Re: [PATCH 1/1] xc_domain_restore: Allow QEMU to increase memory

From: Ian Campbell <ian.campbell@citrix.com>
To: George Dunlap <george.dunlap@eu.citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>,
	Stefano Stabellini <stefano.stabellini@eu.citrix.com>,
	Andrew Cooper <andrew.cooper3@citrix.com>,
	Ian Jackson <ian.jackson@eu.citrix.com>,
	Don Slutz <dslutz@verizon.com>,
	"xen-devel@lists.xen.org" <xen-devel@lists.xen.org>,
	Shriram Rajagopalan <rshriram@cs.ubc.ca>,
	Hongyang Yang <yanghy@cn.fujitsu.com>
Subject: Re: [PATCH 1/1] xc_domain_restore: Allow QEMU to increase memory
Date: Thu, 16 Apr 2015 10:29:15 +0100	[thread overview]
Message-ID: <1429176555.25195.56.camel@citrix.com> (raw)
In-Reply-To: <552F7D8B.7010907@eu.citrix.com>

On Thu, 2015-04-16 at 10:14 +0100, George Dunlap wrote:
> On 04/16/2015 10:00 AM, Ian Campbell wrote:
> > On Wed, 2015-04-15 at 17:52 +0100, Wei Liu wrote:
> >> On Wed, Apr 15, 2015 at 05:45:15PM +0100, Ian Campbell wrote:
> >>> On Wed, 2015-04-15 at 17:36 +0100, Wei Liu wrote:
> >>>> On Wed, Apr 15, 2015 at 03:34:48PM +0100, Ian Campbell wrote:
> >>>>> On Tue, 2015-04-14 at 18:54 +0100, Wei Liu wrote:
> >>>>>> Let's see if we can record this in xc image format. I haven't looked,
> >>>>>> but George mentioned that it might be possible to do so.
> >>>>>
> >>>>> Can this not be done at the save stage in the bit where we update the
> >>>>> JSON to reflect the actual configuration?
> >>>>>
> >>>>
> >>>> Yes, it's possible to do that. The value is not in libxl idl simply
> >>>> because it's always not there.
> >>>
> >>> We can't set the existing maxmem then?
> >>>
> >>
> >> If the "we" in your question means an applications that use libxl's
> >> public interface, then no. It's not in IDL. It's not in xenstore. That
> >> value is currently managed by libxl + libxc + xen.
> > 
> > I was thinking of libxl_domain_build_info.max_memkb, which is in a
> > struct which is marshalled over the wire and which could be updated on
> > save to reflect the actual usage.
> > 
> > Does using that field not work for some reason?
> 
> The problem, I think, is that max_memkb says how much memory is
> *reported to the guest*.  So what happens when the VM on the other side
> reboots?  Won't libxl use this newly-increased max_memkb as the memory
> to be reported *to the guest*?  And then qemu will have to allocate *yet
> more* memory for its roms?

Ah, yes, I think you are right, that's pretty dumb.

> Meaning that the size of the VM will increase by a few kB (MB?) every
> time it reboots?
> 
> This is why I think we should try to start making a clear distinction
> between "what the guest sees as RAM" and "what xen sees as memory"; And
> I proposed using "memory" for what the guest sees, and "pages" for what
> Xen sees.

That might help, as might my remembering about
docs/misc/libxl_memory.txt more often.

Ian.