This patch will prevent guest to start when the maximum vcpus are greater than cpu topology limit. Currently similar checks do exist only during setvcpus. The patch adds the missing check. The c9cb35c255222 reverted similar check to avoid older configs from vanishing. The current patch adds it in domain validation part to be consistent with the original intent. Signed-off-by: Kothapally Madhu Pavan <kmp(a)linux.vnet.ibm.com&gt; --- src/qemu/qemu_process.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c index 184440d..f0d42b8 100644 --- a/src/qemu/qemu_process.c +++ b/src/qemu/qemu_process.c @@ -3738,6 +3738,13 @@ qemuValidateCpuCount(virDomainDefPtr def, return -1; } + if (def->cpu->sockets && virDomainDefGetVcpusMax(def) > + def->cpu->sockets * def->cpu->cores * def->cpu->threads) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("Maximum CPUs greater than topology limit")); + return -1; + }

On 02/07/2017 06:41 PM, Peter Krempa wrote: > On Tue, Feb 07, 2017 at 07:56:29 -0500, Kothapally Madhu Pavan wrote: > > This patch will prevent guest to start when the maximum > > vcpus are greater than cpu topology limit. Currently similar > > checks do exist only during setvcpus. The patch adds the > > missing check. The c9cb35c255222 reverted similar check to > > avoid older configs from vanishing. The current patch adds > > it in domain validation part to be consistent with the > > original intent. > > > > Signed-off-by: Kothapally Madhu Pavan <kmp(a)linux.vnet.ibm.com&gt; > > --- > > src/qemu/qemu_process.c | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c > > index 184440d..f0d42b8 100644 > > --- a/src/qemu/qemu_process.c > > +++ b/src/qemu/qemu_process.c > > @@ -3738,6 +3738,13 @@ qemuValidateCpuCount(virDomainDefPtr def, > > return -1; > > } > > + if (def->cpu->sockets && virDomainDefGetVcpusMax(def) > > > + def->cpu->sockets * def->cpu->cores * def->cpu->threads) { > > + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", > > + _("Maximum CPUs greater than topology limit")); > > + return -1; > > + } > This exact logic is done in qemuDomainDefValidate: > > + if (virDomainDefGetVcpusTopology(def, &topologycpus) == 0 && > + topologycpus != virDomainDefGetVcpusMax(def)) { > + /* presence of query-hotpluggable-cpus should be a good enough witness */ > + if (virQEMUCapsGet(qemuCaps, QEMU_CAPS_QUERY_HOTPLUGGABLE_CPUS)) { > + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", > + _("CPU topology doesn't match maximum vcpu count")); > + goto cleanup; > } > } > > See commit 043ba4a40a4ae26cf616146d0d1c129d65b156b8. The only difference > is that the code is validated only for certain versions of qemu known to > reject such configuration. > > Is that not enough? <vcpu placement='static'>64</vcpu> <cpu> <topology sockets='1' cores='4' threads='8'/> </cpu> # virsh start one-0 error: Failed to start domain one-0 error: internal error: process exited while connecting to monitor: 2017-02-08T06:41:59.615692Z qemu-system-ppc64: cpu topology: sockets (1) * cores (4) * threads (8) < smp_cpus (64) I have qemu version 2.5.0 which doesn't have cpu hotplug capability hence I get this error. Any ways we can improve and error out using libvirt? I tried with qemu-2.4.1 and still see the error, from this I can say smp topology check is maintained by qemu even before cpu hotplug feature is added.

When we had vcpu count validation in XML parser, domains with above mentioned config were rejected irrespective of qemu capabilities.