1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2019 Mellanox Technologies, Ltd
8 #include <rte_malloc.h>
12 #include <mlx5_common.h>
14 #include "mlx5_vdpa_utils.h"
15 #include "mlx5_vdpa.h"
19 mlx5_vdpa_virtq_handler(void *cb_arg)
21 struct mlx5_vdpa_virtq *virtq = cb_arg;
22 struct mlx5_vdpa_priv *priv = virtq->priv;
27 nbytes = read(virtq->intr_handle.fd, &buf, 8);
30 errno == EWOULDBLOCK ||
33 DRV_LOG(ERR, "Failed to read kickfd of virtq %d: %s",
34 virtq->index, strerror(errno));
38 rte_write32(virtq->index, priv->virtq_db_addr);
39 if (virtq->notifier_state == MLX5_VDPA_NOTIFIER_STATE_DISABLED) {
40 if (rte_vhost_host_notifier_ctrl(priv->vid, virtq->index, true))
41 virtq->notifier_state = MLX5_VDPA_NOTIFIER_STATE_ERR;
43 virtq->notifier_state =
44 MLX5_VDPA_NOTIFIER_STATE_ENABLED;
45 DRV_LOG(INFO, "Virtq %u notifier state is %s.", virtq->index,
46 virtq->notifier_state ==
47 MLX5_VDPA_NOTIFIER_STATE_ENABLED ? "enabled" :
50 DRV_LOG(DEBUG, "Ring virtq %u doorbell.", virtq->index);
54 mlx5_vdpa_virtq_unset(struct mlx5_vdpa_virtq *virtq)
57 int retries = MLX5_VDPA_INTR_RETRIES;
60 if (virtq->intr_handle.fd != -1) {
61 while (retries-- && ret == -EAGAIN) {
62 ret = rte_intr_callback_unregister(&virtq->intr_handle,
63 mlx5_vdpa_virtq_handler,
66 DRV_LOG(DEBUG, "Try again to unregister fd %d "
67 "of virtq %d interrupt, retries = %d.",
68 virtq->intr_handle.fd,
69 (int)virtq->index, retries);
70 usleep(MLX5_VDPA_INTR_RETRIES_USEC);
73 virtq->intr_handle.fd = -1;
76 ret = mlx5_vdpa_virtq_stop(virtq->priv, virtq->index);
78 DRV_LOG(WARNING, "Failed to stop virtq %d.",
80 claim_zero(mlx5_devx_cmd_destroy(virtq->virtq));
83 for (i = 0; i < RTE_DIM(virtq->umems); ++i) {
84 if (virtq->umems[i].obj)
85 claim_zero(mlx5_glue->devx_umem_dereg
86 (virtq->umems[i].obj));
87 if (virtq->umems[i].buf)
88 rte_free(virtq->umems[i].buf);
90 memset(&virtq->umems, 0, sizeof(virtq->umems));
92 mlx5_vdpa_event_qp_destroy(&virtq->eqp);
93 virtq->notifier_state = MLX5_VDPA_NOTIFIER_STATE_DISABLED;
98 mlx5_vdpa_virtqs_release(struct mlx5_vdpa_priv *priv)
101 struct mlx5_vdpa_virtq *virtq;
103 for (i = 0; i < priv->nr_virtqs; i++) {
104 virtq = &priv->virtqs[i];
105 mlx5_vdpa_virtq_unset(virtq);
106 if (virtq->counters) {
107 claim_zero(mlx5_devx_cmd_destroy(virtq->counters));
108 virtq->counters = NULL;
109 memset(&virtq->reset, 0, sizeof(virtq->reset));
111 memset(virtq->err_time, 0, sizeof(virtq->err_time));
115 claim_zero(mlx5_devx_cmd_destroy(priv->tis));
119 claim_zero(mlx5_devx_cmd_destroy(priv->td));
122 if (priv->virtq_db_addr) {
123 claim_zero(munmap(priv->virtq_db_addr, priv->var->length));
124 priv->virtq_db_addr = NULL;
131 mlx5_vdpa_virtq_modify(struct mlx5_vdpa_virtq *virtq, int state)
