Stowage/go
2
0
Fork 0
mirror of https://github.com/golang/go.git synced 2025-12-08 06:10:04 +00:00
Code Issues Projects Releases Packages Wiki Activity
go/src/cmd/internal/obj/mips/obj0.go
Cherry Zhang b2482e4817 cmd/internal/obj: mark split-stack prologue nonpreemptible 
When there are both a synchronous preemption request (by
clobbering the stack guard) and an asynchronous one (by signal),
the running goroutine may observe the synchronous request first
in stack bounds check, and go to the path of calling morestack.
If the preemption signal arrives at this point before the call to
morestack, the goroutine will be asynchronously preempted,
entering the scheduler. When it is resumed, the scheduler clears
the preemption request, unclobbers the stack guard. But the
resumed goroutine will still call morestack, as it is already on
its way. morestack will, as there is no preemption request,
double the stack unnecessarily. If this happens multiple times,
the stack may grow too big, although only a small amount is
actually used.

To fix this, we mark the stack bounds check and the call to
morestack async-nonpreemptible, starting after the memory
instruction (mostly a load, on x86 CMP with memory).

Not done for Wasm as it does not support async preemption.

Fixes #35470.

Change-Id: Ibd7f3d935a3649b80f47539116ec9b9556680cf2
Reviewed-on: https://go-review.googlesource.com/c/go/+/207350
Reviewed-by: David Chase <drchase@google.com>
2019-11-27 01:30:19 +00:00

1477 lines
29 KiB
Go
Raw Blame History

// cmd/9l/noop.c, cmd/9l/pass.c, cmd/9l/span.c from Vita Nuova.
//
// 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 © 2000-2008 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 © 2000-2008 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 mips
import (
"cmd/internal/obj"
"cmd/internal/objabi"
"cmd/internal/sys"
"encoding/binary"
"fmt"
"math"
)
func progedit(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) {
c := ctxt0{ctxt: ctxt, newprog: newprog}
p.From.Class = 0
p.To.Class = 0
// Rewrite JMP/JAL to symbol as TYPE_BRANCH.
switch p.As {
case AJMP,
AJAL,
ARET,
obj.ADUFFZERO,
obj.ADUFFCOPY:
if p.To.Sym != nil {
p.To.Type = obj.TYPE_BRANCH
}
}
// Rewrite float constants to values stored in memory.
switch p.As {
case AMOVF:
if p.From.Type == obj.TYPE_FCONST {
f32 := float32(p.From.Val.(float64))
if math.Float32bits(f32) == 0 {
p.As = AMOVW
p.From.Type = obj.TYPE_REG
p.From.Reg = REGZERO
break
}
p.From.Type = obj.TYPE_MEM
p.From.Sym = ctxt.Float32Sym(f32)
p.From.Name = obj.NAME_EXTERN
p.From.Offset = 0
}
case AMOVD:
if p.From.Type == obj.TYPE_FCONST {
f64 := p.From.Val.(float64)
if math.Float64bits(f64) == 0 && c.ctxt.Arch.Family == sys.MIPS64 {
p.As = AMOVV
p.From.Type = obj.TYPE_REG
p.From.Reg = REGZERO
break
}
p.From.Type = obj.TYPE_MEM
p.From.Sym = ctxt.Float64Sym(f64)
p.From.Name = obj.NAME_EXTERN
p.From.Offset = 0
}
// Put >32-bit constants in memory and load them
case AMOVV:
if p.From.Type == obj.TYPE_CONST && p.From.Name == obj.NAME_NONE && p.From.Reg == 0 && int64(int32(p.From.Offset)) != p.From.Offset {
p.From.Type = obj.TYPE_MEM
p.From.Sym = ctxt.Int64Sym(p.From.Offset)
p.From.Name = obj.NAME_EXTERN
p.From.Offset = 0
}
}
// Rewrite SUB constants into ADD.
switch p.As {
case ASUB:
if p.From.Type == obj.TYPE_CONST {
p.From.Offset = -p.From.Offset
p.As = AADD
}
case ASUBU:
if p.From.Type == obj.TYPE_CONST {
p.From.Offset = -p.From.Offset
p.As = AADDU
}
case ASUBV:
if p.From.Type == obj.TYPE_CONST {
p.From.Offset = -p.From.Offset
p.As = AADDV
}
case ASUBVU:
if p.From.Type == obj.TYPE_CONST {
p.From.Offset = -p.From.Offset
p.As = AADDVU
}
}
}
func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
// TODO(minux): add morestack short-cuts with small fixed frame-size.
c := ctxt0{ctxt: ctxt, newprog: newprog, cursym: cursym}
// a switch for enabling/disabling instruction scheduling
nosched := true
if c.cursym.Func.Text == nil || c.cursym.Func.Text.Link == nil {
return
}
p := c.cursym.Func.Text
textstksiz := p.To.Offset
if textstksiz == -ctxt.FixedFrameSize() {
// Historical way to mark NOFRAME.
