On Mon, May 12, 2025 at 07:13:08AM -0700, Sean Christopherson wrote:
>On Mon, May 12, 2025, Chao Gao wrote:
>> @@ -535,10 +538,20 @@ void fpstate_init_user(struct fpstate *fpstate)
>>
>> static void __fpstate_reset(struct fpstate *fpstate, u64 xfd)
>> {
>> - /* Initialize sizes and feature masks */
>> - fpstate->size = fpu_kernel_cfg.default_size;
>> + /*
>> + * Initialize sizes and feature masks. Supervisor features and
>> + * sizes may diverge between guest FPUs and host FPUs, whereas
>> + * user features and sizes are always identical the same.
>
>Pick of of "identical" or "the same" :-)
Sure.
>
>And maybe explain why supervisor features can diverge, while the kernel ensures
>user features are identical? Ditto for the XFD divergence. E.g. I think this
>would be accurate (though I may be reading too much into user features):
>
> /*
> * Supervisor features (and thus sizes) may diverge between guest FPUs
> * and host FPUs, as some supervisor features are supported for guests
> * despite not being utilized by the host. User features and sizes are
> * always identical, which allows for common guest and userspace ABI.
> *
> * For the host, set XFD to the kernel's desired initialization value.
> * For guests, set XFD to its architectural RESET value.
> */
Yea, this looks much better.
