1 // Copyright 2014 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
11 "runtime/internal/atomic"
12 "runtime/internal/sys"
16 // set using cmd/go/internal/modload.ModInfoProg
19 // Goroutine scheduler
20 // The scheduler's job is to distribute ready-to-run goroutines over worker threads.
22 // The main concepts are:
24 // M - worker thread, or machine.
25 // P - processor, a resource that is required to execute Go code.
26 // M must have an associated P to execute Go code, however it can be
27 // blocked or in a syscall w/o an associated P.
29 // Design doc at https://golang.org/s/go11sched.
31 // Worker thread parking/unparking.
32 // We need to balance between keeping enough running worker threads to utilize
33 // available hardware parallelism and parking excessive running worker threads
34 // to conserve CPU resources and power. This is not simple for two reasons:
35 // (1) scheduler state is intentionally distributed (in particular, per-P work
36 // queues), so it is not possible to compute global predicates on fast paths;
37 // (2) for optimal thread management we would need to know the future (don't park
38 // a worker thread when a new goroutine will be readied in near future).
40 // Three rejected approaches that would work badly:
41 // 1. Centralize all scheduler state (would inhibit scalability).
42 // 2. Direct goroutine handoff. That is, when we ready a new goroutine and there
43 // is a spare P, unpark a thread and handoff it the thread and the goroutine.
44 // This would lead to thread state thrashing, as the thread that readied the
45 // goroutine can be out of work the very next moment, we will need to park it.
46 // Also, it would destroy locality of computation as we want to preserve
47 // dependent goroutines on the same thread; and introduce additional latency.
48 // 3. Unpark an additional thread whenever we ready a goroutine and there is an
49 // idle P, but don't do handoff. This would lead to excessive thread parking/
50 // unparking as the additional threads will instantly park without discovering
53 // The current approach:
55 // This approach applies to three primary sources of potential work: readying a
56 // goroutine, new/modified-earlier timers, and idle-priority GC. See below for
57 // additional details.
59 // We unpark an additional thread when we submit work if (this is wakep()):
60 // 1. There is an idle P, and
61 // 2. There are no "spinning" worker threads.
63 // A worker thread is considered spinning if it is out of local work and did
64 // not find work in the global run queue or netpoller; the spinning state is
65 // denoted in m.spinning and in sched.nmspinning. Threads unparked this way are
66 // also considered spinning; we don't do goroutine handoff so such threads are
67 // out of work initially. Spinning threads spin on looking for work in per-P
68 // run queues and timer heaps or from the GC before parking. If a spinning
69 // thread finds work it takes itself out of the spinning state and proceeds to
70 // execution. If it does not find work it takes itself out of the spinning
71 // state and then parks.
73 // If there is at least one spinning thread (sched.nmspinning>1), we don't
74 // unpark new threads when submitting work. To compensate for that, if the last
75 // spinning thread finds work and stops spinning, it must unpark a new spinning
76 // thread. This approach smooths out unjustified spikes of thread unparking,
77 // but at the same time guarantees eventual maximal CPU parallelism
80 // The main implementation complication is that we need to be very careful
81 // during spinning->non-spinning thread transition. This transition can race
82 // with submission of new work, and either one part or another needs to unpark
83 // another worker thread. If they both fail to do that, we can end up with
84 // semi-persistent CPU underutilization.
86 // The general pattern for submission is:
87 // 1. Submit work to the local run queue, timer heap, or GC state.
88 // 2. #StoreLoad-style memory barrier.
89 // 3. Check sched.nmspinning.
91 // The general pattern for spinning->non-spinning transition is:
92 // 1. Decrement nmspinning.
93 // 2. #StoreLoad-style memory barrier.
94 // 3. Check all per-P work queues and GC for new work.
96 // Note that all this complexity does not apply to global run queue as we are
97 // not sloppy about thread unparking when submitting to global queue. Also see
98 // comments for nmspinning manipulation.
100 // How these different sources of work behave varies, though it doesn't affect
101 // the synchronization approach:
102 // * Ready goroutine: this is an obvious source of work; the goroutine is
103 // immediately ready and must run on some thread eventually.
104 // * New/modified-earlier timer: The current timer implementation (see time.go)
105 // uses netpoll in a thread with no work available to wait for the soonest
106 // timer. If there is no thread waiting, we want a new spinning thread to go
108 // * Idle-priority GC: The GC wakes a stopped idle thread to contribute to
109 // background GC work (note: currently disabled per golang.org/issue/19112).
110 // Also see golang.org/issue/44313, as this should be extended to all GC
120 //go:linkname runtime_inittask runtime..inittask
121 var runtime_inittask initTask
123 //go:linkname main_inittask main..inittask
124 var main_inittask initTask
126 // main_init_done is a signal used by cgocallbackg that initialization
127 // has been completed. It is made before _cgo_notify_runtime_init_done,
128 // so all cgo calls can rely on it existing. When main_init is complete,
129 // it is closed, meaning cgocallbackg can reliably receive from it.
130 var main_init_done chan bool
132 //go:linkname main_main main.main
135 // mainStarted indicates that the main M has started.
138 // runtimeInitTime is the nanotime() at which the runtime started.
139 var runtimeInitTime int64
141 // Value to use for signal mask for newly created M's.
142 var initSigmask sigset
144 // The main goroutine.
148 // Racectx of m0->g0 is used only as the parent of the main goroutine.
149 // It must not be used for anything else.
152 // Max stack size is 1 GB on 64-bit, 250 MB on 32-bit.
153 // Using decimal instead of binary GB and MB because
154 // they look nicer in the stack overflow failure message.
155 if goarch.PtrSize == 8 {
156 maxstacksize = 1000000000
158 maxstacksize = 250000000
161 // An upper limit for max stack size. Used to avoid random crashes
162 // after calling SetMaxStack and trying to allocate a stack that is too big,
163 // since stackalloc works with 32-bit sizes.
164 maxstackceiling = 2 * maxstacksize
166 // Allow newproc to start new Ms.
169 if GOARCH != "wasm" { // no threads on wasm yet, so no sysmon
171 newm(sysmon, nil, -1)
175 // Lock the main goroutine onto this, the main OS thread,
176 // during initialization. Most programs won't care, but a few
177 // do require certain calls to be made by the main thread.
178 // Those can arrange for main.main to run in the main thread
179 // by calling runtime.LockOSThread during initialization
180 // to preserve the lock.
184 throw("runtime.main not on m0")
187 // Record when the world started.
188 // Must be before doInit for tracing init.
189 runtimeInitTime = nanotime()
190 if runtimeInitTime == 0 {
191 throw("nanotime returning zero")
194 if debug.inittrace != 0 {
195 inittrace.id = getg().goid
196 inittrace.active = true
199 doInit(&runtime_inittask) // Must be before defer.
201 // Defer unlock so that runtime.Goexit during init does the unlock too.
211 main_init_done = make(chan bool)
213 if _cgo_thread_start == nil {
214 throw("_cgo_thread_start missing")
216 if GOOS != "windows" {
217 if _cgo_setenv == nil {
218 throw("_cgo_setenv missing")
220 if _cgo_unsetenv == nil {
221 throw("_cgo_unsetenv missing")
224 if _cgo_notify_runtime_init_done == nil {
225 throw("_cgo_notify_runtime_init_done missing")
227 // Start the template thread in case we enter Go from
228 // a C-created thread and need to create a new thread.
229 startTemplateThread()
230 cgocall(_cgo_notify_runtime_init_done, nil)
233 doInit(&main_inittask)
235 // Disable init tracing after main init done to avoid overhead
236 // of collecting statistics in malloc and newproc
237 inittrace.active = false
239 close(main_init_done)
244 if isarchive || islibrary {
245 // A program compiled with -buildmode=c-archive or c-shared
246 // has a main, but it is not executed.
249 fn := main_main // make an indirect call, as the linker doesn't know the address of the main package when laying down the runtime
252 runExitHooks(0) // run hooks now, since racefini does not return
256 // Make racy client program work: if panicking on
257 // another goroutine at the same time as main returns,
258 // let the other goroutine finish printing the panic trace.
259 // Once it does, it will exit. See issues 3934 and 20018.
260 if runningPanicDefers.Load() != 0 {
261 // Running deferred functions should not take long.
262 for c := 0; c < 1000; c++ {
263 if runningPanicDefers.Load() == 0 {
269 if panicking.Load() != 0 {
270 gopark(nil, nil, waitReasonPanicWait, traceEvGoStop, 1)
281 // os_beforeExit is called from os.Exit(0).
283 //go:linkname os_beforeExit os.runtime_beforeExit
284 func os_beforeExit(exitCode int) {
285 runExitHooks(exitCode)
286 if exitCode == 0 && raceenabled {
291 // start forcegc helper goroutine
296 func forcegchelper() {
298 lockInit(&forcegc.lock, lockRankForcegc)
301 if forcegc.idle.Load() {
302 throw("forcegc: phase error")
304 forcegc.idle.Store(true)
305 goparkunlock(&forcegc.lock, waitReasonForceGCIdle, traceEvGoBlock, 1)
306 // this goroutine is explicitly resumed by sysmon
307 if debug.gctrace > 0 {
310 // Time-triggered, fully concurrent.
311 gcStart(gcTrigger{kind: gcTriggerTime, now: nanotime()})
315 // Gosched yields the processor, allowing other goroutines to run. It does not
316 // suspend the current goroutine, so execution resumes automatically.
324 // goschedguarded yields the processor like gosched, but also checks
325 // for forbidden states and opts out of the yield in those cases.
328 func goschedguarded() {
329 mcall(goschedguarded_m)
332 // goschedIfBusy yields the processor like gosched, but only does so if
333 // there are no idle Ps or if we're on the only P and there's nothing in
334 // the run queue. In both cases, there is freely available idle time.
337 func goschedIfBusy() {
339 // Call gosched if gp.preempt is set; we may be in a tight loop that
340 // doesn't otherwise yield.
341 if !gp.preempt && sched.npidle.Load() > 0 {
347 // Puts the current goroutine into a waiting state and calls unlockf on the
350 // If unlockf returns false, the goroutine is resumed.
352 // unlockf must not access this G's stack, as it may be moved between
353 // the call to gopark and the call to unlockf.
355 // Note that because unlockf is called after putting the G into a waiting
356 // state, the G may have already been readied by the time unlockf is called
357 // unless there is external synchronization preventing the G from being
358 // readied. If unlockf returns false, it must guarantee that the G cannot be
359 // externally readied.
361 // Reason explains why the goroutine has been parked. It is displayed in stack
362 // traces and heap dumps. Reasons should be unique and descriptive. Do not
363 // re-use reasons, add new ones.
364 func gopark(unlockf func(*g, unsafe.Pointer) bool, lock unsafe.Pointer, reason waitReason, traceEv byte, traceskip int) {
365 if reason != waitReasonSleep {
366 checkTimeouts() // timeouts may expire while two goroutines keep the scheduler busy
370 status := readgstatus(gp)
371 if status != _Grunning && status != _Gscanrunning {
372 throw("gopark: bad g status")
375 mp.waitunlockf = unlockf
376 gp.waitreason = reason
377 mp.waittraceev = traceEv
378 mp.waittraceskip = traceskip
380 // can't do anything that might move the G between Ms here.
384 // Puts the current goroutine into a waiting state and unlocks the lock.
385 // The goroutine can be made runnable again by calling goready(gp).
386 func goparkunlock(lock *mutex, reason waitReason, traceEv byte, traceskip int) {
387 gopark(parkunlock_c, unsafe.Pointer(lock), reason, traceEv, traceskip)
390 func goready(gp *g, traceskip int) {
392 ready(gp, traceskip, true)
397 func acquireSudog() *sudog {
398 // Delicate dance: the semaphore implementation calls
399 // acquireSudog, acquireSudog calls new(sudog),
400 // new calls malloc, malloc can call the garbage collector,
401 // and the garbage collector calls the semaphore implementation
403 // Break the cycle by doing acquirem/releasem around new(sudog).
404 // The acquirem/releasem increments m.locks during new(sudog),
405 // which keeps the garbage collector from being invoked.
408 if len(pp.sudogcache) == 0 {
409 lock(&sched.sudoglock)
410 // First, try to grab a batch from central cache.
411 for len(pp.sudogcache) < cap(pp.sudogcache)/2 && sched.sudogcache != nil {
412 s := sched.sudogcache
413 sched.sudogcache = s.next
415 pp.sudogcache = append(pp.sudogcache, s)
417 unlock(&sched.sudoglock)
418 // If the central cache is empty, allocate a new one.
419 if len(pp.sudogcache) == 0 {
420 pp.sudogcache = append(pp.sudogcache, new(sudog))
423 n := len(pp.sudogcache)
424 s := pp.sudogcache[n-1]
425 pp.sudogcache[n-1] = nil
426 pp.sudogcache = pp.sudogcache[:n-1]
428 throw("acquireSudog: found s.elem != nil in cache")
435 func releaseSudog(s *sudog) {
437 throw("runtime: sudog with non-nil elem")
440 throw("runtime: sudog with non-false isSelect")
443 throw("runtime: sudog with non-nil next")
446 throw("runtime: sudog with non-nil prev")
448 if s.waitlink != nil {
449 throw("runtime: sudog with non-nil waitlink")
452 throw("runtime: sudog with non-nil c")
456 throw("runtime: releaseSudog with non-nil gp.param")
458 mp := acquirem() // avoid rescheduling to another P
460 if len(pp.sudogcache) == cap(pp.sudogcache) {
461 // Transfer half of local cache to the central cache.
462 var first, last *sudog
463 for len(pp.sudogcache) > cap(pp.sudogcache)/2 {
464 n := len(pp.sudogcache)
465 p := pp.sudogcache[n-1]
466 pp.sudogcache[n-1] = nil
467 pp.sudogcache = pp.sudogcache[:n-1]
475 lock(&sched.sudoglock)
476 last.next = sched.sudogcache
477 sched.sudogcache = first
478 unlock(&sched.sudoglock)
480 pp.sudogcache = append(pp.sudogcache, s)
484 // called from assembly.
485 func badmcall(fn func(*g)) {
486 throw("runtime: mcall called on m->g0 stack")
489 func badmcall2(fn func(*g)) {
490 throw("runtime: mcall function returned")
493 func badreflectcall() {
494 panic(plainError("arg size to reflect.call more than 1GB"))
498 //go:nowritebarrierrec
499 func badmorestackg0() {
500 writeErrStr("fatal: morestack on g0\n")
504 //go:nowritebarrierrec
505 func badmorestackgsignal() {
506 writeErrStr("fatal: morestack on gsignal\n")
514 func lockedOSThread() bool {
516 return gp.lockedm != 0 && gp.m.lockedg != 0
520 // allgs contains all Gs ever created (including dead Gs), and thus
523 // Access via the slice is protected by allglock or stop-the-world.
524 // Readers that cannot take the lock may (carefully!) use the atomic
529 // allglen and allgptr are atomic variables that contain len(allgs) and
530 // &allgs[0] respectively. Proper ordering depends on totally-ordered
531 // loads and stores. Writes are protected by allglock.
