* Borislav Petkov (bp@xxxxxxxxx) wrote: > On Wed, Oct 20, 2021 at 11:10:23AM -0500, Michael Roth wrote: > > At which point we then switch to using the CPUID table? But at that > > point all the previous CPUID checks, both SEV-related/non-SEV-related, > > are now possibly not consistent with what's in the CPUID table. Do we > > then revalidate? > > Well, that's a tough question. That's basically the same question as, > does Linux support heterogeneous cores and can it handle hardware > features which get enabled after boot. The perfect example is, late > microcode loading which changes CPUID bits and adds new functionality. > > And the answer to that is, well, hard. You need to decide this on a > case-by-case basis. I can imagine a malicious hypervisor trying to return different cpuid answers to different threads or even the same thread at different times. > But isn't it that the SNP CPUID page will be parsed early enough anyway > so that kernel proper will see only SNP CPUID info and init properly > using that? > > > Even a non-malicious hypervisor might provide inconsistent values > > between the two sources due to bugs, or SNP validation suppressing > > certain feature bits that hypervisor otherwise exposes, etc. > > There's also migration, lemme point to a very recent example: > > https://lore.kernel.org/r/20211021104744.24126-1-jane.malalane@xxxxxxxxxx Ewww. > which is exactly what you say - a non-malicious HV taking care of its > migration pool. So how do you handle that? Well, the spec (AMD 56860 SEV spec) says: 'If firmware encounters a CPUID function that is in the standard or extended ranges, then the firmware performs a check to ensure that the provided output would not lead to an insecure guest state' so I take that 'firmware' to be the PSP; that wording doesn't say that it checks that the CPUID is identical, just that it 'would not lead to an insecure guest' - so a hypervisor could hide any 'no longer affected by' flag for all the CPUs in it's migration pool and the firmware shouldn't complain; so it should be OK to pessimise. Dave > > Now all the code after sme_enable() can potentially take unexpected > > execution paths, where post-sme_enable() code makes assumptions about > > pre-sme_enable() checks that may no longer hold true. > > So as I said above, if you parse SNP CPUID page early enough, you don't > have to worry about feature rediscovery. Early enough means, before > identify_boot_cpu(). > > -- > Regards/Gruss, > Boris. > > https://people.kernel.org/tglx/notes-about-netiquette > -- Dr. David Alan Gilbert / dgilbert@xxxxxxxxxx / Manchester, UK