Re: [PATCH v4 10/26] sched/deadline: Release the bandwidth of a task that leaves the class asleep

[Mathieu Desnoyers]

On 2026-11-06 21:23, sashiko-bot@xxxxxxxxxx wrote:
> dl_zero_lag() mixes PI donor state with kept task params

The 0-lag time here comes from the deadline and runtime the task ran on
during the boost, against its own bandwidth. task_non_contending() uses
the same values upstream, for a queued task at the deboost and for a
boosted task that blocks. This patch only adds the path of a task that
is not queued, which leaked before.

If the time is ahead, the runtime left fits the task's own bandwidth, so
a task that comes back gains nothing. If it is past, the bandwidth goes
with the parameters. The cost is the 0-lag time from before the boost:
switched_to_dl() cancels the timer that held it. So the release can come
up to one period of the task early, or late until the donor's deadline.
Keeping that time across a boost would change how the inactive timer
treats a boosted task. I would leave that out of this series.

Thanks,

Mathieu

--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com