func progedit(ctxt *liblink.Link, p *liblink.Prog) {
var literal string
var s *liblink.LSym
var q *liblink.Prog
// See obj6.c for discussion of TLS.
if canuselocaltls(ctxt) != 0 {
// Reduce TLS initial exec model to TLS local exec model.
// Sequences like
// MOVL TLS, BX
// ... off(BX)(TLS*1) ...
// become
// NOP
// ... off(TLS) ...
if p.As == AMOVL && p.From.Type_ == D_TLS && D_AX <= p.To.Type_ && p.To.Type_ <= D_DI {
p.As = ANOP
p.From.Type_ = D_NONE
p.To.Type_ = D_NONE
}
if p.From.Index == D_TLS && D_INDIR+D_AX <= p.From.Type_ && p.From.Type_ <= D_INDIR+D_DI {
p.From.Type_ = D_INDIR + D_TLS
p.From.Scale = 0
p.From.Index = D_NONE
}
if p.To.Index == D_TLS && D_INDIR+D_AX <= p.To.Type_ && p.To.Type_ <= D_INDIR+D_DI {
p.To.Type_ = D_INDIR + D_TLS
p.To.Scale = 0
p.To.Index = D_NONE
}
} else {
// As a courtesy to the C compilers, rewrite TLS local exec load as TLS initial exec load.
// The instruction
// MOVL off(TLS), BX
// becomes the sequence
// MOVL TLS, BX
// MOVL off(BX)(TLS*1), BX
// This allows the C compilers to emit references to m and g using the direct off(TLS) form.
if p.As == AMOVL && p.From.Type_ == D_INDIR+D_TLS && D_AX <= p.To.Type_ && p.To.Type_ <= D_DI {
q = liblink.Appendp(ctxt, p)
q.As = p.As
q.From = p.From
q.From.Type_ = D_INDIR + p.To.Type_
q.From.Index = D_TLS
q.From.Scale = 2 // TODO: use 1
q.To = p.To
p.From.Type_ = D_TLS
p.From.Index = D_NONE
p.From.Offset = 0
}
}
// TODO: Remove.
if ctxt.Headtype == liblink.Hplan9 {
if p.From.Scale == 1 && p.From.Index == D_TLS {
p.From.Scale = 2
}
if p.To.Scale == 1 && p.To.Index == D_TLS {
p.To.Scale = 2
}
}
// Rewrite CALL/JMP/RET to symbol as D_BRANCH.
switch p.As {
case ACALL,
AJMP,
ARET:
if (p.To.Type_ == D_EXTERN || p.To.Type_ == D_STATIC) && p.To.Sym != nil {
p.To.Type_ = D_BRANCH
}
break
}
// Rewrite float constants to values stored in memory.
switch p.As {
// Convert AMOVSS $(0), Xx to AXORPS Xx, Xx
case AMOVSS:
if p.From.Type_ == D_FCONST {
if p.From.U.Dval == 0 {
if p.To.Type_ >= D_X0 {
if p.To.Type_ <= D_X7 {
p.As = AXORPS
p.From.Type_ = p.To.Type_
p.From.Index = p.To.Index
break
}
}
}
}
fallthrough
// fallthrough
case AFMOVF,
AFADDF,
AFSUBF,
AFSUBRF,
AFMULF,
AFDIVF,
AFDIVRF,
AFCOMF,
AFCOMFP,
AADDSS,
ASUBSS,
AMULSS,
ADIVSS,
ACOMISS,
AUCOMISS:
if p.From.Type_ == D_FCONST {
var i32 uint32
var f32 float32
f32 = float32(p.From.U.Dval)
i32 = math.Float32bits(f32)
literal = fmt.Sprintf("$f32.%08x", i32)
s = liblink.Linklookup(ctxt, literal, 0)
if s.Type_ == 0 {
s.Type_ = liblink.SRODATA
liblink.Adduint32(ctxt, s, i32)
s.Reachable = 0
}
p.From.Type_ = D_EXTERN
p.From.Sym = s
p.From.Offset = 0
}
// Convert AMOVSD $(0), Xx to AXORPS Xx, Xx
case AMOVSD:
if p.From.Type_ == D_FCONST {
if p.From.U.Dval == 0 {
if p.To.Type_ >= D_X0 {
if p.To.Type_ <= D_X7 {
p.As = AXORPS
p.From.Type_ = p.To.Type_
p.From.Index = p.To.Index
break
}
}
}
}
fallthrough
// fallthrough
case AFMOVD,
AFADDD,
AFSUBD,
AFSUBRD,
AFMULD,
AFDIVD,
AFDIVRD,
AFCOMD,
AFCOMDP,
AADDSD,
ASUBSD,
AMULSD,
ADIVSD,
ACOMISD,
AUCOMISD:
if p.From.Type_ == D_FCONST {
var i64 uint64
i64 = math.Float64bits(p.From.U.Dval)
literal = fmt.Sprintf("$f64.%016x", i64)
s = liblink.Linklookup(ctxt, literal, 0)
if s.Type_ == 0 {
s.Type_ = liblink.SRODATA
liblink.Adduint64(ctxt, s, i64)
s.Reachable = 0
}
p.From.Type_ = D_EXTERN
p.From.Sym = s
p.From.Offset = 0
}
break
}
}
func progedit(ctxt *liblink.Link, p *liblink.Prog) {
var literal string
var s *liblink.LSym
var tlsfallback *liblink.LSym
p.From.Class = 0
p.To.Class = 0
// Rewrite B/BL to symbol as D_BRANCH.
