On 16.08.2024 13:12, Sergiy Kibrik wrote:
> Do not compile handlers of guest access to AMD-specific MSRs when 
> CONFIG_AMD=n.

What I'm missing in the description is clarification on how boundaries were
drawn. In guest_rdmsr() there is, for example, also handling of 
MSR_AMD_PATCHLEVEL.
Which I'm okay to leave aside for now, as long as it's clear why that is.

> --- a/xen/arch/x86/msr.c
> +++ b/xen/arch/x86/msr.c
> @@ -219,6 +219,7 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t 
> *val)
>          *val = msrs->tsc_aux;
>          break;
>  
> +#ifdef CONFIG_AMD
>      case MSR_K8_SYSCFG:
>      case MSR_K8_TOP_MEM1:
>      case MSR_K8_TOP_MEM2:
> @@ -281,6 +282,7 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t 
> *val)
>                                     ? 0 : (msr - MSR_AMD64_DR1_ADDRESS_MASK + 
> 1),
>                                     ARRAY_SIZE(msrs->dr_mask))];
>          break;
> +#endif /* CONFIG_AMD */
>  
>          /*
>           * TODO: Implement when we have better topology representation.
> @@ -552,6 +554,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t 
> val)
>              wrmsr_tsc_aux(val);
>          break;
>  
> +#ifdef CONFIG_AMD
>      case MSR_VIRT_SPEC_CTRL:
>          if ( !cp->extd.virt_ssbd )
>              goto gp_fault;
> @@ -598,6 +601,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t 
> val)
>          if ( v == curr && (curr->arch.dr7 & DR7_ACTIVE_MASK) )
>              wrmsrl(msr, val);
>          break;
> +#endif /* CONFIG_AMD */
>  
>      default:
>          return X86EMUL_UNHANDLEABLE;

Is just adding #ifdef-s actually correct? That results in different behavior on
e.g. Intel hardware, I think, depending on whether AMD=y or AMD=n. In the latter
case the function will now return X86EMUL_UNHANDLEABLE, while in the former case
it would return X86EMUL_EXCEPTION.

Jan

Reply via email to