From: Xueming Li Date: Tue, 27 Oct 2020 08:28:45 +0000 (+0000) Subject: vdpa/mlx5: handle hardware error X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=0474419bae7c70117010e77c5278965fff9cbf9b;p=dpdk.git vdpa/mlx5: handle hardware error When hardware error happens, vdpa didn't get such information and leave driver in silent: working state but no response. This patch subscribes firmware virtq error event and try to recover max 3 times in 3 seconds, stop virtq if max retry number reached. When error happens, PMD log in warning level. If failed to recover, outputs error log. Query virtq statistics to get error counters report. Acked-by: Matan Azrad Signed-off-by: Xueming Li Reviewed-by: Maxime Coquelin --- diff --git a/doc/guides/rel_notes/release_20_11.rst b/doc/guides/rel_notes/release_20_11.rst index a0b5337bb1..fb241af23b 100644 --- a/doc/guides/rel_notes/release_20_11.rst +++ b/doc/guides/rel_notes/release_20_11.rst @@ -195,6 +195,10 @@ New Features * Added support of Age action query. * Added support of multi-ports hairpin. + Updated Mellanox mlx5 vDPA driver: + + * Added support of vDPA VirtQ error handling. + * **Updated Solarflare network PMD.** Updated the Solarflare ``sfc_efx`` driver with changes including: diff --git a/doc/guides/vdpadevs/mlx5.rst b/doc/guides/vdpadevs/mlx5.rst index 9d10fee2b6..3a6d88362d 100644 --- a/doc/guides/vdpadevs/mlx5.rst +++ b/doc/guides/vdpadevs/mlx5.rst @@ -134,3 +134,11 @@ Driver options driver to no-traffic mode. In this mode the timer events are stopped and interrupts are configured to the device in order to notify traffic for the driver. Default value is 2s. + +Error handling +^^^^^^^^^^^^^^ + +Upon potential hardware errors, mlx5 PMD try to recover, give up if failed 3 +times in 3 seconds, virtq will be put in disable state. User should check log +to get error information, or query vdpa statistics counter to know error type +and count report. diff --git a/drivers/vdpa/mlx5/mlx5_vdpa.c b/drivers/vdpa/mlx5/mlx5_vdpa.c index 2d88633bfd..2d65370343 100644 --- a/drivers/vdpa/mlx5/mlx5_vdpa.c +++ b/drivers/vdpa/mlx5/mlx5_vdpa.c @@ -283,6 +283,7 @@ mlx5_vdpa_dev_close(int vid) } if (priv->configured) ret |= mlx5_vdpa_lm_log(priv); + mlx5_vdpa_err_event_unset(priv); mlx5_vdpa_cqe_event_unset(priv); mlx5_vdpa_steer_unset(priv); mlx5_vdpa_virtqs_release(priv); @@ -318,6 +319,7 @@ mlx5_vdpa_dev_config(int vid) DRV_LOG(WARNING, "MTU cannot be set on device %s.", vdev->device->name); if (mlx5_vdpa_pd_create(priv) || mlx5_vdpa_mem_register(priv) || + mlx5_vdpa_err_event_setup(priv) || mlx5_vdpa_virtqs_prepare(priv) || mlx5_vdpa_steer_setup(priv) || mlx5_vdpa_cqe_event_setup(priv)) { mlx5_vdpa_dev_close(vid); diff --git a/drivers/vdpa/mlx5/mlx5_vdpa.h b/drivers/vdpa/mlx5/mlx5_vdpa.h index fcbc12ab0c..0d6886c52c 100644 --- a/drivers/vdpa/mlx5/mlx5_vdpa.h +++ b/drivers/vdpa/mlx5/mlx5_vdpa.h @@ -87,6 +87,7 @@ struct mlx5_vdpa_virtq { uint16_t vq_size; uint8_t notifier_state; bool stopped; + uint32_t version; struct mlx5_vdpa_priv *priv; struct mlx5_devx_obj *virtq; struct mlx5_devx_obj *counters; @@ -97,6 +98,8 @@ struct mlx5_vdpa_virtq { uint32_t size; } umems[3]; struct rte_intr_handle intr_handle; + uint64_t err_time[3]; /* RDTSC time of recent errors. */ + uint32_t n_retry; struct mlx5_devx_virtio_q_couners_attr reset; }; @@ -143,8 +146,10 @@ struct mlx5_vdpa_priv { struct rte_vhost_memory *vmem; uint32_t eqn; struct mlx5dv_devx_event_channel *eventc; + struct mlx5dv_devx_event_channel *err_chnl; struct mlx5dv_devx_uar *uar; struct rte_intr_handle intr_handle; + struct rte_intr_handle err_intr_handle; struct mlx5_devx_obj *td; struct mlx5_devx_obj *tis; uint16_t nr_virtqs; @@ -259,6 +264,25 @@ int mlx5_vdpa_cqe_event_setup(struct mlx5_vdpa_priv *priv); */ void mlx5_vdpa_cqe_event_unset(struct mlx5_vdpa_priv *priv); +/** + * Setup error interrupt handler. + * + * @param[in] priv + * The vdpa driver private structure. + * + * @return + * 0 on success, a negative errno value otherwise and rte_errno is set. + */ +int mlx5_vdpa_err_event_setup(struct mlx5_vdpa_priv *priv); + +/** + * Unset error event handler. + * + * @param[in] priv + * The vdpa driver private structure. + */ +void mlx5_vdpa_err_event_unset(struct mlx5_vdpa_priv *priv); + /** * Release a virtq and all its related resources. * @@ -392,6 +416,19 @@ int mlx5_vdpa_virtq_modify(struct mlx5_vdpa_virtq *virtq, int state); */ int mlx5_vdpa_virtq_stop(struct mlx5_vdpa_priv *priv, int index); +/** + * Query virtq information. + * + * @param[in] priv + * The vdpa driver private structure. + * @param[in] index + * The virtq index. + * + * @return + * 0 on success, a negative value otherwise. + */ +int mlx5_vdpa_virtq_query(struct mlx5_vdpa_priv *priv, int index); + /** * Get virtq statistics. * diff --git a/drivers/vdpa/mlx5/mlx5_vdpa_event.c b/drivers/vdpa/mlx5/mlx5_vdpa_event.c index 8a01e42794..010543cb6a 100644 --- a/drivers/vdpa/mlx5/mlx5_vdpa_event.c +++ b/drivers/vdpa/mlx5/mlx5_vdpa_event.c @@ -15,11 +15,14 @@ #include #include +#include #include "mlx5_vdpa_utils.h" #include "mlx5_vdpa.h" +#define MLX5_VDPA_ERROR_TIME_SEC 3u + void mlx5_vdpa_event_qp_global_release(struct mlx5_vdpa_priv *priv) { @@ -378,6 +381,147 @@ mlx5_vdpa_interrupt_handler(void *cb_arg) pthread_mutex_unlock(&priv->vq_config_lock); } +static void +mlx5_vdpa_err_interrupt_handler(void *cb_arg __rte_unused) +{ +#ifdef HAVE_IBV_DEVX_EVENT + struct mlx5_vdpa_priv *priv = cb_arg; + union { + struct mlx5dv_devx_async_event_hdr event_resp; + uint8_t buf[sizeof(struct mlx5dv_devx_async_event_hdr) + 128]; + } out; + uint32_t vq_index, i, version; + struct mlx5_vdpa_virtq *virtq; + uint64_t sec; + + pthread_mutex_lock(&priv->vq_config_lock); + while (mlx5_glue->devx_get_event(priv->err_chnl, &out.event_resp, + sizeof(out.buf)) >= + (ssize_t)sizeof(out.event_resp.cookie)) { + vq_index = out.event_resp.cookie & UINT32_MAX; + version = out.event_resp.cookie >> 32; + if (vq_index >= priv->nr_virtqs) { + DRV_LOG(ERR, "Invalid device %s error event virtq %d.", + priv->vdev->device->name, vq_index); + continue; + } + virtq = &priv->virtqs[vq_index]; + if (!virtq->enable || virtq->version != version) + continue; + if (rte_rdtsc() / rte_get_tsc_hz() < MLX5_VDPA_ERROR_TIME_SEC) + continue; + virtq->stopped = true; + /* Query error info. */ + if (mlx5_vdpa_virtq_query(priv, vq_index)) + goto log; + /* Disable vq. */ + if (mlx5_vdpa_virtq_enable(priv, vq_index, 0)) { + DRV_LOG(ERR, "Failed to disable virtq %d.", vq_index); + goto log; + } + /* Retry if error happens less than N times in 3 seconds. */ + sec = (rte_rdtsc() - virtq->err_time[0]) / rte_get_tsc_hz(); + if (sec > MLX5_VDPA_ERROR_TIME_SEC) { + /* Retry. */ + if (mlx5_vdpa_virtq_enable(priv, vq_index, 1)) + DRV_LOG(ERR, "Failed to enable virtq %d.", + vq_index); + else + DRV_LOG(WARNING, "Recover virtq %d: %u.", + vq_index, ++virtq->n_retry); + } else { + /* Retry timeout, give up. */ + DRV_LOG(ERR, "Device %s virtq %d failed to recover.", + priv->vdev->device->name, vq_index); + } +log: + /* Shift in current