hw/rdma/rdma_rm.c

   1 /*
   2  * QEMU paravirtual RDMA - Resource Manager Implementation
   3  *
   4  * Copyright (C) 2018 Oracle
   5  * Copyright (C) 2018 Red Hat Inc
   6  *
   7  * Authors:
   8  *     Yuval Shaia <yuval.shaia@oracle.com>
   9  *     Marcel Apfelbaum <marcel@redhat.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  12  * See the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include "qemu/osdep.h"
  17 #include "qapi/error.h"
  18 #include "cpu.h"
  19 #include "monitor/monitor.h"
  20
  21 #include "trace.h"
  22 #include "rdma_utils.h"
  23 #include "rdma_backend.h"
  24 #include "rdma_rm.h"
  25
  26 /* Page directory and page tables */
  27 #define PG_DIR_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
  28 #define PG_TBL_SZ { TARGET_PAGE_SIZE / sizeof(__u64) }
  29
  30 void rdma_dump_device_counters(Monitor *mon, RdmaDeviceResources *dev_res)
  31 {
  32     monitor_printf(mon, "\ttx               : %" PRId64 "\n",
  33                    dev_res->stats.tx);
  34     monitor_printf(mon, "\ttx_len           : %" PRId64 "\n",
  35                    dev_res->stats.tx_len);
  36     monitor_printf(mon, "\ttx_err           : %" PRId64 "\n",
  37                    dev_res->stats.tx_err);
  38     monitor_printf(mon, "\trx_bufs          : %" PRId64 "\n",
  39                    dev_res->stats.rx_bufs);
  40     monitor_printf(mon, "\trx_srq           : %" PRId64 "\n",
  41                    dev_res->stats.rx_srq);
  42     monitor_printf(mon, "\trx_bufs_len      : %" PRId64 "\n",
  43                    dev_res->stats.rx_bufs_len);
  44     monitor_printf(mon, "\trx_bufs_err      : %" PRId64 "\n",
  45                    dev_res->stats.rx_bufs_err);
  46     monitor_printf(mon, "\tcomps            : %" PRId64 "\n",
  47                    dev_res->stats.completions);
  48     monitor_printf(mon, "\tmissing_comps    : %" PRId32 "\n",
  49                    dev_res->stats.missing_cqe);
  50     monitor_printf(mon, "\tpoll_cq (bk)     : %" PRId64 "\n",
  51                    dev_res->stats.poll_cq_from_bk);
  52     monitor_printf(mon, "\tpoll_cq_ppoll_to : %" PRId64 "\n",
  53                    dev_res->stats.poll_cq_ppoll_to);
  54     monitor_printf(mon, "\tpoll_cq (fe)     : %" PRId64 "\n",
  55                    dev_res->stats.poll_cq_from_guest);
  56     monitor_printf(mon, "\tpoll_cq_empty    : %" PRId64 "\n",
  57                    dev_res->stats.poll_cq_from_guest_empty);
  58     monitor_printf(mon, "\tmad_tx           : %" PRId64 "\n",
  59                    dev_res->stats.mad_tx);
  60     monitor_printf(mon, "\tmad_tx_err       : %" PRId64 "\n",
  61                    dev_res->stats.mad_tx_err);
  62     monitor_printf(mon, "\tmad_rx           : %" PRId64 "\n",
  63                    dev_res->stats.mad_rx);
  64     monitor_printf(mon, "\tmad_rx_err       : %" PRId64 "\n",
  65                    dev_res->stats.mad_rx_err);
  66     monitor_printf(mon, "\tmad_rx_bufs      : %" PRId64 "\n",
  67                    dev_res->stats.mad_rx_bufs);
  68     monitor_printf(mon, "\tmad_rx_bufs_err  : %" PRId64 "\n",
  69                    dev_res->stats.mad_rx_bufs_err);
  70     monitor_printf(mon, "\tPDs              : %" PRId32 "\n",
  71                    dev_res->pd_tbl.used);
  72     monitor_printf(mon, "\tMRs              : %" PRId32 "\n",
