[RFT PATCH 0/2] arm64: efi: Call SetVaMap() with a 1:1 mapping

[Nathan Chancellor]

On Tue, Jan 17, 2023 at 04:20:09PM +0100, Ard Biesheuvel wrote:
> On Tue, 17 Jan 2023 at 15:27, Ard Biesheuvel <ardb at kernel.org> wrote:
> >
> > Linux on arm64 is now in the same boat as x86, where supporting laptops
> > that were built to run Windows and never tested beyond what is required
> > for the Windows Logo certification need workarounds for all kinds of
> > bizarre behaviors.
> >
> > On Snapdragon laptops, we cannot call SetVirtualAddressMap() from the
> > stub, because the firmware will crash while trying to access memory via
> > the virtual addresses being installed, which is explicitly unsupported
> > by the EFI spec.
> >
> > However, not calling SetVirtualAddressMap() results in other problems:
> > on Ampere Altra, it causes SetTime() to crash. On Surface and Flex5g
> > Windows-on-ARM laptops, it causes ResetSystem() to crash.
> >
> > So let's try to work around this while not making too much of a mess.
> >
> > First of all, install a 1:1 mapping instead of avoiding SetVaMap()
> > altogether - from the EFI spec pov, this should amount to the same
> > thing.
> >
> > Then, given that we already use a SMBIOS based hack for Altra to force
> > the use of SetVirtualAddressMap(), let's check for Surface systems in
> > the same way.
> >
> > Please test, and please report the SMBIOS type 1 family field for which
> > this workaround is needed.
> >
> > Also, note that these changes will not make a difference if the
> > EFI_RT_PROPERTIES_TABLE lists SetVirtualAddressMap() as not implemented.
> >
> > Nathan, I would appreciate it if you could give this a spin on your
> > Altra box (only patch #1 should make a difference), and for good
> > measure, double check that hwclock still works as it should.
> >
> > Cc: Johan Hovold <johan+linaro at kernel.org>
> > Cc: Maximilian Luz <luzmaximilian at gmail.com>
> > Cc: Nathan Chancellor <nathan at kernel.org>
> > Cc: Steev Klimaszewski <steev at kali.org>
> > Cc: Shawn Guo <shawn.guo at linaro.org>
> >
> > Ard Biesheuvel (2):
> >   arm64: efi: Prefer a flat virtual mapping of the runtime services
> >   arm64: efi: Force use of SetVirtualAddressMap() on MS Surface
> >
> 
> Bah this does not even work on Yoga C630, so this is not going to help us.
> 
> If we want ResetSystem() on these machines, we'll have to retain other
> memory ranges and map the in the EFI runtime map. Yuck.
> 
> Nathan - still interested in whether patch #1 works on Altra,

I applied patch 1 on top of commit 6e50979a9c87 ("Merge tag
'mm-hotfixes-stable-2023-01-16-15-23' of
git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm") in Linus' tree
and everything still appears to be okay with hwclock. If there is any
more specific testing that I should do, please let me know. Feel free to
add

    Tested-by: Nathan Chancellor <nathan at kernel.org>

to patch 1 in future revisions, and I am happy to test anything else
that you might need in this series or future ones.

Mainline:

# uname -mr
6.2.0-rc4-00031-g6e50979a9c87 aarch64

# hwclock
2023-01-17 09:04:58.845411-07:00

Patch:

# uname -mr
6.2.0-rc4-00032-g20165e83052e aarch64

# hwclock
2023-01-17 10:25:38.843788-07:00

Cheers,
Nathan