Merge pull request #11185 from fidencio/topic/reclaim-guest-freed-memory-backport-from-runtime-rs

runtime: clh: Add reclaim_guest_freed_memory [BACKPORT]

src/runtime/config/configuration-clh.toml.in

@@ -185,6 +185,16 @@ block_device_driver = "virtio-blk"
 # Default false
 #block_device_cache_direct = true
 
+# Reclaim guest freed memory.
+# Enabling this will result in the VM balloon device having f_reporting=on set.
+# Then the hypervisor will use it to reclaim guest freed memory.
+# This is useful for reducing the amount of memory used by a VM.
+# Enabling this feature may sometimes reduce the speed of memory access in
+# the VM.
+#
+# Default false
+#reclaim_guest_freed_memory = true
+
 # Enable huge pages for VM RAM, default false
 # Enabling this will result in the VM memory
 # being allocated using huge pages.

src/runtime/pkg/katautils/config-settings.go.in

@@ -76,6 +76,7 @@ const defaultBlockDeviceCacheDirect bool = false
 const defaultBlockDeviceCacheNoflush bool = false
 const defaultEnableIOThreads bool = false
 const defaultEnableMemPrealloc bool = false
+const defaultEnableReclaimGuestFreedMemory bool = false
 const defaultEnableHugePages bool = false
 const defaultEnableIOMMU bool = false
 const defaultEnableIOMMUPlatform bool = false

src/runtime/pkg/katautils/config.go

@@ -147,6 +147,7 @@ type hypervisor struct {
 	VhostUserDeviceReconnect       uint32                    `toml:"vhost_user_reconnect_timeout_sec"`
 	DisableBlockDeviceUse          bool                      `toml:"disable_block_device_use"`
 	MemPrealloc                    bool                      `toml:"enable_mem_prealloc"`
+	ReclaimGuestFreedMemory        bool                      `toml:"reclaim_guest_freed_memory"`
 	HugePages                      bool                      `toml:"enable_hugepages"`
 	VirtioMem                      bool                      `toml:"enable_virtio_mem"`
 	IOMMU                          bool                      `toml:"enable_iommu"`
@@ -1082,6 +1083,7 @@ func newClhHypervisorConfig(h hypervisor) (vc.HypervisorConfig, error) {
 		VirtioFSCacheSize:              h.VirtioFSCacheSize,
 		VirtioFSCache:                  h.VirtioFSCache,
 		MemPrealloc:                    h.MemPrealloc,
+		ReclaimGuestFreedMemory:        h.ReclaimGuestFreedMemory,
 		HugePages:                      h.HugePages,
 		FileBackedMemRootDir:           h.FileBackedMemRootDir,
 		FileBackedMemRootList:          h.FileBackedMemRootList,
@@ -1434,6 +1436,7 @@ func GetDefaultHypervisorConfig() vc.HypervisorConfig {
 		DisableBlockDeviceUse:    defaultDisableBlockDeviceUse,
 		DefaultBridges:           defaultBridgesCount,
 		MemPrealloc:              defaultEnableMemPrealloc,
+		ReclaimGuestFreedMemory:  defaultEnableReclaimGuestFreedMemory,
 		HugePages:                defaultEnableHugePages,
 		IOMMU:                    defaultEnableIOMMU,
 		IOMMUPlatform:            defaultEnableIOMMUPlatform,

src/runtime/pkg/oci/utils.go

@@ -705,6 +705,12 @@ func addHypervisorMemoryOverrides(ocispec specs.Spec, sbConfig *vc.SandboxConfig
 		sbConfig.HypervisorConfig.FileBackedMemRootDir = value
 	}
 
+	if err := newAnnotationConfiguration(ocispec, vcAnnotations.ReclaimGuestFreedMemory).setBool(func(reclaimGuestFreedMemory bool) {
+		sbConfig.HypervisorConfig.ReclaimGuestFreedMemory = reclaimGuestFreedMemory
+	}); err != nil {
+		return err
+	}
+
 	if err := newAnnotationConfiguration(ocispec, vcAnnotations.HugePages).setBool(func(hugePages bool) {
 		sbConfig.HypervisorConfig.HugePages = hugePages
 	}); err != nil {

src/runtime/virtcontainers/clh.go

@@ -452,6 +452,37 @@ func (clh *cloudHypervisor) enableProtection() error {
 	}
 }
 