p.From.Sym.Set(obj.AttrNoFrame, true)
textstksiz = 0
}
if textstksiz < 0 {
c.ctxt.Diag("negative frame size %d - did you mean NOFRAME?", textstksiz)
}
if p.From.Sym.NoFrame() {
if textstksiz != 0 {
c.ctxt.Diag("NOFRAME functions must have a frame size of 0, not %d", textstksiz)
}
}
c.cursym.Func.Args = p.To.Val.(int32)
c.cursym.Func.Locals = int32(textstksiz)
/*
* find leaf subroutines
* strip NOPs
* expand RET
* expand BECOME pseudo
*/
var q *obj.Prog
var q1 *obj.Prog
for p := c.cursym.Func.Text; p != nil; p = p.Link {
switch p.As {
/* too hard, just leave alone */
case obj.ATEXT:
q = p
p.Mark |= LABEL | LEAF | SYNC
if p.Link != nil {
p.Link.Mark |= LABEL
}
/* too hard, just leave alone */
case AMOVW,
AMOVV:
q = p
if p.To.Type == obj.TYPE_REG && p.To.Reg >= REG_SPECIAL {
p.Mark |= LABEL | SYNC
break
}
if p.From.Type == obj.TYPE_REG && p.From.Reg >= REG_SPECIAL {
p.Mark |= LABEL | SYNC
}
/* too hard, just leave alone */
case ASYSCALL,
AWORD,
ATLBWR,
ATLBWI,
ATLBP,
ATLBR:
q = p
p.Mark |= LABEL | SYNC
case ANOR:
q = p
if p.To.Type == obj.TYPE_REG {
if p.To.Reg == REGZERO {
p.Mark |= LABEL | SYNC
}
}
case ABGEZAL,
ABLTZAL,
AJAL,
obj.ADUFFZERO,
obj.ADUFFCOPY:
c.cursym.Func.Text.Mark &^= LEAF
fallthrough
case AJMP,
ABEQ,
ABGEZ,
ABGTZ,
ABLEZ,
ABLTZ,
ABNE,
ABFPT, ABFPF:
if p.As == ABFPT || p.As == ABFPF {
// We don't treat ABFPT and ABFPF as branches here,
// so that we will always fill nop (0x0) in their
// delay slot during assembly.
// This is to workaround a kernel FPU emulator bug
// where it uses the user stack to simulate the
// instruction in the delay slot if it's not 0x0,
// and somehow that leads to SIGSEGV when the kernel
// jump to the stack.
p.Mark |= SYNC
} else {
p.Mark |= BRANCH
}
q = p
q1 = p.Pcond
if q1 != nil {
for q1.As == obj.ANOP {
q1 = q1.Link
p.Pcond = q1
}
if q1.Mark&LEAF == 0 {
q1.Mark |= LABEL
}
}
//else {
// p.Mark |= LABEL
//}
q1 = p.Link
if q1 != nil {
q1.Mark |= LABEL
}
continue
case ARET:
q = p
if p.Link != nil {
p.Link.Mark |= LABEL
}
continue
case obj.ANOP:
q1 = p.Link
q.Link = q1 /* q is non-nop */
q1.Mark |= p.Mark
continue
default:
q = p
continue
}
}
var mov, add obj.As
if c.ctxt.Arch.Family == sys.MIPS64 {
add = AADDV
mov = AMOVV
} else {
add = AADDU
mov = AMOVW
}
autosize := int32(0)
var p1 *obj.Prog
var p2 *obj.Prog
for p := c.cursym.Func.Text; p != nil; p = p.Link {
o := p.As
switch o {
case obj.ATEXT:
autosize = int32(textstksiz)
if p.Mark&LEAF != 0 && autosize == 0 {
// A leaf function with no locals has no frame.
p.From.Sym.Set(obj.AttrNoFrame, true)
}
if !p.From.Sym.NoFrame() {
// If there is a stack frame at all, it includes
// space to save the LR.
autosize += int32(c.ctxt.FixedFrameSize())
}
if autosize&4 != 0 && c.ctxt.Arch.Family == sys.MIPS64 {
autosize += 4
}
if autosize == 0 && c.cursym.Func.Text.Mark&LEAF == 0 {
if c.cursym.Func.Text.From.Sym.NoSplit() {
if ctxt.Debugvlog {
ctxt.Logf("save suppressed in: %s\n", c.cursym.Name)
}
c.cursym.Func.Text.Mark |= LEAF
}
}
p.To.Offset = int64(autosize) - ctxt.FixedFrameSize()
if c.cursym.Func.Text.Mark&LEAF != 0 {
c.cursym.Set(obj.AttrLeaf, true)
if p.From.Sym.NoFrame() {
break
}
}
if !p.From.Sym.NoSplit() {
p = c.stacksplit(p, autosize) // emit split check
}
q = p
if autosize != 0 {
// Make sure to save link register for non-empty frame, even if
// it is a leaf function, so that traceback works.
