go/src/cmd/link/internal/x86/asm.go

770 lines
20 KiB
Go
Raw Normal View History

// Inferno utils/8l/asm.c
// https://bitbucket.org/inferno-os/inferno-os/src/default/utils/8l/asm.c
//
// Copyright © 1994-1999 Lucent Technologies Inc. All rights reserved.
// Portions Copyright © 1995-1997 C H Forsyth (forsyth@terzarima.net)
// Portions Copyright © 1997-1999 Vita Nuova Limited
// Portions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com)
// Portions Copyright © 2004,2006 Bruce Ellis
// Portions Copyright © 2005-2007 C H Forsyth (forsyth@terzarima.net)
// Revisions Copyright © 2000-2007 Lucent Technologies Inc. and others
// Portions Copyright © 2009 The Go Authors. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
package x86
import (
"cmd/internal/objabi"
"cmd/internal/sys"
"cmd/link/internal/ld"
"cmd/link/internal/sym"
"debug/elf"
"log"
)
// Append 4 bytes to s and create a R_CALL relocation targeting t to fill them in.
func addcall(ctxt *ld.Link, s *sym.Symbol, t *sym.Symbol) {
s.Attr |= sym.AttrReachable
i := s.Size
s.Size += 4
s.Grow(s.Size)
r := s.AddRel()
r.Sym = t
r.Off = int32(i)
r.Type = objabi.R_CALL
r.Siz = 4
}
func gentext(ctxt *ld.Link) {
if ctxt.DynlinkingGo() {
// We need get_pc_thunk.
} else {
switch ctxt.BuildMode {
case ld.BuildModeCArchive:
if !ctxt.IsELF {
return
}
case ld.BuildModePIE, ld.BuildModeCShared, ld.BuildModePlugin:
// We need get_pc_thunk.
default:
return
}
}
[dev.ssa] cmd/compile: fix PIC for SSA-generated code Access to globals requires a 2-instruction sequence on PIC 386. MOVL foo(SB), AX is translated by the obj package into: CALL getPCofNextInstructionInTempRegister(SB) MOVL (&foo-&thisInstruction)(tmpReg), AX The call returns the PC of the next instruction in a register. The next instruction then offsets from that register to get the address required. The tricky part is the allocation of the temp register. The legacy compiler always used CX, and forbid the register allocator from allocating CX when in PIC mode. We can't easily do that in SSA because CX is actually a required register for shift instructions. (I think the old backend got away with this because the register allocator never uses CX, only codegen knows that shifts must use CX.) Instead, we allow the temp register to be anything. When the destination of the MOV (or LEA) is an integer register, we can use that register. Otherwise, we make sure to compile the operation using an LEA to reference the global. So MOVL AX, foo(SB) is never generated directly. Instead, SSA generates: LEAL foo(SB), DX MOVL AX, (DX) which is then rewritten by the obj package to: CALL getPcInDX(SB) LEAL (&foo-&thisInstruction)(DX), AX MOVL AX, (DX) So this CL modifies the obj package to use different thunks to materialize the pc into different registers. We use the registers that regalloc chose so that SSA can still allocate the full set of registers. Change-Id: Ie095644f7164a026c62e95baf9d18a8bcaed0bba Reviewed-on: https://go-review.googlesource.com/25442 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-08-03 13:00:49 -07:00
// Generate little thunks that load the PC of the next instruction into a register.
thunks := make([]*sym.Symbol, 0, 7+len(ctxt.Textp))
[dev.ssa] cmd/compile: fix PIC for SSA-generated code Access to globals requires a 2-instruction sequence on PIC 386. MOVL foo(SB), AX is translated by the obj package into: CALL getPCofNextInstructionInTempRegister(SB) MOVL (&foo-&thisInstruction)(tmpReg), AX The call returns the PC of the next instruction in a register. The next instruction then offsets from that register to get the address required. The tricky part is the allocation of the temp register. The legacy compiler always used CX, and forbid the register allocator from allocating CX when in PIC mode. We can't easily do that in SSA because CX is actually a required register for shift instructions. (I think the old backend got away with this because the register allocator never uses CX, only codegen knows that shifts must use CX.) Instead, we allow the temp register to be anything. When the destination of the MOV (or LEA) is an integer register, we can use that register. Otherwise, we make sure to compile the operation using an LEA to reference the global. So MOVL AX, foo(SB) is never generated directly. Instead, SSA generates: LEAL foo(SB), DX MOVL AX, (DX) which is then rewritten by the obj package to: CALL getPcInDX(SB) LEAL (&foo-&thisInstruction)(DX), AX MOVL AX, (DX) So this CL modifies the obj package to use different thunks to materialize the pc into different registers. We use the registers that regalloc chose so that SSA can still allocate the full set of registers. Change-Id: Ie095644f7164a026c62e95baf9d18a8bcaed0bba Reviewed-on: https://go-review.googlesource.com/25442 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-08-03 13:00:49 -07:00
for _, r := range [...]struct {
name string
num uint8
}{
{"ax", 0},
{"cx", 1},
{"dx", 2},
{"bx", 3},
// sp
{"bp", 5},
{"si", 6},
{"di", 7},
} {
thunkfunc := ctxt.Syms.Lookup("__x86.get_pc_thunk."+r.name, 0)
thunkfunc.Type = sym.STEXT
thunkfunc.Attr |= sym.AttrLocal
thunkfunc.Attr |= sym.AttrReachable //TODO: remove?
