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

[dev.simd] all: merge master (9de69f6) into dev.simd

Browse source 
Merge List:

+ 2025-08-20 9de69f6913 errors: mention Is/As in Join docs
+ 2025-08-20 4afd482812 cmd/go/internal/doc: pass URL fragments separately with -http
+ 2025-08-20 509d5f647f internal/poll: don't call Seek for overlapped Windows handles
+ 2025-08-20 853fc12739 internal/poll: set the correct file offset in FD.Seek for Windows overlapped handles
+ 2025-08-19 bd885401d5 runtime: save and restore all fcc registers in async preempt on loong64
+ 2025-08-19 119546ea4f cmd/go: document install outputs to $GOOS_$GOARCH when cross compiling
+ 2025-08-19 ffa882059c unique: deflake TestCanonMap/LoadOrStore/ConcurrentUnsharedKeys
+ 2025-08-19 1f2e8e03e4 os: fix path in MkdirTemp error message
+ 2025-08-19 5024d0d884 cmd/compile: tweak example command in README
+ 2025-08-19 b80ffb64d8 internal/trace: remove redundant info from Event.String
+ 2025-08-19 c7d8bda459 cmd/compile/internal: make function comments match function names
+ 2025-08-19 de2d741667 internal/trace: use RFC3339Nano for wall clock snapshots in Event.String
+ 2025-08-19 c61db5ebd5 syscall: forkAndExecInChild1: don't reuse pid variable
+ 2025-08-19 07ee3bfc63 cmd/go: use modern pprof flags in documentation
+ 2025-08-18 5a56d8848b cmd/compile: ensure we use allowed registers for input-clobbering instructions
+ 2025-08-18 c3927a47f0 runtime: fix comments in tracetype.go
+ 2025-08-15 77f911e31c internal/trace: emit final sync event for generation in Go 1.26+
+ 2025-08-15 786be1d2bf runtime: don't overwrite global stop channel in tests
+ 2025-08-15 4a7fde922f internal/trace: add end-of-generation signal to trace
+ 2025-08-15 cb814bd5bc net: skip TestIPv4WriteMsgUDPAddrPort on plan9
+ 2025-08-15 78a3968c2c runtime/metrics: add metric for current Go-owned thread count
+ 2025-08-15 ab8121a407 runtime/metrics: add metric for total goroutines created
+ 2025-08-15 13df972f68 runtime/metrics: add metrics for goroutine sched states
+ 2025-08-15 bd07fafb0a runtime: disable stack shrinking for all waiting-for-suspendG cases
+ 2025-08-15 a651e2ea47 runtime: remove duff support for amd64
+ 2025-08-15 e4291e484c runtime: remove duff support for arm64
+ 2025-08-15 15d6dbc05c cmd/compile: use generated loops instead of DUFFCOPY on arm64
+ 2025-08-15 bca3e98b8a cmd/go: test barrier actions
+ 2025-08-15 052fcde9fd internal/runtime: cleaner overflow checker
+ 2025-08-15 3871c0d84d syscall: permit nil destination address in sendmsgN{Inet4,Inet6}
+ 2025-08-14 a8564bd412 runtime: make all synctest bubble violations fatal panics

Change-Id: Ibc94566bc69bcb59b1d79b6fa868610ca2d1d223
This commit is contained in:
Cherry Mui 2025-08-20 16:06:42 -04:00
commit 103b6e39ca
68 changed files with 1449 additions and 1363 deletions
Download patch file Download diff file Expand all files Collapse all files

8
src/cmd/compile/README.md
View file

@ -281,11 +281,11 @@ dependencies, so is not suitable for distributed build systems.)
``` ```
$ go install golang.org/x/tools/cmd/toolstash@latest $ go install golang.org/x/tools/cmd/toolstash@latest
$ git clone https://go.googlesource.com/go $ git clone https://go.googlesource.com/go
$ cd go $ export PATH=$PWD/go/bin:$PATH
$ cd go/src
$ git checkout -b mybranch $ git checkout -b mybranch
$ ./src/all.bash # build and confirm good starting point $ ./all.bash # build and confirm good starting point
$ export PATH=$PWD/bin:$PATH $ toolstash save # save current tools
$ toolstash save # save current tools
``` ```
After that, your edit/compile/test cycle can be similar to: After that, your edit/compile/test cycle can be similar to:
``` ```

196
src/cmd/compile/internal/arm64/ssa.go
View file

@ -1162,41 +1162,119 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
// BNE loop // BNE loop
// There's a past-the-end pointer here, any problem with that? // There's a past-the-end pointer here, any problem with that?
case ssa.OpARM64DUFFCOPY:
p := s.Prog(obj.ADUFFCOPY)
p.To.Type = obj.TYPE_MEM
p.To.Name = obj.NAME_EXTERN
p.To.Sym = ir.Syms.Duffcopy
p.To.Offset = v.AuxInt
case ssa.OpARM64LoweredMove: case ssa.OpARM64LoweredMove:
// LDP.P 16(R16), (R25, Rtmp) dstReg := v.Args[0].Reg()
// STP.P (R25, Rtmp), 16(R17) srcReg := v.Args[1].Reg()
// CMP Rarg2, R16 if dstReg == srcReg {
// BLE -3(PC) break
// arg2 is the address of the last element of src }
p := s.Prog(arm64.ALDP) tmpReg1 := int16(arm64.REG_R24)
p.Scond = arm64.C_XPOST tmpReg2 := int16(arm64.REG_R25)
p.From.Type = obj.TYPE_MEM n := v.AuxInt
p.From.Reg = arm64.REG_R16 if n < 16 {
p.From.Offset = 16 v.Fatalf("Move too small %d", n)
p.To.Type = obj.TYPE_REGREG }
p.To.Reg = arm64.REG_R25
p.To.Offset = int64(arm64.REGTMP) // Generate copying instructions.
p2 := s.Prog(arm64.ASTP) var off int64
p2.Scond = arm64.C_XPOST for n >= 16 {
p2.From.Type = obj.TYPE_REGREG // LDP off(srcReg), (tmpReg1, tmpReg2)
p2.From.Reg = arm64.REG_R25 // STP (tmpReg1, tmpReg2), off(dstReg)
p2.From.Offset = int64(arm64.REGTMP) move16(s, srcReg, dstReg, tmpReg1, tmpReg2, off, false)
p2.To.Type = obj.TYPE_MEM off += 16
p2.To.Reg = arm64.REG_R17 n -= 16
p2.To.Offset = 16 }
p3 := s.Prog(arm64.ACMP) if n > 8 {
p3.From.Type = obj.TYPE_REG // MOVD off(srcReg), tmpReg1
p3.From.Reg = v.Args[2].Reg() // MOVD tmpReg1, off(dstReg)
p3.Reg = arm64.REG_R16 move8(s, srcReg, dstReg, tmpReg1, off)
p4 := s.Prog(arm64.ABLE) off += 8
p4.To.Type = obj.TYPE_BRANCH n -= 8
p4.To.SetTarget(p) }
if n != 0 {
// MOVD off+n-8(srcReg), tmpReg1
// MOVD tmpReg1, off+n-8(dstReg)
move8(s, srcReg, dstReg, tmpReg1, off+n-8)
}
case ssa.OpARM64LoweredMoveLoop:
dstReg := v.Args[0].Reg()
srcReg := v.Args[1].Reg()
if dstReg == srcReg {
break
}
countReg := int16(arm64.REG_R23)
tmpReg1 := int16(arm64.REG_R24)
tmpReg2 := int16(arm64.REG_R25)
n := v.AuxInt
loopSize := int64(64)
if n < 3*loopSize {
// - a loop count of 0 won't work.
// - a loop count of 1 is useless.
// - a loop count of 2 is a code size ~tie
// 3 instructions to implement the loop
// 4 instructions in the loop body
// vs
// 8 instructions in the straightline code
// Might as well use straightline code.
v.Fatalf("ZeroLoop size too small %d", n)
}
// Put iteration count in a register.
// MOVD $n, countReg
p := s.Prog(arm64.AMOVD)
p.From.Type = obj.TYPE_CONST
p.From.Offset = n / loopSize
p.To.Type = obj.TYPE_REG
p.To.Reg = countReg
cntInit := p
// Move loopSize bytes starting at srcReg to dstReg.
// Increment srcReg and destReg by loopSize as a side effect.
for range loopSize / 16 {
// LDP.P 16(srcReg), (tmpReg1, tmpReg2)
// STP.P (tmpReg1, tmpReg2), 16(dstReg)
move16(s, srcReg, dstReg, tmpReg1, tmpReg2, 0, true)
}
// Decrement loop count.
// SUB $1, countReg
p = s.Prog(arm64.ASUB)
p.From.Type = obj.TYPE_CONST
p.From.Offset = 1
p.To.Type = obj.TYPE_REG
p.To.Reg = countReg
// Jump to loop header if we're not done yet.
// CBNZ head
p = s.Prog(arm64.ACBNZ)
p.From.Type = obj.TYPE_REG
p.From.Reg = countReg
p.To.Type = obj.TYPE_BRANCH
p.To.SetTarget(cntInit.Link)
// Multiples of the loop size are now done.
n %= loopSize
// Copy any fractional portion.
var off int64
for n >= 16 {
// LDP off(srcReg), (tmpReg1, tmpReg2)
// STP (tmpReg1, tmpReg2), off(dstReg)
move16(s, srcReg, dstReg, tmpReg1, tmpReg2, off, false)
off += 16
n -= 16
}
if n > 8 {
// MOVD off(srcReg), tmpReg1
// MOVD tmpReg1, off(dstReg)
move8(s, srcReg, dstReg, tmpReg1, off)
off += 8
n -= 8
}
if n != 0 {
// MOVD off+n-8(srcReg), tmpReg1
// MOVD tmpReg1, off+n-8(dstReg)
move8(s, srcReg, dstReg, tmpReg1, off+n-8)
}
case ssa.OpARM64CALLstatic, ssa.OpARM64CALLclosure, ssa.OpARM64CALLinter: case ssa.OpARM64CALLstatic, ssa.OpARM64CALLclosure, ssa.OpARM64CALLinter:
s.Call(v) s.Call(v)
case ssa.OpARM64CALLtail: case ssa.OpARM64CALLtail:
@ -1599,3 +1677,53 @@ func zero8(s *ssagen.State, reg int16, off int64) {
p.To.Reg = reg p.To.Reg = reg
p.To.Offset = off p.To.Offset = off
} }
// move16 copies 16 bytes at src+off to dst+off.
// Uses registers tmp1 and tmp2.
// If postInc is true, increment src and dst by 16.
func move16(s *ssagen.State, src, dst, tmp1, tmp2 int16, off int64, postInc bool) {
// LDP off(src), (tmp1, tmp2)
ld := s.Prog(arm64.ALDP)
ld.From.Type = obj.TYPE_MEM
ld.From.Reg = src
ld.From.Offset = off
ld.To.Type = obj.TYPE_REGREG
ld.To.Reg = tmp1
ld.To.Offset = int64(tmp2)
// STP (tmp1, tmp2), off(dst)
st := s.Prog(arm64.ASTP)
st.From.Type = obj.TYPE_REGREG
st.From.Reg = tmp1
st.From.Offset = int64(tmp2)
st.To.Type = obj.TYPE_MEM
st.To.Reg = dst
st.To.Offset = off
if postInc {
if off != 0 {
panic("can't postinc with non-zero offset")
}
ld.Scond = arm64.C_XPOST
st.Scond = arm64.C_XPOST
ld.From.Offset = 16
st.To.Offset = 16
}
}
// move8 copies 8 bytes at src+off to dst+off.
// Uses register tmp.
func move8(s *ssagen.State, src, dst, tmp int16, off int64) {
// MOVD off(src), tmp
ld := s.Prog(arm64.AMOVD)
ld.From.Type = obj.TYPE_MEM
ld.From.Reg = src
ld.From.Offset = off
ld.To.Type = obj.TYPE_REG
ld.To.Reg = tmp
// MOVD tmp, off(dst)
st := s.Prog(arm64.AMOVD)
st.From.Type = obj.TYPE_REG
st.From.Reg = tmp
st.To.Type = obj.TYPE_MEM
st.To.Reg = dst
st.To.Offset = off
}

2
src/cmd/compile/internal/loopvar/loopvar_test.go
View file

@ -50,7 +50,7 @@ var cases = []testcase{
{"1", "", 0, []string{"for_nested.go"}}, {"1", "", 0, []string{"for_nested.go"}},
} }
// TestLoopVar checks that the GOEXPERIMENT and debug flags behave as expected. // TestLoopVarGo1_21 checks that the GOEXPERIMENT and debug flags behave as expected.
func TestLoopVarGo1_21(t *testing.T) { func TestLoopVarGo1_21(t *testing.T) {
switch runtime.GOOS { switch runtime.GOOS {
case "linux", "darwin": case "linux", "darwin":

2
src/cmd/compile/internal/pgoir/irgraph.go
View file

@ -158,7 +158,7 @@ func New(profileFile string) (*Profile, error) {
}, nil }, nil
} }
// initializeIRGraph builds the IRGraph by visiting all the ir.Func in decl list // createIRGraph builds the IRGraph by visiting all the ir.Func in decl list
// of a package. // of a package.
func createIRGraph(namedEdgeMap pgo.NamedEdgeMap) *IRGraph { func createIRGraph(namedEdgeMap pgo.NamedEdgeMap) *IRGraph {
g := &IRGraph{ g := &IRGraph{

35
src/cmd/compile/internal/ssa/_gen/ARM64.rules
View file

@ -462,39 +462,8 @@
(STP [16] dst (Select0 <typ.UInt64> (LDP [16] src mem)) (Select1 <typ.UInt64> (LDP [16] src mem)) (STP [16] dst (Select0 <typ.UInt64> (LDP [16] src mem)) (Select1 <typ.UInt64> (LDP [16] src mem))
(STP dst (Select0 <typ.UInt64> (LDP src mem)) (Select1 <typ.UInt64> (LDP src mem)) mem)))) (STP dst (Select0 <typ.UInt64> (LDP src mem)) (Select1 <typ.UInt64> (LDP src mem)) mem))))
// strip off fractional word move (Move [s] dst src mem) && s > 64 && s < 192 && logLargeCopy(v, s) => (LoweredMove [s] dst src mem)
(Move [s] dst src mem) && s%16 != 0 && s%16 <= 8 && s > 64 => (Move [s] dst src mem) && s >= 192 && logLargeCopy(v, s) => (LoweredMoveLoop [s] dst src mem)
(Move [8]
(OffPtr <dst.Type> dst [s-8])
(OffPtr <src.Type> src [s-8])
(Move [s-s%16] dst src mem))
(Move [s] dst src mem) && s%16 != 0 && s%16 > 8 && s > 64 =>
(Move [16]
(OffPtr <dst.Type> dst [s-16])
(OffPtr <src.Type> src [s-16])
(Move [s-s%16] dst src mem))
// medium move uses a duff device
(Move [s] dst src mem)
&& s > 64 && s <= 16*64 && s%16 == 0
&& logLargeCopy(v, s) =>
(DUFFCOPY [8 * (64 - s/16)] dst src mem)
// 8 is the number of bytes to encode:
//
// LDP.P 16(R16), (R26, R27)
// STP.P (R26, R27), 16(R17)
//
// 64 is number of these blocks. See runtime/duff_arm64.s:duffcopy
// large move uses a loop
(Move [s] dst src mem)
&& s%16 == 0 && s > 16*64
&& logLargeCopy(v, s) =>
(LoweredMove
dst
src
(ADDconst <src.Type> src [s-16])
mem)
// calls // calls
(StaticCall ...) => (CALLstatic ...) (StaticCall ...) => (CALLstatic ...)

51
src/cmd/compile/internal/ssa/_gen/ARM64Ops.go
View file

@ -144,6 +144,8 @@ func init() {
gpspsbg = gpspg | buildReg("SB") gpspsbg = gpspg | buildReg("SB")
fp = buildReg("F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31") fp = buildReg("F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27 F28 F29 F30 F31")
callerSave = gp | fp | buildReg("g") // runtime.setg (and anything calling it) may clobber g callerSave = gp | fp | buildReg("g") // runtime.setg (and anything calling it) may clobber g
r24to25 = buildReg("R24 R25")
r23to25 = buildReg("R23 R24 R25")
rz = buildReg("ZERO") rz = buildReg("ZERO")
first16 = buildReg("R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15") first16 = buildReg("R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15")
) )
@ -568,47 +570,40 @@ func init() {
needIntTemp: true, needIntTemp: true,
}, },
// duffcopy // medium copying
// arg0 = address of dst memory (in R21, changed as side effect) // arg0 = address of dst memory
// arg1 = address of src memory (in R20, changed as side effect) // arg1 = address of src memory
// arg2 = mem // arg2 = mem
// auxint = offset into duffcopy code to start executing // auxint = # of bytes to copy
// returns mem // returns mem
// R20, R21 changed as side effect
// R16 and R17 may be clobbered by linker trampoline.
{ {
name: "DUFFCOPY", name: "LoweredMove",
aux: "Int64", aux: "Int64",
argLength: 3, argLength: 3,
reg: regInfo{ reg: regInfo{
inputs: []regMask{buildReg("R21"), buildReg("R20")}, inputs: []regMask{gp &^ r24to25, gp &^ r24to25},
clobbers: buildReg("R16 R17 R20 R21 R26 R30"), clobbers: r24to25, // TODO: figure out needIntTemp x2
}, },
//faultOnNilArg0: true, // Note: removed for 73748. TODO: reenable at some point faultOnNilArg0: true,
//faultOnNilArg1: true, faultOnNilArg1: true,
unsafePoint: true, // FP maintenance around DUFFCOPY can be clobbered by interrupts
}, },
// large move // large copying
// arg0 = address of dst memory (in R17 aka arm64.REGRT2, changed as side effect) // arg0 = address of dst memory
// arg1 = address of src memory (in R16 aka arm64.REGRT1, changed as side effect) // arg1 = address of src memory
// arg2 = address of the last element of src // arg2 = mem
// arg3 = mem // auxint = # of bytes to copy
// returns mem // returns mem
// LDP.P 16(R16), (R25, Rtmp)
// STP.P (R25, Rtmp), 16(R17)
// CMP Rarg2, R16
// BLE -3(PC)
// Note: the-end-of-src may be not a valid pointer. it's a problem if it is spilled.
// the-end-of-src - 16 is within the area to copy, ok to spill.
