1 files changed, 57 insertions, 44 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
index 05e148db9889..c93492b67788 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c
@@ -14,17 +14,22 @@
 
 struct mlx5_sf_dev_table {
 	struct xarray devices;
-	unsigned int max_sfs;
 	phys_addr_t base_address;
 	u64 sf_bar_length;
 	struct notifier_block nb;
-	struct mutex table_lock; /* Serializes sf life cycle and vhca state change handler */
 	struct workqueue_struct *active_wq;
 	struct work_struct work;
 	u8 stop_active_wq:1;
 	struct mlx5_core_dev *dev;
 };
 
+struct mlx5_sf_dev_active_work_ctx {
+	struct work_struct work;
+	struct mlx5_vhca_state_event event;
+	struct mlx5_sf_dev_table *table;
+	int sf_index;
+};
+
 static bool mlx5_sf_dev_supported(const struct mlx5_core_dev *dev)
 {
 	return MLX5_CAP_GEN(dev, sf) && mlx5_vhca_event_supported(dev);
@@ -110,12 +115,6 @@ static void mlx5_sf_dev_add(struct mlx5_core_dev *dev, u16 sf_index, u16 fn_id,
 	sf_dev->parent_mdev = dev;
 	sf_dev->fn_id = fn_id;
 
-	if (!table->max_sfs) {
-		mlx5_adev_idx_free(id);
-		kfree(sf_dev);
-		err = -EOPNOTSUPP;
-		goto add_err;
-	}
 	sf_dev->bar_base_addr = table->base_address + (sf_index * table->sf_bar_length);
 
 	trace_mlx5_sf_dev_add(dev, sf_dev, id);
@@ -172,7 +171,6 @@ mlx5_sf_dev_state_change_handler(struct notifier_block *nb, unsigned long event_
 		return 0;
 
 	sf_index = event->function_id - base_id;
-	mutex_lock(&table->table_lock);
 	sf_dev = xa_load(&table->devices, sf_index);
 	switch (event->new_vhca_state) {
 	case MLX5_VHCA_STATE_INVALID:
@@ -196,7 +194,6 @@ mlx5_sf_dev_state_change_handler(struct notifier_block *nb, unsigned long event_
 	default:
 		break;
 	}
-	mutex_unlock(&table->table_lock);
 	return 0;
 }
 
@@ -221,15 +218,44 @@ static int mlx5_sf_dev_vhca_arm_all(struct mlx5_sf_dev_table *table)
 	return 0;
 }
 
-static void mlx5_sf_dev_add_active_work(struct work_struct *work)
+static void mlx5_sf_dev_add_active_work(struct work_struct *_work)
 {
-	struct mlx5_sf_dev_table *table = container_of(work, struct mlx5_sf_dev_table, work);
+	struct mlx5_sf_dev_active_work_ctx *work_ctx;
+
+	work_ctx = container_of(_work, struct mlx5_sf_dev_active_work_ctx, work);
+	if (work_ctx->table->stop_active_wq)
+		goto out;
+	/* Don't probe device which is already probe */
+	if (!xa_load(&work_ctx->table->devices, work_ctx->sf_index))
+		mlx5_sf_dev_add(work_ctx->table->dev, work_ctx->sf_index,
+				work_ctx->event.function_id, work_ctx->event.sw_function_id);
+	/* There is a race where SF got inactive after the query
+	 * above. e.g.: the query returns that the state of the
+	 * SF is active, and after that the eswitch manager set it to
+	 * inactive.
+	 * This case cannot be managed in SW, since the probing of the
+	 * SF is on one system, and the inactivation is on a different
+	 * system.
+	 * If the inactive is done after the SF perform init_hca(),
+	 * the SF will fully probe and then removed. If it was
+	 * done before init_hca(), the SF probe will fail.
+	 */
+out:
+	kfree(work_ctx);
+}
+
+/* In case SFs are generated externally, probe active SFs */
+static void mlx5_sf_dev_queue_active_works(struct work_struct *_work)
+{
+	struct mlx5_sf_dev_table *table = container_of(_work, struct mlx5_sf_dev_table, work);
 	u32 out[MLX5_ST_SZ_DW(query_vhca_state_out)] = {};
+	struct mlx5_sf_dev_active_work_ctx *work_ctx;
 	struct mlx5_core_dev *dev = table->dev;
 	u16 max_functions;
 	u16 function_id;
 	u16 sw_func_id;
 	int err = 0;
+	int wq_idx;
 	u8 state;
 	int i;
 