533 // allgptr is updated before allglen. Readers should read allglen
534 // before allgptr to ensure that allglen is always <= len(allgptr). New
535 // Gs appended during the race can be missed. For a consistent view of
536 // all Gs, allglock must be held.
538 // allgptr copies should always be stored as a concrete type or
539 // unsafe.Pointer, not uintptr, to ensure that GC can still reach it
540 // even if it points to a stale array.
545 func allgadd(gp *g) {
546 if readgstatus(gp) == _Gidle {
547 throw("allgadd: bad status Gidle")
551 allgs = append(allgs, gp)
552 if &allgs[0] != allgptr {
553 atomicstorep(unsafe.Pointer(&allgptr), unsafe.Pointer(&allgs[0]))
555 atomic.Storeuintptr(&allglen, uintptr(len(allgs)))
559 // allGsSnapshot returns a snapshot of the slice of all Gs.
561 // The world must be stopped or allglock must be held.
562 func allGsSnapshot() []*g {
563 assertWorldStoppedOrLockHeld(&allglock)
565 // Because the world is stopped or allglock is held, allgadd
566 // cannot happen concurrently with this. allgs grows
567 // monotonically and existing entries never change, so we can
568 // simply return a copy of the slice header. For added safety,
569 // we trim everything past len because that can still change.
570 return allgs[:len(allgs):len(allgs)]
573 // atomicAllG returns &allgs[0] and len(allgs) for use with atomicAllGIndex.
574 func atomicAllG() (**g, uintptr) {
575 length := atomic.Loaduintptr(&allglen)
576 ptr := (**g)(atomic.Loadp(unsafe.Pointer(&allgptr)))
580 // atomicAllGIndex returns ptr[i] with the allgptr returned from atomicAllG.
581 func atomicAllGIndex(ptr **g, i uintptr) *g {
582 return *(**g)(add(unsafe.Pointer(ptr), i*goarch.PtrSize))
585 // forEachG calls fn on every G from allgs.
587 // forEachG takes a lock to exclude concurrent addition of new Gs.
588 func forEachG(fn func(gp *g)) {
590 for _, gp := range allgs {
596 // forEachGRace calls fn on every G from allgs.
598 // forEachGRace avoids locking, but does not exclude addition of new Gs during
599 // execution, which may be missed.
600 func forEachGRace(fn func(gp *g)) {
601 ptr, length := atomicAllG()
602 for i := uintptr(0); i < length; i++ {
603 gp := atomicAllGIndex(ptr, i)
610 // Number of goroutine ids to grab from sched.goidgen to local per-P cache at once.
611 // 16 seems to provide enough amortization, but other than that it's mostly arbitrary number.
615 // cpuinit sets up CPU feature flags and calls internal/cpu.Initialize. env should be the complete
616 // value of the GODEBUG environment variable.
617 func cpuinit(env string) {
619 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
620 cpu.DebugOptions = true
624 // Support cpu feature variables are used in code generated by the compiler
625 // to guard execution of instructions that can not be assumed to be always supported.
628 x86HasPOPCNT = cpu.X86.HasPOPCNT
629 x86HasSSE41 = cpu.X86.HasSSE41
630 x86HasFMA = cpu.X86.HasFMA
633 armHasVFPv4 = cpu.ARM.HasVFPv4
636 arm64HasATOMICS = cpu.ARM64.HasATOMICS
640 // getGodebugEarly extracts the environment variable GODEBUG from the environment on
641 // Unix-like operating systems and returns it. This function exists to extract GODEBUG
642 // early before much of the runtime is initialized.
643 func getGodebugEarly() string {
644 const prefix = "GODEBUG="
647 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
648 // Similar to goenv_unix but extracts the environment value for
650 // TODO(moehrmann): remove when general goenvs() can be called before cpuinit()
652 for argv_index(argv, argc+1+n) != nil {
656 for i := int32(0); i < n; i++ {
657 p := argv_index(argv, argc+1+i)
658 s := unsafe.String(p, findnull(p))
660 if hasPrefix(s, prefix) {
661 env = gostring(p)[len(prefix):]
669 // The bootstrap sequence is:
673 // make & queue new G
674 // call runtime·mstart
676 // The new G calls runtime·main.
678 lockInit(&sched.lock, lockRankSched)
679 lockInit(&sched.sysmonlock, lockRankSysmon)
680 lockInit(&sched.deferlock, lockRankDefer)
681 lockInit(&sched.sudoglock, lockRankSudog)
682 lockInit(&deadlock, lockRankDeadlock)
683 lockInit(&paniclk, lockRankPanic)
684 lockInit(&allglock, lockRankAllg)
685 lockInit(&allpLock, lockRankAllp)
686 lockInit(&reflectOffs.lock, lockRankReflectOffs)
687 lockInit(&finlock, lockRankFin)
688 lockInit(&trace.bufLock, lockRankTraceBuf)
689 lockInit(&trace.stringsLock, lockRankTraceStrings)
690 lockInit(&trace.lock, lockRankTrace)
691 lockInit(&cpuprof.lock, lockRankCpuprof)
692 lockInit(&trace.stackTab.lock, lockRankTraceStackTab)
693 // Enforce that this lock is always a leaf lock.
694 // All of this lock's critical sections should be
696 lockInit(&memstats.heapStats.noPLock, lockRankLeafRank)
698 // raceinit must be the first call to race detector.
699 // In particular, it must be done before mallocinit below calls racemapshadow.
702 gp.racectx, raceprocctx0 = raceinit()
705 sched.maxmcount = 10000
707 // The world starts stopped.
713 godebug := getGodebugEarly()
714 initPageTrace(godebug) // must run after mallocinit but before anything allocates
715 cpuinit(godebug) // must run before alginit
716 alginit() // maps, hash, fastrand must not be used before this call
717 fastrandinit() // must run before mcommoninit
718 mcommoninit(gp.m, -1)
719 modulesinit() // provides activeModules
720 typelinksinit() // uses maps, activeModules
721 itabsinit() // uses activeModules
722 stkobjinit() // must run before GC starts
724 sigsave(&gp.m.sigmask)
725 initSigmask = gp.m.sigmask
732 // if disableMemoryProfiling is set, update MemProfileRate to 0 to turn off memprofile.
733 // Note: parsedebugvars may update MemProfileRate, but when disableMemoryProfiling is
734 // set to true by the linker, it means that nothing is consuming the profile, it is
735 // safe to set MemProfileRate to 0.
736 if disableMemoryProfiling {
741 sched.lastpoll.Store(nanotime())
743 if n, ok := atoi32(gogetenv("GOMAXPROCS")); ok && n > 0 {
746 if procresize(procs) != nil {
747 throw("unknown runnable goroutine during bootstrap")
751 // World is effectively started now, as P's can run.
754 if buildVersion == "" {
755 // Condition should never trigger. This code just serves
756 // to ensure runtime·buildVersion is kept in the resulting binary.
757 buildVersion = "unknown"
759 if len(modinfo) == 1 {
760 // Condition should never trigger. This code just serves
761 // to ensure runtime·modinfo is kept in the resulting binary.
766 func dumpgstatus(gp *g) {
768 print("runtime: gp: gp=", gp, ", goid=", gp.goid, ", gp->atomicstatus=", readgstatus(gp), "\n")
769 print("runtime: getg: g=", thisg, ", goid=", thisg.goid, ", g->atomicstatus=", readgstatus(thisg), "\n")
772 // sched.lock must be held.
774 assertLockHeld(&sched.lock)
776 if mcount() > sched.maxmcount {
777 print("runtime: program exceeds ", sched.maxmcount, "-thread limit\n")
778 throw("thread exhaustion")
782 // mReserveID returns the next ID to use for a new m. This new m is immediately
783 // considered 'running' by checkdead.
785 // sched.lock must be held.
786 func mReserveID() int64 {
787 assertLockHeld(&sched.lock)
789 if sched.mnext+1 < sched.mnext {
790 throw("runtime: thread ID overflow")
798 // Pre-allocated ID may be passed as 'id', or omitted by passing -1.
799 func mcommoninit(mp *m, id int64) {
802 // g0 stack won't make sense for user (and is not necessary unwindable).
804 callers(1, mp.createstack[:])
815 lo := uint32(int64Hash(uint64(mp.id), fastrandseed))
816 hi := uint32(int64Hash(uint64(cputicks()), ^fastrandseed))
820 // Same behavior as for 1.17.
821 // TODO: Simplify this.
822 if goarch.BigEndian {
823 mp.fastrand = uint64(lo)<<32 | uint64(hi)
825 mp.fastrand = uint64(hi)<<32 | uint64(lo)
829 if mp.gsignal != nil {
830 mp.gsignal.stackguard1 = mp.gsignal.stack.lo + _StackGuard
833 // Add to allm so garbage collector doesn't free g->m
834 // when it is just in a register or thread-local storage.
837 // NumCgoCall() iterates over allm w/o schedlock,
838 // so we need to publish it safely.
839 atomicstorep(unsafe.Pointer(&allm), unsafe.Pointer(mp))
842 // Allocate memory to hold a cgo traceback if the cgo call crashes.
843 if iscgo || GOOS == "solaris" || GOOS == "illumos" || GOOS == "windows" {
844 mp.cgoCallers = new(cgoCallers)
848 func (mp *m) becomeSpinning() {
850 sched.nmspinning.Add(1)
851 sched.needspinning.Store(0)
854 var fastrandseed uintptr
856 func fastrandinit() {
857 s := (*[unsafe.Sizeof(fastrandseed)]byte)(unsafe.Pointer(&fastrandseed))[:]
861 // Mark gp ready to run.
862 func ready(gp *g, traceskip int, next bool) {
864 traceGoUnpark(gp, traceskip)
867 status := readgstatus(gp)
870 mp := acquirem() // disable preemption because it can be holding p in a local var
871 if status&^_Gscan != _Gwaiting {
873 throw("bad g->status in ready")
876 // status is Gwaiting or Gscanwaiting, make Grunnable and put on runq
877 casgstatus(gp, _Gwaiting, _Grunnable)
878 runqput(mp.p.ptr(), gp, next)
883 // freezeStopWait is a large value that freezetheworld sets
884 // sched.stopwait to in order to request that all Gs permanently stop.
885 const freezeStopWait = 0x7fffffff
887 // freezing is set to non-zero if the runtime is trying to freeze the
889 var freezing atomic.Bool
891 // Similar to stopTheWorld but best-effort and can be called several times.
892 // There is no reverse operation, used during crashing.
893 // This function must not lock any mutexes.
894 func freezetheworld() {
896 // stopwait and preemption requests can be lost
897 // due to races with concurrently executing threads,
898 // so try several times
899 for i := 0; i < 5; i++ {
900 // this should tell the scheduler to not start any new goroutines
901 sched.stopwait = freezeStopWait
902 sched.gcwaiting.Store(true)
903 // this should stop running goroutines
905 break // no running goroutines
915 // All reads and writes of g's status go through readgstatus, casgstatus
916 // castogscanstatus, casfrom_Gscanstatus.
919 func readgstatus(gp *g) uint32 {
920 return gp.atomicstatus.Load()
923 // The Gscanstatuses are acting like locks and this releases them.
924 // If it proves to be a performance hit we should be able to make these
925 // simple atomic stores but for now we are going to throw if
926 // we see an inconsistent state.
927 func casfrom_Gscanstatus(gp *g, oldval, newval uint32) {
930 // Check that transition is valid.
933 print("runtime: casfrom_Gscanstatus bad oldval gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
935 throw("casfrom_Gscanstatus:top gp->status is not in scan state")
941 if newval == oldval&^_Gscan {
942 success = gp.atomicstatus.CompareAndSwap(oldval, newval)
946 print("runtime: casfrom_Gscanstatus failed gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
948 throw("casfrom_Gscanstatus: gp->status is not in scan state")
950 releaseLockRank(lockRankGscan)
953 // This will return false if the gp is not in the expected status and the cas fails.
954 // This acts like a lock acquire while the casfromgstatus acts like a lock release.
955 func castogscanstatus(gp *g, oldval, newval uint32) bool {
961 if newval == oldval|_Gscan {
962 r := gp.atomicstatus.CompareAndSwap(oldval, newval)
964 acquireLockRank(lockRankGscan)
970 print("runtime: castogscanstatus oldval=", hex(oldval), " newval=", hex(newval), "\n")
971 throw("castogscanstatus")
975 // casgstatusAlwaysTrack is a debug flag that causes casgstatus to always track
976 // various latencies on every transition instead of sampling them.
977 var casgstatusAlwaysTrack = false
979 // If asked to move to or from a Gscanstatus this will throw. Use the castogscanstatus
980 // and casfrom_Gscanstatus instead.
981 // casgstatus will loop if the g->atomicstatus is in a Gscan status until the routine that
982 // put it in the Gscan state is finished.
985 func casgstatus(gp *g, oldval, newval uint32) {
986 if (oldval&_Gscan != 0) || (newval&_Gscan != 0) || oldval == newval {
988 print("runtime: casgstatus: oldval=", hex(oldval), " newval=", hex(newval), "\n")
989 throw("casgstatus: bad incoming values")
993 acquireLockRank(lockRankGscan)
994 releaseLockRank(lockRankGscan)
996 // See https://golang.org/cl/21503 for justification of the yield delay.
997 const yieldDelay = 5 * 1000
1000 // loop if gp->atomicstatus is in a scan state giving
1001 // GC time to finish and change the state to oldval.
1002 for i := 0; !gp.atomicstatus.CompareAndSwap(oldval, newval); i++ {
1003 if oldval == _Gwaiting && gp.atomicstatus.Load() == _Grunnable {
1004 throw("casgstatus: waiting for Gwaiting but is Grunnable")
1007 nextYield = nanotime() + yieldDelay
1009 if nanotime() < nextYield {
1010 for x := 0; x < 10 && gp.atomicstatus.Load() != oldval; x++ {
1015 nextYield = nanotime() + yieldDelay/2
1019 if oldval == _Grunning {
1020 // Track every gTrackingPeriod time a goroutine transitions out of running.
1021 if casgstatusAlwaysTrack || gp.trackingSeq%gTrackingPeriod == 0 {
1030 // Handle various kinds of tracking.
1033 // - Time spent in runnable.
1034 // - Time spent blocked on a sync.Mutex or sync.RWMutex.
1037 // We transitioned out of runnable, so measure how much
1038 // time we spent in this state and add it to
1041 gp.runnableTime += now - gp.trackingStamp
1042 gp.trackingStamp = 0
1044 if !gp.waitreason.isMutexWait() {
1045 // Not blocking on a lock.
1048 // Blocking on a lock, measure it. Note that because we're
1049 // sampling, we have to multiply by our sampling period to get
1050 // a more representative estimate of the absolute value.
1051 // gTrackingPeriod also represents an accurate sampling period
1052 // because we can only enter this state from _Grunning.
1054 sched.totalMutexWaitTime.Add((now - gp.trackingStamp) * gTrackingPeriod)
1055 gp.trackingStamp = 0
1059 if !gp.waitreason.isMutexWait() {
1060 // Not blocking on a lock.