[dev.ssa] cmd/compile: fix PIC for SSA-generated code Access to globals requires a 2-instruction sequence on PIC 386. MOVL foo(SB), AX is translated by the obj package into: CALL getPCofNextInstructionInTempRegister(SB) MOVL (&foo-&thisInstruction)(tmpReg), AX The call returns the PC of the next instruction in a register. The next instruction then offsets from that register to get the address required. The tricky part is the allocation of the temp register. The legacy compiler always used CX, and forbid the register allocator from allocating CX when in PIC mode. We can't easily do that in SSA because CX is actually a required register for shift instructions. (I think the old backend got away with this because the register allocator never uses CX, only codegen knows that shifts must use CX.) Instead, we allow the temp register to be anything. When the destination of the MOV (or LEA) is an integer register, we can use that register. Otherwise, we make sure to compile the operation using an LEA to reference the global. So MOVL AX, foo(SB) is never generated directly. Instead, SSA generates: LEAL foo(SB), DX MOVL AX, (DX) which is then rewritten by the obj package to: CALL getPcInDX(SB) LEAL (&foo-&thisInstruction)(DX), AX MOVL AX, (DX) So this CL modifies the obj package to use different thunks to materialize the pc into different registers. We use the registers that regalloc chose so that SSA can still allocate the full set of registers. Change-Id: Ie095644f7164a026c62e95baf9d18a8bcaed0bba Reviewed-on: https://go-review.googlesource.com/25442 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-08-03 13:00:49 -07:00
o := func(op ...uint8) {
for _, op1 := range op {
thunkfunc.AddUint8(op1)
[dev.ssa] cmd/compile: fix PIC for SSA-generated code Access to globals requires a 2-instruction sequence on PIC 386. MOVL foo(SB), AX is translated by the obj package into: CALL getPCofNextInstructionInTempRegister(SB) MOVL (&foo-&thisInstruction)(tmpReg), AX The call returns the PC of the next instruction in a register. The next instruction then offsets from that register to get the address required. The tricky part is the allocation of the temp register. The legacy compiler always used CX, and forbid the register allocator from allocating CX when in PIC mode. We can't easily do that in SSA because CX is actually a required register for shift instructions. (I think the old backend got away with this because the register allocator never uses CX, only codegen knows that shifts must use CX.) Instead, we allow the temp register to be anything. When the destination of the MOV (or LEA) is an integer register, we can use that register. Otherwise, we make sure to compile the operation using an LEA to reference the global. So MOVL AX, foo(SB) is never generated directly. Instead, SSA generates: LEAL foo(SB), DX MOVL AX, (DX) which is then rewritten by the obj package to: CALL getPcInDX(SB) LEAL (&foo-&thisInstruction)(DX), AX MOVL AX, (DX) So this CL modifies the obj package to use different thunks to materialize the pc into different registers. We use the registers that regalloc chose so that SSA can still allocate the full set of registers. Change-Id: Ie095644f7164a026c62e95baf9d18a8bcaed0bba Reviewed-on: https://go-review.googlesource.com/25442 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-08-03 13:00:49 -07:00
}
}
[dev.ssa] cmd/compile: fix PIC for SSA-generated code Access to globals requires a 2-instruction sequence on PIC 386. MOVL foo(SB), AX is translated by the obj package into: CALL getPCofNextInstructionInTempRegister(SB) MOVL (&foo-&thisInstruction)(tmpReg), AX The call returns the PC of the next instruction in a register. The next instruction then offsets from that register to get the address required. The tricky part is the allocation of the temp register. The legacy compiler always used CX, and forbid the register allocator from allocating CX when in PIC mode. We can't easily do that in SSA because CX is actually a required register for shift instructions. (I think the old backend got away with this because the register allocator never uses CX, only codegen knows that shifts must use CX.) Instead, we allow the temp register to be anything. When the destination of the MOV (or LEA) is an integer register, we can use that register. Otherwise, we make sure to compile the operation using an LEA to reference the global. So MOVL AX, foo(SB) is never generated directly. Instead, SSA generates: LEAL foo(SB), DX MOVL AX, (DX) which is then rewritten by the obj package to: CALL getPcInDX(SB) LEAL (&foo-&thisInstruction)(DX), AX MOVL AX, (DX) So this CL modifies the obj package to use different thunks to materialize the pc into different registers. We use the registers that regalloc chose so that SSA can still allocate the full set of registers. Change-Id: Ie095644f7164a026c62e95baf9d18a8bcaed0bba Reviewed-on: https://go-review.googlesource.com/25442 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-08-03 13:00:49 -07:00
// 8b 04 24 mov (%esp),%eax
// Destination register is in bits 3-5 of the middle byte, so add that in.
