On Tue, 11 Dec 2007 00:34:33 +0100
Stefano Brivio <stefano.brivio@polimi.it> wrote:

quoted text
> On Tue, 11 Dec 2007 00:04:25 +0100
> Ingo Molnar <mingo@elte.hu> wrote:
> > 
> > * Ingo Molnar <mingo@elte.hu> wrote:
> > 
> > > * Andrew Morton <akpm@linux-foundation.org> wrote:
> > > 
> > > > > what do you think? Right now i've got them queued up for
> > > > > 2.6.25 in both the scheduler-devel and the x86-devel git
> > > > > trees - but can submit them for 2.6.24 if it's better if we
> > > > > did them there. I've got no strong opinion either way.
> > > > 
> > > > printk_clock() doesn't seem terribly important but what's this
> > > > stuff about effects on udelay/mdelay?  That can be serious if
> > > > they're getting shortened.
> > > 
> > > since udelay depends on loops_per_jiffy, which is fixed up 
> > > time_cpufreq_notifier(), i dont see how it could be affected by 
> > > frequency changes. (but that's the theory - practice might be 
> > > different)
> > 
> > Stefano Brivio reported udelay()/mdelay() effects in the b43
> > driver. (and it caused driver failures for him.)
> > 
> > Stefano, could you please try to sum up your experiences with that 
> > issue? Is it reproducable, and the 5 patches i did fix it? (if yes, 
> > could you try to re-do the mdelay verifications perhaps, to make
> > sure it's not some other effect interacting here. In theory
> > sched-clock scaling has no effect on udelay behavior.)
> 
> Sorry for disappearing. Anyway, yes, those patches fixed it.
> Precision in delays isn't that good when using my crappy unstable TSC
> (mdelay(2000) causes delays between 2 and 2.9 seconds) but it's not
> depending on frequency changes anymore. So I'd say it's fixed, but
> please tell me if you want me to do any other test so as to be sure
> it is.
> 
> 
I'm still quite concerned about this in dual/quad core scenarios;
the frequency of both cores is the maximum of what linux sets each core to;
this means that if you're THIS sensitive to that there still is quite a nasty issue there.

I wonder if the various delay functions (maybe only in .25) should use the maximum observed loops_per_jiffie instead always (across cpus) to be super safe here.

-- 
If you want to reach me at my work email, use arjan@linux.intel.com
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/