Re: 3.4.70+ kernel WARNING spew dysfunction on failed migration

[David Vrabel <david.vrabel@citrix.com>]

On 07/01/14 18:55, Ian Jackson wrote:
> I did the following test:
> 
>    mv /etc/xen/scripts/block /etc/xen/scripts/block.aside
>    xl migrate debian.guest.osstest localhost
> 
> xl did what appears to be the right thing: it did most of the
> migration, failed to run the block scripts at the end of the
> migration, and destroyed the destination domain and instead resumed
> the source guest.
> 
> However, the source guest immediately went mad spewing WARNINGs and
> was after that no longer contactable via the network and not
> apparently responsive on the console.  See below.
> 
> This is with:
> 
>   [    0.000000] Linux version 3.4.70+ (osstest@rice-weevil) (gcc
>   version 4.4.5 (Debian 4.4.5-8) ) #1 SMP Wed Dec 4 03:14:51 GMT 2013
> 
> For reasons I don't understand it doesn't seem to print the actual
> kernel git hash in dmesg, but I think it was that from flight 22264,
> i.e.  234d96ee0f3b8e49501d068a2a3165aa4db60903.  It's i386, on a
> 64-bit Xen.
> 
> Thanks,
> Ian.
> 
> debian login: [  124.595658] PM: freeze of devices complete after 2.980 msecs
> [  124.595991] PM: late freeze of devices complete after 0.013 msecs
> [  124.600919] PM: noirq freeze of devices complete after 4.884 msecs
> [  124.601105] Grant tables using version 2 layout.
> [  124.601105] ------------[ cut here ]------------
> [  124.601105] kernel BUG at drivers/xen/events.c:1582!
> [  124.601105] invalid opcode: 0000 [#1] SMP 
> [  124.601105] Modules linked in: [last unloaded: scsi_wait_scan]
> [  124.601105] 
> [  124.601105] Pid: 6, comm: migration/0 Not tainted 3.4.70+ #1  
> [  124.601105] EIP: 0061:[<c12f5d25>] EFLAGS: 00010082 CPU: 0
> [  124.601105] EIP is at xen_irq_resume+0x215/0x370

We shouldn't be calling xen_irq_resume() when resuming the source VM.
The EVTCHNOP_bind_irq is failing because the VIRQ is still bound.

This would suggest that the suspend hypercall has not correctly returned
the cancelled state.

Could this be because of the tools issue mentioned by Ian C?

David