Handle many vCPUs -- backport

backend/manager/modules/bll/src/test/java/org/ovirt/engine/core/bll/UpdateVmCommandTest.java

@@ -182,6 +182,9 @@ public static Stream<MockConfigDescriptor<?>> mockConfiguration() {
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfCpuPerSocket, version, 16),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfThreadsPerCpu, version, 8),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmCpus, version, createMaxNumberOfVmCpusMap()),
+                MockConfigDescriptor.of(ConfigValues.MaxNumOfCpusCoefficient, 2),
+                MockConfigDescriptor.of(ConfigValues.ManyVmCpus, 128),
+                MockConfigDescriptor.of(ConfigValues.UefiBigVmMemoryGB, 16),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmSockets, version, 16),
                 MockConfigDescriptor.of(ConfigValues.VM32BitMaxMemorySizeInMB, version, 20480),
                 MockConfigDescriptor.of(ConfigValues.VM64BitMaxMemorySizeInMB, version, 4194304),

backend/manager/modules/bll/src/test/java/org/ovirt/engine/core/bll/utils/CompatibilityVersionUpdaterTest.java

@@ -61,6 +61,9 @@ public static Stream<MockConfigDescriptor<?>> mockConfiguration() {
 
                 MockConfigDescriptor.of(ConfigValues.VM64BitMaxMemorySizeInMB, Version.getLast(), MAX_MEM),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmCpus, Version.getLast(), createMaxNumberOfVmCpusMap()),
+                MockConfigDescriptor.of(ConfigValues.MaxNumOfCpusCoefficient, 2),
+                MockConfigDescriptor.of(ConfigValues.ManyVmCpus, 128),
+                MockConfigDescriptor.of(ConfigValues.UefiBigVmMemoryGB, 16),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmSockets, Version.getLast(), 16),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfCpuPerSocket, Version.getLast(), 254),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfThreadsPerCpu, Version.getLast(), 8),

backend/manager/modules/bll/src/test/java/org/ovirt/engine/core/bll/validator/VmValidatorTest.java

@@ -78,6 +78,9 @@ public static Stream<MockConfigDescriptor<?>> mockConfiguration() {
         return Stream.of(
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmCpus, COMPAT_VERSION_FOR_CPU_SOCKET_TEST,
                         createMaxNumberOfVmCpusMap()),
+                MockConfigDescriptor.of(ConfigValues.MaxNumOfCpusCoefficient, 2),
+                MockConfigDescriptor.of(ConfigValues.ManyVmCpus, 128),
+                MockConfigDescriptor.of(ConfigValues.UefiBigVmMemoryGB, 16),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmSockets,
                         COMPAT_VERSION_FOR_CPU_SOCKET_TEST,
                         MAX_NUM_SOCKETS),

backend/manager/modules/bll/src/test/java/org/ovirt/engine/core/utils/ovf/OvfManagerTest.java

@@ -96,7 +96,10 @@ public static Stream<MockConfigDescriptor<?>> mockConfiguration() {
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmSockets, Version.v4_3, 16),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmCpus, Version.v4_3, createMaxNumberOfVmCpusMap()),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmSockets, Version.getLast(), 16),
-                MockConfigDescriptor.of(ConfigValues.MaxNumOfVmCpus, Version.getLast(), createMaxNumberOfVmCpusMap())
+                MockConfigDescriptor.of(ConfigValues.MaxNumOfVmCpus, Version.getLast(), createMaxNumberOfVmCpusMap()),
+                MockConfigDescriptor.of(ConfigValues.MaxNumOfCpusCoefficient, 2),
+                MockConfigDescriptor.of(ConfigValues.ManyVmCpus, 128),
+                MockConfigDescriptor.of(ConfigValues.UefiBigVmMemoryGB, 16)
         );
     }
 

backend/manager/modules/common/src/main/java/org/ovirt/engine/core/common/config/ConfigValues.java

@@ -206,6 +206,10 @@ public enum ConfigValues {
     @TypeConverterAttribute(Integer.class)
     MaxNumOfCpusCoefficient(ClientAccessLevel.User),
     @TypeConverterAttribute(Integer.class)
+    ManyVmCpus(ClientAccessLevel.User),
+    @TypeConverterAttribute(Integer.class)
+    UefiBigVmMemoryGB(ClientAccessLevel.User),
+    @TypeConverterAttribute(Integer.class)
     NumberVmRefreshesBeforeSave,
     @TypeConverterAttribute(Integer.class)
     NumberVdsRefreshesBeforeTryToStartUnknownVms,

backend/manager/modules/common/src/main/java/org/ovirt/engine/core/common/utils/VmCpuCountHelper.java

