Re: [PATCH] [RFC] Kbuild: change CONFIG_FRAME_WARN for 32-bit

From: Nathan Chancellor <nathan@kernel.org>
To: Arnd Bergmann <arnd@kernel.org>
Cc: "Arnd Bergmann" <arnd@arndb.de>,
	"Uwe Kleine-König" <u.kleine-koenig@pengutronix.de>,
	"Masahiro Yamada" <masahiroy@kernel.org>,
	"Kees Cook" <keescook@chromium.org>,
	linux-kbuild@vger.kernel.org,
	"Andrew Morton" <akpm@linux-foundation.org>,
	"Peter Zijlstra" <peterz@infradead.org>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] [RFC] Kbuild: change CONFIG_FRAME_WARN for 32-bit
Date: Fri, 17 Jun 2022 14:35:54 -0700	[thread overview]
Message-ID: <Yqzzuk+Wv5q1zIKm@dev-arch.thelio-3990X> (raw)
In-Reply-To: <20220617150922.1878926-1-arnd@kernel.org>

Hi Arnd,

On Fri, Jun 17, 2022 at 05:08:53PM +0200, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> The introduction of CONFIG_GCC_PLUGIN_LATENT_ENTROPY raised the
> warning limit for 32-bit architectures to a much higher value in
> 2016. Initially this had no effect for Arm machines as their kernels
> tend to be cross-compiled, and the feature detection for the plugin did
> not work with common cross compilers.
> 
> I could not find the original regression report that led to the warning
> limit getting raised, but I have been unable to reproduce this with gcc-12
> and linux-5.18 -- all frame sizes appear to be be below the normal 1024
> byte limit on at least arm32 and i386.
> 
> However, the KASAN feature introduced later on does raise the frame size
> of a number of functions above the warning limit, in almost all cases to
> somewhere below 1280 bytes. The effect is that an arm allmodconfig build
> produces no warnings because of the check for the latent entropy plugin,
> but a normal KASAN based build with gcc does produce build warnings or
> failures when CONFIG_WERROR is set.
> 
> The 1280 byte limit is an arbitrary choice here, as it disables almost
> all warnings but leaves a few outliers that are probably better addressed
> with code changes. There are a number of functions that are just below the
> normal 1024 byte limit at the moment, and an extra 256 bytes for those
> is within the expectation, but more than that is probably an indication
> of a bad design or a real bug.
> 
> Cc: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> Cc: Masahiro Yamada <masahiroy@kernel.org>
> Cc: Kees Cook <keescook@chromium.org>
> Cc: linux-kbuild@vger.kernel.org
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  lib/Kconfig.debug | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
> index 3da5f9acb966..8a3afd837e99 100644
> --- a/lib/Kconfig.debug
> +++ b/lib/Kconfig.debug
> @@ -387,9 +387,9 @@ endif # DEBUG_INFO
>  config FRAME_WARN
>  	int "Warn for stack frames larger than"
>  	range 0 8192
> -	default 2048 if GCC_PLUGIN_LATENT_ENTROPY
>  	default 2048 if PARISC
>  	default 1536 if (!64BIT && XTENSA)
> +	default 1280 if (!64BIT && KASAN_STACK)

For what it's worth, if this was changed to just KASAN, this would
improve the situation for clang and ARCH=arm allmodconfig as well. It
leaves just two -Wframe-larger-than warnings in
drivers/gpu/drm/amd/display, which can be worked around with a kernel
side change:

https://github.com/ClangBuiltLinux/linux/issues/1455#issuecomment-995337569

Cheers,
Nathan