// Store link register before decrement SP, so if a signal comes
// during the execution of the function prologue, the traceback
// code will not see a half-updated stack frame.
// This sequence is not async preemptible, as if we open a frame
// at the current SP, it will clobber the saved LR.
q = c.ctxt.StartUnsafePoint(q, c.newprog)
q = obj.Appendp(q, newprog)
q.As = mov
q.Pos = p.Pos
q.From.Type = obj.TYPE_REG
q.From.Reg = REGLINK
q.To.Type = obj.TYPE_MEM
q.To.Offset = int64(-autosize)
q.To.Reg = REGSP
q = obj.Appendp(q, newprog)
q.As = add
q.Pos = p.Pos
q.From.Type = obj.TYPE_CONST
q.From.Offset = int64(-autosize)
q.To.Type = obj.TYPE_REG
q.To.Reg = REGSP
q.Spadj = +autosize
q = c.ctxt.EndUnsafePoint(q, c.newprog, -1)
}
if c.cursym.Func.Text.From.Sym.Wrapper() && c.cursym.Func.Text.Mark&LEAF == 0 {
// if(g->panic != nil && g->panic->argp == FP) g->panic->argp = bottom-of-frame
//
// MOV g_panic(g), R1
// BEQ R1, end
// MOV panic_argp(R1), R2
// ADD $(autosize+FIXED_FRAME), R29, R3
// BNE R2, R3, end
// ADD $FIXED_FRAME, R29, R2
// MOV R2, panic_argp(R1)
// end:
// NOP
//
// The NOP is needed to give the jumps somewhere to land.
// It is a liblink NOP, not an mips NOP: it encodes to 0 instruction bytes.
//
// We don't generate this for leafs because that means the wrapped
// function was inlined into the wrapper.
q = obj.Appendp(q, newprog)
q.As = mov
q.From.Type = obj.TYPE_MEM
q.From.Reg = REGG
q.From.Offset = 4 * int64(c.ctxt.Arch.PtrSize) // G.panic
q.To.Type = obj.TYPE_REG
q.To.Reg = REG_R1
q = obj.Appendp(q, newprog)
q.As = ABEQ
q.From.Type = obj.TYPE_REG
q.From.Reg = REG_R1
q.To.Type = obj.TYPE_BRANCH
q.Mark |= BRANCH
p1 = q
q = obj.Appendp(q, newprog)
q.As = mov
q.From.Type = obj.TYPE_MEM
q.From.Reg = REG_R1
q.From.Offset = 0 // Panic.argp
q.To.Type = obj.TYPE_REG
q.To.Reg = REG_R2
q = obj.Appendp(q, newprog)
q.As = add
q.From.Type = obj.TYPE_CONST
q.From.Offset = int64(autosize) + ctxt.FixedFrameSize()
q.Reg = REGSP
q.To.Type = obj.TYPE_REG
q.To.Reg = REG_R3
q = obj.Appendp(q, newprog)
q.As = ABNE
q.From.Type = obj.TYPE_REG
q.From.Reg = REG_R2
q.Reg = REG_R3
q.To.Type = obj.TYPE_BRANCH
q.Mark |= BRANCH
p2 = q
q = obj.Appendp(q, newprog)
q.As = add
q.From.Type = obj.TYPE_CONST
q.From.Offset = ctxt.FixedFrameSize()
q.Reg = REGSP
q.To.Type = obj.TYPE_REG
q.To.Reg = REG_R2
q = obj.Appendp(q, newprog)
q.As = mov
q.From.Type = obj.TYPE_REG
q.From.Reg = REG_R2
q.To.Type = obj.TYPE_MEM
q.To.Reg = REG_R1
q.To.Offset = 0 // Panic.argp
q = obj.Appendp(q, newprog)
q.As = obj.ANOP
p1.Pcond = q
p2.Pcond = q
}
case ARET:
if p.From.Type == obj.TYPE_CONST {
ctxt.Diag("using BECOME (%v) is not supported!", p)
break
}
retSym := p.To.Sym
p.To.Name = obj.NAME_NONE // clear fields as we may modify p to other instruction
p.To.Sym = nil
if c.cursym.Func.Text.Mark&LEAF != 0 {
if autosize == 0 {
p.As = AJMP
p.From = obj.Addr{}
if retSym != nil { // retjmp
p.To.Type = obj.TYPE_BRANCH
p.To.Name = obj.NAME_EXTERN
p.To.Sym = retSym
} else {
p.To.Type = obj.TYPE_MEM
p.To.Reg = REGLINK
p.To.Offset = 0
}
p.Mark |= BRANCH
break
}
p.As = add
p.From.Type = obj.TYPE_CONST
p.From.Offset = int64(autosize)
p.To.Type = obj.TYPE_REG
p.To.Reg = REGSP
p.Spadj = -autosize
q = c.newprog()
q.As = AJMP
q.Pos = p.Pos
