Re: [PATCH 11/15] x86: Rewrite cyc2ns to avoid the need to disable IRQs

[Peter Zijlstra]

On Mon, Jun 16, 2014 at 10:43:38PM +0530, Viresh Kumar wrote:
> Cc'ing Mauro/Rafael/Srivatsa..
> 
> On Thu, Dec 12, 2013 at 7:38 PM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > Use a ring-buffer like multi-version object structure which allows
> > always having a coherent object; we use this to avoid having to
> > disable IRQs while reading sched_clock() and avoids a problem when
> > getting an NMI while changing the cyc2ns data.
> >
> >                         MAINLINE   PRE        POST
> >
> >     sched_clock_stable: 1          1          1
> >     (cold) sched_clock: 329841     331312     257223
> >     (cold) local_clock: 301773     310296     309889
> >     (warm) sched_clock: 38375      38247      25280
> >     (warm) local_clock: 100371     102713     85268
> >     (warm) rdtsc:       27340      27289      24247
> >     sched_clock_stable: 0          0          0
> >     (cold) sched_clock: 382634     372706     301224
> >     (cold) local_clock: 396890     399275     399870
> >     (warm) sched_clock: 38194      38124      25630
> >     (warm) local_clock: 143452     148698     129629
> >     (warm) rdtsc:       27345      27365      24307
> >
> > Signed-off-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > ---
> >  arch/x86/include/asm/timer.h     |   23 +++
> >  arch/x86/kernel/cpu/perf_event.c |   14 +-
> >  arch/x86/kernel/tsc.c            |  229 ++++++++++++++++++++++++++++++++++-----
> >  3 files changed, 236 insertions(+), 30 deletions(-)
> 
> Hi Peter,
> 
> We have been following a bug around cpufreq changes in 3.14
> (https://bugzilla.kernel.org/show_bug.cgi?id=77201) and our tests
> narrowed it down to this patch after which things aren't working
> as expected.

> Mauro (registosites@xxxxxxxxxxx) have reported that since 3.14
> after doing hot-unplug/hotplug of CPU1, cpufreq POSTCHANGE notifications
> for every second frequency change hangs his machine ..

Ah, just a freeze?

> Last working tag was: 3.13.8 and it started failing from 3.14 and fails
> for latest 3.15 as well..
> 
> model name : AMD Turion(tm) 64 X2 Mobile Technology TL-64
> CPUFreq driver: powernow-k8 ..
> CPUs: Only two CPUs, sharing clock line

What's specific to this particular CPU?

> Can you give any pointers, so that he can get it fixed ?

So what you can try is force a cyc2ns read before the write in
set_cyc2ns_scale(). I think its possible that if we do not do the read,
the write will wait for a 'free' slot indefinitely.

Attachment: pgpec33pPrjXk.pgp
Description: PGP signature