Merge e29e65c103 into 9379a18c8a

src/runtime/Makefile

@@ -260,6 +260,7 @@ DEFVIRTIOFSQUEUESIZE ?= 1024
 # Make sure you quote args.
 DEFVIRTIOFSEXTRAARGS ?= [\"--thread-pool-size=1\", \"--announce-submounts\"]
 DEFENABLEIOTHREADS := false
+DEFINDEPIOTHREADS := 0
 DEFENABLEVHOSTUSERSTORE := false
 DEFVHOSTUSERSTOREPATH := $(PKGRUNDIR)/vhost-user
 DEFVALIDVHOSTUSERSTOREPATHS := [\"$(DEFVHOSTUSERSTOREPATH)\"]
@@ -731,6 +732,7 @@ USER_VARS += DEFVIRTIOFSEXTRAARGS
 USER_VARS += DEFENABLEANNOTATIONS
 USER_VARS += DEFENABLEANNOTATIONSTEE
 USER_VARS += DEFENABLEIOTHREADS
+USER_VARS += DEFINDEPIOTHREADS
 USER_VARS += DEFSECCOMPSANDBOXPARAM
 USER_VARS += DEFENABLEVHOSTUSERSTORE
 USER_VARS += DEFVHOSTUSERSTOREPATH

src/runtime/config/configuration-qemu-coco-dev.toml.in

@@ -259,11 +259,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

src/runtime/config/configuration-qemu-nvidia-gpu-snp.toml.in

@@ -268,11 +268,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

src/runtime/config/configuration-qemu-nvidia-gpu-tdx.toml.in

@@ -252,11 +252,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

src/runtime/config/configuration-qemu-nvidia-gpu.toml.in

@@ -257,11 +257,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

src/runtime/config/configuration-qemu-se.toml.in

@@ -243,11 +243,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

src/runtime/config/configuration-qemu-snp.toml.in

@@ -268,11 +268,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

src/runtime/config/configuration-qemu-tdx.toml.in

@@ -253,11 +253,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

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

@@ -258,11 +258,15 @@ block_device_aio = "@DEFBLOCKDEVICEAIO_QEMU@"
 #block_device_cache_noflush = true
 
 # Enable iothreads (data-plane) to be used. This causes IO to be
-# handled in a separate IO thread. This is currently only implemented
+# handled in a separate IO thread. This is currently implemented
-# for SCSI.
+# for virtio-scsi and virtio-blk.
 #
 enable_iothreads = @DEFENABLEIOTHREADS@
 
+# Independent IOThreads enables IO to be processed in a separate thread, it is
+# for QEMU hotplug device attach to iothread, like virtio-blk.
+indep_iothreads = @DEFINDEPIOTHREADS@
+
 # Enable pre allocation of VM RAM, default false
 # Enabling this will result in lower container density
 # as all of the memory will be allocated and locked

src/runtime/pkg/govmm/qemu/qmp.go

@@ -1102,7 +1102,7 @@ func (q *QMP) ExecuteDeviceDel(ctx context.Context, devID string) error {
 // disableModern indicates if virtio version 1.0 should be replaced by the
 // former version 0.9, as there is a KVM bug that occurs when using virtio
 // 1.0 in nested environments.
-func (q *QMP) ExecutePCIDeviceAdd(ctx context.Context, blockdevID, devID, driver, addr, bus, romfile string, queues int, shared, disableModern bool) error {
+func (q *QMP) ExecutePCIDeviceAdd(ctx context.Context, blockdevID, devID, driver, addr, bus, romfile string, queues int, shared, disableModern bool, iothreadID string) error {
 	args := map[string]interface{}{
 		"id":     devID,
 		"driver": driver,
@@ -1129,6 +1129,10 @@ func (q *QMP) ExecutePCIDeviceAdd(ctx context.Context, blockdevID, devID, driver
 		}
 	}
 