1063 // Blocking on a lock. Write down the timestamp.
1065 gp.trackingStamp = now
1067 // We just transitioned into runnable, so record what
1068 // time that happened.
1070 gp.trackingStamp = now
1072 // We're transitioning into running, so turn off
1073 // tracking and record how much time we spent in
1076 sched.timeToRun.record(gp.runnableTime)
1081 // casGToWaiting transitions gp from old to _Gwaiting, and sets the wait reason.
1083 // Use this over casgstatus when possible to ensure that a waitreason is set.
1084 func casGToWaiting(gp *g, old uint32, reason waitReason) {
1085 // Set the wait reason before calling casgstatus, because casgstatus will use it.
1086 gp.waitreason = reason
1087 casgstatus(gp, old, _Gwaiting)
1090 // casgstatus(gp, oldstatus, Gcopystack), assuming oldstatus is Gwaiting or Grunnable.
1091 // Returns old status. Cannot call casgstatus directly, because we are racing with an
1092 // async wakeup that might come in from netpoll. If we see Gwaiting from the readgstatus,
1093 // it might have become Grunnable by the time we get to the cas. If we called casgstatus,
1094 // it would loop waiting for the status to go back to Gwaiting, which it never will.
1097 func casgcopystack(gp *g) uint32 {
1099 oldstatus := readgstatus(gp) &^ _Gscan
1100 if oldstatus != _Gwaiting && oldstatus != _Grunnable {
1101 throw("copystack: bad status, not Gwaiting or Grunnable")
1103 if gp.atomicstatus.CompareAndSwap(oldstatus, _Gcopystack) {
1109 // casGToPreemptScan transitions gp from _Grunning to _Gscan|_Gpreempted.
1111 // TODO(austin): This is the only status operation that both changes
1112 // the status and locks the _Gscan bit. Rethink this.
1113 func casGToPreemptScan(gp *g, old, new uint32) {
1114 if old != _Grunning || new != _Gscan|_Gpreempted {
1115 throw("bad g transition")
1117 acquireLockRank(lockRankGscan)
1118 for !gp.atomicstatus.CompareAndSwap(_Grunning, _Gscan|_Gpreempted) {
1122 // casGFromPreempted attempts to transition gp from _Gpreempted to
1123 // _Gwaiting. If successful, the caller is responsible for
1124 // re-scheduling gp.
1125 func casGFromPreempted(gp *g, old, new uint32) bool {
1126 if old != _Gpreempted || new != _Gwaiting {
1127 throw("bad g transition")
1129 gp.waitreason = waitReasonPreempted
1130 return gp.atomicstatus.CompareAndSwap(_Gpreempted, _Gwaiting)
1133 // stopTheWorld stops all P's from executing goroutines, interrupting
1134 // all goroutines at GC safe points and records reason as the reason
1135 // for the stop. On return, only the current goroutine's P is running.
1136 // stopTheWorld must not be called from a system stack and the caller
1137 // must not hold worldsema. The caller must call startTheWorld when
1138 // other P's should resume execution.
1140 // stopTheWorld is safe for multiple goroutines to call at the
1141 // same time. Each will execute its own stop, and the stops will
1144 // This is also used by routines that do stack dumps. If the system is
1145 // in panic or being exited, this may not reliably stop all
1147 func stopTheWorld(reason string) {
1148 semacquire(&worldsema)
1150 gp.m.preemptoff = reason
1151 systemstack(func() {
1152 // Mark the goroutine which called stopTheWorld preemptible so its
1153 // stack may be scanned.
1154 // This lets a mark worker scan us while we try to stop the world
1155 // since otherwise we could get in a mutual preemption deadlock.
1156 // We must not modify anything on the G stack because a stack shrink
1157 // may occur. A stack shrink is otherwise OK though because in order
1158 // to return from this function (and to leave the system stack) we
1159 // must have preempted all goroutines, including any attempting
1160 // to scan our stack, in which case, any stack shrinking will
1161 // have already completed by the time we exit.
1162 // Don't provide a wait reason because we're still executing.
1163 casGToWaiting(gp, _Grunning, waitReasonStoppingTheWorld)
1164 stopTheWorldWithSema()
1165 casgstatus(gp, _Gwaiting, _Grunning)
1169 // startTheWorld undoes the effects of stopTheWorld.
1170 func startTheWorld() {
1171 systemstack(func() { startTheWorldWithSema(false) })
1173 // worldsema must be held over startTheWorldWithSema to ensure
1174 // gomaxprocs cannot change while worldsema is held.
1176 // Release worldsema with direct handoff to the next waiter, but
1177 // acquirem so that semrelease1 doesn't try to yield our time.
1179 // Otherwise if e.g. ReadMemStats is being called in a loop,
1180 // it might stomp on other attempts to stop the world, such as
1181 // for starting or ending GC. The operation this blocks is
1182 // so heavy-weight that we should just try to be as fair as
1185 // We don't want to just allow us to get preempted between now
1186 // and releasing the semaphore because then we keep everyone
1187 // (including, for example, GCs) waiting longer.
1190 semrelease1(&worldsema, true, 0)
1194 // stopTheWorldGC has the same effect as stopTheWorld, but blocks
1195 // until the GC is not running. It also blocks a GC from starting
1196 // until startTheWorldGC is called.
1197 func stopTheWorldGC(reason string) {
1199 stopTheWorld(reason)
1202 // startTheWorldGC undoes the effects of stopTheWorldGC.
1203 func startTheWorldGC() {
1208 // Holding worldsema grants an M the right to try to stop the world.
1209 var worldsema uint32 = 1
1211 // Holding gcsema grants the M the right to block a GC, and blocks
1212 // until the current GC is done. In particular, it prevents gomaxprocs
1213 // from changing concurrently.
1215 // TODO(mknyszek): Once gomaxprocs and the execution tracer can handle
1216 // being changed/enabled during a GC, remove this.
1217 var gcsema uint32 = 1
1219 // stopTheWorldWithSema is the core implementation of stopTheWorld.
1220 // The caller is responsible for acquiring worldsema and disabling
1221 // preemption first and then should stopTheWorldWithSema on the system
1224 // semacquire(&worldsema, 0)
1225 // m.preemptoff = "reason"
1226 // systemstack(stopTheWorldWithSema)
1228 // When finished, the caller must either call startTheWorld or undo
1229 // these three operations separately:
1231 // m.preemptoff = ""
1232 // systemstack(startTheWorldWithSema)
1233 // semrelease(&worldsema)
1235 // It is allowed to acquire worldsema once and then execute multiple
1236 // startTheWorldWithSema/stopTheWorldWithSema pairs.
1237 // Other P's are able to execute between successive calls to
1238 // startTheWorldWithSema and stopTheWorldWithSema.
1239 // Holding worldsema causes any other goroutines invoking
1240 // stopTheWorld to block.
1241 func stopTheWorldWithSema() {
1244 // If we hold a lock, then we won't be able to stop another M
1245 // that is blocked trying to acquire the lock.
1247 throw("stopTheWorld: holding locks")
1251 sched.stopwait = gomaxprocs
1252 sched.gcwaiting.Store(true)
1255 gp.m.p.ptr().status = _Pgcstop // Pgcstop is only diagnostic.
1257 // try to retake all P's in Psyscall status
1258 for _, pp := range allp {
1260 if s == _Psyscall && atomic.Cas(&pp.status, s, _Pgcstop) {
1272 pp, _ := pidleget(now)
1276 pp.status = _Pgcstop
1279 wait := sched.stopwait > 0
1282 // wait for remaining P's to stop voluntarily
1285 // wait for 100us, then try to re-preempt in case of any races
1286 if notetsleep(&sched.stopnote, 100*1000) {
1287 noteclear(&sched.stopnote)
1296 if sched.stopwait != 0 {
1297 bad = "stopTheWorld: not stopped (stopwait != 0)"
1299 for _, pp := range allp {
1300 if pp.status != _Pgcstop {
1301 bad = "stopTheWorld: not stopped (status != _Pgcstop)"
1305 if freezing.Load() {
1306 // Some other thread is panicking. This can cause the
1307 // sanity checks above to fail if the panic happens in
1308 // the signal handler on a stopped thread. Either way,
1309 // we should halt this thread.
1320 func startTheWorldWithSema(emitTraceEvent bool) int64 {
1321 assertWorldStopped()
1323 mp := acquirem() // disable preemption because it can be holding p in a local var
1324 if netpollinited() {
1325 list := netpoll(0) // non-blocking
1335 p1 := procresize(procs)
1336 sched.gcwaiting.Store(false)
1337 if sched.sysmonwait.Load() {
1338 sched.sysmonwait.Store(false)
1339 notewakeup(&sched.sysmonnote)
1352 throw("startTheWorld: inconsistent mp->nextp")
1355 notewakeup(&mp.park)
1357 // Start M to run P. Do not start another M below.
1362 // Capture start-the-world time before doing clean-up tasks.
1363 startTime := nanotime()
1368 // Wakeup an additional proc in case we have excessive runnable goroutines
1369 // in local queues or in the global queue. If we don't, the proc will park itself.
1370 // If we have lots of excessive work, resetspinning will unpark additional procs as necessary.
1378 // usesLibcall indicates whether this runtime performs system calls
1380 func usesLibcall() bool {
1382 case "aix", "darwin", "illumos", "ios", "solaris", "windows":
1385 return GOARCH == "386" || GOARCH == "amd64" || GOARCH == "arm" || GOARCH == "arm64"
1390 // mStackIsSystemAllocated indicates whether this runtime starts on a
1391 // system-allocated stack.
1392 func mStackIsSystemAllocated() bool {
1394 case "aix", "darwin", "plan9", "illumos", "ios", "solaris", "windows":
1398 case "386", "amd64", "arm", "arm64":
1405 // mstart is the entry-point for new Ms.
1406 // It is written in assembly, uses ABI0, is marked TOPFRAME, and calls mstart0.
1409 // mstart0 is the Go entry-point for new Ms.
1410 // This must not split the stack because we may not even have stack
1411 // bounds set up yet.
1413 // May run during STW (because it doesn't have a P yet), so write
1414 // barriers are not allowed.
1417 //go:nowritebarrierrec
1421 osStack := gp.stack.lo == 0
1423 // Initialize stack bounds from system stack.
1424 // Cgo may have left stack size in stack.hi.
1425 // minit may update the stack bounds.
1427 // Note: these bounds may not be very accurate.
1428 // We set hi to &size, but there are things above
1429 // it. The 1024 is supposed to compensate this,
1430 // but is somewhat arbitrary.
1433 size = 8192 * sys.StackGuardMultiplier
1435 gp.stack.hi = uintptr(noescape(unsafe.Pointer(&size)))
1436 gp.stack.lo = gp.stack.hi - size + 1024
1438 // Initialize stack guard so that we can start calling regular
1440 gp.stackguard0 = gp.stack.lo + _StackGuard
1441 // This is the g0, so we can also call go:systemstack
1442 // functions, which check stackguard1.
1443 gp.stackguard1 = gp.stackguard0
1446 // Exit this thread.
1447 if mStackIsSystemAllocated() {
1448 // Windows, Solaris, illumos, Darwin, AIX and Plan 9 always system-allocate
1449 // the stack, but put it in gp.stack before mstart,
1450 // so the logic above hasn't set osStack yet.
1456 // The go:noinline is to guarantee the getcallerpc/getcallersp below are safe,
1457 // so that we can set up g0.sched to return to the call of mstart1 above.
1464 throw("bad runtime·mstart")
1467 // Set up m.g0.sched as a label returning to just
1468 // after the mstart1 call in mstart0 above, for use by goexit0 and mcall.
1469 // We're never coming back to mstart1 after we call schedule,
1470 // so other calls can reuse the current frame.
1471 // And goexit0 does a gogo that needs to return from mstart1
1472 // and let mstart0 exit the thread.
1473 gp.sched.g = guintptr(unsafe.Pointer(gp))
1474 gp.sched.pc = getcallerpc()
1475 gp.sched.sp = getcallersp()
1480 // Install signal handlers; after minit so that minit can
1481 // prepare the thread to be able to handle the signals.
1486 if fn := gp.m.mstartfn; fn != nil {
1491 acquirep(gp.m.nextp.ptr())
1497 // mstartm0 implements part of mstart1 that only runs on the m0.
1499 // Write barriers are allowed here because we know the GC can't be
1500 // running yet, so they'll be no-ops.
1502 //go:yeswritebarrierrec
1504 // Create an extra M for callbacks on threads not created by Go.
1505 // An extra M is also needed on Windows for callbacks created by
1506 // syscall.NewCallback. See issue #6751 for details.
1507 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
1514 // mPark causes a thread to park itself, returning once woken.
1519 notesleep(&gp.m.park)
1520 noteclear(&gp.m.park)
1523 // mexit tears down and exits the current thread.
1525 // Don't call this directly to exit the thread, since it must run at
1526 // the top of the thread stack. Instead, use gogo(&gp.m.g0.sched) to
1527 // unwind the stack to the point that exits the thread.
1529 // It is entered with m.p != nil, so write barriers are allowed. It
1530 // will release the P before exiting.
1532 //go:yeswritebarrierrec
1533 func mexit(osStack bool) {
1537 // This is the main thread. Just wedge it.
1539 // On Linux, exiting the main thread puts the process
1540 // into a non-waitable zombie state. On Plan 9,
1541 // exiting the main thread unblocks wait even though
1542 // other threads are still running. On Solaris we can
1543 // neither exitThread nor return from mstart. Other
1544 // bad things probably happen on other platforms.
1546 // We could try to clean up this M more before wedging
1547 // it, but that complicates signal handling.
1548 handoffp(releasep())
1554 throw("locked m0 woke up")
1560 // Free the gsignal stack.
1561 if mp.gsignal != nil {
1562 stackfree(mp.gsignal.stack)
1563 // On some platforms, when calling into VDSO (e.g. nanotime)
1564 // we store our g on the gsignal stack, if there is one.
1565 // Now the stack is freed, unlink it from the m, so we
1566 // won't write to it when calling VDSO code.
1570 // Remove m from allm.
1572 for pprev := &allm; *pprev != nil; pprev = &(*pprev).alllink {
1578 throw("m not found in allm")
1580 // Delay reaping m until it's done with the stack.
1582 // Put mp on the free list, though it will not be reaped while freeWait
1583 // is freeMWait. mp is no longer reachable via allm, so even if it is
1584 // on an OS stack, we must keep a reference to mp alive so that the GC
1585 // doesn't free mp while we are still using it.
1587 // Note that the free list must not be linked through alllink because
1588 // some functions walk allm without locking, so may be using alllink.
1589 mp.freeWait.Store(freeMWait)
1590 mp.freelink = sched.freem
1594 atomic.Xadd64(&ncgocall, int64(mp.ncgocall))
1597 handoffp(releasep())
1598 // After this point we must not have write barriers.
