net/mlx5: Use xarray to store and manage completion IRQs
authorMaher Sanalla <msanalla@nvidia.com>
Sun, 18 Jun 2023 16:23:24 +0000 (19:23 +0300)
committerSaeed Mahameed <saeedm@nvidia.com>
Mon, 7 Aug 2023 17:53:50 +0000 (10:53 -0700)
Use xarray to store the completion IRQs instead of a fixed-size allocated
array as not all completion IRQs will be requested on driver load, but
rather on demand when an EQ is created. The xarray offers more scalability,
reduced memory overhead, and provides the ability to dynamically resize the
array when needed.

Signed-off-by: Maher Sanalla <msanalla@nvidia.com>
Reviewed-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
drivers/net/ethernet/mellanox/mlx5/core/eq.c

index 268fd61..ac22d4d 100644 (file)
@@ -61,7 +61,7 @@ struct mlx5_eq_table {
        int                     curr_comp_eqs;
        int                     max_comp_eqs;
        struct mlx5_irq_table   *irq_table;
-       struct mlx5_irq         **comp_irqs;
+       struct xarray           comp_irqs;
        struct mlx5_irq         *ctrl_irq;
        struct cpu_rmap         *rmap;
        struct cpumask          used_cpus;
@@ -455,14 +455,18 @@ int mlx5_eq_table_init(struct mlx5_core_dev *dev)
 
        eq_table->irq_table = mlx5_irq_table_get(dev);
        cpumask_clear(&eq_table->used_cpus);
+       xa_init(&eq_table->comp_irqs);
        eq_table->curr_comp_eqs = 0;
        return 0;
 }
 
 void mlx5_eq_table_cleanup(struct mlx5_core_dev *dev)
 {
+       struct mlx5_eq_table *table = dev->priv.eq_table;
+
        mlx5_eq_debugfs_cleanup(dev);
-       kvfree(dev->priv.eq_table);
+       xa_destroy(&table->comp_irqs);
+       kvfree(table);
 }
 
 /* Async EQs */
@@ -810,10 +814,13 @@ EXPORT_SYMBOL(mlx5_eq_update_ci);
 static void comp_irqs_release_pci(struct mlx5_core_dev *dev)
 {
        struct mlx5_eq_table *table = dev->priv.eq_table;
-       int i;
+       struct mlx5_irq *irq;
+       unsigned long index;
 
-       for (i = 0; i < table->max_comp_eqs; i++)
-               mlx5_irq_release_vector(table->comp_irqs[i]);
+       xa_for_each(&table->comp_irqs, index, irq) {
+               xa_erase(&table->comp_irqs, index);
+               mlx5_irq_release_vector(irq);
+       }
 }
 
 static int comp_irqs_request_pci(struct mlx5_core_dev *dev)
@@ -849,7 +856,8 @@ spread_done:
                if (IS_ERR(irq))
                        break;
 
-               table->comp_irqs[i] = irq;
+               if (xa_err(xa_store(&table->comp_irqs, i, irq, GFP_KERNEL)))
+                       break;
        }
 
        kfree(cpus);
@@ -859,10 +867,13 @@ spread_done:
 static void comp_irqs_release_sf(struct mlx5_core_dev *dev)
 {
        struct mlx5_eq_table *table = dev->priv.eq_table;
-       int i;
+       struct mlx5_irq *irq;
+       unsigned long index;
 
-       for (i = 0; i < table->max_comp_eqs; i++)
-               mlx5_irq_affinity_irq_release(dev, table->comp_irqs[i]);
+       xa_for_each(&table->comp_irqs, index, irq) {
+               xa_erase(&table->comp_irqs, index);
+               mlx5_irq_affinity_irq_release(dev, irq);
+       }
 }
 
 static int comp_irqs_request_sf(struct mlx5_core_dev *dev)
@@ -876,7 +887,8 @@ static int comp_irqs_request_sf(struct mlx5_core_dev *dev)
                if (IS_ERR(irq))
                        break;
 
-               table->comp_irqs[i] = irq;
+               if (xa_err(xa_store(&table->comp_irqs, i, irq, GFP_KERNEL)))
+                       break;
        }
 
        return i ? i : PTR_ERR(irq);
@@ -884,31 +896,14 @@ static int comp_irqs_request_sf(struct mlx5_core_dev *dev)
 
 static void comp_irqs_release(struct mlx5_core_dev *dev)
 {
-       struct mlx5_eq_table *table = dev->priv.eq_table;
-
        mlx5_core_is_sf(dev) ? comp_irqs_release_sf(dev) :
                               comp_irqs_release_pci(dev);
-
-       kfree(table->comp_irqs);
 }
 
 static int comp_irqs_request(struct mlx5_core_dev *dev)
 {
-       struct mlx5_eq_table *table = dev->priv.eq_table;
-       int ncomp_eqs;
-       int ret;
-
-       ncomp_eqs = table->max_comp_eqs;
-       table->comp_irqs = kcalloc(ncomp_eqs, sizeof(*table->comp_irqs), GFP_KERNEL);
-       if (!table->comp_irqs)
-               return -ENOMEM;
-
-       ret = mlx5_core_is_sf(dev) ? comp_irqs_request_sf(dev) :
-                                    comp_irqs_request_pci(dev);
-       if (ret < 0)
-               kfree(table->comp_irqs);
-
-       return ret;
+       return mlx5_core_is_sf(dev) ? comp_irqs_request_sf(dev) :
+                                     comp_irqs_request_pci(dev);
 }
 
 #ifdef CONFIG_RFS_ACCEL
@@ -983,10 +978,11 @@ static int create_comp_eqs(struct mlx5_core_dev *dev)
 {
        struct mlx5_eq_table *table = dev->priv.eq_table;
        struct mlx5_eq_comp *eq;
+       struct mlx5_irq *irq;
+       unsigned long index;
        int ncomp_eqs;
        int nent;
        int err;
-       int i;
 
        err = alloc_rmap(dev);
        if (err)
@@ -1002,7 +998,8 @@ static int create_comp_eqs(struct mlx5_core_dev *dev)
        INIT_LIST_HEAD(&table->comp_eqs_list);
        nent = comp_eq_depth_devlink_param_get(dev);
 
-       for (i = 0; i < ncomp_eqs; i++) {
+       xa_for_each(&table->comp_irqs, index, irq)
+       {
                struct mlx5_eq_param param = {};
 
                eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, dev->priv.numa_node);
@@ -1018,7 +1015,7 @@ static int create_comp_eqs(struct mlx5_core_dev *dev)
 
                eq->irq_nb.notifier_call = mlx5_eq_comp_int;
                param = (struct mlx5_eq_param) {
-                       .irq = table->comp_irqs[i],
+                       .irq = irq,
                        .nent = nent,
                };