q.To.Type = obj.TYPE_MEM
q.To.Offset = 0
q.To.Reg = REGLINK
q.Mark |= BRANCH
q.Spadj = +autosize
q.Link = p.Link
p.Link = q
break
}
p.As = mov
p.From.Type = obj.TYPE_MEM
p.From.Offset = 0
p.From.Reg = REGSP
p.To.Type = obj.TYPE_REG
p.To.Reg = REGLINK
if autosize != 0 {
q = c.newprog()
q.As = add
q.Pos = p.Pos
q.From.Type = obj.TYPE_CONST
q.From.Offset = int64(autosize)
q.To.Type = obj.TYPE_REG
q.To.Reg = REGSP
q.Spadj = -autosize
q.Link = p.Link
p.Link = q
}
q1 = c.newprog()
q1.As = AJMP
q1.Pos = p.Pos
if retSym != nil { // retjmp
q1.To.Type = obj.TYPE_BRANCH
q1.To.Name = obj.NAME_EXTERN
q1.To.Sym = retSym
} else {
q1.To.Type = obj.TYPE_MEM
q1.To.Offset = 0
q1.To.Reg = REGLINK
}
q1.Mark |= BRANCH
q1.Spadj = +autosize
q1.Link = q.Link
q.Link = q1
case AADD,
AADDU,
AADDV,
AADDVU:
if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSP && p.From.Type == obj.TYPE_CONST {
p.Spadj = int32(-p.From.Offset)
}
case obj.AGETCALLERPC:
if cursym.Leaf() {
/* MOV LR, Rd */
p.As = mov
p.From.Type = obj.TYPE_REG
p.From.Reg = REGLINK
} else {
/* MOV (RSP), Rd */
p.As = mov
p.From.Type = obj.TYPE_MEM
p.From.Reg = REGSP
}
}
}
if c.ctxt.Arch.Family == sys.MIPS {
// rewrite MOVD into two MOVF in 32-bit mode to avoid unaligned memory access
for p = c.cursym.Func.Text; p != nil; p = p1 {
p1 = p.Link
if p.As != AMOVD {
continue
}
if p.From.Type != obj.TYPE_MEM && p.To.Type != obj.TYPE_MEM {
continue
}
p.As = AMOVF
q = c.newprog()
*q = *p
q.Link = p.Link
p.Link = q
p1 = q.Link
var addrOff int64
if c.ctxt.Arch.ByteOrder == binary.BigEndian {
addrOff = 4 // swap load/save order
}
if p.From.Type == obj.TYPE_MEM {
reg := REG_F0 + (p.To.Reg-REG_F0)&^1
p.To.Reg = reg
q.To.Reg = reg + 1
p.From.Offset += addrOff
q.From.Offset += 4 - addrOff
} else if p.To.Type == obj.TYPE_MEM {
reg := REG_F0 + (p.From.Reg-REG_F0)&^1
p.From.Reg = reg
q.From.Reg = reg + 1
p.To.Offset += addrOff
q.To.Offset += 4 - addrOff
}
}
}
if nosched {
// if we don't do instruction scheduling, simply add
// NOP after each branch instruction.
for p = c.cursym.Func.Text; p != nil; p = p.Link {
if p.Mark&BRANCH != 0 {
c.addnop(p)
}
}
return
}
// instruction scheduling
q = nil // p - 1
q1 = c.cursym.Func.Text // top of block
o := 0 // count of instructions
for p = c.cursym.Func.Text; p != nil; p = p1 {
p1 = p.Link
o++
if p.Mark&NOSCHED != 0 {
if q1 != p {
c.sched(q1, q)
}
for ; p != nil; p = p.Link {
if p.Mark&NOSCHED == 0 {
break
}
q = p
}
p1 = p
q1 = p
o = 0
continue
}
if p.Mark&(LABEL|SYNC) != 0 {
if q1 != p {
c.sched(q1, q)
}
q1 = p
o = 1
}
if p.Mark&(BRANCH|SYNC) != 0 {
c.sched(q1, p)
q1 = p1
o = 0
}
if o >= NSCHED {
c.sched(q1, p)
q1 = p1
o = 0
}
q = p
}
}
func (c *ctxt0) stacksplit(p *obj.Prog, framesize int32) *obj.Prog {
var mov, add, sub obj.As
if c.ctxt.Arch.Family == sys.MIPS64 {
add = AADDV
mov = AMOVV
sub = ASUBVU
} else {
add = AADDU
mov = AMOVW
sub = ASUBU
}
// MOV g_stackguard(g), R1
p = obj.Appendp(p, c.newprog)
p.As = mov
p.From.Type = obj.TYPE_MEM
p.From.Reg = REGG
p.From.Offset = 2 * int64(c.ctxt.Arch.PtrSize) // G.stackguard0
if c.cursym.CFunc() {
p.From.Offset = 3 * int64(c.ctxt.Arch.PtrSize) // G.stackguard1
}
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R1
// Mark the stack bound check and morestack call async nonpreemptible.
