hostname[SMC_MAX_HOSTNAME_LEN + 1]; 228 char smc_seid[SMC_MAX_EID_LEN + 1]; 229 struct nlattr *attrs; 230 u8 *seid = NULL; 231 u8 *host = NULL; 232 void *nlh; 233 234 nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 235 &smc_gen_nl_family, NLM_F_MULTI, 236 SMC_NETLINK_GET_SYS_INFO); 237 if (!nlh) 238 goto errmsg; 239 if (cb_ctx->pos[0]) 240 goto errout; 241 attrs = nla_nest_start(skb, SMC_GEN_SYS_INFO); 242 if (!attrs) 243 goto errout; 244 if (nla_put_u8(skb, SMC_NLA_SYS_VER, SMC_V2)) 245 goto errattr; 246 if (nla_put_u8(skb, SMC_NLA_SYS_REL, SMC_RELEASE)) 247 goto errattr; 248 if (nla_put_u8(skb, SMC_NLA_SYS_IS_ISM_V2, smc_ism_is_v2_capable())) 249 goto errattr; 250 if (nla_put_u8(skb, SMC_NLA_SYS_IS_SMCR_V2, true)) 251 goto errattr; 252 smc_clc_get_hostname(&host); 253 if (host) { 254 memcpy(hostname, host, SMC_MAX_HOSTNAME_LEN); 255 hostname[SMC_MAX_HOSTNAME_LEN] = 0; 256 if (nla_put_string(skb, SMC_NLA_SYS_LOCAL_HOST, hostname)) 257 goto errattr; 258 } 259 if (smc_ism_is_v2_capable()) { 260 smc_ism_get_system_eid(&seid); 261 memcpy(smc_seid, seid, SMC_MAX_EID_LEN); 262 smc_seid[SMC_MAX_EID_LEN] = 0; 263 if (nla_put_string(skb, SMC_NLA_SYS_SEID, smc_seid)) 264 goto errattr; 265 } 266 nla_nest_end(skb, attrs); 267 genlmsg_end(skb, nlh); 268 cb_ctx->pos[0] = 1; 269 return skb->len; 270 271 errattr: 272 nla_nest_cancel(skb, attrs); 273 errout: 274 genlmsg_cancel(skb, nlh); 275 errmsg: 276 return skb->len; 277 } 278 279 /* Fill SMC_NLA_LGR_D_V2_COMMON/SMC_NLA_LGR_R_V2_COMMON nested attributes */ 280 static int smc_nl_fill_lgr_v2_common(struct smc_link_group *lgr, 281 struct sk_buff *skb, 282 struct netlink_callback *cb, 283 struct nlattr *v2_attrs) 284 { 285 char smc_host[SMC_MAX_HOSTNAME_LEN + 1]; 286 char smc_eid[SMC_MAX_EID_LEN + 1]; 287 288 if (nla_put_u8(skb, SMC_NLA_LGR_V2_VER, lgr->smc_version)) 289 goto errv2attr; 290 if (nla_put_u8(skb, SMC_NLA_LGR_V2_REL, lgr->peer_smc_release)) 291 goto errv2attr; 292 if (nla_put_u8(skb, SMC_NLA_LGR_V2_OS, lgr->peer_os)) 293 goto errv2attr; 294 memcpy(smc_host, lgr->peer_hostname, SMC_MAX_HOSTNAME_LEN); 295 smc_host[SMC_MAX_HOSTNAME_LEN] = 0; 296 if (nla_put_string(skb, SMC_NLA_LGR_V2_PEER_HOST, smc_host)) 297 goto errv2attr; 298 memcpy(smc_eid, lgr->negotiated_eid, SMC_MAX_EID_LEN); 299 smc_eid[SMC_MAX_EID_LEN] = 0; 300 if (nla_put_string(skb, SMC_NLA_LGR_V2_NEG_EID, smc_eid)) 301 goto errv2attr; 302 303 nla_nest_end(skb, v2_attrs); 304 return 0; 305 306 errv2attr: 307 nla_nest_cancel(skb, v2_attrs); 308 return -EMSGSIZE; 309 } 310 311 static int smc_nl_fill_smcr_lgr_v2(struct smc_link_group *lgr, 312 struct sk_buff *skb, 313 struct netlink_callback *cb) 314 { 315 struct nlattr *v2_attrs; 316 317 v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2); 318 if (!v2_attrs) 319 goto errattr; 320 if (nla_put_u8(skb, SMC_NLA_LGR_R_V2_DIRECT, !lgr->uses_gateway)) 321 goto errv2attr; 322 323 nla_nest_end(skb, v2_attrs); 324 return 0; 325 326 errv2attr: 327 nla_nest_cancel(skb, v2_attrs); 328 errattr: 329 return -EMSGSIZE; 330 } 331 332 static int smc_nl_fill_lgr(struct smc_link_group *lgr, 333 struct sk_buff *skb, 334 struct netlink_callback *cb) 335 { 336 char smc_target[SMC_MAX_PNETID_LEN + 1]; 337 struct nlattr *attrs, *v2_attrs; 338 339 attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCR); 340 if (!attrs) 341 goto errout; 342 343 if (nla_put_u32(skb, SMC_NLA_LGR_R_ID, *((u32 *)&lgr->id))) 344 goto errattr; 345 if (nla_put_u32(skb, SMC_NLA_LGR_R_CONNS_NUM, lgr->conns_num)) 346 goto errattr; 347 if (nla_put_u8(skb, SMC_NLA_LGR_R_ROLE, lgr->role)) 348 goto errattr; 349 if (nla_put_u8(skb, SMC_NLA_LGR_R_TYPE, lgr->type)) 350 goto errattr; 351 if (nla_put_u8(skb, SMC_NLA_LGR_R_VLAN_ID, lgr->vlan_id)) 352 goto errattr; 353 memcpy(smc_target, lgr->pnet_id, SMC_MAX_PNETID_LEN); 354 smc_target[SMC_MAX_PNETID_LEN] = 0; 355 if (nla_put_string(skb, SMC_NLA_LGR_R_PNETID, smc_target)) 356 goto errattr; 357 if (lgr->smc_version > SMC_V1) { 358 v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2_COMMON); 359 if (!v2_attrs) 360 goto errattr; 361 if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs)) 362 goto errattr; 363 if (smc_nl_fill_smcr_lgr_v2(lgr, skb, cb)) 364 goto errattr; 365 } 366 367 nla_nest_end(skb, attrs); 368 return 0; 369 errattr: 370 nla_nest_cancel(skb, attrs); 371 errout: 372 return -EMSGSIZE; 373 } 374 375 static int smc_nl_fill_lgr_link(struct smc_link_group *lgr, 376 struct smc_link *link, 377 struct sk_buff *skb, 378 struct netlink_callback *cb) 379 { 380 char smc_ibname[IB_DEVICE_NAME_MAX]; 381 u8 smc_gid_target[41]; 382 struct nlattr *attrs; 383 u32 link_uid = 0; 384 void *nlh; 385 386 nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 387 &smc_gen_nl_family, NLM_F_MULTI, 388 SMC_NETLINK_GET_LINK_SMCR); 389 if (!nlh) 390 goto errmsg; 391 392 attrs = nla_nest_start(skb, SMC_GEN_LINK_SMCR); 393 if (!attrs) 394 goto errout; 395 396 if (nla_put_u8(skb, SMC_NLA_LINK_ID, link->link_id)) 397 goto errattr; 398 if (nla_put_u32(skb, SMC_NLA_LINK_STATE, link->state)) 399 goto errattr; 400 if (nla_put_u32(skb, SMC_NLA_LINK_CONN_CNT, 401 atomic_read(&link->conn_cnt))) 402 goto errattr; 403 if (nla_put_u8(skb, SMC_NLA_LINK_IB_PORT, link->ibport)) 404 goto errattr; 405 if (nla_put_u32(skb, SMC_NLA_LINK_NET_DEV, link->ndev_ifidx)) 406 goto errattr; 407 snprintf(smc_ibname, sizeof(smc_ibname), "%s", link->ibname); 408 if (nla_put_string(skb, SMC_NLA_LINK_IB_DEV, smc_ibname)) 409 goto errattr; 410 memcpy(&link_uid, link->link_uid, sizeof(link_uid)); 411 if (nla_put_u32(skb, SMC_NLA_LINK_UID, link_uid)) 412 goto errattr; 413 memcpy(&link_uid, link->peer_link_uid, sizeof(link_uid)); 414 if (nla_put_u32(skb, SMC_NLA_LINK_PEER_UID, link_uid)) 415 goto errattr; 416 memset(smc_gid_target, 0, sizeof(smc_gid_target)); 417 smc_gid_be16_convert(smc_gid_target, link->gid); 418 if (nla_put_string(skb, SMC_NLA_LINK_GID, smc_gid_target)) 419 goto errattr; 420 memset(smc_gid_target, 0, sizeof(smc_gid_target)); 421 smc_gid_be16_convert(smc_gid_target, link->peer_gid); 422 if (nla_put_string(skb, SMC_NLA_LINK_PEER_GID, smc_gid_target)) 423 goto errattr; 424 425 nla_nest_end(skb, attrs); 426 genlmsg_end(skb, nlh); 427 return 0; 428 errattr: 429 nla_nest_cancel(skb, attrs); 430 errout: 431 genlmsg_cancel(skb, nlh); 432 errmsg: 433 return -EMSGSIZE; 434 } 435 436 static int smc_nl_handle_lgr(struct smc_link_group *lgr, 437 struct sk_buff *skb, 438 struct netlink_callback *cb, 439 bool list_links) 440 { 441 void *nlh; 442 int i; 443 444 nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 445 &smc_gen_nl_family, NLM_F_MULTI, 446 SMC_NETLINK_GET_LGR_SMCR); 447 if (!nlh) 448 goto errmsg; 449 if (smc_nl_fill_lgr(lgr, skb, cb)) 450 goto errout; 451 452 genlmsg_end(skb, nlh); 453 if (!list_links) 454 goto out; 455 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 456 if (!smc_link_usable(&lgr->lnk[i])) 457 continue; 458 if (smc_nl_fill_lgr_link(lgr, &lgr->lnk[i], skb, cb)) 459 goto errout; 460 } 461 out: 462 return 0; 463 464 errout: 465 genlmsg_cancel(skb, nlh); 466 errmsg: 467 return -EMSGSIZE; 468 } 469 470 static void smc_nl_fill_lgr_list(struct smc_lgr_list *smc_lgr, 471 struct sk_buff *skb, 472 struct netlink_callback *cb, 473 bool list_links) 474 { 475 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 476 struct smc_link_group *lgr; 477 int snum = cb_ctx->pos[0]; 478 int num = 0; 479 480 spin_lock_bh(&smc_lgr->lock); 481 list_for_each_entry(lgr, &smc_lgr->list, list) { 482 if (num < snum) 483 goto next; 484 if (smc_nl_handle_lgr(lgr, skb, cb, list_links)) 485 goto errout; 486 next: 487 num++; 488 } 489 errout: 490 spin_unlock_bh(&smc_lgr->lock); 491 cb_ctx->pos[0] = num; 492 } 493 494 static int smc_nl_fill_smcd_lgr(struct smc_link_group *lgr, 495 struct sk_buff *skb, 496 struct netlink_callback *cb) 497 { 498 char smc_pnet[SMC_MAX_PNETID_LEN + 1]; 499 struct nlattr *attrs; 500 void *nlh; 501 502 nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 503 &smc_gen_nl_family, NLM_F_MULTI, 504 SMC_NETLINK_GET_LGR_SMCD); 505 if (!nlh) 506 goto errmsg; 507 508 attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCD); 509 if (!attrs) 510 goto errout; 511 512 if (nla_put_u32(skb, SMC_NLA_LGR_D_ID, *((u32 *)&lgr->id))) 513 goto errattr; 514 if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_GID, lgr->smcd->local_gid, 