Re: [PATCH 1/2] libxl: set stdvga=1 by default when creating a hvm guest

[Ian Campbell <Ian.Campbell@citrix.com>]

On Wed, 2012-06-27 at 02:45 +0100, Ren, Yongjie wrote:
> > -----Original Message-----
> > From: Stefano Stabellini [mailto:stefano.stabellini@eu.citrix.com]
> > Sent: Wednesday, June 27, 2012 1:00 AM
> > To: Ian Campbell
> > Cc: Ren, Yongjie; xen-devel@lists.xen.org; Ian Jackson; Tim (Xen.org);
> > Stefano Stabellini
> > Subject: Re: [PATCH 1/2] libxl: set stdvga=1 by default when creating a hvm
> > guest
> > 
> > On Tue, 26 Jun 2012, Ian Campbell wrote:
> > > On Tue, 2012-06-26 at 07:02 +0100, Ren, Yongjie wrote:
> > > > libxl: set stdvga=1 by default when creating a hvm guest
> > > >
> > > > Most of the modern OSes (e.g. Windows XP, Windows 7, RHEL6.x,
> > Ubuntu, Fedora) support VBE 2.0 or later.
> > > > So, select a standard VGA card with VBE as the default emulated
> > graphics device.
> > >
> > > I'm not expert on the graphics side of HVM, but on the face of it
> > > switching the default to something more modern seems like a
> > reasonable
> > > idea, although I'm not sure if we should be doing this for 4.2 at this
> > > point.
> > >
> > > I've CCd Tim and Stefano for input from the HVM and QEMU sides.
> > 
> > I think it is a good thing.
> > The only thing to keep in mind is that QEMU upstream is switching to
> > 16MB of videoram for stdvga. So at some point in the near future
> > upstream QEMU will stop working correctly with xen 4.2, unless we bump
> > the videoram to 16MB too.
> > 
> Yes, we should pay attention to this when using upstream QEMU.
> 
> > 
> > > > It's also a workaround for the following bug.
> > > > http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1812
> > >
> > > Do you understand the root cause of that bug?
> > >
> I don't understand the root cause.
> 
> > > It's hard to see how detaching a VF relates to the VGA emulation in use.
> > > Can you explain it? Are you sure you aren't just masking the real issue
> > > here?
> > 
> It's strange that detaching a VF may break the graphics display.
> As it only happens when 'stdvga=0', it might not be a normal usage.
> 
> > Indeed. We cannot possibly accept the patch on the basis that it looks
> > like it is masking an unrelated pci-passthrough bug.
> >
> I don't want to mask that bug, either. :-)
> The following sentence is quoted from xl.cfg man page.
> "If your guest supports VBE 2.0 or later (e.g. Windows XP onwards) 
> then you should enable this (stdvga option)."
> If we set 'stdvga=1', we will not meet the bug (#1812).
> I assume many Xen users (including me) are not very familiar with 'stdvga' and 
> will leave it as default (it's 0 before my patch). 
> If then, users may meet something *strange* like bug #1812.
> I don't think it's friendly to end users, so I set the default value
> of 'stdvga' to '1'.

There are good reasons which justify setting stdvga=1 by default. But
anything to do with #1812 is not one of them. It is obviously wrong to
justify making an unrelated configuration change on the basis that it
hides a bug by default without first understanding the reasons for the
bug.

I certainly hope you are not planning to close #1812 and stop
investigating it if we change the stdvga default.

Ian.