Re: [PATCH v5 3/7] x86/fpu/xstate: Differentiate default features for host and guest FPUs

[Xin Li <xin@zytor.com>]

On 4/27/2025 10:51 PM, Xin Li wrote:
> On 4/25/2025 9:09 AM, Edgecombe, Rick P wrote:
>>>    And that will create a bit of a
>>>   :snafu if Linux does gain support for SSS.
>>>
>>> *:https://lore.kernel.org/kvm/ZM1jV3UPL0AMpVDI@google.com/
>> I chatted with Xin about this a few weeks ago. It sounds like FRED 
>> bare metal
>> SSS will not need CET_S state, but it wasn't 100% clear.
> 
> FRED reuses one CET_S MSR IA32_PL0_SSP, and give it an alias
> IA32_FRED_SSP0.

Native use of IA32_FRED_SSP0 is very much like IA32_FRED_RSP0:

1) Both are per-task constants.

2) Both are only used for delivering events when running userspace.


IA32_FRED_RSP0 is set on return to userspace:

https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=fe85ee391966c4cf3bfe1c405314e894c951f521


I suppose we'll likely apply the same approach to IA32_FRED_SSP0 if we
plan to enable SSS for the kernel.  This won't add any extra maintenance
cost, as both x86 and KVM maintainers are well aware.

Thanks!
     Xin