// If we get preempted here, when resumed the preemption request is
// cleared, but we'll still call morestack, which will double the stack
// unnecessarily. See issue #35470.
p = c.ctxt.StartUnsafePoint(p, c.newprog)
var q *obj.Prog
if framesize <= objabi.StackSmall {
// small stack: SP < stackguard
// AGTU SP, stackguard, R1
p = obj.Appendp(p, c.newprog)
p.As = ASGTU
p.From.Type = obj.TYPE_REG
p.From.Reg = REGSP
p.Reg = REG_R1
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R1
} else if framesize <= objabi.StackBig {
// large stack: SP-framesize < stackguard-StackSmall
// ADD $-(framesize-StackSmall), SP, R2
// SGTU R2, stackguard, R1
p = obj.Appendp(p, c.newprog)
p.As = add
p.From.Type = obj.TYPE_CONST
p.From.Offset = -(int64(framesize) - objabi.StackSmall)
p.Reg = REGSP
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R2
p = obj.Appendp(p, c.newprog)
p.As = ASGTU
p.From.Type = obj.TYPE_REG
p.From.Reg = REG_R2
p.Reg = REG_R1
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R1
} else {
// Such a large stack we need to protect against wraparound.
// If SP is close to zero:
// SP-stackguard+StackGuard <= framesize + (StackGuard-StackSmall)
// The +StackGuard on both sides is required to keep the left side positive:
// SP is allowed to be slightly below stackguard. See stack.h.
//
// Preemption sets stackguard to StackPreempt, a very large value.
// That breaks the math above, so we have to check for that explicitly.
// // stackguard is R1
// MOV $StackPreempt, R2
// BEQ R1, R2, label-of-call-to-morestack
// ADD $StackGuard, SP, R2
// SUB R1, R2
// MOV $(framesize+(StackGuard-StackSmall)), R1
// SGTU R2, R1, R1
p = obj.Appendp(p, c.newprog)
p.As = mov
p.From.Type = obj.TYPE_CONST
p.From.Offset = objabi.StackPreempt
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R2
p = obj.Appendp(p, c.newprog)
q = p
p.As = ABEQ
p.From.Type = obj.TYPE_REG
p.From.Reg = REG_R1
p.Reg = REG_R2
p.To.Type = obj.TYPE_BRANCH
p.Mark |= BRANCH
p = obj.Appendp(p, c.newprog)
p.As = add
p.From.Type = obj.TYPE_CONST
p.From.Offset = int64(objabi.StackGuard)
p.Reg = REGSP
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R2
p = obj.Appendp(p, c.newprog)
p.As = sub
p.From.Type = obj.TYPE_REG
p.From.Reg = REG_R1
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R2
p = obj.Appendp(p, c.newprog)
p.As = mov
p.From.Type = obj.TYPE_CONST
p.From.Offset = int64(framesize) + int64(objabi.StackGuard) - objabi.StackSmall
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R1
p = obj.Appendp(p, c.newprog)
p.As = ASGTU
p.From.Type = obj.TYPE_REG
p.From.Reg = REG_R2
p.Reg = REG_R1
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R1
}
// q1: BNE R1, done
p = obj.Appendp(p, c.newprog)
q1 := p
p.As = ABNE
p.From.Type = obj.TYPE_REG
p.From.Reg = REG_R1
p.To.Type = obj.TYPE_BRANCH
p.Mark |= BRANCH
// MOV LINK, R3
p = obj.Appendp(p, c.newprog)
p.As = mov
p.From.Type = obj.TYPE_REG
p.From.Reg = REGLINK
p.To.Type = obj.TYPE_REG
p.To.Reg = REG_R3
if q != nil {
q.Pcond = p
p.Mark |= LABEL
}
p = c.ctxt.EmitEntryStackMap(c.cursym, p, c.newprog)
// JAL runtime.morestack(SB)
p = obj.Appendp(p, c.newprog)
p.As = AJAL
