fix bugs

Author: vishesh92

api/src/main/java/org/apache/cloudstack/gpu/GpuService.java

@@ -81,7 +81,15 @@ public interface GpuService extends Manager {
      *
      * @param vmId The ID of the VM to deallocate GPU devices for.
      */
-    void deallocateGpuDevicesForVmOnHost(long vmId);
+    void deallocateAllGpuDevicesForVm(long vmId);
+
+
+    /**
+     * Deallocate GPU devices for a VM on a host.
+     *
+     * @param vmId The ID of the VM to deallocate GPU devices for.
+     */
+    void deallocateGpuDevicesForVmOnHost(long vmId, long hostId);
 
     /**
      * Deallocate existing GPU devices for a VM on a host and allocate new GPU devices to the VM.
@@ -122,7 +130,7 @@ public interface GpuService extends Manager {
      * @param gpuCount    The number of GPU devices to get.
      * @return The GPU devices.
      */
-    GPUDeviceTO getGPUDevice(VirtualMachine vm, VgpuProfile vgpuProfile, int gpuCount);
+    GPUDeviceTO getGPUDevice(VirtualMachine vm, long hostId, VgpuProfile vgpuProfile, int gpuCount);
 
     /**
      * Gets the GPU group details from the GPU devices on a host.

engine/components-api/src/main/java/com/cloud/resource/ResourceManager.java

@@ -208,7 +208,7 @@ public interface ResourceManager extends ResourceService, Configurable {
      * @param gpuCount
      * @return GPUDeviceTO[]
      */
-    GPUDeviceTO getGPUDevice(VirtualMachine vm, VgpuProfileVO vgpuProfile, int gpuCount);
+    GPUDeviceTO getGPUDevice(VirtualMachine vm, long hostId, VgpuProfileVO vgpuProfile, int gpuCount);
 
     /**
      * Get available GPU device

engine/orchestration/src/main/java/com/cloud/vm/VirtualMachineManagerImpl.java

@@ -2356,7 +2356,7 @@ public void destroy(final String vmUuid, final boolean expunge) throws AgentUnav
 
         deleteVMSnapshots(vm, expunge);
 
-        gpuService.deallocateGpuDevicesForVmOnHost(vm.getId());
+        gpuService.deallocateAllGpuDevicesForVm(vm.getId());
 
         Transaction.execute(new TransactionCallbackWithExceptionNoReturn<CloudRuntimeException>() {
             @Override
@@ -2928,6 +2928,8 @@ protected void migrate(final VMInstanceVO vm, final long srcHostId, final Deploy
                 logger.info("Migration was unsuccessful.  Cleaning up: {}", vm);
                 _networkMgr.rollbackNicForMigration(vmSrc, profile);
                 volumeMgr.release(vm.getId(), dstHostId);
+                // deallocate GPU devices for the VM on the destination host
+                gpuService.deallocateGpuDevicesForVmOnHost(vm.getId(), dstHostId);
 
                 _alertMgr.sendAlert(alertType, fromHost.getDataCenterId(), fromHost.getPodId(),
                         "Unable to migrate vm " + vm.getInstanceName() + " from host " + fromHost.getName() + " in zone " + dest.getDataCenter().getName() + " and pod " +
@@ -2946,6 +2948,8 @@ protected void migrate(final VMInstanceVO vm, final long srcHostId, final Deploy
             } else {
                 _networkMgr.commitNicForMigration(vmSrc, profile);
                 volumeMgr.release(vm.getId(), srcHostId);
+                // deallocate GPU devices for the VM on the src host after migration is complete
+                gpuService.deallocateGpuDevicesForVmOnHost(vm.getId(), srcHostId);
                 _networkMgr.setHypervisorHostname(profile, dest, true);
                 recreateCheckpointsKvmOnVmAfterMigration(vm, dstHostId);
 

engine/schema/src/main/java/com/cloud/gpu/dao/GpuDeviceDaoImpl.java

@@ -60,6 +60,7 @@ public GpuDeviceDaoImpl() {
         allFieldSearch = createSearchBuilder();
         allFieldSearch.and(IDS, allFieldSearch.entity().getId(), SearchCriteria.Op.IN);
         allFieldSearch.and(HOST_ID, allFieldSearch.entity().getHostId(), SearchCriteria.Op.EQ);
+        allFieldSearch.and(CARD_ID, allFieldSearch.entity().getCardId(), SearchCriteria.Op.EQ);
         allFieldSearch.and(BUS_ADDRESS, allFieldSearch.entity().getBusAddress(), SearchCriteria.Op.EQ);
         allFieldSearch.and(STATE, allFieldSearch.entity().getState(), SearchCriteria.Op.EQ);
         allFieldSearch.and(VGPU_PROFILE_ID, allFieldSearch.entity().getVgpuProfileId(), SearchCriteria.Op.EQ);

engine/schema/src/main/java/com/cloud/gpu/dao/VgpuProfileDao.java

@@ -34,6 +34,8 @@ public interface VgpuProfileDao extends GenericDao<VgpuProfileVO, Long> {
 