133 struct mlx5_devx_virtq_attr attr = {
134 .type = MLX5_VIRTQ_MODIFY_TYPE_STATE,
135 .state = state ? MLX5_VIRTQ_STATE_RDY :
136 MLX5_VIRTQ_STATE_SUSPEND,
137 .queue_index = virtq->index,
140 return mlx5_devx_cmd_modify_virtq(virtq->virtq, &attr);
144 mlx5_vdpa_virtq_stop(struct mlx5_vdpa_priv *priv, int index)
146 struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index];
151 ret = mlx5_vdpa_virtq_modify(virtq, 0);
154 virtq->stopped = true;
155 DRV_LOG(DEBUG, "vid %u virtq %u was stopped.", priv->vid, index);
156 return mlx5_vdpa_virtq_query(priv, index);
160 mlx5_vdpa_virtq_query(struct mlx5_vdpa_priv *priv, int index)
162 struct mlx5_devx_virtq_attr attr = {0};
163 struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index];
166 if (mlx5_devx_cmd_query_virtq(virtq->virtq, &attr)) {
167 DRV_LOG(ERR, "Failed to query virtq %d.", index);
170 DRV_LOG(INFO, "Query vid %d vring %d: hw_available_idx=%d, "
171 "hw_used_index=%d", priv->vid, index,
172 attr.hw_available_index, attr.hw_used_index);
173 ret = rte_vhost_set_vring_base(priv->vid, index,
174 attr.hw_available_index,
177 DRV_LOG(ERR, "Failed to set virtq %d base.", index);
180 if (attr.state == MLX5_VIRTQ_STATE_ERROR)
181 DRV_LOG(WARNING, "vid %d vring %d hw error=%hhu",
182 priv->vid, index, attr.error_type);
187 mlx5_vdpa_hva_to_gpa(struct rte_vhost_memory *mem, uint64_t hva)
189 struct rte_vhost_mem_region *reg;
193 for (i = 0; i < mem->nregions; i++) {
194 reg = &mem->regions[i];
195 if (hva >= reg->host_user_addr &&
196 hva < reg->host_user_addr + reg->size) {
197 gpa = hva - reg->host_user_addr + reg->guest_phys_addr;
205 mlx5_vdpa_virtq_setup(struct mlx5_vdpa_priv *priv, int index)
207 struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index];
208 struct rte_vhost_vring vq;
209 struct mlx5_devx_virtq_attr attr = {0};
213 uint16_t last_avail_idx;
214 uint16_t last_used_idx;
215 uint16_t event_num = MLX5_EVENT_TYPE_OBJECT_CHANGE;
218 ret = rte_vhost_get_vhost_vring(priv->vid, index, &vq);
221 virtq->index = index;
222 virtq->vq_size = vq.size;
223 attr.tso_ipv4 = !!(priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO4));
224 attr.tso_ipv6 = !!(priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO6));
225 attr.tx_csum = !!(priv->features & (1ULL << VIRTIO_NET_F_CSUM));
226 attr.rx_csum = !!(priv->features & (1ULL << VIRTIO_NET_F_GUEST_CSUM));
227 attr.virtio_version_1_0 = !!(priv->features & (1ULL <<
228 VIRTIO_F_VERSION_1));
229 attr.type = (priv->features & (1ULL << VIRTIO_F_RING_PACKED)) ?
230 MLX5_VIRTQ_TYPE_PACKED : MLX5_VIRTQ_TYPE_SPLIT;
232 * No need event QPs creation when the guest in poll mode or when the
233 * capability allows it.