515 SMC_NLA_LGR_D_PAD)) 516 goto errattr; 517 if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_PEER_GID, lgr->peer_gid, 518 SMC_NLA_LGR_D_PAD)) 519 goto errattr; 520 if (nla_put_u8(skb, SMC_NLA_LGR_D_VLAN_ID, lgr->vlan_id)) 521 goto errattr; 522 if (nla_put_u32(skb, SMC_NLA_LGR_D_CONNS_NUM, lgr->conns_num)) 523 goto errattr; 524 if (nla_put_u32(skb, SMC_NLA_LGR_D_CHID, smc_ism_get_chid(lgr->smcd))) 525 goto errattr; 526 memcpy(smc_pnet, lgr->smcd->pnetid, SMC_MAX_PNETID_LEN); 527 smc_pnet[SMC_MAX_PNETID_LEN] = 0; 528 if (nla_put_string(skb, SMC_NLA_LGR_D_PNETID, smc_pnet)) 529 goto errattr; 530 if (lgr->smc_version > SMC_V1) { 531 struct nlattr *v2_attrs; 532 533 v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_D_V2_COMMON); 534 if (!v2_attrs) 535 goto errattr; 536 if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs)) 537 goto errattr; 538 } 539 nla_nest_end(skb, attrs); 540 genlmsg_end(skb, nlh); 541 return 0; 542 543 errattr: 544 nla_nest_cancel(skb, attrs); 545 errout: 546 genlmsg_cancel(skb, nlh); 547 errmsg: 548 return -EMSGSIZE; 549 } 550 551 static int smc_nl_handle_smcd_lgr(struct smcd_dev *dev, 552 struct sk_buff *skb, 553 struct netlink_callback *cb) 554 { 555 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 556 struct smc_link_group *lgr; 557 int snum = cb_ctx->pos[1]; 558 int rc = 0, num = 0; 559 560 spin_lock_bh(&dev->lgr_lock); 561 list_for_each_entry(lgr, &dev->lgr_list, list) { 562 if (!lgr->is_smcd) 563 continue; 564 if (num < snum) 565 goto next; 566 rc = smc_nl_fill_smcd_lgr(lgr, skb, cb); 567 if (rc) 568 goto errout; 569 next: 570 num++; 571 } 572 errout: 573 spin_unlock_bh(&dev->lgr_lock); 574 cb_ctx->pos[1] = num; 575 return rc; 576 } 577 578 static int smc_nl_fill_smcd_dev(struct smcd_dev_list *dev_list, 579 struct sk_buff *skb, 580 struct netlink_callback *cb) 581 { 582 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 583 struct smcd_dev *smcd_dev; 584 int snum = cb_ctx->pos[0]; 585 int rc = 0, num = 0; 586 587 mutex_lock(&dev_list->mutex); 588 list_for_each_entry(smcd_dev, &dev_list->list, list) { 589 if (list_empty(&smcd_dev->lgr_list)) 590 continue; 591 if (num < snum) 592 goto next; 593 rc = smc_nl_handle_smcd_lgr(smcd_dev, skb, cb); 594 if (rc) 595 goto errout; 596 next: 597 num++; 598 } 599 errout: 600 mutex_unlock(&dev_list->mutex); 601 cb_ctx->pos[0] = num; 602 return rc; 603 } 604 605 int smcr_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb) 606 { 607 bool list_links = false; 608 609 smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links); 610 return skb->len; 611 } 612 613 int smcr_nl_get_link(struct sk_buff *skb, struct netlink_callback *cb) 614 { 615 bool list_links = true; 616 617 smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links); 618 return skb->len; 619 } 620 621 int smcd_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb) 622 { 623 smc_nl_fill_smcd_dev(&smcd_dev_list, skb, cb); 624 return skb->len; 625 } 626 627 void smc_lgr_cleanup_early(struct smc_link_group *lgr) 628 { 629 spinlock_t *lgr_lock; 630 631 if (!lgr) 632 return; 633 634 smc_lgr_list_head(lgr, &lgr_lock); 635 spin_lock_bh(lgr_lock); 636 /* do not use this link group for new connections */ 637 if (!list_empty(&lgr->list)) 638 list_del_init(&lgr->list); 639 spin_unlock_bh(lgr_lock); 640 __smc_lgr_terminate(lgr, true); 641 } 642 643 static void smcr_lgr_link_deactivate_all(struct smc_link_group *lgr) 644 { 645 int i; 646 647 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 648 struct smc_link *lnk = &lgr->lnk[i]; 649 650 if (smc_link_sendable(lnk)) 651 lnk->state = SMC_LNK_INACTIVE; 652 } 653 wake_up_all(&lgr->llc_msg_waiter); 654 wake_up_all(&lgr->llc_flow_waiter); 655 } 656 657 static void smc_lgr_free(struct smc_link_group *lgr); 658 659 static void smc_lgr_free_work(struct work_struct *work) 660 { 661 struct smc_link_group *lgr = container_of(to_delayed_work(work), 662 struct smc_link_group, 663 free_work); 664 spinlock_t *lgr_lock; 665 bool conns; 666 667 smc_lgr_list_head(lgr, &lgr_lock); 668 spin_lock_bh(lgr_lock); 669 if (lgr->freeing) { 670 spin_unlock_bh(lgr_lock); 671 return; 672 } 673 read_lock_bh(&lgr->conns_lock); 674 conns = RB_EMPTY_ROOT(&lgr->conns_all); 675 read_unlock_bh(&lgr->conns_lock); 676 if (!conns) { /* number of lgr connections is no longer zero */ 677 spin_unlock_bh(lgr_lock); 678 return; 679 } 680 list_del_init(&lgr->list); /* remove from smc_lgr_list */ 681 lgr->freeing = 1; /* this instance does the freeing, no new schedule */ 682 spin_unlock_bh(lgr_lock); 683 cancel_delayed_work(&lgr->free_work); 684 685 if (!lgr->is_smcd && !lgr->terminating) 686 smc_llc_send_link_delete_all(lgr, true, 687 SMC_LLC_DEL_PROG_INIT_TERM); 688 if (lgr->is_smcd && !lgr->terminating) 689 smc_ism_signal_shutdown(lgr); 690 if (!lgr->is_smcd) 691 smcr_lgr_link_deactivate_all(lgr); 692 smc_lgr_free(lgr); 693 } 694 695 static void smc_lgr_terminate_work(struct work_struct *work) 696 { 697 struct smc_link_group *lgr = container_of(work, struct smc_link_group, 698 terminate_work); 699 700 __smc_lgr_terminate(lgr, true); 701 } 702 703 /* return next unique link id for the lgr */ 704 static u8 smcr_next_link_id(struct smc_link_group *lgr) 705 { 706 u8 link_id; 707 int i; 708 709 while (1) { 710 again: 711 link_id = ++lgr->next_link_id; 712 if (!link_id) /* skip zero as link_id */ 713 link_id = ++lgr->next_link_id; 714 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 715 if (smc_link_usable(&lgr->lnk[i]) && 716 lgr->lnk[i].link_id == link_id) 717 goto again; 718 } 719 break; 720 } 721 return link_id; 722 } 723 724 static void smcr_copy_dev_info_to_link(struct smc_link *link) 725 { 726 struct smc_ib_device *smcibdev = link->smcibdev; 727 728 snprintf(link->ibname, sizeof(link->ibname), "%s", 729 smcibdev->ibdev->name); 730 link->ndev_ifidx = smcibdev->ndev_ifidx[link->ibport - 1]; 731 } 732 733 int smcr_link_init(struct smc_link_group *lgr, struct smc_link *lnk, 734 u8 link_idx, struct smc_init_info *ini) 735 { 736 struct smc_ib_device *smcibdev; 737 u8 rndvec[3]; 738 int rc; 739 740 if (lgr->smc_version == SMC_V2) { 741 lnk->smcibdev = ini->smcrv2.ib_dev_v2; 742 lnk->ibport = ini->smcrv2.ib_port_v2; 743 } else { 744 lnk->smcibdev = ini->ib_dev; 745 lnk->ibport = ini->ib_port; 746 } 747 get_device(&lnk->smcibdev->ibdev->dev); 748 atomic_inc(&lnk->smcibdev->lnk_cnt); 749 lnk->path_mtu = lnk->smcibdev->pattr[lnk->ibport - 1].active_mtu; 750 lnk->link_id = smcr_next_link_id(lgr); 751 lnk->lgr = lgr; 752 lnk->link_idx = link_idx; 753 smc_ibdev_cnt_inc(lnk); 754 smcr_copy_dev_info_to_link(lnk); 755 atomic_set(&lnk->conn_cnt, 0); 756 smc_llc_link_set_uid(lnk); 757 INIT_WORK(&lnk->link_down_wrk, smc_link_down_work); 758 if (!lnk->smcibdev->initialized) { 759 rc = (int)smc_ib_setup_per_ibdev(lnk->smcibdev); 760 if (rc) 761 goto out; 762 } 763 get_random_bytes(rndvec, sizeof(rndvec)); 764 lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) + 765 (rndvec[2] << 16); 766 rc = smc_ib_determine_gid(lnk->smcibdev, lnk->ibport, 767 ini->vlan_id, lnk->gid, &lnk->sgid_index, 768 lgr->smc_version == SMC_V2 ? 