From bb7e22a8ab00ff9ba911a45ba8784cef9e6d6f7a Mon Sep 17 00:00:00 2001 From: Huy Nguyen Date: Tue, 18 Dec 2018 14:15:56 +0200 Subject: [PATCH] IB/mlx5: Fix long EEH recover time with NVMe offloads On NVMe offloads connection with many IO queues, EEH takes long time to recover. The culprit is the synchronize_srcu in the destroy_mkey. The 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 Reviewed-by: Daniel Jurgens Signed-off-by: Leon Romanovsky Signed-off-by: Jason Gunthorpe --- 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 fd6ea1f75085..1bd8c1b1dba1 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ 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 return err; @@ -237,6 +238,9 @@ static void remove_keys(struct mlx5_ib_dev *dev, int c, int num) { struct mlx5_mr_cache *cache = &dev->cache; struct mlx5_cache_ent *ent = &cache->ent[c]; +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + bool odp_mkey_exist = false; +#endif struct mlx5_ib_mr *tmp_mr; struct mlx5_ib_mr *mr; LIST_HEAD(del_list); @@ -249,6 +253,10 @@ static void remove_keys(struct mlx5_ib_dev *dev, int c, int num) break; } mr = list_first_entry(&ent->head, struct mlx5_ib_mr, list); +#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING + if (mr->umem && mr->umem->is_odp) + odp_mkey_exist = true; +#endif list_move(&mr->list, &del_list); ent->cur--; ent->size--; @@ -257,7 +265,8 @@ static void remove_keys(struct mlx5_ib_dev *dev, int c, int num) } #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - synchronize_srcu(&dev->mr_srcu); + if (odp_mkey_exist) + synchronize_srcu(&dev->mr_srcu); #endif list_for_each_entry_safe(mr, tmp_mr, &del_list, list) { @@ -572,6 +581,7 @@ static void clean_keys(struct mlx5_ib_dev *dev, int c) { struct mlx5_mr_cache *cache = &dev->cache; struct mlx5_cache_ent *ent = &cache->ent[c]; + bool odp_mkey_exist = false; struct mlx5_ib_mr *tmp_mr; struct mlx5_ib_mr *mr; LIST_HEAD(del_list); @@ -584,6 +594,8 @@ static void clean_keys(struct mlx5_ib_dev *dev, int c) break; } mr = list_first_entry(&ent->head, struct mlx5_ib_mr, list); + if (mr->umem && mr->umem->is_odp) + odp_mkey_exist = true; list_move(&mr->list, &del_list); ent->cur--; ent->size--; @@ -592,7 +604,8 @@ static void clean_keys(struct mlx5_ib_dev *dev, int c) } #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING - synchronize_srcu(&dev->mr_srcu); + if (odp_mkey_exist) + synchronize_srcu(&dev->mr_srcu); #endif list_for_each_entry_safe(mr, tmp_mr, &del_list, list) {