Re: [PATCH 3/6] cpuset write throttle

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Thu, 2007-10-04 at 00:56 -0700, Paul Jackson wrote:
> Peter wrote:
> > Perhaps you can keep a proportion in the cpu-set, and do a similar trick
> > that the process proportions do.
> 
> Beware -- the following comment is made by someone who has been
> basically zero attention to this thread, so could be --way-- off
> the mark.
> 
> Be that as it may, avoid putting anything in the cpuset that you need
> to get to frequently.  Access to all cpusets in the system is guarded
> by a single global mutex.  The current performance assumption is that
> about the only things that need to access the contents of a cpuset are:
>  1) explicit user file operations on the special files in the cpuset
>     file system, and
>  2) some exceptional situations from slow code paths, such as memory
>     shortage or cpu hotplug events.
> 
> Well ... almost.  If you don't mind the occassional access to the wrong
> cpuset, then just taking the task_lock on the current task will guarantee
> you that the cpuset pointer in the task struct points to --some-- cpuset,
> usually the right one.  This can be (and is) used for some statistic
> gathering purposes (look for 'fmeter' in kernel/cpuset.c), where exact
> counts are not required.  Perhaps that applies here as well.

Ugh, yeah. Its a statistical thing, but task_lock is quite a big lock to
take. Are cpusets RCU freed? In which case we could just rcu_deref the
cpuset pointer and do whatever needs done to whatever we find.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>