kvm-tls-2:add the preservation of user-TLS in the Arm64 kvm platform

This patch load/save TLS for the container application.

Related issue: full context-switch supporting for Arm64 #1238

COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gvisor/pull/2761 from lubinszARM:pr_tls_2 cb5dbca1c9c3f378002406da7a58887f9b5032b3
PiperOrigin-RevId: 322887044

pkg/sentry/arch/arch_aarch64.go

@@ -28,7 +28,14 @@ import (
 )
 
 // Registers represents the CPU registers for this architecture.
-type Registers = linux.PtraceRegs
+//
+// +stateify savable
+type Registers struct {
+	linux.PtraceRegs
+
+	// TPIDR_EL0 is the EL0 Read/Write Software Thread ID Register.
+	TPIDR_EL0 uint64
+}
 
 const (
 	// SyscallWidth is the width of insturctions.
@@ -101,9 +108,6 @@ type State struct {
 	// Our floating point state.
 	aarch64FPState `state:"wait"`
 
-	// TLS pointer
-	TPValue uint64
-
 	// FeatureSet is a pointer to the currently active feature set.
 	FeatureSet *cpuid.FeatureSet
 
@@ -157,7 +161,6 @@ func (s *State) Fork() State {
 	return State{
 		Regs:           s.Regs,
 		aarch64FPState: s.aarch64FPState.fork(),
-		TPValue:        s.TPValue,
 		FeatureSet:     s.FeatureSet,
 		OrigR0:         s.OrigR0,
 	}
@@ -241,18 +244,18 @@ func (s *State) ptraceGetRegs() Registers {
 	return s.Regs
 }
 
-var registersSize = (*Registers)(nil).SizeBytes()
+var ptraceRegistersSize = (*linux.PtraceRegs)(nil).SizeBytes()
 
 // PtraceSetRegs implements Context.PtraceSetRegs.
 func (s *State) PtraceSetRegs(src io.Reader) (int, error) {
 	var regs Registers
-	buf := make([]byte, registersSize)
+	buf := make([]byte, ptraceRegistersSize)
 	if _, err := io.ReadFull(src, buf); err != nil {
 		return 0, err
 	}
 	regs.UnmarshalUnsafe(buf)
 	s.Regs = regs
-	return registersSize, nil
+	return ptraceRegistersSize, nil
 }
 
 // PtraceGetFPRegs implements Context.PtraceGetFPRegs.
@@ -278,7 +281,7 @@ const (
 func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) {
 	switch regset {
 	case _NT_PRSTATUS:
-		if maxlen < registersSize {
+		if maxlen < ptraceRegistersSize {
 			return 0, syserror.EFAULT
 		}
 		return s.PtraceGetRegs(dst)
@@ -291,7 +294,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int,
 func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) {
 	switch regset {
 	case _NT_PRSTATUS:
-		if maxlen < registersSize {
+		if maxlen < ptraceRegistersSize {
 			return 0, syserror.EFAULT
 		}
 		return s.PtraceSetRegs(src)

pkg/sentry/arch/arch_amd64.go

@@ -300,7 +300,7 @@ func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) {
 	// PTRACE_PEEKUSER and PTRACE_POKEUSER are only effective on regs and
 	// u_debugreg, returning 0 or silently no-oping for other fields
 	// respectively.
-	if addr < uintptr(registersSize) {
+	if addr < uintptr(ptraceRegistersSize) {
 		regs := c.ptraceGetRegs()
 		buf := make([]byte, regs.SizeBytes())
 		regs.MarshalUnsafe(buf)
@@ -315,7 +315,7 @@ func (c *context64) PtracePokeUser(addr, data uintptr) error {
 	if addr&7 != 0 || addr >= userStructSize {
 		return syscall.EIO
 	}
-	if addr < uintptr(registersSize) {
+	if addr < uintptr(ptraceRegistersSize) {
 		regs := c.ptraceGetRegs()
 		buf := make([]byte, regs.SizeBytes())
 		regs.MarshalUnsafe(buf)

pkg/sentry/arch/arch_arm64.go

@@ -142,7 +142,7 @@ func (c *context64) SetStack(value uintptr) {
 
 // TLS returns the current TLS pointer.
 func (c *context64) TLS() uintptr {
-	return uintptr(c.TPValue)
+	return uintptr(c.Regs.TPIDR_EL0)
 }
 
 // SetTLS sets the current TLS pointer. Returns false if value is invalid.
@@ -151,7 +151,7 @@ func (c *context64) SetTLS(value uintptr) bool {
 		return false
 	}
 
-	c.TPValue = uint64(value)
+	c.Regs.TPIDR_EL0 = uint64(value)
 	return true
 }
 

pkg/sentry/arch/arch_x86.go

@@ -31,7 +31,11 @@ import (
 )
 
 // Registers represents the CPU registers for this architecture.
-type Registers = linux.PtraceRegs
+//
+// +stateify savable
+type Registers struct {
+	linux.PtraceRegs
+}
 
 // System-related constants for x86.
 const (
@@ -311,12 +315,12 @@ func (s *State) ptraceGetRegs() Registers {
 	return regs
 }
 
-var registersSize = (*Registers)(nil).SizeBytes()
+var ptraceRegistersSize = (*linux.PtraceRegs)(nil).SizeBytes()
 
 // PtraceSetRegs implements Context.PtraceSetRegs.
 func (s *State) PtraceSetRegs(src io.Reader) (int, error) {
 	var regs Registers
-	buf := make([]byte, registersSize)
+	buf := make([]byte, ptraceRegistersSize)
 	if _, err := io.ReadFull(src, buf); err != nil {
 		return 0, err
 	}
@@ -374,7 +378,7 @@ func (s *State) PtraceSetRegs(src io.Reader) (int, error) {
 	}
 	regs.Eflags = (s.Regs.Eflags &^ eflagsPtraceMutable) | (regs.Eflags & eflagsPtraceMutable)
 	s.Regs = regs
-	return registersSize, nil
+	return ptraceRegistersSize, nil
 }
 
 // isUserSegmentSelector returns true if the given segment selector specifies a
@@ -543,7 +547,7 @@ const (
 func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) {
 	switch regset {
 	case _NT_PRSTATUS:
-		if maxlen < registersSize {
+		if maxlen < ptraceRegistersSize {
 			return 0, syserror.EFAULT
 		}
 		return s.PtraceGetRegs(dst)
@@ -563,7 +567,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int,
 func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) {
 	switch regset {
 	case _NT_PRSTATUS:
-		if maxlen < registersSize {
+		if maxlen < ptraceRegistersSize {
 			return 0, syserror.EFAULT
 		}
 		return s.PtraceSetRegs(src)

pkg/sentry/platform/ring0/kernel_arm64.go

@@ -58,7 +58,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) {
 
 	regs.Pstate &= ^uint64(UserFlagsClear)
 	regs.Pstate |= UserFlagsSet
+
+	SetTLS(regs.TPIDR_EL0)
+
 	kernelExitToEl0()
+
+	regs.TPIDR_EL0 = GetTLS()
+
 	vector = c.vecCode
 
 	// Perform the switch.