Message ID | 1507051546-11677-1-git-send-email-mark.rutland@arm.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Tue, Oct 03, 2017 at 06:25:46PM +0100, Mark Rutland wrote: > AddressSanitizer instrumentation can significantly bloat the stack, and > with GCC 7 this can result in stack overflows at boot time in some > configurations. > > We can avoid this by doubling our stack size when KASAN is in use, as is > already done on x86 (and has been since KASAN was introduced). > Regardless of other patches to decrease KASAN's stack utilization, > kernels built with KASAN will always require more stack space than those > built without, and we should take this into account. > > Signed-off-by: Mark Rutland <mark.rutland@arm.com> > Cc: Catalin Marinas <catalin.marinas@arm.com> > Cc: Will Deacon <will.deacon@arm.com> > Cc: Suzuki K Poulose <suzuki.poulose@arm.com> > --- > arch/arm64/include/asm/memory.h | 9 ++++++--- > 1 file changed, 6 insertions(+), 3 deletions(-) > > In local testing (with KASAN and other options enabled), I've seen a stack > depth of 23928 bytes while booting to a prompt with an Ubuntu 14.04 filesystem > mounted over NFS. Without KASAN (but with other debug options enabled), this > was 7992 bytes. > > I wonder if we want to go for 4x rather than 2x... Let's hope it's fine for now with 32K stacks. We may run into allocation issues with 64K stacks (and without VMAP_STACKS). Do you want this patch for 4.14?
On Wed, Oct 04, 2017 at 12:31:31PM +0100, Catalin Marinas wrote: > On Tue, Oct 03, 2017 at 06:25:46PM +0100, Mark Rutland wrote: > > AddressSanitizer instrumentation can significantly bloat the stack, and > > with GCC 7 this can result in stack overflows at boot time in some > > configurations. > > > > We can avoid this by doubling our stack size when KASAN is in use, as is > > already done on x86 (and has been since KASAN was introduced). > > Regardless of other patches to decrease KASAN's stack utilization, > > kernels built with KASAN will always require more stack space than those > > built without, and we should take this into account. > > > > Signed-off-by: Mark Rutland <mark.rutland@arm.com> > > Cc: Catalin Marinas <catalin.marinas@arm.com> > > Cc: Will Deacon <will.deacon@arm.com> > > Cc: Suzuki K Poulose <suzuki.poulose@arm.com> > > --- > > arch/arm64/include/asm/memory.h | 9 ++++++--- > > 1 file changed, 6 insertions(+), 3 deletions(-) > > > > In local testing (with KASAN and other options enabled), I've seen a stack > > depth of 23928 bytes while booting to a prompt with an Ubuntu 14.04 filesystem > > mounted over NFS. Without KASAN (but with other debug options enabled), this > > was 7992 bytes. > > > > I wonder if we want to go for 4x rather than 2x... > > Let's hope it's fine for now with 32K stacks. We may run into allocation > issues with 64K stacks (and without VMAP_STACKS). > > Do you want this patch for 4.14? If you're happy to take it, that would be great, but otherwise it can wait for v4.15. Arguably this fixes the introduction of KASAN to arm64, but I'm not sure how much we care about backporting this to stable. Thanks, Mark.
On Wed, Oct 04, 2017 at 12:39:09PM +0100, Mark Rutland wrote: > On Wed, Oct 04, 2017 at 12:31:31PM +0100, Catalin Marinas wrote: > > On Tue, Oct 03, 2017 at 06:25:46PM +0100, Mark Rutland wrote: > > > AddressSanitizer instrumentation can significantly bloat the stack, and > > > with GCC 7 this can result in stack overflows at boot time in some > > > configurations. > > > > > > We can avoid this by doubling our stack size when KASAN is in use, as is > > > already done on x86 (and has been since KASAN was introduced). > > > Regardless of other patches to decrease KASAN's stack utilization, > > > kernels built with KASAN will always require more stack space than those > > > built without, and we should take this into account. > > > > > > Signed-off-by: Mark Rutland <mark.rutland@arm.com> > > > Cc: Catalin Marinas <catalin.marinas@arm.com> > > > Cc: Will Deacon <will.deacon@arm.com> > > > Cc: Suzuki K Poulose <suzuki.poulose@arm.com> > > > --- > > > arch/arm64/include/asm/memory.h | 9 ++++++--- > > > 1 file changed, 6 insertions(+), 3 deletions(-) > > > > > > In local testing (with KASAN and other options enabled), I've seen a stack > > > depth of 23928 bytes while booting to a prompt with an Ubuntu 14.04 filesystem > > > mounted over NFS. Without KASAN (but with other debug options enabled), this > > > was 7992 bytes. > > > > > > I wonder if we want to go for 4x rather than 2x... > > > > Let's hope it's fine for now with 32K stacks. We may run into allocation > > issues with 64K stacks (and without VMAP_STACKS). > > > > Do you want this patch for 4.14? > > If you're happy to take it, that would be great, but otherwise it can > wait for v4.15. > > Arguably this fixes the introduction of KASAN to arm64, but I'm not sure > how much we care about backporting this to stable. I queued it for 4.14. If anyone thinks it's really important for older kernels, we can send it to stable later (I don't think it is though).
diff --git a/arch/arm64/include/asm/memory.h b/arch/arm64/include/asm/memory.h index 3585a5e..f7c4d21 100644 --- a/arch/arm64/include/asm/memory.h +++ b/arch/arm64/include/asm/memory.h @@ -95,16 +95,19 @@ #define KERNEL_END _end /* - * The size of the KASAN shadow region. This should be 1/8th of the - * size of the entire kernel virtual address space. + * KASAN requires 1/8th of the kernel virtual address space for the shadow + * region. KASAN can bloat the stack significantly, so double the (minimum) + * stack size when KASAN is in use. */ #ifdef CONFIG_KASAN #define KASAN_SHADOW_SIZE (UL(1) << (VA_BITS - 3)) +#define KASAN_THREAD_SHIFT 1 #else #define KASAN_SHADOW_SIZE (0) +#define KASAN_THREAD_SHIFT 0 #endif -#define MIN_THREAD_SHIFT 14 +#define MIN_THREAD_SHIFT (14 + KASAN_THREAD_SHIFT) /* * VMAP'd stacks are allocated at page granularity, so we must ensure that such
AddressSanitizer instrumentation can significantly bloat the stack, and with GCC 7 this can result in stack overflows at boot time in some configurations. We can avoid this by doubling our stack size when KASAN is in use, as is already done on x86 (and has been since KASAN was introduced). Regardless of other patches to decrease KASAN's stack utilization, kernels built with KASAN will always require more stack space than those built without, and we should take this into account. Signed-off-by: Mark Rutland <mark.rutland@arm.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Suzuki K Poulose <suzuki.poulose@arm.com> --- arch/arm64/include/asm/memory.h | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) In local testing (with KASAN and other options enabled), I've seen a stack depth of 23928 bytes while booting to a prompt with an Ubuntu 14.04 filesystem mounted over NFS. Without KASAN (but with other debug options enabled), this was 7992 bytes. I wonder if we want to go for 4x rather than 2x... Thanks, Mark.