ceph/src/spdk/dpdk/drivers/vdpa/mlx5/mlx5_vdpa_virtq.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2019 Mellanox Technologies, Ltd
   3  */
   4 #include <string.h>
   5 #include <unistd.h>
   6 #include <sys/mman.h>
   7
   8 #include <rte_malloc.h>
   9 #include <rte_errno.h>
  10 #include <rte_io.h>
  11
  12 #include <mlx5_common.h>
  13
  14 #include "mlx5_vdpa_utils.h"
  15 #include "mlx5_vdpa.h"
  16
  17
  18 static void
  19 mlx5_vdpa_virtq_handler(void *cb_arg)
  20 {
  21         struct mlx5_vdpa_virtq *virtq = cb_arg;
  22         struct mlx5_vdpa_priv *priv = virtq->priv;
  23         uint64_t buf;
  24         int nbytes;
  25
  26         do {
  27                 nbytes = read(virtq->intr_handle.fd, &buf, 8);
  28                 if (nbytes < 0) {
  29                         if (errno == EINTR ||
  30                             errno == EWOULDBLOCK ||
  31                             errno == EAGAIN)
  32                                 continue;
  33                         DRV_LOG(ERR,  "Failed to read kickfd of virtq %d: %s",
  34                                 virtq->index, strerror(errno));
  35                 }
  36                 break;
  37         } while (1);
  38         rte_write32(virtq->index, priv->virtq_db_addr);
  39         DRV_LOG(DEBUG, "Ring virtq %u doorbell.", virtq->index);
  40 }
  41
  42 static int
  43 mlx5_vdpa_virtq_unset(struct mlx5_vdpa_virtq *virtq)
  44 {
  45         unsigned int i;
  46         int retries = MLX5_VDPA_INTR_RETRIES;
  47         int ret = -EAGAIN;
  48
  49         if (virtq->intr_handle.fd != -1) {
  50                 while (retries-- && ret == -EAGAIN) {
  51                         ret = rte_intr_callback_unregister(&virtq->intr_handle,
  52                                                         mlx5_vdpa_virtq_handler,
  53                                                         virtq);
  54                         if (ret == -EAGAIN) {
  55                                 DRV_LOG(DEBUG, "Try again to unregister fd %d "
  56                                         "of virtq %d interrupt, retries = %d.",
  57                                         virtq->intr_handle.fd,
  58                                         (int)virtq->index, retries);
  59                                 usleep(MLX5_VDPA_INTR_RETRIES_USEC);
  60                         }
  61                 }
  62                 virtq->intr_handle.fd = -1;
  63         }
  64         if (virtq->virtq)
  65                 claim_zero(mlx5_devx_cmd_destroy(virtq->virtq));
  66         virtq->virtq = NULL;
  67         for (i = 0; i < RTE_DIM(virtq->umems); ++i) {
  68                 if (virtq->umems[i].obj)
  69                         claim_zero(mlx5_glue->devx_umem_dereg
  70                                                          (virtq->umems[i].obj));
  71                 if (virtq->umems[i].buf)
  72                         rte_free(virtq->umems[i].buf);
  73         }
  74         memset(&virtq->umems, 0, sizeof(virtq->umems));
  75         if (virtq->eqp.fw_qp)
  76                 mlx5_vdpa_event_qp_destroy(&virtq->eqp);
  77         return 0;
  78 }
  79
  80 void
  81 mlx5_vdpa_virtqs_release(struct mlx5_vdpa_priv *priv)
  82 {
  83         int i;
  84
  85         for (i = 0; i < priv->nr_virtqs; i++) {
  86                 mlx5_vdpa_virtq_unset(&priv->virtqs[i]);
  87                 priv->virtqs[i].enable = 0;
  88         }
  89         if (priv->tis) {
  90                 claim_zero(mlx5_devx_cmd_destroy(priv->tis));
  91                 priv->tis = NULL;
  92         }
  93         if (priv->td) {
  94                 claim_zero(mlx5_devx_cmd_destroy(priv->td));
  95                 priv->td = NULL;
  96         }
  97         if (priv->virtq_db_addr) {
