Re: [PATCH] arm64: cpufeature: Export Armv8.6 Matrix feature to userspace

[Julien Grall]

Hi Will,

On 29/10/2019 11:15, Will Deacon wrote:
> On Fri, Oct 25, 2019 at 06:10:56PM +0100, Julien Grall wrote:
> > This patch provides support for reporting the presence of Armv8.6
> > Matrix and its optional features to userspace.
>
> Are you sure this is 8.6 and not earlier?

This was introduced by Armv8.6 see [1] but allowed to be used by Armv8.2 and 
onwards.

> > This based on [1] + commit ec52c7134b1f "arm64: cpufeature: Treat
> > ID_AA64ZFR0_EL1 as RAZ when SVE is not enabled" (taken from v5.4-rc4).
> >
> > [1]  arm64/for-next/elf-hwcap-docs
> > ---
> >   Documentation/arm64/cpu-feature-registers.rst |  8 ++++++++
> >   Documentation/arm64/elf_hwcaps.rst            | 15 +++++++++++++++
> >   arch/arm64/include/asm/hwcap.h                |  4 ++++
> >   arch/arm64/include/asm/sysreg.h               |  7 +++++++
> >   arch/arm64/include/uapi/asm/hwcap.h           |  4 ++++
> >   arch/arm64/kernel/cpufeature.c                | 11 +++++++++++
> >   arch/arm64/kernel/cpuinfo.c                   |  4 ++++
> >   7 files changed, 53 insertions(+)
> >
> > diff --git a/Documentation/arm64/cpu-feature-registers.rst b/Documentation/arm64/cpu-feature-registers.rst
> > index ffcf4e2c71ef..d1d6d56a7b08 100644
> > --- a/Documentation/arm64/cpu-feature-registers.rst
> > +++ b/Documentation/arm64/cpu-feature-registers.rst
> > @@ -193,6 +193,8 @@ infrastructure:
> >        +------------------------------+---------+---------+
> >        | Name                         |  bits   | visible |
> >        +------------------------------+---------+---------+
> > +     | I8MM                         | [52-55] |    y    |
> > +     +------------------------------+---------+---------+
>
> Looking at:
>
> https://developer.arm.com/docs/ddi0601/latest/aarch64-system-registers/id_aa64isar1_el1
>
> Then I8MM is advertised as "Armv8.2", alongside other fields that we haven't
> listed here such as BF16 and SPECRES.
>
> So we probably want a patch bringing all of this up to speed, rather than
> randomly advertising some features and not others.
>
> >        | SB                           | [36-39] |    y    |
> >        +------------------------------+---------+---------+
> >        | FRINTTS                      | [32-35] |    y    |
> > @@ -227,6 +229,12 @@ infrastructure:
> >        +------------------------------+---------+---------+
> >        | Name                         |  bits   | visible |
> >        +------------------------------+---------+---------+
> > +     | F64MM                        | [56-59] |    y    |
> > +     +------------------------------+---------+---------+
> > +     | F32MM                        | [52-55] |    y    |
> > +     +------------------------------+---------+---------+
> > +     | I8MM                         | [44-47] |    y    |
> > +     +------------------------------+---------+---------+
>
> Urgh, we're inconsistent in our bitfields. Some are [lo-hi] whilst others
> are [hi-lo]. Please can you fix that in a preparatory patch? I prefer
> [hi-lo] and it matches the arch docs.

Sure.

Cheers,

[1] 
https://community.arm.com/developer/ip-products/processors/b/processors-ip-blog/posts/arm-architecture-developments-armv8-6-a

> Will

--
Julien Grall