Re: [PATCHv2 0/6] efi: detect erroneous firmware IRQ manipulation

[Matt Fleming <matt-mF/unelCI9GS6iBeEJttW/XRex20P6io@public.gmane.org>]

On Fri, 22 Apr, at 04:12:59PM, Ard Biesheuvel wrote:
> On 22 April 2016 at 15:51, Mark Rutland <mark.rutland-5wv7dgnIgG8@public.gmane.org> wrote:
> > Some firmware erroneously unmask IRQs (and potentially other architecture
> > specific exceptions) during runtime services functions, in violation of both
> > common sense and the UEFI specification. This can result in a number of issues
> > if said exceptions are taken when they are expected to be masked, and
> > additionally can confuse IRQ tracing if the original mask state is not
> > restored prior to returning from firmware.
> >
> > In practice it's difficult to check that firmware never unmasks exceptions, but
> > we can at least check that the IRQ flags are at least consistent upon entry to
> > and return from a runtime services function call. This series implements said
> > check in the shared EFI runtime wrappers code, after an initial round of
> > refactoring such that this can be generic.
> >
> > I have left ia64 as-is, without this check, as ia64 doesn't currently use the
> > generic runtime wrappers, has many special cases for the runtime calls which
> > don't fit well with the generic code, and I don't expect a new, buggy ia64
> > firmware to appear soon.
> >
> > The first time corruption of the IRQ flags is detected, we dump a stack trace,
> > and set TAINT_FIRMWARE_WORKAROUND. Additionally, and in all subsequent cases,
> > we log (with ratelimiting) the specific corruption of the flags, and restore
> > the expected flags to avoid redundant warnings elsewhere.
> >
> > Since v1 [1]:
> > * Fix thinko: s/local_irq_save/local_save_flags/
> > * Remove ifdefs after conversion
> > * Remove reundant semicolon from x86 patch
> > * Move efi_call_virt_check_flags before first use
> > * Add Acked-bys and Reviewed-bys
> >
> > Ard, I assume that your Reviewed-by still stands for the final patch, even
> > though efi_call_virt_check_flags moved. Please shout if that's not the case!
> >
> 
> No, that's fine. Thanks for respinning so quickly.
> 
> > Hopefully you're also happy to extend that to the new patch removing the
> > ifdefs once they become superfluous.
> >
> 
> Matt: in case your review bandwidth is limited atm, I'd much prefer
> this series making v4.7 than the GOP stuff or the other stuff i have
> been posting over the past weeks.

I like this series a lot (well, ignoring the fact that the firmware is
trying to eat itself). The runtime call code is much cleaner now, and
this is a great precedent for any future multi-architecture quirks we
may need.

Queued for v4.7, thanks everyone!