  98                 claim_zero(munmap(priv->virtq_db_addr, priv->var->length));
  99                 priv->virtq_db_addr = NULL;
 100         }
 101         priv->features = 0;
 102         priv->nr_virtqs = 0;
 103 }
 104
 105 int
 106 mlx5_vdpa_virtq_modify(struct mlx5_vdpa_virtq *virtq, int state)
 107 {
 108         struct mlx5_devx_virtq_attr attr = {
 109                         .type = MLX5_VIRTQ_MODIFY_TYPE_STATE,
 110                         .state = state ? MLX5_VIRTQ_STATE_RDY :
 111                                          MLX5_VIRTQ_STATE_SUSPEND,
 112                         .queue_index = virtq->index,
 113         };
 114
 115         return mlx5_devx_cmd_modify_virtq(virtq->virtq, &attr);
 116 }
 117
 118 int
 119 mlx5_vdpa_virtq_stop(struct mlx5_vdpa_priv *priv, int index)
 120 {
 121         struct mlx5_devx_virtq_attr attr = {0};
 122         struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index];
 123         int ret = mlx5_vdpa_virtq_modify(virtq, 0);
 124
 125         if (ret)
 126                 return -1;
 127         if (mlx5_devx_cmd_query_virtq(virtq->virtq, &attr)) {
 128                 DRV_LOG(ERR, "Failed to query virtq %d.", index);
 129                 return -1;
 130         }
 131         DRV_LOG(INFO, "Query vid %d vring %d: hw_available_idx=%d, "
 132                 "hw_used_index=%d", priv->vid, index,
 133                 attr.hw_available_index, attr.hw_used_index);
 134         ret = rte_vhost_set_vring_base(priv->vid, index,
 135                                        attr.hw_available_index,
 136                                        attr.hw_used_index);
 137         if (ret) {
 138                 DRV_LOG(ERR, "Failed to set virtq %d base.", index);
 139                 return -1;
 140         }
 141         return 0;
 142 }
 143
 144 static uint64_t
 145 mlx5_vdpa_hva_to_gpa(struct rte_vhost_memory *mem, uint64_t hva)
 146 {
 147         struct rte_vhost_mem_region *reg;
 148         uint32_t i;
 149         uint64_t gpa = 0;
 150
 151         for (i = 0; i < mem->nregions; i++) {
 152                 reg = &mem->regions[i];
 153                 if (hva >= reg->host_user_addr &&
 154                     hva < reg->host_user_addr + reg->size) {
 155                         gpa = hva - reg->host_user_addr + reg->guest_phys_addr;
 156                         break;
 157                 }
 158         }
 159         return gpa;
 160 }
 161
 162 static int
 163 mlx5_vdpa_virtq_setup(struct mlx5_vdpa_priv *priv, int index)
 164 {
 165         struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index];
 166         struct rte_vhost_vring vq;
 167         struct mlx5_devx_virtq_attr attr = {0};
 168         uint64_t gpa;
 169         int ret;
 170         unsigned int i;
 171         uint16_t last_avail_idx;
 172         uint16_t last_used_idx;
 173
 174         ret = rte_vhost_get_vhost_vring(priv->vid, index, &vq);
 175         if (ret)
 176                 return -1;
 177         virtq->index = index;
 178         virtq->vq_size = vq.size;
 179         attr.tso_ipv4 = !!(priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO4));
 180         attr.tso_ipv6 = !!(priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO6));
 181         attr.tx_csum = !!(priv->features & (1ULL << VIRTIO_NET_F_CSUM));
 182         attr.rx_csum = !!(priv->features & (1ULL << VIRTIO_NET_F_GUEST_CSUM));
 183         attr.virtio_version_1_0 = !!(priv->features & (1ULL <<
 184                                                         VIRTIO_F_VERSION_1));
 185         attr.type = (priv->features & (1ULL << VIRTIO_F_RING_PACKED)) ?
 186                         MLX5_VIRTQ_TYPE_PACKED : MLX5_VIRTQ_TYPE_SPLIT;
 187         /*
 188          * No need event QPs creation when the guest in poll mode or when the
 189          * capability allows it.
