IB/mlx5: Reset flow support for IB kernel ULPs
The driver exposes interfaces that directly relate to HW state.
Upon fatal error, consumers of these interfaces (ULPs) that rely
on completion of all their posted work-request could hang, thereby
introducing dependencies in shutdown order. To prevent this from
happening, we manage the relevant resources (CQs, QPs) that are used
by the device. Upon a fatal error, we now generate simulated
completions for outstanding WQEs that were not completed at the
time the HW was reset.
It includes invoking the completion event handler for all involved
CQs so that the ULPs will poll those CQs. When polled we return
simulated CQEs with IB_WC_WR_FLUSH_ERR return code enabling ULPs
to clean up their resources and not wait forever for completions
upon receiving remove_one.
The above change requires an extra check in the data path to make
sure that when device is in error state, the simulated CQEs will
be returned and no further WQEs will be posted.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 21acee4..9b6d283 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -1980,6 +1980,65 @@
mutex_unlock(&ports->devr->mutex);
}
+static void mlx5_ib_handle_internal_error(struct mlx5_ib_dev *ibdev)
+{
+ struct mlx5_ib_qp *mqp;
+ struct mlx5_ib_cq *send_mcq, *recv_mcq;
+ struct mlx5_core_cq *mcq;
+ struct list_head cq_armed_list;
+ unsigned long flags_qp;
+ unsigned long flags_cq;
+ unsigned long flags;
+
+ INIT_LIST_HEAD(&cq_armed_list);
+
+ /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
+ spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
+ list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
+ spin_lock_irqsave(&mqp->sq.lock, flags_qp);
+ if (mqp->sq.tail != mqp->sq.head) {
+ send_mcq = to_mcq(mqp->ibqp.send_cq);
+ spin_lock_irqsave(&send_mcq->lock, flags_cq);
+ if (send_mcq->mcq.comp &&
+ mqp->ibqp.send_cq->comp_handler) {
+ if (!send_mcq->mcq.reset_notify_added) {
+ send_mcq->mcq.reset_notify_added = 1;
+ list_add_tail(&send_mcq->mcq.reset_notify,
+ &cq_armed_list);
+ }
+ }
+ spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
+ }
+ spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
+ spin_lock_irqsave(&mqp->rq.lock, flags_qp);
+ /* no handling is needed for SRQ */
+ if (!mqp->ibqp.srq) {
+ if (mqp->rq.tail != mqp->rq.head) {
+ recv_mcq = to_mcq(mqp->ibqp.recv_cq);
+ spin_lock_irqsave(&recv_mcq->lock, flags_cq);
+ if (recv_mcq->mcq.comp &&
+ mqp->ibqp.recv_cq->comp_handler) {
+ if (!recv_mcq->mcq.reset_notify_added) {
+ recv_mcq->mcq.reset_notify_added = 1;
+ list_add_tail(&recv_mcq->mcq.reset_notify,
+ &cq_armed_list);
+ }
+ }
+ spin_unlock_irqrestore(&recv_mcq->lock,
+ flags_cq);
+ }
+ }
+ spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
+ }
+ /*At that point all inflight post send were put to be executed as of we
+ * lock/unlock above locks Now need to arm all involved CQs.
+ */
+ list_for_each_entry(mcq, &cq_armed_list, reset_notify) {
+ mcq->comp(mcq);
+ }
+ spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
+}
+
static void mlx5_ib_event(struct mlx5_core_dev *dev, void *context,
enum mlx5_dev_event event, unsigned long param)
{
@@ -1992,6 +2051,7 @@
case MLX5_DEV_EVENT_SYS_ERROR:
ibdev->ib_active = false;
ibev.event = IB_EVENT_DEVICE_FATAL;
+ mlx5_ib_handle_internal_error(ibdev);
break;
case MLX5_DEV_EVENT_PORT_UP:
@@ -2595,6 +2655,8 @@
mutex_init(&dev->flow_db.lock);
mutex_init(&dev->cap_mask_mutex);
+ INIT_LIST_HEAD(&dev->qp_list);
+ spin_lock_init(&dev->reset_flow_resource_lock);
if (ll == IB_LINK_LAYER_ETHERNET) {
err = mlx5_enable_roce(dev);