[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] x86/HPET: mask interrupt while changing affinity



>>> On 20.03.13 at 00:24, Andrew Cooper <andrew.cooper3@xxxxxxxxxx> wrote:
> On 19/03/13 22:48, Sander Eikelenboom wrote:
>> Tuesday, March 19, 2013, 5:00:33 PM, you wrote:
>>
>>>>>> On 19.03.13 at 16:53, Sander Eikelenboom <linux@xxxxxxxxxxxxxx> wrote:
>>>> Could this change have a averse affect on AMD systems ?
>>> It shouldn't, ...
>>>> With this patch booting the dom0 kernel slowly seems to come to a halt 
>>>> (sometime trying to mount rootfs, sometimes a little further trying to 
>>>> bring 
> 
>>>> networking up.)
>>> ... but apparently does (apart from also having the intended effect
>>> of eliminating vector-without-IRQ warnings). But it's not obvious
>>> how that would be - after all, the two added calls should be pretty
>>> benign performance wise.
>>>> I don't see any evident warnings or errors, reverting this commit makes 
>>>> the 
>>>> system boot OK again.
>>> Does the watchdog work on it? If so, could you see whether enabling
>>> that catches something? Or else, do the debug keys still work when
>>> the box stopped?
>> Yes they still work, and somehow using the debug keys seems to make it 
> continu for a bit (slowly and ending up in infinite loop printing firewall 
> messages)
>>
>> Serial log attached, hope i have used the debug keys you are interested in, 
> if not please do specify ...
>>
>> --
>> Sander
> 
> hpet_msi_mask() flips the Timer Interrupt Enable bit, which causes the
> timer not to generate interrupts, but to continue running.
> 
> Are we by any chance suffering from a bad interaction of oneshot timers
> not being re-armed for exceedingly long periods of time ?

See the response/patch I just sent - I don't think it's the lack of
re-arming (which is only an indirect effect), but the missing of
interrupts. Apparently this can be close to negligible (on the two
machines I tested this on) to having a severe effect (on Sander's)
system.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.