769 &ini->smcrv2 : NULL); 770 if (rc) 771 goto out; 772 rc = smc_llc_link_init(lnk); 773 if (rc) 774 goto out; 775 rc = smc_wr_alloc_link_mem(lnk); 776 if (rc) 777 goto clear_llc_lnk; 778 rc = smc_ib_create_protection_domain(lnk); 779 if (rc) 780 goto free_link_mem; 781 rc = smc_ib_create_queue_pair(lnk); 782 if (rc) 783 goto dealloc_pd; 784 rc = smc_wr_create_link(lnk); 785 if (rc) 786 goto destroy_qp; 787 lnk->state = SMC_LNK_ACTIVATING; 788 return 0; 789 790 destroy_qp: 791 smc_ib_destroy_queue_pair(lnk); 792 dealloc_pd: 793 smc_ib_dealloc_protection_domain(lnk); 794 free_link_mem: 795 smc_wr_free_link_mem(lnk); 796 clear_llc_lnk: 797 smc_llc_link_clear(lnk, false); 798 out: 799 smc_ibdev_cnt_dec(lnk); 800 put_device(&lnk->smcibdev->ibdev->dev); 801 smcibdev = lnk->smcibdev; 802 memset(lnk, 0, sizeof(struct smc_link)); 803 lnk->state = SMC_LNK_UNUSED; 804 if (!atomic_dec_return(&smcibdev->lnk_cnt)) 805 wake_up(&smcibdev->lnks_deleted); 806 return rc; 807 } 808 809 /* create a new SMC link group */ 810 static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini) 811 { 812 struct smc_link_group *lgr; 813 struct list_head *lgr_list; 814 struct smc_link *lnk; 815 spinlock_t *lgr_lock; 816 u8 link_idx; 817 int rc = 0; 818 int i; 819 820 if (ini->is_smcd && ini->vlan_id) { 821 if (smc_ism_get_vlan(ini->ism_dev[ini->ism_selected], 822 ini->vlan_id)) { 823 rc = SMC_CLC_DECL_ISMVLANERR; 824 goto out; 825 } 826 } 827 828 lgr = kzalloc(sizeof(*lgr), GFP_KERNEL); 829 if (!lgr) { 830 rc = SMC_CLC_DECL_MEM; 831 goto ism_put_vlan; 832 } 833 lgr->tx_wq = alloc_workqueue("smc_tx_wq-%*phN", 0, 0, 834 SMC_LGR_ID_SIZE, &lgr->id); 835 if (!lgr->tx_wq) { 836 rc = -ENOMEM; 837 goto free_lgr; 838 } 839 lgr->is_smcd = ini->is_smcd; 840 lgr->sync_err = 0; 841 lgr->terminating = 0; 842 lgr->freeing = 0; 843 lgr->vlan_id = ini->vlan_id; 844 mutex_init(&lgr->sndbufs_lock); 845 mutex_init(&lgr->rmbs_lock); 846 rwlock_init(&lgr->conns_lock); 847 for (i = 0; i < SMC_RMBE_SIZES; i++) { 848 INIT_LIST_HEAD(&lgr->sndbufs[i]); 849 INIT_LIST_HEAD(&lgr->rmbs[i]); 850 } 851 lgr->next_link_id = 0; 852 smc_lgr_list.num += SMC_LGR_NUM_INCR; 853 memcpy(&lgr->id, (u8 *)&smc_lgr_list.num, SMC_LGR_ID_SIZE); 854 INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work); 855 INIT_WORK(&lgr->terminate_work, smc_lgr_terminate_work); 856 lgr->conns_all = RB_ROOT; 857 if (ini->is_smcd) { 858 /* SMC-D specific settings */ 859 get_device(&ini->ism_dev[ini->ism_selected]->dev); 860 lgr->peer_gid = ini->ism_peer_gid[ini->ism_selected]; 861 lgr->smcd = ini->ism_dev[ini->ism_selected]; 862 lgr_list = &ini->ism_dev[ini->ism_selected]->lgr_list; 863 lgr_lock = &lgr->smcd->lgr_lock; 864 lgr->smc_version = ini->smcd_version; 865 lgr->peer_shutdown = 0; 866 atomic_inc(&ini->ism_dev[ini->ism_selected]->lgr_cnt); 867 } else { 868 /* SMC-R specific settings */ 869 struct smc_ib_device *ibdev; 870 int ibport; 871 872 lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT; 873 lgr->smc_version = ini->smcr_version; 874 memcpy(lgr->peer_systemid, ini->peer_systemid, 875 SMC_SYSTEMID_LEN); 876 if (lgr->smc_version == SMC_V2) { 877 ibdev = ini->smcrv2.ib_dev_v2; 878 ibport = ini->smcrv2.ib_port_v2; 879 lgr->saddr = ini->smcrv2.saddr; 880 lgr->uses_gateway = ini->smcrv2.uses_gateway; 881 memcpy(lgr->nexthop_mac, ini->smcrv2.nexthop_mac, 882 ETH_ALEN); 883 } else { 884 ibdev = ini->ib_dev; 885 ibport = ini->ib_port; 886 } 887 memcpy(lgr->pnet_id, ibdev->pnetid[ibport - 1], 888 SMC_MAX_PNETID_LEN); 889 if (smc_wr_alloc_lgr_mem(lgr)) 890 goto free_wq; 891 smc_llc_lgr_init(lgr, smc); 892 893 link_idx = SMC_SINGLE_LINK; 894 lnk = &lgr->lnk[link_idx]; 895 rc = smcr_link_init(lgr, lnk, link_idx, ini); 896 if (rc) { 897 smc_wr_free_lgr_mem(lgr); 898 goto free_wq; 899 } 900 lgr_list = &smc_lgr_list.list; 901 lgr_lock = &smc_lgr_list.lock; 902 atomic_inc(&lgr_cnt); 903 } 904 smc->conn.lgr = lgr; 905 spin_lock_bh(lgr_lock); 906 list_add_tail(&lgr->list, lgr_list); 907 spin_unlock_bh(lgr_lock); 908 return 0; 909 910 free_wq: 911 destroy_workqueue(lgr->tx_wq); 912 free_lgr: 913 kfree(lgr); 914 ism_put_vlan: 915 if (ini->is_smcd && ini->vlan_id) 916 smc_ism_put_vlan(ini->ism_dev[ini->ism_selected], ini->vlan_id); 917 out: 918 if (rc < 0) { 919 if (rc == -ENOMEM) 920 rc = SMC_CLC_DECL_MEM; 921 else 922 rc = SMC_CLC_DECL_INTERR; 923 } 924 return rc; 925 } 926 927 static int smc_write_space(struct smc_connection *conn) 928 { 929 int buffer_len = conn->peer_rmbe_size; 930 union smc_host_cursor prod; 931 union smc_host_cursor cons; 932 int space; 933 934 smc_curs_copy(&prod, &conn->local_tx_ctrl.prod, conn); 935 smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn); 936 /* determine rx_buf space */ 937 space = buffer_len - smc_curs_diff(buffer_len, &cons, &prod); 938 return space; 939 } 940 941 static int smc_switch_cursor(struct smc_sock *smc, struct smc_cdc_tx_pend *pend, 942 struct smc_wr_buf *wr_buf) 943 { 944 struct smc_connection *conn = &smc->conn; 945 union smc_host_cursor cons, fin; 946 int rc = 0; 947 int diff; 948 949 smc_curs_copy(&conn->tx_curs_sent, &conn->tx_curs_fin, conn); 950 smc_curs_copy(&fin, &conn->local_tx_ctrl_fin, conn); 951 /* set prod cursor to old state, enforce tx_rdma_writes() */ 952 smc_curs_copy(&conn->local_tx_ctrl.prod, &fin, conn); 953 smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn); 954 955 if (smc_curs_comp(conn->peer_rmbe_size, &cons, &fin) < 0) { 956 /* cons cursor advanced more than fin, and prod was set 957 * fin above, so now prod is smaller than cons. Fix that. 958 */ 959 diff = smc_curs_diff(conn->peer_rmbe_size, &fin, &cons); 960 smc_curs_add(conn->sndbuf_desc->len, 961 &conn->tx_curs_sent, diff); 962 smc_curs_add(conn->sndbuf_desc->len, 963 &conn->tx_curs_fin, diff); 964 965 smp_mb__before_atomic(); 966 atomic_add(diff, &conn->sndbuf_space); 967 smp_mb__after_atomic(); 968 969 smc_curs_add(conn->peer_rmbe_size, 970 &conn->local_tx_ctrl.prod, diff); 971 smc_curs_add(conn->peer_rmbe_size, 972 &conn->local_tx_ctrl_fin, diff); 973 } 974 /* recalculate, value is used by tx_rdma_writes() */ 975 atomic_set(&smc->conn.peer_rmbe_space, smc_write_space(conn)); 976 977 if (smc->sk.sk_state != SMC_INIT && 978 smc->sk.sk_state != SMC_CLOSED) { 979 rc = smcr_cdc_msg_send_validation(conn, pend, wr_buf); 980 if (!rc) { 981 queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work, 0); 982 smc->sk.sk_data_ready(&smc->sk); 983 } 984 } else { 985 smc_wr_tx_put_slot(conn->lnk, 986 (struct smc_wr_tx_pend_priv *)pend); 987 } 988 return rc; 989 } 990 991 void smc_switch_link_and_count(struct smc_connection *conn, 992 struct smc_link *to_lnk) 993 { 994 atomic_dec(&conn->lnk->conn_cnt); 995 conn->lnk = to_lnk; 996 atomic_inc(&conn->lnk->conn_cnt); 997 } 998 999 struct smc_link *smc_switch_conns(struct smc_link_group *lgr, 1000 struct smc_link *from_lnk, bool is_dev_err) 1001 { 1002 struct smc_link *to_lnk = NULL; 1003 struct smc_cdc_tx_pend *pend; 1004 struct smc_connection *conn; 1005 struct smc_wr_buf *wr_buf; 1006 struct smc_sock *smc; 1007 struct rb_node *node; 1008 int i, rc = 0; 1009 1010 /* link is inactive, wake up tx waiters */ 1011 smc_wr_wakeup_tx_wait(from_lnk); 1012 1013 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1014 if (!smc_link_active(&lgr->lnk[i]) || i == from_lnk->link_idx) 1015 continue; 1016 if (is_dev_err && from_lnk->smcibdev == lgr->lnk[i].smcibdev && 1017 from_lnk->ibport == lgr->lnk[i].ibport) { 1018 continue; 1019 } 1020 to_lnk = &lgr->lnk[i]; 1021 break; 1022 } 1023 if (!to_lnk || !smc_wr_tx_link_hold(to_lnk)) { 1024 smc_lgr_terminate_sched(lgr); 1025 return NULL; 1026 } 1027 again: 1028 read_lock_bh(&lgr->conns_lock); 1029 for (node = rb_first(&lgr->conns_all); node; node = rb_next(node)) { 1030 conn = rb_entry(node, struct smc_connection, alert_node); 1031 if (conn->lnk != from_lnk) 1032 continue; 1033 smc = container_of(conn, struct smc_sock, conn); 1034 /* conn->lnk not yet set in SMC_INIT state */ 1035 if (smc->sk.sk_state == SMC_INIT) 1036 continue; 1037 if (smc->sk.sk_state == SMC_CLOSED || 1038 smc->sk.sk_state == SMC_PEERCLOSEWAIT1 || 1039 smc->sk.sk_state == SMC_PEERCLOSEWAIT2 || 1040 smc->sk.sk_state == SMC_APPFINCLOSEWAIT || 1041 smc->sk.sk_state == SMC_APPCLOSEWAIT1 || 1042 smc->sk.sk_state == SMC_APPCLOSEWAIT2 || 1043 smc->sk.sk_state == SMC_PEERFINCLOSEWAIT || 1044 smc->sk.sk_state == SMC_PEERABORTWAIT || 1045 smc->sk.sk_state == SMC_PROCESSABORT) { 1046 spin_lock_bh(&conn->send_lock); 1047 smc_switch_link_and_count(conn, to_lnk); 1048 spin_unlock_bh(&conn->send_lock); 1049 continue; 1050 } 1051 sock_hold(&smc->sk); 1052 read_unlock_bh(&lgr->conns_lock); 1053 /* pre-fetch buffer outside of send_lock, might sleep */ 1054 rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend); 1055 if (rc) 1056 goto err_out; 1057 /* avoid race with smcr_tx_sndbuf_nonempty() */ 1058 spin_lock_bh(&conn->send_lock); 1059 smc_switch_link_and_count(conn, to_lnk); 1060 rc = smc_switch_cursor(smc, pend, wr_buf); 1061 spin_unlock_bh(&conn->send_lock); 1062 sock_put(&smc->sk); 1063 if (rc) 1064 goto err_out; 1065 goto again; 1066 } 1067 read_unlock_bh(&lgr->conns_lock); 1068 smc_wr_tx_link_put(to_lnk); 1069 return to_lnk; 1070 1071 err_out: 1072 smcr_link_down_cond_sched(to_lnk); 1073 smc_wr_tx_link_put(to_lnk); 1074 return NULL; 1075 } 1076 1077 static void smcr_buf_unuse(struct smc_buf_desc *rmb_desc, 1078 struct smc_link_group *lgr) 1079 { 1080 int rc; 1081 1082 if (rmb_desc->is_conf_rkey && !list_empty(&lgr->list)) { 1083 /* unregister rmb with peer */ 1084 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 1085 if (!rc) { 1086 /* protect against smc_llc_cli_rkey_exchange() */ 1087 mutex_lock(&lgr->llc_conf_mutex); 1088 smc_llc_do_delete_rkey(lgr, rmb_desc); 1089 rmb_desc->is_conf_rkey = false; 1090 mutex_unlock(&lgr->llc_conf_mutex); 1091 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 1092 } 1093 } 1094 1095 if (rmb_desc->is_reg_err) { 1096 /* buf registration failed, reuse not possible */ 1097 mutex_lock(&lgr->rmbs_lock); 1098 list_del(&rmb_desc->list); 1099 mutex_unlock(&lgr->rmbs_lock); 1100 1101 smc_buf_free(lgr, true, rmb_desc); 1102 } else { 1103 rmb_desc->used = 0; 1104 } 1105 } 1106 1107 static void smc_buf_unuse(struct smc_connection *conn, 1108 struct smc_link_group *lgr) 1109 { 1110 if (conn->sndbuf_desc) 1111 conn->sndbuf_desc->used = 0; 1112 if (conn->rmb_desc && lgr->is_smcd) 1113 conn->rmb_desc->used = 0; 1114 else if (conn->rmb_desc) 1115 smcr_buf_unuse(conn->rmb_desc, lgr); 1116 } 1117 1118 /* remove a finished connection from its link group */ 1119 void smc_conn_free(struct smc_connection *conn) 1120 { 1121 struct smc_link_group *lgr = conn->lgr; 1122 1123 if (!lgr) 1124 return; 1125 if (lgr->is_smcd) { 1126 if (!list_empty(&lgr->list)) 1127 smc_ism_unset_conn(conn); 1128 tasklet_kill(&conn->rx_tsklet); 1129 } else { 1130 smc_cdc_wait_pend_tx_wr(conn); 1131 if (current_work() != &conn->abort_work) 1132 cancel_work_sync(&conn->abort_work); 1133 } 1134 if (!list_empty(&lgr->list)) { 1135 smc_lgr_unregister_conn(conn); 1136 smc_buf_unuse(conn, lgr); /* allow buffer reuse */ 1137 } 1138 1139 if (!lgr->conns_num) 1140 smc_lgr_schedule_free_work(lgr); 1141 } 1142 1143 /* unregister a link from a buf_desc */ 1144 static void smcr_buf_unmap_link(struct smc_buf_desc *buf_desc, bool is_rmb, 1145 struct smc_link *lnk) 1146 { 1147 if (is_rmb) 1148 buf_desc->is_reg_mr[lnk->link_idx] = false; 1149 if (!buf_desc->is_map_ib[lnk->link_idx]) 1150 return; 1151 if (is_rmb) { 1152 if (buf_desc->mr_rx[lnk->link_idx]) { 1153 smc_ib_put_memory_region( 1154 buf_desc->mr_rx[lnk->link_idx]); 1155 buf_desc->mr_rx[lnk->link_idx] = NULL; 1156 } 1157 smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_FROM_DEVICE); 1158 } else { 1159 smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_TO_DEVICE); 1160 } 1161 sg_free_table(&buf_desc->sgt[lnk->link_idx]); 1162 buf_desc->is_map_ib[lnk->link_idx] = false; 1163 } 1164 1165 /* unmap all buffers of lgr for a deleted link */ 1166 static void smcr_buf_unmap_lgr(struct smc_link *lnk) 1167 { 1168 struct smc_link_group *lgr = lnk->lgr; 1169 struct smc_buf_desc *buf_desc, *bf; 1170 int i; 1171 1172 for (i = 0; i < SMC_RMBE_SIZES; i++) { 1173 mutex_lock(&lgr->rmbs_lock); 1174 list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list) 1175 smcr_buf_unmap_link(buf_desc, true, lnk); 1176 mutex_unlock(&lgr->rmbs_lock); 1177 mutex_lock(&lgr->sndbufs_lock); 1178 list_for_each_entry_safe(buf_desc, bf, &lgr->sndbufs[i], 1179 list) 1180 smcr_buf_unmap_link(buf_desc, false, lnk); 1181 mutex_unlock(&lgr->sndbufs_lock); 1182 } 1183 } 1184 1185 static void smcr_rtoken_clear_link(struct smc_link *lnk) 1186 { 1187 struct smc_link_group *lgr = lnk->lgr; 1188 int i; 1189 1190 for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) { 1191 lgr->rtokens[i][lnk->link_idx].rkey = 0; 1192 lgr->rtokens[i][lnk->link_idx].dma_addr = 0; 1193 } 1194 } 1195 1196 /* must be called under lgr->llc_conf_mutex lock */ 1197 void smcr_link_clear(struct smc_link *lnk, bool log) 1198 { 1199 struct smc_ib_device *smcibdev; 1200 1201 if (!lnk->lgr || lnk->state == SMC_LNK_UNUSED) 1202 return; 1203 lnk->peer_qpn = 0; 1204 smc_llc_link_clear(lnk, log); 1205 smcr_buf_unmap_lgr(lnk); 1206 smcr_rtoken_clear_link(lnk); 1207 smc_ib_modify_qp_error(lnk); 1208 smc_wr_free_link(lnk); 1209 smc_ib_destroy_queue_pair(lnk); 1210 smc_ib_dealloc_protection_domain(lnk); 1211 smc_wr_free_link_mem(lnk); 1212 smc_ibdev_cnt_dec(lnk); 1213 put_device(&lnk->smcibdev->ibdev->dev); 1214 smcibdev = lnk->smcibdev; 1215 memset(lnk, 0, sizeof(struct smc_link)); 1216 lnk->state = SMC_LNK_UNUSED; 1217 if (!atomic_dec_return(&smcibdev->lnk_cnt)) 1218 wake_up(&smcibdev->lnks_deleted); 1219 } 1220 1221 static void smcr_buf_free(struct smc_link_group *lgr, bool is_rmb, 1222 struct smc_buf_desc *buf_desc) 1223 { 1224 int i; 1225 1226 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) 1227 smcr_buf_unmap_link(buf_desc, is_rmb, &lgr->lnk[i]); 1228 1229 if (buf_desc->pages) 1230 __free_pages(buf_desc->pages, buf_desc->order); 1231 kfree(buf_desc); 1232 } 1233 1234 static void smcd_buf_free(struct smc_link_group *lgr, bool is_dmb, 1235 struct smc_buf_desc *buf_desc) 1236 { 1237 if (is_dmb) { 1238 /* restore original buf len */ 1239 buf_desc->len += sizeof(struct smcd_cdc_msg); 1240 smc_ism_unregister_dmb(lgr->smcd, buf_desc); 1241 } else { 1242 kfree(buf_desc->cpu_addr); 1243 } 1244 kfree(buf_desc); 1245 } 1246 1247 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb, 1248 struct smc_buf_desc *buf_desc) 1249 { 1250 if (lgr->is_smcd) 1251 smcd_buf_free(lgr, is_rmb, buf_desc); 1252 else 1253 smcr_buf_free(lgr, is_rmb, buf_desc); 1254 } 1255 1256 static void __smc_lgr_free_bufs(struct smc_link_group *lgr, bool is_rmb) 1257 { 1258 struct smc_buf_desc *buf_desc, *bf_desc; 1259 struct list_head *buf_list; 1260 int i; 1261 1262 for (i = 0; i < SMC_RMBE_SIZES; i++) { 1263 if (is_rmb) 1264 buf_list = &lgr->rmbs[i]; 1265 else 1266 buf_list = &lgr->sndbufs[i]; 1267 list_for_each_entry_safe(buf_desc, bf_desc, buf_list, 1268 list) { 1269 list_del(&buf_desc->list); 1270 smc_buf_free(lgr, is_rmb, buf_desc); 1271 } 1272 } 1273 } 1274 1275 static void smc_lgr_free_bufs(struct smc_link_group *lgr) 1276 { 1277 /* free send buffers */ 1278 __smc_lgr_free_bufs(lgr, false); 1279 /* free rmbs */ 1280 __smc_lgr_free_bufs(lgr, true); 1281 } 1282 1283 /* remove a link group */ 1284 static void smc_lgr_free(struct smc_link_group *lgr) 1285 { 1286 int i; 1287 1288 if (!lgr->is_smcd) { 1289 mutex_lock(&lgr->llc_conf_mutex); 1290 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1291 if (lgr->lnk[i].state != SMC_LNK_UNUSED) 1292 smcr_link_clear(&lgr->lnk[i], false); 1293 } 1294 mutex_unlock(&lgr->llc_conf_mutex); 1295 smc_llc_lgr_clear(lgr); 1296 } 1297 1298 smc_lgr_free_bufs(lgr); 1299 destroy_workqueue(lgr->tx_wq); 1300 if (lgr->is_smcd) { 1301 smc_ism_put_vlan(lgr->smcd, lgr->vlan_id); 1302 put_device(&lgr->smcd->dev); 1303 if (!atomic_dec_return(&lgr->smcd->lgr_cnt)) 1304 wake_up(&lgr->smcd->lgrs_deleted); 1305 } else { 1306 smc_wr_free_lgr_mem(lgr); 1307 if (!atomic_dec_return(&lgr_cnt)) 1308 wake_up(&lgrs_deleted); 1309 } 1310 kfree(lgr); 1311 } 1312 1313 static void smc_sk_wake_ups(struct smc_sock *smc) 1314 { 1315 smc->sk.sk_write_space(&smc->sk); 1316 smc->sk.sk_data_ready(&smc->sk); 1317 smc->sk.sk_state_change(&smc->sk); 1318 } 1319 1320 /* kill a connection */ 1321 static void smc_conn_kill(struct smc_connection *conn, bool soft) 1322 { 1323 struct smc_sock *smc = container_of(conn, struct smc_sock, conn); 1324 1325 if (conn->lgr->is_smcd && conn->lgr->peer_shutdown) 1326 conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1; 1327 else 1328 smc_close_abort(conn); 1329 conn->killed = 1; 1330 smc->sk.sk_err = ECONNABORTED; 1331 smc_sk_wake_ups(smc); 1332 if (conn->lgr->is_smcd) { 1333 smc_ism_unset_conn(conn); 1334 if (soft) 1335 tasklet_kill(&conn->rx_tsklet); 1336 else 1337 tasklet_unlock_wait(&conn->rx_tsklet); 1338 } else { 1339 smc_cdc_wait_pend_tx_wr(conn); 1340 } 1341 smc_lgr_unregister_conn(conn); 1342 smc_close_active_abort(smc); 1343 } 1344 1345 static void smc_lgr_cleanup(struct smc_link_group *lgr) 1346 { 1347 if (lgr->is_smcd) { 1348 smc_ism_signal_shutdown(lgr); 1349 } else { 1350 u32 rsn = lgr->llc_termination_rsn; 1351 1352 if (!rsn) 1353 rsn = SMC_LLC_DEL_PROG_INIT_TERM; 1354 smc_llc_send_link_delete_all(lgr, false, rsn); 1355 smcr_lgr_link_deactivate_all(lgr); 1356 } 1357 } 1358 1359 /* terminate link group 1360 * @soft: true if link group shutdown can take its time 1361 * false if immediate link group shutdown is required 1362 */ 1363 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft) 1364 { 1365 struct smc_connection *conn; 1366 struct smc_sock *smc; 1367 struct rb_node *node; 1368 1369 if (lgr->terminating) 1370 return; /* lgr already terminating */ 1371 /* cancel free_work sync, will terminate when lgr->freeing is set */ 1372 cancel_delayed_work_sync(&lgr->free_work); 1373 lgr->terminating = 1; 1374 1375 /* kill remaining link group connections */ 1376 read_lock_bh(&lgr->conns_lock); 1377 node = rb_first(&lgr->conns_all); 1378 while (node) { 1379 read_unlock_bh(&lgr->conns_lock); 1380 conn = rb_entry(node, struct smc_connection, alert_node); 1381 smc = container_of(conn, struct smc_sock, conn); 1382 sock_hold(&smc->sk); /* sock_put below */ 1383 lock_sock(&smc->sk); 1384 smc_conn_kill(conn, soft); 1385 release_sock(&smc->sk); 1386 sock_put(&smc->sk); /* sock_hold above */ 1387 read_lock_bh(&lgr->conns_lock); 1388 node = rb_first(&lgr->conns_all); 1389 } 1390 read_unlock_bh(&lgr->conns_lock); 1391 smc_lgr_cleanup(lgr); 1392 smc_lgr_free(lgr); 1393 } 1394 1395 /* unlink link group and schedule termination */ 1396 void smc_lgr_terminate_sched(struct smc_link_group *lgr) 1397 { 1398 spinlock_t *lgr_lock; 1399 1400 smc_lgr_list_head(lgr, &lgr_lock); 1401 spin_lock_bh(lgr_lock); 1402 if (list_empty(&lgr->list) || lgr->terminating || lgr->freeing) { 1403 spin_unlock_bh(lgr_lock); 1404 return; /* lgr already terminating */ 1405 } 1406 list_del_init(&lgr->list); 1407 lgr->freeing = 1; 1408 spin_unlock_bh(lgr_lock); 1409 schedule_work(&lgr->terminate_work); 1410 } 1411 1412 /* Called when peer lgr shutdown (regularly or abnormally) is received */ 1413 void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid, unsigned short vlan) 1414 { 1415 struct smc_link_group *lgr, *l; 1416 LIST_HEAD(lgr_free_list); 1417 1418 /* run common cleanup function and build free list */ 1419 spin_lock_bh(&dev->lgr_lock); 1420 list_for_each_entry_safe(lgr, l, &dev->lgr_list, list) { 1421 if ((!peer_gid || lgr->peer_gid == peer_gid) && 1422 (vlan == VLAN_VID_MASK || lgr->vlan_id == vlan)) { 1423 if (peer_gid) /* peer triggered termination */ 1424 lgr->peer_shutdown = 1; 1425 list_move(&lgr->list, &lgr_free_list); 1426 lgr->freeing = 1; 1427 } 1428 } 1429 spin_unlock_bh(&dev->lgr_lock); 1430 1431 /* cancel the regular free workers and actually free lgrs */ 1432 list_for_each_entry_safe(lgr, l, &lgr_free_list, list) { 1433 list_del_init(&lgr->list); 1434 schedule_work(&lgr->terminate_work); 1435 } 1436 } 1437 1438 /* Called when an SMCD device is removed or the smc module is unloaded */ 1439 void smc_smcd_terminate_all(struct smcd_dev *smcd) 1440 { 1441 struct smc_link_group *lgr, *lg; 1442 LIST_HEAD(lgr_free_list); 1443 1444 spin_lock_bh(&smcd->lgr_lock); 1445 list_splice_init(&smcd->lgr_list, &lgr_free_list); 1446 list_for_each_entry(lgr, &lgr_free_list, list) 1447 lgr->freeing = 1; 1448 spin_unlock_bh(&smcd->lgr_lock); 1449 1450 list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) { 1451 list_del_init(&lgr->list); 1452 __smc_lgr_terminate(lgr, false); 1453 } 1454 1455 if (atomic_read(&smcd->lgr_cnt)) 1456 wait_event(smcd->lgrs_deleted, !atomic_read(&smcd->lgr_cnt)); 1457 } 1458 1459 /* Called when an SMCR device is removed or the smc module is unloaded. 1460 * If smcibdev is given, all SMCR link groups using this device are terminated. 1461 * If smcibdev is NULL, all SMCR link groups are terminated. 1462 * 1463 * We must wait here for QPs been destroyed before we destroy the CQs, 1464 * or we won't received any CQEs and cdc_pend_tx_wr cannot reach 0 thus 1465 * smc_sock cannot be released. 1466 */ 1467 void smc_smcr_terminate_all(struct smc_ib_device *smcibdev) 1468 { 1469 struct smc_link_group *lgr, *lg; 1470 LIST_HEAD(lgr_free_list); 1471 LIST_HEAD(lgr_linkdown_list); 1472 int i; 1473 1474 spin_lock_bh(&smc_lgr_list.lock); 1475 if (!smcibdev) { 1476 list_splice_init(&smc_lgr_list.list, &lgr_free_list); 1477 list_for_each_entry(lgr, &lgr_free_list, list) 1478 lgr->freeing = 1; 1479 } else { 1480 list_for_each_entry_safe(lgr, lg, &smc_lgr_list.list, list) { 1481 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1482 if (lgr->lnk[i].smcibdev == smcibdev) 1483 list_move_tail(&lgr->list, &lgr_linkdown_list); 1484 } 1485 } 1486 } 1487 spin_unlock_bh(&smc_lgr_list.lock); 1488 1489 list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) { 1490 list_del_init(&lgr->list); 1491 smc_llc_set_termination_rsn(lgr, SMC_LLC_DEL_OP_INIT_TERM); 1492 __smc_lgr_terminate(lgr, false); 1493 } 1494 1495 list_for_each_entry_safe(lgr, lg, &lgr_linkdown_list, list) { 1496 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1497 if (lgr->lnk[i].smcibdev == smcibdev) { 1498 mutex_lock(&lgr->llc_conf_mutex); 1499 smcr_link_down_cond(&lgr->lnk[i]); 1500 mutex_unlock(&lgr->llc_conf_mutex); 1501 } 1502 } 1503 } 1504 1505 if (smcibdev) { 1506 if (atomic_read(&smcibdev->lnk_cnt)) 1507 wait_event(smcibdev->lnks_deleted, 1508 !atomic_read(&smcibdev->lnk_cnt)); 1509 } else { 1510 if (atomic_read(&lgr_cnt)) 1511 wait_event(lgrs_deleted, !atomic_read(&lgr_cnt)); 1512 } 1513 } 1514 1515 /* set new lgr type and clear all asymmetric link tagging */ 1516 void smcr_lgr_set_type(struct smc_link_group *lgr, enum smc_lgr_type new_type) 1517 { 1518 char *lgr_type = ""; 1519 int i; 1520 1521 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) 1522 if (smc_link_usable(&lgr->lnk[i])) 1523 lgr->lnk[i].link_is_asym = false; 1524 if (lgr->type == new_type) 1525 return; 1526 lgr->type = new_type; 1527 1528 switch (lgr->type) { 1529 case SMC_LGR_NONE: 1530 lgr_type = "NONE"; 1531 break; 1532 case SMC_LGR_SINGLE: 1533 lgr_type = "SINGLE"; 1534 break; 1535 case SMC_LGR_SYMMETRIC: 1536 lgr_type = "SYMMETRIC"; 1537 break; 1538 case SMC_LGR_ASYMMETRIC_PEER: 1539 lgr_type = "ASYMMETRIC_PEER"; 1540 break; 1541 case SMC_LGR_ASYMMETRIC_LOCAL: 1542 lgr_type = "ASYMMETRIC_LOCAL"; 1543 break; 1544 } 1545 pr_warn_ratelimited("smc: SMC-R lg %*phN state changed: " 1546 "%s, pnetid %.16s\n", SMC_LGR_ID_SIZE, &lgr->id, 1547 lgr_type, lgr->pnet_id); 1548 } 1549 1550 /* set new lgr type and tag a link as asymmetric */ 1551 void smcr_lgr_set_type_asym(struct smc_link_group *lgr, 1552 enum smc_lgr_type new_type, int asym_lnk_idx) 1553 { 1554 smcr_lgr_set_type(lgr, new_type); 1555 lgr->lnk[asym_lnk_idx].link_is_asym = true; 1556 } 1557 1558 /* abort connection, abort_work scheduled from tasklet context */ 1559 static void smc_conn_abort_work(struct work_struct *work) 1560 { 1561 struct smc_connection *conn = container_of(work, 1562 struct smc_connection, 1563 abort_work); 1564 struct smc_sock *smc = container_of(conn, struct smc_sock, conn); 1565 1566 lock_sock(&smc->sk); 1567 smc_conn_kill(conn, true); 1568 release_sock(&smc->sk); 1569 sock_put(&smc->sk); /* sock_hold done by schedulers of abort_work */ 1570 } 1571 1572 void smcr_port_add(struct smc_ib_device *smcibdev, u8 ibport) 1573 { 1574 struct smc_link_group *lgr, *n; 1575 1576 list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) { 1577 struct smc_link *link; 1578 1579 if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id, 1580 SMC_MAX_PNETID_LEN) || 1581 lgr->type == SMC_LGR_SYMMETRIC || 1582 lgr->type == SMC_LGR_ASYMMETRIC_PEER) 1583 continue; 1584 1585 /* trigger local add link processing */ 1586 link = smc_llc_usable_link(lgr); 1587 if (link) 1588 smc_llc_add_link_local(link); 1589 } 1590 } 1591 1592 /* link is down - switch connections to alternate link, 1593 * must be called under lgr->llc_conf_mutex lock 1594 */ 1595 static void smcr_link_down(struct smc_link *lnk) 1596 { 1597 struct smc_link_group *lgr = lnk->lgr; 1598 struct smc_link *to_lnk; 1599 int del_link_id; 1600 1601 if (!lgr || lnk->state == SMC_LNK_UNUSED || list_empty(&lgr->list)) 1602 return; 1603 1604 to_lnk = smc_switch_conns(lgr, lnk, true); 1605 if (!to_lnk) { /* no backup link available */ 1606 smcr_link_clear(lnk, true); 1607 return; 1608 } 1609 smcr_lgr_set_type(lgr, SMC_LGR_SINGLE); 1610 del_link_id = lnk->link_id; 1611 1612 if (lgr->role == SMC_SERV) { 1613 /* trigger local delete link processing */ 1614 smc_llc_srv_delete_link_local(to_lnk, del_link_id); 1615 } else { 1616 if (lgr->llc_flow_lcl.type != SMC_LLC_FLOW_NONE) { 1617 /* another llc task is ongoing */ 1618 mutex_unlock(&lgr->llc_conf_mutex); 1619 wait_event_timeout(lgr->llc_flow_waiter, 1620 (list_empty(&lgr->list) || 1621 