From: Andrey Konovalov <[email protected]>
Disable KASAN instrumentation of arch/arm64/kernel/stacktrace.c.
This speeds up Generic KASAN by 5-20%.
As a side-effect, KASAN is now unable to detect bugs in the stack trace
collection code. This is taken as an acceptable downside.
Also replace READ_ONCE_NOCHECK() with READ_ONCE() in stacktrace.c.
As the file is now not instrumented, there is no need to use the
NOCHECK version of READ_ONCE().
Suggested-by: Mark Rutland <[email protected]>
Signed-off-by: Andrey Konovalov <[email protected]>
---
arch/arm64/kernel/Makefile | 3 +++
arch/arm64/kernel/stacktrace.c | 4 ++--
2 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/arch/arm64/kernel/Makefile b/arch/arm64/kernel/Makefile
index fa7981d0d917..da8cf6905c76 100644
--- a/arch/arm64/kernel/Makefile
+++ b/arch/arm64/kernel/Makefile
@@ -14,6 +14,9 @@ CFLAGS_REMOVE_return_address.o = $(CC_FLAGS_FTRACE)
CFLAGS_REMOVE_syscall.o = -fstack-protector -fstack-protector-strong
CFLAGS_syscall.o += -fno-stack-protector
+# Do not instrument to improve performance.
+KASAN_SANITIZE_stacktrace.o := n
+
# It's not safe to invoke KCOV when portions of the kernel environment aren't
# available or are out-of-sync with HW state. Since `noinstr` doesn't always
# inhibit KCOV instrumentation, disable it for the entire compilation unit.
diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
index e4103e085681..33e96ae4b15f 100644
--- a/arch/arm64/kernel/stacktrace.c
+++ b/arch/arm64/kernel/stacktrace.c
@@ -110,8 +110,8 @@ static int notrace unwind_frame(struct task_struct *tsk,
* Record this frame record's values and location. The prev_fp and
* prev_type are only meaningful to the next unwind_frame() invocation.
*/
- frame->fp = READ_ONCE_NOCHECK(*(unsigned long *)(fp));
- frame->pc = READ_ONCE_NOCHECK(*(unsigned long *)(fp + 8));
+ frame->fp = READ_ONCE(*(unsigned long *)(fp));
+ frame->pc = READ_ONCE(*(unsigned long *)(fp + 8));
frame->prev_fp = fp;
frame->prev_type = info.type;
--
2.25.1
On Sun, May 22, 2022 at 01:50:58AM +0200, [email protected] wrote:
> From: Andrey Konovalov <[email protected]>
>
> Disable KASAN instrumentation of arch/arm64/kernel/stacktrace.c.
>
> This speeds up Generic KASAN by 5-20%.
>
> As a side-effect, KASAN is now unable to detect bugs in the stack trace
> collection code. This is taken as an acceptable downside.
>
> Also replace READ_ONCE_NOCHECK() with READ_ONCE() in stacktrace.c.
> As the file is now not instrumented, there is no need to use the
> NOCHECK version of READ_ONCE().
>
> Suggested-by: Mark Rutland <[email protected]>
> Signed-off-by: Andrey Konovalov <[email protected]>
> ---
> arch/arm64/kernel/Makefile | 3 +++
> arch/arm64/kernel/stacktrace.c | 4 ++--
> 2 files changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/arch/arm64/kernel/Makefile b/arch/arm64/kernel/Makefile
> index fa7981d0d917..da8cf6905c76 100644
> --- a/arch/arm64/kernel/Makefile
> +++ b/arch/arm64/kernel/Makefile
> @@ -14,6 +14,9 @@ CFLAGS_REMOVE_return_address.o = $(CC_FLAGS_FTRACE)
> CFLAGS_REMOVE_syscall.o = -fstack-protector -fstack-protector-strong
> CFLAGS_syscall.o += -fno-stack-protector
>
> +# Do not instrument to improve performance.
> +KASAN_SANITIZE_stacktrace.o := n
Can we make that a little more descriptive? e.g.
# When KASAN is enabled, a stacktrace is recorded for every alloc/free, which
# can significantly impact performance. Avoid instrumenting the stacktrace code
# to minimize this impact.
KASAN_SANITIZE_stacktrace.o := n
With that:
Acked-by: Mark Rutland <[email protected]>
Mark.
> +
> # It's not safe to invoke KCOV when portions of the kernel environment aren't
> # available or are out-of-sync with HW state. Since `noinstr` doesn't always
> # inhibit KCOV instrumentation, disable it for the entire compilation unit.
> diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
> index e4103e085681..33e96ae4b15f 100644
> --- a/arch/arm64/kernel/stacktrace.c
> +++ b/arch/arm64/kernel/stacktrace.c
> @@ -110,8 +110,8 @@ static int notrace unwind_frame(struct task_struct *tsk,
> * Record this frame record's values and location. The prev_fp and
> * prev_type are only meaningful to the next unwind_frame() invocation.
> */
> - frame->fp = READ_ONCE_NOCHECK(*(unsigned long *)(fp));
> - frame->pc = READ_ONCE_NOCHECK(*(unsigned long *)(fp + 8));
> + frame->fp = READ_ONCE(*(unsigned long *)(fp));
> + frame->pc = READ_ONCE(*(unsigned long *)(fp + 8));
> frame->prev_fp = fp;
> frame->prev_type = info.type;
>
> --
> 2.25.1
>