  73                    dev_res->mr_tbl.used);
  74     monitor_printf(mon, "\tUCs              : %" PRId32 "\n",
  75                    dev_res->uc_tbl.used);
  76     monitor_printf(mon, "\tQPs              : %" PRId32 "\n",
  77                    dev_res->qp_tbl.used);
  78     monitor_printf(mon, "\tCQs              : %" PRId32 "\n",
  79                    dev_res->cq_tbl.used);
  80     monitor_printf(mon, "\tCEQ_CTXs         : %" PRId32 "\n",
  81                    dev_res->cqe_ctx_tbl.used);
  82 }
  83
  84 static inline void res_tbl_init(const char *name, RdmaRmResTbl *tbl,
  85                                 uint32_t tbl_sz, uint32_t res_sz)
  86 {
  87     tbl->tbl = g_malloc(tbl_sz * res_sz);
  88
  89     strncpy(tbl->name, name, MAX_RM_TBL_NAME);
  90     tbl->name[MAX_RM_TBL_NAME - 1] = 0;
  91
  92     tbl->bitmap = bitmap_new(tbl_sz);
  93     tbl->tbl_sz = tbl_sz;
  94     tbl->res_sz = res_sz;
  95     tbl->used = 0;
  96     qemu_mutex_init(&tbl->lock);
  97 }
  98
  99 static inline void res_tbl_free(RdmaRmResTbl *tbl)
 100 {
 101     if (!tbl->bitmap) {
 102         return;
 103     }
 104     qemu_mutex_destroy(&tbl->lock);
 105     g_free(tbl->tbl);
 106     g_free(tbl->bitmap);
 107 }
 108
 109 static inline void *rdma_res_tbl_get(RdmaRmResTbl *tbl, uint32_t handle)
 110 {
 111     trace_rdma_res_tbl_get(tbl->name, handle);
 112
 113     if ((handle < tbl->tbl_sz) && (test_bit(handle, tbl->bitmap))) {
 114         return tbl->tbl + handle * tbl->res_sz;
 115     } else {
 116         rdma_error_report("Table %s, invalid handle %d", tbl->name, handle);
 117         return NULL;
 118     }
 119 }
 120
 121 static inline void *rdma_res_tbl_alloc(RdmaRmResTbl *tbl, uint32_t *handle)
 122 {
 123     qemu_mutex_lock(&tbl->lock);
 124
 125     *handle = find_first_zero_bit(tbl->bitmap, tbl->tbl_sz);
 126     if (*handle > tbl->tbl_sz) {
 127         rdma_error_report("Table %s, failed to allocate, bitmap is full",
 128                           tbl->name);
 129         qemu_mutex_unlock(&tbl->lock);
 130         return NULL;
 131     }
 132
 133     set_bit(*handle, tbl->bitmap);
 134
 135     tbl->used++;
 136
 137     qemu_mutex_unlock(&tbl->lock);
 138
 139     memset(tbl->tbl + *handle * tbl->res_sz, 0, tbl->res_sz);
 140
 141     trace_rdma_res_tbl_alloc(tbl->name, *handle);
 142
 143     return tbl->tbl + *handle * tbl->res_sz;
 144 }
 145
 146 static inline void rdma_res_tbl_dealloc(RdmaRmResTbl *tbl, uint32_t handle)
 147 {
 148     trace_rdma_res_tbl_dealloc(tbl->name, handle);
 149
 150     QEMU_LOCK_GUARD(&tbl->lock);
 151
 152     if (handle < tbl->tbl_sz) {
 153         clear_bit(handle, tbl->bitmap);
 154         tbl->used--;
 155     }
 156
 157 }
 158
 159 int rdma_rm_alloc_pd(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 160                      uint32_t *pd_handle, uint32_t ctx_handle)
 161 {
 162     RdmaRmPD *pd;
 163     int ret = -ENOMEM;
 164
 165     pd = rdma_res_tbl_alloc(&dev_res->pd_tbl, pd_handle);
 166     if (!pd) {
 167         goto out;
 168     }
 169
 170     ret = rdma_backend_create_pd(backend_dev, &pd->backend_pd);
 171     if (ret) {
 172         ret = -EIO;
 173         goto out_tbl_dealloc;
 174     }
 175
 176     pd->ctx_handle = ctx_handle;
 177
 178     return 0;
 179
 180 out_tbl_dealloc:
