From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id B56946713 for ; Tue, 6 Sep 2022 16:08:15 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id A69A6C433D6; Tue, 6 Sep 2022 16:08:14 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1662480495; bh=zCc7fKyX2vVh00Skn6XiWidfTXu0QhuXY1caqeayhMM=; h=Date:From:To:Cc:Subject:References:In-Reply-To:From; b=ba9F9jP6xg9kM/goDxi/ZxniQ8EZq0lcpVWdjBcc+ux0XgBUqqCPRAweRuoyu2z9e r+joBj/rUf8bX8zUIFs8W7qJieAyaT8X1CqQ2gu5KWXzD4xgONbx78+Y0F6U9F5OKa Bq732ZHvCASpV2p3JitatXHhcrNM4qh+UAfssAQKAHDRKplbHOH5YZPP40mbCZZIzP AIbTA6WW+f9wd4/x1ZppyWmCupDzBZolui7qTwPrIpXidy/y9fNW4Dq1r5A5MOLO1h yABXALX0TfV6e8RHoeC8VlqKW6V/huyMWAKqmENRC1vdxA1g1jS4s9cYQizU2E2QRo i7Z1dJk+JwaTQ== Date: Tue, 6 Sep 2022 09:08:12 -0700 From: Nathan Chancellor To: Masahiro Yamada Cc: Nick Desaulniers , Michal Marek , Tom Rix , Linux Kbuild mailing list , Linux Kernel Mailing List , clang-built-linux , X86 ML , Dmitrii Bundin , Fangrui Song , Alexey Alexandrov , Bill Wendling , Greg Thelen Subject: Re: [PATCH v2 2/5] Makefile.compiler: Use KBUILD_AFLAGS for as-option Message-ID: References: <20220831184408.2778264-1-ndesaulniers@google.com> <20220831184408.2778264-3-ndesaulniers@google.com> Precedence: bulk X-Mailing-List: llvm@lists.linux.dev List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: On Mon, Sep 05, 2022 at 06:09:28PM +0900, Masahiro Yamada wrote: > On Thu, Sep 1, 2022 at 4:53 AM Nathan Chancellor wrote: > > > > On Wed, Aug 31, 2022 at 11:44:05AM -0700, Nick Desaulniers wrote: > > > as-instr uses KBUILD_AFLAGS, but as-option uses KBUILD_CFLAGS. This can > > > cause as-option to fail unexpectedly because clang will emit > > > -Werror,-Wunused-command-line-argument for various -m and -f flags for > > > assembler sources. > > > > Now that I am looking closer at it, where does that '-Werror' come from? > > The related commit is > c3f0d0bc5b01ad90c45276952802455750444b4f > > The previous discussion with Arnd is > https://lore.kernel.org/linux-kbuild/20170314213724.3836900-1-arnd@arndb.de/ Right, although this is for cc-option, not as-option, no? > > For cc-option, we add it to elevate clang's warnings about unused '-f', > > '-m', and '-W' flags to errors so that we do not add those flags. > > However, I do not see '-Werror' in as-option. I am going to assume it > > came from CONFIG_WERROR, as I believe Android has that turned on by > > default. > > CONFIG_WERROR is added to CFLAGS. > But, I guess it is more correct to do likewise for others. > (https://patchwork.kernel.org/project/linux-kbuild/patch/20220905083619.672091-1-masahiroy@kernel.org/) Ack. > > I think that is the real problem: without '-Werror', the only > > error that should come from as-option is when an option isn't supported > > by the assembler, as clang will still warn but those will not be fatal > > but with '-Werror', those warnings turn fatal, causing all subsequent > > as-option calls to fail. > > Presumably, it is correct to add -Werror to as-option as well. > We have no reason to add it to cc-option, but not to as-option. > > I also believe '-x assembler' should be changed to > '-x assembler-with-cpp'. > > As I mentioned somewhere before, our assembly code (*.S) is always > preprocessed. There is no *.s file in the kernel source tree. > > So, '-x assembler-with-cpp' matches the real situation. All good points, I think that is a good fix as well. > One interesting thing is, clang does not warn > [-Wunused-command-line-argument] for *.S files. > > $ clang -fomit-frame-pointer -c -x assembler /dev/null -o /dev/null > clang: warning: argument unused during compilation: > '-fomit-frame-pointer' [-Wunused-command-line-argument] > > $ clang -fomit-frame-pointer -c -x assembler-with-cpp /dev/null -o /dev/null Interesting... I suspect that is likely intentional, as some compiler flags could be used during preprocessing (it's come up before: https://github.com/llvm/llvm-project/issues/55656) but I was not able to figure out exactly where in clang the flags were consumed but Driver.cpp is quite large and I didn't look too hard :) More importantly, it still allows us to catch invalid assembler arguments: $ clang -c -x assembler-with-cpp /dev/null -o /dev/null -Wa,-foo clang-16: error: unsupported argument '-foo' to option '-Wa,' $ clang -c -x assembler-with-cpp /dev/null -o /dev/null -Wa,--noexecstack > The root cause is we are using '-x assembler', which > never happens in the kernel tree. > > To sum up, the code I think correct is: > > as-option = $(call try-run,\ > $(CC) -Werror $(KBUILD_AFLAGS) $(1) -c -x assembler-with-cpp > /dev/null -o "$$TMP",$(1),$(2)) > Agreed. Thank you as always for your feedback! Cheers, Nathan