Fix kernel flags handling and add missing vectors.

PiperOrigin-RevId: 199877174
Change-Id: I9d19ea301608c2b989df0a6123abb1e779427853

5 files changed, 70 insertions, 31 deletions

diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go
index 7fcb7451f..7ac289756 100644
--- a/pkg/sentry/platform/kvm/machine_amd64.go
+++ b/pkg/sentry/platform/kvm/machine_amd64.go
@@ -150,13 +150,20 @@ func (c *vCPU) fault(signal int32) (*arch.SignalInfo, usermem.AccessType, error)
 		// the code provided here. We need to re-execute.
 		return nil, usermem.NoAccess, platform.ErrContextInterrupt
 	}
-	info := &arch.SignalInfo{Signo: signal}
+	info := &arch.SignalInfo{
+		Signo: signal,
+	}
 	info.SetAddr(uint64(faultAddr))
 	accessType := usermem.AccessType{
 		Read:    code&(1<<1) == 0,
 		Write:   code&(1<<1) != 0,
 		Execute: code&(1<<4) != 0,
 	}
+	if !accessType.Write && !accessType.Execute {
+		info.Code = 1 // SEGV_MAPERR.
+	} else {
+		info.Code = 2 // SEGV_ACCERR.
+	}
 	return info, accessType, platform.ErrContextSignal
 }
 
@@ -191,30 +198,55 @@ func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts) (*arch.SignalInfo, user
 		return c.fault(int32(syscall.SIGSEGV))
 
 	case ring0.Debug, ring0.Breakpoint:
-		info := &arch.SignalInfo{Signo: int32(syscall.SIGTRAP)}
+		info := &arch.SignalInfo{
+			Signo: int32(syscall.SIGTRAP),
+			Code:  1, // TRAP_BRKPT (breakpoint).
+		}
+		info.SetAddr(switchOpts.Registers.Rip) // Include address.
 		return info, usermem.AccessType{}, platform.ErrContextSignal
 
 	case ring0.GeneralProtectionFault:
-		if !ring0.IsCanonical(switchOpts.Registers.Rip) {
-			// If the RIP is non-canonical, it's a SEGV.
-			info := &arch.SignalInfo{Signo: int32(syscall.SIGSEGV)}
-			return info, usermem.AccessType{}, platform.ErrContextSignal
+		info := &arch.SignalInfo{
+			Signo: int32(syscall.SIGSEGV),
+			Code:  arch.SignalInfoKernel,
 		}
-		// Otherwise, we deliver a SIGBUS.
-		info := &arch.SignalInfo{Signo: int32(syscall.SIGBUS)}
+		info.SetAddr(switchOpts.Registers.Rip) // Include address.
 		return info, usermem.AccessType{}, platform.ErrContextSignal
 
 	case ring0.InvalidOpcode:
-		info := &arch.SignalInfo{Signo: int32(syscall.SIGILL)}
+		info := &arch.SignalInfo{
+			Signo: int32(syscall.SIGILL),
+			Code:  1, // ILL_ILLOPC (illegal opcode).
+		}
+		info.SetAddr(switchOpts.Registers.Rip) // Include address.
+		return info, usermem.AccessType{}, platform.ErrContextSignal
+
+	case ring0.DivideByZero:
+		info := &arch.SignalInfo{
+			Signo: int32(syscall.SIGFPE),
+			Code:  1, // FPE_INTDIV (divide by zero).
+		}
+		info.SetAddr(switchOpts.Registers.Rip) // Include address.
 		return info, usermem.AccessType{}, platform.ErrContextSignal
 
 	case ring0.X87FloatingPointException:
-		info := &arch.SignalInfo{Signo: int32(syscall.SIGFPE)}
+		info := &arch.SignalInfo{
+			Signo: int32(syscall.SIGFPE),
+			Code:  7, // FPE_FLTINV (invalid operation).
+		}
+		info.SetAddr(switchOpts.Registers.Rip) // Include address.
 		return info, usermem.AccessType{}, platform.ErrContextSignal
 
 	case ring0.Vector(bounce):
 		return nil, usermem.NoAccess, platform.ErrContextInterrupt
 
+	case ring0.AlignmentCheck:
+		info := &arch.SignalInfo{
+			Signo: int32(syscall.SIGBUS),
+			Code:  2, // BUS_ADRERR (physical address does not exist).
+		}
+		return info, usermem.NoAccess, platform.ErrContextSignal
+
 	case ring0.NMI:
 		// An NMI is generated only when a fault is not servicable by
 		// KVM itself, so we think some mapping is writeable but it's
diff --git a/pkg/sentry/platform/ring0/entry_amd64.s b/pkg/sentry/platform/ring0/entry_amd64.s
index e8638133b..08c15ad65 100644
--- a/pkg/sentry/platform/ring0/entry_amd64.s
+++ b/pkg/sentry/platform/ring0/entry_amd64.s
@@ -248,10 +248,12 @@ TEXT ·exception(SB),NOSPLIT,$0
 
 user:
 	SWAP_GS()
-	XCHGQ CPU_REGISTERS+PTRACE_RAX(GS), AX // Swap for AX (regs).
-	REGISTERS_SAVE(AX, 0)                  // Save all except IP, FLAGS, SP, AX.
-	MOVQ CPU_REGISTERS+PTRACE_RAX(GS), BX  // Load saved AX value.
-	MOVQ BX, PTRACE_RAX(AX)                // Save everything else.
+	ADDQ $-8, SP                            // Adjust for flags.
+	MOVQ $_KERNEL_FLAGS, 0(SP); BYTE $0x9d; // Reset flags (POPFQ).
+	XCHGQ CPU_REGISTERS+PTRACE_RAX(GS), AX  // Swap for user regs.
+	REGISTERS_SAVE(AX, 0)                   // Save all except IP, FLAGS, SP, AX.
+	MOVQ CPU_REGISTERS+PTRACE_RAX(GS), BX   // Restore original AX.
+	MOVQ BX, PTRACE_RAX(AX)                 // Save it.
 	MOVQ BX, PTRACE_ORIGRAX(AX)
 	MOVQ 16(SP), BX; MOVQ BX, PTRACE_RIP(AX)
 	MOVQ 24(SP), CX; MOVQ CX, PTRACE_CS(AX)
diff --git a/pkg/sentry/platform/ring0/kernel_amd64.go b/pkg/sentry/platform/ring0/kernel_amd64.go
index 37d5484e1..c828af654 100644
--- a/pkg/sentry/platform/ring0/kernel_amd64.go
+++ b/pkg/sentry/platform/ring0/kernel_amd64.go
@@ -20,20 +20,6 @@ import (
 	"encoding/binary"
 )
 
-const (
-	// KernelFlagsSet should always be set in the kernel.
-	KernelFlagsSet = _RFLAGS_RESERVED
-
-	// UserFlagsSet are always set in userspace.
-	UserFlagsSet = _RFLAGS_RESERVED | _RFLAGS_IF
-
-	// KernelFlagsClear should always be clear in the kernel.
-	KernelFlagsClear = _RFLAGS_IF | _RFLAGS_NT | _RFLAGS_IOPL
-
-	// UserFlagsClear are always cleared in userspace.
-	UserFlagsClear = _RFLAGS_NT | _RFLAGS_IOPL
-)
-
 // init initializes architecture-specific state.
 func (k *Kernel) init(opts KernelOpts) {
 	// Save the root page tables.
@@ -85,6 +71,9 @@ func (c *CPU) init() {
 	c.registers.Ss = uint64(Kdata)
 	c.registers.Fs = uint64(Kdata)
 	c.registers.Gs = uint64(Kdata)
+
+	// Set mandatory flags.
+	c.registers.Eflags = KernelFlagsSet
 }
 