p.To.Type = obj.TYPE_BRANCH
if c.cursym.CFunc() {
p.To.Sym = c.ctxt.Lookup("runtime.morestackc")
} else if !c.cursym.Func.Text.From.Sym.NeedCtxt() {
p.To.Sym = c.ctxt.Lookup("runtime.morestack_noctxt")
} else {
p.To.Sym = c.ctxt.Lookup("runtime.morestack")
}
p.Mark |= BRANCH
p = c.ctxt.EndUnsafePoint(p, c.newprog, -1)
// JMP start
p = obj.Appendp(p, c.newprog)
p.As = AJMP
p.To.Type = obj.TYPE_BRANCH
p.Pcond = c.cursym.Func.Text.Link
p.Mark |= BRANCH
// placeholder for q1's jump target
p = obj.Appendp(p, c.newprog)
p.As = obj.ANOP // zero-width place holder
q1.Pcond = p
return p
}
func (c *ctxt0) addnop(p *obj.Prog) {
q := c.newprog()
q.As = ANOOP
q.Pos = p.Pos
q.Link = p.Link
p.Link = q
}
const (
E_HILO = 1 << 0
E_FCR = 1 << 1
E_MCR = 1 << 2
E_MEM = 1 << 3
E_MEMSP = 1 << 4 /* uses offset and size */
E_MEMSB = 1 << 5 /* uses offset and size */
ANYMEM = E_MEM | E_MEMSP | E_MEMSB
//DELAY = LOAD|BRANCH|FCMP
DELAY = BRANCH /* only schedule branch */
)
type Dep struct {
ireg uint32
freg uint32
cc uint32
}
type Sch struct {
p obj.Prog
set Dep
used Dep
soffset int32
size uint8
nop uint8
comp bool
}
func (c *ctxt0) sched(p0, pe *obj.Prog) {
var sch [NSCHED]Sch
/*
* build side structure
*/
s := sch[:]
for p := p0; ; p = p.Link {
s[0].p = *p
c.markregused(&s[0])
if p == pe {
break
}
s = s[1:]
}
se := s
for i := cap(sch) - cap(se); i >= 0; i-- {
s = sch[i:]
if s[0].p.Mark&DELAY == 0 {
continue
}
if -cap(s) < -cap(se) {
if !conflict(&s[0], &s[1]) {
continue
}
}
var t []Sch
var j int
for j = cap(sch) - cap(s) - 1; j >= 0; j-- {
t = sch[j:]
if t[0].comp {
if s[0].p.Mark&BRANCH != 0 {
continue
}
}
if t[0].p.Mark&DELAY != 0 {
if -cap(s) >= -cap(se) || conflict(&t[0], &s[1]) {
continue
}
}
for u := t[1:]; -cap(u) <= -cap(s); u = u[1:] {
if c.depend(&u[0], &t[0]) {
continue
}
}
goto out2
}
if s[0].p.Mark&BRANCH != 0 {
s[0].nop = 1
}
continue
out2:
// t[0] is the instruction being moved to fill the delay
stmp := t[0]
copy(t[:i-j], t[1:i-j+1])
s[0] = stmp
if t[i-j-1].p.Mark&BRANCH != 0 {
// t[i-j] is being put into a branch delay slot
// combine its Spadj with the branch instruction
t[i-j-1].p.Spadj += t[i-j].p.Spadj
t[i-j].p.Spadj = 0
}
i--
}
/*
* put it all back
*/
var p *obj.Prog
var q *obj.Prog
for s, p = sch[:], p0; -cap(s) <= -cap(se); s, p = s[1:], q {
q = p.Link
if q != s[0].p.Link {
*p = s[0].p
p.Link = q
}
for s[0].nop != 0 {
s[0].nop--
c.addnop(p)
}
}
}
func (c *ctxt0) markregused(s *Sch) {
p := &s.p
s.comp = c.compound(p)
s.nop = 0
if s.comp {
s.set.ireg |= 1 << (REGTMP - REG_R0)
s.used.ireg |= 1 << (REGTMP - REG_R0)
}
ar := 0 /* dest is really reference */
ad := 0 /* source/dest is really address */
ld := 0 /* opcode is load instruction */
sz := 20 /* size of load/store for overlap computation */
/*
* flags based on opcode
*/
switch p.As {
case obj.ATEXT:
c.autosize = int32(p.To.Offset + 8)
ad = 1
case AJAL:
r := p.Reg
if r == 0 {
r = REGLINK
}
s.set.ireg |= 1 << uint(r-REG_R0)
ar = 1
ad = 1
case ABGEZAL,
ABLTZAL:
s.set.ireg |= 1 << (REGLINK - REG_R0)
fallthrough
case ABEQ,
ABGEZ,
ABGTZ,
ABLEZ,
ABLTZ,
ABNE:
ar = 1
ad = 1
case ABFPT,
ABFPF:
ad = 1
s.used.cc |= E_FCR
case ACMPEQD,
