Re: suspend regression in 4.1-rc1

[Stephane Eranian]

Hi,

On Mon, May 18, 2015 at 4:05 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> On Mon, May 18, 2015 at 06:56:46AM -0400, Ulrich Obergfell wrote:
>> > Subject: watchdog: Fix merge 'conflict'
>> >
>> > Two watchdog changes that came through different trees had a non
>> > conflicting conflict, that is, one changed the semantics of a variable
>> > but no actual code conflict happened. So the merge appeared fine, but
>> > the resulting code did not behave as expected.
>> >
>> > Commit 195daf665a62 ("watchdog: enable the new user interface of the
>> > watchdog mechanism") changes the semantics of watchdog_user_enabled,
>> > which thereafter is only used by the functions introduced by
>> > b3738d293233 ("watchdog: Add watchdog enable/disable all functions").
>>
>> Don and I already posted a patch in April to address this:
>>
>> https://lkml.org/lkml/2015/4/22/306
>> http://ozlabs.org/~akpm/mmots/broken-out/watchdog-fix-watchdog_nmi_enable_all.patch
>
> Yeah, but it seems to have gotten lost on its way to Linus.
>
>> > There further appears to be a distinct lack of serialization between
>> > setting and using watchdog_enabled, so perhaps we should wrap the
>> > {en,dis}able_all() things in watchdog_proc_mutex.
>>
>> As I understand it, the {en,dis}able_all() functions are only called early
>> at kernel startup, so I do not see how they could be racing with watchdog
>> code that is executed in the context of write() system calls to parameters
>> in /proc/sys/kernel. Please see also my earlier reply to Michal for further
>> details: http://marc.info/?l=linux-pm&m=143194387208250&w=2
>>
>> Do we really need synchronization here?
>
> Same argument as in my previous email; its best to implement exposed
> functions fully and correctly, irrespective of their usage sites.
>
> It costs little extra and might safe a few hairs down the lined. None of
> this is performance critical.

I cannot reproduce this problem on my T430s running tip.git at 4.1-rc3.

The thing about b37609c30e41 is that is introduces a deferred initcall
for perf_events. It adds an subsys_initcall after the default initialization
of perf_events The reason is that the fixup_ht_bug() needs to wait until
cpu topology is setup before proceeding. Thus by the time
watchdog_nmi_disable_all() is called from that function, the kernel
may be multi-cpu already. Thus, there may be a race.




commit b37609c30e41264c4df4acff78abfc894499a49b
Author: Stephane Eranian <eranian@xxxxxxxxxx>
Date:   Mon Nov 17 20:07:04 2014 +0100
   perf/x86/intel: Make the HT bug workaround conditional on HT enabled
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/