235 attr.event_mode = vq.callfd != -1 || !(priv->caps.event_mode & (1 <<
236 MLX5_VIRTQ_EVENT_MODE_NO_MSIX)) ?
237 MLX5_VIRTQ_EVENT_MODE_QP :
238 MLX5_VIRTQ_EVENT_MODE_NO_MSIX;
239 if (attr.event_mode == MLX5_VIRTQ_EVENT_MODE_QP) {
240 ret = mlx5_vdpa_event_qp_create(priv, vq.size, vq.callfd,
243 DRV_LOG(ERR, "Failed to create event QPs for virtq %d.",
247 attr.qp_id = virtq->eqp.fw_qp->id;
249 DRV_LOG(INFO, "Virtq %d is, for sure, working by poll mode, no"
250 " need event QPs and event mechanism.", index);
252 if (priv->caps.queue_counters_valid) {
253 if (!virtq->counters)
254 virtq->counters = mlx5_devx_cmd_create_virtio_q_counters
256 if (!virtq->counters) {
257 DRV_LOG(ERR, "Failed to create virtq couners for virtq"
261 attr.counters_obj_id = virtq->counters->id;
263 /* Setup 3 UMEMs for each virtq. */
264 for (i = 0; i < RTE_DIM(virtq->umems); ++i) {
265 virtq->umems[i].size = priv->caps.umems[i].a * vq.size +
266 priv->caps.umems[i].b;
267 virtq->umems[i].buf = rte_zmalloc(__func__,
268 virtq->umems[i].size, 4096);
269 if (!virtq->umems[i].buf) {
270 DRV_LOG(ERR, "Cannot allocate umem %d memory for virtq"
274 virtq->umems[i].obj = mlx5_glue->devx_umem_reg(priv->ctx,
276 virtq->umems[i].size,
277 IBV_ACCESS_LOCAL_WRITE);
278 if (!virtq->umems[i].obj) {
279 DRV_LOG(ERR, "Failed to register umem %d for virtq %u.",
283 attr.umems[i].id = virtq->umems[i].obj->umem_id;
284 attr.umems[i].offset = 0;
285 attr.umems[i].size = virtq->umems[i].size;
287 if (attr.type == MLX5_VIRTQ_TYPE_SPLIT) {
288 gpa = mlx5_vdpa_hva_to_gpa(priv->vmem,
289 (uint64_t)(uintptr_t)vq.desc);
291 DRV_LOG(ERR, "Failed to get descriptor ring GPA.");
294 attr.desc_addr = gpa;
295 gpa = mlx5_vdpa_hva_to_gpa(priv->vmem,
296 (uint64_t)(uintptr_t)vq.used);
298 DRV_LOG(ERR, "Failed to get GPA for used ring.");
301 attr.used_addr = gpa;
302 gpa = mlx5_vdpa_hva_to_gpa(priv->vmem,
303 (uint64_t)(uintptr_t)vq.avail);
305 DRV_LOG(ERR, "Failed to get GPA for available ring.");
308 attr.available_addr = gpa;
310 ret = rte_vhost_get_vring_base(priv->vid, index, &last_avail_idx,
315 DRV_LOG(WARNING, "Couldn't get vring base, idx are set to 0");
317 DRV_LOG(INFO, "vid %d: Init last_avail_idx=%d, last_used_idx=%d for "
318 "virtq %d.", priv->vid, last_avail_idx,
319 last_used_idx, index);
321 attr.hw_available_index = last_avail_idx;
322 attr.hw_used_index = last_used_idx;
323 attr.q_size = vq.size;
324 attr.mkey = priv->gpa_mkey_index;
325 attr.tis_id = priv->tis->id;
326 attr.queue_index = index;
328 virtq->virtq = mlx5_devx_cmd_create_virtq(priv->ctx, &attr);
332 claim_zero(rte_vhost_enable_guest_notification(priv->vid, index, 1));
333 if (mlx5_vdpa_virtq_modify(virtq, 1))
336 rte_write32(virtq->index, priv->virtq_db_addr);
337 /* Setup doorbell mapping. */
338 virtq->intr_handle.fd = vq.kickfd;
339 if (virtq->intr_handle.fd == -1) {
340 DRV_LOG(WARNING, "Virtq %d kickfd is invalid.", index);
342 virtq->intr_handle.type = RTE_INTR_HANDLE_EXT;
343 if (rte_intr_callback_register(&virtq->intr_handle,
344 mlx5_vdpa_virtq_handler,
346 virtq->intr_handle.fd = -1;
347 DRV_LOG(ERR, "Failed to register virtq %d interrupt.",
351 DRV_LOG(DEBUG, "Register fd %d interrupt for virtq %d.",
352 virtq->intr_handle.fd, index);
355 /* Subscribe virtq error event. */
357 cookie = ((uint64_t)virtq->version << 32) + index;
