Re: [PATCH v2 REPOST 12/12] x86/hvm/ioreq: add a new mappable resource type...

From: George Dunlap <george.dunlap@citrix.com>
To: Paul Durrant <Paul.Durrant@citrix.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>,
	Wei Liu <wei.liu2@citrix.com>,
	Andrew Cooper <Andrew.Cooper3@citrix.com>,
	"Tim (Xen.org)" <tim@xen.org>, Jan Beulich <jbeulich@suse.com>,
	"xen-devel@lists.xenproject.org" <xen-devel@lists.xenproject.org>,
	Ian Jackson <Ian.Jackson@citrix.com>,
	Roger Pau Monne <roger.pau@citrix.com>
Subject: Re: [PATCH v2 REPOST 12/12] x86/hvm/ioreq: add a new mappable resource type...
Date: Tue, 29 Aug 2017 15:38:07 +0100	[thread overview]
Message-ID: <CAFLBxZbcVyrrCg=DPPcw_GNjwoAY5CnVVnyN6Bs9BXmhmc-ZGg@mail.gmail.com> (raw)
In-Reply-To: <2751163196024f0d96a457c073badf61@AMSPEX02CL03.citrite.net>

On Tue, Aug 29, 2017 at 3:31 PM, Paul Durrant <Paul.Durrant@citrix.com> wrote:
>> -----Original Message-----
> [snip]
>> >> I'm not terribly happy with allocating out-of-band pages either.  One
>> >> of the advantages of the way things are done now (with the page
>> >> allocated to the guest VM) is that it is more resilient to unexpected
>> >> events:  If the domain dies before the emulator is done, you have a
>> >> "zombie" domain until the process exits.  But once the process exits
>> >> for any reason -- whether crashing or whatever -- the ref is freed and
>> >> the domain can finish dying.
>> >>
>> >> What happens in this case if the dm process in dom0 is killed /
>> >> segfaults before it can unmap the page?  Will the page be properly
>> >> freed, or will it just leak?
>> >
>> > The page is referenced by the ioreq server in the target domain, so it will
>> be freed when the target domain is destroyed.
>>
>> I don't understand how you're using the terms... I would have
>> interpreted 'target domain' to me means the guest VM to which emulated
>> devices are being provided, and 'ioreq server' means the process
>> (perhaps in dom0, perhaps in a stubdomain) which is providing the
>> emulated devices.
>>
>> Did you mean that it's referenced by the ioreq_server struct in the
>> target domain, and so a put_page() will happen when the guest is
>> destroyed?
>
> Terminology issues :-) By 'ioreq server' I mean the infrastructure in Xen, centred around struct ioreq_server. I refer to the dom0 process/stub domain/xengt module/whatever as the 'emulator'.
> So, yes, the fact that the page is referenced in the ioreq server of the target domain means that a put_page() will happen when that domain is destroyed.

OK; in that case:

Acked-by: George Dunlap <george.dunlap@citrix.com>

I think that's the only other Ack you need from me.  Thanks for doing this work.

 -George

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