On 01.09.26 08:21, Vincent Guittot wrote:
> On Mon, 31 Aug 2026 at 18:54, Yury Norov <ynorov@nvidia.com> wrote:
>>
>> On Mon, Aug 31, 2026 at 05:09:41PM +0200, Vincent Guittot wrote:
>>
>>>> Dietmar/Vincent,
>>>> Do you think it makes sense to enable the driver on ARM64 now?
>>>> Or you think it is better to delay it and once the feature is stable
>>>> ARM ecosystem can enable it?
>>>
>>> It's always better to support all arch by default, unless something is
>>> missing which is not the case here.
>>
>> ARM64 testing is obviously missed.
> 
> But the cpumask is already available not like if you need to create a new one

IMHO, when testing the steal_governor on arm64 w/o
`allow_mismatched_32bit_el0`, I wouldn't expect much difference in this
respect compared to the architectures already tested.

Also, AFAIK, `allow_mismatched_32bit_el0` was mainly relevant for
Android devices up to Android 13 that supported 32-bit userspace, so I
wouldn't expect it to be commonly enabled on recent devices.

As Vincent pointed out, `task_cpu_possible_mask(p)` (which defaults to
`cpu_possible_mask`) is already used in `kernel/sched/core.c` and
`kernel/cgroup/cpuset.c` to support `allow_mismatched_32bit_el0`.