[PATCH 0/2] powercap: idle_inject: Create threads on demand

[Thomas Gleixner]

On Thu, Nov 05 2026 at 10:58, Andrew Morton wrote:
> On Thu, 5 Nov 2026 19:46:13 +0100 Daniel Lezcano <daniel.lezcano at oss.qualcomm.com> wrote:
>
>> On 11/3/26 00:09, Daniel Lezcano wrote:
>> > Idle injection currently registers per-CPU kernel threads during early
>> > initialization, including on CPUs for which no idle injection device
>> > will be registered.
>> > 
>> > This series makes thread creation follow the lifetime and CPU mask of
>> > each idle injection device.
>> > 
>> > The first patch adds an optional CPU mask to smpboot thread descriptors.
>> > The mask restricts thread creation, CPU hotplug handling and teardown,
>> > while preserving the existing behavior for descriptors without a mask.
>> > 
>> > The second patch uses this support to register idle injection threads
>> > when a device registers and destroy them when it unregisters. CPU
>> > ownership claims, registration rollback and thread teardown are
>> > serialized to prevent overlapping registrations from sharing threads.
>> > 
>> > Validation includes W=1 builds of the affected objects, whitespace
>> > checks and static review of the teardown locking paths. Concurrent
>> > registration and CPU hotplug runtime testing remains outstanding.
>> > 
>> > The series was tested on Rockchip RK3588.
>> 
>> One thing not specified in this cover is which tree these changes have 
>> to go through ?
>
> I'd volunteer Rafael's, after peterz/tglx review.
>
> Please!  Removing one kernel thread per cpu makes the world a better
> place.

I'll look at it in the next days with brain awake...