qemu: tdx: Adapt to the TDX 1.5 stack

QEMU for TDX 1.5 makes use of private memory map/unmap.
Make changes to govmm to support this. Support for private backing fd
for memory is added as knob to the qemu config.

Userspace's map/unmap operations are done by fallocate() ioctl on the
backing store fd.
Reference:
https://lore.kernel.org/linux-mm/20220519153713.819591-1-chao.p.peng@linux.intel.com/

Fixes: #7770

Signed-off-by: Archana Shinde <archana.m.shinde@intel.com>
Signed-off-by: Fabiano Fidêncio <fabiano.fidencio@intel.com>

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

@@ -141,9 +141,16 @@ const (
 func isDimmSupported(config *Config) bool {
 	switch runtime.GOARCH {
 	case "amd64", "386", "ppc64le", "arm64":
-		if config != nil && config.Machine.Type == MachineTypeMicrovm {
+		if config != nil {
-			// microvm does not support NUMA
+			if config.Machine.Type == MachineTypeMicrovm {
-			return false
+				// microvm does not support NUMA
+				return false
+			}
+			if config.Knobs.MemFDPrivate {
+				// TDX guests rely on MemFD Private, which
+				// does not have NUMA support yet
+				return false
+			}
 		}
 		return true
 	default:
@@ -2628,6 +2635,9 @@ type Knobs struct {
 	// MemPrealloc will allocate all the RAM upfront
 	MemPrealloc bool
 
+	// Private Memory FD meant for private memory map/unmap.
+	MemFDPrivate bool
+
 	// FileBackedMem requires Memory.Size and Memory.Path of the VM to
 	// be set.
 	FileBackedMem bool
@@ -2992,10 +3002,13 @@ func (config *Config) appendMemoryKnobs() {
 		return
 	}
 	var objMemParam, numaMemParam string
+
 	dimmName := "dimm1"
 	if config.Knobs.HugePages {
 		objMemParam = "memory-backend-file,id=" + dimmName + ",size=" + config.Memory.Size + ",mem-path=/dev/hugepages"
 		numaMemParam = "node,memdev=" + dimmName
+	} else if config.Knobs.MemFDPrivate {
+		objMemParam = "memory-backend-memfd-private,id=" + dimmName + ",size=" + config.Memory.Size
 	} else if config.Knobs.FileBackedMem && config.Memory.Path != "" {
 		objMemParam = "memory-backend-file,id=" + dimmName + ",size=" + config.Memory.Size + ",mem-path=" + config.Memory.Path
 		numaMemParam = "node,memdev=" + dimmName

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

@@ -632,6 +632,29 @@ func TestAppendMemoryFileBackedMemPrealloc(t *testing.T) {
 	testConfigAppend(conf, knobs, memString+" "+knobsString, t)
 }
 
+func TestAppendMemoryBackedMemFdPrivate(t *testing.T) {
+	conf := &Config{
+		Memory: Memory{
+			Size:  "1G",
+			Slots: 8,
+		},
+	}
+	memString := "-m 1G,slots=8"
+	testConfigAppend(conf, conf.Memory, memString, t)
+
+	knobs := Knobs{
+		MemFDPrivate: true,
+		MemShared:    false,
+	}
+	objMemString := "-object memory-backend-memfd-private,id=dimm1,size=1G"
+	memBackendString := "-machine memory-backend=dimm1"
+
+	knobsString := objMemString + " "
+	knobsString += memBackendString
+
+	testConfigAppend(conf, knobs, memString+" "+knobsString, t)
+}
+
 func TestNoRebootKnob(t *testing.T) {
 	conf := &Config{}
 

src/runtime/virtcontainers/qemu.go

@@ -604,6 +604,23 @@ func (q *qemu) CreateVM(ctx context.Context, id string, network Network, hypervi
 		return err
 	}
 
+	if q.config.ConfidentialGuest {
+		// At this point we're safe to just check for the protection field
+		// on the hypervisor specific code, as availableGuestProtection()
+		// has been called earlier and we know we have the value stored.
+		if q.arch.getProtection() == tdxProtection {
+			knobs.MemFDPrivate = true
+
+			// In case Nydus or VirtioFS is used, which may become a reality
+			// in the future, whenever we get those hardened for TDX, those
+			// knobs below would be automatically set.  Let's make sure we
+			// pre-emptively disable them, and with that we can avoid some
+			// headaches in the future.
+			knobs.FileBackedMem = false
+			knobs.MemShared = false
+		}
+	}
+
 	kernelPath, err := q.config.KernelAssetPath()
 	if err != nil {
 		return err

src/runtime/virtcontainers/qemu_arch_base.go

@@ -71,6 +71,9 @@ type qemuArch interface {
 	// memoryTopology returns the memory topology using the given amount of memoryMb and hostMemoryMb
 	memoryTopology(memoryMb, hostMemoryMb uint64, slots uint8) govmmQemu.Memory
 
+	// protection returns platform protection
+	getProtection() guestProtection
+
 	// appendConsole appends a console to devices
 	appendConsole(ctx context.Context, devices []govmmQemu.Device, path string) ([]govmmQemu.Device, error)
 
@@ -280,6 +283,10 @@ func (q *qemuArchBase) machine() govmmQemu.Machine {
 	return q.qemuMachine
 }
 
+func (q *qemuArchBase) getProtection() guestProtection {
+	return q.protection
+}
+
 func (q *qemuArchBase) qemuPath() string {
 	return q.qemuExePath
 }