delve/pkg/proc/amd64util/xsave.go

package amd64util

import (
	"bytes"
	"encoding/binary"
	"fmt"

	"github.com/go-delve/delve/pkg/proc"
)

// AMD64Xstate represents amd64 XSAVE area. See Section 13.1 (and
// following) of Intel® 64 and IA-32 Architectures Software Developer’s
// Manual, Volume 1: Basic Architecture.
type AMD64Xstate struct {
	AMD64PtraceFpRegs
	Xsave       []byte // raw xsave area
	AvxState    bool   // contains AVX state
	YmmSpace    [256]byte
	Avx512State bool // contains AVX512 state
	ZmmSpace    [512]byte
}

// AMD64PtraceFpRegs tracks user_fpregs_struct in /usr/include/x86_64-linux-gnu/sys/user.h
type AMD64PtraceFpRegs struct {
	Cwd      uint16
	Swd      uint16
	Ftw      uint16
	Fop      uint16
	Rip      uint64
	Rdp      uint64
	Mxcsr    uint32
	MxcrMask uint32
	StSpace  [32]uint32
	XmmSpace [256]byte
	Padding  [24]uint32
}

// Decode decodes an XSAVE area to a list of name/value pairs of registers.
func (xstate *AMD64Xstate) Decode() []proc.Register {
	var regs []proc.Register
	// x87 registers
	regs = proc.AppendUint64Register(regs, "CW", uint64(xstate.Cwd))
	regs = proc.AppendUint64Register(regs, "SW", uint64(xstate.Swd))
	regs = proc.AppendUint64Register(regs, "TW", uint64(xstate.Ftw))
	regs = proc.AppendUint64Register(regs, "FOP", uint64(xstate.Fop))
	regs = proc.AppendUint64Register(regs, "FIP", xstate.Rip)
	regs = proc.AppendUint64Register(regs, "FDP", xstate.Rdp)

	for i := 0; i < len(xstate.StSpace); i += 4 {
		var buf bytes.Buffer
		binary.Write(&buf, binary.LittleEndian, uint64(xstate.StSpace[i+1])<<32|uint64(xstate.StSpace[i]))
		binary.Write(&buf, binary.LittleEndian, uint16(xstate.StSpace[i+2]))
		regs = proc.AppendBytesRegister(regs, fmt.Sprintf("ST(%d)", i/4), buf.Bytes())
	}

	// SSE registers
	regs = proc.AppendUint64Register(regs, "MXCSR", uint64(xstate.Mxcsr))
	regs = proc.AppendUint64Register(regs, "MXCSR_MASK", uint64(xstate.MxcrMask))

	for i := 0; i < len(xstate.XmmSpace); i += 16 {
		n := i / 16
		regs = proc.AppendBytesRegister(regs, fmt.Sprintf("XMM%d", n), xstate.XmmSpace[i:i+16])
		if xstate.AvxState {
			regs = proc.AppendBytesRegister(regs, fmt.Sprintf("YMM%d", n), xstate.YmmSpace[i:i+16])
			if xstate.Avx512State {
				regs = proc.AppendBytesRegister(regs, fmt.Sprintf("ZMM%d", n), xstate.ZmmSpace[n*32:(n+1)*32])
			}
		}
	}

	return regs
}

const (
	_XSTATE_MAX_KNOWN_SIZE = 2969

	_XSAVE_XMM_REGION_START        = 160
	_XSAVE_HEADER_START            = 512
	_XSAVE_HEADER_LEN              = 64
	_XSAVE_EXTENDED_REGION_START   = 576
	_XSAVE_SSE_REGION_LEN          = 416
)