switch p.As {
case AB,
ABL,
ADUFFZERO,
ADUFFCOPY:
if p.To.Type_ == D_OREG && (p.To.Name == D_EXTERN || p.To.Name == D_STATIC) && p.To.Sym != nil {
p.To.Type_ = D_BRANCH
}
break
}
// Replace TLS register fetches on older ARM procesors.
switch p.As {
// Treat MRC 15, 0, <reg>, C13, C0, 3 specially.
case AMRC:
if p.To.Offset&0xffff0fff == 0xee1d0f70 {
// Because the instruction might be rewriten to a BL which returns in R0
// the register must be zero.
if p.To.Offset&0xf000 != 0 {
ctxt.Diag("%L: TLS MRC instruction must write to R0 as it might get translated into a BL instruction", p.Lineno)
}
if ctxt.Goarm < 7 {
// Replace it with BL runtime.read_tls_fallback(SB) for ARM CPUs that lack the tls extension.
if tlsfallback == nil {
tlsfallback = liblink.Linklookup(ctxt, "runtime.read_tls_fallback", 0)
}
// MOVW LR, R11
p.As = AMOVW
p.From.Type_ = D_REG
p.From.Reg = REGLINK
p.To.Type_ = D_REG
p.To.Reg = REGTMP
// BL runtime.read_tls_fallback(SB)
p = liblink.Appendp(ctxt, p)
p.As = ABL
p.To.Type_ = D_BRANCH
p.To.Sym = tlsfallback
p.To.Offset = 0
// MOVW R11, LR
p = liblink.Appendp(ctxt, p)
p.As = AMOVW
p.From.Type_ = D_REG
p.From.Reg = REGTMP
p.To.Type_ = D_REG
p.To.Reg = REGLINK
break
}
}
// Otherwise, MRC/MCR instructions need no further treatment.
p.As = AWORD
break
}
// Rewrite float constants to values stored in memory.
switch p.As {
case AMOVF:
if p.From.Type_ == D_FCONST && chipfloat5(ctxt, p.From.U.Dval) < 0 && (chipzero5(ctxt, p.From.U.Dval) < 0 || p.Scond&C_SCOND != C_SCOND_NONE) {
var i32 uint32
var f32 float32
f32 = float32(p.From.U.Dval)
i32 = math.Float32bits(f32)
literal = fmt.Sprintf("$f32.%08x", i32)
s = liblink.Linklookup(ctxt, literal, 0)
if s.Type_ == 0 {
s.Type_ = liblink.SRODATA
liblink.Adduint32(ctxt, s, i32)
s.Reachable = 0
}
p.From.Type_ = D_OREG
p.From.Sym = s
p.From.Name = D_EXTERN
p.From.Offset = 0
}
case AMOVD:
if p.From.Type_ == D_FCONST && chipfloat5(ctxt, p.From.U.Dval) < 0 && (chipzero5(ctxt, p.From.U.Dval) < 0 || p.Scond&C_SCOND != C_SCOND_NONE) {
var i64 uint64
i64 = math.Float64bits(p.From.U.Dval)
literal = fmt.Sprintf("$f64.%016x", i64)
s = liblink.Linklookup(ctxt, literal, 0)
if s.Type_ == 0 {
s.Type_ = liblink.SRODATA
liblink.Adduint64(ctxt, s, i64)
s.Reachable = 0
}
p.From.Type_ = D_OREG
p.From.Sym = s
p.From.Name = D_EXTERN
p.From.Offset = 0
}
break
}
if ctxt.Flag_shared != 0 {
// Shared libraries use R_ARM_TLS_IE32 instead of
// R_ARM_TLS_LE32, replacing the link time constant TLS offset in
// runtime.tlsg with an address to a GOT entry containing the
// offset. Rewrite $runtime.tlsg(SB) to runtime.tlsg(SB) to
// compensate.
if ctxt.Tlsg == nil {
ctxt.Tlsg = liblink.Linklookup(ctxt, "runtime.tlsg", 0)
}
if p.From.Type_ == D_CONST && p.From.Name == D_EXTERN && p.From.Sym == ctxt.Tlsg {
p.From.Type_ = D_OREG
}
if p.To.Type_ == D_CONST && p.To.Name == D_EXTERN && p.To.Sym == ctxt.Tlsg {
p.To.Type_ = D_OREG
}
}
}
func progedit(ctxt *liblink.Link, p *liblink.Prog) {
var literal string
var s *liblink.LSym
var q *liblink.Prog
// Thread-local storage references use the TLS pseudo-register.
// As a register, TLS refers to the thread-local storage base, and it
// can only be loaded into another register:
//
// MOVQ TLS, AX
//
// An offset from the thread-local storage base is written off(reg)(TLS*1).
// Semantically it is off(reg), but the (TLS*1) annotation marks this as
// indexing from the loaded TLS base. This emits a relocation so that
// if the linker needs to adjust the offset, it can. For example:
//
// MOVQ TLS, AX
// MOVQ 8(AX)(TLS*1), CX // load m into CX
//
// On systems that support direct access to the TLS memory, this
// pair of instructions can be reduced to a direct TLS memory reference:
//
// MOVQ 8(TLS), CX // load m into CX
//
// The 2-instruction and 1-instruction forms correspond roughly to
// ELF TLS initial exec mode and ELF TLS local exec mode, respectively.
//
// We applies this rewrite on systems that support the 1-instruction form.
// The decision is made using only the operating system (and probably
// the -shared flag, eventually), not the link mode. If some link modes
// on a particular operating system require the 2-instruction form,
// then all builds for that operating system will use the 2-instruction
// form, so that the link mode decision can be delayed to link time.
//
// In this way, all supported systems use identical instructions to
// access TLS, and they are rewritten appropriately first here in
// liblink and then finally using relocations in the linker.
if canuselocaltls(ctxt) != 0 {
// Reduce TLS initial exec model to TLS local exec model.
// Sequences like
// MOVQ TLS, BX
// ... off(BX)(TLS*1) ...
// become
// NOP
// ... off(TLS) ...
//
// TODO(rsc): Remove the Hsolaris special case. It exists only to
// guarantee we are producing byte-identical binaries as before this code.
// But it should be unnecessary.
if (p.As == AMOVQ || p.As == AMOVL) && p.From.Type_ == D_TLS && D_AX <= p.To.Type_ && p.To.Type_ <= D_R15 && ctxt.Headtype != liblink.Hsolaris {
nopout(p)
}
if p.From.Index == D_TLS && D_INDIR+D_AX <= p.From.Type_ && p.From.Type_ <= D_INDIR+D_R15 {
p.From.Type_ = D_INDIR + D_TLS
p.From.Scale = 0
p.From.Index = D_NONE
}
if p.To.Index == D_TLS && D_INDIR+D_AX <= p.To.Type_ && p.To.Type_ <= D_INDIR+D_R15 {
p.To.Type_ = D_INDIR + D_TLS
p.To.Scale = 0
p.To.Index = D_NONE
}
} else {
// As a courtesy to the C compilers, rewrite TLS local exec load as TLS initial exec load.