time to error time log end. */ + for (i = 1; i < RTE_DIM(virtq->err_time); i++) + virtq->err_time[i - 1] = virtq->err_time[i]; + virtq->err_time[RTE_DIM(virtq->err_time) - 1] = rte_rdtsc(); + } + pthread_mutex_unlock(&priv->vq_config_lock); +#endif +} + +int +mlx5_vdpa_err_event_setup(struct mlx5_vdpa_priv *priv) +{ + int ret; + int flags; + + /* Setup device event channel. */ + priv->err_chnl = mlx5_glue->devx_create_event_channel(priv->ctx, 0); + if (!priv->err_chnl) { + rte_errno = errno; + DRV_LOG(ERR, "Failed to create device event channel %d.", + rte_errno); + goto error; + } + flags = fcntl(priv->err_chnl->fd, F_GETFL); + ret = fcntl(priv->err_chnl->fd, F_SETFL, flags | O_NONBLOCK); + if (ret) { + DRV_LOG(ERR, "Failed to change device event channel FD."); + goto error; + } + priv->err_intr_handle.fd = priv->err_chnl->fd; + priv->err_intr_handle.type = RTE_INTR_HANDLE_EXT; + if (rte_intr_callback_register(&priv->err_intr_handle, + mlx5_vdpa_err_interrupt_handler, + priv)) { + priv->err_intr_handle.fd = 0; + DRV_LOG(ERR, "Failed to register error interrupt for device %d.", + priv->vid); + goto error; + } else { + DRV_LOG(DEBUG, "Registered error interrupt for device%d.", + priv->vid); + } + return 0; +error: + mlx5_vdpa_err_event_unset(priv); + return -1; +} + +void +mlx5_vdpa_err_event_unset(struct mlx5_vdpa_priv *priv) +{ + int retries = MLX5_VDPA_INTR_RETRIES; + int ret = -EAGAIN; + + if (!priv->err_intr_handle.fd) + return; + while (retries-- && ret == -EAGAIN) { + ret = rte_intr_callback_unregister(&priv->err_intr_handle, + mlx5_vdpa_err_interrupt_handler, + priv); + if (ret == -EAGAIN) { + DRV_LOG(DEBUG, "Try again to unregister fd %d " + "of error interrupt, retries = %d.", + priv->err_intr_handle.fd, retries); + rte_pause(); + } + } + memset(&priv->err_intr_handle, 0, sizeof(priv->err_intr_handle)); + if (priv->err_chnl) { +#ifdef HAVE_IBV_DEVX_EVENT + union { + struct mlx5dv_devx_async_event_hdr event_resp; + uint8_t buf[sizeof(struct mlx5dv_devx_async_event_hdr) + + 128]; + } out; + + /* Clean all pending events. */ + while (mlx5_glue->devx_get_event(priv->err_chnl, + &out.event_resp, sizeof(out.buf)) >= + (ssize_t)sizeof(out.event_resp.cookie)) + ; +#endif + mlx5_glue->devx_destroy_event_channel(priv->err_chnl); + priv->err_chnl = NULL; + } +} + int mlx5_vdpa_cqe_event_setup(struct mlx5_vdpa_priv *priv) { diff --git a/drivers/vdpa/mlx5/mlx5_vdpa_virtq.c b/drivers/vdpa/mlx5/mlx5_vdpa_virtq.c index 17e71cf4f4..d5ac040544 100644 --- a/drivers/vdpa/mlx5/mlx5_vdpa_virtq.c +++ b/drivers/vdpa/mlx5/mlx5_vdpa_virtq.c @@ -88,11 +88,6 @@ mlx5_vdpa_virtq_unset(struct mlx5_vdpa_virtq *virtq) rte_free(virtq->umems[i].buf); } memset(&virtq->umems, 0, sizeof(virtq->umems)); - if (virtq->counters) { - claim_zero(mlx5_devx_cmd_destroy(virtq->counters)); - virtq->counters = NULL; - } - memset(&virtq->reset, 0, sizeof(virtq->reset)); if (virtq->eqp.fw_qp) mlx5_vdpa_event_qp_destroy(&virtq->eqp); virtq->notifier_state = MLX5_VDPA_NOTIFIER_STATE_DISABLED; @@ -103,9 +98,19 @@ void mlx5_vdpa_virtqs_release(struct mlx5_vdpa_priv *priv) { int i; + struct mlx5_vdpa_virtq *virtq; - for (i = 0; i < priv->nr_virtqs; i++) - mlx5_vdpa_virtq_unset(&priv->virtqs[i]); + for (i = 0; i < priv->nr_virtqs; i++) { + virtq = &priv->virtqs[i]; + mlx5_vdpa_virtq_unset(virtq); + if (virtq->counters) { + claim_zero(mlx5_devx_cmd_destroy(virtq->counters)); + virtq->counters = NULL; + memset(&virtq->reset, 0, sizeof(virtq->reset)); + } + memset(virtq->err_time, 