     VgpuProfileVO findByNameAndCardId(String name, long cardId);
 
+    int removeByCardId(long cardId);
+
     Pair<List<VgpuProfileVO>, Integer> searchAndCountVgpuProfiles(Long id, String name, String keyword, Long gpuCardId,
             boolean activeOnly, Long startIndex, Long pageSize);
 }

engine/schema/src/main/java/com/cloud/gpu/dao/VgpuProfileDaoImpl.java

@@ -57,6 +57,13 @@ public VgpuProfileVO findByNameAndCardId(String name, long cardId) {
         return findOneBy(sc);
     }
 
+    @Override
+    public int removeByCardId(long cardId) {
+        SearchCriteria<VgpuProfileVO> sc = allFieldSearch.create();
+        sc.setParameters("cardId", cardId);
+        return remove(sc);
+    }
+
     @Override
     public Pair<List<VgpuProfileVO>, Integer> searchAndCountVgpuProfiles(Long id, String name, String keyword,
             Long gpuCardId, boolean activeOnly, Long startIndex, Long pageSize) {

engine/schema/src/main/resources/META-INF/db/schema-42010to42100.sql

@@ -238,11 +238,11 @@ CREATE TABLE IF NOT EXISTS `cloud`.`gui_themes_details` (
 CREATE TABLE IF NOT EXISTS `cloud`.`gpu_card` (
   `id` bigint unsigned NOT NULL AUTO_INCREMENT COMMENT 'id',
   `uuid` varchar(40) NOT NULL UNIQUE,
-  `device_id` varchar(255) NOT NULL COMMENT 'device id of the GPU card',
+  `device_id` varchar(4) NOT NULL COMMENT 'device id of the GPU card',
   `device_name` varchar(255) NOT NULL COMMENT 'device name of the GPU card',
   `name` varchar(255) NOT NULL COMMENT 'name of the GPU card',
   `vendor_name` varchar(255) NOT NULL COMMENT 'vendor name of the GPU card',
-  `vendor_id` varchar(255) NOT NULL COMMENT 'vendor id of the GPU card',
+  `vendor_id` varchar(4) NOT NULL COMMENT 'vendor id of the GPU card',
   `created` datetime NOT NULL COMMENT 'date created',
   PRIMARY KEY (`id`),
   UNIQUE KEY (`vendor_id`, `device_id`)

plugins/hypervisors/kvm/src/main/java/com/cloud/hypervisor/kvm/resource/wrapper/LibvirtMigrateCommandWrapper.java

@@ -23,6 +23,7 @@
 import java.io.InputStream;
 import java.net.URISyntaxException;
 import java.nio.charset.StandardCharsets;
+import java.util.ArrayList;
 import java.util.HashSet;
 import java.util.List;
 import java.util.Map;
@@ -40,9 +41,13 @@
 import javax.xml.parsers.ParserConfigurationException;
 import javax.xml.transform.TransformerException;
 
+import com.cloud.agent.api.VgpuTypesInfo;
+import com.cloud.agent.api.to.GPUDeviceTO;
+import com.cloud.hypervisor.kvm.resource.LibvirtGpuDef;
 import com.cloud.hypervisor.kvm.resource.LibvirtXMLParser;
 import org.apache.cloudstack.utils.security.ParserUtils;
 import org.apache.commons.collections.MapUtils;
+import org.apache.commons.collections4.CollectionUtils;
 import org.apache.commons.io.FilenameUtils;
 import org.apache.commons.io.IOUtils;
 import org.apache.commons.lang3.StringUtils;
@@ -226,6 +231,8 @@ Use VIR_DOMAIN_XML_SECURE (value = 1) prior to v1.0.0.
 
             xmlDesc = updateVmSharesIfNeeded(command, xmlDesc, libvirtComputingResource);
 
+            xmlDesc = updateGpuDevicesIfNeeded(command, xmlDesc, libvirtComputingResource);
+
             dconn = libvirtUtilitiesHelper.retrieveQemuConnection(destinationUri);
 
             if (to.getType() == VirtualMachine.Type.User) {
@@ -417,6 +424,116 @@ protected Set<String> getMigrateStorageDeviceLabels(List<DiskDef> diskDefinition
         return setOfLabels;
     }
 