 181     rdma_res_tbl_dealloc(&dev_res->pd_tbl, *pd_handle);
 182
 183 out:
 184     return ret;
 185 }
 186
 187 RdmaRmPD *rdma_rm_get_pd(RdmaDeviceResources *dev_res, uint32_t pd_handle)
 188 {
 189     return rdma_res_tbl_get(&dev_res->pd_tbl, pd_handle);
 190 }
 191
 192 void rdma_rm_dealloc_pd(RdmaDeviceResources *dev_res, uint32_t pd_handle)
 193 {
 194     RdmaRmPD *pd = rdma_rm_get_pd(dev_res, pd_handle);
 195
 196     if (pd) {
 197         rdma_backend_destroy_pd(&pd->backend_pd);
 198         rdma_res_tbl_dealloc(&dev_res->pd_tbl, pd_handle);
 199     }
 200 }
 201
 202 int rdma_rm_alloc_mr(RdmaDeviceResources *dev_res, uint32_t pd_handle,
 203                      uint64_t guest_start, uint64_t guest_length,
 204                      void *host_virt, int access_flags, uint32_t *mr_handle,
 205                      uint32_t *lkey, uint32_t *rkey)
 206 {
 207     RdmaRmMR *mr;
 208     int ret = 0;
 209     RdmaRmPD *pd;
 210
 211     pd = rdma_rm_get_pd(dev_res, pd_handle);
 212     if (!pd) {
 213         return -EINVAL;
 214     }
 215
 216     mr = rdma_res_tbl_alloc(&dev_res->mr_tbl, mr_handle);
 217     if (!mr) {
 218         return -ENOMEM;
 219     }
 220     trace_rdma_rm_alloc_mr(*mr_handle, host_virt, guest_start, guest_length,
 221                            access_flags);
 222
 223     if (host_virt) {
 224         mr->virt = host_virt;
 225         mr->start = guest_start;
 226         mr->length = guest_length;
 227         mr->virt += (mr->start & (TARGET_PAGE_SIZE - 1));
 228
 229         ret = rdma_backend_create_mr(&mr->backend_mr, &pd->backend_pd, mr->virt,
 230                                      mr->length, guest_start, access_flags);
 231         if (ret) {
 232             ret = -EIO;
 233             goto out_dealloc_mr;
 234         }
 235 #ifdef LEGACY_RDMA_REG_MR
 236         /* We keep mr_handle in lkey so send and recv get get mr ptr */
 237         *lkey = *mr_handle;
 238 #else
 239         *lkey = rdma_backend_mr_lkey(&mr->backend_mr);
 240 #endif
 241     }
 242
 243     *rkey = -1;
 244
 245     mr->pd_handle = pd_handle;
 246
 247     return 0;
 248
 249 out_dealloc_mr:
 250     rdma_res_tbl_dealloc(&dev_res->mr_tbl, *mr_handle);
 251
 252     return ret;
 253 }
 254
 255 RdmaRmMR *rdma_rm_get_mr(RdmaDeviceResources *dev_res, uint32_t mr_handle)
 256 {
 257     return rdma_res_tbl_get(&dev_res->mr_tbl, mr_handle);
 258 }
 259
 260 void rdma_rm_dealloc_mr(RdmaDeviceResources *dev_res, uint32_t mr_handle)
 261 {
 262     RdmaRmMR *mr = rdma_rm_get_mr(dev_res, mr_handle);
 263
 264     if (mr) {
 265         rdma_backend_destroy_mr(&mr->backend_mr);
 266         trace_rdma_rm_dealloc_mr(mr_handle, mr->start);
 267         if (mr->start) {
 268             mr->virt -= (mr->start & (TARGET_PAGE_SIZE - 1));
 269             munmap(mr->virt, mr->length);
 270         }
 271         rdma_res_tbl_dealloc(&dev_res->mr_tbl, mr_handle);
 272     }
 273 }
 274
 275 int rdma_rm_alloc_uc(RdmaDeviceResources *dev_res, uint32_t pfn,
 276                      uint32_t *uc_handle)
 277 {
 278     RdmaRmUC *uc;
 279
 280     /* TODO: Need to make sure pfn is between bar start address and
 281      * bsd+RDMA_BAR2_UAR_SIZE
 282     if (pfn > RDMA_BAR2_UAR_SIZE) {
 283         rdma_error_report("pfn out of range (%d > %d)", pfn,
 284                           RDMA_BAR2_UAR_SIZE);