358 ret = mlx5_glue->devx_subscribe_devx_event(priv->err_chnl,
363 DRV_LOG(ERR, "Failed to subscribe device %d virtq %d error event.",
368 virtq->stopped = false;
369 DRV_LOG(DEBUG, "vid %u virtq %u was created successfully.", priv->vid,
373 mlx5_vdpa_virtq_unset(virtq);
378 mlx5_vdpa_features_validate(struct mlx5_vdpa_priv *priv)
380 if (priv->features & (1ULL << VIRTIO_F_RING_PACKED)) {
381 if (!(priv->caps.virtio_queue_type & (1 <<
382 MLX5_VIRTQ_TYPE_PACKED))) {
383 DRV_LOG(ERR, "Failed to configur PACKED mode for vdev "
384 "%d - it was not reported by HW/driver"
385 " capability.", priv->vid);
389 if (priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO4)) {
390 if (!priv->caps.tso_ipv4) {
391 DRV_LOG(ERR, "Failed to enable TSO4 for vdev %d - TSO4"
392 " was not reported by HW/driver capability.",
397 if (priv->features & (1ULL << VIRTIO_NET_F_HOST_TSO6)) {
398 if (!priv->caps.tso_ipv6) {
399 DRV_LOG(ERR, "Failed to enable TSO6 for vdev %d - TSO6"
400 " was not reported by HW/driver capability.",
405 if (priv->features & (1ULL << VIRTIO_NET_F_CSUM)) {
406 if (!priv->caps.tx_csum) {
407 DRV_LOG(ERR, "Failed to enable CSUM for vdev %d - CSUM"
408 " was not reported by HW/driver capability.",
413 if (priv->features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
414 if (!priv->caps.rx_csum) {
415 DRV_LOG(ERR, "Failed to enable GUEST CSUM for vdev %d"
416 " GUEST CSUM was not reported by HW/driver "
417 "capability.", priv->vid);
421 if (priv->features & (1ULL << VIRTIO_F_VERSION_1)) {
422 if (!priv->caps.virtio_version_1_0) {
423 DRV_LOG(ERR, "Failed to enable version 1 for vdev %d "
424 "version 1 was not reported by HW/driver"
425 " capability.", priv->vid);
433 mlx5_vdpa_virtqs_prepare(struct mlx5_vdpa_priv *priv)
435 struct mlx5_devx_tis_attr tis_attr = {0};
437 uint16_t nr_vring = rte_vhost_get_vring_num(priv->vid);
438 int ret = rte_vhost_get_negotiated_features(priv->vid, &priv->features);
440 if (ret || mlx5_vdpa_features_validate(priv)) {
441 DRV_LOG(ERR, "Failed to configure negotiated features.");
444 if (nr_vring > priv->caps.max_num_virtio_queues * 2) {
445 DRV_LOG(ERR, "Do not support more than %d virtqs(%d).",
446 (int)priv->caps.max_num_virtio_queues * 2,
450 /* Always map the entire page. */
451 priv->virtq_db_addr = mmap(NULL, priv->var->length, PROT_READ |
452 PROT_WRITE, MAP_SHARED, priv->ctx->cmd_fd,
453 priv->var->mmap_off);
454 if (priv->virtq_db_addr == MAP_FAILED) {
455 DRV_LOG(ERR, "Failed to map doorbell page %u.", errno);
456 priv->virtq_db_addr = NULL;
459 DRV_LOG(DEBUG, "VAR address of doorbell mapping is %p.",
460 priv->virtq_db_addr);
462 priv->td = mlx5_devx_cmd_create_td(priv->ctx);
464 DRV_LOG(ERR, "Failed to create transport domain.");
467 tis_attr.transport_domain = priv->td->id;
468 priv->tis = mlx5_devx_cmd_create_tis(priv->ctx, &tis_attr);
470 DRV_LOG(ERR, "Failed to create TIS.");
473 priv->nr_virtqs = nr_vring;
474 for (i = 0; i < nr_vring; i++)
475 if (priv->virtqs[i].enable && mlx5_vdpa_virtq_setup(priv, i))
479 mlx5_vdpa_virtqs_release(priv);
484 mlx5_vdpa_virtq_is_modified(struct mlx5_vdpa_priv *priv,
485 struct mlx5_vdpa_virtq *virtq)
487 struct rte_vhost_vring vq;
488 int ret = rte_vhost_get_vhost_vring(priv->vid, virtq->index, &vq);
492 if (vq.size != virtq->vq_size || vq.kickfd != virtq->intr_handle.fd)
494 if (virtq->eqp.cq.cq) {
495 if (vq.callfd != virtq->eqp.cq.callfd)
