+JKim (for FreeBSD's perspective) > -----Original Message----- > From: Nick Desaulniers <ndesaulni...@google.com> > Sent: Monday, June 1, 2020 4:18 PM > To: Moore, Robert <robert.mo...@intel.com>; Kaneda, Erik > <erik.kan...@intel.com>; Wysocki, Rafael J <rafael.j.wyso...@intel.com>; > Len Brown <l...@kernel.org> > Cc: Ard Biesheuvel <a...@kernel.org>; dvyu...@google.com; > gli...@google.com; guohan...@huawei.com; linux-arm- > ker...@lists.infradead.org; linux-kernel@vger.kernel.org; > lorenzo.pieral...@arm.com; mark.rutl...@arm.com; > ndesaulni...@google.com; p...@google.com; r...@rjwysocki.net; > w...@kernel.org; sta...@vger.kernel.org; linux-a...@vger.kernel.org; > de...@acpica.org > Subject: [PATCH] ACPICA: fix UBSAN warning using __builtin_offsetof > > 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 > Hi,
> Looks like the emulated offsetof macro ACPI_OFFSET is causing these. We > can avoid this by using the compiler builtin, __builtin_offsetof. > This doesn't really fly because __builtin_offsetof is a compiler extension. It looks like a lot of stddef.h files do this: #define offsetof(a,b) __builtin_offset(a,b) So does anyone have objections to ACPI_OFFSET being defined to offsetof()? This will allow a host OS project project to use their own definitions of offsetof in place of ACPICA's. If they don't have a definition for offsetof, we can supply the old one as a fallback. Here's a patch: --- a/include/acpi/actypes.h +++ b/include/acpi/actypes.h @@ -504,11 +504,17 @@ typedef u64 acpi_integer; #define ACPI_SUB_PTR(t, a, b) ACPI_CAST_PTR (t, (ACPI_CAST_PTR (u8, (a)) - (acpi_size)(b))) #define ACPI_PTR_DIFF(a, b) ((acpi_size) (ACPI_CAST_PTR (u8, (a)) - ACPI_CAST_PTR (u8, (b)))) +/* Use an existing definiton for offsetof */ + +#ifndef offsetof +#define offsetof(d,f) ACPI_PTR_DIFF (&(((d *) 0)->f), (void *) 0) +#endif + /* Pointer/Integer type conversions */ #define ACPI_TO_POINTER(i) ACPI_CAST_PTR (void, (acpi_size) (i)) #define ACPI_TO_INTEGER(p) ACPI_PTR_DIFF (p, (void *) 0) -#define ACPI_OFFSET(d, f) ACPI_PTR_DIFF (&(((d *) 0)->f), (void *) 0) +#define ACPI_OFFSET(d, f) offsetof (d,f) #define ACPI_PHYSADDR_TO_PTR(i) ACPI_TO_POINTER(i) #define ACPI_PTR_TO_PHYSADDR(i) ACPI_TO_INTEGER(i) Thanks, Erik > The non-kernel runtime of UBSAN would print: > runtime error: member access within null pointer of type for this macro. > > Link: https://lore.kernel.org/lkml/20200521100952.GA5360@willie-the-truck/ > Cc: sta...@vger.kernel.org > Reported-by: Will Deacon <w...@kernel.org> > Suggested-by: Ard Biesheuvel <a...@kernel.org> > Signed-off-by: Nick Desaulniers <ndesaulni...@google.com> > --- > include/acpi/actypes.h | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/include/acpi/actypes.h b/include/acpi/actypes.h index > 4defed58ea33..04359c70b198 100644 > --- a/include/acpi/actypes.h > +++ b/include/acpi/actypes.h > @@ -508,7 +508,7 @@ typedef u64 acpi_integer; > > #define ACPI_TO_POINTER(i) ACPI_CAST_PTR (void, (acpi_size) (i)) > #define ACPI_TO_INTEGER(p) ACPI_PTR_DIFF (p, (void *) 0) > -#define ACPI_OFFSET(d, f) ACPI_PTR_DIFF (&(((d *) 0)->f), > (void *) > 0) > +#define ACPI_OFFSET(d, f) __builtin_offsetof(d, f) > #define ACPI_PHYSADDR_TO_PTR(i) ACPI_TO_POINTER(i) > #define ACPI_PTR_TO_PHYSADDR(i) ACPI_TO_INTEGER(i) > > -- > 2.27.0.rc2.251.g90737beb825-goog
