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

Re: [Xen-devel] Xen-unstable: pci-passthrough "irq 16: nobody cared" on HVM guest shutdown on irq of device not passed through.

>>> On 26.09.14 at 11:18, <linux@xxxxxxxxxxxxxx> wrote:
> Friday, September 26, 2014, 8:59:22 AM, you wrote:
>>>>> On 26.09.14 at 00:09, <linux@xxxxxxxxxxxxxx> wrote:
>>> - Tried switching off the onboard soundcard in the bios. Now irq16 is not 
>>> bound 
>>>   to any device, but the machine still freezes without any error (on serial 
>>>   console with sync-console on, triple ctrl-a also doesn't work anymore)
>> I suppose that hang is with irqpoll still in use? Ctrl-a not working anymore
>> makes me wonder whether you use a PCI serial card sitting on that same
>> IRQ for the Xen console... 
> With and without irqpoll in use. 

Oh, even without. That's worrying indeed. But again, the main thing
to understand is who sets up and unmasks IRQ 16 when and for
what reason.

> I don't know what happens if there is a race / ordering problem in say 
> xen/iommu/pciback pulling 
> the device from the guest on shutdown while there are still irq's pending ?
> (the code of xen_pciback's release and resetting function for instance seems 
> to do a different 
> ordering compared to vfio-pci's)

That's a different aspect, as long as I recall correctly that the passed
through device isn't itself sitting on IRQ 16.

>> Furthermore in that mode (with supposedly no
>> handler set up for IRQ 16) monitoring (with a little bit of debugging code)
>> how/when IRQ 16 gets setup and unmasked may provide further hints.
> I don't know if i made it clear enough, but without the device occupying 
> irq16 
> it doesn't give the irq16 nobody cared (or any such message or error), it 
> just 
> freezes.

Because likely it doesn't even get that far.


Xen-devel mailing list



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