Re: [REGRESSION] Linux 6.6.64 crashes when booting as PVH domU

[Marek Marczykowski-Górecki <marmarek@xxxxxxxxxxxxxxxxxxxxxx>]

On Wed, Dec 11, 2024 at 01:37:52PM +0100, Ard Biesheuvel wrote:
> (cc Greg)
> 
> On Wed, 11 Dec 2024 at 13:30, Marek Marczykowski-Górecki
> <marmarek@xxxxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Wed, Dec 11, 2024 at 01:24:08PM +0100, Ard Biesheuvel wrote:
> > > On Wed, 11 Dec 2024 at 12:53, Jürgen Groß <jgross@xxxxxxxx> wrote:
> > > >
> > > > Jason, Ard,
> > > >
> > > > I guess there are some prereq patches missing in stable 6.6.y branch?
> > > >
> > > >
> > > > Juergen
> > > >
> > > > On 11.12.24 12:41, Marek Marczykowski-Górecki wrote:
> > > > > Hi,
> > > > >
> > > > > With Linux 6.6.64 I get the following crash on domU boot:
> > > > >
> > > > > (XEN) d5v0 Triple fault - invoking HVM shutdown action 1
> > > > > (XEN) *** Dumping Dom5 vcpu#0 state: ***
> > > > > (XEN) ----[ Xen-4.19.0  x86_64  debug=n  Tainted:  M     ]----
> ...
> > > > >
> > > > > Linux 6.6.63 works fine.
> > > > >
> > > > > Looking at the changes, I suspect one of those:
> > > > >
> > > > >      83d123e27623 x86/pvh: Call C code via the kernel virtual mapping
> > > > >      f662b4a69e1d x86/pvh: Set phys_base when calling 
> > > > > xen_prepare_pvh()
> > > > >
> > >
> > > The second patch shouldn't have been backported. It is unnecessary,
> > > given that in the old situation, the kernel image needs to be loaded
> > > at a fixed address. And it assumes  that %rbp is set to the physical
> > > load offset, but those patches were not backported.
> >
> > It has this tag:
> >
> >     Stable-dep-of: e8fbc0d9cab6 ("x86/pvh: Call C code via the kernel 
> > virtual mapping")
> >
> 
> That was added by the stable maintainers - someone grabbed a patch
> from the middle of an unrelated series to make e8fbc0d9cab6 apply
> without lexical conflicts.
> 
> > Does it mean neither of them should be backported?
> >
> > But then, the e8fbc0d9cab6 has "Fixes:" tag (pointing at very old
> > commit).
> >
> 
> If someone thinks e8fbc0d9cab6 should be backported, they should
> rebase it onto v6.6.y, not backport random other patches until
> git-apply stops complaining. And ideally, someone would build and boot
> the result to check whether it works.
> 
> For now, it would be better to revert both.

I can confirm that reverting both commits fixes the issue.

-- 
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab

Attachment: signature.asc
Description: PGP signature