v2:
  - It turns out it works around the v1 cpuset missing offline cpu
    problem better than I originally thought. So the patch description
    is updated accordingly.

This small patch series makes the cpuset hotplug a bit more efficient
by eliminating unnecessary task iteration and cpu/node masks update
when a cpu hotplug event (online/offline) happens.

It can also largely work around the known problem of missing previously
offlined cpus in v1 cpuset with some exceptions.

Waiman Long (2):
  cgroup/cpuset: Skip task update if hotplug doesn't affect current
    cpuset
  cgroup/cpuset: Don't update tasks' cpumasks for cpu offline events

 kernel/cgroup/cpuset.c | 31 +++++++++++++++++++++++--------
 1 file changed, 23 insertions(+), 8 deletions(-)

-- 
2.31.1

Hi Waiman,

On Thu, Feb 02, 2023 at 09:31:58AM -0500, Waiman Long wrote:
>  v2:
>   - It turns out it works around the v1 cpuset missing offline cpu
>     problem better than I originally thought. So the patch description
>     is updated accordingly.
> 
> This small patch series makes the cpuset hotplug a bit more efficient
> by eliminating unnecessary task iteration and cpu/node masks update
> when a cpu hotplug event (online/offline) happens.
> 
> It can also largely work around the known problem of missing previously
> offlined cpus in v1 cpuset with some exceptions.

I can't tell whether this is intended to address the issue I initially
reported at [1], so I gave it a spin anyway and it doesn't seem to help.

If it's unrelated, then sorry for the noise!

Will

[1] https://lore.kernel.org/r/20230117160825.GA17756@willie-the-truck

On Thu, Feb 02, 2023 at 09:31:58AM -0500, Waiman Long wrote:
>  v2:
>   - It turns out it works around the v1 cpuset missing offline cpu
>     problem better than I originally thought. So the patch description
>     is updated accordingly.
> 
> This small patch series makes the cpuset hotplug a bit more efficient
> by eliminating unnecessary task iteration and cpu/node masks update
> when a cpu hotplug event (online/offline) happens.
> 
> It can also largely work around the known problem of missing previously
> offlined cpus in v1 cpuset with some exceptions.

Imma hold on these patches for now. Let's see how the regression fix goes
and revisit after the merge window.

Thanks.

-- 
tejun