[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20241108174638.GC2564051@thelio-3990X>
Date: Fri, 8 Nov 2024 10:46:38 -0700
From: Nathan Chancellor <nathan@...nel.org>
To: Rong Xu <xur@...gle.com>
Cc: Masahiro Yamada <masahiroy@...nel.org>,
Nicolas Schier <nicolas@...sle.eu>, Han Shen <shenhan@...gle.com>,
Nick Desaulniers <ndesaulniers@...gle.com>,
Bill Wendling <morbo@...gle.com>,
Justin Stitt <justinstitt@...gle.com>, linux-kbuild@...r.kernel.org,
linux-kernel@...r.kernel.org, llvm@...ts.linux.dev
Subject: Re: [PATCH] kbuild: Fix Propeller build option
Hi Rong,
Thanks for the quick patch!
On Thu, Nov 07, 2024 at 09:10:30PM -0800, Rong Xu wrote:
> The '-fbasic-block-sections=labels' option has been deprecated with
> the following llvm patch:
> https://github.com/llvm/llvm-project/pull/110039
>
> The old option still works, but with a warning like
> clang: warning: argument '-fbasic-block-sections=labels' \
> is deprecated, use '-fbasic-block-address-map' instead [-Wdeprecated]
>
> Currently, the option change is only in the ToT clang (v20) and not yet
> released in v19.
I think this whole block could be a little easier to read and understand
with some more standard kernel commit message practices. I would combine
the first and third block into one sentence and I would unwrap the
warning text (it is okay to be long for a little clarity). Perhaps
something like this?
| The '-fbasic-block-sections=labels' option has been deprecated in tip of
| tree clang (20.0.0) [1]. While the option still works, a warning is
| emitted:
|
| clang: warning: argument '-fbasic-block-sections=labels' is deprecated, use '-fbasic-block-address-map' instead [-Wdeprecated]
then putting:
| Link: https://github.com/llvm/llvm-project/pull/110039 [1]
right above your Signoff.
> This patch adds a check to the Makefile to set the proper option.
Try to avoid saying "this patch" in kernel commit messages, it is
implicit. It would be cleaner to say something like:
| Add a version check to set the proper option.
You can see some more commit message tips in sections 4.2.3 to 4.2.6 in
the tip documentation:
https://docs.kernel.org/process/maintainer-tip.html#changelog
> If the option change is later integrated in v19.x, we need to update the
> check accordingly.
I think it is worth dropping this sentence entirely, as I would argue
that it is pretty unlikely that this option gets backported to
release/19.x, since it would basically accelerate the deprecation
timeline, which seems unreasonable to do in a stable release.
> Signed-off-by: Rong Xu <xur@...gle.com>
> Reported-by: Nathan Chancellor <nathan@...nel.org>
> ---
> scripts/Makefile.propeller | 14 ++++++++++++--
> 1 file changed, 12 insertions(+), 2 deletions(-)
>
> diff --git a/scripts/Makefile.propeller b/scripts/Makefile.propeller
> index 344190717e47..1a68ea0d1fe4 100644
> --- a/scripts/Makefile.propeller
> +++ b/scripts/Makefile.propeller
> @@ -5,7 +5,13 @@ ifdef CLANG_PROPELLER_PROFILE_PREFIX
> CFLAGS_PROPELLER_CLANG := -fbasic-block-sections=list=$(CLANG_PROPELLER_PROFILE_PREFIX)_cc_profile.txt -ffunction-sections
> KBUILD_LDFLAGS += --symbol-ordering-file=$(CLANG_PROPELLER_PROFILE_PREFIX)_ld_profile.txt --no-warn-symbol-ordering
> else
> - CFLAGS_PROPELLER_CLANG := -fbasic-block-sections=labels
> + # Staring with clang v20, the '-fbasic-block-sections=labels' option is
> + # deprecated. Use the recommended '-fbasic-block-address-map' option.
Might be worth linking the specific change here for archeology sake.
> + ifeq ($(call clang-min-version, 200000),y)
> + CFLAGS_PROPELLER_CLANG := -fbasic-block-address-map
> + else
> + CFLAGS_PROPELLER_CLANG := -fbasic-block-sections=labels
> + endif
> endif
>
> # Propeller requires debug information to embed module names in the profiles.
> @@ -21,7 +27,11 @@ ifdef CONFIG_LTO_CLANG_THIN
> ifdef CLANG_PROPELLER_PROFILE_PREFIX
> KBUILD_LDFLAGS += --lto-basic-block-sections=$(CLANG_PROPELLER_PROFILE_PREFIX)_cc_profile.txt
> else
> - KBUILD_LDFLAGS += --lto-basic-block-sections=labels
> + ifeq ($(call clang-min-version, 200000),y)
Would it be better/more accurate to check the LLD version here? I doubt
it would really happen but it is possible for someone to have LLVM=1
(using their system wide ld.lld and LLVM tools) and CC=clang-20 to just
use an updated clang. Perhaps:
ifeq ($(call test-ge, $(CONFIG_LLD_VERSION), 200000),y)
here?
> + KBUILD_LDFLAGS += --lto-basic-block-address-map
> + else
> + KBUILD_LDFLAGS += --lto-basic-block-sections=labels
> + endif
> endif
> endif
>
>
> base-commit: 0dcc2d1066150787017a71f035145c566597dec7
> --
> 2.47.0.277.g8800431eea-goog
>
Powered by blists - more mailing lists