{ {
name: "LoweredMove", name: "LoweredMoveLoop",
argLength: 4, aux: "Int64",
argLength: 3,
reg: regInfo{ reg: regInfo{
inputs: []regMask{buildReg("R17"), buildReg("R16"), gp &^ buildReg("R25")}, inputs: []regMask{gp &^ r23to25, gp &^ r23to25},
clobbers: buildReg("R16 R17 R25"), clobbers: r23to25, // TODO: figure out needIntTemp x3
clobbersArg0: true,
clobbersArg1: true,
}, },
clobberFlags: true,
faultOnNilArg0: true, faultOnNilArg0: true,
faultOnNilArg1: true, faultOnNilArg1: true,
}, },

2
src/cmd/compile/internal/ssa/func_test.go
View file

@ -475,7 +475,7 @@ func opcodeMap(f *Func) map[Op]int {
return m return m
} }
// opcodeCounts checks that the number of opcodes listed in m agree with the // checkOpcodeCounts checks that the number of opcodes listed in m agree with the
// number of opcodes that appear in the function. // number of opcodes that appear in the function.
func checkOpcodeCounts(t *testing.T, f *Func, m map[Op]int) { func checkOpcodeCounts(t *testing.T, f *Func, m map[Op]int) {
n := opcodeMap(f) n := opcodeMap(f)

42
src/cmd/compile/internal/ssa/opGen.go
View file

@ -2906,8 +2906,8 @@ const (
OpARM64GreaterEqualNoov OpARM64GreaterEqualNoov
OpARM64LoweredZero OpARM64LoweredZero
OpARM64LoweredZeroLoop OpARM64LoweredZeroLoop
OpARM64DUFFCOPY
OpARM64LoweredMove OpARM64LoweredMove
OpARM64LoweredMoveLoop
OpARM64LoweredGetClosurePtr OpARM64LoweredGetClosurePtr
OpARM64LoweredGetCallerSP OpARM64LoweredGetCallerSP
OpARM64LoweredGetCallerPC OpARM64LoweredGetCallerPC
@ -43040,32 +43040,34 @@ var opcodeTable = [...]opInfo{
clobbersArg0: true, clobbersArg0: true,
}, },
}, },
{
name: "DUFFCOPY",
auxType: auxInt64,
argLen: 3,
unsafePoint: true,
reg: regInfo{
inputs: []inputInfo{
{0, 1048576}, // R21
{1, 524288}, // R20
},
clobbers: 303759360, // R16 R17 R20 R21 R26 R30
},
},
{ {
name: "LoweredMove", name: "LoweredMove",
argLen: 4, auxType: auxInt64,
clobberFlags: true, argLen: 3,
faultOnNilArg0: true, faultOnNilArg0: true,
faultOnNilArg1: true, faultOnNilArg1: true,
reg: regInfo{ reg: regInfo{
inputs: []inputInfo{ inputs: []inputInfo{
{0, 131072}, // R17 {0, 310378495}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R26 R30
{1, 65536}, // R16 {1, 310378495}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R26 R30
{2, 318767103}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R26 R30
}, },
clobbers: 16973824, // R16 R17 R25 clobbers: 25165824, // R24 R25
},
},
{
name: "LoweredMoveLoop",
auxType: auxInt64,
argLen: 3,
faultOnNilArg0: true,
faultOnNilArg1: true,
reg: regInfo{
inputs: []inputInfo{
{0, 306184191}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R26 R30
{1, 306184191}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R26 R30
},
clobbers: 29360128, // R23 R24 R25
clobbersArg0: true,
clobbersArg1: true,
}, },
}, },
{ {

3
src/cmd/compile/internal/ssa/regalloc.go
View file

@ -1756,10 +1756,9 @@ func (s *regAllocState) regalloc(f *Func) {
// spilling the value with the most distant next use. // spilling the value with the most distant next use.
continue continue
} }
// Copy input to a new clobberable register. // Copy input to a different register that won't be clobbered.
c := s.allocValToReg(v.Args[i], m, true, v.Pos) c := s.allocValToReg(v.Args[i], m, true, v.Pos)
s.copies[c] = false s.copies[c] = false
args[i] = c
} }
// Pick a temporary register if needed. // Pick a temporary register if needed.

74
src/cmd/compile/internal/ssa/rewriteARM64.go
View file

@ -19688,87 +19688,35 @@ func rewriteValueARM64_OpMove(v *Value) bool {
return true return true
} }
// match: (Move [s] dst src mem) // match: (Move [s] dst src mem)
// cond: s%16 != 0 && s%16 <= 8 && s > 64 // cond: s > 64 && s < 192 && logLargeCopy(v, s)
// result: (Move [8] (OffPtr <dst.Type> dst [s-8]) (OffPtr <src.Type> src [s-8]) (Move [s-s%16] dst src mem)) // result: (LoweredMove [s] dst src mem)
for { for {
s := auxIntToInt64(v.AuxInt) s := auxIntToInt64(v.AuxInt)
dst := v_0 dst := v_0
src := v_1 src := v_1
mem := v_2 mem := v_2
if !(s%16 != 0 && s%16 <= 8 && s > 64) { if !(s > 64 && s < 192 && logLargeCopy(v, s)) {
break break
} }
v.reset(OpMove) v.reset(OpARM64LoweredMove)
v.AuxInt = int64ToAuxInt(8) v.AuxInt = int64ToAuxInt(s)
v0 := b.NewValue0(v.Pos, OpOffPtr, dst.Type)
v0.AuxInt = int64ToAuxInt(s - 8)
v0.AddArg(dst)
v1 := b.NewValue0(v.Pos, OpOffPtr, src.Type)
v1.AuxInt = int64ToAuxInt(s - 8)
v1.AddArg(src)
v2 := b.NewValue0(v.Pos, OpMove, types.TypeMem)
v2.AuxInt = int64ToAuxInt(s - s%16)
v2.AddArg3(dst, src, mem)
v.AddArg3(v0, v1, v2)
return true
}
// match: (Move [s] dst src mem)
// cond: s%16 != 0 && s%16 > 8 && s > 64
// result: (Move [16] (OffPtr <dst.Type> dst [s-16]) (OffPtr <src.Type> src [s-16]) (Move [s-s%16] dst src mem))
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !(s%16 != 0 && s%16 > 8 && s > 64) {
break
}
v.reset(OpMove)
v.AuxInt = int64ToAuxInt(16)
v0 := b.NewValue0(v.Pos, OpOffPtr, dst.Type)
v0.AuxInt = int64ToAuxInt(s - 16)
v0.AddArg(dst)
v1 := b.NewValue0(v.Pos, OpOffPtr, src.Type)
v1.AuxInt = int64ToAuxInt(s - 16)
v1.AddArg(src)
v2 := b.NewValue0(v.Pos, OpMove, types.TypeMem)
v2.AuxInt = int64ToAuxInt(s - s%16)
v2.AddArg3(dst, src, mem)
v.AddArg3(v0, v1, v2)
return true
}
// match: (Move [s] dst src mem)
// cond: s > 64 && s <= 16*64 && s%16 == 0 && logLargeCopy(v, s)
// result: (DUFFCOPY [8 * (64 - s/16)] dst src mem)
for {
s := auxIntToInt64(v.AuxInt)
dst := v_0
src := v_1
mem := v_2
if !(s > 64 && s <= 16*64 && s%16 == 0 && logLargeCopy(v, s)) {
break
}
v.reset(OpARM64DUFFCOPY)
v.AuxInt = int64ToAuxInt(8 * (64 - s/16))
v.AddArg3(dst, src, mem) v.AddArg3(dst, src, mem)
return true return true
} }
// match: (Move [s] dst src mem) // match: (Move [s] dst src mem)
// cond: s%16 == 0 && s > 16*64 && logLargeCopy(v, s) // cond: s >= 192 && logLargeCopy(v, s)
// result: (LoweredMove dst src (ADDconst <src.Type> src [s-16]) mem) // result: (LoweredMoveLoop [s] dst src mem)
for { for {
s := auxIntToInt64(v.AuxInt) s := auxIntToInt64(v.AuxInt)
dst := v_0 dst := v_0
src := v_1 src := v_1
mem := v_2 mem := v_2
if !(s%16 == 0 && s > 16*64 && logLargeCopy(v, s)) { if !(s >= 192 && logLargeCopy(v, s)) {
break break
} }
v.reset(OpARM64LoweredMove) v.reset(OpARM64LoweredMoveLoop)
v0 := b.NewValue0(v.Pos, OpARM64ADDconst, src.Type) v.AuxInt = int64ToAuxInt(s)
v0.AuxInt = int64ToAuxInt(s - 16) v.AddArg3(dst, src, mem)
v0.AddArg(src)
v.AddArg4(dst, src, v0, mem)
return true return true
} }
return false return false

6
src/cmd/go/alldocs.go
View file

@ -758,6 +758,8 @@
// variable, which defaults to $GOPATH/bin or $HOME/go/bin if the GOPATH // variable, which defaults to $GOPATH/bin or $HOME/go/bin if the GOPATH
// environment variable is not set. Executables in $GOROOT // environment variable is not set. Executables in $GOROOT
// are installed in $GOROOT/bin or $GOTOOLDIR instead of $GOBIN. // are installed in $GOROOT/bin or $GOTOOLDIR instead of $GOBIN.
// Cross compiled binaries are installed in $GOOS_$GOARCH subdirectories
// of the above.
// //
// If the arguments have version suffixes (like @latest or @v1.0.0), "go install" // If the arguments have version suffixes (like @latest or @v1.0.0), "go install"
// builds packages in module-aware mode, ignoring the go.mod file in the current // builds packages in module-aware mode, ignoring the go.mod file in the current
@ -3231,8 +3233,8 @@
// //
// Several of the flags control profiling and write an execution profile // Several of the flags control profiling and write an execution profile
// suitable for "go tool pprof"; run "go tool pprof -h" for more // suitable for "go tool pprof"; run "go tool pprof -h" for more
// information. The --alloc_space, --alloc_objects, and --show_bytes // information. The -sample_index=alloc_space, -sample_index=alloc_objects,
// options of pprof control how the information is presented. // and -show_bytes options of pprof control how the information is presented.
// //
// The following flags are recognized by the 'go test' command and // The following flags are recognized by the 'go test' command and
// control the execution of any test: // control the execution of any test:

20
src/cmd/go/internal/doc/doc.go
View file

@ -212,16 +212,16 @@ func do(writer io.Writer, flagSet *flag.FlagSet, args []string) (err error) {
mod, err := runCmd(append(os.Environ(), "GOWORK=off"), "go", "list", "-m") mod, err := runCmd(append(os.Environ(), "GOWORK=off"), "go", "list", "-m")
if err == nil && mod != "" && mod != "command-line-arguments" { if err == nil && mod != "" && mod != "command-line-arguments" {
// If there's a module, go to the module's doc page. // If there's a module, go to the module's doc page.
return doPkgsite(mod) return doPkgsite(mod, "")
} }
gowork, err := runCmd(nil, "go", "env", "GOWORK") gowork, err := runCmd(nil, "go", "env", "GOWORK")
if err == nil && gowork != "" { if err == nil && gowork != "" {
// Outside a module, but in a workspace, go to the home page // Outside a module, but in a workspace, go to the home page
// with links to each of the modules' pages. // with links to each of the modules' pages.
return doPkgsite("") return doPkgsite("", "")
} }
// Outside a module or workspace, go to the documentation for the standard library. // Outside a module or workspace, go to the documentation for the standard library.
return doPkgsite("std") return doPkgsite("std", "")
} }
// If args are provided, we need to figure out which page to open on the pkgsite // If args are provided, we need to figure out which page to open on the pkgsite
@ -282,11 +282,11 @@ func do(writer io.Writer, flagSet *flag.FlagSet, args []string) (err error) {
} }
if found { if found {
if serveHTTP { if serveHTTP {
path, err := objectPath(userPath, pkg, symbol, method) path, fragment, err := objectPath(userPath, pkg, symbol, method)
if err != nil { if err != nil {
return err return err
} }
return doPkgsite(path) return doPkgsite(path, fragment)
} }
return nil return nil
} }
@ -305,7 +305,8 @@ func runCmd(env []string, cmdline ...string) (string, error) {
return strings.TrimSpace(stdout.String()), nil return strings.TrimSpace(stdout.String()), nil
} }
func objectPath(userPath string, pkg *Package, symbol, method string) (string, error) { // returns a path followed by a fragment (or an error)
func objectPath(userPath string, pkg *Package, symbol, method string) (string, string, error) {
var err error var err error
path := pkg.build.ImportPath path := pkg.build.ImportPath
if path == "." { if path == "." {
@ -314,7 +315,7 @@ func objectPath(userPath string, pkg *Package, symbol, method string) (string, e
// go list to get the import path. // go list to get the import path.
path, err = runCmd(nil, "go", "list", userPath) path, err = runCmd(nil, "go", "list", userPath)
if err != nil { if err != nil {
return "", err return "", "", err
} }
} }
@ -322,10 +323,7 @@ func objectPath(userPath string, pkg *Package, symbol, method string) (string, e
if symbol != "" && method != "" { if symbol != "" && method != "" {
object = symbol + "." + method object = symbol + "." + method
} }
if object != "" { return path, object, nil
path = path + "#" + object
}
return path, nil
} }
// failMessage creates a nicely formatted error message when there is no result to show. // failMessage creates a nicely formatted error message when there is no result to show.

5
src/cmd/go/internal/doc/pkgsite.go
View file

@ -34,7 +34,7 @@ func pickUnusedPort() (int, error) {
return port, nil return port, nil
} }
func doPkgsite(urlPath string) error { func doPkgsite(urlPath, fragment string) error {
port, err := pickUnusedPort() port, err := pickUnusedPort()
if err != nil { if err != nil {
return fmt.Errorf("failed to find port for documentation server: %v", err) return fmt.Errorf("failed to find port for documentation server: %v", err)
@ -44,6 +44,9 @@ func doPkgsite(urlPath string) error {
if err != nil { if err != nil {
return fmt.Errorf("internal error: failed to construct url: %v", err) return fmt.Errorf("internal error: failed to construct url: %v", err)
} }
if fragment != "" {
path += "#" + fragment
}
// Turn off the default signal handler for SIGINT (and SIGQUIT on Unix) // Turn off the default signal handler for SIGINT (and SIGQUIT on Unix)
// and instead wait for the child process to handle the signal and // and instead wait for the child process to handle the signal and

2
src/cmd/go/internal/doc/pkgsite_bootstrap.go
View file

@ -8,4 +8,4 @@
package doc package doc
func doPkgsite(string) error { return nil } func doPkgsite(string, string) error { return nil }

41
src/cmd/go/internal/test/test.go
View file

@ -186,8 +186,8 @@ and flags that apply to the resulting test binary.
Several of the flags control profiling and write an execution profile Several of the flags control profiling and write an execution profile
suitable for "go tool pprof"; run "go tool pprof -h" for more suitable for "go tool pprof"; run "go tool pprof -h" for more
information. The --alloc_space, --alloc_objects, and --show_bytes information. The -sample_index=alloc_space, -sample_index=alloc_objects,
options of pprof control how the information is presented. and -show_bytes options of pprof control how the information is presented.
The following flags are recognized by the 'go test' command and The following flags are recognized by the 'go test' command and
control the execution of any test: control the execution of any test:
@ -1044,11 +1044,36 @@ func runTest(ctx context.Context, cmd *base.Command, args []string) {
prints = append(prints, printTest) prints = append(prints, printTest)
} }
// Order runs for coordinating start JSON prints. // Order runs for coordinating start JSON prints via two mechanisms:
// 1. Channel locking forces runTest actions to start in-order.
// 2. Barrier tasks force runTest actions to be scheduled in-order.
// We need both for performant behavior, as channel locking without the barrier tasks starves the worker pool,
// and barrier tasks without channel locking doesn't guarantee start in-order behavior alone.
var prevBarrier *work.Action
ch := make(chan struct{}) ch := make(chan struct{})
close(ch) close(ch)
for _, a := range runs { for _, a := range runs {
if r, ok := a.Actor.(*runTestActor); ok { if r, ok := a.Actor.(*runTestActor); ok {
// Inject a barrier task between the run action and its dependencies.
// This barrier task wil also depend on the previous barrier task.
// This prevents the run task from being scheduled until all previous run dependencies have finished.
// The build graph will be augmented to look roughly like this:
// build("a") build("b") build("c")
// | | |
// barrier("a.test") -> barrier("b.test") -> barrier("c.test")
// | | |
// run("a.test") run("b.test") run("c.test")
barrier := &work.Action{
Mode: "test barrier",
Deps: slices.Clip(a.Deps),
}
if prevBarrier != nil {
barrier.Deps = append(barrier.Deps, prevBarrier)
}
a.Deps = []*work.Action{barrier}
prevBarrier = barrier
r.prev = ch r.prev = ch
ch = make(chan struct{}) ch = make(chan struct{})
r.next = ch r.next = ch
@ -1400,6 +1425,8 @@ func (lockedStdout) Write(b []byte) (int, error) {
func (r *runTestActor) Act(b *work.Builder, ctx context.Context, a *work.Action) error { func (r *runTestActor) Act(b *work.Builder, ctx context.Context, a *work.Action) error {
sh := b.Shell(a) sh := b.Shell(a)
barrierAction := a.Deps[0]
buildAction := barrierAction.Deps[0]
// Wait for previous test to get started and print its first json line. // Wait for previous test to get started and print its first json line.
select { select {
@ -1530,7 +1557,7 @@ func (r *runTestActor) Act(b *work.Builder, ctx context.Context, a *work.Action)
// we have different link inputs but the same final binary, // we have different link inputs but the same final binary,
// we still reuse the cached test result. // we still reuse the cached test result.
// c.saveOutput will store the result under both IDs. // c.saveOutput will store the result under both IDs.
r.c.tryCacheWithID(b, a, a.Deps[0].BuildContentID()) r.c.tryCacheWithID(b, a, buildAction.BuildContentID())
} }
if r.c.buf != nil { if r.c.buf != nil {
if stdout != &buf { if stdout != &buf {
@ -1581,7 +1608,7 @@ func (r *runTestActor) Act(b *work.Builder, ctx context.Context, a *work.Action)
// fresh copies of tools to test as part of the testing. // fresh copies of tools to test as part of the testing.
addToEnv = "GOCOVERDIR=" + gcd addToEnv = "GOCOVERDIR=" + gcd
} }
args := str.StringList(execCmd, a.Deps[0].BuiltTarget(), testlogArg, panicArg, fuzzArg, coverdirArg, testArgs) args := str.StringList(execCmd, buildAction.BuiltTarget(), testlogArg, panicArg, fuzzArg, coverdirArg, testArgs)
if testCoverProfile != "" { if testCoverProfile != "" {
// Write coverage to temporary profile, for merging later. // Write coverage to temporary profile, for merging later.
@ -1741,8 +1768,8 @@ func (r *runTestActor) Act(b *work.Builder, ctx context.Context, a *work.Action)
// tryCache is called just before the link attempt, // tryCache is called just before the link attempt,
// to see if the test result is cached and therefore the link is unneeded. // to see if the test result is cached and therefore the link is unneeded.
