Re: [PATCH] ACPICA: fix UBSAN warning using __builtin_offsetof

[Will Deacon]

Hey Nick,

On Tue, Jun 02, 2020 at 11:46:31AM -0700, Nick Desaulniers wrote:
> On Mon, Jun 1, 2020 at 5:03 PM Kaneda, Erik <erik.kaneda@xxxxxxxxx> wrote:
> > > Will reported UBSAN warnings:
> > > UBSAN: null-ptr-deref in drivers/acpi/acpica/tbfadt.c:459:37
> > > UBSAN: null-ptr-deref in arch/arm64/kernel/smp.c:596:6
> > >
> > > Looks like the emulated offsetof macro ACPI_OFFSET is causing these. We
> > > can avoid this by using the compiler builtin, __builtin_offsetof.
> >
> > I'll take a look at this tomorrow
> > >
> > > The non-kernel runtime of UBSAN would print:
> > > runtime error: member access within null pointer of type for this macro.
> >
> > actypes.h is owned by ACPICA so we typically do not allow compiler-specific
> > extensions because the code is intended to be compiled using the C99 standard
> > without compiler extensions. We could allow this sort of thing in a Linux-specific
> > header file like include/acpi/platform/aclinux.h but I'll take a look at the error as well..
> 
> If I'm not allowed to touch that header, it looks like I can include
> <linux/stddef.h> (rather than my host's <stddef.h>) to get a
> definition of `offsetof` thats implemented in terms of
> `__builtin_offsetof`.  I should be able to use that to replace uses of
> ACPI_OFFSET.  Are any of these off limits?

It's not so much about not being allowed to touch the header, but rather
that the kernel imports the code from a different project:

https://acpica.org/community

> $ grep -rn ACPI_OFFSET
> arch/arm64/include/asm/acpi.h:34:#define ACPI_MADT_GICC_MIN_LENGTH
> ACPI_OFFSET(  \
> arch/arm64/include/asm/acpi.h:41:#define ACPI_MADT_GICC_SPE
> (ACPI_OFFSET(struct acpi_madt_generic_interrupt, \

I'm happy to take patches to the stuff under arch/arm64/, fwiw.

Will