Re: [edk2-devel] ArmVirtPkg: non-executable EFI_LOADER_DATA breaks GRUB on Ubuntu 22.04

["Pedro Falcato" <pedro.falcato@gmail.com>]

On Thu, Jul 13, 2023 at 6:20 PM Ard Biesheuvel <ardb@kernel.org> wrote:
>
> On Thu, 13 Jul 2023 at 18:57, Pedro Falcato <pedro.falcato@gmail.com> wrote:
> >
> > On Tue, Jul 11, 2023 at 7:58 AM Gerd Hoffmann <kraxel@redhat.com> wrote:
> > >
> > > On Mon, Jul 10, 2023 at 04:58:15PM +0100, Pedro Falcato wrote:
> > > > On Mon, Jul 10, 2023 at 2:28 PM <osy@turing.llc> wrote:
> > > > >
> > > > > I have an existing install of Ubuntu 22.04 on a QEMU virtual machine which I've decided to update the UEFI firmware. After doing so, GRUB no longer boots ("Synchronous Exception" message seen). After a git bisect session, I found the problematic 2997ae38739756ecba9b0de19e86032ebc689ef9. The comment says GRUB should have been fixed in 2017, but for one reason or another, my VM which was built in 2022 still had the issue. Regardless, I don't think it's a good idea to break GRUB, even if it's fixed in 2017. In the very least, a better error message would be preferable to crashing with an "Synchronous Exception." Googling this error message shows that other people may be hitting this issue as well but the vague error symptom means its impossible to know if it's the same issue or not.
> > > >
> > > > +CC Some of the folks involved in the original discussion
> > > >
> > > > In the original thread, people discussed some alternative behavior to
> > > > just crashing on a NX fault. Is this still an alternative?
> > >
> > > The idea is: Improve page fault handler to (a) print a big'n'fat
> > > warning, and (b) loosening up memory permissions for the faulting
> > > page address.
> > >
> > > No patch for that emerged (yet?).
> >
> > Ack. I can work on that.
> >
> > > > I'm kind of thinking this should be addressed by distros anyway....
> > > > How is $CURRENT_YEAR Ubuntu still shipping bad GRUBs? I know the
> > > > situation around GRUB and distro patching is complicated but...
> > > > Do we have any idea of how many distros/GRUBs are affected by this?
> > >
> > > Too many :(
> >
> > Ugh, even the latest releases?
> >
> > > > Personally, I would like to avoid loosening up memory permissions.
> > >
> > > Well, you can't have both.  You have to pick between strict nx handling
> > > and grub bug compatibility ...
> >
> > Yes. IMO it should be ok to add a hack around NX handling if there's a
> > solid plan for dealing with this from the distros' side (and phasing
> > this out). And I'm assuming upstream GRUB has this fixed.
> > This whole situation is kind of messy as firmware people add new
> > restrictions that weren't really there in the first place.
> >
> > Also, what's the situation on this for x86? I assume it's a lot worse there?
> >
>
> To be honest, I have little sympathy for the gigantic mess that the
> distros have created for themselves with GRUB, shim, etc. Mainline
> GRUB works fine with mainline EDK2, and secure boot in a arm64 VM is
> rather pointless, given that the [emulated] NOR flash is writable by
> the guest OS. The breakage is in the downstream GRUB changes that make
> it interoperate with shim, and its hacked up PE loader.
>
> If we are going to accommodate every broken GRUB build that the
> distros ever released, we won't be able to make any progress on this
> front. I understand that the distros need to support their existing
> user bases, so I am willing to consider facilities that make it easier
> to create builds that work around such issues.
>
> However, just turning off NX support is not one of the options.
> Upstream is not what the distros are shipping, this applies to GRUB
> and shim as well as EDK2: so if their downstream GRUB breaks EDK2,
> they can fix it in their EDK2 builds, either by carrying a code
> change, or by enabling an upstream build flag that is off by default.

I understand your sentiment, but I don't see how this can work. Let's
say Fedora has a fixed GRUB (I have no idea if this is the case), so
they have a fully-NX'd edk2. Then someone tries to boot Ubuntu 22.04 -
it crashes because Ubuntu's GRUB is borked; what now? I don't know if
this case is super prevalent in virtualization, but it's definitely a
problem (and it seems to have happened to osy here?).

I agree that turning off NX sucks, but so does not being able to boot
into distros as recent as Ubuntu 22.04. It might be that a single
Sleep(10); +  a nice loud error message gets the idea across? maybe
over a stable tag or so, then we remove the hack and add full-NX. What
do you think?

-- 
Pedro