RDMA/mlx5: Create ODP EQ only when ODP MR is created

There is no need to create the ODP EQ if the user doesn't use ODP MRs.
Hence, create it only when the first ODP MR is created. This EQ will be
destroyed only when the device is unloaded.
This will decrease the number of EQs created per device. for example: If
we creates 1K devices (SF/VF/etc'), than we will decrease the num of EQs
by 1K.

Link: https://lore.kernel.org/r/20210314125418.179716-1-leon@kernel.org
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 03fe99deea147dbd8273f942f57798994086f060..d6bd744aba6be0e1992b14c53f7321b820335823 100644 (file)
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -1080,6 +1080,7 @@ struct mlx5_ib_dev {
        struct mutex                    slow_path_mutex;
        struct ib_odp_caps      odp_caps;
        u64                     odp_max_size;
+       struct mutex            odp_eq_mutex;
        struct mlx5_ib_pf_eq    odp_pf_eq;
 
        struct xarray           odp_mkeys;
@@ -1358,6 +1359,7 @@ struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev);
 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
+int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq);
 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev);
 int __init mlx5_ib_odp_init(void);
 void mlx5_ib_odp_cleanup(void);
@@ -1377,6 +1379,11 @@ static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev)
 }
 
 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
+static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev,
+                                     struct mlx5_ib_pf_eq *eq)
+{
+       return 0;
+}
 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {}
 static inline int mlx5_ib_odp_init(void) { return 0; }
 static inline void mlx5_ib_odp_cleanup(void)                               {}

diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
index 9dcb9fb4eeaa59f27b467cf721e02d72e68fb124..eeb9f7e15634f8c61cc93b5f1f3b0fbfda1f2d0d 100644 (file)
--- a/drivers/infiniband/hw/mlx5/mr.c
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -1500,6 +1500,9 @@ static struct ib_mr *create_user_odp_mr(struct ib_pd *pd, u64 start, u64 length,
        if (!IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING))
                return ERR_PTR(-EOPNOTSUPP);
 
+       err = mlx5r_odp_create_eq(dev, &dev->odp_pf_eq);
+       if (err)
+               return ERR_PTR(err);
        if (!start && length == U64_MAX) {
                if (iova != 0)
                        return ERR_PTR(-EINVAL);

diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 3008d1539ad41aacbba2b90f0e6c981de574e977..a0b9111b508a555e736570f695ca686e5dc35c8e 100644 (file)
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -1531,20 +1531,24 @@ enum {
        MLX5_IB_NUM_PF_DRAIN    = 64,
 };
 
-static int
-mlx5_ib_create_pf_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)
+int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)
 {
        struct mlx5_eq_param param = {};
-       int err;
+       int err = 0;
 
+       mutex_lock(&dev->odp_eq_mutex);
+       if (eq->core)
+               goto unlock;
        INIT_WORK(&eq->work, mlx5_ib_eq_pf_action);
        spin_lock_init(&eq->lock);
        eq->dev = dev;
 
        eq->pool = mempool_create_kmalloc_pool(MLX5_IB_NUM_PF_DRAIN,
                                               sizeof(struct mlx5_pagefault));
-       if (!eq->pool)
-               return -ENOMEM;
+       if (!eq->pool) {
+               err = -ENOMEM;
+               goto unlock;
+       }
 
        eq->wq = alloc_workqueue("mlx5_ib_page_fault",
                                 WQ_HIGHPRI | WQ_UNBOUND | WQ_MEM_RECLAIM,
@@ -1555,7 +1559,7 @@ mlx5_ib_create_pf_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)
        }
 
        eq->irq_nb.notifier_call = mlx5_ib_eq_pf_int;
-       param = (struct mlx5_eq_param) {
+       param = (struct mlx5_eq_param){
                .irq_index = 0,
                .nent = MLX5_IB_NUM_PF_EQE,
        };
@@ -1571,21 +1575,27 @@ mlx5_ib_create_pf_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)
                goto err_eq;
        }
 
+       mutex_unlock(&dev->odp_eq_mutex);
        return 0;
 err_eq:
        mlx5_eq_destroy_generic(dev->mdev, eq->core);
 err_wq:
+       eq->core = NULL;
        destroy_workqueue(eq->wq);
 err_mempool:
        mempool_destroy(eq->pool);
+unlock:
+       mutex_unlock(&dev->odp_eq_mutex);
        return err;
 }
 
 static int
-mlx5_ib_destroy_pf_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)
+mlx5_ib_odp_destroy_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)
 {
        int err;
 
+       if (!eq->core)
+               return 0;
        mlx5_eq_disable(dev->mdev, eq->core, &eq->irq_nb);
        err = mlx5_eq_destroy_generic(dev->mdev, eq->core);
        cancel_work_sync(&eq->work);
@@ -1642,8 +1652,7 @@ int mlx5_ib_odp_init_one(struct mlx5_ib_dev *dev)
                }
        }
 
-       ret = mlx5_ib_create_pf_eq(dev, &dev->odp_pf_eq);
-
+       mutex_init(&dev->odp_eq_mutex);
        return ret;
 }
 
@@ -1652,7 +1661,7 @@ void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *dev)
        if (!(dev->odp_caps.general_caps & IB_ODP_SUPPORT))
                return;
 
-       mlx5_ib_destroy_pf_eq(dev, &dev->odp_pf_eq);
+       mlx5_ib_odp_destroy_eq(dev, &dev->odp_pf_eq);
 }
 
 int mlx5_ib_odp_init(void)