Re: [PATCH 0/6] Introduce ID_PFR2 and other CPU feature changes

[Peter Maydell]

On Fri, 14 Feb 2020 at 04:23, Anshuman Khandual
<anshuman.khandual@xxxxxxx> wrote:
>
>
>
> On 01/28/2020 06:09 PM, Anshuman Khandual wrote:
> > This series is primarily motivated from an adhoc list from Mark Rutland
> > during our ID_ISAR6 discussion [1]. Besides, it also includes a patch
> > which does macro replacement for various open bits shift encodings in
> > various CPU ID registers. This series is based on linux-next 20200124.
> >
> > [1] https://patchwork.kernel.org/patch/11287805/
> >
> > Is there anything else apart from these changes which can be accommodated
> > in this series, please do let me know. Thank you.
>
> Just a gentle ping. Any updates, does this series looks okay ? Is there
> anything else related to CPU ID register feature bits, which can be added
> up here. FWIW, the series still applies on v5.6-rc1.

I just ran into some "32-bit KVM doesn't expose all the ID
registers to userspace via the ONE_REG API" issues today.
I don't know if they'd be reasonable as something to include
in this patchset or if they're unrelated.

Anyway, missing stuff I have noticed specifically:
 * MVFR2
 * ID_MMFR4
 * ID_ISAR6

More generally I would have expected all these 32-bit registers
to exist and read-as-zero for the purpose of the ONE_REG APIs,
because that's what the architecture says is supposed to happen
and it means we have compatibility and QEMU doesn't gradually
build up lots of "kernel doesn't support this yet" conditionals...
I think we get this right for 64-bit KVM, but can we do it for
32-bit as well?

thanks
-- PMM