// It reports whether the result can be satisfied from cache. // It reports whether the result can be satisfied from cache.
func (c *runCache) tryCache(b *work.Builder, a *work.Action) bool { func (c *runCache) tryCache(b *work.Builder, a *work.Action, linkAction *work.Action) bool {
return c.tryCacheWithID(b, a, a.Deps[0].BuildActionID()) return c.tryCacheWithID(b, a, linkAction.BuildActionID())
} }
func (c *runCache) tryCacheWithID(b *work.Builder, a *work.Action, id string) bool { func (c *runCache) tryCacheWithID(b *work.Builder, a *work.Action, id string) bool {

2
src/cmd/go/internal/work/action.go
View file

@ -92,7 +92,7 @@ type Action struct {
buggyInstall bool // is this a buggy install (see -linkshared)? buggyInstall bool // is this a buggy install (see -linkshared)?
TryCache func(*Builder, *Action) bool // callback for cache bypass TryCache func(*Builder, *Action, *Action) bool // callback for cache bypass
CacheExecutable bool // Whether to cache executables produced by link steps CacheExecutable bool // Whether to cache executables produced by link steps

2
src/cmd/go/internal/work/build.go
View file

@ -568,6 +568,8 @@ Executables are installed in the directory named by the GOBIN environment
variable, which defaults to $GOPATH/bin or $HOME/go/bin if the GOPATH variable, which defaults to $GOPATH/bin or $HOME/go/bin if the GOPATH
environment variable is not set. Executables in $GOROOT environment variable is not set. Executables in $GOROOT
are installed in $GOROOT/bin or $GOTOOLDIR instead of $GOBIN. are installed in $GOROOT/bin or $GOTOOLDIR instead of $GOBIN.
Cross compiled binaries are installed in $GOOS_$GOARCH subdirectories
of the above.
If the arguments have version suffixes (like @latest or @v1.0.0), "go install" If the arguments have version suffixes (like @latest or @v1.0.0), "go install"
builds packages in module-aware mode, ignoring the go.mod file in the current builds packages in module-aware mode, ignoring the go.mod file in the current

21
src/cmd/go/internal/work/buildid.go
View file

@ -401,6 +401,25 @@ var (
stdlibRecompiledIncOnce = sync.OnceFunc(stdlibRecompiled.Inc) stdlibRecompiledIncOnce = sync.OnceFunc(stdlibRecompiled.Inc)
) )
// testRunAction returns the run action for a test given the link action
// for the test binary, if the only (non-test-barrier) action that depend
// on the link action is the run action.
func testRunAction(a *Action) *Action {
if len(a.triggers) != 1 || a.triggers[0].Mode != "test barrier" {
return nil
}
var runAction *Action
for _, t := range a.triggers[0].triggers {
if t.Mode == "test run" {
if runAction != nil {
return nil
}
runAction = t
}
}
return runAction
}
// useCache tries to satisfy the action a, which has action ID actionHash, // useCache tries to satisfy the action a, which has action ID actionHash,
// by using a cached result from an earlier build. // by using a cached result from an earlier build.
// If useCache decides that the cache can be used, it sets a.buildID // If useCache decides that the cache can be used, it sets a.buildID
@ -526,7 +545,7 @@ func (b *Builder) useCache(a *Action, actionHash cache.ActionID, target string,
// then to avoid the link step, report the link as up-to-date. // then to avoid the link step, report the link as up-to-date.
// We avoid the nested build ID problem in the previous special case // We avoid the nested build ID problem in the previous special case
// by recording the test results in the cache under the action ID half. // by recording the test results in the cache under the action ID half.
if len(a.triggers) == 1 && a.triggers[0].TryCache != nil && a.triggers[0].TryCache(b, a.triggers[0]) { if ra := testRunAction(a); ra != nil && ra.TryCache != nil && ra.TryCache(b, ra, a) {
// Best effort attempt to display output from the compile and link steps. // Best effort attempt to display output from the compile and link steps.
// If it doesn't work, it doesn't work: reusing the test result is more // If it doesn't work, it doesn't work: reusing the test result is more
// important than reprinting diagnostic information. // important than reprinting diagnostic information.

5
src/cmd/go/internal/work/cover.go
View file

@ -36,8 +36,9 @@ func (b *Builder) CovData(a *Action, cmdargs ...any) ([]byte, error) {
// but will be empty; in this case the return is an empty string. // but will be empty; in this case the return is an empty string.
func BuildActionCoverMetaFile(runAct *Action) (string, error) { func BuildActionCoverMetaFile(runAct *Action) (string, error) {
p := runAct.Package p := runAct.Package
for i := range runAct.Deps { barrierAct := runAct.Deps[0]
pred := runAct.Deps[i] for i := range barrierAct.Deps {
pred := barrierAct.Deps[i]
if pred.Mode != "build" || pred.Package == nil { if pred.Mode != "build" || pred.Package == nil {
continue continue
} }

16
src/cmd/go/internal/work/exec.go
View file

@ -183,7 +183,21 @@ func (b *Builder) Do(ctx context.Context, root *Action) {
for _, a0 := range a.triggers { for _, a0 := range a.triggers {
if a.Failed != nil { if a.Failed != nil {
a0.Failed = a.Failed if a0.Mode == "test barrier" {
// If this action was triggered by a test, there
// will be a test barrier action in between the test
// and the true trigger. But there will be other
// triggers that are other barriers that are waiting
// for this one. Propagate the failure to the true
// trigger, but not to the other barriers.
for _, bt := range a0.triggers {
if bt.Mode != "test barrier" {
bt.Failed = a.Failed
}
}
} else {
a0.Failed = a.Failed
}
} }
if a0.pending--; a0.pending == 0 { if a0.pending--; a0.pending == 0 {
b.ready.push(a0) b.ready.push(a0)

8
src/cmd/internal/obj/arm64/asm7.go
View file

@ -892,8 +892,6 @@ var optab = []Optab{
{obj.ANOP, C_LCON, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0}, // nop variants, see #40689 {obj.ANOP, C_LCON, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0}, // nop variants, see #40689
{obj.ANOP, C_ZREG, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0}, {obj.ANOP, C_ZREG, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0},
{obj.ANOP, C_VREG, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0}, {obj.ANOP, C_VREG, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0},
{obj.ADUFFZERO, C_NONE, C_NONE, C_NONE, C_SBRA, C_NONE, 5, 4, 0, 0, 0}, // same as AB/ABL
{obj.ADUFFCOPY, C_NONE, C_NONE, C_NONE, C_SBRA, C_NONE, 5, 4, 0, 0, 0}, // same as AB/ABL
{obj.APCALIGN, C_LCON, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0}, // align code {obj.APCALIGN, C_LCON, C_NONE, C_NONE, C_NONE, C_NONE, 0, 0, 0, 0, 0}, // align code
{obj.APCALIGNMAX, C_LCON, C_NONE, C_NONE, C_LCON, C_NONE, 0, 0, 0, 0, 0}, // align code, conditional {obj.APCALIGNMAX, C_LCON, C_NONE, C_NONE, C_LCON, C_NONE, 0, 0, 0, 0, 0}, // align code, conditional
} }
@ -3297,9 +3295,7 @@ func buildop(ctxt *obj.Link) {
obj.AFUNCDATA, obj.AFUNCDATA,
obj.APCALIGN, obj.APCALIGN,
obj.APCALIGNMAX, obj.APCALIGNMAX,
obj.APCDATA, obj.APCDATA:
obj.ADUFFZERO,
obj.ADUFFCOPY:
break break
} }
} }
@ -6971,7 +6967,7 @@ func (c *ctxt7) opbra(p *obj.Prog, a obj.As) uint32 {
case AB: case AB:
return 0<<31 | 5<<26 /* imm26 */ return 0<<31 | 5<<26 /* imm26 */
case obj.ADUFFZERO, obj.ADUFFCOPY, ABL: case ABL:
return 1<<31 | 5<<26 return 1<<31 | 5<<26
} }

146
src/cmd/internal/obj/arm64/obj7.go
View file

@ -319,11 +319,7 @@ func progedit(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) {
// Rewrite BR/BL to symbol as TYPE_BRANCH. // Rewrite BR/BL to symbol as TYPE_BRANCH.
switch p.As { switch p.As {
case AB, case AB, ABL, obj.ARET:
ABL,
obj.ARET,
obj.ADUFFZERO,
obj.ADUFFCOPY:
if p.To.Sym != nil { if p.To.Sym != nil {
p.To.Type = obj.TYPE_BRANCH p.To.Type = obj.TYPE_BRANCH
} }
@ -400,39 +396,6 @@ func progedit(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) {
// Rewrite p, if necessary, to access global data via the global offset table. // Rewrite p, if necessary, to access global data via the global offset table.
func (c *ctxt7) rewriteToUseGot(p *obj.Prog) { func (c *ctxt7) rewriteToUseGot(p *obj.Prog) {
if p.As == obj.ADUFFCOPY || p.As == obj.ADUFFZERO {
// ADUFFxxx $offset
// becomes
// MOVD runtime.duffxxx@GOT, REGTMP
// ADD $offset, REGTMP
// CALL REGTMP
var sym *obj.LSym
if p.As == obj.ADUFFZERO {
sym = c.ctxt.LookupABI("runtime.duffzero", obj.ABIInternal)
} else {
sym = c.ctxt.LookupABI("runtime.duffcopy", obj.ABIInternal)
}
offset := p.To.Offset
p.As = AMOVD
p.From.Type = obj.TYPE_MEM
p.From.Name = obj.NAME_GOTREF
p.From.Sym = sym
p.To.Type = obj.TYPE_REG
p.To.Reg = REGTMP
p.To.Name = obj.NAME_NONE
p.To.Offset = 0
p.To.Sym = nil
p1 := obj.Appendp(p, c.newprog)
p1.As = AADD
p1.From.Type = obj.TYPE_CONST
p1.From.Offset = offset
p1.To.Type = obj.TYPE_REG
p1.To.Reg = REGTMP
p2 := obj.Appendp(p1, c.newprog)
p2.As = obj.ACALL
p2.To.Type = obj.TYPE_REG
p2.To.Reg = REGTMP
}
// We only care about global data: NAME_EXTERN means a global // We only care about global data: NAME_EXTERN means a global
// symbol in the Go sense, and p.Sym.Local is true for a few // symbol in the Go sense, and p.Sym.Local is true for a few
@ -543,9 +506,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
case obj.ATEXT: case obj.ATEXT:
p.Mark |= LEAF p.Mark |= LEAF
case ABL, case ABL:
obj.ADUFFZERO,
obj.ADUFFCOPY:
c.cursym.Func().Text.Mark &^= LEAF c.cursym.Func().Text.Mark &^= LEAF
} }
} }
@ -912,110 +873,7 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
p.From.Type = obj.TYPE_MEM p.From.Type = obj.TYPE_MEM
p.From.Reg = REGSP p.From.Reg = REGSP
} }
case obj.ADUFFCOPY:
// ADR ret_addr, R27
// STP (FP, R27), -24(SP)
// SUB 24, SP, FP
// DUFFCOPY
// ret_addr:
// SUB 8, SP, FP
q1 := p
// copy DUFFCOPY from q1 to q4
q4 := obj.Appendp(p, c.newprog)
q4.Pos = p.Pos
q4.As = obj.ADUFFCOPY
q4.To = p.To
q1.As = AADR
q1.From.Type = obj.TYPE_BRANCH
q1.To.Type = obj.TYPE_REG
q1.To.Reg = REG_R27
q2 := obj.Appendp(q1, c.newprog)
q2.Pos = p.Pos
q2.As = ASTP
q2.From.Type = obj.TYPE_REGREG
q2.From.Reg = REGFP
q2.From.Offset = int64(REG_R27)
q2.To.Type = obj.TYPE_MEM
q2.To.Reg = REGSP
q2.To.Offset = -24
// maintain FP for DUFFCOPY
q3 := obj.Appendp(q2, c.newprog)
q3.Pos = p.Pos
q3.As = ASUB
q3.From.Type = obj.TYPE_CONST
q3.From.Offset = 24
q3.Reg = REGSP
q3.To.Type = obj.TYPE_REG
q3.To.Reg = REGFP
q5 := obj.Appendp(q4, c.newprog)
q5.Pos = p.Pos
q5.As = ASUB
q5.From.Type = obj.TYPE_CONST
q5.From.Offset = 8
q5.Reg = REGSP
q5.To.Type = obj.TYPE_REG
q5.To.Reg = REGFP
q1.From.SetTarget(q5)
p = q5
case obj.ADUFFZERO:
// ADR ret_addr, R27
// STP (FP, R27), -24(SP)
// SUB 24, SP, FP
// DUFFZERO
// ret_addr:
// SUB 8, SP, FP
q1 := p
// copy DUFFZERO from q1 to q4
q4 := obj.Appendp(p, c.newprog)
q4.Pos = p.Pos
q4.As = obj.ADUFFZERO
q4.To = p.To
q1.As = AADR
q1.From.Type = obj.TYPE_BRANCH
q1.To.Type = obj.TYPE_REG
q1.To.Reg = REG_R27
q2 := obj.Appendp(q1, c.newprog)
q2.Pos = p.Pos
q2.As = ASTP
q2.From.Type = obj.TYPE_REGREG
q2.From.Reg = REGFP
q2.From.Offset = int64(REG_R27)
q2.To.Type = obj.TYPE_MEM
q2.To.Reg = REGSP
q2.To.Offset = -24
// maintain FP for DUFFZERO
q3 := obj.Appendp(q2, c.newprog)
q3.Pos = p.Pos
q3.As = ASUB
q3.From.Type = obj.TYPE_CONST
q3.From.Offset = 24
q3.Reg = REGSP
q3.To.Type = obj.TYPE_REG
q3.To.Reg = REGFP
q5 := obj.Appendp(q4, c.newprog)
q5.Pos = p.Pos
q5.As = ASUB
q5.From.Type = obj.TYPE_CONST
q5.From.Offset = 8
q5.Reg = REGSP
q5.To.Type = obj.TYPE_REG
q5.To.Reg = REGFP
q1.From.SetTarget(q5)
p = q5
} }
if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSP && p.Spadj == 0 { if p.To.Type == obj.TYPE_REG && p.To.Reg == REGSP && p.Spadj == 0 {
f := c.cursym.Func() f := c.cursym.Func()
if f.FuncFlag&abi.FuncFlagSPWrite == 0 { if f.FuncFlag&abi.FuncFlagSPWrite == 0 {

25
src/cmd/internal/obj/x86/asm6.go
View file

@ -4013,15 +4013,6 @@ func (ab *AsmBuf) mediaop(ctxt *obj.Link, o *Optab, op int, osize int, z int) in
return z return z
} }
var bpduff1 = []byte{
0x48, 0x89, 0x6c, 0x24, 0xf0, // MOVQ BP, -16(SP)
0x48, 0x8d, 0x6c, 0x24, 0xf0, // LEAQ -16(SP), BP
}
var bpduff2 = []byte{
0x48, 0x8b, 0x6d, 0x00, // MOVQ 0(BP), BP
}
// asmevex emits EVEX pregis and opcode byte. // asmevex emits EVEX pregis and opcode byte.
// In addition to asmvex r/m, vvvv and reg fields also requires optional // In addition to asmvex r/m, vvvv and reg fields also requires optional
// K-masking register. // K-masking register.
@ -4859,16 +4850,6 @@ func (ab *AsmBuf) doasm(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog) {
ctxt.Diag("directly calling duff when dynamically linking Go") ctxt.Diag("directly calling duff when dynamically linking Go")
} }
if yt.zcase == Zcallduff && ctxt.Arch.Family == sys.AMD64 {
// Maintain BP around call, since duffcopy/duffzero can't do it
// (the call jumps into the middle of the function).
// This makes it possible to see call sites for duffcopy/duffzero in
// BP-based profiling tools like Linux perf (which is the
// whole point of maintaining frame pointers in Go).
// MOVQ BP, -16(SP)
// LEAQ -16(SP), BP
ab.Put(bpduff1)
}
ab.Put1(byte(op)) ab.Put1(byte(op))
cursym.AddRel(ctxt, obj.Reloc{ cursym.AddRel(ctxt, obj.Reloc{
Type: objabi.R_CALL, Type: objabi.R_CALL,
@ -4879,12 +4860,6 @@ func (ab *AsmBuf) doasm(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog) {
}) })
ab.PutInt32(0) ab.PutInt32(0)
if yt.zcase == Zcallduff && ctxt.Arch.Family == sys.AMD64 {
// Pop BP pushed above.
// MOVQ 0(BP), BP
ab.Put(bpduff2)
}
// TODO: jump across functions needs reloc // TODO: jump across functions needs reloc
case Zbr, Zjmp, Zloop: case Zbr, Zjmp, Zloop:
if p.As == AXBEGIN { if p.As == AXBEGIN {

1
src/errors/join.go
View file

@ -16,6 +16,7 @@ import (
// between each string. // between each string.
// //
// A non-nil error returned by Join implements the Unwrap() []error method. // A non-nil error returned by Join implements the Unwrap() []error method.
// The errors may be inspected with [Is] and [As].
func Join(errs ...error) error { func Join(errs ...error) error {
n := 0 n := 0
for _, err := range errs { for _, err := range errs {

46
src/internal/poll/fd_windows.go
View file

@ -622,12 +622,22 @@ func (fd *FD) Pread(b []byte, off int64) (int, error) {
fd.l.Lock() fd.l.Lock()
defer fd.l.Unlock() defer fd.l.Unlock()
curoffset, err := syscall.Seek(fd.Sysfd, 0, io.SeekCurrent) if fd.isBlocking {
if err != nil { curoffset, err := syscall.Seek(fd.Sysfd, 0, io.SeekCurrent)
return 0, err if err != nil {
return 0, err
}
defer syscall.Seek(fd.Sysfd, curoffset, io.SeekStart)
defer fd.setOffset(curoffset)
} else {
// Overlapped handles don't have the file pointer updated
// when performing I/O operations, so there is no need to
// call Seek to reset the file pointer.
// Also, some overlapped file handles don't support seeking.
// See https://go.dev/issues/74951.
curoffset := fd.offset
defer fd.setOffset(curoffset)
} }
defer syscall.Seek(fd.Sysfd, curoffset, io.SeekStart)
defer fd.setOffset(curoffset)
o := &fd.rop o := &fd.rop
o.InitBuf(b) o.InitBuf(b)
fd.setOffset(off) fd.setOffset(off)
@ -847,12 +857,22 @@ func (fd *FD) Pwrite(buf []byte, off int64) (int, error) {
fd.l.Lock() fd.l.Lock()
defer fd.l.Unlock() defer fd.l.Unlock()
curoffset, err := syscall.Seek(fd.Sysfd, 0, io.SeekCurrent) if fd.isBlocking {
if err != nil { curoffset, err := syscall.Seek(fd.Sysfd, 0, io.SeekCurrent)
return 0, err if err != nil {
return 0, err
}
defer syscall.Seek(fd.Sysfd, curoffset, io.SeekStart)
defer fd.setOffset(curoffset)
} else {
// Overlapped handles don't have the file pointer updated
// when performing I/O operations, so there is no need to
// call Seek to reset the file pointer.