 285         return -ENOMEM;
 286     }
 287     */
 288
 289     uc = rdma_res_tbl_alloc(&dev_res->uc_tbl, uc_handle);
 290     if (!uc) {
 291         return -ENOMEM;
 292     }
 293
 294     return 0;
 295 }
 296
 297 RdmaRmUC *rdma_rm_get_uc(RdmaDeviceResources *dev_res, uint32_t uc_handle)
 298 {
 299     return rdma_res_tbl_get(&dev_res->uc_tbl, uc_handle);
 300 }
 301
 302 void rdma_rm_dealloc_uc(RdmaDeviceResources *dev_res, uint32_t uc_handle)
 303 {
 304     RdmaRmUC *uc = rdma_rm_get_uc(dev_res, uc_handle);
 305
 306     if (uc) {
 307         rdma_res_tbl_dealloc(&dev_res->uc_tbl, uc_handle);
 308     }
 309 }
 310
 311 RdmaRmCQ *rdma_rm_get_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle)
 312 {
 313     return rdma_res_tbl_get(&dev_res->cq_tbl, cq_handle);
 314 }
 315
 316 int rdma_rm_alloc_cq(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 317                      uint32_t cqe, uint32_t *cq_handle, void *opaque)
 318 {
 319     int rc;
 320     RdmaRmCQ *cq;
 321
 322     cq = rdma_res_tbl_alloc(&dev_res->cq_tbl, cq_handle);
 323     if (!cq) {
 324         return -ENOMEM;
 325     }
 326
 327     cq->opaque = opaque;
 328     cq->notify = CNT_CLEAR;
 329
 330     rc = rdma_backend_create_cq(backend_dev, &cq->backend_cq, cqe);
 331     if (rc) {
 332         rc = -EIO;
 333         goto out_dealloc_cq;
 334     }
 335
 336     return 0;
 337
 338 out_dealloc_cq:
 339     rdma_rm_dealloc_cq(dev_res, *cq_handle);
 340
 341     return rc;
 342 }
 343
 344 void rdma_rm_req_notify_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle,
 345                            bool notify)
 346 {
 347     RdmaRmCQ *cq;
 348
 349     cq = rdma_rm_get_cq(dev_res, cq_handle);
 350     if (!cq) {
 351         return;
 352     }
 353
 354     if (cq->notify != CNT_SET) {
 355         cq->notify = notify ? CNT_ARM : CNT_CLEAR;
 356     }
 357 }
 358
 359 void rdma_rm_dealloc_cq(RdmaDeviceResources *dev_res, uint32_t cq_handle)
 360 {
 361     RdmaRmCQ *cq;
 362
 363     cq = rdma_rm_get_cq(dev_res, cq_handle);
 364     if (!cq) {
 365         return;
 366     }
 367
 368     rdma_backend_destroy_cq(&cq->backend_cq);
 369
 370     rdma_res_tbl_dealloc(&dev_res->cq_tbl, cq_handle);
 371 }
 372
 373 RdmaRmQP *rdma_rm_get_qp(RdmaDeviceResources *dev_res, uint32_t qpn)
 374 {
 375     GBytes *key = g_bytes_new(&qpn, sizeof(qpn));
 376
 377     RdmaRmQP *qp = g_hash_table_lookup(dev_res->qp_hash, key);
 378
 379     g_bytes_unref(key);
 380
 381     if (!qp) {
 382         rdma_error_report("Invalid QP handle %d", qpn);
 383     }
 384
 385     return qp;
 386 }
 387
 388 int rdma_rm_alloc_qp(RdmaDeviceResources *dev_res, uint32_t pd_handle,
 389                      uint8_t qp_type, uint32_t max_send_wr,
 390                      uint32_t max_send_sge, uint32_t send_cq_handle,
 391                      uint32_t max_recv_wr, uint32_t max_recv_sge,
 392                      uint32_t recv_cq_handle, void *opaque, uint32_t *qpn,
 393                      uint8_t is_srq, uint32_t srq_handle)
 394 {
 395     int rc;
 396     RdmaRmQP *qp;
 397     RdmaRmCQ *scq, *rcq;
 398     RdmaRmPD *pd;
 399     RdmaRmSRQ *srq = NULL;
 400     uint32_t rm_qpn;
 401
 402     pd = rdma_rm_get_pd(dev_res, pd_handle);
 403     if (!pd) {
 404         return -EINVAL;
 405     }
