lkml.org 
[lkml]   [2014]   [Mar]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC] [PATCH] Pre-emption control for userspace
On 03/04/2014 02:12 PM, H. Peter Anvin wrote:
>
> Shades of the Android wakelocks, no?
>
> This seems to effectively give userspace an option to turn preemptive
> multitasking into cooperative multitasking, which of course is
> unacceptable for a privileged process (the same reason why unprivileged
> processes aren't allowed to run at above-normal priority, including RT
> priority.)
>
> I have several issues with this interface:
>
> 1. First, a process needs to know if it *should* have been preempted
> before it calls sched_yield(). So there needs to be a second flag set
> by the scheduler when granting amnesty.

Good idea. I like it. I will add it.

>
> 2. A process which fails to call sched_yield() after being granted
> amnesty must be penalized.

I agree. Is it fair to say that such a process sees the penalty by being
charged that extra timeslice and being pushed to the right side of RB
tree since its p->se.vruntime would have gone up, which then delays the
time when it can get CPU again? I am open to adding a more explicit
penalty - maybe deny its next preemption delay request if it failed to
call sched_yield() the last time when it should have?

>
> 3. I'm not keen on occupying a full page for this. I'm wondering if
> doing a pointer into user space, futex-style, might make more sense.
> The downside, of course, is what happens if the page being pointed to is
> swapped out.

Using a full page for what is effectively a single bit flag does not sit
well with me either. Doing it through proc forces minimum size of a page
(please correct me there if I am wrong). I will explore your idea some
more to see if that can be made to work.

>
> Keep in mind this HAS to be per thread.
>

Thanks, hpa!

--
Khalid



\
 
 \ /
  Last update: 2014-03-05 00:42    [W:2.314 / U:0.652 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site