lgr->llc_flow_lcl.type == SMC_LLC_FLOW_NONE), 1622 SMC_LLC_WAIT_TIME); 1623 mutex_lock(&lgr->llc_conf_mutex); 1624 } 1625 if (!list_empty(&lgr->list)) { 1626 smc_llc_send_delete_link(to_lnk, del_link_id, 1627 SMC_LLC_REQ, true, 1628 SMC_LLC_DEL_LOST_PATH); 1629 smcr_link_clear(lnk, true); 1630 } 1631 wake_up(&lgr->llc_flow_waiter); /* wake up next waiter */ 1632 } 1633 } 1634 1635 /* must be called under lgr->llc_conf_mutex lock */ 1636 void smcr_link_down_cond(struct smc_link *lnk) 1637 { 1638 if (smc_link_downing(&lnk->state)) { 1639 trace_smcr_link_down(lnk, __builtin_return_address(0)); 1640 smcr_link_down(lnk); 1641 } 1642 } 1643 1644 /* will get the lgr->llc_conf_mutex lock */ 1645 void smcr_link_down_cond_sched(struct smc_link *lnk) 1646 { 1647 if (smc_link_downing(&lnk->state)) { 1648 trace_smcr_link_down(lnk, __builtin_return_address(0)); 1649 schedule_work(&lnk->link_down_wrk); 1650 } 1651 } 1652 1653 void smcr_port_err(struct smc_ib_device *smcibdev, u8 ibport) 1654 { 1655 struct smc_link_group *lgr, *n; 1656 int i; 1657 1658 list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) { 1659 if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id, 1660 SMC_MAX_PNETID_LEN)) 1661 continue; /* lgr is not affected */ 1662 if (list_empty(&lgr->list)) 1663 continue; 1664 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1665 struct smc_link *lnk = &lgr->lnk[i]; 1666 1667 if (smc_link_usable(lnk) && 1668 lnk->smcibdev == smcibdev && lnk->ibport == ibport) 1669 smcr_link_down_cond_sched(lnk); 1670 } 1671 } 1672 } 1673 1674 static void smc_link_down_work(struct work_struct *work) 1675 { 1676 struct smc_link *link = container_of(work, struct smc_link, 1677 link_down_wrk); 1678 struct smc_link_group *lgr = link->lgr; 1679 1680 if (list_empty(&lgr->list)) 1681 return; 1682 wake_up_all(&lgr->llc_msg_waiter); 1683 mutex_lock(&lgr->llc_conf_mutex); 1684 smcr_link_down(link); 1685 mutex_unlock(&lgr->llc_conf_mutex); 1686 } 1687 1688 static int smc_vlan_by_tcpsk_walk(struct net_device *lower_dev, 1689 struct netdev_nested_priv *priv) 1690 { 1691 unsigned short *vlan_id = (unsigned short *)priv->data; 1692 1693 if (is_vlan_dev(lower_dev)) { 1694 *vlan_id = vlan_dev_vlan_id(lower_dev); 1695 return 1; 1696 } 1697 1698 return 0; 1699 } 1700 1701 /* Determine vlan of internal TCP socket. */ 1702 int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini) 1703 { 1704 struct dst_entry *dst = sk_dst_get(clcsock->sk); 1705 struct netdev_nested_priv priv; 1706 struct net_device *ndev; 1707 int rc = 0; 1708 1709 ini->vlan_id = 0; 1710 if (!dst) { 1711 rc = -ENOTCONN; 1712 goto out; 1713 } 1714 if (!dst->dev) { 1715 rc = -ENODEV; 1716 goto out_rel; 1717 } 1718 1719 ndev = dst->dev; 1720 if (is_vlan_dev(ndev)) { 1721 ini->vlan_id = vlan_dev_vlan_id(ndev); 1722 goto out_rel; 1723 } 1724 1725 priv.data = (void *)&ini->vlan_id; 1726 rtnl_lock(); 1727 netdev_walk_all_lower_dev(ndev, smc_vlan_by_tcpsk_walk, &priv); 1728 rtnl_unlock(); 1729 1730 out_rel: 1731 dst_release(dst); 1732 out: 1733 return rc; 1734 } 1735 1736 static bool smcr_lgr_match(struct smc_link_group *lgr, u8 smcr_version, 1737 u8 peer_systemid[], 1738 u8 peer_gid[], 1739 u8 peer_mac_v1[], 1740 enum smc_lgr_role role, u32 clcqpn) 1741 { 1742 int i; 1743 1744 if (memcmp(lgr->peer_systemid, peer_systemid, SMC_SYSTEMID_LEN) || 1745 lgr->role != role) 1746 return false; 1747 1748 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1749 if (!smc_link_active(&lgr->lnk[i])) 1750 continue; 1751 if ((lgr->role == SMC_SERV || lgr->lnk[i].peer_qpn == clcqpn) && 1752 !memcmp(lgr->lnk[i].peer_gid, peer_gid, SMC_GID_SIZE) && 1753 (smcr_version == SMC_V2 || 1754 !memcmp(lgr->lnk[i].peer_mac, peer_mac_v1, ETH_ALEN))) 1755 return true; 1756 } 1757 return false; 1758 } 1759 1760 static bool smcd_lgr_match(struct smc_link_group *lgr, 1761 struct smcd_dev *smcismdev, u64 peer_gid) 1762 { 1763 return lgr->peer_gid == peer_gid && lgr->smcd == smcismdev; 1764 } 1765 1766 /* create a new SMC connection (and a new link group if necessary) */ 1767 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini) 1768 { 1769 struct smc_connection *conn = &smc->conn; 1770 struct list_head *lgr_list; 1771 struct smc_link_group *lgr; 1772 enum smc_lgr_role role; 1773 spinlock_t *lgr_lock; 1774 int rc = 0; 1775 1776 lgr_list = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_list : 1777 &smc_lgr_list.list; 1778 lgr_lock = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_lock : 1779 &smc_lgr_list.lock; 1780 ini->first_contact_local = 1; 1781 role = smc->listen_smc ? SMC_SERV : SMC_CLNT; 1782 if (role == SMC_CLNT && ini->first_contact_peer) 1783 /* create new link group as well */ 1784 goto create; 1785 1786 /* determine if an existing link group can be reused */ 1787 spin_lock_bh(lgr_lock); 1788 list_for_each_entry(lgr, lgr_list, list) { 1789 write_lock_bh(&lgr->conns_lock); 1790 if ((ini->is_smcd ? 1791 smcd_lgr_match(lgr, ini->ism_dev[ini->ism_selected], 1792 ini->ism_peer_gid[ini->ism_selected]) : 1793 smcr_lgr_match(lgr, ini->smcr_version, 1794 ini->peer_systemid, 1795 ini->peer_gid, ini->peer_mac, role, 1796 ini->ib_clcqpn)) && 1797 !lgr->sync_err && 1798 (ini->smcd_version == SMC_V2 || 1799 lgr->vlan_id == ini->vlan_id) && 1800 (role == SMC_CLNT || ini->is_smcd || 1801 lgr->conns_num < SMC_RMBS_PER_LGR_MAX)) { 1802 /* link group found */ 1803 ini->first_contact_local = 0; 1804 conn->lgr = lgr; 1805 rc = smc_lgr_register_conn(conn, false); 1806 write_unlock_bh(&lgr->conns_lock); 1807 if (!rc && delayed_work_pending(&lgr->free_work)) 1808 cancel_delayed_work(&lgr->free_work); 1809 break; 1810 } 1811 write_unlock_bh(&lgr->conns_lock); 1812 } 1813 spin_unlock_bh(lgr_lock); 1814 if (rc) 1815 return rc; 1816 1817 if (role == SMC_CLNT && !ini->first_contact_peer && 1818 ini->first_contact_local) { 1819 /* Server reuses a link group, but Client wants to start 1820 * a new one 1821 * send out_of_sync decline, reason synchr. error 1822 */ 1823 return SMC_CLC_DECL_SYNCERR; 1824 } 1825 1826 create: 1827 if (ini->first_contact_local) { 1828 rc = smc_lgr_create(smc, ini); 1829 if (rc) 1830 goto out; 1831 lgr = conn->lgr; 1832 write_lock_bh(&lgr->conns_lock); 1833 rc = smc_lgr_register_conn(conn, true); 1834 write_unlock_bh(&lgr->conns_lock); 1835 if (rc) { 1836 smc_lgr_cleanup_early(lgr); 1837 goto out; 1838 } 1839 } 1840 conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE; 1841 conn->local_tx_ctrl.len = SMC_WR_TX_SIZE; 1842 conn->urg_state = SMC_URG_READ; 1843 init_waitqueue_head(&conn->cdc_pend_tx_wq); 1844 INIT_WORK(&smc->conn.abort_work, smc_conn_abort_work); 1845 if (ini->is_smcd) { 1846 conn->rx_off = sizeof(struct smcd_cdc_msg); 1847 smcd_cdc_rx_init(conn); /* init tasklet for this conn */ 1848 } else { 1849 conn->rx_off = 0; 1850 } 1851 #ifndef KERNEL_HAS_ATOMIC64 1852 spin_lock_init(&conn->acurs_lock); 1853 #endif 1854 1855 out: 1856 return rc; 1857 } 1858 1859 #define SMCD_DMBE_SIZES 6 /* 0 -> 16KB, 1 -> 32KB, .. 6 -> 1MB */ 1860 #define SMCR_RMBE_SIZES 5 /* 0 -> 16KB, 1 -> 32KB, .. 5 -> 512KB */ 1861 1862 /* convert the RMB size into the compressed notation (minimum 16K, see 1863 * SMCD/R_DMBE_SIZES. 