+	if iothreadID != "" {
+		args["iothread"] = iothreadID
+	}
+
 	return q.executeCommand(ctx, "device_add", args, nil)
 }
 

src/runtime/pkg/govmm/qemu/qmp_test.go

@@ -1066,7 +1066,7 @@ func TestQMPPCIDeviceAdd(t *testing.T) {
 	blockdevID := fmt.Sprintf("drive_%s", volumeUUID)
 	devID := fmt.Sprintf("device_%s", volumeUUID)
 	err := q.ExecutePCIDeviceAdd(context.Background(), blockdevID, devID,
-		"virtio-blk-pci", "0x1", "", "", 1, true, false)
+		"virtio-blk-pci", "0x1", "", "", 1, true, false, "")
 	if err != nil {
 		t.Fatalf("Unexpected error %v", err)
 	}

src/runtime/pkg/katatestutils/utils.go

@@ -233,6 +233,7 @@ type RuntimeConfigOptions struct {
 	DefaultMemSize        uint32
 	DefaultMaxMemorySize  uint64
 	DefaultMsize9p        uint32
+	DefaultIndepIOThreads uint32
 	DisableBlock          bool
 	EnableIOThreads       bool
 	DisableNewNetNs       bool
@@ -318,6 +319,7 @@ func MakeRuntimeConfigFileData(config RuntimeConfigOptions) string {
 	default_memory = ` + strconv.FormatUint(uint64(config.DefaultMemSize), 10) + `
 	disable_block_device_use =  ` + strconv.FormatBool(config.DisableBlock) + `
 	enable_iothreads =  ` + strconv.FormatBool(config.EnableIOThreads) + `
+	indep_iothreads = ` + strconv.FormatUint(uint64(config.DefaultIndepIOThreads), 10) + `
 	cold_plug_vfio =  "` + config.ColdPlugVFIO.String() + `"
 	hot_plug_vfio =   "` + config.HotPlugVFIO.String() + `"
 	pcie_root_port = ` + strconv.FormatUint(uint64(config.PCIeRootPort), 10) + `

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

@@ -75,6 +75,7 @@ const defaultBlockDeviceCacheSet bool = false
 const defaultBlockDeviceCacheDirect bool = false
 const defaultBlockDeviceCacheNoflush bool = false
 const defaultEnableIOThreads bool = false
+const defaultIndepIOThreads uint32 = 0
 const defaultEnableMemPrealloc bool = false
 const defaultEnableReclaimGuestFreedMemory bool = false
 const defaultEnableHugePages bool = false

src/runtime/pkg/katautils/config.go

@@ -155,6 +155,7 @@ type hypervisor struct {
 	Debug                          bool                      `toml:"enable_debug"`
 	DisableNestingChecks           bool                      `toml:"disable_nesting_checks"`
 	EnableIOThreads                bool                      `toml:"enable_iothreads"`
+	IndepIOThreads                 uint32                    `toml:"indep_iothreads"`
 	DisableImageNvdimm             bool                      `toml:"disable_image_nvdimm"`
 	HotPlugVFIO                    config.PCIePort           `toml:"hot_plug_vfio"`
 	ColdPlugVFIO                   config.PCIePort           `toml:"cold_plug_vfio"`
@@ -614,6 +615,14 @@ func (h hypervisor) msize9p() uint32 {
 	return h.Msize9p
 }
 
