On 5/14/2018 6:09 PM, Rohit Zambre wrote:
Great, thanks!
-Rohit
The PR was merged, thanks.
On Mon, May 14, 2018 at 9:53 AM, Yishai Hadas
<yishaih@xxxxxxxxxxxxxxxxxx> wrote:
On 5/2/2018 2:17 AM, Rohit Zambre wrote:
The primary motivation of this series is to disable the lock on a QP that
is created under a Thread Domain.
Currently, the only way to disable any locking is through the global
MLX5_SINGLE_THREADED environment variable, which is unhelpful in
disabling locks of a subset of verbs objects while enabling locks on
others. The first patch enables the driver to control locking on each
individual verbs object.
The second patch uses the changes from the first to disable locking on the
QP when it is created within a Thread Domain.
Changes since v1:
- introduce mlx5_spinlock_init_pd to compute whether or not lock is needed
based on the Protection Domain passed by the object
- replace mlx5_spinlock_init with mlx5_spinlock_init_pd for all objects
that are associated with a Protection Domain
- update commit message of patch 2 to describe mlx5_spinlock_init_pd
Rohit Zambre (2):
mlx5: Allow individual locking-control for different verbs objects
mlx5: Disable locking on a QP if it is assigned to a thread domain
providers/mlx5/mlx5.c | 4 ++--
providers/mlx5/mlx5.h | 22 +++++++++++++++++++---
providers/mlx5/verbs.c | 12 ++++++------
3 files changed, 27 insertions(+), 11 deletions(-)
I have opened a PR [1] which includes your series and the fix from Eli
regarding the 'need_uuar_lock' when there are no medium bfregs that you
reported on.
I plan to merge it tomorrow if no comments will come.
[1] https://github.com/linux-rdma/rdma-core/pull/327
Note: The Travis error doesn't look related to this series.
Yishai
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
