Re: [Xen-devel] Xen-unstable: irq.c:xxxx: dom1: forcing unbind of pirq 40

[Sander Eikelenboom <linux@xxxxxxxxxxxxxx>]

Tuesday, December 9, 2014, 6:29:08 PM, you wrote:

>>>> On 09.12.14 at 17:24, <linux@xxxxxxxxxxxxxx> wrote:
>> Currently i'm running Xen-unstable with Stefano's libxl fixup patches (not 
>> doing 
>> reset due to qmp race, switching order of xc_domain_irq_permission and 
>> xc_physdev_unmap_pirq)
>> and with a kernel with Konrad's 3.19 patch series. 
>> 
>> Although i have seen these messages before, so i don't know if it is a 
>> recent 
>> regression (don't think so). 
>> 
>> I still see these in xl dmesg when shutting down guests with pci 
>> passthrough:
>> "irq.c:xxxx: dom1: forcing unbind of pirq 40"
>> 
>> It seems to cleanup ok afterwards, but from the sound warning it shouldn't 
>> get
>> to that point. I get these for both PV and HVM guests.

> These messages simply indicate that the guest didn't do some
> expected cleanup itself.

OK, the thing is that at present is does this on every shutdown of guests with 
pci-passthrough, all running a recent kernel. That hasn't always been that way 
as far as i can remember. I will see if i have some old 
guest kernels and see if i get the same there.

> I see these too when Dom0 shuts down,
> at least for certain drivers. For HVM arguably qemu might know
> better and tear things down properly, but in no event are these
> messages - when occurring in connection with a guest going
> down - indications of a problem. We could even see to lower their
> severity or suppress them altogether when the domain is known
> to be dying (you may want to check whether ->is_dying is already
> set at that point).

Just tested, but for both PV and HVM when it enters "unmap_domain_pirq" 
d->is_dying ? 1 : 0 returns 0.

> Jan




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