diff options
| author | Yishai Hadas <yishaih@nvidia.com> | 2021-02-02 09:13:09 +0200 |
|---|---|---|
| committer | Jason Gunthorpe <jgg@nvidia.com> | 2021-02-08 20:31:11 -0400 |
| commit | db72438c9319cfd37e3c237a7754ca862ae12d63 (patch) | |
| tree | c5fe461d002f4fb853f9b944da24d4423cc66bec /drivers/infiniband/hw/mlx5/mlx5_ib.h | |
| parent | a5887d62072e09a9d3b01549ac5d555bbeb9be3d (diff) | |
RDMA/mlx5: Cleanup the synchronize_srcu() from the ODP flow
Cleanup the synchronize_srcu() from the ODP flow as it was found to be a
very heavy time consumer as part of dereg_mr.
For example de-registration of 10000 ODP MRs each with size of 2M hugepage
took 19.6 sec comparing de-registration of same number of non ODP MRs that
took 172 ms.
The new locking scheme uses the wait_event() mechanism which follows the
use count of the MR instead of using synchronize_srcu().
By that change, the time required for the above test took 95 ms which is
even better than the non ODP flow.
Once fully dropped the srcu usage, had to come with a lock to protect the
XA access.
As part of using the above mechanism we could also clean the
num_deferred_work stuff and follow the use count instead.
Link: https://lore.kernel.org/r/20210202071309.2057998-1-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Diffstat (limited to 'drivers/infiniband/hw/mlx5/mlx5_ib.h')
| -rw-r--r-- | drivers/infiniband/hw/mlx5/mlx5_ib.h | 31 |
1 files changed, 23 insertions, 8 deletions
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index ab52083634e6..88cc26e008fc 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -684,11 +684,8 @@ struct mlx5_ib_mr { u64 pi_iova; /* For ODP and implicit */ - atomic_t num_deferred_work; - wait_queue_head_t q_deferred_work; struct xarray implicit_children; union { - struct rcu_head rcu; struct list_head elm; struct work_struct work; } odp_destroy; @@ -1068,11 +1065,6 @@ struct mlx5_ib_dev { u64 odp_max_size; struct mlx5_ib_pf_eq odp_pf_eq; - /* - * Sleepable RCU that prevents destruction of MRs while they are still - * being used by a page fault handler. - */ - struct srcu_struct odp_srcu; struct xarray odp_mkeys; u32 null_mkey; @@ -1599,6 +1591,29 @@ static inline bool mlx5_ib_can_reconfig_with_umr(struct mlx5_ib_dev *dev, return true; } +static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev, + struct mlx5_core_mkey *mmkey) +{ + refcount_set(&mmkey->usecount, 1); + + return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key), + mmkey, GFP_KERNEL)); +} + +/* deref an mkey that can participate in ODP flow */ +static inline void mlx5r_deref_odp_mkey(struct mlx5_core_mkey *mmkey) +{ + if (refcount_dec_and_test(&mmkey->usecount)) + wake_up(&mmkey->wait); +} + +/* deref an mkey that can participate in ODP flow and wait for relese */ +static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_core_mkey *mmkey) +{ + mlx5r_deref_odp_mkey(mmkey); + wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0); +} + int mlx5_ib_test_wc(struct mlx5_ib_dev *dev); static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev) |