1600 // Invoke the deadlock detector. This must happen after
1601 // handoffp because it may have started a new M to take our
1608 if GOOS == "darwin" || GOOS == "ios" {
1609 // Make sure pendingPreemptSignals is correct when an M exits.
1611 if mp.signalPending.Load() != 0 {
1612 pendingPreemptSignals.Add(-1)
1616 // Destroy all allocated resources. After this is called, we may no
1617 // longer take any locks.
1621 // No more uses of mp, so it is safe to drop the reference.
1622 mp.freeWait.Store(freeMRef)
1624 // Return from mstart and let the system thread
1625 // library free the g0 stack and terminate the thread.
1629 // mstart is the thread's entry point, so there's nothing to
1630 // return to. Exit the thread directly. exitThread will clear
1631 // m.freeWait when it's done with the stack and the m can be
1633 exitThread(&mp.freeWait)
1636 // forEachP calls fn(p) for every P p when p reaches a GC safe point.
1637 // If a P is currently executing code, this will bring the P to a GC
1638 // safe point and execute fn on that P. If the P is not executing code
1639 // (it is idle or in a syscall), this will call fn(p) directly while
1640 // preventing the P from exiting its state. This does not ensure that
1641 // fn will run on every CPU executing Go code, but it acts as a global
1642 // memory barrier. GC uses this as a "ragged barrier."
1644 // The caller must hold worldsema.
1647 func forEachP(fn func(*p)) {
1649 pp := getg().m.p.ptr()
1652 if sched.safePointWait != 0 {
1653 throw("forEachP: sched.safePointWait != 0")
1655 sched.safePointWait = gomaxprocs - 1
1656 sched.safePointFn = fn
1658 // Ask all Ps to run the safe point function.
1659 for _, p2 := range allp {
1661 atomic.Store(&p2.runSafePointFn, 1)
1666 // Any P entering _Pidle or _Psyscall from now on will observe
1667 // p.runSafePointFn == 1 and will call runSafePointFn when
1668 // changing its status to _Pidle/_Psyscall.
1670 // Run safe point function for all idle Ps. sched.pidle will
1671 // not change because we hold sched.lock.
1672 for p := sched.pidle.ptr(); p != nil; p = p.link.ptr() {
1673 if atomic.Cas(&p.runSafePointFn, 1, 0) {
1675 sched.safePointWait--
1679 wait := sched.safePointWait > 0
1682 // Run fn for the current P.
1685 // Force Ps currently in _Psyscall into _Pidle and hand them
1686 // off to induce safe point function execution.
1687 for _, p2 := range allp {
1689 if s == _Psyscall && p2.runSafePointFn == 1 && atomic.Cas(&p2.status, s, _Pidle) {
1699 // Wait for remaining Ps to run fn.
1702 // Wait for 100us, then try to re-preempt in
1703 // case of any races.
1705 // Requires system stack.
1706 if notetsleep(&sched.safePointNote, 100*1000) {
1707 noteclear(&sched.safePointNote)
1713 if sched.safePointWait != 0 {
1714 throw("forEachP: not done")
1716 for _, p2 := range allp {
1717 if p2.runSafePointFn != 0 {
1718 throw("forEachP: P did not run fn")
1723 sched.safePointFn = nil
1728 // runSafePointFn runs the safe point function, if any, for this P.
1729 // This should be called like
1731 // if getg().m.p.runSafePointFn != 0 {
1735 // runSafePointFn must be checked on any transition in to _Pidle or
1736 // _Psyscall to avoid a race where forEachP sees that the P is running
1737 // just before the P goes into _Pidle/_Psyscall and neither forEachP
1738 // nor the P run the safe-point function.
1739 func runSafePointFn() {
1740 p := getg().m.p.ptr()
1741 // Resolve the race between forEachP running the safe-point
1742 // function on this P's behalf and this P running the
1743 // safe-point function directly.
1744 if !atomic.Cas(&p.runSafePointFn, 1, 0) {
1747 sched.safePointFn(p)
1749 sched.safePointWait--
1750 if sched.safePointWait == 0 {
1751 notewakeup(&sched.safePointNote)
1756 // When running with cgo, we call _cgo_thread_start
1757 // to start threads for us so that we can play nicely with
1759 var cgoThreadStart unsafe.Pointer
1761 type cgothreadstart struct {
1767 // Allocate a new m unassociated with any thread.
1768 // Can use p for allocation context if needed.
1769 // fn is recorded as the new m's m.mstartfn.
1770 // id is optional pre-allocated m ID. Omit by passing -1.
1772 // This function is allowed to have write barriers even if the caller
1773 // isn't because it borrows pp.
1775 //go:yeswritebarrierrec
1776 func allocm(pp *p, fn func(), id int64) *m {
1779 // The caller owns pp, but we may borrow (i.e., acquirep) it. We must
1780 // disable preemption to ensure it is not stolen, which would make the
1781 // caller lose ownership.
1786 acquirep(pp) // temporarily borrow p for mallocs in this function
1789 // Release the free M list. We need to do this somewhere and
1790 // this may free up a stack we can use.
1791 if sched.freem != nil {
1794 for freem := sched.freem; freem != nil; {
1795 wait := freem.freeWait.Load()
1796 if wait == freeMWait {
1797 next := freem.freelink
1798 freem.freelink = newList
1803 // Free the stack if needed. For freeMRef, there is
1804 // nothing to do except drop freem from the sched.freem
1806 if wait == freeMStack {
1807 // stackfree must be on the system stack, but allocm is
1808 // reachable off the system stack transitively from
1810 systemstack(func() {
1811 stackfree(freem.g0.stack)
1814 freem = freem.freelink
1816 sched.freem = newList
1824 // In case of cgo or Solaris or illumos or Darwin, pthread_create will make us a stack.
1825 // Windows and Plan 9 will layout sched stack on OS stack.
1826 if iscgo || mStackIsSystemAllocated() {
1829 mp.g0 = malg(8192 * sys.StackGuardMultiplier)
1833 if pp == gp.m.p.ptr() {
1838 allocmLock.runlock()
1842 // needm is called when a cgo callback happens on a
1843 // thread without an m (a thread not created by Go).
1844 // In this case, needm is expected to find an m to use
1845 // and return with m, g initialized correctly.
1846 // Since m and g are not set now (likely nil, but see below)
1847 // needm is limited in what routines it can call. In particular
1848 // it can only call nosplit functions (textflag 7) and cannot
1849 // do any scheduling that requires an m.
1851 // In order to avoid needing heavy lifting here, we adopt
1852 // the following strategy: there is a stack of available m's
1853 // that can be stolen. Using compare-and-swap
1854 // to pop from the stack has ABA races, so we simulate
1855 // a lock by doing an exchange (via Casuintptr) to steal the stack
1856 // head and replace the top pointer with MLOCKED (1).
1857 // This serves as a simple spin lock that we can use even
1858 // without an m. The thread that locks the stack in this way
1859 // unlocks the stack by storing a valid stack head pointer.
1861 // In order to make sure that there is always an m structure
1862 // available to be stolen, we maintain the invariant that there
1863 // is always one more than needed. At the beginning of the
1864 // program (if cgo is in use) the list is seeded with a single m.
1865 // If needm finds that it has taken the last m off the list, its job
1866 // is - once it has installed its own m so that it can do things like
1867 // allocate memory - to create a spare m and put it on the list.
1869 // Each of these extra m's also has a g0 and a curg that are
1870 // pressed into service as the scheduling stack and current
1871 // goroutine for the duration of the cgo callback.
1873 // When the callback is done with the m, it calls dropm to
1874 // put the m back on the list.
1878 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
1879 // Can happen if C/C++ code calls Go from a global ctor.
1880 // Can also happen on Windows if a global ctor uses a
1881 // callback created by syscall.NewCallback. See issue #6751
1884 // Can not throw, because scheduler is not initialized yet.
1885 writeErrStr("fatal error: cgo callback before cgo call\n")
1889 // Save and block signals before getting an M.
1890 // The signal handler may call needm itself,
1891 // and we must avoid a deadlock. Also, once g is installed,
1892 // any incoming signals will try to execute,
1893 // but we won't have the sigaltstack settings and other data
1894 // set up appropriately until the end of minit, which will
1895 // unblock the signals. This is the same dance as when
1896 // starting a new m to run Go code via newosproc.
1901 // Lock extra list, take head, unlock popped list.
1902 // nilokay=false is safe here because of the invariant above,
1903 // that the extra list always contains or will soon contain
1905 mp := lockextra(false)
1907 // Set needextram when we've just emptied the list,
1908 // so that the eventual call into cgocallbackg will
1909 // allocate a new m for the extra list. We delay the
1910 // allocation until then so that it can be done
1911 // after exitsyscall makes sure it is okay to be
1912 // running at all (that is, there's no garbage collection
1913 // running right now).
1914 mp.needextram = mp.schedlink == 0
1916 unlockextra(mp.schedlink.ptr())
1918 // Store the original signal mask for use by minit.
1919 mp.sigmask = sigmask
1921 // Install TLS on some platforms (previously setg
1922 // would do this if necessary).
1925 // Install g (= m->g0) and set the stack bounds
1926 // to match the current stack. We don't actually know
1927 // how big the stack is, like we don't know how big any
1928 // scheduling stack is, but we assume there's at least 32 kB,
1929 // which is more than enough for us.
1932 gp.stack.hi = getcallersp() + 1024
1933 gp.stack.lo = getcallersp() - 32*1024
1934 gp.stackguard0 = gp.stack.lo + _StackGuard
1936 // Initialize this thread to use the m.
1940 // mp.curg is now a real goroutine.
1941 casgstatus(mp.curg, _Gdead, _Gsyscall)
1945 // newextram allocates m's and puts them on the extra list.
1946 // It is called with a working local m, so that it can do things
1947 // like call schedlock and allocate.
1949 c := extraMWaiters.Swap(0)
1951 for i := uint32(0); i < c; i++ {
1955 // Make sure there is at least one extra M.
1956 mp := lockextra(true)
1964 // oneNewExtraM allocates an m and puts it on the extra list.
1965 func oneNewExtraM() {
1966 // Create extra goroutine locked to extra m.
1967 // The goroutine is the context in which the cgo callback will run.
1968 // The sched.pc will never be returned to, but setting it to
1969 // goexit makes clear to the traceback routines where
1970 // the goroutine stack ends.
1971 mp := allocm(nil, nil, -1)
1973 gp.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum
1974 gp.sched.sp = gp.stack.hi
1975 gp.sched.sp -= 4 * goarch.PtrSize // extra space in case of reads slightly beyond frame
1977 gp.sched.g = guintptr(unsafe.Pointer(gp))
1978 gp.syscallpc = gp.sched.pc
1979 gp.syscallsp = gp.sched.sp
1980 gp.stktopsp = gp.sched.sp
1981 // malg returns status as _Gidle. Change to _Gdead before
1982 // adding to allg where GC can see it. We use _Gdead to hide
1983 // this from tracebacks and stack scans since it isn't a
1984 // "real" goroutine until needm grabs it.
1985 casgstatus(gp, _Gidle, _Gdead)
1992 gp.goid = sched.goidgen.Add(1)
1993 gp.sysblocktraced = true
1995 gp.racectx = racegostart(abi.FuncPCABIInternal(newextram) + sys.PCQuantum)
1998 // Trigger two trace events for the locked g in the extra m,
1999 // since the next event of the g will be traceEvGoSysExit in exitsyscall,
2000 // while calling from C thread to Go.
2001 traceGoCreate(gp, 0) // no start pc
2003 traceEvent(traceEvGoInSyscall, -1, gp.goid)
2005 // put on allg for garbage collector
2008 // gp is now on the allg list, but we don't want it to be
2009 // counted by gcount. It would be more "proper" to increment
2010 // sched.ngfree, but that requires locking. Incrementing ngsys
2011 // has the same effect.
2014 // Add m to the extra list.
2015 mnext := lockextra(true)
2016 mp.schedlink.set(mnext)
2021 // dropm is called when a cgo callback has called needm but is now
2022 // done with the callback and returning back into the non-Go thread.
2023 // It puts the current m back onto the extra list.
2025 // The main expense here is the call to signalstack to release the
2026 // m's signal stack, and then the call to needm on the next callback
2027 // from this thread. It is tempting to try to save the m for next time,
2028 // which would eliminate both these costs, but there might not be
2029 // a next time: the current thread (which Go does not control) might exit.
2030 // If we saved the m for that thread, there would be an m leak each time
2031 // such a thread exited. Instead, we acquire and release an m on each
2032 // call. These should typically not be scheduling operations, just a few
2033 // atomics, so the cost should be small.
2035 // TODO(rsc): An alternative would be to allocate a dummy pthread per-thread
2036 // variable using pthread_key_create. Unlike the pthread keys we already use
2037 // on OS X, this dummy key would never be read by Go code. It would exist
2038 // only so that we could register at thread-exit-time destructor.
2039 // That destructor would put the m back onto the extra list.
2040 // This is purely a performance optimization. The current version,
2041 // in which dropm happens on each cgo call, is still correct too.
2042 // We may have to keep the current version on systems with cgo
2043 // but without pthreads, like Windows.
2045 // Clear m and g, and return m to the extra list.
2046 // After the call to setg we can only call nosplit functions
2047 // with no pointer manipulation.
2050 // Return mp.curg to dead state.
2051 casgstatus(mp.curg, _Gsyscall, _Gdead)
2052 mp.curg.preemptStop = false
2055 // Block signals before unminit.
2056 // Unminit unregisters the signal handling stack (but needs g on some systems).
2057 // Setg(nil) clears g, which is the signal handler's cue not to run Go handlers.
2058 // It's important not to try to handle a signal between those two steps.
2059 sigmask := mp.sigmask
2063 mnext := lockextra(true)
2065 mp.schedlink.set(mnext)
2069 // Commit the release of mp.
2072 msigrestore(sigmask)
2075 // A helper function for EnsureDropM.
2076 func getm() uintptr {
2077 return uintptr(unsafe.Pointer(getg().m))
2080 var extram atomic.Uintptr
2081 var extraMCount uint32 // Protected by lockextra
2082 var extraMWaiters atomic.Uint32
2084 // lockextra locks the extra list and returns the list head.
2085 // The caller must unlock the list by storing a new list head
2086 // to extram. If nilokay is true, then lockextra will
2087 // return a nil list head if that's what it finds. If nilokay is false,
2088 // lockextra will keep waiting until the list head is no longer nil.
2091 func lockextra(nilokay bool) *m {
2096 old := extram.Load()
2101 if old == 0 && !nilokay {
2103 // Add 1 to the number of threads
2104 // waiting for an M.
2105 // This is cleared by newextram.
2106 extraMWaiters.Add(1)
2112 if extram.CompareAndSwap(old, locked) {
2113 return (*m)(unsafe.Pointer(old))
2121 func unlockextra(mp *m) {
2122 extram.Store(uintptr(unsafe.Pointer(mp)))
2126 // allocmLock is locked for read when creating new Ms in allocm and their
2127 // addition to allm. Thus acquiring this lock for write blocks the
2128 // creation of new Ms.