o(0x8b, 0x04+r.num<<3, 0x24)
// c3 ret
o(0xc3)
cmd/link: insert trampolines for too-far jumps on ARM ARM direct CALL/JMP instruction has 24 bit offset, which can only encodes jumps within +/-32M. When the target is too far, the top bits get truncated and the program jumps wild. This CL detects too-far jumps and automatically insert trampolines, currently only internal linking on ARM. It is necessary to make the following changes to the linker: - Resolve direct jump relocs when assigning addresses to functions. this allows trampoline insertion without moving all code that already laid down. - Lay down packages in dependency order, so that when resolving a inter-package direct jump reloc, the target address is already known. Intra-package jumps are assumed never too far. - a linker flag -debugtramp is added for debugging trampolines: "-debugtramp=1 -v" prints trampoline debug message "-debugtramp=2" forces all inter-package jump to use trampolines (currently ARM only) "-debugtramp=2 -v" does both - Some data structures are changed for bookkeeping. On ARM, pseudo DIV/DIVU/MOD/MODU instructions now clobber R8 (unfortunate). In the standard library there is no ARM assembly code that uses these instructions, and the compiler no longer emits them (CL 29390). all.bash passes with -debugtramp=2, except a disassembly test (this is unavoidable as we changed the instruction). TBD: debug info of trampolines? Fixes #17028. Change-Id: Idcce347ea7e0af77c4079041a160b2f6e114b474 Reviewed-on: https://go-review.googlesource.com/29397 Reviewed-by: David Crawshaw <crawshaw@golang.org> Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-09-14 14:47:12 -04:00
thunks = append(thunks, thunkfunc)
[dev.ssa] cmd/compile: fix PIC for SSA-generated code Access to globals requires a 2-instruction sequence on PIC 386. MOVL foo(SB), AX is translated by the obj package into: CALL getPCofNextInstructionInTempRegister(SB) MOVL (&foo-&thisInstruction)(tmpReg), AX The call returns the PC of the next instruction in a register. The next instruction then offsets from that register to get the address required. The tricky part is the allocation of the temp register. The legacy compiler always used CX, and forbid the register allocator from allocating CX when in PIC mode. We can't easily do that in SSA because CX is actually a required register for shift instructions. (I think the old backend got away with this because the register allocator never uses CX, only codegen knows that shifts must use CX.) Instead, we allow the temp register to be anything. When the destination of the MOV (or LEA) is an integer register, we can use that register. Otherwise, we make sure to compile the operation using an LEA to reference the global. So MOVL AX, foo(SB) is never generated directly. Instead, SSA generates: LEAL foo(SB), DX MOVL AX, (DX) which is then rewritten by the obj package to: CALL getPcInDX(SB) LEAL (&foo-&thisInstruction)(DX), AX MOVL AX, (DX) So this CL modifies the obj package to use different thunks to materialize the pc into different registers. We use the registers that regalloc chose so that SSA can still allocate the full set of registers. Change-Id: Ie095644f7164a026c62e95baf9d18a8bcaed0bba Reviewed-on: https://go-review.googlesource.com/25442 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-08-03 13:00:49 -07:00
}
cmd/link: insert trampolines for too-far jumps on ARM ARM direct CALL/JMP instruction has 24 bit offset, which can only encodes jumps within +/-32M. When the target is too far, the top bits get truncated and the program jumps wild. This CL detects too-far jumps and automatically insert trampolines, currently only internal linking on ARM. It is necessary to make the following changes to the linker: - Resolve direct jump relocs when assigning addresses to functions. this allows trampoline insertion without moving all code that already laid down. - Lay down packages in dependency order, so that when resolving a inter-package direct jump reloc, the target address is already known. Intra-package jumps are assumed never too far. - a linker flag -debugtramp is added for debugging trampolines: "-debugtramp=1 -v" prints trampoline debug message "-debugtramp=2" forces all inter-package jump to use trampolines (currently ARM only) "-debugtramp=2 -v" does both - Some data structures are changed for bookkeeping. On ARM, pseudo DIV/DIVU/MOD/MODU instructions now clobber R8 (unfortunate). In the standard library there is no ARM assembly code that uses these instructions, and the compiler no longer emits them (CL 29390). all.bash passes with -debugtramp=2, except a disassembly test (this is unavoidable as we changed the instruction). TBD: debug info of trampolines? Fixes #17028. Change-Id: Idcce347ea7e0af77c4079041a160b2f6e114b474 Reviewed-on: https://go-review.googlesource.com/29397 Reviewed-by: David Crawshaw <crawshaw@golang.org> Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-09-14 14:47:12 -04:00
ctxt.Textp = append(thunks, ctxt.Textp...) // keep Textp in dependency order
addmoduledata := ctxt.Syms.Lookup("runtime.addmoduledata", 0)
if addmoduledata.Type == sym.STEXT && ctxt.BuildMode != ld.BuildModePlugin {
// we're linking a module containing the runtime -> no need for
// an init function
return
}
addmoduledata.Attr |= sym.AttrReachable
initfunc := ctxt.Syms.Lookup("go.link.addmoduledata", 0)
initfunc.Type = sym.STEXT
initfunc.Attr |= sym.AttrLocal
initfunc.Attr |= sym.AttrReachable
[dev.ssa] cmd/compile: fix PIC for SSA-generated code Access to globals requires a 2-instruction sequence on PIC 386. MOVL foo(SB), AX is translated by the obj package into: CALL getPCofNextInstructionInTempRegister(SB) MOVL (&foo-&thisInstruction)(tmpReg), AX The call returns the PC of the next instruction in a register. The next instruction then offsets from that register to get the address required. The tricky part is the allocation of the temp register. The legacy compiler always used CX, and forbid the register allocator from allocating CX when in PIC mode. We can't easily do that in SSA because CX is actually a required register for shift instructions. (I think the old backend got away with this because the register allocator never uses CX, only codegen knows that shifts must use CX.) Instead, we allow the temp register to be anything. When the destination of the MOV (or LEA) is an integer register, we can use that register. Otherwise, we make sure to compile the operation using an LEA to reference the global. So MOVL AX, foo(SB) is never generated directly. Instead, SSA generates: LEAL foo(SB), DX MOVL AX, (DX) which is then rewritten by the obj package to: CALL getPcInDX(SB) LEAL (&foo-&thisInstruction)(DX), AX MOVL AX, (DX) So this CL modifies the obj package to use different thunks to materialize the pc into different registers. We use the registers that regalloc chose so that SSA can still allocate the full set of registers. Change-Id: Ie095644f7164a026c62e95baf9d18a8bcaed0bba Reviewed-on: https://go-review.googlesource.com/25442 Run-TryBot: Keith Randall <khr@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: David Chase <drchase@google.com>
2016-08-03 13:00:49 -07:00
o := func(op ...uint8) {
for _, op1 := range op {
initfunc.AddUint8(op1)
}
}
// go.link.addmoduledata:
// 53 push %ebx
// e8 00 00 00 00 call __x86.get_pc_thunk.cx + R_CALL __x86.get_pc_thunk.cx
// 8d 81 00 00 00 00 lea 0x0(%ecx), %eax + R_PCREL ctxt.Moduledata
// 8d 99 00 00 00 00 lea 0x0(%ecx), %ebx + R_GOTPC _GLOBAL_OFFSET_TABLE_
// e8 00 00 00 00 call runtime.addmoduledata@plt + R_CALL runtime.addmoduledata
// 5b pop %ebx
// c3 ret
o(0x53)
o(0xe8)
addcall(ctxt, initfunc, ctxt.Syms.Lookup("__x86.get_pc_thunk.cx", 0))
o(0x8d, 0x81)
initfunc.AddPCRelPlus(ctxt.Arch, ctxt.Moduledata, 6)
o(0x8d, 0x99)
i := initfunc.Size
initfunc.Size += 4
initfunc.Grow(initfunc.Size)
r := initfunc.AddRel()
r.Sym = ctxt.Syms.Lookup("_GLOBAL_OFFSET_TABLE_", 0)
r.Off = int32(i)
r.Type = objabi.R_PCREL
r.Add = 12
r.Siz = 4
o(0xe8)
addcall(ctxt, initfunc, addmoduledata)
o(0x5b)
o(0xc3)
if ctxt.BuildMode == ld.BuildModePlugin {
ctxt.Textp = append(ctxt.Textp, addmoduledata)
}
cmd/link: insert trampolines for too-far jumps on ARM ARM direct CALL/JMP instruction has 24 bit offset, which can only encodes jumps within +/-32M. When the target is too far, the top bits get truncated and the program jumps wild. This CL detects too-far jumps and automatically insert trampolines, currently only internal linking on ARM. It is necessary to make the following changes to the linker: - Resolve direct jump relocs when assigning addresses to functions. this allows trampoline insertion without moving all code that already laid down. - Lay down packages in dependency order, so that when resolving a inter-package direct jump reloc, the target address is already known. Intra-package jumps are assumed never too far. - a linker flag -debugtramp is added for debugging trampolines: "-debugtramp=1 -v" prints trampoline debug message "-debugtramp=2" forces all inter-package jump to use trampolines (currently ARM only) "-debugtramp=2 -v" does both - Some data structures are changed for bookkeeping. On ARM, pseudo DIV/DIVU/MOD/MODU instructions now clobber R8 (unfortunate). In the standard library there is no ARM assembly code that uses these instructions, and the compiler no longer emits them (CL 29390). all.bash passes with -debugtramp=2, except a disassembly test (this is unavoidable as we changed the instruction). TBD: debug info of trampolines? Fixes #17028. Change-Id: Idcce347ea7e0af77c4079041a160b2f6e114b474 Reviewed-on: https://go-review.googlesource.com/29397 Reviewed-by: David Crawshaw <crawshaw@golang.org> Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-09-14 14:47:12 -04:00
ctxt.Textp = append(ctxt.Textp, initfunc)
initarray_entry := ctxt.Syms.Lookup("go.link.addmoduledatainit", 0)
initarray_entry.Attr |= sym.AttrReachable
initarray_entry.Attr |= sym.AttrLocal
initarray_entry.Type = sym.SINITARR
initarray_entry.AddAddr(ctxt.Arch, initfunc)
}
func adddynrel(ctxt *ld.Link, s *sym.Symbol, r *sym.Reloc) bool {
targ := r.Sym
switch r.Type {
default:
if r.Type >= 256 {
ld.Errorf(s, "unexpected relocation type %d (%s)", r.Type, sym.RelocName(ctxt.Arch, r.Type))
return false
}
// Handle relocations found in ELF object files.
case 256 + objabi.RelocType(elf.R_386_PC32):
if targ.Type == sym.SDYNIMPORT {
ld.Errorf(s, "unexpected R_386_PC32 relocation for dynamic symbol %s", targ.Name)
}
// TODO(mwhudson): the test of VisibilityHidden here probably doesn't make
// sense and should be removed when someone has thought about it properly.
if (targ.Type == 0 || targ.Type == sym.SXREF) && !targ.Attr.VisibilityHidden() {
ld.Errorf(s, "unknown symbol %s in pcrel", targ.Name)
}
r.Type = objabi.R_PCREL
r.Add += 4
return true
case 256 + objabi.RelocType(elf.R_386_PLT32):
r.Type = objabi.R_PCREL
r.Add += 4
if targ.Type == sym.SDYNIMPORT {
addpltsym(ctxt, targ)
r.Sym = ctxt.Syms.Lookup(".plt", 0)
r.Add += int64(targ.Plt())
}
return true
case 256 + objabi.RelocType(elf.R_386_GOT32), 256 + objabi.RelocType(elf.R_386_GOT32X):
if targ.Type != sym.SDYNIMPORT {
// have symbol
if r.Off >= 2 && s.P[r.Off-2] == 0x8b {
// turn MOVL of GOT entry into LEAL of symbol address, relative to GOT.
s.P[r.Off-2] = 0x8d
r.Type = objabi.R_GOTOFF
return true
}
if r.Off >= 2 && s.P[r.Off-2] == 0xff && s.P[r.Off-1] == 0xb3 {
// turn PUSHL of GOT entry into PUSHL of symbol itself.
// use unnecessary SS prefix to keep instruction same length.
s.P[r.Off-2] = 0x36
s.P[r.Off-1] = 0x68
r.Type = objabi.R_ADDR
return true
}
ld.Errorf(s, "unexpected GOT reloc for non-dynamic symbol %s", targ.Name)
return false
}
addgotsym(ctxt, targ)
r.Type = objabi.R_CONST // write r->add during relocsym
r.Sym = nil
r.Add += int64(targ.Got())
return true
case 256 + objabi.RelocType(elf.R_386_GOTOFF):
r.Type = objabi.R_GOTOFF
return true
case 256 + objabi.RelocType(elf.R_386_GOTPC):
r.Type = objabi.R_PCREL
r.Sym = ctxt.Syms.Lookup(".got", 0)
r.Add += 4
return true
case 256 + objabi.RelocType(elf.R_386_32):
if targ.Type == sym.SDYNIMPORT {
ld.Errorf(s, "unexpected R_386_32 relocation for dynamic symbol %s", targ.Name)
}
r.Type = objabi.R_ADDR
return true
case 512 + ld.MACHO_GENERIC_RELOC_VANILLA*2 + 0:
r.Type = objabi.R_ADDR
if targ.Type == sym.SDYNIMPORT {
ld.Errorf(s, "unexpected reloc for dynamic symbol %s", targ.Name)
}
return true
case 512 + ld.MACHO_GENERIC_RELOC_VANILLA*2 + 1:
if targ.Type == sym.SDYNIMPORT {
addpltsym(ctxt, targ)
r.Sym = ctxt.Syms.Lookup(".plt", 0)
r.Add = int64(targ.Plt())
r.Type = objabi.R_PCREL
return true
}
r.Type = objabi.R_PCREL
return true
case 512 + ld.MACHO_FAKE_GOTPCREL:
if targ.Type != sym.SDYNIMPORT {
// have symbol
// turn MOVL of GOT entry into LEAL of symbol itself
if r.Off < 2 || s.P[r.Off-2] != 0x8b {
ld.Errorf(s, "unexpected GOT reloc for non-dynamic symbol %s", targ.Name)
return false
}
s.P[r.Off-2] = 0x8d
r.Type = objabi.R_PCREL
return true
}
addgotsym(ctxt, targ)
r.Sym = ctxt.Syms.Lookup(".got", 0)
r.Add += int64(targ.Got())
r.Type = objabi.R_PCREL
return true
}
// Handle references to ELF symbols from our own object files.
if targ.Type != sym.SDYNIMPORT {
return true
}
switch r.Type {
case objabi.R_CALL,
objabi.R_PCREL:
if ctxt.LinkMode == ld.LinkExternal {
// External linker will do this relocation.
return true
}
addpltsym(ctxt, targ)
r.Sym = ctxt.Syms.Lookup(".plt", 0)
r.Add = int64(targ.Plt())
return true
case objabi.R_ADDR:
if s.Type != sym.SDATA {
break
}
if ctxt.IsELF {
ld.Adddynsym(ctxt, targ)
rel := ctxt.Syms.Lookup(".rel", 0)
rel.AddAddrPlus(ctxt.Arch, s, int64(r.Off))
rel.AddUint32(ctxt.Arch, ld.ELF32_R_INFO(uint32(targ.Dynid), uint32(elf.R_386_32)))
r.Type = objabi.R_CONST // write r->add during relocsym
r.Sym = nil
return true
}
if ctxt.HeadType == objabi.Hdarwin && s.Size == int64(ctxt.Arch.PtrSize) && r.Off == 0 {
// Mach-O relocations are a royal pain to lay out.
// They use a compact stateful bytecode representation
// that is too much bother to deal with.
// Instead, interpret the C declaration
// void *_Cvar_stderr = &stderr;
// as making _Cvar_stderr the name of a GOT entry
// for stderr. This is separate from the usual GOT entry,
// just in case the C code assigns to the variable,
// and of course it only works for single pointers,
// but we only need to support cgo and that's all it needs.
ld.Adddynsym(ctxt, targ)
got := ctxt.Syms.Lookup(".got", 0)
s.Type = got.Type
s.Attr |= sym.AttrSubSymbol
s.Outer = got
s.Sub = got.Sub
got.Sub = s
s.Value = got.Size
got.AddUint32(ctxt.Arch, 0)
ctxt.Syms.Lookup(".linkedit.got", 0).AddUint32(ctxt.Arch, uint32(targ.Dynid))
r.Type = 256 // ignore during relocsym
return true
}
}
return false
}
func elfreloc1(ctxt *ld.Link, r *sym.Reloc, sectoff int64) bool {
ctxt.Out.Write32(uint32(sectoff))
elfsym := r.Xsym.ElfsymForReloc()
switch r.Type {
default:
return false
case objabi.R_ADDR:
if r.Siz == 4 {
ctxt.Out.Write32(uint32(elf.R_386_32) | uint32(elfsym)<<8)
} else {
return false
}
case objabi.R_GOTPCREL:
if r.Siz == 4 {
ctxt.Out.Write32(uint32(elf.R_386_GOTPC))
if r.Xsym.Name != "_GLOBAL_OFFSET_TABLE_" {
ctxt.Out.Write32(uint32(sectoff))
ctxt.Out.Write32(uint32(elf.R_386_GOT32) | uint32(elfsym)<<8)
}
} else {
return false
}
case objabi.R_CALL:
if r.Siz == 4 {
if r.Xsym.Type == sym.SDYNIMPORT {
ctxt.Out.Write32(uint32(elf.R_386_PLT32) | uint32(elfsym)<<8)
} else {
ctxt.Out.Write32(uint32(elf.R_386_PC32) | uint32(elfsym)<<8)
}
} else {
return false
}
case objabi.R_PCREL:
if r.Siz == 4 {
ctxt.Out.Write32(uint32(elf.R_386_PC32) | uint32(elfsym)<<8)
} else {
return false
}
case objabi.R_TLS_LE:
if r.Siz == 4 {
ctxt.Out.Write32(uint32(elf.R_386_TLS_LE) | uint32(elfsym)<<8)
} else {
return false
}
case objabi.R_TLS_IE:
if r.Siz == 4 {
ctxt.Out.Write32(uint32(elf.R_386_GOTPC))
ctxt.Out.Write32(uint32(sectoff))
ctxt.Out.Write32(uint32(elf.R_386_TLS_GOTIE) | uint32(elfsym)<<8)
} else {
return false
}
}
return true
}
func machoreloc1(arch *sys.Arch, out *ld.OutBuf, s *sym.Symbol, r *sym.Reloc, sectoff int64) bool {
var v uint32
rs := r.Xsym
if rs.Type == sym.SHOSTOBJ || r.Type == objabi.R_CALL {
if rs.Dynid < 0 {
ld.Errorf(s, "reloc %d (%s) to non-macho symbol %s type=%d (%s)", r.Type, sym.RelocName(arch, r.Type), rs.Name, rs.Type, rs.Type)
return false
}
v = uint32(rs.Dynid)
v |= 1 << 27 // external relocation
} else {
v = uint32(rs.Sect.Extnum)
if v == 0 {
ld.Errorf(s, "reloc %d (%s) to symbol %s in non-macho section %s type=%d (%s)", r.Type, sym.RelocName(arch, r.Type), rs.Name, rs.Sect.Name, rs.Type, rs.Type)
return false
}
}
switch r.Type {
default:
return false
case objabi.R_ADDR:
v |= ld.MACHO_GENERIC_RELOC_VANILLA << 28
case objabi.R_CALL,
objabi.R_PCREL:
v |= 1 << 24 // pc-relative bit
v |= ld.MACHO_GENERIC_RELOC_VANILLA << 28
}
switch r.Siz {
default:
return false
case 1:
v |= 0 << 25
case 2:
v |= 1 << 25
case 4:
v |= 2 << 25
case 8:
v |= 3 << 25
}
out.Write32(uint32(sectoff))
out.Write32(v)
return true
}
func pereloc1(arch *sys.Arch, out *ld.OutBuf, s *sym.Symbol, r *sym.Reloc, sectoff int64) bool {
var v uint32
rs := r.Xsym
if rs.Dynid < 0 {
ld.Errorf(s, "reloc %d (%s) to non-coff symbol %s type=%d (%s)", r.Type, sym.RelocName(arch, r.Type), rs.Name, rs.Type, rs.Type)
return false
}
out.Write32(uint32(sectoff))
out.Write32(uint32(rs.Dynid))
switch r.Type {
default:
return false
case objabi.R_DWARFSECREF:
v = ld.IMAGE_REL_I386_SECREL
case objabi.R_ADDR:
v = ld.IMAGE_REL_I386_DIR32
case objabi.R_CALL,
objabi.R_PCREL:
v = ld.IMAGE_REL_I386_REL32
}
out.Write16(uint16(v))
return true
}
func archreloc(ctxt *ld.Link, r *sym.Reloc, s *sym.Symbol, val int64) (int64, bool) {
if ctxt.LinkMode == ld.LinkExternal {
return val, false
}
switch r.Type {
case objabi.R_CONST:
return r.Add, true
case objabi.R_GOTOFF:
return ld.Symaddr(r.Sym) + r.Add - ld.Symaddr(ctxt.Syms.Lookup(".got", 0)), true
}
return val, false
}
func archrelocvariant(ctxt *ld.Link, r *sym.Reloc, s *sym.Symbol, t int64) int64 {
log.Fatalf("unexpected relocation variant")
return t
}
func elfsetupplt(ctxt *ld.Link) {
plt := ctxt.Syms.Lookup(".plt", 0)
got := ctxt.Syms.Lookup(".got.plt", 0)
if plt.Size == 0 {
// pushl got+4
plt.AddUint8(0xff)
plt.AddUint8(0x35)
plt.AddAddrPlus(ctxt.Arch, got, 4)
// jmp *got+8
plt.AddUint8(0xff)
plt.AddUint8(0x25)
plt.AddAddrPlus(ctxt.Arch, got, 8)
// zero pad
plt.AddUint32(ctxt.Arch, 0)
// assume got->size == 0 too
got.AddAddrPlus(ctxt.Arch, ctxt.Syms.Lookup(".dynamic", 0), 0)
got.AddUint32(ctxt.Arch, 0)
got.AddUint32(ctxt.Arch, 0)
}
}
func addpltsym(ctxt *ld.Link, s *sym.Symbol) {
if s.Plt() >= 0 {
return
}
ld.Adddynsym(ctxt, s)
if ctxt.IsELF {
plt := ctxt.Syms.Lookup(".plt", 0)
got := ctxt.Syms.Lookup(".got.plt", 0)
rel := ctxt.Syms.Lookup(".rel.plt", 0)
if plt.Size == 0 {
elfsetupplt(ctxt)
}
// jmpq *got+size
plt.AddUint8(0xff)
plt.AddUint8(0x25)
plt.AddAddrPlus(ctxt.Arch, got, got.Size)
// add to got: pointer to current pos in plt
got.AddAddrPlus(ctxt.Arch, plt, plt.Size)
// pushl $x
plt.AddUint8(0x68)
plt.AddUint32(ctxt.Arch, uint32(rel.Size))
// jmp .plt
plt.AddUint8(0xe9)
plt.AddUint32(ctxt.Arch, uint32(-(plt.Size + 4)))
// rel
rel.AddAddrPlus(ctxt.Arch, got, got.Size-4)
rel.AddUint32(ctxt.Arch, ld.ELF32_R_INFO(uint32(s.Dynid), uint32(elf.R_386_JMP_SLOT)))
s.SetPlt(int32(plt.Size - 16))
} else if ctxt.HeadType == objabi.Hdarwin {
// Same laziness as in 6l.
plt := ctxt.Syms.Lookup(".plt", 0)
addgotsym(ctxt, s)
ctxt.Syms.Lookup(".linkedit.plt", 0).AddUint32(ctxt.Arch, uint32(s.Dynid))
// jmpq *got+size(IP)
s.SetPlt(int32(plt.Size))
plt.AddUint8(0xff)
plt.AddUint8(0x25)
plt.AddAddrPlus(ctxt.Arch, ctxt.Syms.Lookup(".got", 0), int64(s.Got()))
} else {
ld.Errorf(s, "addpltsym: unsupported binary format")
}
}
func addgotsym(ctxt *ld.Link, s *sym.Symbol) {
if s.Got() >= 0 {
return
}
ld.Adddynsym(ctxt, s)
got := ctxt.Syms.Lookup(".got", 0)
s.SetGot(int32(got.Size))
got.AddUint32(ctxt.Arch, 0)
if ctxt.IsELF {
rel := ctxt.Syms.Lookup(".rel", 0)
rel.AddAddrPlus(ctxt.Arch, got, int64(s.Got()))
rel.AddUint32(ctxt.Arch, ld.ELF32_R_INFO(uint32(s.Dynid), uint32(elf.R_386_GLOB_DAT)))
} else if ctxt.HeadType == objabi.Hdarwin {
ctxt.Syms.Lookup(".linkedit.got", 0).AddUint32(ctxt.Arch, uint32(s.Dynid))
} else {
ld.Errorf(s, "addgotsym: unsupported binary format")
}
}
func asmb(ctxt *ld.Link) {
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f asmb\n", ld.Cputime())
}
if ctxt.IsELF {
ld.Asmbelfsetup()
}
sect := ld.Segtext.Sections[0]
ctxt.Out.SeekSet(int64(sect.Vaddr - ld.Segtext.Vaddr + ld.Segtext.Fileoff))
// 0xCC is INT $3 - breakpoint instruction