+func (h hypervisor) indepiothreads() uint32 {
+	if h.IndepIOThreads == 0 {
+		return defaultIndepIOThreads
+	}
+
+	return h.IndepIOThreads
+}
+
 func (h hypervisor) guestHookPath() string {
 	if h.GuestHookPath == "" {
 		return defaultGuestHookPath
@@ -810,6 +819,7 @@ func newFirecrackerHypervisorConfig(h hypervisor) (vc.HypervisorConfig, error) {
 		DisableNestingChecks:  h.DisableNestingChecks,
 		BlockDeviceDriver:     blockDriver,
 		EnableIOThreads:       h.EnableIOThreads,
+		IndepIOThreads:        h.indepiothreads(),
 		DisableVhostNet:       true, // vhost-net backend is not supported in Firecracker
 		GuestHookPath:         h.guestHookPath(),
 		RxRateLimiterMaxRate:  rxRateLimiterMaxRate,
@@ -964,6 +974,7 @@ func newQemuHypervisorConfig(h hypervisor) (vc.HypervisorConfig, error) {
 		BlockDeviceCacheDirect:   h.BlockDeviceCacheDirect,
 		BlockDeviceCacheNoflush:  h.BlockDeviceCacheNoflush,
 		EnableIOThreads:          h.EnableIOThreads,
+		IndepIOThreads:           h.indepiothreads(),
 		Msize9p:                  h.msize9p(),
 		DisableImageNvdimm:       h.DisableImageNvdimm,
 		HotPlugVFIO:              h.hotPlugVFIO(),
@@ -1094,6 +1105,7 @@ func newClhHypervisorConfig(h hypervisor) (vc.HypervisorConfig, error) {
 		BlockDeviceCacheSet:            h.BlockDeviceCacheSet,
 		BlockDeviceCacheDirect:         h.BlockDeviceCacheDirect,
 		EnableIOThreads:                h.EnableIOThreads,
+		IndepIOThreads:                 h.indepiothreads(),
 		Msize9p:                        h.msize9p(),
 		DisableImageNvdimm:             h.DisableImageNvdimm,
 		ColdPlugVFIO:                   h.coldPlugVFIO(),
@@ -1452,6 +1464,7 @@ func GetDefaultHypervisorConfig() vc.HypervisorConfig {
 		BlockDeviceCacheDirect:   defaultBlockDeviceCacheDirect,
 		BlockDeviceCacheNoflush:  defaultBlockDeviceCacheNoflush,
 		EnableIOThreads:          defaultEnableIOThreads,
+		IndepIOThreads:           defaultIndepIOThreads,
 		Msize9p:                  defaultMsize9p,
 		ColdPlugVFIO:             defaultColdPlugVFIO,
 		HotPlugVFIO:              defaultHotPlugVFIO,

src/runtime/pkg/oci/utils.go

@@ -840,6 +840,16 @@ func addHypervisorBlockOverrides(ocispec specs.Spec, sbConfig *vc.SandboxConfig)
 		return err
 	}
 
+	if err := newAnnotationConfiguration(ocispec, vcAnnotations.IndepIOThreads).setUintWithCheck(func(indepiothreads uint64) error {
+		if indepiothreads < 0 {
+			return fmt.Errorf("Error parsing annotation for indepiothreads, please specify positive numeric value")
+		}
+		sbConfig.HypervisorConfig.IndepIOThreads = uint32(indepiothreads)
+		return nil
+		}); err != nil {
+		return err
+	}
+
 	if err := newAnnotationConfiguration(ocispec, vcAnnotations.BlockDeviceCacheSet).setBool(func(blockDeviceCacheSet bool) {
 		sbConfig.HypervisorConfig.BlockDeviceCacheSet = blockDeviceCacheSet
 	}); err != nil {

src/runtime/virtcontainers/documentation/api/1.0/api.md

@@ -246,9 +246,13 @@ type HypervisorConfig struct {
 	DisableBlockDeviceUse bool
 
 	// EnableIOThreads enables IO to be processed in a separate thread.
-	// Supported currently for virtio-scsi driver.
+	// Supported currently for virtio-scsi and virtio-blk(depend on IndepIOThreads) driver.
 	EnableIOThreads bool
 
+	// Independent IOThreads enables IO to be processed in a separate thread, it is
+	// for QEMU hotplug device attach to iothread, like virtio-blk.
+	IndepIOThreads uint32
+
 	// Debug changes the default hypervisor and kernel parameters to
 	// enable debug output where available.
 	Debug bool

src/runtime/virtcontainers/hypervisor.go

@@ -602,9 +602,12 @@ type HypervisorConfig struct {
 	DisableBlockDeviceUse bool
 
 	// EnableIOThreads enables IO to be processed in a separate thread.
-	// Supported currently for virtio-scsi driver.
+	// Supported currently for virtio-scsi driver and virtio-blk(depend on IndepIOThreads) driver.
 	EnableIOThreads bool
 
