lkml.org 
[lkml]   [2015]   [Jan]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v4 10/11] perf/x86/intel: Perform rotation on Intel CQM RMIDs
On Fri, Jan 09, 2015 at 03:24:42PM +0000, Matt Fleming wrote:
> On Fri, 09 Jan, at 02:02:50PM, Peter Zijlstra wrote:
> > On Fri, Jan 09, 2015 at 12:14:01PM +0000, Matt Fleming wrote:
> > > On Tue, 06 Jan, at 06:17:12PM, Peter Zijlstra wrote:
> > > >
> > > > afaict the again label will try and steal yet another rmid, if rmids
> > > > don't decay fast enough, we could end up with all rmids on the limbo
> > > > list and none active. Or am I missing something here?
> > >
> > > You're not missing anything, that's true, we will try and steal more
> > > RMIDs. We could perhaps put a limit on how many RMIDs we're willing to
> > > steal, but I think it should definitely be > 1 because RMIDs can
> > > stabilize out of order.
> > >
> > > It's worth pointing out that we only steal more RMIDs if the ones on the
> > > limbo list have been queued for the "minimum queue time" - it really is
> > > a last resort.
> >
> > Do we really care? Why not just hold up everything until the one(s) we
> > have are low enough?
> >
> > Yes it all blows, but would not some active be better than none active,
> > just because the stupid lines aren't clearing fast enough?
>
> Right, but now we need a "steal limit", so we know when to stop stealing
> active RMIDs.
>
> (cqm_max_rmid + 1) / 4 ?
>
> I guess any limit is better than no limit.

Yeah, that'll work, when the free+limbo count is 1/4th the total we
should stop pulling more plugs.


\
 
 \ /
  Last update: 2015-01-09 17:21    [W:0.237 / U:0.452 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site