On 05/10/2011 02:26 AM, Lai Jiangshan wrote:
+int qemuMonitorJSONInjectNMI(qemuMonitorPtr mon) +{ + int ret; + virJSONValuePtr cmd; + virJSONValuePtr reply = NULL; + + cmd = qemuMonitorJSONMakeCommand("inject-nmi", NULL); + if (!cmd) + return -1; + + if ((ret = qemuMonitorJSONCommand(mon, cmd, &reply)) < 0) + goto cleanup; + + if (qemuMonitorJSONHasError(reply, "CommandNotFound") && + qemuMonitorCheckHMP(mon, "inject-nmi")) { + VIR_DEBUG0("inject-nmi command not found, trying HMP"); + ret = qemuMonitorTextInjectNMI(mon);
+ +int qemuMonitorTextInjectNMI(qemuMonitorPtr mon) +{ + const char *cmd = "inject-nmi"; + char *reply = NULL; + + if (qemuMonitorHMPCommand(mon, cmd, &reply) < 0) + goto fail; + + if (strstr(reply, "unknown command") != NULL) { + VIR_FREE(reply); + + /* fallback to 'nmi' if qemu has not supported "inject-nmi" yet. */ + cmd = "nmi 0"; + reply = NULL; + if (qemuMonitorHMPCommand(mon, cmd, &reply) < 0)
I just noticed that upstream qemu now has QMP 'inject-nmi' (commit a404666), and that as a result, HMP 'nmi' command no longer takes an integer option (commit e9b4b43). If libvirt uses QMP, then this is not an issue, but if libvirt uses HMP while talking to a newer qemu, is this attempt to do 'nmi 0' going to cause a syntax error? That is, do we need a patch to libvirt to detect a usage error with 'nmi 0' on hmp that falls back to trying plain 'nmi' before giving up altogether? -- Eric Blake eblake@redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org