// Also, some overlapped file handles don't support seeking.
// See https://go.dev/issues/74951.
curoffset := fd.offset
defer fd.setOffset(curoffset)
} }
defer syscall.Seek(fd.Sysfd, curoffset, io.SeekStart)
defer fd.setOffset(curoffset)
var ntotal int var ntotal int
for { for {
@ -1107,6 +1127,12 @@ func (fd *FD) Seek(offset int64, whence int) (int64, error) {
fd.l.Lock() fd.l.Lock()
defer fd.l.Unlock() defer fd.l.Unlock()
if !fd.isBlocking && whence == io.SeekCurrent {
// Windows doesn't keep the file pointer for overlapped file handles.
// We do it ourselves in case to account for any read or write
// operations that may have occurred.
offset += fd.offset
}
n, err := syscall.Seek(fd.Sysfd, offset, whence) n, err := syscall.Seek(fd.Sysfd, offset, whence)
fd.setOffset(n) fd.setOffset(n)
return n, err return n, err

4
src/internal/runtime/maps/table.go
View file

@ -8,6 +8,7 @@ package maps
import ( import (
"internal/abi" "internal/abi"
"internal/goarch" "internal/goarch"
"internal/runtime/math"
"unsafe" "unsafe"
) )
@ -127,8 +128,7 @@ func (t *table) maxGrowthLeft() uint16 {
// single-group tables, we could fill all slots. // single-group tables, we could fill all slots.
return t.capacity - 1 return t.capacity - 1
} else { } else {
if t.capacity*maxAvgGroupLoad < t.capacity { if t.capacity > math.MaxUint16/maxAvgGroupLoad {
// TODO(prattmic): Do something cleaner.
panic("overflow") panic("overflow")
} }
return (t.capacity * maxAvgGroupLoad) / abi.MapGroupSlots return (t.capacity * maxAvgGroupLoad) / abi.MapGroupSlots

1
src/internal/runtime/math/math.go
View file

@ -7,6 +7,7 @@ package math
import "internal/goarch" import "internal/goarch"
const ( const (
MaxUint16 = ^uint16(0)
MaxUint32 = ^uint32(0) MaxUint32 = ^uint32(0)
MaxUint64 = ^uint64(0) MaxUint64 = ^uint64(0)
MaxUintptr = ^uintptr(0) MaxUintptr = ^uintptr(0)

83
src/internal/synctest/synctest_test.go
View file

@ -383,57 +383,59 @@ func TestChannelMovedOutOfBubble(t *testing.T) {
for _, test := range []struct { for _, test := range []struct {
desc string desc string
f func(chan struct{}) f func(chan struct{})
wantPanic string wantFatal string
}{{ }{{
desc: "receive", desc: "receive",
f: func(ch chan struct{}) { f: func(ch chan struct{}) {
<-ch <-ch
}, },
wantPanic: "receive on synctest channel from outside bubble", wantFatal: "receive on synctest channel from outside bubble",
}, { }, {
desc: "send", desc: "send",
f: func(ch chan struct{}) { f: func(ch chan struct{}) {
ch <- struct{}{} ch <- struct{}{}
}, },
wantPanic: "send on synctest channel from outside bubble", wantFatal: "send on synctest channel from outside bubble",
}, { }, {
desc: "close", desc: "close",
f: func(ch chan struct{}) { f: func(ch chan struct{}) {
close(ch) close(ch)
}, },
wantPanic: "close of synctest channel from outside bubble", wantFatal: "close of synctest channel from outside bubble",
}} { }} {
t.Run(test.desc, func(t *testing.T) { t.Run(test.desc, func(t *testing.T) {
// Bubbled channel accessed from outside any bubble. // Bubbled channel accessed from outside any bubble.
t.Run("outside_bubble", func(t *testing.T) { t.Run("outside_bubble", func(t *testing.T) {
donec := make(chan struct{}) wantFatal(t, test.wantFatal, func() {
ch := make(chan chan struct{}) donec := make(chan struct{})
go func() { ch := make(chan chan struct{})
defer close(donec) go func() {
defer wantPanic(t, test.wantPanic) defer close(donec)
test.f(<-ch) test.f(<-ch)
}() }()
synctest.Run(func() { synctest.Run(func() {
ch <- make(chan struct{}) ch <- make(chan struct{})
})
<-donec
}) })
<-donec
}) })
// Bubbled channel accessed from a different bubble. // Bubbled channel accessed from a different bubble.
t.Run("different_bubble", func(t *testing.T) { t.Run("different_bubble", func(t *testing.T) {
donec := make(chan struct{}) wantFatal(t, test.wantFatal, func() {
ch := make(chan chan struct{}) donec := make(chan struct{})
go func() { ch := make(chan chan struct{})
defer close(donec) go func() {
c := <-ch defer close(donec)
c := <-ch
synctest.Run(func() {
test.f(c)
})
}()
synctest.Run(func() { synctest.Run(func() {
defer wantPanic(t, test.wantPanic) ch <- make(chan struct{})
test.f(c)
}) })
}() <-donec
synctest.Run(func() {
ch <- make(chan struct{})
}) })
<-donec
}) })
}) })
} }
@ -443,39 +445,40 @@ func TestTimerFromInsideBubble(t *testing.T) {
for _, test := range []struct { for _, test := range []struct {
desc string desc string
f func(tm *time.Timer) f func(tm *time.Timer)
wantPanic string wantFatal string
}{{ }{{
desc: "read channel", desc: "read channel",
f: func(tm *time.Timer) { f: func(tm *time.Timer) {
<-tm.C <-tm.C
}, },
wantPanic: "receive on synctest channel from outside bubble", wantFatal: "receive on synctest channel from outside bubble",
}, { }, {
desc: "Reset", desc: "Reset",
f: func(tm *time.Timer) { f: func(tm *time.Timer) {
tm.Reset(1 * time.Second) tm.Reset(1 * time.Second)
}, },
wantPanic: "reset of synctest timer from outside bubble", wantFatal: "reset of synctest timer from outside bubble",
}, { }, {
desc: "Stop", desc: "Stop",
f: func(tm *time.Timer) { f: func(tm *time.Timer) {
tm.Stop() tm.Stop()
}, },
wantPanic: "stop of synctest timer from outside bubble", wantFatal: "stop of synctest timer from outside bubble",
}} { }} {
t.Run(test.desc, func(t *testing.T) { t.Run(test.desc, func(t *testing.T) {
donec := make(chan struct{}) wantFatal(t, test.wantFatal, func() {
ch := make(chan *time.Timer) donec := make(chan struct{})
go func() { ch := make(chan *time.Timer)
defer close(donec) go func() {
defer wantPanic(t, test.wantPanic) defer close(donec)
test.f(<-ch) test.f(<-ch)
}() }()
synctest.Run(func() { synctest.Run(func() {
tm := time.NewTimer(1 * time.Second) tm := time.NewTimer(1 * time.Second)
ch <- tm ch <- tm
})
<-donec
}) })
<-donec
}) })
} }
} }

7
src/internal/trace/batch.go
View file

@ -44,6 +44,10 @@ func (b *batch) isSyncBatch(ver version.Version) bool {
(tracev2.EventType(b.data[0]) == tracev2.EvSync && ver >= version.Go125)) (tracev2.EventType(b.data[0]) == tracev2.EvSync && ver >= version.Go125))
} }
func (b *batch) isEndOfGeneration() bool {
return b.exp == tracev2.NoExperiment && len(b.data) > 0 && tracev2.EventType(b.data[0]) == tracev2.EvEndOfGeneration
}
// readBatch reads the next full batch from r. // readBatch reads the next full batch from r.
func readBatch(r interface { func readBatch(r interface {
io.Reader io.Reader
@ -54,6 +58,9 @@ func readBatch(r interface {
if err != nil { if err != nil {
return batch{}, 0, err return batch{}, 0, err
} }
if typ := tracev2.EventType(b); typ == tracev2.EvEndOfGeneration {
return batch{m: NoThread, exp: tracev2.NoExperiment, data: []byte{b}}, 0, nil
}
if typ := tracev2.EventType(b); typ != tracev2.EvEventBatch && typ != tracev2.EvExperimentalBatch { if typ := tracev2.EventType(b); typ != tracev2.EvEventBatch && typ != tracev2.EvExperimentalBatch {
return batch{}, 0, fmt.Errorf("expected batch event, got event %d", typ) return batch{}, 0, fmt.Errorf("expected batch event, got event %d", typ)
} }

4
src/internal/trace/event.go
View file

@ -843,7 +843,6 @@ func (e Event) String() string {
fmt.Fprintf(&sb, " Task=%d Category=%q Message=%q", l.Task, l.Category, l.Message) fmt.Fprintf(&sb, " Task=%d Category=%q Message=%q", l.Task, l.Category, l.Message)
case EventStateTransition: case EventStateTransition:
s := e.StateTransition() s := e.StateTransition()
fmt.Fprintf(&sb, " Resource=%s Reason=%q", s.Resource, s.Reason)
switch s.Resource.Kind { switch s.Resource.Kind {
case ResourceGoroutine: case ResourceGoroutine:
id := s.Resource.Goroutine() id := s.Resource.Goroutine()
@ -854,6 +853,7 @@ func (e Event) String() string {
old, new := s.Proc() old, new := s.Proc()
fmt.Fprintf(&sb, " ProcID=%d %s->%s", id, old, new) fmt.Fprintf(&sb, " ProcID=%d %s->%s", id, old, new)
} }
fmt.Fprintf(&sb, " Reason=%q", s.Reason)
if s.Stack != NoStack { if s.Stack != NoStack {
fmt.Fprintln(&sb) fmt.Fprintln(&sb)
fmt.Fprintln(&sb, "TransitionStack=") fmt.Fprintln(&sb, "TransitionStack=")
@ -879,7 +879,7 @@ func (e Event) String() string {
fmt.Fprintf(&sb, " Trace=%d Mono=%d Wall=%s", fmt.Fprintf(&sb, " Trace=%d Mono=%d Wall=%s",
s.ClockSnapshot.Trace, s.ClockSnapshot.Trace,
s.ClockSnapshot.Mono, s.ClockSnapshot.Mono,
s.ClockSnapshot.Wall.Format(time.RFC3339), s.ClockSnapshot.Wall.Format(time.RFC3339Nano),
) )
} }
} }

119
src/internal/trace/generation.go
View file

@ -9,6 +9,7 @@ import (
"bytes" "bytes"
"cmp" "cmp"
"encoding/binary" "encoding/binary"
"errors"
"fmt" "fmt"
"io" "io"
"slices" "slices"
@ -32,22 +33,102 @@ type generation struct {
*evTable *evTable
} }
// readGeneration buffers and decodes the structural elements of a trace generation
// out of r.
func readGeneration(r *bufio.Reader, ver version.Version) (*generation, error) {
if ver < version.Go126 {
return nil, errors.New("internal error: readGeneration called for <1.26 trace")
}
g := &generation{
evTable: &evTable{
pcs: make(map[uint64]frame),
},
batches: make(map[ThreadID][]batch),
}
// Read batches one at a time until we either hit the next generation.
for {
b, gen, err := readBatch(r)
if err == io.EOF {
if len(g.batches) != 0 {
return nil, errors.New("incomplete generation found; trace likely truncated")
}
return nil, nil // All done.
}
if err != nil {
return nil, err
}
if g.gen == 0 {
// Initialize gen.
g.gen = gen
}
if b.isEndOfGeneration() {
break
}
if gen == 0 {
// 0 is a sentinel used by the runtime, so we'll never see it.
return nil, fmt.Errorf("invalid generation number %d", gen)
}
if gen != g.gen {
return nil, fmt.Errorf("broken trace: missing end-of-generation event, or generations are interleaved")
}
if g.minTs == 0 || b.time < g.minTs {
g.minTs = b.time
}
if err := processBatch(g, b, ver); err != nil {
return nil, err
}
}
// Check some invariants.
if g.freq == 0 {
return nil, fmt.Errorf("no frequency event found")
}
if !g.hasClockSnapshot {
return nil, fmt.Errorf("no clock snapshot event found")
}
// N.B. Trust that the batch order is correct. We can't validate the batch order
// by timestamp because the timestamps could just be plain wrong. The source of
// truth is the order things appear in the trace and the partial order sequence
// numbers on certain events. If it turns out the batch order is actually incorrect
// we'll very likely fail to advance a partial order from the frontier.
// Compactify stacks and strings for better lookup performance later.
g.stacks.compactify()
g.strings.compactify()
// Validate stacks.
if err := validateStackStrings(&g.stacks, &g.strings, g.pcs); err != nil {
return nil, err
}
// Now that we have the frequency, fix up CPU samples.
fixUpCPUSamples(g.cpuSamples, g.freq)
return g, nil
}
// spilledBatch represents a batch that was read out for the next generation, // spilledBatch represents a batch that was read out for the next generation,
// while reading the previous one. It's passed on when parsing the next // while reading the previous one. It's passed on when parsing the next
// generation. // generation.
//
// Used only for trace versions < Go126.
type spilledBatch struct { type spilledBatch struct {
gen uint64 gen uint64
*batch *batch
} }
// readGeneration buffers and decodes the structural elements of a trace generation // readGenerationWithSpill buffers and decodes the structural elements of a trace generation
// out of r. spill is the first batch of the new generation (already buffered and // out of r. spill is the first batch of the new generation (already buffered and
// parsed from reading the last generation). Returns the generation and the first // parsed from reading the last generation). Returns the generation and the first
// batch read of the next generation, if any. // batch read of the next generation, if any.
// //
// If gen is non-nil, it is valid and must be processed before handling the returned // If gen is non-nil, it is valid and must be processed before handling the returned
// error. // error.
func readGeneration(r *bufio.Reader, spill *spilledBatch, ver version.Version) (*generation, *spilledBatch, error) { func readGenerationWithSpill(r *bufio.Reader, spill *spilledBatch, ver version.Version) (*generation, *spilledBatch, error) {
if ver >= version.Go126 {
return nil, nil, errors.New("internal error: readGenerationWithSpill called for Go 1.26+ trace")
}
g := &generation{ g := &generation{
evTable: &evTable{ evTable: &evTable{
pcs: make(map[uint64]frame), pcs: make(map[uint64]frame),
@ -56,6 +137,7 @@ func readGeneration(r *bufio.Reader, spill *spilledBatch, ver version.Version) (
} }
// Process the spilled batch. // Process the spilled batch.
if spill != nil { if spill != nil {
// Process the spilled batch, which contains real data.
g.gen = spill.gen g.gen = spill.gen
g.minTs = spill.batch.time g.minTs = spill.batch.time
if err := processBatch(g, *spill.batch, ver); err != nil { if err := processBatch(g, *spill.batch, ver); err != nil {
@ -63,8 +145,7 @@ func readGeneration(r *bufio.Reader, spill *spilledBatch, ver version.Version) (
} }
spill = nil spill = nil
} }
// Read batches one at a time until we either hit EOF or // Read batches one at a time until we either hit the next generation.
// the next generation.
var spillErr error var spillErr error
for { for {
b, gen, err := readBatch(r) b, gen, err := readBatch(r)
@ -73,7 +154,7 @@ func readGeneration(r *bufio.Reader, spill *spilledBatch, ver version.Version) (
} }
if err != nil { if err != nil {
if g.gen != 0 { if g.gen != 0 {
// This is an error reading the first batch of the next generation. // This may be an error reading the first batch of the next generation.
// This is fine. Let's forge ahead assuming that what we've got so // This is fine. Let's forge ahead assuming that what we've got so
// far is fine. // far is fine.
spillErr = err spillErr = err
@ -89,7 +170,8 @@ func readGeneration(r *bufio.Reader, spill *spilledBatch, ver version.Version) (
// Initialize gen. // Initialize gen.
g.gen = gen g.gen = gen
} }
if gen == g.gen+1 { // TODO: advance this the same way the runtime does. if gen == g.gen+1 {
// TODO: Increment the generation with wraparound the same way the runtime does.
spill = &spilledBatch{gen: gen, batch: &b} spill = &spilledBatch{gen: gen, batch: &b}
break break
} }
@ -134,15 +216,8 @@ func readGeneration(r *bufio.Reader, spill *spilledBatch, ver version.Version) (
return nil, nil, err return nil, nil, err
} }
// Fix up the CPU sample timestamps, now that we have freq. // Now that we have the frequency, fix up CPU samples.
for i := range g.cpuSamples { fixUpCPUSamples(g.cpuSamples, g.freq)
s := &g.cpuSamples[i]
s.time = g.freq.mul(timestamp(s.time))
}
// Sort the CPU samples.
slices.SortFunc(g.cpuSamples, func(a, b cpuSample) int {
return cmp.Compare(a.time, b.time)
})
return g, spill, spillErr return g, spill, spillErr
} }
@ -174,6 +249,8 @@ func processBatch(g *generation, b batch, ver version.Version) error {
if err := addExperimentalBatch(g.expBatches, b); err != nil { if err := addExperimentalBatch(g.expBatches, b); err != nil {
return err return err
} }
case b.isEndOfGeneration():
return errors.New("internal error: unexpectedly processing EndOfGeneration; broken trace?")
default: default:
if _, ok := g.batches[b.m]; !ok { if _, ok := g.batches[b.m]; !ok {
g.batchMs = append(g.batchMs, b.m) g.batchMs = append(g.batchMs, b.m)
@ -512,3 +589,15 @@ func addExperimentalBatch(expBatches map[tracev2.Experiment][]ExperimentalBatch,
}) })
return nil return nil
} }
func fixUpCPUSamples(samples []cpuSample, freq frequency) {
// Fix up the CPU sample timestamps.
for i := range samples {
s := &samples[i]
s.time = freq.mul(timestamp(s.time))
}
// Sort the CPU samples.
slices.SortFunc(samples, func(a, b cpuSample) int {
return cmp.Compare(a.time, b.time)
})
}

8
src/internal/trace/internal/testgen/trace.go
View file

@ -322,6 +322,14 @@ func (g *Generation) writeEventsTo(tw *raw.TextWriter) {
} }
} }
b.writeEventsTo(tw) b.writeEventsTo(tw)
// Write end-of-generation event if necessary.
if g.trace.ver >= version.Go126 {
tw.WriteEvent(raw.Event{
Version: g.trace.ver,
Ev: tracev2.EvEndOfGeneration,
})
}
} }
func (g *Generation) newStructuralBatch() *Batch { func (g *Generation) newStructuralBatch() *Batch {

161
src/internal/trace/reader.go
View file

@ -6,6 +6,7 @@ package trace
import ( import (
"bufio" "bufio"
"errors"
"fmt" "fmt"
"io" "io"
"slices" "slices"
@ -22,18 +23,29 @@ import (
// event as the first event, and a Sync event as the last event. // event as the first event, and a Sync event as the last event.
