On Tue, Dec 18, 2018 at 02:15:56PM +0200, Leon Romanovsky wrote:
> From: Huy Nguyen <huyn@xxxxxxxxxxxx>
>
> On NVMe offloads connection with many IO queues, EEH takes long time to
> recover. The culprit is the synchronize_srcu in the destroy_mkey. Solution
> is to use synchronize_srcu only for ODP mkey.
>
> Fixes: b4cfe447d47b ("IB/mlx5: Implement on demand paging by adding support for MMU notifiers")
> Signed-off-by: Huy Nguyen <huyn@xxxxxxxxxxxx>
> Reviewed-by: Daniel Jurgens <danielj@xxxxxxxxxxxx>
> Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxxxx>
> drivers/infiniband/hw/mlx5/mr.c | 19 ++++++++++++++++---
> 1 file changed, 16 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
> index d45a9163c198..0f01059724e9 100644
> +++ b/drivers/infiniband/hw/mlx5/mr.c
> @@ -73,7 +73,8 @@ static int destroy_mkey(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
>
> #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
> /* Wait until all page fault handlers using the mr complete. */
> - synchronize_srcu(&dev->mr_srcu);
> + if (mr->umem && mr->umem->is_odp)
> + synchronize_srcu(&dev->mr_srcu);
> #endif
We don't need any of these gross #ifdef's..
Add a
static inline bool is_odp_mr(struct mlx5_ib_mr *mr)
{
return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem && mr->umem->is_odp;
}
And use it in place of all the is_odp tests.
Since it returns the constant false if ODP is disabled the compiler
will drop all the branches naturally without the horrible #ifdef mess.
Someone should send a cleanup patch for ODP to get rid of more
#ifdefs.
Jason