1864 * In contrast to plain ilog2, this rounds towards the next power of 2, 1865 * so the socket application gets at least its desired sndbuf / rcvbuf size. 1866 */ 1867 static u8 smc_compress_bufsize(int size, bool is_smcd, bool is_rmb) 1868 { 1869 const unsigned int max_scat = SG_MAX_SINGLE_ALLOC * PAGE_SIZE; 1870 u8 compressed; 1871 1872 if (size <= SMC_BUF_MIN_SIZE) 1873 return 0; 1874 1875 size = (size - 1) >> 14; /* convert to 16K multiple */ 1876 compressed = min_t(u8, ilog2(size) + 1, 1877 is_smcd ? SMCD_DMBE_SIZES : SMCR_RMBE_SIZES); 1878 1879 if (!is_smcd && is_rmb) 1880 /* RMBs are backed by & limited to max size of scatterlists */ 1881 compressed = min_t(u8, compressed, ilog2(max_scat >> 14)); 1882 1883 return compressed; 1884 } 1885 1886 /* convert the RMB size from compressed notation into integer */ 1887 int smc_uncompress_bufsize(u8 compressed) 1888 { 1889 u32 size; 1890 1891 size = 0x00000001 << (((int)compressed) + 14); 1892 return (int)size; 1893 } 1894 1895 /* try to reuse a sndbuf or rmb description slot for a certain 1896 * buffer size; if not available, return NULL 1897 */ 1898 static struct smc_buf_desc *smc_buf_get_slot(int compressed_bufsize, 1899 struct mutex *lock, 1900 struct list_head *buf_list) 1901 { 1902 struct smc_buf_desc *buf_slot; 1903 1904 mutex_lock(lock); 1905 list_for_each_entry(buf_slot, buf_list, list) { 1906 if (cmpxchg(&buf_slot->used, 0, 1) == 0) { 1907 mutex_unlock(lock); 1908 return buf_slot; 1909 } 1910 } 1911 mutex_unlock(lock); 1912 return NULL; 1913 } 1914 1915 /* one of the conditions for announcing a receiver's current window size is 1916 * that it "results in a minimum increase in the window size of 10% of the 1917 * receive buffer space" [RFC7609] 1918 */ 1919 static inline int smc_rmb_wnd_update_limit(int rmbe_size) 1920 { 1921 return min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2); 1922 } 1923 1924 /* map an rmb buf to a link */ 1925 static int smcr_buf_map_link(struct smc_buf_desc *buf_desc, bool is_rmb, 1926 struct smc_link *lnk) 1927 { 1928 int rc; 1929 1930 if (buf_desc->is_map_ib[lnk->link_idx]) 1931 return 0; 1932 1933 rc = sg_alloc_table(&buf_desc->sgt[lnk->link_idx], 1, GFP_KERNEL); 1934 if (rc) 1935 return rc; 1936 sg_set_buf(buf_desc->sgt[lnk->link_idx].sgl, 1937 buf_desc->cpu_addr, buf_desc->len); 1938 1939 /* map sg table to DMA address */ 1940 rc = smc_ib_buf_map_sg(lnk, buf_desc, 1941 is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE); 1942 /* SMC protocol depends on mapping to one DMA address only */ 1943 if (rc != 1) { 1944 rc = -EAGAIN; 1945 goto free_table; 1946 } 1947 1948 /* create a new memory region for the RMB */ 1949 if (is_rmb) { 1950 rc = smc_ib_get_memory_region(lnk->roce_pd, 1951 IB_ACCESS_REMOTE_WRITE | 1952 IB_ACCESS_LOCAL_WRITE, 1953 buf_desc, lnk->link_idx); 1954 if (rc) 1955 goto buf_unmap; 1956 smc_ib_sync_sg_for_device(lnk, buf_desc, DMA_FROM_DEVICE); 1957 } 1958 buf_desc->is_map_ib[lnk->link_idx] = true; 1959 return 0; 1960 1961 buf_unmap: 1962 smc_ib_buf_unmap_sg(lnk, buf_desc, 1963 is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE); 1964 free_table: 1965 sg_free_table(&buf_desc->sgt[lnk->link_idx]); 1966 return rc; 1967 } 1968 1969 /* register a new rmb on IB device, 1970 * must be called under lgr->llc_conf_mutex lock 1971 */ 1972 int smcr_link_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc) 1973 { 1974 if (list_empty(&link->lgr->list)) 1975 return -ENOLINK; 1976 if (!rmb_desc->is_reg_mr[link->link_idx]) { 1977 /* register memory region for new rmb */ 1978 if (smc_wr_reg_send(link, rmb_desc->mr_rx[link->link_idx])) { 1979 rmb_desc->is_reg_err = true; 1980 return -EFAULT; 1981 } 1982 rmb_desc->is_reg_mr[link->link_idx] = true; 1983 } 1984 return 0; 1985 } 1986 1987 static int _smcr_buf_map_lgr(struct smc_link *lnk, struct mutex *lock, 1988 struct list_head *lst, bool is_rmb) 1989 { 1990 struct smc_buf_desc *buf_desc, *bf; 1991 int rc = 0; 1992 1993 mutex_lock(lock); 1994 list_for_each_entry_safe(buf_desc, bf, lst, list) { 1995 if (!buf_desc->used) 1996 continue; 1997 rc = smcr_buf_map_link(buf_desc, is_rmb, lnk); 1998 if (rc) 1999 goto out; 2000 } 2001 out: 2002 mutex_unlock(lock); 2003 return rc; 2004 } 2005 2006 /* map all used buffers of lgr for a new link */ 2007 int smcr_buf_map_lgr(struct smc_link *lnk) 2008 { 2009 struct smc_link_group *lgr = lnk->lgr; 2010 int i, rc = 0; 2011 2012 for (i = 0; i < SMC_RMBE_SIZES; i++) { 2013 rc = _smcr_buf_map_lgr(lnk, &lgr->rmbs_lock, 2014 &lgr->rmbs[i], true); 2015 if (rc) 2016 return rc; 2017 rc = _smcr_buf_map_lgr(lnk, &lgr->sndbufs_lock, 2018 &lgr->sndbufs[i], false); 2019 if (rc) 2020 return rc; 2021 } 2022 return 0; 2023 } 2024 2025 /* register all used buffers of lgr for a new link, 2026 * must be called under lgr->llc_conf_mutex lock 2027 */ 2028 int smcr_buf_reg_lgr(struct smc_link *lnk) 2029 { 2030 struct smc_link_group *lgr = lnk->lgr; 2031 struct smc_buf_desc *buf_desc, *bf; 2032 int i, rc = 0; 2033 2034 mutex_lock(&lgr->rmbs_lock); 2035 for (i = 0; i < SMC_RMBE_SIZES; i++) { 2036 list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list) { 2037 if (!buf_desc->used) 2038 continue; 2039 rc = smcr_link_reg_rmb(lnk, buf_desc); 2040 if (rc) 2041 goto out; 2042 } 2043 } 2044 out: 2045 mutex_unlock(&lgr->rmbs_lock); 2046 return rc; 2047 } 2048 2049 static struct smc_buf_desc *smcr_new_buf_create(struct smc_link_group *lgr, 2050 bool is_rmb, int bufsize) 2051 { 2052 struct smc_buf_desc *buf_desc; 2053 2054 /* try to alloc a new buffer */ 2055 buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL); 2056 if (!buf_desc) 2057 return ERR_PTR(-ENOMEM); 2058 2059 buf_desc->order = get_order(bufsize); 2060 buf_desc->pages = alloc_pages(GFP_KERNEL | __GFP_NOWARN | 2061 __GFP_NOMEMALLOC | __GFP_COMP | 2062 __GFP_NORETRY | __GFP_ZERO, 2063 buf_desc->order); 2064 if (!buf_desc->pages) { 2065 kfree(buf_desc); 2066 return ERR_PTR(-EAGAIN); 2067 } 2068 buf_desc->cpu_addr = (void *)page_address(buf_desc->pages); 2069 buf_desc->len = bufsize; 2070 return buf_desc; 2071 } 2072 2073 /* map buf_desc on all usable links, 2074 * unused buffers stay mapped as long as the link is up 2075 */ 2076 static int smcr_buf_map_usable_links(struct smc_link_group *lgr, 2077 struct smc_buf_desc *buf_desc, bool is_rmb) 2078 { 2079 int i, rc = 0; 2080 2081 /* protect against parallel link reconfiguration */ 2082 mutex_lock(&lgr->llc_conf_mutex); 2083 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 2084 struct smc_link *lnk = &lgr->lnk[i]; 2085 2086 if (!smc_link_usable(lnk)) 2087 continue; 2088 if (smcr_buf_map_link(buf_desc, is_rmb, lnk)) { 2089 rc = -ENOMEM; 2090 goto out; 2091 } 2092 } 2093 out: 2094 mutex_unlock(&lgr->llc_conf_mutex); 2095 return rc; 2096 } 2097 2098 static struct smc_buf_desc *smcd_new_buf_create(struct smc_link_group *lgr, 2099 bool is_dmb, int bufsize) 2100 { 2101 struct smc_buf_desc *buf_desc; 2102 int rc; 2103 2104 /* try to alloc a new DMB */ 2105 buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL); 2106 if (!buf_desc) 2107 return ERR_PTR(-ENOMEM); 2108 if (is_dmb) { 2109 rc = smc_ism_register_dmb(lgr, bufsize, buf_desc); 2110 if (rc) { 2111 kfree(buf_desc); 2112 if (rc == -ENOMEM) 2113 return ERR_PTR(-EAGAIN); 2114 if (rc == -ENOSPC) 2115 return ERR_PTR(-ENOSPC); 2116 return ERR_PTR(-EIO); 2117 } 2118 buf_desc->pages = virt_to_page(buf_desc->cpu_addr); 2119 /* CDC header stored in buf. So, pretend it was smaller */ 2120 buf_desc->len = bufsize - sizeof(struct smcd_cdc_msg); 2121 } else { 2122 buf_desc->cpu_addr = kzalloc(bufsize, GFP_KERNEL | 2123 __GFP_NOWARN | __GFP_NORETRY | 2124 __GFP_NOMEMALLOC); 2125 if (!buf_desc->cpu_addr) { 2126 kfree(buf_desc); 2127 return ERR_PTR(-EAGAIN); 2128 } 2129 buf_desc->len = bufsize; 2130 } 2131 return buf_desc; 2132 } 2133 2134 static int __smc_buf_create(struct smc_sock *smc, bool is_smcd, bool is_rmb) 2135 { 2136 struct smc_buf_desc *buf_desc = ERR_PTR(-ENOMEM); 2137 struct smc_connection *conn = &smc->conn; 2138 struct smc_link_group *lgr = conn->lgr; 2139 struct list_head *buf_list; 2140 int bufsize, bufsize_short; 2141 bool is_dgraded = false; 2142 struct mutex *lock; /* lock buffer list */ 2143 int sk_buf_size; 2144 2145 if (is_rmb) 2146 /* use socket recv buffer size (w/o overhead) as start value */ 2147 sk_buf_size = smc->sk.sk_rcvbuf / 2; 2148 else 2149 /* use socket send buffer size (w/o overhead) as start value */ 2150 sk_buf_size = smc->sk.sk_sndbuf / 2; 2151 2152 for (bufsize_short = smc_compress_bufsize(sk_buf_size, is_smcd, is_rmb); 2153 bufsize_short >= 0; bufsize_short--) { 2154 if (is_rmb) { 2155 lock = &lgr->rmbs_lock; 2156 buf_list = &lgr->rmbs[bufsize_short]; 2157 } else { 2158 lock = &lgr->sndbufs_lock; 2159 buf_list = &lgr->sndbufs[bufsize_short]; 2160 } 2161 bufsize = smc_uncompress_bufsize(bufsize_short); 2162 2163 /* check for reusable slot in the link group */ 2164 buf_desc = smc_buf_get_slot(bufsize_short, lock, buf_list); 2165 if (buf_desc) { 2166 SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize); 2167 SMC_STAT_BUF_REUSE(smc, is_smcd, is_rmb); 2168 memset(buf_desc->cpu_addr, 0, bufsize); 2169 break; /* found reusable slot */ 2170 } 2171 2172 if (is_smcd) 2173 buf_desc = smcd_new_buf_create(lgr, is_rmb, bufsize); 2174 else 2175 buf_desc = smcr_new_buf_create(lgr, is_rmb, bufsize); 2176 2177 if (PTR_ERR(buf_desc) == -ENOMEM) 2178 break; 2179 if (IS_ERR(buf_desc)) { 2180 if (!is_dgraded) { 