IB/mlx5: Page faults handling infrastructure

* Refactor MR registration and cleanup, and fix reg_pages accounting.
* Create a work queue to handle page fault events in a kthread context.
* Register a fault handler to get events from the core for each QP.

The registered fault handler is empty in this patch, and only a later
patch implements it.

Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>

1 files changed, 145 insertions, 0 deletions

diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 66c39ee..63bbdba 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -32,6 +32,8 @@
 
 #include "mlx5_ib.h"
 
+struct workqueue_struct *mlx5_ib_page_fault_wq;
+
 #define COPY_ODP_BIT_MLX_TO_IB(reg, ib_caps, field_name, bit_name) do {	\
 	if (be32_to_cpu(reg.field_name) & MLX5_ODP_SUPPORT_##bit_name)	\
 		ib_caps->field_name |= IB_ODP_SUPPORT_##bit_name;	\
@@ -58,3 +60,146 @@ int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev *dev)
 out:
 	return err;
 }
+
+static struct mlx5_ib_mr *mlx5_ib_odp_find_mr_lkey(struct mlx5_ib_dev *dev,
+						   u32 key)
+{
+	u32 base_key = mlx5_base_mkey(key);
+	struct mlx5_core_mr *mmr = __mlx5_mr_lookup(dev->mdev, base_key);
+
+	if (!mmr || mmr->key != key)
+		return NULL;
+
+	return container_of(mmr, struct mlx5_ib_mr, mmr);
+}
+
+static void mlx5_ib_page_fault_resume(struct mlx5_ib_qp *qp,
+				      struct mlx5_ib_pfault *pfault,
+				      int error) {
+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device);
+	int ret = mlx5_core_page_fault_resume(dev->mdev, qp->mqp.qpn,
+					      pfault->mpfault.flags,
+					      error);
+	if (ret)
+		pr_err("Failed to resolve the page fault on QP 0x%x\n",
+		       qp->mqp.qpn);
+}
+
+void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp,
+			       struct mlx5_ib_pfault *pfault)
+{
+	u8 event_subtype = pfault->mpfault.event_subtype;
+
+	switch (event_subtype) {
+	default:
+		pr_warn("Invalid page fault event subtype: 0x%x\n",
+			event_subtype);
+		mlx5_ib_page_fault_resume(qp, pfault, 1);
+		break;
+	}
+}
+
+static void mlx5_ib_qp_pfault_action(struct work_struct *work)
+{
+	struct mlx5_ib_pfault *pfault = container_of(work,
+						     struct mlx5_ib_pfault,
+						     work);
+	enum mlx5_ib_pagefault_context context =
+		mlx5_ib_get_pagefault_context(&pfault->mpfault);
+	struct mlx5_ib_qp *qp = container_of(pfault, struct mlx5_ib_qp,
+					     pagefaults[context]);
+	mlx5_ib_mr_pfault_handler(qp, pfault);
+}
+
+void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->disable_page_faults_lock, flags);
+	qp->disable_page_faults = 1;
+	spin_unlock_irqrestore(&qp->disable_page_faults_lock, flags);
+
+	/*
+	 * Note that at this point, we are guarenteed that no more
+	 * work queue elements will be posted to the work queue with
+	 * the QP we are closing.
+	 */
+	flush_workqueue(mlx5_ib_page_fault_wq);
+}
+
+void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->disable_page_faults_lock, flags);
+	qp->disable_page_faults = 0;
+	spin_unlock_irqrestore(&qp->disable_page_faults_lock, flags);
+}
+
+static void mlx5_ib_pfault_handler(struct mlx5_core_qp *qp,
+				   struct mlx5_pagefault *pfault)
+{
+	/*
+	 * Note that we will only get one fault event per QP per context
+	 * (responder/initiator, read/write), until we resolve the page fault
+	 * with the mlx5_ib_page_fault_resume command. Since this function is
+	 * called from within the work element, there is no risk of missing
+	 * events.
+	 */
+	struct mlx5_ib_qp *mibqp = to_mibqp(qp);
+	enum mlx5_ib_pagefault_context context =
+		mlx5_ib_get_pagefault_context(pfault);
+	struct mlx5_ib_pfault *qp_pfault = &mibqp->pagefaults[context];
+
+	qp_pfault->mpfault = *pfault;
+
+	/* No need to stop interrupts here since we are in an interrupt */
+	spin_lock(&mibqp->disable_page_faults_lock);
+	if (!mibqp->disable_page_faults)
+		queue_work(mlx5_ib_page_fault_wq, &qp_pfault->work);
+	spin_unlock(&mibqp->disable_page_faults_lock);
+}
+
+void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp)
+{
+	int i;
+
+	qp->disable_page_faults = 1;
+	spin_lock_init(&qp->disable_page_faults_lock);
+
+	qp->mqp.pfault_handler	= mlx5_ib_pfault_handler;
+
+	for (i = 0; i < MLX5_IB_PAGEFAULT_CONTEXTS; ++i)
+		INIT_WORK(&qp->pagefaults[i].work, mlx5_ib_qp_pfault_action);
+}
+
+int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev)
+{
+	int ret;
+
+	ret = init_srcu_struct(&ibdev->mr_srcu);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev)
+{
+	cleanup_srcu_struct(&ibdev->mr_srcu);
+}
+
+int __init mlx5_ib_odp_init(void)
+{
+	mlx5_ib_page_fault_wq =
+		create_singlethread_workqueue("mlx5_ib_page_faults");
+	if (!mlx5_ib_page_fault_wq)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void mlx5_ib_odp_cleanup(void)
+{
+	destroy_workqueue(mlx5_ib_page_fault_wq);
+}