2131 // execLock serializes exec and clone to avoid bugs or unspecified
2132 // behaviour around exec'ing while creating/destroying threads. See
2137 // These errors are reported (via writeErrStr) by some OS-specific
2138 // versions of newosproc and newosproc0.
2140 failthreadcreate = "runtime: failed to create new OS thread\n"
2141 failallocatestack = "runtime: failed to allocate stack for the new OS thread\n"
2144 // newmHandoff contains a list of m structures that need new OS threads.
2145 // This is used by newm in situations where newm itself can't safely
2146 // start an OS thread.
2147 var newmHandoff struct {
2150 // newm points to a list of M structures that need new OS
2151 // threads. The list is linked through m.schedlink.
2154 // waiting indicates that wake needs to be notified when an m
2155 // is put on the list.
2159 // haveTemplateThread indicates that the templateThread has
2160 // been started. This is not protected by lock. Use cas to set
2162 haveTemplateThread uint32
2165 // Create a new m. It will start off with a call to fn, or else the scheduler.
2166 // fn needs to be static and not a heap allocated closure.
2167 // May run with m.p==nil, so write barriers are not allowed.
2169 // id is optional pre-allocated m ID. Omit by passing -1.
2171 //go:nowritebarrierrec
2172 func newm(fn func(), pp *p, id int64) {
2173 // allocm adds a new M to allm, but they do not start until created by
2174 // the OS in newm1 or the template thread.
2176 // doAllThreadsSyscall requires that every M in allm will eventually
2177 // start and be signal-able, even with a STW.
2179 // Disable preemption here until we start the thread to ensure that
2180 // newm is not preempted between allocm and starting the new thread,
2181 // ensuring that anything added to allm is guaranteed to eventually
2185 mp := allocm(pp, fn, id)
2187 mp.sigmask = initSigmask
2188 if gp := getg(); gp != nil && gp.m != nil && (gp.m.lockedExt != 0 || gp.m.incgo) && GOOS != "plan9" {
2189 // We're on a locked M or a thread that may have been
2190 // started by C. The kernel state of this thread may
2191 // be strange (the user may have locked it for that
2192 // purpose). We don't want to clone that into another
2193 // thread. Instead, ask a known-good thread to create
2194 // the thread for us.
2196 // This is disabled on Plan 9. See golang.org/issue/22227.
2198 // TODO: This may be unnecessary on Windows, which
2199 // doesn't model thread creation off fork.
2200 lock(&newmHandoff.lock)
2201 if newmHandoff.haveTemplateThread == 0 {
2202 throw("on a locked thread with no template thread")
2204 mp.schedlink = newmHandoff.newm
2205 newmHandoff.newm.set(mp)
2206 if newmHandoff.waiting {
2207 newmHandoff.waiting = false
2208 notewakeup(&newmHandoff.wake)
2210 unlock(&newmHandoff.lock)
2211 // The M has not started yet, but the template thread does not
2212 // participate in STW, so it will always process queued Ms and
2213 // it is safe to releasem.
2223 var ts cgothreadstart
2224 if _cgo_thread_start == nil {
2225 throw("_cgo_thread_start missing")
2228 ts.tls = (*uint64)(unsafe.Pointer(&mp.tls[0]))
2229 ts.fn = unsafe.Pointer(abi.FuncPCABI0(mstart))
2231 msanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2234 asanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2236 execLock.rlock() // Prevent process clone.
2237 asmcgocall(_cgo_thread_start, unsafe.Pointer(&ts))
2241 execLock.rlock() // Prevent process clone.
2246 // startTemplateThread starts the template thread if it is not already
2249 // The calling thread must itself be in a known-good state.
2250 func startTemplateThread() {
2251 if GOARCH == "wasm" { // no threads on wasm yet
2255 // Disable preemption to guarantee that the template thread will be
2256 // created before a park once haveTemplateThread is set.
2258 if !atomic.Cas(&newmHandoff.haveTemplateThread, 0, 1) {
2262 newm(templateThread, nil, -1)
2266 // templateThread is a thread in a known-good state that exists solely
2267 // to start new threads in known-good states when the calling thread
2268 // may not be in a good state.
2270 // Many programs never need this, so templateThread is started lazily
2271 // when we first enter a state that might lead to running on a thread
2272 // in an unknown state.
2274 // templateThread runs on an M without a P, so it must not have write
2277 //go:nowritebarrierrec
2278 func templateThread() {
2285 lock(&newmHandoff.lock)
2286 for newmHandoff.newm != 0 {
2287 newm := newmHandoff.newm.ptr()
2288 newmHandoff.newm = 0
2289 unlock(&newmHandoff.lock)
2291 next := newm.schedlink.ptr()
2296 lock(&newmHandoff.lock)
2298 newmHandoff.waiting = true
2299 noteclear(&newmHandoff.wake)
2300 unlock(&newmHandoff.lock)
2301 notesleep(&newmHandoff.wake)
2305 // Stops execution of the current m until new work is available.
2306 // Returns with acquired P.
2310 if gp.m.locks != 0 {
2311 throw("stopm holding locks")
2314 throw("stopm holding p")
2317 throw("stopm spinning")
2324 acquirep(gp.m.nextp.ptr())
2329 // startm's caller incremented nmspinning. Set the new M's spinning.
2330 getg().m.spinning = true
2333 // Schedules some M to run the p (creates an M if necessary).
2334 // If p==nil, tries to get an idle P, if no idle P's does nothing.
2335 // May run with m.p==nil, so write barriers are not allowed.
2336 // If spinning is set, the caller has incremented nmspinning and must provide a
2337 // P. startm will set m.spinning in the newly started M.
2339 // Callers passing a non-nil P must call from a non-preemptible context. See
2340 // comment on acquirem below.
2342 // Must not have write barriers because this may be called without a P.
2344 //go:nowritebarrierrec
2345 func startm(pp *p, spinning bool) {
2346 // Disable preemption.
2348 // Every owned P must have an owner that will eventually stop it in the
2349 // event of a GC stop request. startm takes transient ownership of a P
2350 // (either from argument or pidleget below) and transfers ownership to
2351 // a started M, which will be responsible for performing the stop.
2353 // Preemption must be disabled during this transient ownership,
2354 // otherwise the P this is running on may enter GC stop while still
2355 // holding the transient P, leaving that P in limbo and deadlocking the
2358 // Callers passing a non-nil P must already be in non-preemptible
2359 // context, otherwise such preemption could occur on function entry to
2360 // startm. Callers passing a nil P may be preemptible, so we must
2361 // disable preemption before acquiring a P from pidleget below.
2366 // TODO(prattmic): All remaining calls to this function
2367 // with _p_ == nil could be cleaned up to find a P
2368 // before calling startm.
2369 throw("startm: P required for spinning=true")
2380 // No M is available, we must drop sched.lock and call newm.
2381 // However, we already own a P to assign to the M.
2383 // Once sched.lock is released, another G (e.g., in a syscall),
2384 // could find no idle P while checkdead finds a runnable G but
2385 // no running M's because this new M hasn't started yet, thus
2386 // throwing in an apparent deadlock.
2388 // Avoid this situation by pre-allocating the ID for the new M,
2389 // thus marking it as 'running' before we drop sched.lock. This
2390 // new M will eventually run the scheduler to execute any
2397 // The caller incremented nmspinning, so set m.spinning in the new M.
2401 // Ownership transfer of pp committed by start in newm.
2402 // Preemption is now safe.
2408 throw("startm: m is spinning")
2411 throw("startm: m has p")
2413 if spinning && !runqempty(pp) {
2414 throw("startm: p has runnable gs")
2416 // The caller incremented nmspinning, so set m.spinning in the new M.
2417 nmp.spinning = spinning
2419 notewakeup(&nmp.park)
2420 // Ownership transfer of pp committed by wakeup. Preemption is now
2425 // Hands off P from syscall or locked M.
2426 // Always runs without a P, so write barriers are not allowed.
2428 //go:nowritebarrierrec
2429 func handoffp(pp *p) {
2430 // handoffp must start an M in any situation where
2431 // findrunnable would return a G to run on pp.
2433 // if it has local work, start it straight away
2434 if !runqempty(pp) || sched.runqsize != 0 {
2438 // if there's trace work to do, start it straight away
2439 if (trace.enabled || trace.shutdown) && traceReaderAvailable() != nil {
2443 // if it has GC work, start it straight away
2444 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) {
2448 // no local work, check that there are no spinning/idle M's,
2449 // otherwise our help is not required
2450 if sched.nmspinning.Load()+sched.npidle.Load() == 0 && sched.nmspinning.CompareAndSwap(0, 1) { // TODO: fast atomic
2451 sched.needspinning.Store(0)
2456 if sched.gcwaiting.Load() {
2457 pp.status = _Pgcstop
2459 if sched.stopwait == 0 {
2460 notewakeup(&sched.stopnote)
2465 if pp.runSafePointFn != 0 && atomic.Cas(&pp.runSafePointFn, 1, 0) {
2466 sched.safePointFn(pp)
2467 sched.safePointWait--
2468 if sched.safePointWait == 0 {
2469 notewakeup(&sched.safePointNote)
2472 if sched.runqsize != 0 {
2477 // If this is the last running P and nobody is polling network,
2478 // need to wakeup another M to poll network.
2479 if sched.npidle.Load() == gomaxprocs-1 && sched.lastpoll.Load() != 0 {
2485 // The scheduler lock cannot be held when calling wakeNetPoller below
2486 // because wakeNetPoller may call wakep which may call startm.
2487 when := nobarrierWakeTime(pp)
2496 // Tries to add one more P to execute G's.
2497 // Called when a G is made runnable (newproc, ready).
2498 // Must be called with a P.
2500 // Be conservative about spinning threads, only start one if none exist
2502 if sched.nmspinning.Load() != 0 || !sched.nmspinning.CompareAndSwap(0, 1) {
2506 // Disable preemption until ownership of pp transfers to the next M in
2507 // startm. Otherwise preemption here would leave pp stuck waiting to
2510 // See preemption comment on acquirem in startm for more details.
2515 pp, _ = pidlegetSpinning(0)
2517 if sched.nmspinning.Add(-1) < 0 {
2518 throw("wakep: negative nmspinning")
2524 // Since we always have a P, the race in the "No M is available"
2525 // comment in startm doesn't apply during the small window between the
2526 // unlock here and lock in startm. A checkdead in between will always
2527 // see at least one running M (ours).
2535 // Stops execution of the current m that is locked to a g until the g is runnable again.
2536 // Returns with acquired P.
2537 func stoplockedm() {
2540 if gp.m.lockedg == 0 || gp.m.lockedg.ptr().lockedm.ptr() != gp.m {
2541 throw("stoplockedm: inconsistent locking")
2544 // Schedule another M to run this p.
2549 // Wait until another thread schedules lockedg again.
2551 status := readgstatus(gp.m.lockedg.ptr())
2552 if status&^_Gscan != _Grunnable {
2553 print("runtime:stoplockedm: lockedg (atomicstatus=", status, ") is not Grunnable or Gscanrunnable\n")
2554 dumpgstatus(gp.m.lockedg.ptr())
2555 throw("stoplockedm: not runnable")
2557 acquirep(gp.m.nextp.ptr())
2561 // Schedules the locked m to run the locked gp.
2562 // May run during STW, so write barriers are not allowed.
2564 //go:nowritebarrierrec
2565 func startlockedm(gp *g) {
2566 mp := gp.lockedm.ptr()
2568 throw("startlockedm: locked to me")
2571 throw("startlockedm: m has p")
2573 // directly handoff current P to the locked m
2577 notewakeup(&mp.park)
2581 // Stops the current m for stopTheWorld.
2582 // Returns when the world is restarted.
2586 if !sched.gcwaiting.Load() {
2587 throw("gcstopm: not waiting for gc")
2590 gp.m.spinning = false
2591 // OK to just drop nmspinning here,
2592 // startTheWorld will unpark threads as necessary.
2593 if sched.nmspinning.Add(-1) < 0 {
2594 throw("gcstopm: negative nmspinning")
2599 pp.status = _Pgcstop
2601 if sched.stopwait == 0 {
2602 notewakeup(&sched.stopnote)
2608 // Schedules gp to run on the current M.
2609 // If inheritTime is true, gp inherits the remaining time in the
2610 // current time slice. Otherwise, it starts a new time slice.
2613 // Write barriers are allowed because this is called immediately after
2614 // acquiring a P in several places.
2616 //go:yeswritebarrierrec
2617 func execute(gp *g, inheritTime bool) {
2620 if goroutineProfile.active {
2621 // Make sure that gp has had its stack written out to the goroutine
2622 // profile, exactly as it was when the goroutine profiler first stopped
2624 tryRecordGoroutineProfile(gp, osyield)
2627 // Assign gp.m before entering _Grunning so running Gs have an
2631 casgstatus(gp, _Grunnable, _Grunning)
2634 gp.stackguard0 = gp.stack.lo + _StackGuard
2636 mp.p.ptr().schedtick++
2639 // Check whether the profiler needs to be turned on or off.
2640 hz := sched.profilehz
2641 if mp.profilehz != hz {
2642 setThreadCPUProfiler(hz)
2646 // GoSysExit has to happen when we have a P, but before GoStart.
2647 // So we emit it here.
2648 if gp.syscallsp != 0 && gp.sysblocktraced {
2649 traceGoSysExit(gp.sysexitticks)
2657 // Finds a runnable goroutine to execute.
2658 // Tries to steal from other P's, get g from local or global queue, poll network.
2659 // tryWakeP indicates that the returned goroutine is not normal (GC worker, trace
2660 // reader) so the caller should try to wake a P.
2661 func findRunnable() (gp *g, inheritTime, tryWakeP bool) {
2664 // The conditions here and in handoffp must agree: if
2665 // findrunnable would return a G to run, handoffp must start
2670 if sched.gcwaiting.Load() {
2674 if pp.runSafePointFn != 0 {
2678 // now and pollUntil are saved for work stealing later,
2679 // which may steal timers. It's important that between now
2680 // and then, nothing blocks, so these numbers remain mostly
2682 now, pollUntil, _ := checkTimers(pp, 0)
2684 // Try to schedule the trace reader.
2685 if trace.enabled || trace.shutdown {
2688 casgstatus(gp, _Gwaiting, _Grunnable)
2689 traceGoUnpark(gp, 0)
2690 return gp, false, true
2694 // Try to schedule a GC worker.
2695 if gcBlackenEnabled != 0 {
2696 gp, tnow := gcController.findRunnableGCWorker(pp, now)
2698 return gp, false, true
2703 // Check the global runnable queue once in a while to ensure fairness.
2704 // Otherwise two goroutines can completely occupy the local runqueue
2705 // by constantly respawning each other.
2706 if pp.schedtick%61 == 0 && sched.runqsize > 0 {
2708 gp := globrunqget(pp, 1)
2711 return gp, false, false
2715 // Wake up the finalizer G.