+	// Independent IOThreads enables IO to be processed in a separate thread.
+	IndepIOThreads uint32
+
 	// Debug changes the default hypervisor and kernel parameters to
 	// enable debug output where available.
 	Debug bool

src/runtime/virtcontainers/persist.go

@@ -235,6 +235,7 @@ func (s *Sandbox) dumpConfig(ss *persistapi.SandboxState) {
 		BlockDeviceCacheNoflush: sconfig.HypervisorConfig.BlockDeviceCacheNoflush,
 		DisableBlockDeviceUse:   sconfig.HypervisorConfig.DisableBlockDeviceUse,
 		EnableIOThreads:         sconfig.HypervisorConfig.EnableIOThreads,
+		IndepIOThreads:          sconfig.HypervisorConfig.IndepIOThreads,
 		Debug:                   sconfig.HypervisorConfig.Debug,
 		MemPrealloc:             sconfig.HypervisorConfig.MemPrealloc,
 		HugePages:               sconfig.HypervisorConfig.HugePages,
@@ -473,6 +474,7 @@ func loadSandboxConfig(id string) (*SandboxConfig, error) {
 		BlockDeviceCacheNoflush: hconf.BlockDeviceCacheNoflush,
 		DisableBlockDeviceUse:   hconf.DisableBlockDeviceUse,
 		EnableIOThreads:         hconf.EnableIOThreads,
+		IndepIOThreads:          hconf.IndepIOThreads,
 		Debug:                   hconf.Debug,
 		MemPrealloc:             hconf.MemPrealloc,
 		HugePages:               hconf.HugePages,

src/runtime/virtcontainers/persist/api/config.go

@@ -161,9 +161,13 @@ type HypervisorConfig struct {
 	DisableBlockDeviceUse bool
 
 	// EnableIOThreads enables IO to be processed in a separate thread.
-	// Supported currently for virtio-scsi driver.
+	// Supported currently for virtio-scsi and virtio-blk(depend on IndepIOThreads) driver.
 	EnableIOThreads bool
 
+	// Independent IOThreads enables IO to be processed in a separate thread, it is
+	// for QEMU hotplug device attach to iothread, like virtio-blk.
+	IndepIOThreads uint32
+
 	// Debug changes the default hypervisor and kernel parameters to
 	// enable debug output where available.
 	Debug bool

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

@@ -218,9 +218,14 @@ const (
 	DisableBlockDeviceUse = kataAnnotHypervisorPrefix + "disable_block_device_use"
 
 	// EnableIOThreads is a sandbox annotation to enable IO to be processed in a separate thread.
-	// Supported currently for virtio-scsi driver.
+	// Supported currently for virtio-scsi and virtio-blk(depend on IndepIOThreads) driver.
 	EnableIOThreads = kataAnnotHypervisorPrefix + "enable_iothreads"
 
+	// Independent IOThreads enables IO to be processed in a separate thread, it is
+	// for QEMU hotplug device attach to iothread, like virtio-blk.
+	IndepIOThreads = kataAnnotHypervisorPrefix + "indep_iothreads"
+
+
 	// BlockDeviceCacheSet is a sandbox annotation that specifies cache-related options will be set to block devices or not.
 	BlockDeviceCacheSet = kataAnnotHypervisorPrefix + "block_device_cache_set"
 

src/runtime/virtcontainers/qemu.go

@@ -143,6 +143,8 @@ const (
 	qemuStopSandboxTimeoutSecs = 15
 
 	qomPathPrefix = "/machine/peripheral/"
+
+	indepIOThreadsPrefix = "indep_iothread"
 )
 
