fix(host): cpu hotplug compatible with high qemu version (#24759)

pkg/hostman/guestman/guesttasks.go

@@ -2606,6 +2606,7 @@ type SGuestHotplugCpuMemTask struct {
 	addedCpuCount    int
 	addedVcpuIds     []int
 	cpuNumaPin       []*desc.SCpuNumaPin
+	cpuList          []monitor.HotpluggableCPU
 
 	addedMemSize     int
 	memSlotNewIndex  *int
@@ -2643,6 +2644,12 @@ func NewGuestHotplugCpuMemTask(
 // First at all add cpu count, second add mem size
 func (task *SGuestHotplugCpuMemTask) Start() {
 	if task.addCpuCount > 0 {
+		res, err := task.getHotpluggableCPUList()
+		if err != nil {
+			task.onFail(err.Error())
+			return
+		}
+		task.cpuList = res
 		task.startAddCpu()
 	} else if task.addMemSize > 0 {
 		task.startAddMem()
@@ -2682,16 +2689,12 @@ func (task *SGuestHotplugCpuMemTask) buildVcpusMap() {
 }
 
 func (task *SGuestHotplugCpuMemTask) startAddCpusWithFreeVcpuSet(vcpuSet []int) {
-	if task.addedCpuCount >= task.addCpuCount {
-		task.startAddMem()
-		return
-	}
-
-	vcpuId := vcpuSet[0]
-	cb := func(reason string) {
-		if len(reason) > 0 {
-			log.Errorln(reason)
-			task.onFail(reason)
+	for i := range vcpuSet {
+		vcpuId := vcpuSet[i]
+		err := task.AddCpu(task.cpuList, vcpuSet[i])
+		if err != nil {
+			log.Errorf("failed add cpu %d: %s", vcpuSet[i], err)
+			task.onFail(err.Error())
 			return
 		}
 
@@ -2726,9 +2729,9 @@ func (task *SGuestHotplugCpuMemTask) startAddCpusWithFreeVcpuSet(vcpuSet []int)
 		}
 
 		task.addedCpuCount += 1
-		task.startAddCpusWithFreeVcpuSet(vcpuSet[1:])
 	}
-	task.Monitor.AddCpu(vcpuId, cb)
+
+	task.startAddMem()
 }
 
 func (task *SGuestHotplugCpuMemTask) onGetCpuCount(count int) {
@@ -2738,7 +2741,12 @@ func (task *SGuestHotplugCpuMemTask) onGetCpuCount(count int) {
 
 func (task *SGuestHotplugCpuMemTask) doAddCpu() {
 	if task.addedCpuCount < task.addCpuCount {
-		task.Monitor.AddCpu(task.originalCpuCount+task.addedCpuCount, task.onAddCpu)
+		err := task.AddCpu(task.cpuList, task.originalCpuCount+task.addedCpuCount)
+		ret := ""
+		if err != nil {
+			ret = err.Error()
+		}
+		task.onAddCpu(ret)
 	} else {
 		task.startAddMem()
 	}

pkg/hostman/guestman/qemu-kvm.go

@@ -3061,8 +3061,57 @@ func (s *SKVMGuestInstance) doBlockIoThrottle() {
 	}
 }
 
-func (s *SKVMGuestInstance) startHotPlugVcpus(vcpuSet []int) error {
+func (s *SKVMGuestInstance) AddCpu(cpuList []monitor.HotpluggableCPU, cpu int) error {
 	var c = make(chan error)
+	cb := func(res string) {
+		var e error = nil
+		if len(res) > 0 {
+			e = errors.Errorf("failed add cpu %d: %s", cpu, res)
+		}
+		c <- e
+	}
+	if version.GT(s.QemuVersion, "4.0.2") {
+		var found = false
+		for i := range cpuList {
+			if cpuList[i].Props.SocketID == nil || cpuList[i].Props.CoreID == nil {
+				continue
+			}
+			if cpuList[i].QomPath != nil {
+				// cpu present
+				continue
+			}
+			cpusPerSocket := int(s.Desc.CpuDesc.MaxCpus / s.Desc.CpuDesc.Sockets)
+			coreId := int(*cpuList[i].Props.CoreID) + cpusPerSocket*int(*cpuList[i].Props.SocketID)
+			if coreId == cpu {
+				found = true
+				params := map[string]interface{}{
+					"id":        fmt.Sprintf("cpu-%d", cpu),
+					"core-id":   cpu % cpusPerSocket,
+					"socket-id": *cpuList[i].Props.SocketID,
+				}
+				if cpuList[i].Props.ThreadID != nil {
+					params["thread-id"] = *cpuList[i].Props.ThreadID
+				}
+				s.Monitor.DeviceAddCpu(cpuList[i].Type, params, cb)
+				break
+			}
+		}
+
+		if !found {
+			return errors.Errorf("hot pluggable cpu %d not found", cpu)
+		}
+	} else {
+		s.Monitor.AddCpu(cpu, cb)
+	}
+	err, _ := <-c
+	return err
+}
+
+func (s *SKVMGuestInstance) startHotPlugVcpus(vcpuSet []int) error {
+	cpuList, err := s.getHotpluggableCPUList()
+	if err != nil {
+		return errors.Wrap(err, "getHotpluggableCPUList")
+	}
 
 	for i := range vcpuSet {
 		if vcpuSet[i] == 0 {
@@ -3070,14 +3119,8 @@ func (s *SKVMGuestInstance) startHotPlugVcpus(vcpuSet []int) error {
 			continue
 		}
 
-		s.Monitor.AddCpu(vcpuSet[i], func(res string) {
-			var e error = nil
-			if len(res) > 0 {
-				e = errors.Errorf("failed add cpu %d: %s", vcpuSet[i], res)
-			}
-			c <- e
-		})
-		if err, _ := <-c; err != nil {
+		err = s.AddCpu(cpuList, vcpuSet[i])
+		if err != nil {
 			return err
 		}
 	}

pkg/hostman/monitor/monitor.go

@@ -220,6 +220,7 @@ type Monitor interface {
 	ObjectAdd(objectType string, params map[string]string, callback StringCallback)
 	DriveAdd(bus, node string, params map[string]string, callback StringCallback)
 	DeviceAdd(dev string, params map[string]string, callback StringCallback)
+	DeviceAddCpu(dev string, params map[string]interface{}, callback StringCallback)
 
 	XBlockdevChange(parent, node, child string, callback StringCallback)
 	BlockStream(drive string, callback StringCallback)

pkg/hostman/monitor/qmp.go

@@ -553,6 +553,27 @@ func (m *QmpMonitor) DeviceAdd(dev string, params map[string]string, callback St
 	m.Query(cmd, cb)
 }
 
+func (m *QmpMonitor) DeviceAddCpu(dev string, params map[string]interface{}, callback StringCallback) {
+	args := map[string]interface{}{
+		"driver": dev,
+	}
+
+	for k, v := range params {
+		args[k] = v
+	}
+
+	cmd := &Command{
+		Execute: "device_add",
+		Args:    args,
+	}
+
+	cb := func(res *Response) {
+		callback(m.actionResult(res))
+	}
+
+	m.Query(cmd, cb)
+}
+
 func (m *QmpMonitor) MigrateSetDowntime(dtSec float64, callback StringCallback) {
 	m.HumanMonitorCommand(fmt.Sprintf("migrate_set_downtime %f", dtSec), callback)
 }