2181 is_dgraded = true; 2182 SMC_STAT_RMB_DOWNGRADED(smc, is_smcd, is_rmb); 2183 } 2184 continue; 2185 } 2186 2187 SMC_STAT_RMB_ALLOC(smc, is_smcd, is_rmb); 2188 SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize); 2189 buf_desc->used = 1; 2190 mutex_lock(lock); 2191 list_add(&buf_desc->list, buf_list); 2192 mutex_unlock(lock); 2193 break; /* found */ 2194 } 2195 2196 if (IS_ERR(buf_desc)) 2197 return PTR_ERR(buf_desc); 2198 2199 if (!is_smcd) { 2200 if (smcr_buf_map_usable_links(lgr, buf_desc, is_rmb)) { 2201 smcr_buf_unuse(buf_desc, lgr); 2202 return -ENOMEM; 2203 } 2204 } 2205 2206 if (is_rmb) { 2207 conn->rmb_desc = buf_desc; 2208 conn->rmbe_size_short = bufsize_short; 2209 smc->sk.sk_rcvbuf = bufsize * 2; 2210 atomic_set(&conn->bytes_to_rcv, 0); 2211 conn->rmbe_update_limit = 2212 smc_rmb_wnd_update_limit(buf_desc->len); 2213 if (is_smcd) 2214 smc_ism_set_conn(conn); /* map RMB/smcd_dev to conn */ 2215 } else { 2216 conn->sndbuf_desc = buf_desc; 2217 smc->sk.sk_sndbuf = bufsize * 2; 2218 atomic_set(&conn->sndbuf_space, bufsize); 2219 } 2220 return 0; 2221 } 2222 2223 void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn) 2224 { 2225 if (!conn->lgr || conn->lgr->is_smcd || !smc_link_active(conn->lnk)) 2226 return; 2227 smc_ib_sync_sg_for_cpu(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE); 2228 } 2229 2230 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn) 2231 { 2232 if (!conn->lgr || conn->lgr->is_smcd || !smc_link_active(conn->lnk)) 2233 return; 2234 smc_ib_sync_sg_for_device(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE); 2235 } 2236 2237 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn) 2238 { 2239 int i; 2240 2241 if (!conn->lgr || conn->lgr->is_smcd) 2242 return; 2243 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 2244 if (!smc_link_active(&conn->lgr->lnk[i])) 2245 continue; 2246 smc_ib_sync_sg_for_cpu(&conn->lgr->lnk[i], conn->rmb_desc, 2247 DMA_FROM_DEVICE); 2248 } 2249 } 2250 2251 void smc_rmb_sync_sg_for_device(struct smc_connection *conn) 2252 { 2253 int i; 2254 2255 if (!conn->lgr || conn->lgr->is_smcd) 2256 return; 2257 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 2258 if (!smc_link_active(&conn->lgr->lnk[i])) 2259 continue; 2260 smc_ib_sync_sg_for_device(&conn->lgr->lnk[i], conn->rmb_desc, 2261 DMA_FROM_DEVICE); 2262 } 2263 } 2264 2265 /* create the send and receive buffer for an SMC socket; 2266 * receive buffers are called RMBs; 2267 * (even though the SMC protocol allows more than one RMB-element per RMB, 2268 * the Linux implementation uses just one RMB-element per RMB, i.e. uses an 2269 * extra RMB for every connection in a link group 2270 */ 2271 int smc_buf_create(struct smc_sock *smc, bool is_smcd) 2272 { 2273 int rc; 2274 2275 /* create send buffer */ 2276 rc = __smc_buf_create(smc, is_smcd, false); 2277 if (rc) 2278 return rc; 2279 /* create rmb */ 2280 rc = __smc_buf_create(smc, is_smcd, true); 2281 if (rc) { 2282 mutex_lock(&smc->conn.lgr->sndbufs_lock); 2283 list_del(&smc->conn.sndbuf_desc->list); 2284 mutex_unlock(&smc->conn.lgr->sndbufs_lock); 2285 smc_buf_free(smc->conn.lgr, false, smc->conn.sndbuf_desc); 2286 smc->conn.sndbuf_desc = NULL; 2287 } 2288 return rc; 2289 } 2290 2291 static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr) 2292 { 2293 int i; 2294 2295 for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) { 2296 if (!test_and_set_bit(i, lgr->rtokens_used_mask)) 2297 return i; 2298 } 2299 return -ENOSPC; 2300 } 2301 2302 static int smc_rtoken_find_by_link(struct smc_link_group *lgr, int lnk_idx, 2303 u32 rkey) 2304 { 2305 int i; 2306 2307 for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) { 2308 if (test_bit(i, lgr->rtokens_used_mask) && 2309 lgr->rtokens[i][lnk_idx].rkey == rkey) 2310 return i; 2311 } 2312 return -ENOENT; 2313 } 2314 2315 /* set rtoken for a new link to an existing rmb */ 2316 void smc_rtoken_set(struct smc_link_group *lgr, int link_idx, int link_idx_new, 2317 __be32 nw_rkey_known, __be64 nw_vaddr, __be32 nw_rkey) 2318 { 2319 int rtok_idx; 2320 2321 rtok_idx = smc_rtoken_find_by_link(lgr, link_idx, ntohl(nw_rkey_known)); 2322 if (rtok_idx == -ENOENT) 2323 return; 2324 lgr->rtokens[rtok_idx][link_idx_new].rkey = ntohl(nw_rkey); 2325 lgr->rtokens[rtok_idx][link_idx_new].dma_addr = be64_to_cpu(nw_vaddr); 2326 } 2327 2328 /* set rtoken for a new link whose link_id is given */ 2329 void smc_rtoken_set2(struct smc_link_group *lgr, int rtok_idx, int link_id, 2330 __be64 nw_vaddr, __be32 nw_rkey) 2331 { 2332 u64 dma_addr = be64_to_cpu(nw_vaddr); 2333 u32 rkey = ntohl(nw_rkey); 2334 bool found = false; 2335 int link_idx; 2336 2337 for (link_idx = 0; link_idx < SMC_LINKS_PER_LGR_MAX; link_idx++) { 2338 if (lgr->lnk[link_idx].link_id == link_id) { 2339 found = true; 2340 break; 2341 } 2342 } 2343 if (!found) 2344 return; 2345 lgr->rtokens[rtok_idx][link_idx].rkey = rkey; 2346 lgr->rtokens[rtok_idx][link_idx].dma_addr = dma_addr; 2347 } 2348 2349 /* add a new rtoken from peer */ 2350 int smc_rtoken_add(struct smc_link *lnk, __be64 nw_vaddr, __be32 nw_rkey) 2351 { 2352 struct smc_link_group *lgr = smc_get_lgr(lnk); 2353 u64 dma_addr = be64_to_cpu(nw_vaddr); 2354 u32 rkey = ntohl(nw_rkey); 2355 int i; 2356 2357 for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) { 2358 if (lgr->rtokens[i][lnk->link_idx].rkey == rkey && 2359 lgr->rtokens[i][lnk->link_idx].dma_addr == dma_addr && 2360 test_bit(i, lgr->rtokens_used_mask)) { 2361 /* already in list */ 2362 return i; 2363 } 2364 } 2365 i = smc_rmb_reserve_rtoken_idx(lgr); 2366 if (i < 0) 2367 return i; 2368 lgr->rtokens[i][lnk->link_idx].rkey = rkey; 2369 lgr->rtokens[i][lnk->link_idx].dma_addr = dma_addr; 2370 return i; 2371 } 2372 2373 /* delete an rtoken from all links */ 2374 int smc_rtoken_delete(struct smc_link *lnk, __be32 nw_rkey) 2375 { 2376 struct smc_link_group *lgr = smc_get_lgr(lnk); 2377 u32 rkey = ntohl(nw_rkey); 2378 int i, j; 2379 2380 for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) { 2381 if (lgr->rtokens[i][lnk->link_idx].rkey == rkey && 2382 test_bit(i, lgr->rtokens_used_mask)) { 2383 for (j = 0; j < SMC_LINKS_PER_LGR_MAX; j++) { 2384 lgr->rtokens[i][j].rkey = 0; 2385 lgr->rtokens[i][j].dma_addr = 0; 2386 } 2387 clear_bit(i, lgr->rtokens_used_mask); 2388 return 0; 2389 } 2390 } 2391 return -ENOENT; 2392 } 2393 2394 /* save rkey and dma_addr received from peer during clc handshake */ 2395 int smc_rmb_rtoken_handling(struct smc_connection *conn, 2396 struct smc_link *lnk, 2397 struct smc_clc_msg_accept_confirm *clc) 2398 { 2399 conn->rtoken_idx = smc_rtoken_add(lnk, clc->r0.rmb_dma_addr, 2400 clc->r0.rmb_rkey); 2401 if (conn->rtoken_idx < 0) 2402 return conn->rtoken_idx; 2403 return 0; 2404 } 2405 2406 static void smc_core_going_away(void) 2407 { 2408 struct smc_ib_device *smcibdev; 2409 struct smcd_dev *smcd; 2410 2411 mutex_lock(&smc_ib_devices.mutex); 2412 list_for_each_entry(smcibdev, &smc_ib_devices.list, list) { 2413 int i; 2414 2415 for (i = 0; i < SMC_MAX_PORTS; i++) 2416 set_bit(i, smcibdev->ports_going_away); 2417 } 2418 mutex_unlock(&smc_ib_devices.mutex); 2419 2420 mutex_lock(&smcd_dev_list.mutex); 2421 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2422 smcd->going_away = 1; 2423 } 2424 mutex_unlock(&smcd_dev_list.mutex); 2425 } 2426 2427 /* Clean up all SMC link groups */ 2428 static void smc_lgrs_shutdown(void) 2429 { 2430 struct smcd_dev *smcd; 2431 2432 smc_core_going_away(); 2433 2434 smc_smcr_terminate_all(NULL); 2435 2436 mutex_lock(&smcd_dev_list.mutex); 2437 list_for_each_entry(smcd, &smcd_dev_list.list, list) 2438 smc_smcd_terminate_all(smcd); 2439 mutex_unlock(&smcd_dev_list.mutex); 2440 } 2441 2442 static int smc_core_reboot_event(struct notifier_block *this, 2443 unsigned long event, void *ptr) 2444 { 2445 smc_lgrs_shutdown(); 2446 smc_ib_unregister_client(); 2447 return 0; 2448 } 2449 2450 static struct notifier_block smc_reboot_notifier = { 2451 .notifier_call = smc_core_reboot_event, 2452 }; 2453 2454 int __init smc_core_init(void) 2455 { 2456 return register_reboot_notifier(&smc_reboot_notifier); 2457 } 2458 2459 /* Called (from smc_exit) when module is removed */ 2460 void smc_core_exit(void) 2461 { 2462 unregister_reboot_notifier(&smc_reboot_notifier); 2463 smc_lgrs_shutdown(); 2464 } 2465