@@ -98,18 +98,20 @@ public static Integer calcMaxVCpu(VmBase vm, Version compatibilityVersion, Archi
 
         int threadsPerCore = vm.getThreadsPerCpu();
         int cpuPerSocket = vm.getCpuPerSocket();
-        if (compatibilityVersion.greaterOrEquals(Version.v4_6) && maxSockets > LEGACY_MAX_SOCKETS) {
-            // Maximum vCPUs, even when unused, take some memory from the RAM available to
-            // the guest OS.
-            // So it's desirable not to set the maximum number of vCPUs unnecessarily high,
-            // in order to not waste memory. This concerns primarily configurations with a
-            // limit of CPU sockets higher than the one of 16, used for ages and no longer
-            // needed.
-            Integer maxVCpusCoefficient = Config.getValue(ConfigValues.MaxNumOfCpusCoefficient);
+        if (compatibilityVersion.greaterOrEquals(Version.v4_6)) {
+            // Maximum vCPUs, even when unused, take some memory from
+            // the RAM available to the guest OS. Moreover, many vCPUs
+            // can possibly cause other problems (see e.g.
+            // https://bugzilla.redhat.com/2074149). That means it's
+            // better not to set the maximum number of vCPUs
+            // unnecessarily high.
+            final Integer maxVCpusCoefficient = Config.getValue(ConfigValues.MaxNumOfCpusCoefficient);
             final int threadsPerSocket = cpuPerSocket * threadsPerCore;
             final int maxComputedSockets = Math.min(maxSockets,
                     maxVCpusCoefficient * vm.getNumOfCpus() / threadsPerSocket);
-            maxVCpus = Math.min(maxVCpus, Math.max(maxComputedSockets, LEGACY_MAX_SOCKETS) * threadsPerSocket);
+            final Integer manyVmCpus = Config.getValue(ConfigValues.ManyVmCpus);
+            final int maxLegacySockets = Math.min(LEGACY_MAX_SOCKETS, (manyVmCpus - 1) / threadsPerSocket);
+            maxVCpus = Math.min(maxVCpus, Math.max(maxComputedSockets, maxLegacySockets) * threadsPerSocket);
         }
         final BiosType biosType = vm.getBiosType();
         return calcMaxVCpu(architectureFamily, maxSockets, maxVCpus, threadsPerCore, cpuPerSocket, biosType);

backend/manager/modules/utils/src/test/java/org/ovirt/engine/core/utils/ovf/HostedEngineOvfWriterTest.java

@@ -50,7 +50,10 @@ public static Stream<MockConfigDescriptor<?>> mockConfiguration() {
         return Stream.of(
                 MockConfigDescriptor.of(ConfigValues.VdcVersion, Version.getLast().getValue()),
                 MockConfigDescriptor.of(ConfigValues.MaxNumOfVmCpus, Version.getLast(), createMaxNumberOfVmCpusMap()),
-                MockConfigDescriptor.of(ConfigValues.MaxNumOfVmSockets, Version.getLast(), 4)
+                MockConfigDescriptor.of(ConfigValues.MaxNumOfVmSockets, Version.getLast(), 4),
+                MockConfigDescriptor.of(ConfigValues.MaxNumOfCpusCoefficient, 2),
+                MockConfigDescriptor.of(ConfigValues.ManyVmCpus, 128),
+                MockConfigDescriptor.of(ConfigValues.UefiBigVmMemoryGB, 16)
         );
     }
 

backend/manager/modules/vdsbroker/src/main/java/org/ovirt/engine/core/vdsbroker/builder/vminfo/LibvirtVmXmlBuilder.java

@@ -778,7 +778,8 @@ private void writeFeatures() {
         boolean acpiEnabled = vm.getAcpiEnable();
         boolean kaslrEnabled = vmInfoBuildUtils.isKASLRDumpEnabled(vm.getVmOsId());
         boolean secureBootEnabled = vm.getBiosType() == BiosType.Q35_SECURE_BOOT;
-        if (!acpiEnabled && !hypervEnabled && !kaslrEnabled && !secureBootEnabled) {
+        Integer tsegSize = vmInfoBuildUtils.tsegSizeMB(vm, hostDevicesSupplier);
+        if (!acpiEnabled && !hypervEnabled && !kaslrEnabled && !secureBootEnabled && tsegSize == null) {
             return;
         }
 
@@ -858,9 +859,15 @@ private void writeFeatures() {
             writer.writeElement("vmcoreinfo");
         }
 
-        if (secureBootEnabled) {
+        if (secureBootEnabled || tsegSize != null) {
             writer.writeStartElement("smm");
             writer.writeAttributeString("state", "on");
+            if (tsegSize != null) {
+                writer.writeStartElement("tseg");
+                writer.writeAttributeString("unit", "MiB");
+                writer.writeRaw(tsegSize.toString());
+                writer.writeEndElement();
+            }
             writer.writeEndElement();
         }
 

backend/manager/modules/vdsbroker/src/main/java/org/ovirt/engine/core/vdsbroker/builder/vminfo/VmInfoBuildUtils.java

@@ -180,6 +180,10 @@ public class VmInfoBuildUtils {
             Pattern.compile(String.format(BLOCK_DOMAIN_DISK_PATH, ValidationUtils.GUID,
                     ValidationUtils.GUID, ValidationUtils.GUID));
 
+    private static final int DEFAULT_TSEG_SIZE_MB = 16;
+    private static final int MANY_VCPUS_TSEG_SIZE_INCREASE_MB = 24;
+    private static final int PER_TB_TSEG_SIZE_INCREASE_MB = 8;
+
     public static final int NVDIMM_LABEL_SIZE = 128 * 1024;
 
