On Tue, Nov 16, 2021 at 2:50 AM Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: ... > However, the pmu_emulation test does not really need nmi_watchdog=0; > it is only needed by the PMU counters test because Linux reserves one > counter if nmi_watchdog=1, but the pmu_emulation test does not > allocate all counters in the same way. By removing the counters > tests from pmu_emulation, the check on nmi_watchdog=0 can be > removed. If Linux reserves a counter, shouldn't KVM_GET_SUPPORTED_CPUID remove that counter from the available set reported in CPUID.0AH? On Intel, that is. On AMD, we're just SOL.
