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

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

[Nathan Chancellor]

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
> 

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

[Michael Ellerman]

Nathan Chancellor <nathan@kernel.org> writes:
> 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>

I guess yeah.

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

Thanks.

>> ---
>>  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...

Aha OK, I missed that the warning was new in 16.

I'll sit on this for now then until we know if that change will make it
into clang 16.

cheers

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

[Nathan Chancellor]

On Fri, Mar 03, 2023 at 10:53:02AM +1100, Michael Ellerman wrote:
> Nathan Chancellor <nathan@kernel.org> writes:
> > 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>
> 
> I guess yeah.
> 
> >> BugLink: https://github.com/ClangBuiltLinux/linux/issues/1799
> >> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
> >
> > Reviewed-by: Nathan Chancellor <nathan@kernel.org>
> 
> Thanks.
> 
> >> ---
> >>  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...
> 
> Aha OK, I missed that the warning was new in 16.
> 
> I'll sit on this for now then until we know if that change will make it
> into clang 16.

It was merged into release/16.x a few hours ago:

https://github.com/llvm/llvm-project/issues/61128

https://github.com/llvm/llvm-project/commit/9b2e09e9fb1aa3bbe3668d7cc585188a3014d1b9

So I think this particular workaround is no longer needed :)

Cheers,
Nathan