On 10/29/13 11:50, Peter Lieven wrote: > On 29.10.2013 11:48, Paolo Bonzini wrote: >> Il 29/10/2013 11:40, Peter Lieven ha scritto: >>>> The KVM signature should be at CPUID leaf 0x40000100. >>> If I enable hyperv for all vServers the signature is at >>> KVM_CPUID_SIGNATURE_NEXT (0x40000100) otherwise >>> at KVM_CPUID_SIGNATURE (0x0). >> KVM_CPU_ID_SIGNATURE is 0x40000000. >> >>> Does this matter to Linux? >> For recent versions it doesn't. Older versions will not be able to use >> kvmclock (and other PV enhancements for KVM such as steal time or PV >> EOI). > Ok, so this is not an option today - maybe later... > > Any other idea to detect Windows is running or trying to start?
I don't know what I'm talking about. But: - Maybe tracing MSR accesses could give you a "profile". - Windows' ACPI parser is super cranky. You could pass in a custom (but standardized) ACPI table on the command line (-acpitable) that only triggers some warnings in Linux's port of ACPICA, but crashes Windows (BSOD). Like, write & compile a simple table to AML, then mess it up (eg. Package encoding or some such) with a hex editor. This would take some experimentation as well, but searching existing bug reports could help. Laszlo