 190          */
 191         attr.event_mode = vq.callfd != -1 || !(priv->caps.event_mode & (1 <<
 192                                                MLX5_VIRTQ_EVENT_MODE_NO_MSIX)) ?
 193                                                       MLX5_VIRTQ_EVENT_MODE_QP :
 194                                                   MLX5_VIRTQ_EVENT_MODE_NO_MSIX;
 195         if (attr.event_mode == MLX5_VIRTQ_EVENT_MODE_QP) {
 196                 ret = mlx5_vdpa_event_qp_create(priv, vq.size, vq.callfd,
 197                                                 &virtq->eqp);
 198                 if (ret) {
 199                         DRV_LOG(ERR, "Failed to create event QPs for virtq %d.",
 200                                 index);
 201                         return -1;
 202                 }
 203                 attr.qp_id = virtq->eqp.fw_qp->id;
 204         } else {
 205                 DRV_LOG(INFO, "Virtq %d is, for sure, working by poll mode, no"
 206                         " need event QPs and event mechanism.", index);
 207         }
 208         /* Setup 3 UMEMs for each virtq. */
 209         for (i = 0; i < RTE_DIM(virtq->umems); ++i) {
 210                 virtq->umems[i].size = priv->caps.umems[i].a * vq.size +
 211                                                           priv->caps.umems[i].b;
 212                 virtq->umems[i].buf = rte_zmalloc(__func__,
 213                                                   virtq->umems[i].size, 4096);
 214                 if (!virtq->umems[i].buf) {
 215                         DRV_LOG(ERR, "Cannot allocate umem %d memory for virtq"
 216                                 " %u.", i, index);
 217                         goto error;
 218                 }
 219                 virtq->umems[i].obj = mlx5_glue->devx_umem_reg(priv->ctx,
 220                                                         virtq->umems[i].buf,
 221                                                         virtq->umems[i].size,
 222                                                         IBV_ACCESS_LOCAL_WRITE);
 223                 if (!virtq->umems[i].obj) {
 224                         DRV_LOG(ERR, "Failed to register umem %d for virtq %u.",
 225                                 i, index);
 226                         goto error;
 227                 }
 228                 attr.umems[i].id = virtq->umems[i].obj->umem_id;
 229                 attr.umems[i].offset = 0;
 230                 attr.umems[i].size = virtq->umems[i].size;
 231         }
 232         if (attr.type == MLX5_VIRTQ_TYPE_SPLIT) {
 233                 gpa = mlx5_vdpa_hva_to_gpa(priv->vmem,
 234                                            (uint64_t)(uintptr_t)vq.desc);
 235                 if (!gpa) {
 236                         DRV_LOG(ERR, "Failed to get descriptor ring GPA.");
 237                         goto error;
 238                 }
 239                 attr.desc_addr = gpa;
 240                 gpa = mlx5_vdpa_hva_to_gpa(priv->vmem,
 241                                            (uint64_t)(uintptr_t)vq.used);
 242                 if (!gpa) {
 243                         DRV_LOG(ERR, "Failed to get GPA for used ring.");
 244                         goto error;
 245                 }
 246                 attr.used_addr = gpa;
 247                 gpa = mlx5_vdpa_hva_to_gpa(priv->vmem,
 248                                            (uint64_t)(uintptr_t)vq.avail);
 249                 if (!gpa) {
 250                         DRV_LOG(ERR, "Failed to get GPA for available ring.");
 251                         goto error;
 252                 }
 253                 attr.available_addr = gpa;
 254         }
 255         ret = rte_vhost_get_vring_base(priv->vid, index, &last_avail_idx,
 256                                  &last_used_idx);
 257         if (ret) {
 258                 last_avail_idx = 0;
 259                 last_used_idx = 0;
 260                 DRV_LOG(WARNING, "Couldn't get vring base, idx are set to 0");
 261         } else {
 262                 DRV_LOG(INFO, "vid %d: Init last_avail_idx=%d, last_used_idx=%d for "
 263                                 "virtq %d.", priv->vid, last_avail_idx,
 264                                 last_used_idx, index);
 265         }
 266         attr.hw_available_index = last_avail_idx;
 267         attr.hw_used_index = last_used_idx;
 268         attr.q_size = vq.size;
