Re: [Xen-devel] blkback stuck in infinite loop in xen_blkbk_discard()

[Jeremy Fitzhardinge <jeremy@xxxxxxxx>]

On 09/29/2011 02:55 AM, Jan Beulich wrote:
>>>> On 28.09.11 at 23:40, Jeremy Fitzhardinge <jeremy@xxxxxxxx> wrote:
>> I just "xl destroy"d a domain, and now my dom0 kernel is sitting there
>> infinitely spewing:
>>
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>> vbd vbd-83-51712: 5 reading type
>>
>> which seems to be coming from the error case in
>> drivers/block/xen-blkback/xenbus.c:xen_blkbk_discard().
>>
>> I don't know what the backtrace is.  The system seems overall fine,
>> despite the console spew, though I have a pile of dying domains sitting
>> in odd states rather than being cleaned up.
> I wonder how the function gets called during destroy in the first place,
> and how that would prevent destroying a domain (unless there's a
> feedback loop due to the use of xenbus_dev_fatal() here, causing
> some xenstore entry to get written over and over again, triggering
> the respective watch that blkback has active).
>
> But irrespective of this I would think the function should bail without
> doing anything if blkif->blk_backend_type was already set (or
> couldn't get set). But that would then also indicate a more general
> problem in connect(), as that function shouldn't do anything either
> when a domain gets destroyed.

Well, it could have got into that state before I destroyed the domain -
it was a test kernel that crashed very early (before setting up its own
frontend) after booting from pvgrub, so perhaps there was some race in
that handoff.

    J

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel