feat: add device filtering for GPU selection

Signed-off-by: wangshuai <1090646861@qq.com>

Author: wangshuai

doc/config.md

@@ -56,3 +56,8 @@ String type, `hami-core` for using hami-core for container resource limitation,
 Integer type, device memory oversubscription on that node
 * `devicecorescaling`: 
 Integer type, device core oversubscription on that node 
+* `devicesplitcount`: Allowed number of tasks sharing a device.
+* `filterdevices`: Devices that are not registered to HAMi.
+  * `uuid`: UUIDs of devices to ignore
+  * `index`: Indexes of devices to ignore.
+  * A device is ignored by HAMi if it's in `uuid` or `index` list.

pkg/plugin/vgpu/config/config.go

@@ -146,3 +146,48 @@ type DevicePluginConfigs struct {
 		FilterDevice        *FilterDevice `json:"filterdevices"`
 	} `json:"nodeconfig"`
 }
+
+var (
+	filterOnce sync.Once
+	uuidMap    map[string]struct{}
+	indexMap   map[uint]struct{}
+)
+
+func FilterDeviceToRegister(uuid string, index int) bool {
+	filterOnce.Do(initFilter)
+	if len(uuidMap) == 0 && len(indexMap) == 0 {
+		return false
+	}
+
+	if _, ok := uuidMap[uuid]; ok {
+		return true
+	}
+
+	if _, ok := indexMap[uint(index)]; ok {
+		return true
+	}
+
+	return false
+}
+
+func initFilter() {
+	uuidMap = make(map[string]struct{})
+	indexMap = make(map[uint]struct{})
+	if DevicePluginFilterDevice == nil {
+		return
+	}
+
+	if len(DevicePluginFilterDevice.UUID) > 0 {
+		uuidMap = make(map[string]struct{}, len(DevicePluginFilterDevice.UUID))
+		for _, u := range DevicePluginFilterDevice.UUID {
+			uuidMap[u] = struct{}{}
+		}
+	}
+
+	if len(DevicePluginFilterDevice.Index) > 0 {
+		indexMap = make(map[uint]struct{}, len(DevicePluginFilterDevice.Index))
+		for _, idx := range DevicePluginFilterDevice.Index {
+			indexMap[idx] = struct{}{}
+		}
+	}
+}

pkg/plugin/vgpu/nvidia.go

@@ -96,6 +96,14 @@ func (g *GpuDeviceManager) Devices() []*Device {
 		d, ret := config.Nvml().DeviceGetHandleByIndex(i)
 		check(ret)
 
+		uuid, ret := d.GetUUID()
+		check(ret)
+		// Filter GPU device
+		if config.FilterDeviceToRegister(uuid, i) {
+			klog.V(5).Infof("Filtering out GPU device index=%d, uuid=%s", i, uuid)
+			continue
+		}
+
 		migMode, _, ret := d.GetMigMode()
 		if ret != nvml.SUCCESS {
 			if ret == nvml.ERROR_NOT_SUPPORTED {