 406
 407     scq = rdma_rm_get_cq(dev_res, send_cq_handle);
 408     rcq = rdma_rm_get_cq(dev_res, recv_cq_handle);
 409
 410     if (!scq || !rcq) {
 411         rdma_error_report("Invalid send_cqn or recv_cqn (%d, %d)",
 412                           send_cq_handle, recv_cq_handle);
 413         return -EINVAL;
 414     }
 415
 416     if (is_srq) {
 417         srq = rdma_rm_get_srq(dev_res, srq_handle);
 418         if (!srq) {
 419             rdma_error_report("Invalid srqn %d", srq_handle);
 420             return -EINVAL;
 421         }
 422
 423         srq->recv_cq_handle = recv_cq_handle;
 424     }
 425
 426     if (qp_type == IBV_QPT_GSI) {
 427         scq->notify = CNT_SET;
 428         rcq->notify = CNT_SET;
 429     }
 430
 431     qp = rdma_res_tbl_alloc(&dev_res->qp_tbl, &rm_qpn);
 432     if (!qp) {
 433         return -ENOMEM;
 434     }
 435
 436     qp->qpn = rm_qpn;
 437     qp->qp_state = IBV_QPS_RESET;
 438     qp->qp_type = qp_type;
 439     qp->send_cq_handle = send_cq_handle;
 440     qp->recv_cq_handle = recv_cq_handle;
 441     qp->opaque = opaque;
 442     qp->is_srq = is_srq;
 443
 444     rc = rdma_backend_create_qp(&qp->backend_qp, qp_type, &pd->backend_pd,
 445                                 &scq->backend_cq, &rcq->backend_cq,
 446                                 is_srq ? &srq->backend_srq : NULL,
 447                                 max_send_wr, max_recv_wr, max_send_sge,
 448                                 max_recv_sge);
 449
 450     if (rc) {
 451         rc = -EIO;
 452         goto out_dealloc_qp;
 453     }
 454
 455     *qpn = rdma_backend_qpn(&qp->backend_qp);
 456     trace_rdma_rm_alloc_qp(rm_qpn, *qpn, qp_type);
 457     g_hash_table_insert(dev_res->qp_hash, g_bytes_new(qpn, sizeof(*qpn)), qp);
 458
 459     return 0;
 460
 461 out_dealloc_qp:
 462     rdma_res_tbl_dealloc(&dev_res->qp_tbl, qp->qpn);
 463
 464     return rc;
 465 }
 466
 467 int rdma_rm_modify_qp(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 468                       uint32_t qp_handle, uint32_t attr_mask, uint8_t sgid_idx,
 469                       union ibv_gid *dgid, uint32_t dqpn,
 470                       enum ibv_qp_state qp_state, uint32_t qkey,
 471                       uint32_t rq_psn, uint32_t sq_psn)
 472 {
 473     RdmaRmQP *qp;
 474     int ret;
 475
 476     qp = rdma_rm_get_qp(dev_res, qp_handle);
 477     if (!qp) {
 478         return -EINVAL;
 479     }
 480
 481     if (qp->qp_type == IBV_QPT_SMI) {
 482         rdma_error_report("Got QP0 request");
 483         return -EPERM;
 484     } else if (qp->qp_type == IBV_QPT_GSI) {
 485         return 0;
 486     }
 487
 488     trace_rdma_rm_modify_qp(qp_handle, attr_mask, qp_state, sgid_idx);
 489
 490     if (attr_mask & IBV_QP_STATE) {
 491         qp->qp_state = qp_state;
 492
 493         if (qp->qp_state == IBV_QPS_INIT) {
 494             ret = rdma_backend_qp_state_init(backend_dev, &qp->backend_qp,
 495                                              qp->qp_type, qkey);
 496             if (ret) {
 497                 return -EIO;
 498             }
 499         }
 500
 501         if (qp->qp_state == IBV_QPS_RTR) {
 502             /* Get backend gid index */
 503             sgid_idx = rdma_rm_get_backend_gid_index(dev_res, backend_dev,
 504                                                      sgid_idx);
