Re: [PATCH] arm64: vdso32: force vdso32 to be compiled as -marm

[Will Deacon]

On Wed, May 27, 2020 at 11:17:33AM -0700, Nick Desaulniers wrote:
> On Wed, May 27, 2020 at 11:08 AM Will Deacon <will@xxxxxxxxxx> wrote:
> >
> > On Wed, May 27, 2020 at 10:55:24AM -0700, Nick Desaulniers wrote:
> > > On Wed, May 27, 2020 at 6:45 AM Robin Murphy <robin.murphy@xxxxxxx> wrote:
> > > >
> > > > On 2020-05-26 18:31, Nick Desaulniers wrote:
> > > > > Custom toolchains that modify the default target to -mthumb cannot
> > > > > compile the arm64 compat vdso32, as
> > > > > arch/arm64/include/asm/vdso/compat_gettimeofday.h
> > > > > contains assembly that's invalid in -mthumb.  Force the use of -marm,
> > > > > always.
> > > >
> > > > FWIW, this seems suspicious - the only assembly instructions I see there
> > > > are SWI(SVC), MRRC, and a MOV, all of which exist in Thumb for the
> > > > -march=armv7a baseline that we set.
> > > >
> > > > On a hunch, I've just bodged "VDSO_CFLAGS += -mthumb" into my tree and
> > > > built a Thumb VDSO quite happily with Ubuntu 19.04's
> > > > gcc-arm-linux-gnueabihf. What was the actual failure you saw?
> > >
> > > From the link in the commit message: `write to reserved register 'R7'`
> > > https://godbolt.org/z/zwr7iZ
> > > IIUC r7 is reserved for the frame pointer in THUMB?
> > >
> > > What is the implicit default of your gcc-arm-linux-gnueabihf at -O2?
> > > -mthumb, or -marm?
> >
> > Hmm, but this *is* weird because if I build a 32-bit kernel then I get
> > either an ARM or a Thumb-2 VDSO depending on CONFIG_THUMB2_KERNEL. I'm
> > not sure if that's deliberate, but both build and appear to work.
> 
> That's because there's 3 VDSO's when it comes to ARM:
> arm64's 64b vdso: arch/arm64/kernel/vdso
> arm64's 32b vdso: arch/arm64/kernel/vdso32/
> arm's 32b vdso: arch/arm/kernel/vdso.c

Yes, I know that :)

> When you build a 32b kernel, you're only making use of the last of
> those three; the arm64 vdso and vdso32 code is irrelevant.
> This patch is specific to the second case, which is the 32b compat
> vdso for a 64b kernel.

Sure, but if you can build a Thumb-2 vDSO object for arch/arm/ using then
we should be able to build a Thumb-2 compat vDSO for arch/arm64, and your
patch is papering over a deeper issue. Generally, having the compat vDSO
behave differently to the arch/arm/ vDSO is indicative of something being
broken.

In other words, if your patch was correct (not sure that it is) then I
would expect a corresponding change to arch/arm/ to pass -marm when
CONFIG_THUMB2_KERNEL=y. Make sense?

Will