2716 if fingStatus.Load()&(fingWait|fingWake) == fingWait|fingWake {
2717 if gp := wakefing(); gp != nil {
2721 if *cgo_yield != nil {
2722 asmcgocall(*cgo_yield, nil)
2726 if gp, inheritTime := runqget(pp); gp != nil {
2727 return gp, inheritTime, false
2731 if sched.runqsize != 0 {
2733 gp := globrunqget(pp, 0)
2736 return gp, false, false
2741 // This netpoll is only an optimization before we resort to stealing.
2742 // We can safely skip it if there are no waiters or a thread is blocked
2743 // in netpoll already. If there is any kind of logical race with that
2744 // blocked thread (e.g. it has already returned from netpoll, but does
2745 // not set lastpoll yet), this thread will do blocking netpoll below
2747 if netpollinited() && netpollWaiters.Load() > 0 && sched.lastpoll.Load() != 0 {
2748 if list := netpoll(0); !list.empty() { // non-blocking
2751 casgstatus(gp, _Gwaiting, _Grunnable)
2753 traceGoUnpark(gp, 0)
2755 return gp, false, false
2759 // Spinning Ms: steal work from other Ps.
2761 // Limit the number of spinning Ms to half the number of busy Ps.
2762 // This is necessary to prevent excessive CPU consumption when
2763 // GOMAXPROCS>>1 but the program parallelism is low.
2764 if mp.spinning || 2*sched.nmspinning.Load() < gomaxprocs-sched.npidle.Load() {
2769 gp, inheritTime, tnow, w, newWork := stealWork(now)
2771 // Successfully stole.
2772 return gp, inheritTime, false
2775 // There may be new timer or GC work; restart to
2781 if w != 0 && (pollUntil == 0 || w < pollUntil) {
2782 // Earlier timer to wait for.
2787 // We have nothing to do.
2789 // If we're in the GC mark phase, can safely scan and blacken objects,
2790 // and have work to do, run idle-time marking rather than give up the P.
2791 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) && gcController.addIdleMarkWorker() {
2792 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
2794 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
2796 casgstatus(gp, _Gwaiting, _Grunnable)
2798 traceGoUnpark(gp, 0)
2800 return gp, false, false
2802 gcController.removeIdleMarkWorker()
2806 // If a callback returned and no other goroutine is awake,
2807 // then wake event handler goroutine which pauses execution
2808 // until a callback was triggered.
2809 gp, otherReady := beforeIdle(now, pollUntil)
2811 casgstatus(gp, _Gwaiting, _Grunnable)
2813 traceGoUnpark(gp, 0)
2815 return gp, false, false
2821 // Before we drop our P, make a snapshot of the allp slice,
2822 // which can change underfoot once we no longer block
2823 // safe-points. We don't need to snapshot the contents because
2824 // everything up to cap(allp) is immutable.
2825 allpSnapshot := allp
2826 // Also snapshot masks. Value changes are OK, but we can't allow
2827 // len to change out from under us.
2828 idlepMaskSnapshot := idlepMask
2829 timerpMaskSnapshot := timerpMask
2831 // return P and block
2833 if sched.gcwaiting.Load() || pp.runSafePointFn != 0 {
2837 if sched.runqsize != 0 {
2838 gp := globrunqget(pp, 0)
2840 return gp, false, false
2842 if !mp.spinning && sched.needspinning.Load() == 1 {
2843 // See "Delicate dance" comment below.
2848 if releasep() != pp {
2849 throw("findrunnable: wrong p")
2851 now = pidleput(pp, now)
2854 // Delicate dance: thread transitions from spinning to non-spinning
2855 // state, potentially concurrently with submission of new work. We must
2856 // drop nmspinning first and then check all sources again (with
2857 // #StoreLoad memory barrier in between). If we do it the other way
2858 // around, another thread can submit work after we've checked all
2859 // sources but before we drop nmspinning; as a result nobody will
2860 // unpark a thread to run the work.
2862 // This applies to the following sources of work:
2864 // * Goroutines added to a per-P run queue.
2865 // * New/modified-earlier timers on a per-P timer heap.
2866 // * Idle-priority GC work (barring golang.org/issue/19112).
2868 // If we discover new work below, we need to restore m.spinning as a
2869 // signal for resetspinning to unpark a new worker thread (because
2870 // there can be more than one starving goroutine).
2872 // However, if after discovering new work we also observe no idle Ps
2873 // (either here or in resetspinning), we have a problem. We may be
2874 // racing with a non-spinning M in the block above, having found no
2875 // work and preparing to release its P and park. Allowing that P to go
2876 // idle will result in loss of work conservation (idle P while there is
2877 // runnable work). This could result in complete deadlock in the
2878 // unlikely event that we discover new work (from netpoll) right as we
2879 // are racing with _all_ other Ps going idle.
2881 // We use sched.needspinning to synchronize with non-spinning Ms going
2882 // idle. If needspinning is set when they are about to drop their P,
2883 // they abort the drop and instead become a new spinning M on our
2884 // behalf. If we are not racing and the system is truly fully loaded
2885 // then no spinning threads are required, and the next thread to
2886 // naturally become spinning will clear the flag.
2888 // Also see "Worker thread parking/unparking" comment at the top of the
2890 wasSpinning := mp.spinning
2893 if sched.nmspinning.Add(-1) < 0 {
2894 throw("findrunnable: negative nmspinning")
2897 // Note the for correctness, only the last M transitioning from
2898 // spinning to non-spinning must perform these rechecks to
2899 // ensure no missed work. However, the runtime has some cases
2900 // of transient increments of nmspinning that are decremented
2901 // without going through this path, so we must be conservative
2902 // and perform the check on all spinning Ms.
2904 // See https://go.dev/issue/43997.
2906 // Check all runqueues once again.
2907 pp := checkRunqsNoP(allpSnapshot, idlepMaskSnapshot)
2914 // Check for idle-priority GC work again.
2915 pp, gp := checkIdleGCNoP()
2920 // Run the idle worker.
2921 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
2922 casgstatus(gp, _Gwaiting, _Grunnable)
2924 traceGoUnpark(gp, 0)
2926 return gp, false, false
2929 // Finally, check for timer creation or expiry concurrently with
2930 // transitioning from spinning to non-spinning.
2932 // Note that we cannot use checkTimers here because it calls
2933 // adjusttimers which may need to allocate memory, and that isn't
2934 // allowed when we don't have an active P.
2935 pollUntil = checkTimersNoP(allpSnapshot, timerpMaskSnapshot, pollUntil)
2938 // Poll network until next timer.
2939 if netpollinited() && (netpollWaiters.Load() > 0 || pollUntil != 0) && sched.lastpoll.Swap(0) != 0 {
2940 sched.pollUntil.Store(pollUntil)
2942 throw("findrunnable: netpoll with p")
2945 throw("findrunnable: netpoll with spinning")
2951 delay = pollUntil - now
2957 // When using fake time, just poll.
2960 list := netpoll(delay) // block until new work is available
2961 sched.pollUntil.Store(0)
2962 sched.lastpoll.Store(now)
2963 if faketime != 0 && list.empty() {
2964 // Using fake time and nothing is ready; stop M.
2965 // When all M's stop, checkdead will call timejump.
2970 pp, _ := pidleget(now)
2979 casgstatus(gp, _Gwaiting, _Grunnable)
2981 traceGoUnpark(gp, 0)
2983 return gp, false, false
2990 } else if pollUntil != 0 && netpollinited() {
2991 pollerPollUntil := sched.pollUntil.Load()
2992 if pollerPollUntil == 0 || pollerPollUntil > pollUntil {
3000 // pollWork reports whether there is non-background work this P could
3001 // be doing. This is a fairly lightweight check to be used for
3002 // background work loops, like idle GC. It checks a subset of the
3003 // conditions checked by the actual scheduler.
3004 func pollWork() bool {
3005 if sched.runqsize != 0 {
3008 p := getg().m.p.ptr()
3012 if netpollinited() && netpollWaiters.Load() > 0 && sched.lastpoll.Load() != 0 {
3013 if list := netpoll(0); !list.empty() {
3021 // stealWork attempts to steal a runnable goroutine or timer from any P.
3023 // If newWork is true, new work may have been readied.
3025 // If now is not 0 it is the current time. stealWork returns the passed time or
3026 // the current time if now was passed as 0.
3027 func stealWork(now int64) (gp *g, inheritTime bool, rnow, pollUntil int64, newWork bool) {
3028 pp := getg().m.p.ptr()
3032 const stealTries = 4
3033 for i := 0; i < stealTries; i++ {
3034 stealTimersOrRunNextG := i == stealTries-1
3036 for enum := stealOrder.start(fastrand()); !enum.done(); enum.next() {
3037 if sched.gcwaiting.Load() {
3038 // GC work may be available.
3039 return nil, false, now, pollUntil, true
3041 p2 := allp[enum.position()]
3046 // Steal timers from p2. This call to checkTimers is the only place
3047 // where we might hold a lock on a different P's timers. We do this
3048 // once on the last pass before checking runnext because stealing
3049 // from the other P's runnext should be the last resort, so if there
3050 // are timers to steal do that first.
3052 // We only check timers on one of the stealing iterations because
3053 // the time stored in now doesn't change in this loop and checking
3054 // the timers for each P more than once with the same value of now
3055 // is probably a waste of time.
3057 // timerpMask tells us whether the P may have timers at all. If it
3058 // can't, no need to check at all.
3059 if stealTimersOrRunNextG && timerpMask.read(enum.position()) {
3060 tnow, w, ran := checkTimers(p2, now)
3062 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3066 // Running the timers may have
3067 // made an arbitrary number of G's
3068 // ready and added them to this P's
3069 // local run queue. That invalidates
3070 // the assumption of runqsteal
3071 // that it always has room to add
3072 // stolen G's. So check now if there
3073 // is a local G to run.
3074 if gp, inheritTime := runqget(pp); gp != nil {
3075 return gp, inheritTime, now, pollUntil, ranTimer
3081 // Don't bother to attempt to steal if p2 is idle.
3082 if !idlepMask.read(enum.position()) {
3083 if gp := runqsteal(pp, p2, stealTimersOrRunNextG); gp != nil {
3084 return gp, false, now, pollUntil, ranTimer
3090 // No goroutines found to steal. Regardless, running a timer may have
3091 // made some goroutine ready that we missed. Indicate the next timer to
3093 return nil, false, now, pollUntil, ranTimer
3096 // Check all Ps for a runnable G to steal.
3098 // On entry we have no P. If a G is available to steal and a P is available,
3099 // the P is returned which the caller should acquire and attempt to steal the
3101 func checkRunqsNoP(allpSnapshot []*p, idlepMaskSnapshot pMask) *p {
3102 for id, p2 := range allpSnapshot {
3103 if !idlepMaskSnapshot.read(uint32(id)) && !runqempty(p2) {
3105 pp, _ := pidlegetSpinning(0)
3107 // Can't get a P, don't bother checking remaining Ps.
3116 // No work available.
3120 // Check all Ps for a timer expiring sooner than pollUntil.
3122 // Returns updated pollUntil value.
3123 func checkTimersNoP(allpSnapshot []*p, timerpMaskSnapshot pMask, pollUntil int64) int64 {
3124 for id, p2 := range allpSnapshot {
3125 if timerpMaskSnapshot.read(uint32(id)) {
3126 w := nobarrierWakeTime(p2)
3127 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3136 // Check for idle-priority GC, without a P on entry.
3138 // If some GC work, a P, and a worker G are all available, the P and G will be
3139 // returned. The returned P has not been wired yet.
3140 func checkIdleGCNoP() (*p, *g) {
3141 // N.B. Since we have no P, gcBlackenEnabled may change at any time; we
3142 // must check again after acquiring a P. As an optimization, we also check
3143 // if an idle mark worker is needed at all. This is OK here, because if we
3144 // observe that one isn't needed, at least one is currently running. Even if
3145 // it stops running, its own journey into the scheduler should schedule it
3146 // again, if need be (at which point, this check will pass, if relevant).
3147 if atomic.Load(&gcBlackenEnabled) == 0 || !gcController.needIdleMarkWorker() {
3150 if !gcMarkWorkAvailable(nil) {
3154 // Work is available; we can start an idle GC worker only if there is
3155 // an available P and available worker G.
3157 // We can attempt to acquire these in either order, though both have
3158 // synchronization concerns (see below). Workers are almost always
3159 // available (see comment in findRunnableGCWorker for the one case
3160 // there may be none). Since we're slightly less likely to find a P,
3161 // check for that first.
3163 // Synchronization: note that we must hold sched.lock until we are
3164 // committed to keeping it. Otherwise we cannot put the unnecessary P
3165 // back in sched.pidle without performing the full set of idle
3166 // transition checks.
3168 // If we were to check gcBgMarkWorkerPool first, we must somehow handle
3169 // the assumption in gcControllerState.findRunnableGCWorker that an
3170 // empty gcBgMarkWorkerPool is only possible if gcMarkDone is running.
3172 pp, now := pidlegetSpinning(0)
3178 // Now that we own a P, gcBlackenEnabled can't change (as it requires STW).
3179 if gcBlackenEnabled == 0 || !gcController.addIdleMarkWorker() {
3185 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
3189 gcController.removeIdleMarkWorker()
3195 return pp, node.gp.ptr()
3198 // wakeNetPoller wakes up the thread sleeping in the network poller if it isn't
3199 // going to wake up before the when argument; or it wakes an idle P to service
3200 // timers and the network poller if there isn't one already.
3201 func wakeNetPoller(when int64) {
3202 if sched.lastpoll.Load() == 0 {
3203 // In findrunnable we ensure that when polling the pollUntil
3204 // field is either zero or the time to which the current
3205 // poll is expected to run. This can have a spurious wakeup
3206 // but should never miss a wakeup.
3207 pollerPollUntil := sched.pollUntil.Load()
3208 if pollerPollUntil == 0 || pollerPollUntil > when {
3212 // There are no threads in the network poller, try to get
3213 // one there so it can handle new timers.
3214 if GOOS != "plan9" { // Temporary workaround - see issue #42303.
3220 func resetspinning() {
3223 throw("resetspinning: not a spinning m")
3225 gp.m.spinning = false
3226 nmspinning := sched.nmspinning.Add(-1)
3228 throw("findrunnable: negative nmspinning")
3230 // M wakeup policy is deliberately somewhat conservative, so check if we
3231 // need to wakeup another P here. See "Worker thread parking/unparking"
3232 // comment at the top of the file for details.
3236 // injectglist adds each runnable G on the list to some run queue,
3237 // and clears glist. If there is no current P, they are added to the
3238 // global queue, and up to npidle M's are started to run them.
3239 // Otherwise, for each idle P, this adds a G to the global queue
3240 // and starts an M. Any remaining G's are added to the current P's
3242 // This may temporarily acquire sched.lock.
3243 // Can run concurrently with GC.
3244 func injectglist(glist *gList) {
3249 for gp := glist.head.ptr(); gp != nil; gp = gp.schedlink.ptr() {
3250 traceGoUnpark(gp, 0)
3254 // Mark all the goroutines as runnable before we put them
3255 // on the run queues.