// (There may also be any number of Sync events in the middle, too.) // (There may also be any number of Sync events in the middle, too.)
type Reader struct { type Reader struct {
version version.Version version version.Version
r *bufio.Reader r *bufio.Reader
lastTs Time lastTs Time
gen *generation gen *generation
frontier []*batchCursor
cpuSamples []cpuSample
order ordering
syncs int
readGenErr error
done bool
// Spill state.
//
// Traces before Go 1.26 had no explicit end-of-generation signal, and
// so the first batch of the next generation needed to be parsed to identify
// a new generation. This batch is the "spilled" so we don't lose track
// of it when parsing the next generation.
//
// This is unnecessary after Go 1.26 because of an explicit end-of-generation
// signal.
spill *spilledBatch spill *spilledBatch
spillErr error // error from reading spill spillErr error // error from reading spill
spillErrSync bool // whether we emitted a Sync before reporting spillErr spillErrSync bool // whether we emitted a Sync before reporting spillErr
frontier []*batchCursor
cpuSamples []cpuSample
order ordering
syncs int
done bool
v1Events *traceV1Converter v1Events *traceV1Converter
} }
@ -54,7 +66,7 @@ func NewReader(r io.Reader) (*Reader, error) {
return &Reader{ return &Reader{
v1Events: convertV1Trace(tr), v1Events: convertV1Trace(tr),
}, nil }, nil
case version.Go122, version.Go123, version.Go125: case version.Go122, version.Go123, version.Go125, version.Go126:
return &Reader{ return &Reader{
version: v, version: v,
r: br, r: br,
@ -139,52 +151,23 @@ func (r *Reader) ReadEvent() (e Event, err error) {
// Check if we need to refresh the generation. // Check if we need to refresh the generation.
if len(r.frontier) == 0 && len(r.cpuSamples) == 0 { if len(r.frontier) == 0 && len(r.cpuSamples) == 0 {
if r.spillErr != nil { if r.version < version.Go126 {
if r.spillErrSync { return r.nextGenWithSpill()
return Event{}, r.spillErr }
} if r.readGenErr != nil {
r.spillErrSync = true return Event{}, r.readGenErr
}
gen, err := readGeneration(r.r, r.version)
if err != nil {
// Before returning an error, emit the sync event
// for the current generation and queue up the error
// for the next call.
r.readGenErr = err
r.gen = nil
r.syncs++ r.syncs++
return syncEvent(nil, r.lastTs, r.syncs), nil return syncEvent(nil, r.lastTs, r.syncs), nil
} }
if r.gen != nil && r.spill == nil { return r.installGen(gen)
// If we have a generation from the last read,
// and there's nothing left in the frontier, and
// there's no spilled batch, indicating that there's
// no further generation, it means we're done.
// Emit the final sync event.
r.done = true
r.syncs++
return syncEvent(nil, r.lastTs, r.syncs), nil
}
// Read the next generation.
r.gen, r.spill, r.spillErr = readGeneration(r.r, r.spill, r.version)
if r.gen == nil {
r.spillErrSync = true
r.syncs++
return syncEvent(nil, r.lastTs, r.syncs), nil
}
// Reset CPU samples cursor.
r.cpuSamples = r.gen.cpuSamples
// Reset frontier.
for _, m := range r.gen.batchMs {
batches := r.gen.batches[m]
bc := &batchCursor{m: m}
ok, err := bc.nextEvent(batches, r.gen.freq)
if err != nil {
return Event{}, err
}
if !ok {
// Turns out there aren't actually any events in these batches.
continue
}
r.frontier = heapInsert(r.frontier, bc)
}
r.syncs++
// Always emit a sync event at the beginning of the generation.
return syncEvent(r.gen.evTable, r.gen.freq.mul(r.gen.minTs), r.syncs), nil
} }
tryAdvance := func(i int) (bool, error) { tryAdvance := func(i int) (bool, error) {
bc := r.frontier[i] bc := r.frontier[i]
@ -251,6 +234,78 @@ func (r *Reader) ReadEvent() (e Event, err error) {
return ev, nil return ev, nil
} }
// nextGenWithSpill reads the generation and calls nextGen while
// also handling any spilled batches.
func (r *Reader) nextGenWithSpill() (Event, error) {
if r.version >= version.Go126 {
return Event{}, errors.New("internal error: nextGenWithSpill called for Go 1.26+ trace")
}
if r.spillErr != nil {
if r.spillErrSync {
return Event{}, r.spillErr
}
r.spillErrSync = true
r.syncs++
return syncEvent(nil, r.lastTs, r.syncs), nil
}
if r.gen != nil && r.spill == nil {
// If we have a generation from the last read,
// and there's nothing left in the frontier, and
// there's no spilled batch, indicating that there's
// no further generation, it means we're done.
// Emit the final sync event.
r.done = true
r.syncs++
return syncEvent(nil, r.lastTs, r.syncs), nil
}
// Read the next generation.
var gen *generation
gen, r.spill, r.spillErr = readGenerationWithSpill(r.r, r.spill, r.version)
if gen == nil {
r.gen = nil
r.spillErrSync = true
r.syncs++
return syncEvent(nil, r.lastTs, r.syncs), nil
}
return r.installGen(gen)
}
// installGen installs the new generation into the Reader and returns
// a Sync event for the new generation.
func (r *Reader) installGen(gen *generation) (Event, error) {
if gen == nil {
// Emit the final sync event.
r.gen = nil
r.done = true
r.syncs++
return syncEvent(nil, r.lastTs, r.syncs), nil
}
r.gen = gen
// Reset CPU samples cursor.
r.cpuSamples = r.gen.cpuSamples
// Reset frontier.
for _, m := range r.gen.batchMs {
batches := r.gen.batches[m]
bc := &batchCursor{m: m}
ok, err := bc.nextEvent(batches, r.gen.freq)
if err != nil {
return Event{}, err
}
if !ok {
// Turns out there aren't actually any events in these batches.
continue
}
r.frontier = heapInsert(r.frontier, bc)
}
r.syncs++
// Always emit a sync event at the beginning of the generation.
return syncEvent(r.gen.evTable, r.gen.freq.mul(r.gen.minTs), r.syncs), nil
}
func dumpFrontier(frontier []*batchCursor) string { func dumpFrontier(frontier []*batchCursor) string {
var sb strings.Builder var sb strings.Builder
for _, bc := range frontier { for _, bc := range frontier {

4
src/internal/trace/tracev2/events.go
View file

@ -87,8 +87,8 @@ const (
EvSync // start of a sync batch [...EvFrequency|EvClockSnapshot] EvSync // start of a sync batch [...EvFrequency|EvClockSnapshot]
EvClockSnapshot // snapshot of trace, mono and wall clocks [timestamp, mono, sec, nsec] EvClockSnapshot // snapshot of trace, mono and wall clocks [timestamp, mono, sec, nsec]
// Reserved internal in-band end-of-generation signal. Must never appear in the trace. Added in Go 1.25. // In-band end-of-generation signal. Added in Go 1.26.
// This could be used as an explicit in-band end-of-generation signal in the future. // Used in Go 1.25 only internally.
EvEndOfGeneration EvEndOfGeneration
NumEvents NumEvents

6
src/internal/trace/version/version.go
View file

@ -21,7 +21,8 @@ const (
Go122 Version = 22 // v2 Go122 Version = 22 // v2
Go123 Version = 23 // v2 Go123 Version = 23 // v2
Go125 Version = 25 // v2 Go125 Version = 25 // v2
Current = Go125 Go126 Version = 26 // v2
Current = Go126
) )
var versions = map[Version][]tracev2.EventSpec{ var versions = map[Version][]tracev2.EventSpec{
@ -33,7 +34,8 @@ var versions = map[Version][]tracev2.EventSpec{
Go122: tracev2.Specs()[:tracev2.EvUserLog+1], // All events after are Go 1.23+. Go122: tracev2.Specs()[:tracev2.EvUserLog+1], // All events after are Go 1.23+.
Go123: tracev2.Specs()[:tracev2.EvExperimentalBatch+1], // All events after are Go 1.25+. Go123: tracev2.Specs()[:tracev2.EvExperimentalBatch+1], // All events after are Go 1.25+.
Go125: tracev2.Specs(), Go125: tracev2.Specs()[:tracev2.EvClockSnapshot+1], // All events after are Go 1.26+.
Go126: tracev2.Specs(),
} }
// Specs returns the set of event.Specs for this version. // Specs returns the set of event.Specs for this version.

5
src/net/udpsock_test.go
View file

@ -710,6 +710,11 @@ func TestIPv6WriteMsgUDPAddrPortTargetAddrIPVersion(t *testing.T) {
// WriteMsgUDPAddrPort accepts IPv4 and IPv4-mapped IPv6 destination addresses, // WriteMsgUDPAddrPort accepts IPv4 and IPv4-mapped IPv6 destination addresses,
// and rejects IPv6 destination addresses on a "udp4" connection. // and rejects IPv6 destination addresses on a "udp4" connection.
func TestIPv4WriteMsgUDPAddrPortTargetAddrIPVersion(t *testing.T) { func TestIPv4WriteMsgUDPAddrPortTargetAddrIPVersion(t *testing.T) {
switch runtime.GOOS {
case "plan9":
t.Skipf("not supported on %s", runtime.GOOS)
}
if !testableNetwork("udp4") { if !testableNetwork("udp4") {
t.Skipf("skipping: udp4 not available") t.Skipf("skipping: udp4 not available")
} }

66
src/os/os_windows_test.go
View file

@ -1845,6 +1845,72 @@ func TestFile(t *testing.T) {
} }
} }
func TestFileOverlappedSeek(t *testing.T) {
t.Parallel()
name := filepath.Join(t.TempDir(), "foo")
f := newFileOverlapped(t, name, true)
content := []byte("foo")
if _, err := f.Write(content); err != nil {
t.Fatal(err)
}
// Check that the file pointer is at the expected offset.
n, err := f.Seek(0, io.SeekCurrent)
if err != nil {
t.Fatal(err)
}
if n != int64(len(content)) {
t.Errorf("expected file pointer to be at offset %d, got %d", len(content), n)
}
// Set the file pointer to the start of the file.
if _, err := f.Seek(0, io.SeekStart); err != nil {
t.Fatal(err)
}
// Read the first byte.
var buf [1]byte
if _, err := f.Read(buf[:]); err != nil {
t.Fatal(err)
}
if !bytes.Equal(buf[:], content[:len(buf)]) {
t.Errorf("expected %q, got %q", content[:len(buf)], buf[:])
}
// Check that the file pointer is at the expected offset.
n, err = f.Seek(0, io.SeekCurrent)
if err != nil {
t.Fatal(err)
}
if n != int64(len(buf)) {
t.Errorf("expected file pointer to be at offset %d, got %d", len(buf), n)
}
}
func TestFileOverlappedReadAtVolume(t *testing.T) {
// Test that we can use File.ReadAt with an overlapped volume handle.
// See https://go.dev/issues/74951.
t.Parallel()
name := `\\.\` + filepath.VolumeName(t.TempDir())
namep, err := syscall.UTF16PtrFromString(name)
if err != nil {
t.Fatal(err)
}
h, err := syscall.CreateFile(namep,
syscall.GENERIC_READ|syscall.GENERIC_WRITE,
syscall.FILE_SHARE_WRITE|syscall.FILE_SHARE_READ,
nil, syscall.OPEN_ALWAYS, syscall.FILE_FLAG_OVERLAPPED, 0)
if err != nil {
if errors.Is(err, syscall.ERROR_ACCESS_DENIED) {
t.Skip("skipping test: access denied")
}
t.Fatal(err)
}
f := os.NewFile(uintptr(h), name)
defer f.Close()
var buf [0]byte
if _, err := f.ReadAt(buf[:], 0); err != nil {
t.Fatal(err)
}
}
func TestPipe(t *testing.T) { func TestPipe(t *testing.T) {
t.Parallel() t.Parallel()
r, w, err := os.Pipe() r, w, err := os.Pipe()

2
src/os/tempfile.go
View file

@ -105,7 +105,7 @@ func MkdirTemp(dir, pattern string) (string, error) {
if try++; try < 10000 { if try++; try < 10000 {
continue continue
} }
return "", &PathError{Op: "mkdirtemp", Path: dir + string(PathSeparator) + prefix + "*" + suffix, Err: ErrExist} return "", &PathError{Op: "mkdirtemp", Path: prefix + "*" + suffix, Err: ErrExist}
} }
if IsNotExist(err) { if IsNotExist(err) {
if _, err := Stat(dir); IsNotExist(err) { if _, err := Stat(dir); IsNotExist(err) {

10
src/runtime/chan.go
View file

@ -191,7 +191,7 @@ func chansend(c *hchan, ep unsafe.Pointer, block bool, callerpc uintptr) bool {
} }
if c.bubble != nil && getg().bubble != c.bubble { if c.bubble != nil && getg().bubble != c.bubble {
panic(plainError("send on synctest channel from outside bubble")) fatal("send on synctest channel from outside bubble")
} }
// Fast path: check for failed non-blocking operation without acquiring the lock. // Fast path: check for failed non-blocking operation without acquiring the lock.
@ -318,7 +318,7 @@ func chansend(c *hchan, ep unsafe.Pointer, block bool, callerpc uintptr) bool {
func send(c *hchan, sg *sudog, ep unsafe.Pointer, unlockf func(), skip int) { func send(c *hchan, sg *sudog, ep unsafe.Pointer, unlockf func(), skip int) {
if c.bubble != nil && getg().bubble != c.bubble { if c.bubble != nil && getg().bubble != c.bubble {
unlockf() unlockf()
panic(plainError("send on synctest channel from outside bubble")) fatal("send on synctest channel from outside bubble")
} }
if raceenabled { if raceenabled {
if c.dataqsiz == 0 { if c.dataqsiz == 0 {
@ -416,7 +416,7 @@ func closechan(c *hchan) {
panic(plainError("close of nil channel")) panic(plainError("close of nil channel"))
} }
if c.bubble != nil && getg().bubble != c.bubble { if c.bubble != nil && getg().bubble != c.bubble {
panic(plainError("close of synctest channel from outside bubble")) fatal("close of synctest channel from outside bubble")
} }
lock(&c.lock) lock(&c.lock)
@ -538,7 +538,7 @@ func chanrecv(c *hchan, ep unsafe.Pointer, block bool) (selected, received bool)
} }
if c.bubble != nil && getg().bubble != c.bubble { if c.bubble != nil && getg().bubble != c.bubble {
panic(plainError("receive on synctest channel from outside bubble")) fatal("receive on synctest channel from outside bubble")
} }
if c.timer != nil { if c.timer != nil {
@ -702,7 +702,7 @@ func chanrecv(c *hchan, ep unsafe.Pointer, block bool) (selected, received bool)
func recv(c *hchan, sg *sudog, ep unsafe.Pointer, unlockf func(), skip int) { func recv(c *hchan, sg *sudog, ep unsafe.Pointer, unlockf func(), skip int) {
if c.bubble != nil && getg().bubble != c.bubble { if c.bubble != nil && getg().bubble != c.bubble {
unlockf() unlockf()
panic(plainError("receive on synctest channel from outside bubble")) fatal("receive on synctest channel from outside bubble")
} }
if c.dataqsiz == 0 { if c.dataqsiz == 0 {
if raceenabled { if raceenabled {

2
src/runtime/debug.go
View file

@ -177,7 +177,7 @@ func totalMutexWaitTimeNanos() int64 {
// NumGoroutine returns the number of goroutines that currently exist. // NumGoroutine returns the number of goroutines that currently exist.
func NumGoroutine() int { func NumGoroutine() int {
return int(gcount()) return int(gcount(false))
} }
//go:linkname debug_modinfo runtime/debug.modinfo //go:linkname debug_modinfo runtime/debug.modinfo

427
src/runtime/duff_amd64.s
View file

@ -1,427 +0,0 @@
// Code generated by mkduff.go; DO NOT EDIT.
// Run go generate from src/runtime to update.
// See mkduff.go for comments.
#include "textflag.h"
TEXT runtime·duffzero<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-0
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
MOVUPS X15,(DI)
MOVUPS X15,16(DI)
MOVUPS X15,32(DI)
MOVUPS X15,48(DI)
LEAQ 64(DI),DI
RET
TEXT runtime·duffcopy<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-0
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
MOVUPS (SI), X0
ADDQ $16, SI
MOVUPS X0, (DI)
ADDQ $16, DI
RET

267
src/runtime/duff_arm64.s
View file

@ -1,267 +0,0 @@
// Code generated by mkduff.go; DO NOT EDIT.
// Run go generate from src/runtime to update.
// See mkduff.go for comments.
#include "textflag.h"
TEXT runtime·duffzero<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-0
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP.P (ZR, ZR), 16(R20)
STP (ZR, ZR), (R20)
RET
TEXT runtime·duffcopy<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-0
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
LDP.P 16(R20), (R26, R27)
STP.P (R26, R27), 16(R21)
RET

134
src/runtime/metrics.go
View file

@ -8,6 +8,7 @@ package runtime
import ( import (
"internal/godebugs" "internal/godebugs"
"internal/runtime/atomic"
"internal/runtime/gc" "internal/runtime/gc"
"unsafe" "unsafe"
) )
@ -465,9 +466,45 @@ func initMetrics() {
}, },
}, },
"/sched/goroutines:goroutines": { "/sched/goroutines:goroutines": {
compute: func(_ *statAggregate, out *metricValue) { deps: makeStatDepSet(schedStatsDep),
compute: func(in *statAggregate, out *metricValue) {
out.kind = metricKindUint64 out.kind = metricKindUint64
out.scalar = uint64(gcount()) out.scalar = uint64(in.schedStats.gTotal)
},
},
"/sched/goroutines/not-in-go:goroutines": {
deps: makeStatDepSet(schedStatsDep),
compute: func(in *statAggregate, out *metricValue) {
out.kind = metricKindUint64
out.scalar = uint64(in.schedStats.gNonGo)
},
},
"/sched/goroutines/running:goroutines": {
deps: makeStatDepSet(schedStatsDep),
compute: func(in *statAggregate, out *metricValue) {
out.kind = metricKindUint64
out.scalar = uint64(in.schedStats.gRunning)
},
},
"/sched/goroutines/runnable:goroutines": {
deps: makeStatDepSet(schedStatsDep),
compute: func(in *statAggregate, out *metricValue) {
out.kind = metricKindUint64
out.scalar = uint64(in.schedStats.gRunnable)
},
},
"/sched/goroutines/waiting:goroutines": {
deps: makeStatDepSet(schedStatsDep),
compute: func(in *statAggregate, out *metricValue) {
out.kind = metricKindUint64
out.scalar = uint64(in.schedStats.gWaiting)
},
},
"/sched/goroutines-created:goroutines": {
deps: makeStatDepSet(schedStatsDep),
compute: func(in *statAggregate, out *metricValue) {
out.kind = metricKindUint64
out.scalar = uint64(in.schedStats.gCreated)
}, },
}, },
"/sched/latencies:seconds": { "/sched/latencies:seconds": {
@ -495,6 +532,13 @@ func initMetrics() {
sched.stwTotalTimeOther.write(out) sched.stwTotalTimeOther.write(out)
}, },
}, },
"/sched/threads/total:threads": {
deps: makeStatDepSet(schedStatsDep),
compute: func(in *statAggregate, out *metricValue) {
out.kind = metricKindUint64
out.scalar = uint64(in.schedStats.threads)
},
},
"/sync/mutex/wait/total:seconds": { "/sync/mutex/wait/total:seconds": {
compute: func(_ *statAggregate, out *metricValue) { compute: func(_ *statAggregate, out *metricValue) {
out.kind = metricKindFloat64 out.kind = metricKindFloat64
@ -547,6 +591,7 @@ const (
cpuStatsDep // corresponds to cpuStatsAggregate cpuStatsDep // corresponds to cpuStatsAggregate
gcStatsDep // corresponds to gcStatsAggregate gcStatsDep // corresponds to gcStatsAggregate
finalStatsDep // corresponds to finalStatsAggregate finalStatsDep // corresponds to finalStatsAggregate
schedStatsDep // corresponds to schedStatsAggregate
numStatsDeps numStatsDeps
) )
@ -740,6 +785,88 @@ func (a *finalStatsAggregate) compute() {
a.cleanupsQueued, a.cleanupsExecuted = gcCleanups.readQueueStats() a.cleanupsQueued, a.cleanupsExecuted = gcCleanups.readQueueStats()
} }
// schedStatsAggregate contains stats about the scheduler, including
// an approximate count of goroutines in each state.
type schedStatsAggregate struct {
gTotal uint64
gRunning uint64
gRunnable uint64
gNonGo uint64
gWaiting uint64
gCreated uint64
threads uint64
}
// compute populates the schedStatsAggregate with values from the runtime.
func (a *schedStatsAggregate) compute() {
// Lock the scheduler so the global run queue can't change and
// the number of Ps can't change. This doesn't prevent the
// local run queues from changing, so the results are still
// approximate.
lock(&sched.lock)
// The total count of threads owned by Go is the number of Ms
// minus extra Ms on the list or in use.
a.threads = uint64(mcount()) - uint64(extraMInUse.Load()) - uint64(extraMLength.Load())
// Collect running/runnable from per-P run queues.
a.gCreated += sched.goroutinesCreated.Load()
for _, p := range allp {
if p == nil || p.status == _Pdead {
break
}
a.gCreated += p.goroutinesCreated
switch p.status {
case _Prunning:
a.gRunning++
case _Psyscall:
a.gNonGo++
case _Pgcstop:
// The world is stopping or stopped.
// This is fine. The results will be
// slightly odd since nothing else
// is running, but it will be accurate.
}
for {
h := atomic.Load(&p.runqhead)
t := atomic.Load(&p.runqtail)
next := atomic.Loaduintptr((*uintptr)(&p.runnext))
runnable := int32(t - h)
if atomic.Load(&p.runqhead) != h || runnable < 0 {
continue
}
if next != 0 {
runnable++
}
a.gRunnable += uint64(runnable)
break
}
}
// Global run queue.
a.gRunnable += uint64(sched.runq.size)
// Account for Gs that are in _Gsyscall without a P in _Psyscall.
nGsyscallNoP := sched.nGsyscallNoP.Load()
// nGsyscallNoP can go negative during temporary races.
if nGsyscallNoP >= 0 {
a.gNonGo += uint64(nGsyscallNoP)
}
// Compute the number of blocked goroutines. We have to
// include system goroutines in this count because we included
// them above.
a.gTotal = uint64(gcount(true))
a.gWaiting = a.gTotal - (a.gRunning + a.gRunnable + a.gNonGo)
if a.gWaiting < 0 {
a.gWaiting = 0
}
unlock(&sched.lock)
}
// nsToSec takes a duration in nanoseconds and converts it to seconds as // nsToSec takes a duration in nanoseconds and converts it to seconds as
// a float64. // a float64.
func nsToSec(ns int64) float64 { func nsToSec(ns int64) float64 {
@ -758,6 +885,7 @@ type statAggregate struct {
cpuStats cpuStatsAggregate cpuStats cpuStatsAggregate
gcStats gcStatsAggregate gcStats gcStatsAggregate
finalStats finalStatsAggregate finalStats finalStatsAggregate
schedStats schedStatsAggregate
} }
// ensure populates statistics aggregates determined by deps if they // ensure populates statistics aggregates determined by deps if they
@ -782,6 +910,8 @@ func (a *statAggregate) ensure(deps *statDepSet) {
a.gcStats.compute() a.gcStats.compute()
case finalStatsDep: case finalStatsDep:
a.finalStats.compute() a.finalStats.compute()
case schedStatsDep:
a.schedStats.compute()
} }
} }
a.ensured = a.ensured.union(missing) a.ensured = a.ensured.union(missing)

31
src/runtime/metrics/description.go
View file

@ -437,6 +437,32 @@ var allDesc = []Description{
Description: "The current runtime.GOMAXPROCS setting, or the number of operating system threads that can execute user-level Go code simultaneously.", Description: "The current runtime.GOMAXPROCS setting, or the number of operating system threads that can execute user-level Go code simultaneously.",
Kind: KindUint64, Kind: KindUint64,
}, },
{
Name: "/sched/goroutines-created:goroutines",
Description: "Count of goroutines created since program start.",
Cumulative: true,
Kind: KindUint64,
},
{
Name: "/sched/goroutines/not-in-go:goroutines",
Description: "Approximate count of goroutines running or blocked in a system call or cgo call. Not guaranteed to add up to /sched/goroutines:goroutines with other goroutine metrics.",
Kind: KindUint64,
},
{
Name: "/sched/goroutines/runnable:goroutines",
Description: "Approximate count of goroutines ready to execute, but not executing. Not guaranteed to add up to /sched/goroutines:goroutines with other goroutine metrics.",
Kind: KindUint64,
},
{
Name: "/sched/goroutines/running:goroutines",
Description: "Approximate count of goroutines executing. Always less than or equal to /sched/gomaxprocs:threads. Not guaranteed to add up to /sched/goroutines:goroutines with other goroutine metrics.",
Kind: KindUint64,
},
{
Name: "/sched/goroutines/waiting:goroutines",
Description: "Approximate count of goroutines waiting on a resource (I/O or sync primitives). Not guaranteed to add up to /sched/goroutines:goroutines with other goroutine metrics.",
Kind: KindUint64,
},
{ {
Name: "/sched/goroutines:goroutines", Name: "/sched/goroutines:goroutines",
Description: "Count of live goroutines.", Description: "Count of live goroutines.",
@ -472,6 +498,11 @@ var allDesc = []Description{
Kind: KindFloat64Histogram, Kind: KindFloat64Histogram,
Cumulative: true, Cumulative: true,
}, },
{
Name: "/sched/threads/total:threads",
Description: "The current count of live threads that are owned by the Go runtime.",
Kind: KindUint64,
},
{ {
Name: "/sync/mutex/wait/total:seconds", Name: "/sync/mutex/wait/total:seconds",
Description: "Approximate cumulative time goroutines have spent blocked on a sync.Mutex, sync.RWMutex, or runtime-internal lock. This metric is useful for identifying global changes in lock contention. Collect a mutex or block profile using the runtime/pprof package for more detailed contention data.", Description: "Approximate cumulative time goroutines have spent blocked on a sync.Mutex, sync.RWMutex, or runtime-internal lock. This metric is useful for identifying global changes in lock contention. Collect a mutex or block profile using the runtime/pprof package for more detailed contention data.",

27
src/runtime/metrics/doc.go
View file

@ -509,6 +509,29 @@ Below is the full list of supported metrics, ordered lexicographically.
operating system threads that can execute user-level Go code operating system threads that can execute user-level Go code
simultaneously. simultaneously.
/sched/goroutines-created:goroutines
Count of goroutines created since program start.
/sched/goroutines/not-in-go:goroutines
Approximate count of goroutines running or blocked in
a system call or cgo call. Not guaranteed to add up to
/sched/goroutines:goroutines with other goroutine metrics.
/sched/goroutines/runnable:goroutines
Approximate count of goroutines ready to execute,
but not executing. Not guaranteed to add up to
/sched/goroutines:goroutines with other goroutine metrics.
/sched/goroutines/running:goroutines
Approximate count of goroutines executing. Always less than or
equal to /sched/gomaxprocs:threads. Not guaranteed to add up to
/sched/goroutines:goroutines with other goroutine metrics.
/sched/goroutines/waiting:goroutines
Approximate count of goroutines waiting on a resource
(I/O or sync primitives). Not guaranteed to add up to
/sched/goroutines:goroutines with other goroutine metrics.
/sched/goroutines:goroutines /sched/goroutines:goroutines
Count of live goroutines. Count of live goroutines.
@ -549,6 +572,10 @@ Below is the full list of supported metrics, ordered lexicographically.
/sched/pauses/stopping/other:seconds). Bucket counts increase /sched/pauses/stopping/other:seconds). Bucket counts increase
monotonically. monotonically.
/sched/threads/total:threads
The current count of live threads that are owned by the Go
runtime.
/sync/mutex/wait/total:seconds /sync/mutex/wait/total:seconds
Approximate cumulative time goroutines have spent blocked on a Approximate cumulative time goroutines have spent blocked on a
sync.Mutex, sync.RWMutex, or runtime-internal lock. This metric sync.Mutex, sync.RWMutex, or runtime-internal lock. This metric

217
src/runtime/metrics_test.go
View file

@ -22,6 +22,7 @@ import (
"strings" "strings"
"sync" "sync"
"sync/atomic" "sync/atomic"
"syscall"
"testing" "testing"
"time" "time"
"unsafe" "unsafe"
@ -1575,3 +1576,219 @@ func TestReadMetricsFinalizers(t *testing.T) {
t.Errorf("expected %s difference to be exactly %d, got %d -> %d", before[1].Name, N, v0, v1) t.Errorf("expected %s difference to be exactly %d, got %d -> %d", before[1].Name, N, v0, v1)
} }
} }
func TestReadMetricsSched(t *testing.T) {
const (
notInGo = iota
runnable
running
waiting
created
threads
numSamples
)
var s [numSamples]metrics.Sample
s[notInGo].Name = "/sched/goroutines/not-in-go:goroutines"
s[runnable].Name = "/sched/goroutines/runnable:goroutines"
s[running].Name = "/sched/goroutines/running:goroutines"
s[waiting].Name = "/sched/goroutines/waiting:goroutines"
s[created].Name = "/sched/goroutines-created:goroutines"
s[threads].Name = "/sched/threads/total:threads"
logMetrics := func(t *testing.T, s []metrics.Sample) {
for i := range s {
t.Logf("%s: %d", s[i].Name, s[i].Value.Uint64())
}
}
// generalSlack is the amount of goroutines we allow ourselves to be
// off by in any given category, either due to background system
// goroutines or testing package goroutines.
const generalSlack = 4
// waitingSlack is the max number of blocked goroutines left
// from other tests, the testing package, or system
// goroutines.
const waitingSlack = 100
// threadsSlack is the maximum number of threads left over
// from other tests and the runtime (sysmon, the template thread, etc.)
const threadsSlack = 20
// Make sure GC isn't running, since GC workers interfere with
// expected counts.
defer debug.SetGCPercent(debug.SetGCPercent(-1))
runtime.GC()
check := func(t *testing.T, s *metrics.Sample, min, max uint64) {
val := s.Value.Uint64()
if val < min {
t.Errorf("%s too low; %d < %d", s.Name, val, min)
}
if val > max {
t.Errorf("%s too high; %d > %d", s.Name, val, max)
}
}
checkEq := func(t *testing.T, s *metrics.Sample, value uint64) {
check(t, s, value, value)
}
spinUntil := func(f func() bool, timeout time.Duration) bool {
start := time.Now()
for time.Since(start) < timeout {
if f() {
return true
}
time.Sleep(time.Millisecond)
}
return false
}
// Check base values.
t.Run("base", func(t *testing.T) {
defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(1))
metrics.Read(s[:])
logMetrics(t, s[:])
check(t, &s[notInGo], 0, generalSlack)
check(t, &s[runnable], 0, generalSlack)
checkEq(t, &s[running], 1)
check(t, &s[waiting], 0, waitingSlack)
})
metrics.Read(s[:])
createdAfterBase := s[created].Value.Uint64()
// Force Running count to be high. We'll use these goroutines
// for Runnable, too.
const count = 10
var ready, exit atomic.Uint32
for i := 0; i < count-1; i++ {
go func() {
ready.Add(1)
for exit.Load() == 0 {
// Spin to get us and keep us running, but check
// the exit condition so we exit out early if we're
// done.
start := time.Now()
for time.Since(start) < 10*time.Millisecond && exit.Load() == 0 {
}
runtime.Gosched()
}
}()
}
for ready.Load() < count-1 {
runtime.Gosched()
}
// Be careful. We've entered a dangerous state for platforms
// that do not return back to the underlying system unless all
// goroutines are blocked, like js/wasm, since we have a bunch
// of runnable goroutines all spinning. We cannot write anything
// out.
if testenv.HasParallelism() {
t.Run("created", func(t *testing.T) {
metrics.Read(s[:])
logMetrics(t, s[:])
checkEq(t, &s[created], createdAfterBase+count)
})
t.Run("running", func(t *testing.T) {
defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(count + 4))
// It can take a little bit for the scheduler to
// distribute the goroutines to Ps, so retry for a
// while.
spinUntil(func() bool {
metrics.Read(s[:])
return s[running].Value.Uint64() >= count
}, time.Second)
logMetrics(t, s[:])
check(t, &s[running], count, count+4)
check(t, &s[threads], count, count+4+threadsSlack)
})
// Force runnable count to be high.
t.Run("runnable", func(t *testing.T) {
defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(1))
metrics.Read(s[:])
logMetrics(t, s[:])
checkEq(t, &s[running], 1)
check(t, &s[runnable], count-1, count+generalSlack)
})
// Done with the running/runnable goroutines.
exit.Store(1)
} else {
// Read metrics and then exit all the other goroutines,
// so that system calls may proceed.
metrics.Read(s[:])
// Done with the running/runnable goroutines.
exit.Store(1)
// Now we can check our invariants.
t.Run("created", func(t *testing.T) {
// Look for count-1 goroutines because we read metrics
// *before* t.Run goroutine was created for this sub-test.
checkEq(t, &s[created], createdAfterBase+count-1)
})
t.Run("running", func(t *testing.T) {
logMetrics(t, s[:])
checkEq(t, &s[running], 1)
checkEq(t, &s[threads], 1)
})
t.Run("runnable", func(t *testing.T) {
logMetrics(t, s[:])
check(t, &s[runnable], count-1, count+generalSlack)
})
}
// Force not-in-go count to be high. This is a little tricky since
// we try really hard not to let things block in system calls.
// We have to drop to the syscall package to do this reliably.
t.Run("not-in-go", func(t *testing.T) {
// Block a bunch of goroutines on an OS pipe.
pr, pw, err := pipe()
if err != nil {
switch runtime.GOOS {
case "js", "wasip1":
t.Skip("creating pipe:", err)
}
t.Fatal("creating pipe:", err)
}
for i := 0; i < count; i++ {
go syscall.Read(pr, make([]byte, 1))
}
// Let the goroutines block.
spinUntil(func() bool {
metrics.Read(s[:])
return s[notInGo].Value.Uint64() >= count
}, time.Second)
metrics.Read(s[:])
logMetrics(t, s[:])
check(t, &s[notInGo], count, count+generalSlack)
syscall.Close(pw)
syscall.Close(pr)
})
t.Run("waiting", func(t *testing.T) {
// Force waiting count to be high.
const waitingCount = 1000
stop := make(chan bool)
for i := 0; i < waitingCount; i++ {
go func() { <-stop }()
}
// Let the goroutines block.
spinUntil(func() bool {
metrics.Read(s[:])
return s[waiting].Value.Uint64() >= waitingCount
}, time.Second)
metrics.Read(s[:])
logMetrics(t, s[:])
check(t, &s[waiting], waitingCount, waitingCount+waitingSlack)
close(stop)
})
}

19
src/runtime/mgc.go
View file

@ -1521,18 +1521,15 @@ func gcBgMarkWorker(ready chan struct{}) {
} }
systemstack(func() { systemstack(func() {
// Mark our goroutine preemptible so its stack // Mark our goroutine preemptible so its stack can be scanned or observed
// can be scanned or observed by the execution // by the execution tracer. This, for example, lets two mark workers scan
// tracer. This, for example, lets two mark workers // each other (otherwise, they would deadlock).
// scan each other (otherwise, they would
// deadlock). We must not modify anything on
// the G stack. However, stack shrinking is
// disabled for mark workers, so it is safe to
// read from the G stack.
// //
// N.B. The execution tracer is not aware of this status // casGToWaitingForSuspendG marks the goroutine as ineligible for a
// transition and handles it specially based on the // stack shrink, effectively pinning the stack in memory for the duration.
