kvm: prefault a floating point state before restoring it

If physical pages of a memory region are not mapped yet, the kernel will
trigger KVM_EXIT_MMIO and we will map physical pages in bluepillHandler().

An instruction that triggered a fault will not be re-executed, it
will be emulated in the kernel, but it can't  emulate complex
instructions like xsave, xrstor. We can touch the memory with
simple instructions to workaround this problem.

3 files changed, 13 insertions, 13 deletions

diff --git a/pkg/sentry/arch/arch.go b/pkg/sentry/arch/arch.go
index dd2effdf9..3443b9e1b 100644
--- a/pkg/sentry/arch/arch.go
+++ b/pkg/sentry/arch/arch.go
@@ -54,7 +54,7 @@ func (a Arch) String() string {
 // We rely on the individual arch implementations to meet all the necessary
 // requirements. For example, on x86 the region must be 16-byte aligned and 512
 // bytes in size.
-type FloatingPointData byte
+type FloatingPointData []byte
 
 // Context provides architecture-dependent information for a specific thread.
 //
@@ -187,7 +187,7 @@ type Context interface {
 	ClearSingleStep()
 
 	// FloatingPointData will be passed to underlying save routines.
-	FloatingPointData() *FloatingPointData
+	FloatingPointData() FloatingPointData
 
 	// NewMmapLayout returns a layout for a new MM, where MinAddr for the
 	// returned layout must be no lower than min, and MaxAddr for the returned
diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go
index fd73751e7..6b81e9708 100644
--- a/pkg/sentry/arch/arch_aarch64.go
+++ b/pkg/sentry/arch/arch_aarch64.go
@@ -88,15 +88,15 @@ func (f aarch64FPState) fork() aarch64FPState {
 }
 
 // FloatingPointData returns the raw data pointer.
-func (f aarch64FPState) FloatingPointData() *FloatingPointData {
-	return (*FloatingPointData)(&f[0])
+func (f aarch64FPState) FloatingPointData() FloatingPointData {
+	return ([]byte)(f)
 }
 
 // NewFloatingPointData returns a new floating point data blob.
 //
 // This is primarily for use in tests.
-func NewFloatingPointData() *FloatingPointData {
-	return (*FloatingPointData)(&(newAarch64FPState()[0]))
+func NewFloatingPointData() FloatingPointData {
+	return ([]byte)(newAarch64FPState())
 }
 
 // State contains the common architecture bits for aarch64 (the build tag of this
diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go
index 641ada92f..91edf0703 100644
--- a/pkg/sentry/arch/arch_x86.go
+++ b/pkg/sentry/arch/arch_x86.go
@@ -115,7 +115,7 @@ var (
 type x86FPState []byte
 
 // initX86FPState (defined in asm files) sets up initial state.
-func initX86FPState(data *FloatingPointData, useXsave bool)
+func initX86FPState(data *byte, useXsave bool)
 
 func newX86FPStateSlice() []byte {
 	size, align := cpuid.HostFeatureSet().ExtendedStateSize()
@@ -139,7 +139,7 @@ func newX86FPStateSlice() []byte {
 // CPUID we must ensure it does not contain any sentry state.
 func newX86FPState() x86FPState {
 	f := x86FPState(newX86FPStateSlice())
-	initX86FPState(f.FloatingPointData(), cpuid.HostFeatureSet().UseXsave())
+	initX86FPState(&f.FloatingPointData()[0], cpuid.HostFeatureSet().UseXsave())
 	return f
 }
 
@@ -151,15 +151,15 @@ func (f x86FPState) fork() x86FPState {
 }
 
 // FloatingPointData returns the raw data pointer.
-func (f x86FPState) FloatingPointData() *FloatingPointData {
-	return (*FloatingPointData)(&f[0])
+func (f x86FPState) FloatingPointData() FloatingPointData {
+	return []byte(f)
 }
 
 // NewFloatingPointData returns a new floating point data blob.
 //
 // This is primarily for use in tests.
-func NewFloatingPointData() *FloatingPointData {
-	return (*FloatingPointData)(&(newX86FPState()[0]))
+func NewFloatingPointData() FloatingPointData {
+	return (FloatingPointData)(newX86FPState())
 }
 
 // Proto returns a protobuf representation of the system registers in State.
@@ -442,7 +442,7 @@ func sanitizeMXCSR(f x86FPState) {
 	mxcsr := usermem.ByteOrder.Uint32(f[mxcsrOffset:])
 	initMXCSRMask.Do(func() {
 		temp := x86FPState(alignedBytes(uint(ptraceFPRegsSize), 16))
-		initX86FPState(temp.FloatingPointData(), false /* useXsave */)
+		initX86FPState(&temp.FloatingPointData()[0], false /* useXsave */)
 		mxcsrMask = usermem.ByteOrder.Uint32(temp[mxcsrMaskOffset:])
 		if mxcsrMask == 0 {
 			// "If the value of the MXCSR_MASK field is 00000000H, then the