Re: [PATCH v3 1/2] x86/mwait: Add support for idle via umwait

[Artem Bityutskiy <dedekind1@gmail.com>]

On Fri, 2023-06-30 at 00:04 +0200, Thomas Gleixner wrote:
> > +#ifdef CONFIG_X86_64
> > +/*
> > + * Monitor a memory address at 'rcx' using the 'umonitor' instruction.
> > + */
> > +static inline void __umonitor(const void *rcx)
> > +{
> > +       /* "umonitor %rcx" */
> > +#ifdef CONFIG_AS_TPAUSE
> 
> Are you sure that the instruction check for TPAUSE is sufficient to also
> include UMONITOR on all toolchains which support TPAUSE?

I've verified by building the kernel with gcc/binutils and clang/LLVM.
Builds, boots, umwait works, C0.2 happens with both.

I inspected gcc, binutils, and clang/llvm git logs: support for
'tpause' and 'umwait' arrived in the same commit. Details below.

'tpause' and 'umwait' instructions are very similar, arrived together,
guarded together by CPUID.7's "MWAITPKG" bit. Based on this, I'd generally
expect toolchains to support both or none.

I can add a note about this in the commit message too.

Details on commits in the projects I checked.

1. binutils-gcc git tree:
   de89d0a34d5 Enable Intel WAITPKG instructions.

2. gcc git tree:
   55f31ed10fd i386-common.c (OPTION_MASK_ISA_WAITPKG_SET, [...]): New defines.

3. llvm-project git tree:
   2e02579a76cf [OpenMP] Add use of TPAUSE

It'll take some time to re-test and and partially re-measure power/perf,
so I'll send new version a bit later.

Thanks,
Artem.