On 3/19/2022 12:55 AM, Isaku Yamahata wrote:
On Thu, Mar 17, 2022 at 09:58:45PM +0800, Xiaoyao Li <[email protected]> wrote:
...
+void tdx_get_supported_cpuid(uint32_t function, uint32_t index, int reg, + uint32_t *ret) +{ + switch (function) { + case 1: + if (reg == R_ECX) { + *ret &= ~CPUID_EXT_VMX; + } + break; + case 0xd: + if (index == 0) { + if (reg == R_EAX) { + *ret &= (uint32_t)tdx_caps->xfam_fixed0 & XCR0_MASK; + *ret |= (uint32_t)tdx_caps->xfam_fixed1 & XCR0_MASK; + } else if (reg == R_EDX) { + *ret &= (tdx_caps->xfam_fixed0 & XCR0_MASK) >> 32; + *ret |= (tdx_caps->xfam_fixed1 & XCR0_MASK) >> 32; + } + } else if (index == 1) { + /* TODO: Adjust XSS when it's supported. */ + } + break; + case KVM_CPUID_FEATURES: + if (reg == R_EAX) { + *ret &= ~((1ULL << KVM_FEATURE_CLOCKSOURCE) | + (1ULL << KVM_FEATURE_CLOCKSOURCE2) | + (1ULL << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | + (1ULL << KVM_FEATURE_ASYNC_PF) | + (1ULL << KVM_FEATURE_ASYNC_PF_VMEXIT) | + (1ULL << KVM_FEATURE_ASYNC_PF_INT));Because new feature bit may be introduced in future (it's unlikely though), *ret &= (supported_bits) is better than *ret &= ~(unsupported_bits)
Good point, I will introduce supported_kvm_features for it.
