http://bugzilla.opensuse.org/show_bug.cgi?id=1202541 http://bugzilla.opensuse.org/show_bug.cgi?id=1202541#c16 --- Comment #16 from Bernhard Wiedemann <bwiedemann@suse.com> --- Another interesting find: atreju6-suse:~ # cat /proc/cpuinfo |grep lwp|wc -l 16 atreju6-suse:~ # cat /proc/cpuinfo |grep nodeid_msr|wc -l 32 Only half of the cores have the lwp bit set and restarting libvirtd can make it appear or disappear in the virsh capabilities feature flags - probably if the process gets scheduled on the right 50%. http://developer.amd.com/wordpress/media/2012/10/43724.pdf p23 says
LWP is supported on a processor if CPUID Fn8000_0001_ECX[LWP] (bit 15) is set. This bit is identical to the value of CPUID Fn0000_000D_EDX_x0[bit 30], which is bit 62 of the XFeatureSupportedMask and indicates XSAVE support for LWP. A system can check either of those bits to determine if LWP is supported.
With the cpu host-passthrough mode we currently use, this random lwp availability could be the cause of trouble... unless guests never use it. And so far I did not find a single VM that listed the bit. -- You are receiving this mail because: You are on the CC list for the bug.