On Fri, Jan 8, 2021 at 9:31 PM Arnd Bergmann [off-list ref] wrote:
On Fri, Jan 8, 2021 at 7:56 PM Andrey Konovalov [off-list ref] wrote:
quoted
On Fri, Jan 8, 2021 at 5:09 AM Walter Wu [off-list ref] wrote:
quoted
quoted
@@ -2,6 +2,12 @@
CFLAGS_KASAN_NOSANITIZE := -fno-builtin
KASAN_SHADOW_OFFSET ?= $(CONFIG_KASAN_SHADOW_OFFSET)
+ifdef CONFIG_KASAN_STACK
+ stack_enable := 1
+else
+ stack_enable := 0
+endif
+
AFAIR, Arnd wanted to avoid having KASAN_STACK to be enabled by
default when compiling with Clang, since Clang instrumentation leads
to very large kernel stacks, which, in turn, lead to compile-time
warnings. What I don't remember is why there are two configs.
Arnd, is that correct? What was the reason behind having two configs?
I think I just considered it cleaner than defining the extra variable in the
Makefile at the time, as this was the only place that referenced
CONFIG_KASAN_STACK.
The '#if CONFIG_KASAN_STACK' (rather than #ifdef) that got added
later do make my version more confusing though, so I agree that
Walter's second patch improves it.
Acked-by: Arnd Bergmann <arnd@arndb.de>
Got it, thanks!
On a related note: do you have any hope that clang will ever fix
https://bugs.llvm.org/show_bug.cgi?id=38809 and KASAN_STACK
can be enabled by default on clang without risking stack
overflows?
Not sure :(
I've filed this on KASAN bugzilla to not forget:
https://bugzilla.kernel.org/show_bug.cgi?id=211139