 // agnostic list of kernel parameters
@@ -513,6 +515,26 @@ func (q *qemu) setupFileBackedMem(knobs *govmmQemu.Knobs, memory *govmmQemu.Memo
 	memory.Path = target
 }
 
+func (q *qemu) setupIoThread(ioThread *govmmQemu.IOThread) []govmmQemu.IOThread {
+
+	var tmp_threads []govmmQemu.IOThread
+
+	// Add virtio-scsi IOThreads for QEMU
+	if ioThread != nil {
+		tmp_threads = append(tmp_threads, *ioThread)
+	}
+
+	// Add Independent IOThreads for QEMU
+	if q.config.IndepIOThreads > 0 {
+		for i := uint32(0); i < q.config.IndepIOThreads; i++ {
+			id := fmt.Sprintf("%s_%d", indepIOThreadsPrefix, i)
+			tmp_threads = append(tmp_threads, govmmQemu.IOThread{ID: id})
+		}
+	}
+
+	return tmp_threads
+}
+
 func (q *qemu) setConfig(config *HypervisorConfig) error {
 	q.config = *config
 
@@ -734,6 +756,7 @@ func (q *qemu) CreateVM(ctx context.Context, id string, network Network, hypervi
 		return err
 	}
 
+	// Note: Only virtio-SCSI device driver use this ioThread args.
 	devices, ioThread, kernel, err := q.buildDevices(ctx, kernelPath)
 	if err != nil {
 		return err
@@ -802,9 +825,9 @@ func (q *qemu) CreateVM(ctx context.Context, id string, network Network, hypervi
 		return err
 	}
 
-	if ioThread != nil {
+	// Setup iothread for devices.
-		qemuConfig.IOThreads = []govmmQemu.IOThread{*ioThread}
+	qemuConfig.IOThreads = q.setupIoThread(ioThread)
-	}
+
 	// Add RNG device to hypervisor
 	// Skip for s390x (as CPACF is used) or when Confidential Guest is enabled
 	if machine.Type != QemuCCWVirtio && !q.config.ConfidentialGuest {
@@ -1721,7 +1744,13 @@ func (q *qemu) hotplugAddBlockDevice(ctx context.Context, drive *config.BlockDri
 
 		queues := int(q.config.NumVCPUs())
 
-		if err = q.qmpMonitorCh.qmp.ExecutePCIDeviceAdd(q.qmpMonitorCh.ctx, drive.ID, devID, driver, addr, bridge.ID, romFile, queues, true, defaultDisableModern); err != nil {
+		// Make Independent IOThread 0 as the virtio-blk default.
+		var iothreadID string
+		if q.config.EnableIOThreads && q.config.IndepIOThreads > 0 {
+			iothreadID = fmt.Sprintf("%s_%d", indepIOThreadsPrefix, 0)
+		}
+
+		if err = q.qmpMonitorCh.qmp.ExecutePCIDeviceAdd(q.qmpMonitorCh.ctx, drive.ID, devID, driver, addr, bridge.ID, romFile, queues, true, defaultDisableModern, iothreadID); err != nil {
 			return err
 		}
 	case q.config.BlockDeviceDriver == config.VirtioBlockCCW:

src/runtime/virtcontainers/stratovirt.go

@@ -900,7 +900,7 @@ func (s *stratovirt) hotplugBlk(ctx context.Context, drive *config.BlockDrive, o
 		}
 
 		devAddr := fmt.Sprintf("%d", slot)
-		if err := s.qmpMonitorCh.qmp.ExecutePCIDeviceAdd(s.qmpMonitorCh.ctx, drive.ID, drive.ID, driver, devAddr, "", "", 0, false, false); err != nil {
+		if err := s.qmpMonitorCh.qmp.ExecutePCIDeviceAdd(s.qmpMonitorCh.ctx, drive.ID, drive.ID, driver, devAddr, "", "", 0, false, false, ""); err != nil {
 			return err
 		}
 	case RemoveDevice: