[lkml]   [2015]   [Apr]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] hrtimer: Replace cpu_base->active_bases with a direct check of the active list

* Thomas Gleixner <> wrote:

> On Thu, 9 Apr 2015, Ingo Molnar wrote:
> > Btw., does cpu_base->active_bases even make sense? hrtimer bases are
> > fundamentally percpu, and to check whether there are any pending
> > timers is a very simple check:
> >
> > base->active->next != NULL
> >
> > So I'd rather suggest taking a direct look at the head, instead of
> > calculating bit positions, masks, etc.
> >
> > Furthermore, we never actually use cpu_base->active_bases as a
> > 'summary' value (which is the main point of bitmasks in general),
> > so I'd remove that complication altogether.
> >
> > This would speed up various hrtimer primitives like
> > hrtimer_remove()/add and simplify the code. It would be a net code
> > shrink as well.
> Well. You trade a bit more code against touching cache lines to
> figure out whether the clock base has active timers or not. So for a
> lot of scenarios where only clock monotonic is used you touch 3
> cache lines for nothing.

In the (typical) case it will touch one extra cacheline - and removes
a fair bit of complexity which 80 bytes (that touches two cachelines):

7502 427 0 7929 1ef9 hrtimer.o.before
7422 427 0 7849 1ea9 hrtimer.o.after

So even if we were to optimize for cache footprint (which isn't the
only factor we optimize for)it looks like a win-win scenario to me,
even if you ignore the speedup and the simpler code structure...


> I'm about to send out a patch which actually makes better use of the
> active_bases field without creating a code size explosion.

So please lets do this series first - it achieves the same thing, with
less cache used and faster code.



 \ /
  Last update: 2015-04-09 11:41    [W:0.121 / U:1.804 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site