IB/mlx5: Resolve soft lock on massive reg MRs

When calling reg_mr of large MRs (e.g. 4GB) from multiple processes
and MR caches can't supply the required amount of MRs the slow-path
of MR allocation may be used. In this case we need to serialize the
slow-path between the processes to avoid soft lock.

Fixes: e126ba97dba9 ('mlx5: Add driver for Mellanox Connect-IB adapters')
Signed-off-by: Moshe Lazer <moshel@mellanox.com>
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Eli Cohen <eli@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/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index dcdcd195fe53a4dd003b0d22426d146bb81c0380..7d689903c87cfea2a84755f81362095c97167c8d 100644 (file)
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -626,6 +626,8 @@ struct mlx5_ib_dev {
        struct mlx5_ib_resources        devr;
        struct mlx5_mr_cache            cache;
        struct timer_list               delay_timer;
+       /* Prevents soft lock on massive reg MRs */
+       struct mutex                    slow_path_mutex;
        int                             fill_delay;
 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
        struct ib_odp_caps      odp_caps;

diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
index d4ad672b905bf0068b59b2ebffe8de8a85f035e1..4e9012463c37de6381cddde1c527e4c2ff6e84ae 100644 (file)
--- a/drivers/infiniband/hw/mlx5/mr.c
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -610,6 +610,7 @@ int mlx5_mr_cache_init(struct mlx5_ib_dev *dev)
        int err;
        int i;
 
+       mutex_init(&dev->slow_path_mutex);
        cache->wq = alloc_ordered_workqueue("mkey_cache", WQ_MEM_RECLAIM);
        if (!cache->wq) {
                mlx5_ib_warn(dev, "failed to create work queue\n");
@@ -1182,9 +1183,12 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
                goto error;
        }
 
-       if (!mr)
+       if (!mr) {
+               mutex_lock(&dev->slow_path_mutex);
                mr = reg_create(NULL, pd, virt_addr, length, umem, ncont,
                                page_shift, access_flags);
+               mutex_unlock(&dev->slow_path_mutex);
+       }
 
        if (IS_ERR(mr)) {
                err = PTR_ERR(mr);