+func getNonUserDefinedKernelParams(rootfstype string, disableNvdimm bool, dax bool, debug bool, confidential bool, iommu bool) ([]Param, error) {
+	params, err := GetKernelRootParams(rootfstype, disableNvdimm, dax)
+	if err != nil {
+		return []Param{}, err
+	}
+	params = append(params, clhKernelParams...)
+
+	if iommu {
+		params = append(params, Param{"iommu", "pt"})
+	}
+
+	if !debug {
+		// start the guest kernel with 'quiet' in non-debug mode
+		params = append(params, Param{"quiet", ""})
+		return params, nil
+	}
+
+	// In case of debug ...
+
+	// Followed by extra debug parameters if debug enabled in configuration file
+	if confidential {
+		params = append(params, clhDebugConfidentialGuestKernelParams...)
+	} else if runtime.GOARCH == "arm64" {
+		params = append(params, clhArmDebugKernelParams...)
+	} else {
+		params = append(params, clhDebugKernelParams...)
+	}
+	params = append(params, clhDebugKernelParamsCommon...)
+	return params, nil
+}
+
 // For cloudHypervisor this call only sets the internal structure up.
 // The VM will be created and started through StartVM().
 func (clh *cloudHypervisor) CreateVM(ctx context.Context, id string, network Network, hypervisorConfig *HypervisorConfig) error {
@@ -527,34 +558,22 @@ func (clh *cloudHypervisor) CreateVM(ctx context.Context, id string, network Net
 		hotplugSize := clh.config.DefaultMaxMemorySize
 		// OpenAPI only supports int64 values
 		clh.vmconfig.Memory.HotplugSize = func(i int64) *int64 { return &i }(int64((utils.MemUnit(hotplugSize) * utils.MiB).ToBytes()))
+
+		if clh.config.ReclaimGuestFreedMemory {
+			// Create VM with a balloon config so we can enable free page reporting (size of the balloon can be set to zero)
+			clh.vmconfig.Balloon = chclient.NewBalloonConfig(0)
+			// Set the free page reporting flag for ballooning to be true
+			clh.vmconfig.Balloon.SetFreePageReporting(true)
 		}
+	}
+
 	// Set initial amount of cpu's for the virtual machine
 	clh.vmconfig.Cpus = chclient.NewCpusConfig(int32(clh.config.NumVCPUs()), int32(clh.config.DefaultMaxVCPUs))
 
-	params, err := GetKernelRootParams(hypervisorConfig.RootfsType, clh.config.ConfidentialGuest, !clh.config.ConfidentialGuest)
+	params, err := getNonUserDefinedKernelParams(hypervisorConfig.RootfsType, clh.config.ConfidentialGuest, !clh.config.ConfidentialGuest, clh.config.Debug, clh.config.ConfidentialGuest, clh.config.IOMMU)
 	if err != nil {
 		return err
 	}
-	params = append(params, clhKernelParams...)
-
-	// Followed by extra debug parameters if debug enabled in configuration file
-	if clh.config.Debug {
-		if clh.config.ConfidentialGuest {
-			params = append(params, clhDebugConfidentialGuestKernelParams...)
-		} else if runtime.GOARCH == "arm64" {
-			params = append(params, clhArmDebugKernelParams...)
-		} else {
-			params = append(params, clhDebugKernelParams...)
-		}
-		params = append(params, clhDebugKernelParamsCommon...)
-	} else {
-		// start the guest kernel with 'quiet' in non-debug mode
-		params = append(params, Param{"quiet", ""})
-	}
-	if clh.config.IOMMU {
-		params = append(params, Param{"iommu", "pt"})
-	}
-
 	// Followed by extra kernel parameters defined in the configuration file
 	params = append(params, clh.config.KernelParams...)
 

src/runtime/virtcontainers/hypervisor.go

@@ -612,6 +612,9 @@ type HypervisorConfig struct {
 	// MemPrealloc specifies if the memory should be pre-allocated
 	MemPrealloc bool
 
+	// ReclaimGuestFreedMemory is a sandbox annotation that specifies whether the memory freed by the guest will be reclaimed by the hypervisor or not.
+	ReclaimGuestFreedMemory bool
+
 	// HugePages specifies if the memory should be pre-allocated from huge pages
 	HugePages bool
 

src/runtime/virtcontainers/pkg/annotations/annotations.go

@@ -167,6 +167,9 @@ const (
 	// MemPrealloc is a sandbox annotation that specifies the memory space used for nvdimm device by the hypervisor.
 	MemPrealloc = kataAnnotHypervisorPrefix + "enable_mem_prealloc"
 
+	// ReclaimGuestFreedMemory is a sandbox annotation that specifies whether the memory freed by the guest will be reclaimed by the hypervisor or not.
+	ReclaimGuestFreedMemory = kataAnnotHypervisorPrefix + "reclaim_guest_freed_memory"
+
 	// HugePages is a sandbox annotation to specify if the memory should be pre-allocated from huge pages
 	HugePages = kataAnnotHypervisorPrefix + "enable_hugepages"