Description of problem: There's a new CPUId bit 80000021/bit 6 'NullSelectClearsBase - Null segment selector loads also clear the destination segment register base and limit' guests currently detect this behaviour at run time, but there's an upstream fix that starts using this bit and bz 2016959/2016961 request backports; once we start using this bit we should advertise if the cpu has it. Milan apparently does. Rome doesn't advertise the bit but is believed to have actually fixed it but not advertise it. We should advertise the bit in the Milan and Rome models. Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
Wei: Can you point to something that confirms Rome's behaviour on this?
Assigned to Amnon for initial triage per bz process and age of bug created or assigned to virt-maint without triage.
Babu is looking into this. Could you open access to BZ2016959/BZ2016961 to AMD?
(In reply to WEI HUANG from comment #3) > Babu is looking into this. Could you open access to BZ2016959/BZ2016961 to > AMD? Done.
Hi David, The CPUID feature NullSelectClearsBase (Function 80000021/bit 6) is available on Milan and future products. I will plan to add this bit in EPYC-Milan. However, this feature is not available in Rome platform. So, we cannot add it in EPYC-Rome. Thanks Babu
Hi Bandan, Is this one RFE bug according to Commen 0 and Comment 5? If so pleae help add the FutureFeature keywords. Thanks. Best regards Nana
Hi Nitesh, Bandan, According to this bug's Description, 'NullSelectClearsBase - Null segment selector loads also clear the destination segment register base and limit' When migrating VM (specify the cpu to be EPYC) from Rome/Milan to Naples, do you know what migration issue would happen?
(In reply to Li Xiaohui from comment #11) > Hi Nitesh, Bandan, > According to this bug's Description, > 'NullSelectClearsBase - Null segment selector loads also clear the > destination segment register base and limit' > > When migrating VM (specify the cpu to be EPYC) from Rome/Milan to Naples, do > you know what migration issue would happen? First, to respond to Nana, it looks more like a bug fix to me. Regarding migration, I am not entirely sure but from the kernel patch, the assumption is that guest wouldn't be able to detect the virtual cpuid bit when migrating to Naples and so would be aware of the X86_BUG_NULL_SEG bug. Maybe Juan is aware of this ?
After evaluating this issue, there are no plans to address it further or fix it in an upcoming release. Therefore, it is being closed. If plans change such that this issue will be fixed in an upcoming release, then the bug can be reopened.
Hi I redirect this one to Leonardo, he is the one that have looked at CPUid's and AMD machines in recent times. Leonardo?
This series addresses this problem for Milan and Genoa processors. https://lore.kernel.org/kvm/20230504205313.225073-1-babu.moger@amd.com/#r Paolo is planning to pull this series for the next release.