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

Re: [Xen-devel] cpuidle and un-eoid interrupts at the local apic



Andrew Cooper wrote on 2013-09-05:
> On 04/09/2013 20:56, Thimo E. wrote:
>> Hello Andrew,
>> 
>> thanks for your response. At least I've seen the trigger of the new
>> crash (2e) already before, so they seem so belong together.
>> 
>> I can't image that I am the only one on the world who is using a
>> haswell board. And as I haven't seen any other Xen bug/crash reports
>> like mine (and one time you) nor bug reports from users with other
>> operating systems, I ask myself if only my hardware is buggy or if
>> other operating systems handle those "spurious" interrupts in
>> another way ?!?!
>> 
>> What does " ioapic_ack=old" change ?
>> 
>> Best regards
>>   Thimo
> 
> ioapic_ack=old is already in effect - see "Enabled directed EOI with
> ioapic_ack_old on!" in the boot dmesg.
> 
> Originally, it was a bugfix workaround for ancient IO-APIC hardware
> which had a bug on one of the mask bits.  Nowadays, it is used with
> EOI broadcast suppression, which is a APIC transaction performance
> improvement on recent processors.  What it does is affect whether an
> IO-APIC interrupt gets masked when an interrupt is received.
> 
> You could certainly try "ioapic_ack=new" and see whether that makes a
> difference, given a lack of any other ideas.  It will disable EOI
> broadcast suppression.
> 
> ~Andrew
Hi Thimo

Did you see this issue if and only if HVM guest running? If yes, can you try to 
isolate the dom0 VCPUs and HVM guest's VCPUs? For example, pin all dom0's VCPUs 
to some PCPUs and pin all HVM guest's VCPUs to the remain PCPUs.

BTW: you didn't try the device pass-through when the issue occurs?

Best regards,
Yang



_______________________________________________
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®.