 269         attr.mkey = priv->gpa_mkey_index;
 270         attr.tis_id = priv->tis->id;
 271         attr.queue_index = index;
 272         virtq->virtq = mlx5_devx_cmd_create_virtq(priv->ctx, &attr);
 273         virtq->priv = priv;
 274         if (!virtq->virtq)
 275                 goto error;
 276         if (mlx5_vdpa_virtq_modify(virtq, 1))
 277                 goto error;
 278         virtq->priv = priv;
 279         rte_write32(virtq->index, priv->virtq_db_addr);
 280         /* Setup doorbell mapping. */
 281         virtq->intr_handle.fd = vq.kickfd;
 282         if (virtq->intr_handle.fd == -1) {
 283                 DRV_LOG(WARNING, "Virtq %d kickfd is invalid.", index);
 284                 if (!priv->direct_notifier) {
 285                         DRV_LOG(ERR, "Virtq %d cannot be notified.", index);
 286                         goto error;
 287                 }
 288         } else {
 289                 virtq->intr_handle.type = RTE_INTR_HANDLE_EXT;
 290                 if (rte_intr_callback_register(&virtq->intr_handle,
 291                                                mlx5_vdpa_virtq_handler,
 292                                                virtq)) {
 293                         virtq->intr_handle.fd = -1;
 294                         DRV_LOG(ERR, "Failed to register virtq %d interrupt.",
 295                                 index);
 296                         goto error;
 297                 } else {
 298                         DRV_LOG(DEBUG, "Register fd %d interrupt for virtq %d.",
 299                                 virtq->intr_handle.fd, index);
 300                 }
 301         }
 302         return 0;
 303 error:
 304         mlx5_vdpa_virtq_unset(virtq);
 305         return -1;
 306 }
 307
 308 static int
 309 mlx5_vdpa_features_validate(struct mlx5_vdpa_priv *priv)
 310 {
 311         if (priv->features & (1ULL << VIRTIO_F_RING_PACKED)) {
 312                 if (!(priv->caps.virtio_queue_type & (1 <<
 313                                                      MLX5_VIRTQ_TYPE_PACKED))) {
 314                         DRV_LOG(ERR, "Failed to configur PACKED mode for vdev "
 315                                 "%d - it was not reported by HW/driver"
 316                                 " capability.", priv->vid);
 317                         return -ENOTSUP;
 318                 }
 319         }
 320         if (priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO4)) {
 321                 if (!priv->caps.tso_ipv4) {
 322                         DRV_LOG(ERR, "Failed to enable TSO4 for vdev %d - TSO4"
 323                                 " was not reported by HW/driver capability.",
 324                                 priv->vid);
 325                         return -ENOTSUP;
 326                 }
 327         }
 328         if (priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO6)) {
 329                 if (!priv->caps.tso_ipv6) {
 330                         DRV_LOG(ERR, "Failed to enable TSO6 for vdev %d - TSO6"
 331                                 " was not reported by HW/driver capability.",
 332                                 priv->vid);
 333                         return -ENOTSUP;
 334                 }
 335         }
 336         if (priv->features & (1ULL << VIRTIO_NET_F_CSUM)) {
 337                 if (!priv->caps.tx_csum) {
 338                         DRV_LOG(ERR, "Failed to enable CSUM for vdev %d - CSUM"
 339                                 " was not reported by HW/driver capability.",
 340                                 priv->vid);
 341                         return -ENOTSUP;
 342                 }
 343         }
 344         if (priv->features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
 345                 if (!priv->caps.rx_csum) {
 346                         DRV_LOG(ERR, "Failed to enable GUEST CSUM for vdev %d"
 347                                 " GUEST CSUM was not reported by HW/driver "
 348                                 "capability.", priv->vid);
 349                         return -ENOTSUP;
 350                 }
 351         }
 352         if (priv->features & (1ULL << VIRTIO_F_VERSION_1)) {
 353                 if (!priv->caps.virtio_version_1_0) {
 354                         DRV_LOG(ERR, "Failed to enable version 1 for vdev %d "
 355                                 "version 1 was not reported by HW/driver"