3256 head := glist.head.ptr()
3259 for gp := head; gp != nil; gp = gp.schedlink.ptr() {
3262 casgstatus(gp, _Gwaiting, _Grunnable)
3265 // Turn the gList into a gQueue.
3271 startIdle := func(n int) {
3272 for i := 0; i < n; i++ {
3273 mp := acquirem() // See comment in startm.
3276 pp, _ := pidlegetSpinning(0)
3289 pp := getg().m.p.ptr()
3292 globrunqputbatch(&q, int32(qsize))
3298 npidle := int(sched.npidle.Load())
3301 for n = 0; n < npidle && !q.empty(); n++ {
3307 globrunqputbatch(&globq, int32(n))
3314 runqputbatch(pp, &q, qsize)
3318 // One round of scheduler: find a runnable goroutine and execute it.
3324 throw("schedule: holding locks")
3327 if mp.lockedg != 0 {
3329 execute(mp.lockedg.ptr(), false) // Never returns.
3332 // We should not schedule away from a g that is executing a cgo call,
3333 // since the cgo call is using the m's g0 stack.
3335 throw("schedule: in cgo")
3342 // Safety check: if we are spinning, the run queue should be empty.
3343 // Check this before calling checkTimers, as that might call
3344 // goready to put a ready goroutine on the local run queue.
3345 if mp.spinning && (pp.runnext != 0 || pp.runqhead != pp.runqtail) {
3346 throw("schedule: spinning with local work")
3349 gp, inheritTime, tryWakeP := findRunnable() // blocks until work is available
3351 // This thread is going to run a goroutine and is not spinning anymore,
3352 // so if it was marked as spinning we need to reset it now and potentially
3353 // start a new spinning M.
3358 if sched.disable.user && !schedEnabled(gp) {
3359 // Scheduling of this goroutine is disabled. Put it on
3360 // the list of pending runnable goroutines for when we
3361 // re-enable user scheduling and look again.
3363 if schedEnabled(gp) {
3364 // Something re-enabled scheduling while we
3365 // were acquiring the lock.
3368 sched.disable.runnable.pushBack(gp)
3375 // If about to schedule a not-normal goroutine (a GCworker or tracereader),
3376 // wake a P if there is one.
3380 if gp.lockedm != 0 {
3381 // Hands off own p to the locked m,
3382 // then blocks waiting for a new p.
3387 execute(gp, inheritTime)
3390 // dropg removes the association between m and the current goroutine m->curg (gp for short).
3391 // Typically a caller sets gp's status away from Grunning and then
3392 // immediately calls dropg to finish the job. The caller is also responsible
3393 // for arranging that gp will be restarted using ready at an
3394 // appropriate time. After calling dropg and arranging for gp to be
3395 // readied later, the caller can do other work but eventually should
3396 // call schedule to restart the scheduling of goroutines on this m.
3400 setMNoWB(&gp.m.curg.m, nil)
3401 setGNoWB(&gp.m.curg, nil)
3404 // checkTimers runs any timers for the P that are ready.
3405 // If now is not 0 it is the current time.
3406 // It returns the passed time or the current time if now was passed as 0.
3407 // and the time when the next timer should run or 0 if there is no next timer,
3408 // and reports whether it ran any timers.
3409 // If the time when the next timer should run is not 0,
3410 // it is always larger than the returned time.
3411 // We pass now in and out to avoid extra calls of nanotime.
3413 //go:yeswritebarrierrec
3414 func checkTimers(pp *p, now int64) (rnow, pollUntil int64, ran bool) {
3415 // If it's not yet time for the first timer, or the first adjusted
3416 // timer, then there is nothing to do.
3417 next := pp.timer0When.Load()
3418 nextAdj := pp.timerModifiedEarliest.Load()
3419 if next == 0 || (nextAdj != 0 && nextAdj < next) {
3424 // No timers to run or adjust.
3425 return now, 0, false
3432 // Next timer is not ready to run, but keep going
3433 // if we would clear deleted timers.
3434 // This corresponds to the condition below where
3435 // we decide whether to call clearDeletedTimers.
3436 if pp != getg().m.p.ptr() || int(pp.deletedTimers.Load()) <= int(pp.numTimers.Load()/4) {
3437 return now, next, false
3441 lock(&pp.timersLock)
3443 if len(pp.timers) > 0 {
3444 adjusttimers(pp, now)
3445 for len(pp.timers) > 0 {
3446 // Note that runtimer may temporarily unlock
3448 if tw := runtimer(pp, now); tw != 0 {
3458 // If this is the local P, and there are a lot of deleted timers,
3459 // clear them out. We only do this for the local P to reduce
3460 // lock contention on timersLock.
3461 if pp == getg().m.p.ptr() && int(pp.deletedTimers.Load()) > len(pp.timers)/4 {
3462 clearDeletedTimers(pp)
3465 unlock(&pp.timersLock)
3467 return now, pollUntil, ran
3470 func parkunlock_c(gp *g, lock unsafe.Pointer) bool {
3471 unlock((*mutex)(lock))
3475 // park continuation on g0.
3476 func park_m(gp *g) {
3480 traceGoPark(mp.waittraceev, mp.waittraceskip)
3483 // N.B. Not using casGToWaiting here because the waitreason is
3484 // set by park_m's caller.
3485 casgstatus(gp, _Grunning, _Gwaiting)
3488 if fn := mp.waitunlockf; fn != nil {
3489 ok := fn(gp, mp.waitlock)
3490 mp.waitunlockf = nil
3494 traceGoUnpark(gp, 2)
3496 casgstatus(gp, _Gwaiting, _Grunnable)
3497 execute(gp, true) // Schedule it back, never returns.
3503 func goschedImpl(gp *g) {
3504 status := readgstatus(gp)
3505 if status&^_Gscan != _Grunning {
3507 throw("bad g status")
3509 casgstatus(gp, _Grunning, _Grunnable)
3518 // Gosched continuation on g0.
3519 func gosched_m(gp *g) {
3526 // goschedguarded is a forbidden-states-avoided version of gosched_m.
3527 func goschedguarded_m(gp *g) {
3529 if !canPreemptM(gp.m) {
3530 gogo(&gp.sched) // never return
3539 func gopreempt_m(gp *g) {
3546 // preemptPark parks gp and puts it in _Gpreempted.
3549 func preemptPark(gp *g) {
3551 traceGoPark(traceEvGoBlock, 0)
3553 status := readgstatus(gp)
3554 if status&^_Gscan != _Grunning {
3556 throw("bad g status")
3559 if gp.asyncSafePoint {
3560 // Double-check that async preemption does not
3561 // happen in SPWRITE assembly functions.
3562 // isAsyncSafePoint must exclude this case.
3563 f := findfunc(gp.sched.pc)
3565 throw("preempt at unknown pc")
3567 if f.flag&funcFlag_SPWRITE != 0 {
3568 println("runtime: unexpected SPWRITE function", funcname(f), "in async preempt")
3569 throw("preempt SPWRITE")
3573 // Transition from _Grunning to _Gscan|_Gpreempted. We can't
3574 // be in _Grunning when we dropg because then we'd be running
3575 // without an M, but the moment we're in _Gpreempted,
3576 // something could claim this G before we've fully cleaned it
3577 // up. Hence, we set the scan bit to lock down further
3578 // transitions until we can dropg.
3579 casGToPreemptScan(gp, _Grunning, _Gscan|_Gpreempted)
3581 casfrom_Gscanstatus(gp, _Gscan|_Gpreempted, _Gpreempted)
3585 // goyield is like Gosched, but it:
3586 // - emits a GoPreempt trace event instead of a GoSched trace event
3587 // - puts the current G on the runq of the current P instead of the globrunq
3593 func goyield_m(gp *g) {
3598 casgstatus(gp, _Grunning, _Grunnable)
3600 runqput(pp, gp, false)
3604 // Finishes execution of the current goroutine.
3615 // goexit continuation on g0.
3616 func goexit0(gp *g) {
3620 casgstatus(gp, _Grunning, _Gdead)
3621 gcController.addScannableStack(pp, -int64(gp.stack.hi-gp.stack.lo))
3622 if isSystemGoroutine(gp, false) {
3626 locked := gp.lockedm != 0
3629 gp.preemptStop = false
3630 gp.paniconfault = false
3631 gp._defer = nil // should be true already but just in case.
3632 gp._panic = nil // non-nil for Goexit during panic. points at stack-allocated data.
3634 gp.waitreason = waitReasonZero
3639 if gcBlackenEnabled != 0 && gp.gcAssistBytes > 0 {
3640 // Flush assist credit to the global pool. This gives
3641 // better information to pacing if the application is
3642 // rapidly creating an exiting goroutines.
3643 assistWorkPerByte := gcController.assistWorkPerByte.Load()
3644 scanCredit := int64(assistWorkPerByte * float64(gp.gcAssistBytes))
3645 gcController.bgScanCredit.Add(scanCredit)
3646 gp.gcAssistBytes = 0
3651 if GOARCH == "wasm" { // no threads yet on wasm
3653 schedule() // never returns
3656 if mp.lockedInt != 0 {
3657 print("invalid m->lockedInt = ", mp.lockedInt, "\n")
3658 throw("internal lockOSThread error")
3662 // The goroutine may have locked this thread because
3663 // it put it in an unusual kernel state. Kill it
3664 // rather than returning it to the thread pool.
3666 // Return to mstart, which will release the P and exit
3668 if GOOS != "plan9" { // See golang.org/issue/22227.
3671 // Clear lockedExt on plan9 since we may end up re-using
3679 // save updates getg().sched to refer to pc and sp so that a following
3680 // gogo will restore pc and sp.
3682 // save must not have write barriers because invoking a write barrier
3683 // can clobber getg().sched.
3686 //go:nowritebarrierrec
3687 func save(pc, sp uintptr) {
3690 if gp == gp.m.g0 || gp == gp.m.gsignal {
3691 // m.g0.sched is special and must describe the context
3692 // for exiting the thread. mstart1 writes to it directly.
3693 // m.gsignal.sched should not be used at all.
3694 // This check makes sure save calls do not accidentally
3695 // run in contexts where they'd write to system g's.
3696 throw("save on system g not allowed")
3703 // We need to ensure ctxt is zero, but can't have a write
3704 // barrier here. However, it should always already be zero.
3706 if gp.sched.ctxt != nil {
3711 // The goroutine g is about to enter a system call.
3712 // Record that it's not using the cpu anymore.
3713 // This is called only from the go syscall library and cgocall,
3714 // not from the low-level system calls used by the runtime.
3716 // Entersyscall cannot split the stack: the save must
3717 // make g->sched refer to the caller's stack segment, because
3718 // entersyscall is going to return immediately after.
3720 // Nothing entersyscall calls can split the stack either.
3721 // We cannot safely move the stack during an active call to syscall,
3722 // because we do not know which of the uintptr arguments are
3723 // really pointers (back into the stack).
3724 // In practice, this means that we make the fast path run through
3725 // entersyscall doing no-split things, and the slow path has to use systemstack
3726 // to run bigger things on the system stack.
3728 // reentersyscall is the entry point used by cgo callbacks, where explicitly
3729 // saved SP and PC are restored. This is needed when exitsyscall will be called
3730 // from a function further up in the call stack than the parent, as g->syscallsp
3731 // must always point to a valid stack frame. entersyscall below is the normal
3732 // entry point for syscalls, which obtains the SP and PC from the caller.
3735 // At the start of a syscall we emit traceGoSysCall to capture the stack trace.
3736 // If the syscall does not block, that is it, we do not emit any other events.
3737 // If the syscall blocks (that is, P is retaken), retaker emits traceGoSysBlock;
3738 // when syscall returns we emit traceGoSysExit and when the goroutine starts running
3739 // (potentially instantly, if exitsyscallfast returns true) we emit traceGoStart.
3740 // To ensure that traceGoSysExit is emitted strictly after traceGoSysBlock,
3741 // we remember current value of syscalltick in m (gp.m.syscalltick = gp.m.p.ptr().syscalltick),
3742 // whoever emits traceGoSysBlock increments p.syscalltick afterwards;
3743 // and we wait for the increment before emitting traceGoSysExit.
3744 // Note that the increment is done even if tracing is not enabled,
3745 // because tracing can be enabled in the middle of syscall. We don't want the wait to hang.
3748 func reentersyscall(pc, sp uintptr) {
3751 // Disable preemption because during this function g is in Gsyscall status,
3752 // but can have inconsistent g->sched, do not let GC observe it.
3755 // Entersyscall must not call any function that might split/grow the stack.
3756 // (See details in comment above.)
3757 // Catch calls that might, by replacing the stack guard with something that
3758 // will trip any stack check and leaving a flag to tell newstack to die.
3759 gp.stackguard0 = stackPreempt
3760 gp.throwsplit = true
3762 // Leave SP around for GC and traceback.
3766 casgstatus(gp, _Grunning, _Gsyscall)
3767 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
3768 systemstack(func() {
3769 print("entersyscall inconsistent ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
3770 throw("entersyscall")
3775 systemstack(traceGoSysCall)
3776 // systemstack itself clobbers g.sched.{pc,sp} and we might
3777 // need them later when the G is genuinely blocked in a
3782 if sched.sysmonwait.Load() {
3783 systemstack(entersyscall_sysmon)
3787 if gp.m.p.ptr().runSafePointFn != 0 {
3788 // runSafePointFn may stack split if run on this stack
3789 systemstack(runSafePointFn)
3793 gp.m.syscalltick = gp.m.p.ptr().syscalltick
3794 gp.sysblocktraced = true
3799 atomic.Store(&pp.status, _Psyscall)
3800 if sched.gcwaiting.Load() {
3801 systemstack(entersyscall_gcwait)
3808 // Standard syscall entry used by the go syscall library and normal cgo calls.
3810 // This is exported via linkname to assembly in the syscall package and x/sys.
3813 //go:linkname entersyscall
3814 func entersyscall() {
3815 reentersyscall(getcallerpc(), getcallersp())
3818 func entersyscall_sysmon() {
3820 if sched.sysmonwait.Load() {
3821 sched.sysmonwait.Store(false)
3822 notewakeup(&sched.sysmonnote)
3827 func entersyscall_gcwait() {
3829 pp := gp.m.oldp.ptr()
3832 if sched.stopwait > 0 && atomic.Cas(&pp.status, _Psyscall, _Pgcstop) {
3838 if sched.stopwait--; sched.stopwait == 0 {
3839 notewakeup(&sched.stopnote)
3845 // The same as entersyscall(), but with a hint that the syscall is blocking.
3848 func entersyscallblock() {
3851 gp.m.locks++ // see comment in entersyscall
3852 gp.throwsplit = true
3853 gp.stackguard0 = stackPreempt // see comment in entersyscall
3854 gp.m.syscalltick = gp.m.p.ptr().syscalltick
3855 gp.sysblocktraced = true
3856 gp.m.p.ptr().syscalltick++
3858 // Leave SP around for GC and traceback.