     @Inject
@@ -1677,4 +1681,46 @@ public static boolean isVmWithHighNumberOfX86Vcpus(VM vm) {
         return vm.getClusterArch().getFamily() == ArchitectureType.x86
                 && maxNumberOfVcpus(vm) >= VmCpuCountHelper.HIGH_NUMBER_OF_X86_VCPUS;
     }
+
+    /**
+     * Return a recommended TSEG size in MB or null to use the default value.
+     *
+     * Large VMs with UEFI will not start if they don't have TSEG large enough.
+     * There is no rule what "large" is, it's a VM with "too many" vCPUs or "too large" memory.
+     * There are just some suggestions:
+     * - To use 48 MB for VMs with many vCPUs or large memory:
+     *   https://bugzilla.redhat.com/show_bug.cgi?id=2074149#c22
+     * - The same in a different context:
+     *   https://bugzilla.redhat.com/show_bug.cgi?id=1469338#c30
+     * - To increase the size by unspecified amount for many vCPUs and by 8 MB of each TB of
+     *   address space (which is max RAM + NVDIMM size in our case):
+     *   https://bugzilla.redhat.com/show_bug.cgi?id=1469338#c7
+     * - Confirmation that 48 MB may not be enough for really large VMs:
+     *   https://bugzilla.redhat.com/show_bug.cgi?id=2074149#c28
+     **/
+    public Integer tsegSizeMB(VM vm, MemoizingSupplier<Map<String, HostDevice>> hostDevicesSupplier) {
+        // If UEFI is not used, we should be safe and use the default value.
+        if (vm.getBiosType() == null || !vm.getBiosType().isOvmf()) {
+            return null;
+        }
+        // If the VM is small, we also use the default value, because the TSEG size is taken
+        // from the memory available to the guest.
+        final int cpuLimit = Config.getValue(ConfigValues.ManyVmCpus);
+        final int smallMemory = Config.<Integer> getValue(ConfigValues.UefiBigVmMemoryGB) * 1024;
+        if (maxNumberOfVcpus(vm) < cpuLimit && vm.getMaxMemorySizeMb() < smallMemory) {
+            return null;
+        }
+        // Otherwise, we start with the default value (16 MB currently) and add 24 MB
+        // (which implies at least the recommended value of 48 MB together with the default value
+        // and additional value per memory) for VMs with many vCPUs and 8 MB per each (even partial)
+        // TB of RAM. This is probably a bit wasteful but it's better than risking the VM won't
+        // start.
+        int size = DEFAULT_TSEG_SIZE_MB;
+        if (maxNumberOfVcpus(vm) >= cpuLimit) {
+            size += MANY_VCPUS_TSEG_SIZE_INCREASE_MB;
+        }
+        final long memorySize = vm.getMaxMemorySizeMb() + getNvdimmTotalSize(vm, hostDevicesSupplier) / (1024 * 1024);
+        size += (memorySize / (1024 * 1024) + 1) * PER_TB_TSEG_SIZE_INCREASE_MB;
+        return size;
+    }
 }

packaging/dbscripts/upgrade/pre_upgrade/0000_config.sql

@@ -335,6 +335,8 @@ select fn_db_add_config_value_for_versions_up_to('MaxNumOfVmCpus', '{"x86":710,"
 select fn_db_add_config_value_for_versions_up_to('MaxNumOfVmSockets', '16', '4.5');
 select fn_db_add_config_value_for_versions_up_to('MaxNumOfVmSockets', '10000', '4.7');
 select fn_db_add_config_value('MaxNumOfCpusCoefficient', '2', 'general');
+select fn_db_add_config_value('ManyVmCpus', '128', 'general');
+select fn_db_add_config_value('UefiBigVmMemoryGB', '16', 'general');
 select fn_db_add_config_value('MaxRerunVmOnVdsCount','3','general');
 select fn_db_add_config_value('MaxStorageVdsDelayCheckSec','5','general');
 select fn_db_add_config_value('MaxStorageVdsTimeoutCheckSec','30','general');

packaging/etc/engine-config/engine-config.properties

@@ -73,6 +73,10 @@ MaxNumOfVmSockets.description="Max Number of Socket per Virtual Machine; depreca
 MaxNumOfVmSockets.type=Integer
 MaxNumOfCpusCoefficient.description="The factor between max VM CPUs and configured VM CPUs; effective since compatibility level 4.6"
 MaxNumOfCpusCoefficient.type=Integer
+ManyVmCpus.description="How many vCPUs are considered to be many in maximum vCPU number calculations"
+ManyVmCpus.type=Integer
+UefiBigVmMemoryGB.description="How much UEFI VM memory is considered big enough when changing TSEG size"
+UefiBigVmMemoryGB.type=Integer
 MaxRerunVmOnVdsCount.description="Max Virtual Machine Rerun Attempts on a Host"
 MaxRerunVmOnVdsCount.type=Integer
 MaxSchedulerWeight.description="Max weight score for a single scheduler weight module"