ACMPEQF,
ACMPGED,
ACMPGEF,
ACMPGTD,
ACMPGTF:
ar = 1
s.set.cc |= E_FCR
p.Mark |= FCMP
case AJMP:
ar = 1
ad = 1
case AMOVB,
AMOVBU:
sz = 1
ld = 1
case AMOVH,
AMOVHU:
sz = 2
ld = 1
case AMOVF,
AMOVW,
AMOVWL,
AMOVWR:
sz = 4
ld = 1
case AMOVD,
AMOVV,
AMOVVL,
AMOVVR:
sz = 8
ld = 1
case ADIV,
ADIVU,
AMUL,
AMULU,
AREM,
AREMU,
ADIVV,
ADIVVU,
AMULV,
AMULVU,
AREMV,
AREMVU:
s.set.cc = E_HILO
fallthrough
case AADD,
AADDU,
AADDV,
AADDVU,
AAND,
ANOR,
AOR,
ASGT,
ASGTU,
ASLL,
ASRA,
ASRL,
ASLLV,
ASRAV,
ASRLV,
ASUB,
ASUBU,
ASUBV,
ASUBVU,
AXOR,
AADDD,
AADDF,
AADDW,
ASUBD,
ASUBF,
ASUBW,
AMULF,
AMULD,
AMULW,
ADIVF,
ADIVD,
ADIVW:
if p.Reg == 0 {
if p.To.Type == obj.TYPE_REG {
p.Reg = p.To.Reg
}
//if(p->reg == NREG)
// print("botch %P\n", p);
}
}
/*
* flags based on 'to' field
*/
cls := int(p.To.Class)
if cls == 0 {
cls = c.aclass(&p.To) + 1
p.To.Class = int8(cls)
}
cls--
switch cls {
default:
fmt.Printf("unknown class %d %v\n", cls, p)
case C_ZCON,
C_SCON,
C_ADD0CON,
C_AND0CON,
C_ADDCON,
C_ANDCON,
C_UCON,
C_LCON,
C_NONE,
C_SBRA,
C_LBRA,
C_ADDR,
C_TEXTSIZE:
break
case C_HI,
C_LO:
s.set.cc |= E_HILO
case C_FCREG:
s.set.cc |= E_FCR
case C_MREG:
s.set.cc |= E_MCR
case C_ZOREG,
C_SOREG,
C_LOREG:
cls = int(p.To.Reg)
s.used.ireg |= 1 << uint(cls-REG_R0)
if ad != 0 {
break
}
s.size = uint8(sz)
s.soffset = c.regoff(&p.To)
m := uint32(ANYMEM)
if cls == REGSB {
m = E_MEMSB
}
if cls == REGSP {
m = E_MEMSP
}
if ar != 0 {
s.used.cc |= m
} else {
s.set.cc |= m
}
case C_SACON,
C_LACON:
s.used.ireg |= 1 << (REGSP - REG_R0)
case C_SECON,
C_LECON:
s.used.ireg |= 1 << (REGSB - REG_R0)
case C_REG:
if ar != 0 {
s.used.ireg |= 1 << uint(p.To.Reg-REG_R0)
} else {
s.set.ireg |= 1 << uint(p.To.Reg-REG_R0)
}
case C_FREG:
if ar != 0 {
s.used.freg |= 1 << uint(p.To.Reg-REG_F0)
} else {
s.set.freg |= 1 << uint(p.To.Reg-REG_F0)
}
if ld != 0 && p.From.Type == obj.TYPE_REG {
p.Mark |= LOAD
}
case C_SAUTO,
C_LAUTO:
s.used.ireg |= 1 << (REGSP - REG_R0)
if ad != 0 {
break
}
s.size = uint8(sz)
s.soffset = c.regoff(&p.To)
if ar != 0 {
s.used.cc |= E_MEMSP
} else {
s.set.cc |= E_MEMSP
}
case C_SEXT,
C_LEXT:
s.used.ireg |= 1 << (REGSB - REG_R0)
if ad != 0 {
break
}
s.size = uint8(sz)
s.soffset = c.regoff(&p.To)
if ar != 0 {
s.used.cc |= E_MEMSB
} else {
s.set.cc |= E_MEMSB
}
}
/*
* flags based on 'from' field
*/
cls = int(p.From.Class)
if cls == 0 {
cls = c.aclass(&p.From) + 1
p.From.Class = int8(cls)
}
cls--
switch cls {
default:
fmt.Printf("unknown class %d %v\n", cls, p)
case C_ZCON,
C_SCON,
C_ADD0CON,
C_AND0CON,
C_ADDCON,
C_ANDCON,
C_UCON,
C_LCON,
C_NONE,
C_SBRA,
C_LBRA,
C_ADDR,
C_TEXTSIZE:
break
case C_HI,
C_LO:
s.used.cc |= E_HILO
case C_FCREG:
s.used.cc |= E_FCR
case C_MREG:
s.used.cc |= E_MCR
case C_ZOREG,
C_SOREG,
C_LOREG:
cls = int(p.From.Reg)
s.used.ireg |= 1 << uint(cls-REG_R0)
if ld != 0 {
p.Mark |= LOAD
}
s.size = uint8(sz)
s.soffset = c.regoff(&p.From)
m := uint32(ANYMEM)
if cls == REGSB {
m = E_MEMSB
}
if cls == REGSP {
m = E_MEMSP
}
s.used.cc |= m
case C_SACON,
C_LACON:
cls = int(p.From.Reg)
if cls == 0 {
cls = REGSP
}
s.used.ireg |= 1 << uint(cls-REG_R0)
case C_SECON,
C_LECON:
s.used.ireg |= 1 << (REGSB - REG_R0)
case C_REG:
