Skip to content

Conversation

yamahata
Copy link
Contributor

Keep the same behavior regarding CPUID MCE, MCA, and MTRR by setting MCA and MTRR bits of PARAVIRT CTLS.

TDX VE reduction changes guest behavior when the host and (L1) guest opt-in by setting TD_CTLS_REDUCE_VE. With REDUCE_VE enabled, the guest has to further opt in for each feature, like MCA and MTRR. Otherwise, such features are disabled. e.g., 0 in CPUID and #GP when accessing related MSRs.

The upstream Linux TDX guest now opts in REDUCE VE and doesn't opt in any features, including MCA and MTRR. The L2 guest may be confused. Opt-in MCA and MTRR to keep the L0/L1 VMM behavior for the L2 guest to unblock it. As REDUCE VE defines other features, they are future TODO.

Keep the same behavior regarding CPUID MCE, MCA, and MTRR by setting
MCA and MTRR bits of PARAVIRT CTLS.

TDX VE reduction changes guest behavior when the host and (L1) guest
opt-in by setting TD_CTLS_REDUCE_VE.  With REDUCE_VE enabled, the
guest has to further opt in for each feature, like MCA and MTRR.
Otherwise, such features are disabled.  e.g., 0 in CPUID and #GP when
accessing related MSRs.

The upstream Linux TDX guest now opts in REDUCE VE and doesn't opt in
any features, including MCA and MTRR.  The L2 guest may be confused.
Opt-in MCA and MTRR to keep the L0/L1 VMM behavior for the L2 guest to
unblock it.  As REDUCE VE defines other features, they are future
TODO.

Signed-off-by: Isaku Yamahata <[email protected]>
@yamahata
Copy link
Contributor Author

@microsoft-github-policy-service agree company="Intel corporation"

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant