Re: pvh dom0: memory leak from iomem map

[Mukesh Rathor <mukesh.rathor@oracle.com>]

On Fri, 6 Jun 2014 11:53:34 +0200
Tim Deegan <tim@xen.org> wrote:

> At 19:12 -0700 on 05 Jun (1401991969), Mukesh Rathor wrote:
> > On Thu, 5 Jun 2014 11:20:32 +0200
> > Tim Deegan <tim@xen.org> wrote:
> > 
> > > At 18:29 -0700 on 03 Jun (1401816588), Mukesh Rathor wrote:
> > > > Hi Tim,
> > > > 
> > > > When building a dom0 pvh, we populate the p2m with 0..N pfns
> > > > upfront. Then in pvh_map_all_iomem, we walk the e820 and map all
> > > > iomem 1:1. As such any iomem range below N would cause those ram
> > > > frames to be silently dropped. Since the holes could be pretty
> > > > big, I am concenred this could result in significant loss of
> > > > frames. 
> > > 
> > > Right.  So, er, don't do that then? :)  You have all the
> > > information you need available at the time that you build dom0's
> > > p2m, so why not
> > 
> > 
> > I thought about that, and wasn't sure how easy it would be to change
> > construct_dom0 for that purpose. It's common for both PV and PVH,
> > and is pretty messy as is. Anyways, Roger has already done the work
> > of reusing the frames via snooping into M2P, and repopulating those
> > frames that got invalidated when mapping iomem 1:1. 
> > 
> > We don't have any case where a frame is used for some special
> > purpose and does not have a mapping in the M2P, right? Otherwise,
> > his code would be broken....
> 
> Not that I know of.  But I can't believe that any path that builds a
> p2m, punches holes in it, and then uses the _m2p_ to figure out where
> the memory went could be the easy option. 
> 
> If doing it right the first time is too fiddly, we do have code that
> moves a frame from one gfn to another (for the xatp call) -- perhaps
> you could do that when you want to make a hole for mmio?

Well, doing it right the first time, meaning in construct_dom0 when we
populate the p2m would be the cleanest IMO. JFYI.. the reason it didn't
happen that way is because initially it didn't need to happen, because
linux guest did the repopulating. But then Roger noticed it could 
be done in xen itself, and the patch got done quickly for pvh, linux
still has the old code for pv fwiw.

Anyways, I'll add to my list.

thanks
Mukesh