 505             if (sgid_idx <= 0) { /* TODO check also less than bk.max_sgid */
 506                 rdma_error_report("Failed to get bk sgid_idx for sgid_idx %d",
 507                                   sgid_idx);
 508                 return -EIO;
 509             }
 510
 511             ret = rdma_backend_qp_state_rtr(backend_dev, &qp->backend_qp,
 512                                             qp->qp_type, sgid_idx, dgid, dqpn,
 513                                             rq_psn, qkey,
 514                                             attr_mask & IBV_QP_QKEY);
 515             if (ret) {
 516                 return -EIO;
 517             }
 518         }
 519
 520         if (qp->qp_state == IBV_QPS_RTS) {
 521             ret = rdma_backend_qp_state_rts(&qp->backend_qp, qp->qp_type,
 522                                             sq_psn, qkey,
 523                                             attr_mask & IBV_QP_QKEY);
 524             if (ret) {
 525                 return -EIO;
 526             }
 527         }
 528     }
 529
 530     return 0;
 531 }
 532
 533 int rdma_rm_query_qp(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 534                      uint32_t qp_handle, struct ibv_qp_attr *attr,
 535                      int attr_mask, struct ibv_qp_init_attr *init_attr)
 536 {
 537     RdmaRmQP *qp;
 538
 539     qp = rdma_rm_get_qp(dev_res, qp_handle);
 540     if (!qp) {
 541         return -EINVAL;
 542     }
 543
 544     return rdma_backend_query_qp(&qp->backend_qp, attr, attr_mask, init_attr);
 545 }
 546
 547 void rdma_rm_dealloc_qp(RdmaDeviceResources *dev_res, uint32_t qp_handle)
 548 {
 549     RdmaRmQP *qp;
 550     GBytes *key;
 551
 552     key = g_bytes_new(&qp_handle, sizeof(qp_handle));
 553     qp = g_hash_table_lookup(dev_res->qp_hash, key);
 554     g_hash_table_remove(dev_res->qp_hash, key);
 555     g_bytes_unref(key);
 556
 557     if (!qp) {
 558         return;
 559     }
 560
 561     rdma_backend_destroy_qp(&qp->backend_qp, dev_res);
 562
 563     rdma_res_tbl_dealloc(&dev_res->qp_tbl, qp->qpn);
 564 }
 565
 566 RdmaRmSRQ *rdma_rm_get_srq(RdmaDeviceResources *dev_res, uint32_t srq_handle)
 567 {
 568     return rdma_res_tbl_get(&dev_res->srq_tbl, srq_handle);
 569 }
 570
 571 int rdma_rm_alloc_srq(RdmaDeviceResources *dev_res, uint32_t pd_handle,
 572                       uint32_t max_wr, uint32_t max_sge, uint32_t srq_limit,
 573                       uint32_t *srq_handle, void *opaque)
 574 {
 575     RdmaRmSRQ *srq;
 576     RdmaRmPD *pd;
 577     int rc;
 578
 579     pd = rdma_rm_get_pd(dev_res, pd_handle);
 580     if (!pd) {
 581         return -EINVAL;
 582     }
 583
 584     srq = rdma_res_tbl_alloc(&dev_res->srq_tbl, srq_handle);
 585     if (!srq) {
 586         return -ENOMEM;
 587     }
 588
 589     rc = rdma_backend_create_srq(&srq->backend_srq, &pd->backend_pd,
 590                                  max_wr, max_sge, srq_limit);
 591     if (rc) {
 592         rc = -EIO;
 593         goto out_dealloc_srq;
 594     }
 595
 596     srq->opaque = opaque;
 597
 598     return 0;
 599
 600 out_dealloc_srq:
 601     rdma_res_tbl_dealloc(&dev_res->srq_tbl, *srq_handle);
 602
 603     return rc;
 604 }
 605
 606 int rdma_rm_query_srq(RdmaDeviceResources *dev_res, uint32_t srq_handle,