// AMD64XstateRead reads a byte array containing an XSAVE area into regset.
// If readLegacy is true regset.PtraceFpRegs will be filled with the
// contents of the legacy region of the XSAVE area.
// See Section 13.1 (and following) of Intel® 64 and IA-32 Architectures
// Software Developer’s Manual, Volume 1: Basic Architecture.
// If xstateZMMHi256Offset is zero, it will be guessed.
func AMD64XstateRead(xstateargs []byte, readLegacy bool, regset *AMD64Xstate, xstateZMMHi256Offset int) error {
	if _XSAVE_HEADER_START+_XSAVE_HEADER_LEN >= len(xstateargs) {
		return nil
	}
	if readLegacy {
		rdr := bytes.NewReader(xstateargs[:_XSAVE_HEADER_START])
		if err := binary.Read(rdr, binary.LittleEndian, &regset.AMD64PtraceFpRegs); err != nil {
			return err
		}
	}
	xsaveheader := xstateargs[_XSAVE_HEADER_START : _XSAVE_HEADER_START+_XSAVE_HEADER_LEN]
	xstate_bv := binary.LittleEndian.Uint64(xsaveheader[0:8])
	xcomp_bv := binary.LittleEndian.Uint64(xsaveheader[8:16])

	if xcomp_bv&(1<<63) != 0 {
		// compact format not supported
		return nil
	}

	if xstate_bv&(1<<2) == 0 {
		// AVX state not present
		return nil
	}

	avxstate := xstateargs[_XSAVE_EXTENDED_REGION_START:]
	regset.AvxState = true
	copy(regset.YmmSpace[:], avxstate[:len(regset.YmmSpace)])

	if xstate_bv&(1<<6) == 0 {
		// AVX512 state not present
		return nil
	}

	if xstateZMMHi256Offset == 0 {
		// Guess ZMM_Hi256 component offset
		// ref: https://github.com/bminor/binutils-gdb/blob/df89bdf0baf106c3b0a9fae53e4e48607a7f3f87/gdb/i387-tdep.c#L916
		if xstate_bv&(1<<9) != 0 && len(xstateargs) == 2440 {
			// AMD CPUs supporting PKRU
			xstateZMMHi256Offset = 896
		} else {
			// Intel CPUs supporting AVX512
			xstateZMMHi256Offset = 1152
		}
	}

	avx512state := xstateargs[xstateZMMHi256Offset:]
	regset.Avx512State = true
	copy(regset.ZmmSpace[:], avx512state[:len(regset.ZmmSpace)])

	// TODO(aarzilli): if xstate_bv&(1<<7) is set then xstateargs[1664:2688]
	// contains ZMM16 through ZMM31, those aren't just the higher 256bits, it's
	// the full register so each is 64 bytes (512bits)

	return nil
}

func (xstate *AMD64Xstate) SetXmmRegister(n int, value []byte) error {
	if n >= 16 {
		return fmt.Errorf("setting register XMM%d not supported", n)
	}
	if len(value) > 64 {
		return fmt.Errorf("value of register XMM%d too large (%d bytes)", n, len(value))
	}

	// Copy least significant 16 bytes to Xsave area

	xmmval := value
	if len(xmmval) > 16 {
		xmmval = xmmval[:16]
	}
	rest := value[len(xmmval):]

	xmmpos := _XSAVE_XMM_REGION_START + (n * 16)
	if xmmpos >= len(xstate.Xsave) {
		return fmt.Errorf("could not set XMM%d: not in XSAVE area", n)
	}

	copy(xstate.Xsave[xmmpos:], xmmval)

	if len(rest) == 0 {
		return nil
	}

	// Copy bytes [16, 32) to Xsave area

	ymmval := rest
	if len(ymmval) > 16 {
		ymmval = ymmval[:16]
	}
	rest = rest[len(ymmval):]

	ymmpos := _XSAVE_EXTENDED_REGION_START + (n * 16)
	if ymmpos >= len(xstate.Xsave) {
		return fmt.Errorf("could not set XMM%d: bytes 16..%d not in XSAVE area", n, 16+len(ymmval))
	}

	copy(xstate.Xsave[ymmpos:], ymmval)

	if len(rest) == 0 {
		return nil
	}

	// Copy bytes [32, 64) to Xsave area

	zmmval := rest
	zmmpos := AMD64XstateZMMHi256Offset() + (n * 32)
	if zmmpos >= len(xstate.Xsave) {
		return fmt.Errorf("could not set XMM%d: bytes 32..%d not in XSAVE area", n, 32+len(zmmval))
	}

	copy(xstate.Xsave[zmmpos:], zmmval)
	return nil
}