497 } else if (vq.callfd != -1) {
504 mlx5_vdpa_virtq_enable(struct mlx5_vdpa_priv *priv, int index, int enable)
506 struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index];
509 DRV_LOG(INFO, "Update virtq %d status %sable -> %sable.", index,
510 virtq->enable ? "en" : "dis", enable ? "en" : "dis");
511 if (!priv->configured) {
512 virtq->enable = !!enable;
515 if (virtq->enable == !!enable) {
518 ret = mlx5_vdpa_virtq_is_modified(priv, virtq);
520 DRV_LOG(ERR, "Virtq %d modify check failed.", index);
525 DRV_LOG(INFO, "Virtq %d was modified, recreate it.", index);
529 if (is_virtq_recvq(virtq->index, priv->nr_virtqs)) {
530 ret = mlx5_vdpa_steer_update(priv);
532 DRV_LOG(WARNING, "Failed to disable steering "
533 "for virtq %d.", index);
535 mlx5_vdpa_virtq_unset(virtq);
538 ret = mlx5_vdpa_virtq_setup(priv, index);
540 DRV_LOG(ERR, "Failed to setup virtq %d.", index);
544 if (is_virtq_recvq(virtq->index, priv->nr_virtqs)) {
545 ret = mlx5_vdpa_steer_update(priv);
547 DRV_LOG(WARNING, "Failed to enable steering "
548 "for virtq %d.", index);
555 mlx5_vdpa_virtq_stats_get(struct mlx5_vdpa_priv *priv, int qid,
556 struct rte_vdpa_stat *stats, unsigned int n)
558 struct mlx5_vdpa_virtq *virtq = &priv->virtqs[qid];
559 struct mlx5_devx_virtio_q_couners_attr attr = {0};
562 if (!virtq->counters) {
563 DRV_LOG(ERR, "Failed to read virtq %d statistics - virtq "
567 ret = mlx5_devx_cmd_query_virtio_q_counters(virtq->counters, &attr);
569 DRV_LOG(ERR, "Failed to read virtq %d stats from HW.", qid);
572 ret = (int)RTE_MIN(n, (unsigned int)MLX5_VDPA_STATS_MAX);
573 if (ret == MLX5_VDPA_STATS_RECEIVED_DESCRIPTORS)
575 stats[MLX5_VDPA_STATS_RECEIVED_DESCRIPTORS] = (struct rte_vdpa_stat) {
576 .id = MLX5_VDPA_STATS_RECEIVED_DESCRIPTORS,
577 .value = attr.received_desc - virtq->reset.received_desc,
579 if (ret == MLX5_VDPA_STATS_COMPLETED_DESCRIPTORS)
581 stats[MLX5_VDPA_STATS_COMPLETED_DESCRIPTORS] = (struct rte_vdpa_stat) {
582 .id = MLX5_VDPA_STATS_COMPLETED_DESCRIPTORS,
583 .value = attr.completed_desc - virtq->reset.completed_desc,
585 if (ret == MLX5_VDPA_STATS_BAD_DESCRIPTOR_ERRORS)
587 stats[MLX5_VDPA_STATS_BAD_DESCRIPTOR_ERRORS] = (struct rte_vdpa_stat) {
588 .id = MLX5_VDPA_STATS_BAD_DESCRIPTOR_ERRORS,
589 .value = attr.bad_desc_errors - virtq->reset.bad_desc_errors,
591 if (ret == MLX5_VDPA_STATS_EXCEED_MAX_CHAIN)
593 stats[MLX5_VDPA_STATS_EXCEED_MAX_CHAIN] = (struct rte_vdpa_stat) {
594 .id = MLX5_VDPA_STATS_EXCEED_MAX_CHAIN,
595 .value = attr.exceed_max_chain - virtq->reset.exceed_max_chain,
597 if (ret == MLX5_VDPA_STATS_INVALID_BUFFER)
599 stats[MLX5_VDPA_STATS_INVALID_BUFFER] = (struct rte_vdpa_stat) {
600 .id = MLX5_VDPA_STATS_INVALID_BUFFER,
601 .value = attr.invalid_buffer - virtq->reset.invalid_buffer,
603 if (ret == MLX5_VDPA_STATS_COMPLETION_ERRORS)
605 stats[MLX5_VDPA_STATS_COMPLETION_ERRORS] = (struct rte_vdpa_stat) {
606 .id = MLX5_VDPA_STATS_COMPLETION_ERRORS,
607 .value = attr.error_cqes - virtq->reset.error_cqes,
613 mlx5_vdpa_virtq_stats_reset(struct mlx5_vdpa_priv *priv, int qid)
615 struct mlx5_vdpa_virtq *virtq = &priv->virtqs[qid];
618 if (!virtq->counters) {
619 DRV_LOG(ERR, "Failed to read virtq %d statistics - virtq "
623 ret = mlx5_devx_cmd_query_virtio_q_counters(virtq->counters,
626 DRV_LOG(ERR, "Failed to read virtq %d reset stats from HW.",