ld.CodeblkPad(ctxt, int64(sect.Vaddr), int64(sect.Length), []byte{0xCC})
for _, sect = range ld.Segtext.Sections[1:] {
ctxt.Out.SeekSet(int64(sect.Vaddr - ld.Segtext.Vaddr + ld.Segtext.Fileoff))
ld.Datblk(ctxt, int64(sect.Vaddr), int64(sect.Length))
}
if ld.Segrodata.Filelen > 0 {
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f rodatblk\n", ld.Cputime())
}
ctxt.Out.SeekSet(int64(ld.Segrodata.Fileoff))
ld.Datblk(ctxt, int64(ld.Segrodata.Vaddr), int64(ld.Segrodata.Filelen))
}
if ld.Segrelrodata.Filelen > 0 {
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f relrodatblk\n", ld.Cputime())
}
ctxt.Out.SeekSet(int64(ld.Segrelrodata.Fileoff))
ld.Datblk(ctxt, int64(ld.Segrelrodata.Vaddr), int64(ld.Segrelrodata.Filelen))
}
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f datblk\n", ld.Cputime())
}
ctxt.Out.SeekSet(int64(ld.Segdata.Fileoff))
ld.Datblk(ctxt, int64(ld.Segdata.Vaddr), int64(ld.Segdata.Filelen))
ctxt.Out.SeekSet(int64(ld.Segdwarf.Fileoff))
ld.Dwarfblk(ctxt, int64(ld.Segdwarf.Vaddr), int64(ld.Segdwarf.Filelen))
machlink := uint32(0)
if ctxt.HeadType == objabi.Hdarwin {
machlink = uint32(ld.Domacholink(ctxt))
}
ld.Symsize = 0
ld.Spsize = 0
ld.Lcsize = 0
symo := uint32(0)
if !*ld.FlagS {
// TODO: rationalize
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f sym\n", ld.Cputime())
}
switch ctxt.HeadType {
default:
if ctxt.IsELF {
symo = uint32(ld.Segdwarf.Fileoff + ld.Segdwarf.Filelen)
symo = uint32(ld.Rnd(int64(symo), int64(*ld.FlagRound)))
}
case objabi.Hplan9:
symo = uint32(ld.Segdata.Fileoff + ld.Segdata.Filelen)
case objabi.Hdarwin:
symo = uint32(ld.Segdwarf.Fileoff + uint64(ld.Rnd(int64(ld.Segdwarf.Filelen), int64(*ld.FlagRound))) + uint64(machlink))
case objabi.Hwindows:
symo = uint32(ld.Segdwarf.Fileoff + ld.Segdwarf.Filelen)
symo = uint32(ld.Rnd(int64(symo), ld.PEFILEALIGN))
}
ctxt.Out.SeekSet(int64(symo))
switch ctxt.HeadType {
default:
if ctxt.IsELF {
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f elfsym\n", ld.Cputime())
}
ld.Asmelfsym(ctxt)
ctxt.Out.Flush()
ctxt.Out.Write(ld.Elfstrdat)
if ctxt.LinkMode == ld.LinkExternal {
ld.Elfemitreloc(ctxt)
}
}
case objabi.Hplan9:
ld.Asmplan9sym(ctxt)
ctxt.Out.Flush()
sym := ctxt.Syms.Lookup("pclntab", 0)
if sym != nil {
ld.Lcsize = int32(len(sym.P))
ctxt.Out.Write(sym.P)
ctxt.Out.Flush()
}
case objabi.Hwindows:
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f dwarf\n", ld.Cputime())
}
case objabi.Hdarwin:
if ctxt.LinkMode == ld.LinkExternal {
ld.Machoemitreloc(ctxt)
}
}
}
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f headr\n", ld.Cputime())
}
ctxt.Out.SeekSet(0)
switch ctxt.HeadType {
default:
case objabi.Hplan9: /* plan9 */
magic := int32(4*11*11 + 7)
ctxt.Out.Write32b(uint32(magic)) /* magic */
ctxt.Out.Write32b(uint32(ld.Segtext.Filelen)) /* sizes */
ctxt.Out.Write32b(uint32(ld.Segdata.Filelen))
ctxt.Out.Write32b(uint32(ld.Segdata.Length - ld.Segdata.Filelen))
ctxt.Out.Write32b(uint32(ld.Symsize)) /* nsyms */
ctxt.Out.Write32b(uint32(ld.Entryvalue(ctxt))) /* va of entry */
ctxt.Out.Write32b(uint32(ld.Spsize)) /* sp offsets */
ctxt.Out.Write32b(uint32(ld.Lcsize)) /* line offsets */
case objabi.Hdarwin:
ld.Asmbmacho(ctxt)
case objabi.Hlinux,
objabi.Hfreebsd,
objabi.Hnetbsd,
objabi.Hopenbsd,
objabi.Hnacl:
ld.Asmbelf(ctxt, int64(symo))
case objabi.Hwindows:
ld.Asmbpe(ctxt)
}
ctxt.Out.Flush()
}