0, sizeof(virtq->err_time)); + virtq->n_retry = 0; + } if (priv->tis) { claim_zero(mlx5_devx_cmd_destroy(priv->tis)); priv->tis = NULL; @@ -138,7 +143,6 @@ mlx5_vdpa_virtq_modify(struct mlx5_vdpa_virtq *virtq, int state) int mlx5_vdpa_virtq_stop(struct mlx5_vdpa_priv *priv, int index) { - struct mlx5_devx_virtq_attr attr = {0}; struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index]; int ret; @@ -148,6 +152,17 @@ mlx5_vdpa_virtq_stop(struct mlx5_vdpa_priv *priv, int index) if (ret) return -1; virtq->stopped = true; + DRV_LOG(DEBUG, "vid %u virtq %u was stopped.", priv->vid, index); + return mlx5_vdpa_virtq_query(priv, index); +} + +int +mlx5_vdpa_virtq_query(struct mlx5_vdpa_priv *priv, int index) +{ + struct mlx5_devx_virtq_attr attr = {0}; + struct mlx5_vdpa_virtq *virtq = &priv->virtqs[index]; + int ret; + if (mlx5_devx_cmd_query_virtq(virtq->virtq, &attr)) { DRV_LOG(ERR, "Failed to query virtq %d.", index); return -1; @@ -162,7 +177,9 @@ mlx5_vdpa_virtq_stop(struct mlx5_vdpa_priv *priv, int index) DRV_LOG(ERR, "Failed to set virtq %d base.", index); return -1; } - DRV_LOG(DEBUG, "vid %u virtq %u was stopped.", priv->vid, index); + if (attr.state == MLX5_VIRTQ_STATE_ERROR) + DRV_LOG(WARNING, "vid %d vring %d hw error=%hhu", + priv->vid, index, attr.error_type); return 0; } @@ -195,6 +212,8 @@ mlx5_vdpa_virtq_setup(struct mlx5_vdpa_priv *priv, int index) unsigned int i; uint16_t last_avail_idx; uint16_t last_used_idx; + uint16_t event_num = MLX5_EVENT_TYPE_OBJECT_CHANGE; + uint64_t cookie; ret = rte_vhost_get_vhost_vring(priv->vid, index, &vq); if (ret) @@ -231,8 +250,9 @@ mlx5_vdpa_virtq_setup(struct mlx5_vdpa_priv *priv, int index) " need event QPs and event mechanism.", index); } if (priv->caps.queue_counters_valid) { - virtq->counters = mlx5_devx_cmd_create_virtio_q_counters - (priv->ctx); + if (!virtq->counters) + virtq->counters = mlx5_devx_cmd_create_virtio_q_counters + (priv->ctx); if (!virtq->counters) { DRV_LOG(ERR, "Failed to create virtq couners for virtq" " %d.", index); @@ -332,6 +352,19 @@ mlx5_vdpa_virtq_setup(struct mlx5_vdpa_priv *priv, int index) virtq->intr_handle.fd, index); } } + /* Subscribe virtq error event. */ + virtq->version++; + cookie = ((uint64_t)virtq->version << 32) + index; + ret = mlx5_glue->devx_subscribe_devx_event(priv->err_chnl, + virtq->virtq->obj, + sizeof(event_num), + &event_num, cookie); + if (ret) { + DRV_LOG(ERR, "Failed to subscribe device %d virtq %d error event.", + priv->vid, index); + rte_errno = errno; + goto error; + } virtq->stopped = false; DRV_LOG(DEBUG, "vid %u virtq %u was created successfully.", priv->vid, index); @@ -526,12 +559,11 @@ mlx5_vdpa_virtq_stats_get(struct mlx5_vdpa_priv *priv, int qid, struct mlx5_devx_virtio_q_couners_attr attr = {0}; int ret; - if (!virtq->virtq || !virtq->enable) { + if (!virtq->counters) { DRV_LOG(ERR, "Failed to read virtq %d statistics - virtq " "is invalid.", qid); return -EINVAL; } - MLX5_ASSERT(virtq->counters); ret = mlx5_devx_cmd_query_virtio_q_counters(virtq->counters, &attr); if (ret) { DRV_LOG(ERR, "Failed to read virtq %d stats from HW.", qid); @@ -583,12 +615,11 @@ mlx5_vdpa_virtq_stats_reset(struct mlx5_vdpa_priv *priv, int qid) struct mlx5_vdpa_virtq *virtq = &priv->virtqs[qid]; int ret; - if (!virtq->virtq || !virtq->enable) { + if (!virtq->counters) { DRV_LOG(ERR, "Failed to read virtq %d statistics - virtq " "is invalid.", qid); return -EINVAL; } - MLX5_ASSERT(virtq->counters); ret = mlx5_devx_cmd_query_virtio_q_counters(virtq->counters, &virtq->reset); if (ret)