+        String updateGpuDevicesIfNeeded(MigrateCommand migrateCommand, String xmlDesc, LibvirtComputingResource libvirtComputingResource)
+            throws ParserConfigurationException, IOException, SAXException, TransformerException {
+        GPUDeviceTO gpuDevice = migrateCommand.getVirtualMachine().getGpuDevice();
+        if (gpuDevice == null || CollectionUtils.isEmpty(gpuDevice.getGpuDevices())) {
+            logger.debug("No GPU device to update for VM [{}].", migrateCommand.getVmName());
+            return xmlDesc;
+        }
+
+        List<VgpuTypesInfo> devices = gpuDevice.getGpuDevices();
+        logger.info("Updating GPU devices for VM [{}] during migration. Number of devices: {}",
+                    migrateCommand.getVmName(), devices.size());
+
+        // Parse XML and find devices element
+        DocumentBuilderFactory docFactory = ParserUtils.getSaferDocumentBuilderFactory();
+        DocumentBuilder docBuilder = docFactory.newDocumentBuilder();
+        Document document;
+        try (InputStream inputStream = IOUtils.toInputStream(xmlDesc, StandardCharsets.UTF_8)) {
+            document = docBuilder.parse(inputStream);
+        }
+
+        NodeList devicesList = document.getElementsByTagName("devices");
+        if (devicesList.getLength() == 0) {
+            logger.warn("No devices section found in XML for VM [{}]", migrateCommand.getVmName());
+            return xmlDesc;
+        }
+
+        Element devicesElement = (Element) devicesList.item(0);
+
+        // Remove existing GPU hostdev elements and add new ones
+        removeExistingGpuHostdevElements(devicesElement);
+        addNewGpuHostdevElements(document, devicesElement, devices);
+
+        String newXmlDesc = LibvirtXMLParser.getXml(document);
+        logger.debug("Updated XML configuration for VM [{}] with new GPU devices", migrateCommand.getVmName());
+
+        return newXmlDesc;
+    }
+
+    /**
+     * Removes existing GPU hostdev elements from the devices section.
+     * GPU devices are identified as hostdev elements with type='pci' or type='mdev'.
+     */
+    private void removeExistingGpuHostdevElements(Element devicesElement) {
+        NodeList hostdevNodes = devicesElement.getElementsByTagName("hostdev");
+        List<Node> nodesToRemove = new ArrayList<>();
+
+        for (int i = 0; i < hostdevNodes.getLength(); i++) {
+            Node hostdevNode = hostdevNodes.item(i);
+            if (hostdevNode.getNodeType() == Node.ELEMENT_NODE) {
+                Element hostdevElement = (Element) hostdevNode;
+                String hostdevType = hostdevElement.getAttribute("type");
+
+                // Remove hostdev elements that represent GPU devices (type='pci' or type='mdev')
+                if ("pci".equals(hostdevType) || "mdev".equals(hostdevType)) {
+                    // Additional check: ensure this is actually a GPU device by checking mode='subsystem'
+                    String mode = hostdevElement.getAttribute("mode");
+                    if ("subsystem".equals(mode)) {
+                        nodesToRemove.add(hostdevNode);
+                    }
+                }
+            }
+        }
+
+        // Remove the nodes
+        for (Node node : nodesToRemove) {
+            devicesElement.removeChild(node);
+        }
+
+        logger.debug("Removed {} existing GPU hostdev elements", nodesToRemove.size());
+    }
+
+    /**
+     * Adds new GPU hostdev elements to the devices section based on the GPU devices
+     * allocated on the destination host.
+     */
+    private void addNewGpuHostdevElements(Document document, Element devicesElement, List<VgpuTypesInfo> devices)
+            throws ParserConfigurationException, IOException, SAXException {
+        if (devices.isEmpty()) {
+            return;
+        }
+
+        // Reuse parser for efficiency
+        DocumentBuilderFactory factory = ParserUtils.getSaferDocumentBuilderFactory();
+        DocumentBuilder builder = factory.newDocumentBuilder();
+
+        for (VgpuTypesInfo deviceInfo : devices) {
+            Element hostdevElement = createGpuHostdevElement(document, deviceInfo, builder);
+            devicesElement.appendChild(hostdevElement);
+            logger.debug("Added new GPU hostdev element for device: {} (type: {}, busAddress: {})",
+                         deviceInfo.getDeviceName(), deviceInfo.getDeviceType(), deviceInfo.getBusAddress());
+        }
+    }
+
+    /**
+     * Creates a hostdev element for a GPU device using LibvirtGpuDef.
+     */
+    private Element createGpuHostdevElement(Document document, VgpuTypesInfo deviceInfo, DocumentBuilder builder)
+            throws IOException, SAXException {
+        // Generate GPU XML using LibvirtGpuDef
+        LibvirtGpuDef gpuDef = new LibvirtGpuDef();
+        gpuDef.defGpu(deviceInfo);
+        String gpuXml = gpuDef.toString();
+
+        // Parse and import into target document
+        try (InputStream xmlStream = IOUtils.toInputStream(gpuXml, StandardCharsets.UTF_8)) {
+            Document gpuDocument = builder.parse(xmlStream);
+            Element hostdevElement = gpuDocument.getDocumentElement();
+            return (Element) document.importNode(hostdevElement, true);
+        }
+    }
 
     /**
      * Checks if the CPU shares are equal in the source host and destination host.