bug-hurd
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH gnumach] kern/sched_prim: Simplify by not dispatching to idle


From: Samuel Thibault
Subject: Re: [PATCH gnumach] kern/sched_prim: Simplify by not dispatching to idle cpu directly
Date: Sat, 30 Sep 2023 00:45:37 +0200
User-agent: NeoMutt/20170609 (1.8.3)

Damien Zammit, le ven. 29 sept. 2023 05:00:53 +0000, a ecrit:
> This speeds up smp to approximately full speed of uniprocessor!

Ok, but I don't think we want this.

> diff --git a/kern/sched_prim.c b/kern/sched_prim.c
> index 5ee0521a..cdb2ba27 100644
> --- a/kern/sched_prim.c
> +++ b/kern/sched_prim.c
> @@ -1228,8 +1228,7 @@ void update_priority(
>  /*
>   *   thread_setrun:
>   *
> - *   Make thread runnable; dispatch directly onto an idle processor
> - *   if possible.  Else put on appropriate run queue (processor
> + *   Make thread runnable; put on appropriate run queue (processor

Putting the just-woken thread on an idle processor is necessary to
actually get parallelism quickly. Otherwise you'll have to wait for the
idle processor to wake up after its tick period before you'll actually
see parallelism happen.

This looks to me more like the resched-IPI is not properly plumbed and
that's what needs to be fixed.

Of course, people can use this workaround for now to also work on other
parts of SMP questions. But we really do not want to put the idle
question under the carpet.

Samuel



reply via email to

[Prev in Thread] Current Thread [Next in Thread]