@@ -249,27 +275,22 @@ static void mlx5_sf_dev_add_active_work(struct work_struct *work)
 			continue;
 
 		sw_func_id = MLX5_GET(query_vhca_state_out, out, vhca_state_context.sw_function_id);
-		mutex_lock(&table->table_lock);
-		/* Don't probe device which is already probe */
-		if (!xa_load(&table->devices, i))
-			mlx5_sf_dev_add(dev, i, function_id, sw_func_id);
-		/* There is a race where SF got inactive after the query
-		 * above. e.g.: the query returns that the state of the
-		 * SF is active, and after that the eswitch manager set it to
-		 * inactive.
-		 * This case cannot be managed in SW, since the probing of the
-		 * SF is on one system, and the inactivation is on a different
-		 * system.
-		 * If the inactive is done after the SF perform init_hca(),
-		 * the SF will fully probe and then removed. If it was
-		 * done before init_hca(), the SF probe will fail.
-		 */
-		mutex_unlock(&table->table_lock);
+		work_ctx = kzalloc(sizeof(*work_ctx), GFP_KERNEL);
+		if (!work_ctx)
+			return;
+
+		INIT_WORK(&work_ctx->work, &mlx5_sf_dev_add_active_work);
+		work_ctx->event.function_id = function_id;
+		work_ctx->event.sw_function_id = sw_func_id;
+		work_ctx->table = table;
+		work_ctx->sf_index = i;
+		wq_idx = work_ctx->event.function_id % MLX5_DEV_MAX_WQS;
+		mlx5_vhca_events_work_enqueue(dev, wq_idx, &work_ctx->work);
 	}
 }
 
 /* In case SFs are generated externally, probe active SFs */
-static int mlx5_sf_dev_queue_active_work(struct mlx5_sf_dev_table *table)
+static int mlx5_sf_dev_create_active_works(struct mlx5_sf_dev_table *table)
 {
 	if (MLX5_CAP_GEN(table->dev, eswitch_manager))
 		return 0; /* the table is local */
@@ -280,12 +301,12 @@ static int mlx5_sf_dev_queue_active_work(struct mlx5_sf_dev_table *table)
 	table->active_wq = create_singlethread_workqueue("mlx5_active_sf");
 	if (!table->active_wq)
 		return -ENOMEM;
-	INIT_WORK(&table->work, &mlx5_sf_dev_add_active_work);
+	INIT_WORK(&table->work, &mlx5_sf_dev_queue_active_works);
 	queue_work(table->active_wq, &table->work);
 	return 0;
 }
 
-static void mlx5_sf_dev_destroy_active_work(struct mlx5_sf_dev_table *table)
+static void mlx5_sf_dev_destroy_active_works(struct mlx5_sf_dev_table *table)
 {
 	if (table->active_wq) {
 		table->stop_active_wq = true;
@@ -296,7 +317,6 @@ static void mlx5_sf_dev_destroy_active_work(struct mlx5_sf_dev_table *table)
 void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
 {
 	struct mlx5_sf_dev_table *table;
-	unsigned int max_sfs;
 	int err;
 
 	if (!mlx5_sf_dev_supported(dev))
@@ -310,37 +330,30 @@ void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
 
 	table->nb.notifier_call = mlx5_sf_dev_state_change_handler;
 	table->dev = dev;
-	if (MLX5_CAP_GEN(dev, max_num_sf))
-		max_sfs = MLX5_CAP_GEN(dev, max_num_sf);
-	else
-		max_sfs = 1 << MLX5_CAP_GEN(dev, log_max_sf);
 	table->sf_bar_length = 1 << (MLX5_CAP_GEN(dev, log_min_sf_size) + 12);
 	table->base_address = pci_resource_start(dev->pdev, 2);
-	table->max_sfs = max_sfs;
 	xa_init(&table->devices);
-	mutex_init(&table->table_lock);
 	dev->priv.sf_dev_table = table;
 
 	err = mlx5_vhca_event_notifier_register(dev, &table->nb);
 	if (err)
 		goto vhca_err;
 
-	err = mlx5_sf_dev_queue_active_work(table);
+	err = mlx5_sf_dev_create_active_works(table);
 	if (err)
 		goto add_active_err;
 
 	err = mlx5_sf_dev_vhca_arm_all(table);
 	if (err)
 		goto arm_err;
-	mlx5_core_dbg(dev, "SF DEV: max sf devices=%d\n", max_sfs);
 	return;
 
 arm_err:
-	mlx5_sf_dev_destroy_active_work(table);
+	mlx5_sf_dev_destroy_active_works(table);
 add_active_err:
 	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
+	mlx5_vhca_event_work_queues_flush(dev);
 vhca_err:
-	table->max_sfs = 0;
 	kfree(table);
 	dev->priv.sf_dev_table = NULL;
 table_err:
@@ -365,9 +378,9 @@ void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev)
 	if (!table)
 		return;
 
-	mlx5_sf_dev_destroy_active_work(table);
+	mlx5_sf_dev_destroy_active_works(table);
 	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
-	mutex_destroy(&table->table_lock);
+	mlx5_vhca_event_work_queues_flush(dev);
 
 	/* Now that event handler is not running, it is safe to destroy
 	 * the sf device without race.