 // StackTop returns the kernel's stack address.
@@ -119,7 +108,7 @@ func (c *CPU) TSS() (uint64, uint16, *SegmentDescriptor) {
 //
 //go:nosplit
 func (c *CPU) CR0() uint64 {
-	return _CR0_PE | _CR0_PG | _CR0_ET
+	return _CR0_PE | _CR0_PG | _CR0_AM | _CR0_ET
 }
 
 // CR4 returns the CPU's CR4 value.
@@ -240,7 +229,7 @@ func start(c *CPU) {
 
 	// Set the syscall target.
 	wrmsr(_MSR_LSTAR, kernelFunc(sysenter))
-	wrmsr(_MSR_SYSCALL_MASK, _RFLAGS_STEP|_RFLAGS_IF|_RFLAGS_DF|_RFLAGS_IOPL|_RFLAGS_AC|_RFLAGS_NT)
+	wrmsr(_MSR_SYSCALL_MASK, KernelFlagsClear|_RFLAGS_DF)
 
 	// NOTE: This depends on having the 64-bit segments immediately
 	// following the 32-bit user segments. This is simply the way the
diff --git a/pkg/sentry/platform/ring0/offsets_amd64.go b/pkg/sentry/platform/ring0/offsets_amd64.go
index 9acd442ba..ca5fd456b 100644
--- a/pkg/sentry/platform/ring0/offsets_amd64.go
+++ b/pkg/sentry/platform/ring0/offsets_amd64.go
@@ -38,7 +38,8 @@ func Emit(w io.Writer) {
 	fmt.Fprintf(w, "#define CPU_KERNEL_SYSCALL   0x%02x\n", reflect.ValueOf(&c.KernelSyscall).Pointer()-reflect.ValueOf(c).Pointer())
 
 	fmt.Fprintf(w, "\n// Bits.\n")
-	fmt.Fprintf(w, "#define _RFLAGS_IF 0x%02x\n", _RFLAGS_IF)
+	fmt.Fprintf(w, "#define _RFLAGS_IF           0x%02x\n", _RFLAGS_IF)
+	fmt.Fprintf(w, "#define _KERNEL_FLAGS        0x%02x\n", KernelFlagsSet)
 
 	fmt.Fprintf(w, "\n// Vectors.\n")
 	fmt.Fprintf(w, "#define DivideByZero               0x%02x\n", DivideByZero)
diff --git a/pkg/sentry/platform/ring0/x86.go b/pkg/sentry/platform/ring0/x86.go
index 74b140066..3d437a77c 100644
--- a/pkg/sentry/platform/ring0/x86.go
+++ b/pkg/sentry/platform/ring0/x86.go
@@ -24,6 +24,7 @@ import (
 const (
 	_CR0_PE = 1 << 0
 	_CR0_ET = 1 << 4
+	_CR0_AM = 1 << 18
 	_CR0_PG = 1 << 31
 
 	_CR4_PSE        = 1 << 4
@@ -55,6 +56,20 @@ const (
 	_MSR_SYSCALL_MASK = 0xc0000084
 )
 
+const (
+	// KernelFlagsSet should always be set in the kernel.
+	KernelFlagsSet = _RFLAGS_RESERVED
+
+	// UserFlagsSet are always set in userspace.
+	UserFlagsSet = _RFLAGS_RESERVED | _RFLAGS_IF
+
+	// KernelFlagsClear should always be clear in the kernel.
+	KernelFlagsClear = _RFLAGS_STEP | _RFLAGS_IF | _RFLAGS_IOPL | _RFLAGS_AC | _RFLAGS_NT
+
+	// UserFlagsClear are always cleared in userspace.
+	UserFlagsClear = _RFLAGS_NT | _RFLAGS_IOPL
+)
+
 // Vector is an exception vector.
 type Vector uintptr