// wait reason. //
// N.B. The execution tracer is not aware of this status transition and
// handles it specially based on the wait reason.
casGToWaitingForSuspendG(gp, _Grunning, waitReasonGCWorkerActive) casGToWaitingForSuspendG(gp, _Grunning, waitReasonGCWorkerActive)
switch pp.gcMarkWorkerMode { switch pp.gcMarkWorkerMode {
default: default:

2
src/runtime/mkduff.go
View file

@ -32,10 +32,8 @@ import (
) )
func main() { func main() {
gen("amd64", notags, zeroAMD64, copyAMD64)
gen("386", notags, zero386, copy386) gen("386", notags, zero386, copy386)
gen("arm", notags, zeroARM, copyARM) gen("arm", notags, zeroARM, copyARM)
gen("arm64", notags, zeroARM64, copyARM64)
gen("loong64", notags, zeroLOONG64, copyLOONG64) gen("loong64", notags, zeroLOONG64, copyLOONG64)
gen("ppc64x", tagsPPC64x, zeroPPC64x, copyPPC64x) gen("ppc64x", tagsPPC64x, zeroPPC64x, copyPPC64x)
gen("mips64x", tagsMIPS64x, zeroMIPS64x, copyMIPS64x) gen("mips64x", tagsMIPS64x, zeroMIPS64x, copyMIPS64x)

26
src/runtime/mkpreempt.go
View file

@ -610,10 +610,30 @@ func genLoong64(g *gen) {
l.add(movf, reg, regsize) l.add(movf, reg, regsize)
} }
// save/restore FCC0 // Add condition flag register fcc0-fcc7
sv := ""
rs := ""
last := 7
for i := 0; i <= last; i++ {
msb := 7 + (i * 8)
lsb := 0 + (i * 8)
// MOVV FCCx, R4,
// BSTRINSV $msb, R4, $lsb, R5
sv += fmt.Sprintf("%s FCC%d, R4\n", mov, i)
sv += fmt.Sprintf("BSTRINSV $%d, R4, $%d, R5\n", msb, lsb)
// BSTRPICKV $msb, R5, $lsb, R4
// MOVV R4, FCCx
rs += fmt.Sprintf("BSTRPICKV $%d, R5, $%d, R4\n", msb, lsb)
rs += fmt.Sprintf("%s R4, FCC%d", mov, i)
if i != last {
rs += fmt.Sprintf("\n")
}
}
l.addSpecial( l.addSpecial(
mov+" FCC0, R4\n"+mov+" R4, %d(R3)", sv+mov+" R5, %d(R3)",
mov+" %d(R3), R4\n"+mov+" R4, FCC0", mov+" %d(R3), R5\n"+rs,
regsize) regsize)
// allocate frame, save PC of interrupted instruction (in LR) // allocate frame, save PC of interrupted instruction (in LR)

4
src/runtime/mprof.go
View file

@ -1308,7 +1308,7 @@ func goroutineProfileWithLabelsConcurrent(p []profilerecord.StackRecord, labels
// allocation estimate without bothering to STW. As long as // allocation estimate without bothering to STW. As long as
// this is close, then we'll only need to STW once (on the next // this is close, then we'll only need to STW once (on the next
// call). // call).
return int(gcount()), false return int(gcount(false)), false
} }
semacquire(&goroutineProfile.sema) semacquire(&goroutineProfile.sema)
@ -1324,7 +1324,7 @@ func goroutineProfileWithLabelsConcurrent(p []profilerecord.StackRecord, labels
// goroutines that can vary between user and system to ensure that the count // goroutines that can vary between user and system to ensure that the count
// doesn't change during the collection. So, check the finalizer goroutine // doesn't change during the collection. So, check the finalizer goroutine
// and cleanup goroutines in particular. // and cleanup goroutines in particular.
n = int(gcount()) n = int(gcount(false))
if fingStatus.Load()&fingRunningFinalizer != 0 { if fingStatus.Load()&fingRunningFinalizer != 0 {
n++ n++
} }

15
src/runtime/pipe_unix_test.go Normal file
View file

@ -0,0 +1,15 @@
// Copyright 2025 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
//go:build !windows
package runtime_test
import "syscall"
func pipe() (r, w int, err error) {
var p [2]int
err = syscall.Pipe(p[:])
return p[0], p[1], err
}

13
src/runtime/pipe_windows_test.go Normal file
View file

@ -0,0 +1,13 @@
// Copyright 2025 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package runtime_test
import "syscall"
func pipe() (r, w syscall.Handle, err error) {
var p [2]syscall.Handle
err = syscall.Pipe(p[:])
return p[0], p[1], err
}

34
src/runtime/preempt_loong64.s
View file

@ -65,10 +65,40 @@ TEXT ·asyncPreempt(SB),NOSPLIT|NOFRAME,$0-0
MOVD F30, 456(R3) MOVD F30, 456(R3)
MOVD F31, 464(R3) MOVD F31, 464(R3)
MOVV FCC0, R4 MOVV FCC0, R4
MOVV R4, 472(R3) BSTRINSV $7, R4, $0, R5
MOVV FCC1, R4
BSTRINSV $15, R4, $8, R5
MOVV FCC2, R4
BSTRINSV $23, R4, $16, R5
MOVV FCC3, R4
BSTRINSV $31, R4, $24, R5
MOVV FCC4, R4
BSTRINSV $39, R4, $32, R5
MOVV FCC5, R4
BSTRINSV $47, R4, $40, R5
MOVV FCC6, R4
BSTRINSV $55, R4, $48, R5
MOVV FCC7, R4
BSTRINSV $63, R4, $56, R5
MOVV R5, 472(R3)
CALL ·asyncPreempt2(SB) CALL ·asyncPreempt2(SB)
MOVV 472(R3), R4 MOVV 472(R3), R5
BSTRPICKV $7, R5, $0, R4
MOVV R4, FCC0 MOVV R4, FCC0
BSTRPICKV $15, R5, $8, R4
MOVV R4, FCC1
BSTRPICKV $23, R5, $16, R4
MOVV R4, FCC2
BSTRPICKV $31, R5, $24, R4
MOVV R4, FCC3
BSTRPICKV $39, R5, $32, R4
MOVV R4, FCC4
BSTRPICKV $47, R5, $40, R4
MOVV R4, FCC5
BSTRPICKV $55, R5, $48, R4
MOVV R4, FCC6
BSTRPICKV $63, R5, $56, R4
MOVV R4, FCC7
MOVD 464(R3), F31 MOVD 464(R3), F31
MOVD 456(R3), F30 MOVD 456(R3), F30
MOVD 448(R3), F29 MOVD 448(R3), F29

51
src/runtime/proc.go
View file

@ -1007,7 +1007,7 @@ func mcommoninit(mp *m, id int64) {
// when it is just in a register or thread-local storage. // when it is just in a register or thread-local storage.
mp.alllink = allm mp.alllink = allm
// NumCgoCall() and others iterate over allm w/o schedlock, // NumCgoCall and others iterate over allm w/o schedlock,
// so we need to publish it safely. // so we need to publish it safely.
atomicstorep(unsafe.Pointer(&allm), unsafe.Pointer(mp)) atomicstorep(unsafe.Pointer(&allm), unsafe.Pointer(mp))
unlock(&sched.lock) unlock(&sched.lock)
@ -1372,6 +1372,9 @@ func casGToWaiting(gp *g, old uint32, reason waitReason) {
// casGToWaitingForSuspendG transitions gp from old to _Gwaiting, and sets the wait reason. // casGToWaitingForSuspendG transitions gp from old to _Gwaiting, and sets the wait reason.
// The wait reason must be a valid isWaitingForSuspendG wait reason. // The wait reason must be a valid isWaitingForSuspendG wait reason.
// //
// While a goroutine is in this state, it's stack is effectively pinned.
// The garbage collector must not shrink or otherwise mutate the goroutine's stack.
//
// Use this over casgstatus when possible to ensure that a waitreason is set. // Use this over casgstatus when possible to ensure that a waitreason is set.
func casGToWaitingForSuspendG(gp *g, old uint32, reason waitReason) { func casGToWaitingForSuspendG(gp *g, old uint32, reason waitReason) {
if !reason.isWaitingForSuspendG() { if !reason.isWaitingForSuspendG() {
@ -1608,18 +1611,11 @@ func stopTheWorldWithSema(reason stwReason) worldStop {
// stack while we try to stop the world since otherwise we could get // stack while we try to stop the world since otherwise we could get
// in a mutual preemption deadlock. // in a mutual preemption deadlock.
// //
// We must not modify anything on the G stack because a stack shrink // casGToWaitingForSuspendG marks the goroutine as ineligible for a
// may occur, now that we switched to _Gwaiting, specifically if we're // stack shrink, effectively pinning the stack in memory for the duration.
// doing this during the mark phase (mark termination excepted, since
// we know that stack scanning is done by that point). A stack shrink
// is otherwise OK though because in order to return from this function
// (and to leave the system stack) we must have preempted all
// goroutines, including any attempting to scan our stack, in which
// case, any stack shrinking will have already completed by the time we
// exit.
// //
// N.B. The execution tracer is not aware of this status transition and // N.B. The execution tracer is not aware of this status transition and
// andles it specially based on the wait reason. // handles it specially based on the wait reason.
casGToWaitingForSuspendG(getg().m.curg, _Grunning, waitReasonStoppingTheWorld) casGToWaitingForSuspendG(getg().m.curg, _Grunning, waitReasonStoppingTheWorld)
trace := traceAcquire() trace := traceAcquire()
@ -1652,6 +1648,7 @@ func stopTheWorldWithSema(reason stwReason) worldStop {
if trace.ok() { if trace.ok() {
trace.ProcSteal(pp, false) trace.ProcSteal(pp, false)
} }
sched.nGsyscallNoP.Add(1)
pp.syscalltick++ pp.syscalltick++
pp.gcStopTime = nanotime() pp.gcStopTime = nanotime()
sched.stopwait-- sched.stopwait--
@ -2106,16 +2103,11 @@ func forEachP(reason waitReason, fn func(*p)) {
// deadlock as we attempt to preempt a goroutine that's trying // deadlock as we attempt to preempt a goroutine that's trying
// to preempt us (e.g. for a stack scan). // to preempt us (e.g. for a stack scan).
// //
// We must not modify anything on the G stack because a stack shrink // casGToWaitingForSuspendG marks the goroutine as ineligible for a
// may occur. A stack shrink is otherwise OK though because in order // stack shrink, effectively pinning the stack in memory for the duration.
// to return from this function (and to leave the system stack) we
// must have preempted all goroutines, including any attempting
// to scan our stack, in which case, any stack shrinking will
// have already completed by the time we exit.
// //
// N.B. The execution tracer is not aware of this status // N.B. The execution tracer is not aware of this status transition and
// transition and handles it specially based on the // handles it specially based on the wait reason.
// wait reason.
casGToWaitingForSuspendG(gp, _Grunning, reason) casGToWaitingForSuspendG(gp, _Grunning, reason)
forEachPInternal(fn) forEachPInternal(fn)
casgstatus(gp, _Gwaiting, _Grunning) casgstatus(gp, _Gwaiting, _Grunning)
@ -2183,6 +2175,7 @@ func forEachPInternal(fn func(*p)) {
trace.ProcSteal(p2, false) trace.ProcSteal(p2, false)
traceRelease(trace) traceRelease(trace)
} }
sched.nGsyscallNoP.Add(1)
p2.syscalltick++ p2.syscalltick++
handoffp(p2) handoffp(p2)
} else if trace.ok() { } else if trace.ok() {
@ -2456,6 +2449,7 @@ func needm(signal bool) {
// mp.curg is now a real goroutine. // mp.curg is now a real goroutine.
casgstatus(mp.curg, _Gdead, _Gsyscall) casgstatus(mp.curg, _Gdead, _Gsyscall)
sched.ngsys.Add(-1) sched.ngsys.Add(-1)
sched.nGsyscallNoP.Add(1)
if !signal { if !signal {
if trace.ok() { if trace.ok() {
@ -2591,6 +2585,7 @@ func dropm() {
casgstatus(mp.curg, _Gsyscall, _Gdead) casgstatus(mp.curg, _Gsyscall, _Gdead)
mp.curg.preemptStop = false mp.curg.preemptStop = false
sched.ngsys.Add(1) sched.ngsys.Add(1)
sched.nGsyscallNoP.Add(-1)
if !mp.isExtraInSig { if !mp.isExtraInSig {
if trace.ok() { if trace.ok() {
@ -4684,6 +4679,7 @@ func entersyscall_gcwait() {
trace.ProcSteal(pp, true) trace.ProcSteal(pp, true)
traceRelease(trace) traceRelease(trace)
} }
sched.nGsyscallNoP.Add(1)
pp.gcStopTime = nanotime() pp.gcStopTime = nanotime()
pp.syscalltick++ pp.syscalltick++
if sched.stopwait--; sched.stopwait == 0 { if sched.stopwait--; sched.stopwait == 0 {
@ -4716,6 +4712,8 @@ func entersyscallblock() {
gp.m.syscalltick = gp.m.p.ptr().syscalltick gp.m.syscalltick = gp.m.p.ptr().syscalltick
gp.m.p.ptr().syscalltick++ gp.m.p.ptr().syscalltick++
sched.nGsyscallNoP.Add(1)
// Leave SP around for GC and traceback. // Leave SP around for GC and traceback.
pc := sys.GetCallerPC() pc := sys.GetCallerPC()
sp := sys.GetCallerSP() sp := sys.GetCallerSP()
@ -4936,6 +4934,7 @@ func exitsyscallfast_pidle() bool {
} }
unlock(&sched.lock) unlock(&sched.lock)
if pp != nil { if pp != nil {
sched.nGsyscallNoP.Add(-1)
acquirep(pp) acquirep(pp)
return true return true
} }
@ -4962,6 +4961,7 @@ func exitsyscall0(gp *g) {
trace.GoSysExit(true) trace.GoSysExit(true)
traceRelease(trace) traceRelease(trace)
} }
sched.nGsyscallNoP.Add(-1)
dropg() dropg()
lock(&sched.lock) lock(&sched.lock)
var pp *p var pp *p
@ -5262,6 +5262,7 @@ func newproc1(fn *funcval, callergp *g, callerpc uintptr, parked bool, waitreaso
racereleasemergeg(newg, unsafe.Pointer(&labelSync)) racereleasemergeg(newg, unsafe.Pointer(&labelSync))
} }
} }
pp.goroutinesCreated++
releasem(mp) releasem(mp)
return newg return newg
@ -5537,8 +5538,11 @@ func badunlockosthread() {
throw("runtime: internal error: misuse of lockOSThread/unlockOSThread") throw("runtime: internal error: misuse of lockOSThread/unlockOSThread")
} }
func gcount() int32 { func gcount(includeSys bool) int32 {
n := int32(atomic.Loaduintptr(&allglen)) - sched.gFree.stack.size - sched.gFree.noStack.size - sched.ngsys.Load() n := int32(atomic.Loaduintptr(&allglen)) - sched.gFree.stack.size - sched.gFree.noStack.size
if !includeSys {
n -= sched.ngsys.Load()
}
for _, pp := range allp { for _, pp := range allp {
n -= pp.gFree.size n -= pp.gFree.size
} }
@ -5838,6 +5842,8 @@ func (pp *p) destroy() {
pp.gcAssistTime = 0 pp.gcAssistTime = 0
gcCleanups.queued += pp.cleanupsQueued gcCleanups.queued += pp.cleanupsQueued
pp.cleanupsQueued = 0 pp.cleanupsQueued = 0
sched.goroutinesCreated.Add(int64(pp.goroutinesCreated))
pp.goroutinesCreated = 0
pp.xRegs.free() pp.xRegs.free()
pp.status = _Pdead pp.status = _Pdead
} }
@ -6413,6 +6419,7 @@ func retake(now int64) uint32 {
trace.ProcSteal(pp, false) trace.ProcSteal(pp, false)
traceRelease(trace) traceRelease(trace)
} }
sched.nGsyscallNoP.Add(1)
n++ n++
pp.syscalltick++ pp.syscalltick++
handoffp(pp) handoffp(pp)

14
src/runtime/runtime2.go
View file

@ -764,6 +764,9 @@ type p struct {
// gcStopTime is the nanotime timestamp that this P last entered _Pgcstop. // gcStopTime is the nanotime timestamp that this P last entered _Pgcstop.
gcStopTime int64 gcStopTime int64
// goroutinesCreated is the total count of goroutines created by this P.
goroutinesCreated uint64
// xRegs is the per-P extended register state used by asynchronous // xRegs is the per-P extended register state used by asynchronous
// preemption. This is an empty struct on platforms that don't use extended // preemption. This is an empty struct on platforms that don't use extended
// register state. // register state.
@ -792,7 +795,8 @@ type schedt struct {
nmsys int32 // number of system m's not counted for deadlock nmsys int32 // number of system m's not counted for deadlock
nmfreed int64 // cumulative number of freed m's nmfreed int64 // cumulative number of freed m's
ngsys atomic.Int32 // number of system goroutines ngsys atomic.Int32 // number of system goroutines
nGsyscallNoP atomic.Int32 // number of goroutines in syscalls without a P
pidle puintptr // idle p's pidle puintptr // idle p's
npidle atomic.Int32 npidle atomic.Int32
@ -891,6 +895,10 @@ type schedt struct {
// M, but waiting for locks within the runtime. This field stores the value // M, but waiting for locks within the runtime. This field stores the value
// for Ms that have exited. // for Ms that have exited.
totalRuntimeLockWaitTime atomic.Int64 totalRuntimeLockWaitTime atomic.Int64
// goroutinesCreated (plus the value of goroutinesCreated on each P in allp)
// is the sum of all goroutines created by the program.
goroutinesCreated atomic.Uint64
} }
// Values for the flags field of a sigTabT. // Values for the flags field of a sigTabT.
@ -1217,7 +1225,9 @@ var isIdleInSynctest = [len(waitReasonStrings)]bool{
} }
var ( var (
allm *m // Linked-list of all Ms. Written under sched.lock, read atomically.
allm *m
gomaxprocs int32 gomaxprocs int32
numCPUStartup int32 numCPUStartup int32
forcegc forcegcstate forcegc forcegcstate

2
src/runtime/select.go
View file

@ -178,7 +178,7 @@ func selectgo(cas0 *scase, order0 *uint16, pc0 *uintptr, nsends, nrecvs int, blo
if cas.c.bubble != nil { if cas.c.bubble != nil {
if getg().bubble != cas.c.bubble { if getg().bubble != cas.c.bubble {
panic(plainError("select on synctest channel from outside bubble")) fatal("select on synctest channel from outside bubble")
} }
} else { } else {
allSynctest = false allSynctest = false

21
src/runtime/stack.go
View file

@ -1214,15 +1214,18 @@ func isShrinkStackSafe(gp *g) bool {
if gp.parkingOnChan.Load() { if gp.parkingOnChan.Load() {
return false return false
} }
// We also can't copy the stack while tracing is enabled, and // We also can't copy the stack while a gp is in _Gwaiting solely
// gp is in _Gwaiting solely to make itself available to suspendG. // to make itself available to suspendG.
//
// In these cases, the G is actually executing on the system // In these cases, the G is actually executing on the system
// stack, and the execution tracer may want to take a stack trace // stack, and the execution tracer, mutex profiler, etc. may want
// of the G's stack. Note: it's safe to access gp.waitreason here. // to take a stack trace of the G's stack.
// We're only checking if this is true if we took ownership of the //
// Note: it's safe to access gp.waitreason here.