// The instruction
// MOVQ off(TLS), BX
// becomes the sequence
// MOVQ TLS, BX
// MOVQ off(BX)(TLS*1), BX
// This allows the C compilers to emit references to m and g using the direct off(TLS) form.
if (p.As == AMOVQ || p.As == AMOVL) && p.From.Type_ == D_INDIR+D_TLS && D_AX <= p.To.Type_ && p.To.Type_ <= D_R15 {
q = liblink.Appendp(ctxt, p)
q.As = p.As
q.From = p.From
q.From.Type_ = D_INDIR + p.To.Type_
q.From.Index = D_TLS
q.From.Scale = 2 // TODO: use 1
q.To = p.To
p.From.Type_ = D_TLS
p.From.Index = D_NONE
p.From.Offset = 0
}
}
// TODO: Remove.
if ctxt.Headtype == liblink.Hwindows || ctxt.Headtype == liblink.Hplan9 {
if p.From.Scale == 1 && p.From.Index == D_TLS {
p.From.Scale = 2
}
if p.To.Scale == 1 && p.To.Index == D_TLS {
p.To.Scale = 2
}
}
if ctxt.Headtype == liblink.Hnacl {
nacladdr(ctxt, p, &p.From)
nacladdr(ctxt, p, &p.To)
}
// Maintain information about code generation mode.
if ctxt.Mode == 0 {
ctxt.Mode = 64
}
p.Mode = int8(ctxt.Mode)
switch p.As {
case AMODE:
if p.From.Type_ == D_CONST || p.From.Type_ == D_INDIR+D_NONE {
switch int(p.From.Offset) {
case 16,
32,
64:
ctxt.Mode = int(p.From.Offset)
break
}
}
nopout(p)
break
}
// Rewrite CALL/JMP/RET to symbol as D_BRANCH.
switch p.As {
case ACALL,
AJMP,
ARET:
if (p.To.Type_ == D_EXTERN || p.To.Type_ == D_STATIC) && p.To.Sym != nil {
p.To.Type_ = D_BRANCH
}
break
}
// Rewrite float constants to values stored in memory.
switch p.As {
// Convert AMOVSS $(0), Xx to AXORPS Xx, Xx
case AMOVSS:
if p.From.Type_ == D_FCONST {
if p.From.U.Dval == 0 {
if p.To.Type_ >= D_X0 {
if p.To.Type_ <= D_X15 {
p.As = AXORPS
p.From.Type_ = p.To.Type_
p.From.Index = p.To.Index
break
}
}
}
}
fallthrough
// fallthrough
case AFMOVF,
AFADDF,
AFSUBF,
AFSUBRF,
AFMULF,
AFDIVF,
AFDIVRF,
AFCOMF,
AFCOMFP,
AADDSS,
ASUBSS,
AMULSS,
ADIVSS,
ACOMISS,
AUCOMISS:
if p.From.Type_ == D_FCONST {
var i32 uint32
var f32 float32
f32 = float32(p.From.U.Dval)
i32 = math.Float32bits(f32)
literal = fmt.Sprintf("$f32.%08x", i32)
s = liblink.Linklookup(ctxt, literal, 0)
if s.Type_ == 0 {
s.Type_ = liblink.SRODATA
liblink.Adduint32(ctxt, s, i32)
s.Reachable = 0
}
p.From.Type_ = D_EXTERN
p.From.Sym = s
p.From.Offset = 0
}
// Convert AMOVSD $(0), Xx to AXORPS Xx, Xx
case AMOVSD:
if p.From.Type_ == D_FCONST {
if p.From.U.Dval == 0 {
if p.To.Type_ >= D_X0 {
if p.To.Type_ <= D_X15 {
p.As = AXORPS
p.From.Type_ = p.To.Type_
p.From.Index = p.To.Index
break
}
}
}
}
fallthrough
// fallthrough
case AFMOVD,
AFADDD,
AFSUBD,
AFSUBRD,
AFMULD,
AFDIVD,
AFDIVRD,
AFCOMD,
AFCOMDP,
AADDSD,
ASUBSD,
AMULSD,
ADIVSD,
ACOMISD,
AUCOMISD:
if p.From.Type_ == D_FCONST {
var i64 uint64
i64 = math.Float64bits(p.From.U.Dval)
literal = fmt.Sprintf("$f64.%016x", i64)
s = liblink.Linklookup(ctxt, literal, 0)
if s.Type_ == 0 {
s.Type_ = liblink.SRODATA
liblink.Adduint64(ctxt, s, i64)
s.Reachable = 0
}
p.From.Type_ = D_EXTERN
p.From.Sym = s
p.From.Offset = 0
}
break
}
}