Re: [libvirt] [PATCH 1/1] Remove contiguous CPU indexes assumption

On 02/27/2013 05:13 AM, Li Zhang wrote: Without any test case under 'make check' that exposes the failure, I'm a bit worried that this might cause regressions on other setups. I'm not seven sure I even understand the scope of the problem. Is it something specific to running a qemu-system-ppc64 emulator (but can be reproduced on any host architecture), or is it specific to running on a powerpc host (but happens for any version of qemu-* target architecture), or is it something else altogether? We have test framework in place to allow replaying of a QMP JSON response and seeing how qemu_monitor_json.c deals with it; what I'd really like to see is a side-by-side comparison of what the QMP 'query-cpus' command returns for a guest with multiple vcpus on a setup where you are seeing the problem, when compared to a setup that does not have the issue. You can get at this with virsh qemu-monitor-command $dom '{"execute":"query-cpus"}', if that helps. To help you out, here's what I got for a guest using 3 vcpus on my x86_64 host machine and using the qemu-kvm binary: # virsh qemu-monitor-command guest '{"execute":"query-cpus"}' {"return":[{"current":true,"CPU":0,"pc":1025829,"halted":false,"thread_id":5849},{"current":false,"CPU":1,"pc":1005735,"halted":true,"thread_id":5850},{"current":false,"CPU":2,"pc":1005735,"halted":true,"thread_id":5851}],"id":"libvirt-9"} That is, your patch might be right, but I have not yet been convinced that it is right; and while things may currently be broken on ppc, it is not a recent breakage, so being conservative and either proving the fix is right (via testsuite addition) or deferring the fix until post 1.0.3 seems like safer alternatives. Or maybe someone else will chime in and agree to take it now, without waiting for my desired burden of proof. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org