Messages in this thread | | | Date | Fri, 5 Jun 2015 12:07:07 +0200 | From | Ingo Molnar <> | Subject | Re: [BUG, bisect] hrtimer: severe lag after suspend & resume |
| |
* Thomas Gleixner <tglx@linutronix.de> wrote:
> On Thu, 4 Jun 2015, John Stultz wrote: > > On Wed, Jun 3, 2015 at 5:56 PM, Jeremiah Mahler <jmmahler@gmail.com> wrote: > > So I suspect the problem is the change to clock_was_set_seq in > > timekeeping_update is done prior to mirroring the time state to the > > shadow-timekeeper. Thus the next time we do update_wall_time() the > > updated sequence is overwritten by whats in the shadow copy. The > > attached patch moving the modification up seems to avoid the issue for > > me. > > Duh, yes. > > > Thomas: Looking at the problematic change, I'm not a big fan of it. Caching > > timekeeping state here in the hrtimer code has been a source of bugs in the > > past, and I'm not sure I see how avoiding copying 24bytes is that big of a > > win. Especially since it adds more state to the timekeeper and hrtimer base > > that we have to read and mange. > > It's not about copying 24 bytes. It's about touching 3 cache lines for nothing. > In situations where we run high frequency periodic timers on clock monotonic and > nothing is going on in the other clock domains, which is a pretty common > situation, this is measurable in terms of cache utilization. [...]
It's not just about 'touching': it's about _dirtying_ cachelines from a globally executed function (timekeeping), which is then accessed by per-CPU functionality (hrtimers).
That makes it far more expensive, it has similar scalability limiting effects as a global lock - while if we do it smart it can perform as essentially lockless code in most cases.
Thanks,
Ingo
| |