// We're only calling isShrinkStackSafe if we took ownership of the
// G with the _Gscan bit. This prevents the goroutine from transitioning, // G with the _Gscan bit. This prevents the goroutine from transitioning,
// which prevents gp.waitreason from changing. // which prevents gp.waitreason from changing.
if traceEnabled() && readgstatus(gp)&^_Gscan == _Gwaiting && gp.waitreason.isWaitingForSuspendG() { if readgstatus(gp)&^_Gscan == _Gwaiting && gp.waitreason.isWaitingForSuspendG() {
return false return false
} }
return true return true
@ -1258,12 +1261,6 @@ func shrinkstack(gp *g) {
if debug.gcshrinkstackoff > 0 { if debug.gcshrinkstackoff > 0 {
return return
} }
f := findfunc(gp.startpc)
if f.valid() && f.funcID == abi.FuncID_gcBgMarkWorker {
// We're not allowed to shrink the gcBgMarkWorker
// stack (see gcBgMarkWorker for explanation).
return
}
oldsize := gp.stack.hi - gp.stack.lo oldsize := gp.stack.hi - gp.stack.lo
newsize := oldsize / 2 newsize := oldsize / 2

4
src/runtime/time.go
View file

@ -415,7 +415,7 @@ func newTimer(when, period int64, f func(arg any, seq uintptr, delay int64), arg
//go:linkname stopTimer time.stopTimer //go:linkname stopTimer time.stopTimer
func stopTimer(t *timeTimer) bool { func stopTimer(t *timeTimer) bool {
if t.isFake && getg().bubble == nil { if t.isFake && getg().bubble == nil {
panic("stop of synctest timer from outside bubble") fatal("stop of synctest timer from outside bubble")
} }
return t.stop() return t.stop()
} }
@ -430,7 +430,7 @@ func resetTimer(t *timeTimer, when, period int64) bool {
racerelease(unsafe.Pointer(&t.timer)) racerelease(unsafe.Pointer(&t.timer))
} }
if t.isFake && getg().bubble == nil { if t.isFake && getg().bubble == nil {
panic("reset of synctest timer from outside bubble") fatal("reset of synctest timer from outside bubble")
} }
return t.reset(when, period) return t.reset(when, period)
} }

21
src/runtime/trace.go
View file

@ -754,24 +754,7 @@ func traceRegisterLabelsAndReasons(gen uintptr) {
// was on has been returned, ReadTrace returns nil. The caller must copy the // was on has been returned, ReadTrace returns nil. The caller must copy the
// returned data before calling ReadTrace again. // returned data before calling ReadTrace again.
// ReadTrace must be called from one goroutine at a time. // ReadTrace must be called from one goroutine at a time.
func ReadTrace() []byte { func ReadTrace() (buf []byte) {
for {
buf := readTrace()
// Skip over the end-of-generation signal which must not appear
// in the final trace.
if len(buf) == 1 && tracev2.EventType(buf[0]) == tracev2.EvEndOfGeneration {
continue
}
return buf
}
}
// readTrace is the implementation of ReadTrace, except with an additional
// in-band signal as to when the buffer is for a new generation.
//
//go:linkname readTrace runtime/trace.runtime_readTrace
func readTrace() (buf []byte) {
top: top:
var park bool var park bool
systemstack(func() { systemstack(func() {
@ -842,7 +825,7 @@ func readTrace0() (buf []byte, park bool) {
if !trace.headerWritten { if !trace.headerWritten {
trace.headerWritten = true trace.headerWritten = true
unlock(&trace.lock) unlock(&trace.lock)
return []byte("go 1.25 trace\x00\x00\x00"), false return []byte("go 1.26 trace\x00\x00\x00"), false
} }
// Read the next buffer. // Read the next buffer.

79
src/runtime/trace/batch.go
View file

@ -12,72 +12,77 @@ import (
// timestamp is an unprocessed timestamp. // timestamp is an unprocessed timestamp.
type timestamp uint64 type timestamp uint64
// batch represents a batch of trace events.
// It is unparsed except for its header.
type batch struct { type batch struct {
m threadID
time timestamp time timestamp
gen uint64
data []byte data []byte
} }
// threadID is the runtime-internal M structure's ID. This is unique
// for each OS thread.
type threadID int64
// readBatch copies b and parses the trace batch header inside. // readBatch copies b and parses the trace batch header inside.
// Returns the batch, the generation, bytes read, and an error. // Returns the batch, bytes read, and an error.
func readBatch(b []byte) (batch, uint64, uint64, error) { func readBatch(b []byte) (batch, uint64, error) {
if len(b) == 0 { if len(b) == 0 {
return batch{}, 0, 0, fmt.Errorf("batch is empty") return batch{}, 0, fmt.Errorf("batch is empty")
} }
data := make([]byte, len(b)) data := make([]byte, len(b))
if nw := copy(data, b); nw != len(b) { copy(data, b)
return batch{}, 0, 0, fmt.Errorf("unexpected error copying batch")
}
// Read batch header byte.
if typ := tracev2.EventType(b[0]); typ != tracev2.EvEventBatch && typ != tracev2.EvExperimentalBatch {
return batch{}, 0, 1, fmt.Errorf("expected batch event, got event %d", typ)
}
// Read the batch header: gen (generation), thread (M) ID, base timestamp // Read batch header byte.
// for the batch. if typ := tracev2.EventType(b[0]); typ == tracev2.EvEndOfGeneration {
if len(b) != 1 {
return batch{}, 1, fmt.Errorf("unexpected end of generation in batch of size >1")
}
return batch{data: data}, 1, nil
}
if typ := tracev2.EventType(b[0]); typ != tracev2.EvEventBatch && typ != tracev2.EvExperimentalBatch {
return batch{}, 1, fmt.Errorf("expected batch event, got event %d", typ)
}
total := 1 total := 1
b = b[1:] b = b[1:]
// Read the generation
gen, n, err := readUvarint(b) gen, n, err := readUvarint(b)
if err != nil { if err != nil {
return batch{}, gen, uint64(total + n), fmt.Errorf("error reading batch gen: %w", err) return batch{}, uint64(total + n), fmt.Errorf("error reading batch gen: %w", err)
}
total += n
b = b[n:]
m, n, err := readUvarint(b)
if err != nil {
return batch{}, gen, uint64(total + n), fmt.Errorf("error reading batch M ID: %w", err)
}
total += n
b = b[n:]
ts, n, err := readUvarint(b)
if err != nil {
return batch{}, gen, uint64(total + n), fmt.Errorf("error reading batch timestamp: %w", err)
} }
total += n total += n
b = b[n:] b = b[n:]
// Read in the size of the batch to follow. // Read the M (discard it).
_, n, err = readUvarint(b)
if err != nil {
return batch{}, uint64(total + n), fmt.Errorf("error reading batch M ID: %w", err)
}
total += n
b = b[n:]
// Read the timestamp.
ts, n, err := readUvarint(b)
if err != nil {
return batch{}, uint64(total + n), fmt.Errorf("error reading batch timestamp: %w", err)
}
total += n
b = b[n:]
// Read the size of the batch to follow.
size, n, err := readUvarint(b) size, n, err := readUvarint(b)
if err != nil { if err != nil {
return batch{}, gen, uint64(total + n), fmt.Errorf("error reading batch size: %w", err) return batch{}, uint64(total + n), fmt.Errorf("error reading batch size: %w", err)
} }
if size > tracev2.MaxBatchSize { if size > tracev2.MaxBatchSize {
return batch{}, gen, uint64(total + n), fmt.Errorf("invalid batch size %d, maximum is %d", size, tracev2.MaxBatchSize) return batch{}, uint64(total + n), fmt.Errorf("invalid batch size %d, maximum is %d", size, tracev2.MaxBatchSize)
} }
total += n total += n
total += int(size) total += int(size)
if total != len(data) {
return batch{}, uint64(total), fmt.Errorf("expected complete batch")
}
data = data[:total] data = data[:total]
// Return the batch. // Return the batch.
return batch{ return batch{
m: threadID(m), gen: gen,
time: timestamp(ts), time: timestamp(ts),
data: data, data: data,
}, gen, uint64(total), nil }, uint64(total), nil
} }

4
src/runtime/trace/flightrecorder.go
View file

@ -141,9 +141,9 @@ func (fr *FlightRecorder) WriteTo(w io.Writer) (n int64, err error) {
// Write all the data. // Write all the data.
for _, gen := range gens { for _, gen := range gens {
for _, batch := range gen.batches { for _, data := range gen.batches {
// Write batch data. // Write batch data.
nw, err = w.Write(batch.data) nw, err = w.Write(data)
n += int64(nw) n += int64(nw)
if err != nil { if err != nil {
return n, err return n, err

12
src/runtime/trace/recorder.go
View file

@ -41,21 +41,21 @@ func (w *recorder) Write(b []byte) (n int, err error) {
if len(b) == n { if len(b) == n {
return 0, nil return 0, nil
} }
ba, gen, nb, err := readBatch(b[n:]) // Every write from the runtime is guaranteed to be a complete batch. ba, nb, err := readBatch(b[n:]) // Every write from the runtime is guaranteed to be a complete batch.
if err != nil { if err != nil {
return len(b) - int(nb) - n, err return len(b) - int(nb) - n, err
} }
n += int(nb) n += int(nb)
// Append the batch to the current generation. // Append the batch to the current generation.
if r.active.gen == 0 { if ba.gen != 0 && r.active.gen == 0 {
r.active.gen = gen r.active.gen = ba.gen
} }
if r.active.minTime == 0 || r.active.minTime > r.freq.mul(ba.time) { if ba.time != 0 && (r.active.minTime == 0 || r.active.minTime > r.freq.mul(ba.time)) {
r.active.minTime = r.freq.mul(ba.time) r.active.minTime = r.freq.mul(ba.time)
} }
r.active.size += len(ba.data) r.active.size += len(ba.data)
r.active.batches = append(r.active.batches, ba) r.active.batches = append(r.active.batches, ba.data)
return len(b), nil return len(b), nil
} }
@ -99,7 +99,7 @@ type rawGeneration struct {
gen uint64 gen uint64
size int size int
minTime eventTime minTime eventTime
batches []batch batches [][]byte
} }
func traceTimeNow(freq frequency) eventTime { func traceTimeNow(freq frequency) eventTime {

17
src/runtime/trace/subscribe.go
View file

@ -155,7 +155,7 @@ func (t *traceMultiplexer) startLocked() error {
t.subscribersMu.Unlock() t.subscribersMu.Unlock()
go func() { go func() {
header := runtime_readTrace() header := runtime.ReadTrace()
if traceStartWriter != nil { if traceStartWriter != nil {
traceStartWriter.Write(header) traceStartWriter.Write(header)
} }
@ -164,10 +164,16 @@ func (t *traceMultiplexer) startLocked() error {
} }
for { for {
data := runtime_readTrace() data := runtime.ReadTrace()
if data == nil { if data == nil {
break break
} }
if traceStartWriter != nil {
traceStartWriter.Write(data)
}
if flightRecorder != nil {
flightRecorder.Write(data)
}
if len(data) == 1 && tracev2.EventType(data[0]) == tracev2.EvEndOfGeneration { if len(data) == 1 && tracev2.EventType(data[0]) == tracev2.EvEndOfGeneration {
if flightRecorder != nil { if flightRecorder != nil {
flightRecorder.endGeneration() flightRecorder.endGeneration()
@ -187,13 +193,6 @@ func (t *traceMultiplexer) startLocked() error {
if frIsNew { if frIsNew {
flightRecorder.Write(header) flightRecorder.Write(header)
} }
} else {
if traceStartWriter != nil {
traceStartWriter.Write(data)
}
if flightRecorder != nil {
flightRecorder.Write(data)
}
} }
} }
}() }()

4
src/runtime/tracetype.go
View file

@ -2,7 +2,7 @@
// Use of this source code is governed by a BSD-style // Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file. // license that can be found in the LICENSE file.
// Trace stack table and acquisition. // Trace type table.
package runtime package runtime
@ -13,7 +13,7 @@ import (
"unsafe" "unsafe"
) )
// traceTypeTable maps stack traces (arrays of PC's) to unique uint32 ids. // traceTypeTable maps types to unique uint32 ids.
// It is lock-free for reading. // It is lock-free for reading.
type traceTypeTable struct { type traceTypeTable struct {
tab traceMap tab traceMap

10
src/syscall/exec_linux.go
View file

@ -365,11 +365,11 @@ func forkAndExecInChild1(argv0 *byte, argv, envv []*byte, chroot, dir *byte, att
if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(mapPipe[1]), 0, 0); err1 != 0 { if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(mapPipe[1]), 0, 0); err1 != 0 {
goto childerror goto childerror
} }
pid, _, err1 = RawSyscall(SYS_READ, uintptr(mapPipe[0]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2)) c, _, err1 = RawSyscall(SYS_READ, uintptr(mapPipe[0]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
if err1 != 0 { if err1 != 0 {
goto childerror goto childerror
} }
if pid != unsafe.Sizeof(err2) { if c != unsafe.Sizeof(err2) {
err1 = EINVAL err1 = EINVAL
goto childerror goto childerror
} }
@ -427,7 +427,7 @@ func forkAndExecInChild1(argv0 *byte, argv, envv []*byte, chroot, dir *byte, att
if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&psetgroups[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 { if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&psetgroups[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
goto childerror goto childerror
} }
pid, _, err1 = RawSyscall(SYS_WRITE, fd1, uintptr(unsafe.Pointer(&setgroups[0])), uintptr(len(setgroups))) _, _, err1 = RawSyscall(SYS_WRITE, fd1, uintptr(unsafe.Pointer(&setgroups[0])), uintptr(len(setgroups)))
if err1 != 0 { if err1 != 0 {
goto childerror goto childerror
} }
@ -438,7 +438,7 @@ func forkAndExecInChild1(argv0 *byte, argv, envv []*byte, chroot, dir *byte, att
if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&pgid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 { if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&pgid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
goto childerror goto childerror
} }
pid, _, err1 = RawSyscall(SYS_WRITE, fd1, uintptr(unsafe.Pointer(&gidmap[0])), uintptr(len(gidmap))) _, _, err1 = RawSyscall(SYS_WRITE, fd1, uintptr(unsafe.Pointer(&gidmap[0])), uintptr(len(gidmap)))
if err1 != 0 { if err1 != 0 {
goto childerror goto childerror
} }
@ -452,7 +452,7 @@ func forkAndExecInChild1(argv0 *byte, argv, envv []*byte, chroot, dir *byte, att
if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&puid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 { if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&puid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
goto childerror goto childerror
} }
pid, _, err1 = RawSyscall(SYS_WRITE, fd1, uintptr(unsafe.Pointer(&uidmap[0])), uintptr(len(uidmap))) _, _, err1 = RawSyscall(SYS_WRITE, fd1, uintptr(unsafe.Pointer(&uidmap[0])), uintptr(len(uidmap)))
if err1 != 0 { if err1 != 0 {
goto childerror goto childerror
} }

20
src/syscall/syscall_unix.go
View file

@ -410,17 +410,25 @@ func SendmsgN(fd int, p, oob []byte, to Sockaddr, flags int) (n int, err error)
} }
func sendmsgNInet4(fd int, p, oob []byte, to *SockaddrInet4, flags int) (n int, err error) { func sendmsgNInet4(fd int, p, oob []byte, to *SockaddrInet4, flags int) (n int, err error) {
ptr, salen, err := to.sockaddr() var ptr unsafe.Pointer
if err != nil { var salen _Socklen
return 0, err if to != nil {
ptr, salen, err = to.sockaddr()
if err != nil {
return 0, err
}
} }
return sendmsgN(fd, p, oob, ptr, salen, flags) return sendmsgN(fd, p, oob, ptr, salen, flags)
} }
func sendmsgNInet6(fd int, p, oob []byte, to *SockaddrInet6, flags int) (n int, err error) { func sendmsgNInet6(fd int, p, oob []byte, to *SockaddrInet6, flags int) (n int, err error) {
ptr, salen, err := to.sockaddr() var ptr unsafe.Pointer
if err != nil { var salen _Socklen
return 0, err if to != nil {
ptr, salen, err = to.sockaddr()
if err != nil {
return 0, err
}
} }
return sendmsgN(fd, p, oob, ptr, salen, flags) return sendmsgN(fd, p, oob, ptr, salen, flags)
} }

19
src/unique/canonmap_test.go
View file

@ -108,6 +108,25 @@ func testCanonMap(t *testing.T, newMap func() *canonMap[string]) {
wg.Wait() wg.Wait()
} }
// Run an extra GC cycle to de-flake. Sometimes the cleanups
// fail to run in time, despite drainCleanupQueue.
//
// TODO(mknyszek): Figure out why the extra GC is necessary,
// and what is transiently keeping the cleanups live.
// * I have confirmed that they are not completely stuck, and
// they always eventually run.
// * I have also confirmed it's not asynchronous preemption
// keeping them around (though that is a possibility).
// * I have confirmed that they are not simply sitting on
// the queue, and that drainCleanupQueue is just failing
// to actually empty the queue.
// * I have confirmed that it's not a write barrier that's
// keeping it alive, nor is it a weak pointer dereference
// (which shades the object during the GC).
// The corresponding objects do seem to be transiently truly
// reachable, but I have no idea by what path.
runtime.GC()
// Drain cleanups so everything is deleted. // Drain cleanups so everything is deleted.
drainCleanupQueue(t) drainCleanupQueue(t)

75
test/fixedbugs/issue75063.go Normal file
View file

@ -0,0 +1,75 @@
// compile
// Copyright 2025 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package reorder
type Element struct {
A string
B string
C string
D string
E string
Text []string
List []string
Child Elements
F string
G bool
H bool
I string
}
type Elements []Element
func DoesNotCompile(ve Elements) Elements {
aa := Elements{}
bb := Elements{}
cc := Elements{}
dd := Elements{}
ee := Elements{}
ff := Elements{}
gg := Elements{}
hh := Elements{}
ii := Elements{}
if len(ve) != 1 {
return ve
}
for _, e := range ve[0].Child {
if len(e.Text) == 1 && (e.Text[0] == "xx") {
ee = append(ee, e)
} else if len(e.Text) == 1 && e.Text[0] == "yy" {
for _, c := range e.Child {
if len(c.Text) == 1 && c.Text[0] == "zz" {
ii = append(ii, c)
} else {
hh = append(hh, c)
}
}
ii = append(ii, hh...)
e.Child = ii
gg = append(gg, e)
} else if len(e.Text) == 1 && e.Text[0] == "tt" {
for _, entry := range e.Child {
for _, c := range entry.Child {
if len(c.Text) == 1 && c.Text[0] == "ee" {
cc = append(cc, c)
} else {
dd = append(dd, c)
}
}
cc = append(cc, dd...)
entry.Child = cc
bb = append(bb, entry)
cc, dd = Elements{}, Elements{}
}
e.Child = bb
aa = append(aa, e)
} else {
ff = append(ff, e)
}
}
return ve
}