Re: [PATCH 2/2] powerpc/64: Use -mtune=pwr10/9/8 for clang

From: Nathan Chancellor <nathan@kernel.org>
To: Michael Ellerman <mpe@ellerman.id.au>
Cc: linuxppc-dev@lists.ozlabs.org, linux-kbuild@vger.kernel.org,
	llvm@lists.linux.dev, Nick Desaulniers <ndesaulniers@google.com>
Subject: Re: [PATCH 2/2] powerpc/64: Use -mtune=pwr10/9/8 for clang
Date: Thu, 2 Mar 2023 09:43:24 -0700	[thread overview]
Message-ID: <20230302164324.GB3010526@dev-arch.thelio-3990X> (raw)
In-Reply-To: <20230302131656.50626-2-mpe@ellerman.id.au>

Hi Michael,

Thanks for the workaround and sorry this has come to bite us :/

On Fri, Mar 03, 2023 at 12:16:56AM +1100, Michael Ellerman wrote:
> For the -mtune option clang doesn't accept power10/9/8, instead it
> accepts pwr10/9/8. That will be fixed in future versions of clang, but
> the kernel must support the clang versions in the wild.
> 
> So add support for the "pwr" spelling if clang is in use.
> 
> Reported-by: Nathan Chancellor <nathan@kernel.org>

I think that should actually be

Reported-by: Nick Desaulniers <ndesaulniers@google.com>

> BugLink: https://github.com/ClangBuiltLinux/linux/issues/1799
> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

> ---
>  arch/powerpc/platforms/Kconfig.cputype | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> Need to confirm the clang <= 16 statement is correct.

Currently, this is indeed the case. It is possible that Nemanja's patch
will get applied to release/16.x before 16.0.0 final but it might not.
We can always update it later. I think we do want to push to get that
patch applied because I forgot that it is only in 16.0.0 that '-mtune'
starts to do something on PowerPC:

https://github.com/llvm/llvm-project/commit/1dc26b80b872a94c581549a21943756a8c3448a3

Prior to that change, '-mtune' was accepted but did nothing. It is only
once it was hooked up to the backend that we got the spew of warnings. I
think that warrants us trying to get Nemanja's patch into 16.0.0, which
may allow us to drop this workaround altogether...

> diff --git a/arch/powerpc/platforms/Kconfig.cputype b/arch/powerpc/platforms/Kconfig.cputype
> index 7d7477b73951..e4e0e81be7de 100644
> --- a/arch/powerpc/platforms/Kconfig.cputype
> +++ b/arch/powerpc/platforms/Kconfig.cputype
> @@ -278,6 +278,10 @@ config TUNE_CPU
>  	default "-mtune=power10" if POWERPC64_CPU && CC_IS_GCC   && $(cc-option,-mtune=power10)
>  	default "-mtune=power9"  if POWERPC64_CPU && CC_IS_GCC   && $(cc-option,-mtune=power9)
>  	default "-mtune=power8"  if POWERPC64_CPU && CC_IS_GCC   && $(cc-option,-mtune=power8)
> +	# clang <= 16 only supports the "pwr" names
> +	default "-mtune=pwr10"   if POWERPC64_CPU && CC_IS_CLANG && $(cc-option,-mtune=pwr10)
> +	default "-mtune=pwr9"    if POWERPC64_CPU && CC_IS_CLANG && $(cc-option,-mtune=pwr9)
> +	default "-mtune=pwr8"    if POWERPC64_CPU && CC_IS_CLANG && $(cc-option,-mtune=pwr8)
>  
>  config PPC_BOOK3S
>  	def_bool y
> -- 
> 2.39.2
> 