 356                                 " capability.", priv->vid);
 357                         return -ENOTSUP;
 358                 }
 359         }
 360         return 0;
 361 }
 362
 363 int
 364 mlx5_vdpa_virtqs_prepare(struct mlx5_vdpa_priv *priv)
 365 {
 366         struct mlx5_devx_tis_attr tis_attr = {0};
 367         uint32_t i;
 368         uint16_t nr_vring = rte_vhost_get_vring_num(priv->vid);
 369         int ret = rte_vhost_get_negotiated_features(priv->vid, &priv->features);
 370
 371         if (ret || mlx5_vdpa_features_validate(priv)) {
 372                 DRV_LOG(ERR, "Failed to configure negotiated features.");
 373                 return -1;
 374         }
 375         if (nr_vring > priv->caps.max_num_virtio_queues * 2) {
 376                 DRV_LOG(ERR, "Do not support more than %d virtqs(%d).",
 377                         (int)priv->caps.max_num_virtio_queues * 2,
 378                         (int)nr_vring);
 379                 return -1;
 380         }
 381         /* Always map the entire page. */
 382         priv->virtq_db_addr = mmap(NULL, priv->var->length, PROT_READ |
 383                                    PROT_WRITE, MAP_SHARED, priv->ctx->cmd_fd,
 384                                    priv->var->mmap_off);
 385         if (priv->virtq_db_addr == MAP_FAILED) {
 386                 DRV_LOG(ERR, "Failed to map doorbell page %u.", errno);
 387                 priv->virtq_db_addr = NULL;
 388                 goto error;
 389         } else {
 390                 DRV_LOG(DEBUG, "VAR address of doorbell mapping is %p.",
 391                         priv->virtq_db_addr);
 392         }
 393         priv->td = mlx5_devx_cmd_create_td(priv->ctx);
 394         if (!priv->td) {
 395                 DRV_LOG(ERR, "Failed to create transport domain.");
 396                 return -rte_errno;
 397         }
 398         tis_attr.transport_domain = priv->td->id;
 399         priv->tis = mlx5_devx_cmd_create_tis(priv->ctx, &tis_attr);
 400         if (!priv->tis) {
 401                 DRV_LOG(ERR, "Failed to create TIS.");
 402                 goto error;
 403         }
 404         priv->nr_virtqs = nr_vring;
 405         for (i = 0; i < nr_vring; i++) {
 406                 claim_zero(rte_vhost_enable_guest_notification(priv->vid, i,
 407                                                                1));
 408                 if (mlx5_vdpa_virtq_setup(priv, i))
 409                         goto error;
 410         }
 411         return 0;
 412 error:
 413         mlx5_vdpa_virtqs_release(priv);
 414         return -1;
 415 }
 416
 417 int
 418 mlx5_vdpa_virtq_enable(struct mlx5_vdpa_priv *priv, int index, int enable)
 419 {
 420         struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index];
 421         int ret;
 422
 423         DRV_LOG(INFO, "Update virtq %d status %sable -> %sable.", index,
 424                 virtq->enable ? "en" : "dis", enable ? "en" : "dis");
 425         if (virtq->enable == !!enable)
 426                 return 0;
 427         if (!priv->configured) {
 428                 virtq->enable = !!enable;
 429                 return 0;
 430         }
 431         if (enable) {
 432                 /* Configuration might have been updated - reconfigure virtq. */
 433                 if (virtq->virtq) {
 434                         ret = mlx5_vdpa_virtq_stop(priv, index);
 435                         if (ret)
 436                                 DRV_LOG(WARNING, "Failed to stop virtq %d.",
 437                                         index);
 438                         mlx5_vdpa_virtq_unset(virtq);
 439                 }
 440                 ret = mlx5_vdpa_virtq_setup(priv, index);
 441                 if (ret) {
 442                         DRV_LOG(ERR, "Failed to setup virtq %d.", index);
 443                         return ret;
 444                         /* The only case virtq can stay invalid. */
 445                 }
 446         }
 447         virtq->enable = !!enable;
 448         if (is_virtq_recvq(virtq->index, priv->nr_virtqs)) {
 449                 /* Need to add received virtq to the RQT table of the TIRs. */
 450                 ret = mlx5_vdpa_steer_update(priv);
 451                 if (ret) {
 452                         virtq->enable = !enable;
 453                         return ret;
 454                 }
 455         }
 456         return 0;
 457 }