On 11/27/2023 11:43 AM, EwanHai wrote:
Commit 4a910e1 ("target/i386: do not set unsupported VMX secondary
execution controls") implemented a workaround for hosts that have
specific CPUID features but do not support the corresponding VMX
controls, e.g., hosts support RDSEED but do not support RDSEED-Exiting.
In detail, commit 4a910e1 introduced a flag `has_msr_vmx_procbased_clts2`.
If KVM has `MSR_IA32_VMX_PROCBASED_CTLS2` in its msr list, QEMU would
use KVM's settings, avoiding any modifications to this MSR.
However, this commit (4a910e1) didn't account for cases in older Linux
kernels(<5.3) where `MSR_IA32_VMX_PROCBASED_CTLS2` is in
`kvm_feature_msrs`-obtained by ioctl(KVM_GET_MSR_FEATURE_INDEX_LIST),
but not in `kvm_msr_list`-obtained by ioctl(KVM_GET_MSR_INDEX_LIST).
As a result,it did not set the `has_msr_vmx_procbased_clts2` flag based
on `kvm_msr_list` alone, even though KVM maintains the value of this MSR.
This patch supplements the above logic, ensuring that
`has_msr_vmx_procbased_clts2` is correctly set by checking both MSR
lists, thus maintaining compatibility with older kernels.
Signed-off-by: EwanHai <ewanhai-oc@xxxxxxxxxxx>
---
In response to the suggestions from ZhaoLiu(zhao1.liu@xxxxxxxxx),
the following changes have been implemented in v2:
- Adjusted some punctuation in the commit message as per the
suggestions.
- Added comments to the newly added code to indicate that it is a
compatibility fix.
v1 link:
https://lore.kernel.org/all/20230925071453.14908-1-ewanhai-oc@xxxxxxxxxxx/
---
target/i386/kvm/kvm.c | 14 ++++++++++++++
1 file changed, 14 insertions(+)
diff --git a/target/i386/kvm/kvm.c b/target/i386/kvm/kvm.c
index 11b8177eff..c8f6c0b531 100644
--- a/target/i386/kvm/kvm.c
+++ b/target/i386/kvm/kvm.c
@@ -2296,6 +2296,7 @@ void kvm_arch_do_init_vcpu(X86CPU *cpu)
static int kvm_get_supported_feature_msrs(KVMState *s)
{
int ret = 0;
+ int i;
if (kvm_feature_msrs != NULL) {
return 0;
@@ -2330,6 +2331,19 @@ static int kvm_get_supported_feature_msrs(KVMState *s)
return ret;
}
+ /*
+ * Compatibility fix:
+ * Older Linux kernels(<5.3) include the MSR_IA32_VMX_PROCBASED_CTLS2
we can be more accurate, that kernel version 4.17 to 5.2, reports
MSR_IA32_VMX_PROCBASED_CTLS2 in KVM_GET_MSR_FEATURE_INDEX_LIST but not
KVM_GET_MSR_INDEX_LIST.
+ * only in feature msr list, but not in regular msr list. This lead to
+ * an issue in older kernel versions where QEMU, through the regular
+ * MSR list check, assumes the kernel doesn't maintain this msr,
+ * resulting in incorrect settings by QEMU for this msr.
+ */
+ for (i = 0; i < kvm_feature_msrs->nmsrs; i++) {
+ if (kvm_feature_msrs->indices[i] == MSR_IA32_VMX_PROCBASED_CTLS2) {
+ has_msr_vmx_procbased_ctls2 = true;
+ }
+ }
I'm wondering should we move all the initialization of has_msr_*, that
associated with feature MSRs, to here. e.g., has_msr_arch_capabs,
has_msr_vmx_vmfunc,...
return 0;
}