s.used.ireg |= 1 << uint(p.From.Reg-REG_R0)
case C_FREG:
s.used.freg |= 1 << uint(p.From.Reg-REG_F0)
if ld != 0 && p.To.Type == obj.TYPE_REG {
p.Mark |= LOAD
}
case C_SAUTO,
C_LAUTO:
s.used.ireg |= 1 << (REGSP - REG_R0)
if ld != 0 {
p.Mark |= LOAD
}
if ad != 0 {
break
}
s.size = uint8(sz)
s.soffset = c.regoff(&p.From)
s.used.cc |= E_MEMSP
case C_SEXT:
case C_LEXT:
s.used.ireg |= 1 << (REGSB - REG_R0)
if ld != 0 {
p.Mark |= LOAD
}
if ad != 0 {
break
}
s.size = uint8(sz)
s.soffset = c.regoff(&p.From)
s.used.cc |= E_MEMSB
}
cls = int(p.Reg)
if cls != 0 {
if REG_F0 <= cls && cls <= REG_F31 {
s.used.freg |= 1 << uint(cls-REG_F0)
} else {
s.used.ireg |= 1 << uint(cls-REG_R0)
}
}
s.set.ireg &^= (1 << (REGZERO - REG_R0)) /* R0 can't be set */
}
/*
* test to see if two instructions can be
* interchanged without changing semantics
*/
func (c *ctxt0) depend(sa, sb *Sch) bool {
if sa.set.ireg&(sb.set.ireg|sb.used.ireg) != 0 {
return true
}
if sb.set.ireg&sa.used.ireg != 0 {
return true
}
if sa.set.freg&(sb.set.freg|sb.used.freg) != 0 {
return true
}
if sb.set.freg&sa.used.freg != 0 {
return true
}
/*
* special case.
* loads from same address cannot pass.
* this is for hardware fifo's and the like
*/
if sa.used.cc&sb.used.cc&E_MEM != 0 {
if sa.p.Reg == sb.p.Reg {
if c.regoff(&sa.p.From) == c.regoff(&sb.p.From) {
return true
}
}
}
x := (sa.set.cc & (sb.set.cc | sb.used.cc)) | (sb.set.cc & sa.used.cc)
if x != 0 {
/*
* allow SB and SP to pass each other.
* allow SB to pass SB iff doffsets are ok
* anything else conflicts
*/
if x != E_MEMSP && x != E_MEMSB {
return true
}
x = sa.set.cc | sb.set.cc | sa.used.cc | sb.used.cc
if x&E_MEM != 0 {
return true
}
if offoverlap(sa, sb) {
return true
}
}
return false
}
func offoverlap(sa, sb *Sch) bool {
if sa.soffset < sb.soffset {
if sa.soffset+int32(sa.size) > sb.soffset {
return true
}
return false
}
if sb.soffset+int32(sb.size) > sa.soffset {
return true
}
return false
}
/*
* test 2 adjacent instructions
* and find out if inserted instructions
* are desired to prevent stalls.
*/
func conflict(sa, sb *Sch) bool {
if sa.set.ireg&sb.used.ireg != 0 {
return true
}
if sa.set.freg&sb.used.freg != 0 {
return true
}
if sa.set.cc&sb.used.cc != 0 {
return true
}
return false
}
func (c *ctxt0) compound(p *obj.Prog) bool {
o := c.oplook(p)
if o.size != 4 {
return true
}
if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSB {
return true
}
return false
}
var Linkmips64 = obj.LinkArch{
Arch: sys.ArchMIPS64,
Init: buildop,
Preprocess: preprocess,
Assemble: span0,
Progedit: progedit,
DWARFRegisters: MIPSDWARFRegisters,
}
var Linkmips64le = obj.LinkArch{
Arch: sys.ArchMIPS64LE,
Init: buildop,
Preprocess: preprocess,
Assemble: span0,
Progedit: progedit,
DWARFRegisters: MIPSDWARFRegisters,
}
var Linkmips = obj.LinkArch{
Arch: sys.ArchMIPS,
Init: buildop,
Preprocess: preprocess,
Assemble: span0,
Progedit: progedit,
DWARFRegisters: MIPSDWARFRegisters,
}
var Linkmipsle = obj.LinkArch{
Arch: sys.ArchMIPSLE,
Init: buildop,
Preprocess: preprocess,
Assemble: span0,
Progedit: progedit,
DWARFRegisters: MIPSDWARFRegisters,
}
Reference in a new issue View git blame Copy permalink