> > Can you move the call to kvm_cpu_xsave_init() after
> > x86_cpu_enable_xsave_components()?  Is it used anywhere before the CPU is
> > running?
> 
> Yes, this is an "ugly" palce. I did not fully defer the initialization
> of the xstate array earlier also because I observed that both HVF and
> TCG have similar xsave initialization interfaces within their accelerator's
> cpu_instance_init() function.
> 
> I think it might be better to do the same thing for HVF & TCG as well
> (i.e., defer xstate initialization). Otherwise, if we only modify QEMU
> KVM logic, it looks a bit fragmented... What do you think?

Ah, kvm_arch_get_supported_cpuid() currently caches the obtained CPUID.
Delaying xstate initialization would require refreshing the previously
cached old CPUID... This makes the cost of delaying xstate
initialization higher, and I think this way becames not appropriate.

Either keep the current different ways, or back to use host_cpuid for
everything :( . Which do you think is better?

Regards,
Zhao


Reply via email to