The alignment fixup incorrectly decodes faulting ARM VLDn/VSTn instructions (where the optional alignment hint is given but incorrect) as LDR/STR, leading to register corruption. Detect these and correctly treat them as unhandled, so that userspace gets the fault it expects. Reported-by: Simon Hosie <simon.hosie@xxxxxxx> Signed-off-by: Robin Murphy <robin.murphy@xxxxxxx> Cc: <stable@xxxxxxxxxxxxxxx> --- Hi all, This was met with a resounding silence a while back[1], but the bug is still causing annoyance - apparently Clang intrinsics can be a little over-eager with alignment hints. Regardless of whose fault that is, the kernel shouldn't be silently mangling registers unexpectedly. Any objections to me dropping this into the patch system? Robin. [1]:http://www.spinics.net/lists/arm-kernel/msg328889.html arch/arm/mm/alignment.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/arch/arm/mm/alignment.c b/arch/arm/mm/alignment.c index 0c1ab49..83792f4 100644 --- a/arch/arm/mm/alignment.c +++ b/arch/arm/mm/alignment.c @@ -41,6 +41,7 @@ * This code is not portable to processors with late data abort handling. */ #define CODING_BITS(i) (i & 0x0e000000) +#define COND_BITS(i) (i & 0xf0000000) #define LDST_I_BIT(i) (i & (1 << 26)) /* Immediate constant */ #define LDST_P_BIT(i) (i & (1 << 24)) /* Preindex */ @@ -821,6 +822,8 @@ do_alignment(unsigned long addr, unsigned int fsr, struct pt_regs *regs) break; case 0x04000000: /* ldr or str immediate */ + if (COND_BITS(instr) == 0xf0000000) /* NEON VLDn, VSTn */ + goto bad; offset.un = OFFSET_BITS(instr); handler = do_alignment_ldrstr; break; -- 1.9.1 -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html