Re: [PATCH 00/32] x86/msr: Drop 32-bit MSR interfaces

[Jürgen Groß]

On 29.06.26 08:52, Arnd Bergmann wrote:
> On Mon, Jun 29, 2026, at 08:04, Juergen Gross wrote:
> > For accessing the MSR registers on the local CPU, there are 2 types of
> > interfaces: the "modern" 64-bit ones (rdmsrq() etc.) and the 32-bit
> > ones (rdmsr() etc.) which are using the upper and lower 32-bit halves
> > of the 64-bit wide MSR register values.
> >
> > The 32-bit interfaces are not optimal for 3 reasons:
> >
> > - They are based on primitives using 64-bit sized values anyway.
> >
> > - Modern x86 CPUs have added support for MSR access instructions using
> >    an immediate value instead of a register for addressing the MSR,
> >    while the value is in a 64-bit register.
> >
> > - rdmsr() is a macro storing the upper and lower 32-bit halves in
> >    variables specified as macro parameters. This is obscuring variable
> >    assignment through a macro. Additionally rdmsrq() is mimicking this
> >    pattern by being a macro, too, with the target variable specified as
> >    a parameter as well.
> >
> > For those reasons drop the 32-bit interfaces for accessing the x86 MSR
> > registers completely and only use the 64-bit variants.
>
> Hi Jürgen,
>
> I assume this is fine, but since you don't mention it explicitly here,
> please clarify what this means for 32-bit CPUs without the rdmsrq
> instruction. Those will continue using the same instructions as before
> and just change the calling conventions, right?

Yes. I thought this would be clear from the following:

  - They are based on primitives using 64-bit sized values anyway.

> > Note that most patches of this series are independent from each other.
> > Only the patches removing a specific interface (patches 7, 15, 26 and
> > 30) and the last two patches of the series depend on all previous
> > patches.
>
> It looks like you are touching most files twice or more here, to
> first convert from rdmsr to rdmsrq and then to change the
> two-argument rdmsrq() macro to a single-argument inline. If you
> introduce the inline version of rdmsrq() first, you should be
> able to skip the second step (patch 31) as they could be able
> to coexist.

I've discussed how to structure the series with Ingo Molnar before [1]. The
current approach was his preference.


Juergen

[1]: https://lore.kernel.org/lkml/f8d02c78-4681-4043-a5fa-921fa790b1b4@xxxxxxxx/

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature