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

Re: [Xen-devel] [PATCH v8 00/15] execute hotplug scripts from libxl

Ian Campbell wrote:
> On Tue, 2012-07-10 at 07:31 -0400, Roger Pau Monne wrote:
>> Ian Campbell wrote:
>>> (resending as I don't think I had SMTP setup properly on my laptop -- sorry 
>>> if you get this twice!)
>>> On Wed, 2012-07-04 at 07:59 -0400, Roger Pau Monne wrote:
>>>> This new serie (v8) fixes a code refactoring problem that was present
>>>> in v7 (06/15 changed code introduced by 05/15).
>>>  From somewhere in here I'm seeing timeouts waiting for the b/e to go to
>>> state 5 when doing cd-insert on an HVM guest. I suspect because this is
>>> (or should be) turned into a virtual media change rather than an actual
>>> device remove and insert?
>>> BTW libxl_cdrom_insert hasn't been async'd up yet -- I was actually just
>>> looking into that when I noticed this.
> [...]
>> Yes, this is due to the fact that Qemu (traditional at least) doesn't 
>> honour the connection/disconnection protocol, so neither removing the 
>> frontend or setting the backend to "closing" (5), will make Qemu 
>> disconnect the device. I used to have a special "dev->backend_type == 
>> QDISK" to skip the waiting, I've added it to my series again, and it 
>> should solve the waiting problem.
> Actually I think libxl_cdrom_insert is just broken. For an HVM guest
> with an emulated CDROM (i..e the normal case, even if you have PV
> drivers) then the media change protocol is not to remove the device and
> reinsert it. Instead you are supposed to just change the params key. I
> have half a patch to do this (as part of the asyncification of the
> interface) and I think it will make this special case unnecessary, at
> least or the cdrom case.
> Perhaps you also need it for the disk case though, I don't know. Or
> maybe qemu should be taught to honour the protocol?

I'm working on making vfb/vkb async, once that is done I can take a look
on this, but since the device removal is done after Qemu has been killed
I don't have much expectations that this is going to work (at least for
the shutdown/destroy case).

The block-attach/detach case might work with this approach.

Xen-devel mailing list



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