 607                       struct ibv_srq_attr *srq_attr)
 608 {
 609     RdmaRmSRQ *srq;
 610
 611     srq = rdma_rm_get_srq(dev_res, srq_handle);
 612     if (!srq) {
 613         return -EINVAL;
 614     }
 615
 616     return rdma_backend_query_srq(&srq->backend_srq, srq_attr);
 617 }
 618
 619 int rdma_rm_modify_srq(RdmaDeviceResources *dev_res, uint32_t srq_handle,
 620                        struct ibv_srq_attr *srq_attr, int srq_attr_mask)
 621 {
 622     RdmaRmSRQ *srq;
 623
 624     srq = rdma_rm_get_srq(dev_res, srq_handle);
 625     if (!srq) {
 626         return -EINVAL;
 627     }
 628
 629     if ((srq_attr_mask & IBV_SRQ_LIMIT) &&
 630         (srq_attr->srq_limit == 0)) {
 631         return -EINVAL;
 632     }
 633
 634     if ((srq_attr_mask & IBV_SRQ_MAX_WR) &&
 635         (srq_attr->max_wr == 0)) {
 636         return -EINVAL;
 637     }
 638
 639     return rdma_backend_modify_srq(&srq->backend_srq, srq_attr,
 640                                    srq_attr_mask);
 641 }
 642
 643 void rdma_rm_dealloc_srq(RdmaDeviceResources *dev_res, uint32_t srq_handle)
 644 {
 645     RdmaRmSRQ *srq;
 646
 647     srq = rdma_rm_get_srq(dev_res, srq_handle);
 648     if (!srq) {
 649         return;
 650     }
 651
 652     rdma_backend_destroy_srq(&srq->backend_srq, dev_res);
 653     rdma_res_tbl_dealloc(&dev_res->srq_tbl, srq_handle);
 654 }
 655
 656 void *rdma_rm_get_cqe_ctx(RdmaDeviceResources *dev_res, uint32_t cqe_ctx_id)
 657 {
 658     void **cqe_ctx;
 659
 660     cqe_ctx = rdma_res_tbl_get(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
 661     if (!cqe_ctx) {
 662         return NULL;
 663     }
 664
 665     return *cqe_ctx;
 666 }
 667
 668 int rdma_rm_alloc_cqe_ctx(RdmaDeviceResources *dev_res, uint32_t *cqe_ctx_id,
 669                           void *ctx)
 670 {
 671     void **cqe_ctx;
 672
 673     cqe_ctx = rdma_res_tbl_alloc(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
 674     if (!cqe_ctx) {
 675         return -ENOMEM;
 676     }
 677
 678     *cqe_ctx = ctx;
 679
 680     return 0;
 681 }
 682
 683 void rdma_rm_dealloc_cqe_ctx(RdmaDeviceResources *dev_res, uint32_t cqe_ctx_id)
 684 {
 685     rdma_res_tbl_dealloc(&dev_res->cqe_ctx_tbl, cqe_ctx_id);
 686 }
 687
 688 int rdma_rm_add_gid(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 689                     const char *ifname, union ibv_gid *gid, int gid_idx)
 690 {
 691     int rc;
 692
 693     rc = rdma_backend_add_gid(backend_dev, ifname, gid);
 694     if (rc) {
 695         return -EINVAL;
 696     }
 697
 698     memcpy(&dev_res->port.gid_tbl[gid_idx].gid, gid, sizeof(*gid));
 699
 700     return 0;
 701 }
 702
 703 int rdma_rm_del_gid(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 704                     const char *ifname, int gid_idx)
 705 {
 706     int rc;
 707
 708     if (!dev_res->port.gid_tbl[gid_idx].gid.global.interface_id) {
 709         return 0;
 710     }
 711
 712     rc = rdma_backend_del_gid(backend_dev, ifname,
 713                               &dev_res->port.gid_tbl[gid_idx].gid);
 714     if (rc) {
 715         return -EINVAL;
 716     }
 717
 718     memset(dev_res->port.gid_tbl[gid_idx].gid.raw, 0,
 719            sizeof(dev_res->port.gid_tbl[gid_idx].gid));
 720     dev_res->port.gid_tbl[gid_idx].backend_gid_index = -1;
 721
 722     return 0;
 723 }
 724
