Re: [RFC PATCH v1 4/8] sched/proxy: Tell the class about a proxy migration
From: Mathieu Desnoyers
Date: Tue Nov 17 2026 - 08:21:11 EST
On 2026-11-16 15:42, John Stultz wrote:
> So I've got a pending fix from Yu-Ting in my tree (apologies for the
> gerrit link, I've not cleaned up and pushed my tree out yet) that I
> believe duplicates the bit above:
> https://r.android.com/4410518
Fine with me. pelt_migration (case 5) passes on your tree, so please
drop the core.c part of this patch.
>> - if (unlikely(dl_task_bw_elsewhere(p, new_cpu)))
>> + if (unlikely(dl_task_bw_elsewhere(p, new_cpu)) &&
>> + !(p->is_blocked && task_on_rq_migrating(p)))
>> dl_task_move_bw(p, new_cpu);
>
> I may need to pull this bit in separately.
You should not need it. On tip, migrate_task_rq_dl() returns at once
for a task that is not TASK_WAKING. So a proxy migration that calls it
does nothing there. Only my deadline series makes it move bandwidth, so
I will carry this check in that series.
The check leaves the bandwidth where it is. task_woken_dl() moves it
when the task wakes in another root domain. That needs the full wake-up
path for deadline tasks, which dropping the shortcut of patch 2 keeps.
Thanks,
Mathieu
--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com