On 30.03.20 15:30, Janosch Frank wrote: > On 3/30/20 3:15 PM, David Hildenbrand wrote: >> On 30.03.20 15:09, Janosch Frank wrote: >>> On 3/30/20 3:03 PM, Janosch Frank wrote: >>>> On 3/30/20 2:50 PM, David Hildenbrand wrote: >>>>> On 30.03.20 14:20, Janosch Frank wrote: >>>>>> + report(data->vm[0].total_cpus == smp_query_num_cpus(), "cpu # total"); >>>>>> + report(data->vm[0].conf_cpus == smp_query_num_cpus(), "cpu # configured"); >>>>>> + report(data->vm[0].standby_cpus == 0, "cpu # standby"); >>>>>> + report(data->vm[0].reserved_cpus == 0, "cpu # reserved"); >>>>> >>>>> IIRC, using -smp 1,maxcpus=X, you could also test the reported reserved >>>>> CPUs. >>>> >>>> Will try that >>> >>> Just like I thought, QEMU does not manipulate cpu counts and KVM >>> pre-sets standby and reserved to 0. So we have absolutely no change when >>> adding the smp parameter. >> >> Well, for TCG it is properly implemented. Is this a BUG in KVM's STSI code? >> > > KVM tracks online cpus and created cpus, but only reports the online > ones in stsi. > Will QEMU register/create a reserved CPU with KVM? > > To fix this we could also fix-up the cpu reporting in QEMU after KVM > wrote its results. > > @Christian: Guidance? The standby only make sense if we implement the sclp cpu configure things, which we do not. So this can be considered reserved, while standby must be 0. I think we could implement this in QEMUs insert_stsi_3_2_2 function in kvm.c