 725 int rdma_rm_get_backend_gid_index(RdmaDeviceResources *dev_res,
 726                                   RdmaBackendDev *backend_dev, int sgid_idx)
 727 {
 728     if (unlikely(sgid_idx < 0 || sgid_idx >= MAX_PORT_GIDS)) {
 729         rdma_error_report("Got invalid sgid_idx %d", sgid_idx);
 730         return -EINVAL;
 731     }
 732
 733     if (unlikely(dev_res->port.gid_tbl[sgid_idx].backend_gid_index == -1)) {
 734         dev_res->port.gid_tbl[sgid_idx].backend_gid_index =
 735         rdma_backend_get_gid_index(backend_dev,
 736                                    &dev_res->port.gid_tbl[sgid_idx].gid);
 737     }
 738
 739     return dev_res->port.gid_tbl[sgid_idx].backend_gid_index;
 740 }
 741
 742 static void destroy_qp_hash_key(gpointer data)
 743 {
 744     g_bytes_unref(data);
 745 }
 746
 747 static void init_ports(RdmaDeviceResources *dev_res)
 748 {
 749     int i;
 750
 751     memset(&dev_res->port, 0, sizeof(dev_res->port));
 752
 753     dev_res->port.state = IBV_PORT_DOWN;
 754     for (i = 0; i < MAX_PORT_GIDS; i++) {
 755         dev_res->port.gid_tbl[i].backend_gid_index = -1;
 756     }
 757 }
 758
 759 static void fini_ports(RdmaDeviceResources *dev_res,
 760                        RdmaBackendDev *backend_dev, const char *ifname)
 761 {
 762     int i;
 763
 764     dev_res->port.state = IBV_PORT_DOWN;
 765     for (i = 0; i < MAX_PORT_GIDS; i++) {
 766         rdma_rm_del_gid(dev_res, backend_dev, ifname, i);
 767     }
 768 }
 769
 770 int rdma_rm_init(RdmaDeviceResources *dev_res, struct ibv_device_attr *dev_attr)
 771 {
 772     dev_res->qp_hash = g_hash_table_new_full(g_bytes_hash, g_bytes_equal,
 773                                              destroy_qp_hash_key, NULL);
 774     if (!dev_res->qp_hash) {
 775         return -ENOMEM;
 776     }
 777
 778     res_tbl_init("PD", &dev_res->pd_tbl, dev_attr->max_pd, sizeof(RdmaRmPD));
 779     res_tbl_init("CQ", &dev_res->cq_tbl, dev_attr->max_cq, sizeof(RdmaRmCQ));
 780     res_tbl_init("MR", &dev_res->mr_tbl, dev_attr->max_mr, sizeof(RdmaRmMR));
 781     res_tbl_init("QP", &dev_res->qp_tbl, dev_attr->max_qp, sizeof(RdmaRmQP));
 782     res_tbl_init("CQE_CTX", &dev_res->cqe_ctx_tbl, dev_attr->max_qp *
 783                        dev_attr->max_qp_wr, sizeof(void *));
 784     res_tbl_init("UC", &dev_res->uc_tbl, MAX_UCS, sizeof(RdmaRmUC));
 785     res_tbl_init("SRQ", &dev_res->srq_tbl, dev_attr->max_srq,
 786                  sizeof(RdmaRmSRQ));
 787
 788     init_ports(dev_res);
 789
 790     qemu_mutex_init(&dev_res->lock);
 791
 792     memset(&dev_res->stats, 0, sizeof(dev_res->stats));
 793     atomic_set(&dev_res->stats.missing_cqe, 0);
 794
 795     return 0;
 796 }
 797
 798 void rdma_rm_fini(RdmaDeviceResources *dev_res, RdmaBackendDev *backend_dev,
 799                   const char *ifname)
 800 {
 801     qemu_mutex_destroy(&dev_res->lock);
 802
 803     fini_ports(dev_res, backend_dev, ifname);
 804
 805     res_tbl_free(&dev_res->srq_tbl);
 806     res_tbl_free(&dev_res->uc_tbl);
 807     res_tbl_free(&dev_res->cqe_ctx_tbl);
 808     res_tbl_free(&dev_res->qp_tbl);
 809     res_tbl_free(&dev_res->mr_tbl);
 810     res_tbl_free(&dev_res->cq_tbl);
 811     res_tbl_free(&dev_res->pd_tbl);
 812
 813     if (dev_res->qp_hash) {
 814         g_hash_table_destroy(dev_res->qp_hash);
 815     }
 816 }