3862 gp.syscallsp = gp.sched.sp
3863 gp.syscallpc = gp.sched.pc
3864 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
3868 systemstack(func() {
3869 print("entersyscallblock inconsistent ", hex(sp1), " ", hex(sp2), " ", hex(sp3), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
3870 throw("entersyscallblock")
3873 casgstatus(gp, _Grunning, _Gsyscall)
3874 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
3875 systemstack(func() {
3876 print("entersyscallblock inconsistent ", hex(sp), " ", hex(gp.sched.sp), " ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
3877 throw("entersyscallblock")
3881 systemstack(entersyscallblock_handoff)
3883 // Resave for traceback during blocked call.
3884 save(getcallerpc(), getcallersp())
3889 func entersyscallblock_handoff() {
3892 traceGoSysBlock(getg().m.p.ptr())
3894 handoffp(releasep())
3897 // The goroutine g exited its system call.
3898 // Arrange for it to run on a cpu again.
3899 // This is called only from the go syscall library, not
3900 // from the low-level system calls used by the runtime.
3902 // Write barriers are not allowed because our P may have been stolen.
3904 // This is exported via linkname to assembly in the syscall package.
3907 //go:nowritebarrierrec
3908 //go:linkname exitsyscall
3909 func exitsyscall() {
3912 gp.m.locks++ // see comment in entersyscall
3913 if getcallersp() > gp.syscallsp {
3914 throw("exitsyscall: syscall frame is no longer valid")
3918 oldp := gp.m.oldp.ptr()
3920 if exitsyscallfast(oldp) {
3921 // When exitsyscallfast returns success, we have a P so can now use
3923 if goroutineProfile.active {
3924 // Make sure that gp has had its stack written out to the goroutine
3925 // profile, exactly as it was when the goroutine profiler first
3926 // stopped the world.
3927 systemstack(func() {
3928 tryRecordGoroutineProfileWB(gp)
3932 if oldp != gp.m.p.ptr() || gp.m.syscalltick != gp.m.p.ptr().syscalltick {
3933 systemstack(traceGoStart)
3936 // There's a cpu for us, so we can run.
3937 gp.m.p.ptr().syscalltick++
3938 // We need to cas the status and scan before resuming...
3939 casgstatus(gp, _Gsyscall, _Grunning)
3941 // Garbage collector isn't running (since we are),
3942 // so okay to clear syscallsp.
3946 // restore the preemption request in case we've cleared it in newstack
3947 gp.stackguard0 = stackPreempt
3949 // otherwise restore the real _StackGuard, we've spoiled it in entersyscall/entersyscallblock
3950 gp.stackguard0 = gp.stack.lo + _StackGuard
3952 gp.throwsplit = false
3954 if sched.disable.user && !schedEnabled(gp) {
3955 // Scheduling of this goroutine is disabled.
3964 // Wait till traceGoSysBlock event is emitted.
3965 // This ensures consistency of the trace (the goroutine is started after it is blocked).
3966 for oldp != nil && oldp.syscalltick == gp.m.syscalltick {
3969 // We can't trace syscall exit right now because we don't have a P.
3970 // Tracing code can invoke write barriers that cannot run without a P.
3971 // So instead we remember the syscall exit time and emit the event
3972 // in execute when we have a P.
3973 gp.sysexitticks = cputicks()
3978 // Call the scheduler.
3981 // Scheduler returned, so we're allowed to run now.
3982 // Delete the syscallsp information that we left for
3983 // the garbage collector during the system call.
3984 // Must wait until now because until gosched returns
3985 // we don't know for sure that the garbage collector
3988 gp.m.p.ptr().syscalltick++
3989 gp.throwsplit = false
3993 func exitsyscallfast(oldp *p) bool {
3996 // Freezetheworld sets stopwait but does not retake P's.
3997 if sched.stopwait == freezeStopWait {
4001 // Try to re-acquire the last P.
4002 if oldp != nil && oldp.status == _Psyscall && atomic.Cas(&oldp.status, _Psyscall, _Pidle) {
4003 // There's a cpu for us, so we can run.
4005 exitsyscallfast_reacquired()
4009 // Try to get any other idle P.
4010 if sched.pidle != 0 {
4012 systemstack(func() {
4013 ok = exitsyscallfast_pidle()
4014 if ok && trace.enabled {
4016 // Wait till traceGoSysBlock event is emitted.
4017 // This ensures consistency of the trace (the goroutine is started after it is blocked).
4018 for oldp.syscalltick == gp.m.syscalltick {
4032 // exitsyscallfast_reacquired is the exitsyscall path on which this G
4033 // has successfully reacquired the P it was running on before the
4037 func exitsyscallfast_reacquired() {
4039 if gp.m.syscalltick != gp.m.p.ptr().syscalltick {
4041 // The p was retaken and then enter into syscall again (since gp.m.syscalltick has changed).
4042 // traceGoSysBlock for this syscall was already emitted,
4043 // but here we effectively retake the p from the new syscall running on the same p.
4044 systemstack(func() {
4045 // Denote blocking of the new syscall.
4046 traceGoSysBlock(gp.m.p.ptr())
4047 // Denote completion of the current syscall.
4051 gp.m.p.ptr().syscalltick++
4055 func exitsyscallfast_pidle() bool {
4057 pp, _ := pidleget(0)
4058 if pp != nil && sched.sysmonwait.Load() {
4059 sched.sysmonwait.Store(false)
4060 notewakeup(&sched.sysmonnote)
4070 // exitsyscall slow path on g0.
4071 // Failed to acquire P, enqueue gp as runnable.
4073 // Called via mcall, so gp is the calling g from this M.
4075 //go:nowritebarrierrec
4076 func exitsyscall0(gp *g) {
4077 casgstatus(gp, _Gsyscall, _Grunnable)
4081 if schedEnabled(gp) {
4088 // Below, we stoplockedm if gp is locked. globrunqput releases
4089 // ownership of gp, so we must check if gp is locked prior to
4090 // committing the release by unlocking sched.lock, otherwise we
4091 // could race with another M transitioning gp from unlocked to
4093 locked = gp.lockedm != 0
4094 } else if sched.sysmonwait.Load() {
4095 sched.sysmonwait.Store(false)
4096 notewakeup(&sched.sysmonnote)
4101 execute(gp, false) // Never returns.
4104 // Wait until another thread schedules gp and so m again.
4106 // N.B. lockedm must be this M, as this g was running on this M
4107 // before entersyscall.
4109 execute(gp, false) // Never returns.
4112 schedule() // Never returns.
4115 // Called from syscall package before fork.
4117 //go:linkname syscall_runtime_BeforeFork syscall.runtime_BeforeFork
4119 func syscall_runtime_BeforeFork() {
4122 // Block signals during a fork, so that the child does not run
4123 // a signal handler before exec if a signal is sent to the process
4124 // group. See issue #18600.
4126 sigsave(&gp.m.sigmask)
4129 // This function is called before fork in syscall package.
4130 // Code between fork and exec must not allocate memory nor even try to grow stack.
4131 // Here we spoil g->_StackGuard to reliably detect any attempts to grow stack.
4132 // runtime_AfterFork will undo this in parent process, but not in child.
4133 gp.stackguard0 = stackFork
4136 // Called from syscall package after fork in parent.
4138 //go:linkname syscall_runtime_AfterFork syscall.runtime_AfterFork
4140 func syscall_runtime_AfterFork() {
4143 // See the comments in beforefork.
4144 gp.stackguard0 = gp.stack.lo + _StackGuard
4146 msigrestore(gp.m.sigmask)
4151 // inForkedChild is true while manipulating signals in the child process.
4152 // This is used to avoid calling libc functions in case we are using vfork.
4153 var inForkedChild bool
4155 // Called from syscall package after fork in child.
4156 // It resets non-sigignored signals to the default handler, and
4157 // restores the signal mask in preparation for the exec.
4159 // Because this might be called during a vfork, and therefore may be
4160 // temporarily sharing address space with the parent process, this must
4161 // not change any global variables or calling into C code that may do so.
4163 //go:linkname syscall_runtime_AfterForkInChild syscall.runtime_AfterForkInChild
4165 //go:nowritebarrierrec
4166 func syscall_runtime_AfterForkInChild() {
4167 // It's OK to change the global variable inForkedChild here
4168 // because we are going to change it back. There is no race here,
4169 // because if we are sharing address space with the parent process,
4170 // then the parent process can not be running concurrently.
4171 inForkedChild = true
4173 clearSignalHandlers()
4175 // When we are the child we are the only thread running,
4176 // so we know that nothing else has changed gp.m.sigmask.
4177 msigrestore(getg().m.sigmask)
4179 inForkedChild = false
4182 // pendingPreemptSignals is the number of preemption signals
4183 // that have been sent but not received. This is only used on Darwin.
4185 var pendingPreemptSignals atomic.Int32
4187 // Called from syscall package before Exec.
4189 //go:linkname syscall_runtime_BeforeExec syscall.runtime_BeforeExec
4190 func syscall_runtime_BeforeExec() {
4191 // Prevent thread creation during exec.
4194 // On Darwin, wait for all pending preemption signals to
4195 // be received. See issue #41702.
4196 if GOOS == "darwin" || GOOS == "ios" {
4197 for pendingPreemptSignals.Load() > 0 {
4203 // Called from syscall package after Exec.
4205 //go:linkname syscall_runtime_AfterExec syscall.runtime_AfterExec
4206 func syscall_runtime_AfterExec() {
4210 // Allocate a new g, with a stack big enough for stacksize bytes.
4211 func malg(stacksize int32) *g {
4214 stacksize = round2(_StackSystem + stacksize)
4215 systemstack(func() {
4216 newg.stack = stackalloc(uint32(stacksize))
4218 newg.stackguard0 = newg.stack.lo + _StackGuard
4219 newg.stackguard1 = ^uintptr(0)
4220 // Clear the bottom word of the stack. We record g
4221 // there on gsignal stack during VDSO on ARM and ARM64.
4222 *(*uintptr)(unsafe.Pointer(newg.stack.lo)) = 0
4227 // Create a new g running fn.
4228 // Put it on the queue of g's waiting to run.
4229 // The compiler turns a go statement into a call to this.
4230 func newproc(fn *funcval) {
4233 systemstack(func() {
4234 newg := newproc1(fn, gp, pc)
4236 pp := getg().m.p.ptr()
4237 runqput(pp, newg, true)
4245 // Create a new g in state _Grunnable, starting at fn. callerpc is the
4246 // address of the go statement that created this. The caller is responsible
4247 // for adding the new g to the scheduler.
4248 func newproc1(fn *funcval, callergp *g, callerpc uintptr) *g {
4250 fatal("go of nil func value")
4253 mp := acquirem() // disable preemption because we hold M and P in local vars.
4257 newg = malg(_StackMin)
4258 casgstatus(newg, _Gidle, _Gdead)
4259 allgadd(newg) // publishes with a g->status of Gdead so GC scanner doesn't look at uninitialized stack.
4261 if newg.stack.hi == 0 {
4262 throw("newproc1: newg missing stack")
4265 if readgstatus(newg) != _Gdead {
4266 throw("newproc1: new g is not Gdead")
4269 totalSize := uintptr(4*goarch.PtrSize + sys.MinFrameSize) // extra space in case of reads slightly beyond frame
4270 totalSize = alignUp(totalSize, sys.StackAlign)
4271 sp := newg.stack.hi - totalSize
4275 *(*uintptr)(unsafe.Pointer(sp)) = 0
4277 spArg += sys.MinFrameSize
4280 memclrNoHeapPointers(unsafe.Pointer(&newg.sched), unsafe.Sizeof(newg.sched))
4283 newg.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum // +PCQuantum so that previous instruction is in same function
4284 newg.sched.g = guintptr(unsafe.Pointer(newg))
4285 gostartcallfn(&newg.sched, fn)
4286 newg.parentGoid = callergp.goid
4287 newg.gopc = callerpc
4288 newg.ancestors = saveAncestors(callergp)
4289 newg.startpc = fn.fn
4290 if isSystemGoroutine(newg, false) {
4293 // Only user goroutines inherit pprof labels.
4295 newg.labels = mp.curg.labels
4297 if goroutineProfile.active {
4298 // A concurrent goroutine profile is running. It should include
4299 // exactly the set of goroutines that were alive when the goroutine
4300 // profiler first stopped the world. That does not include newg, so
4301 // mark it as not needing a profile before transitioning it from
4303 newg.goroutineProfiled.Store(goroutineProfileSatisfied)
4306 // Track initial transition?
4307 newg.trackingSeq = uint8(fastrand())
4308 if newg.trackingSeq%gTrackingPeriod == 0 {
4309 newg.tracking = true
4311 casgstatus(newg, _Gdead, _Grunnable)
4312 gcController.addScannableStack(pp, int64(newg.stack.hi-newg.stack.lo))
4314 if pp.goidcache == pp.goidcacheend {
4315 // Sched.goidgen is the last allocated id,
4316 // this batch must be [sched.goidgen+1, sched.goidgen+GoidCacheBatch].
4317 // At startup sched.goidgen=0, so main goroutine receives goid=1.
4318 pp.goidcache = sched.goidgen.Add(_GoidCacheBatch)
4319 pp.goidcache -= _GoidCacheBatch - 1
4320 pp.goidcacheend = pp.goidcache + _GoidCacheBatch
4322 newg.goid = pp.goidcache
4325 newg.racectx = racegostart(callerpc)
4326 if newg.labels != nil {
4327 // See note in proflabel.go on labelSync's role in synchronizing
4328 // with the reads in the signal handler.
4329 racereleasemergeg(newg, unsafe.Pointer(&labelSync))
4333 traceGoCreate(newg, newg.startpc)
4340 // saveAncestors copies previous ancestors of the given caller g and
4341 // includes info for the current caller into a new set of tracebacks for
4342 // a g being created.
4343 func saveAncestors(callergp *g) *[]ancestorInfo {
4344 // Copy all prior info, except for the root goroutine (goid 0).
4345 if debug.tracebackancestors <= 0 || callergp.goid == 0 {
4348 var callerAncestors []ancestorInfo
4349 if callergp.ancestors != nil {
4350 callerAncestors = *callergp.ancestors
4352 n := int32(len(callerAncestors)) + 1
4353 if n > debug.tracebackancestors {
4354 n = debug.tracebackancestors
4356 ancestors := make([]ancestorInfo, n)
4357 copy(ancestors[1:], callerAncestors)
4359 var pcs [tracebackInnerFrames]uintptr
4360 npcs := gcallers(callergp, 0, pcs[:])
4361 ipcs := make([]uintptr, npcs)
4363 ancestors[0] = ancestorInfo{
4365 goid: callergp.goid,
4366 gopc: callergp.gopc,
4369 ancestorsp := new([]ancestorInfo)
4370 *ancestorsp = ancestors
4374 // Put on gfree list.
4375 // If local list is too long, transfer a batch to the global list.
4376 func gfput(pp *p, gp *g) {
4377 if readgstatus(gp) != _Gdead {
4378 throw("gfput: bad status (not Gdead)")
4381 stksize := gp.stack.hi - gp.stack.lo
4383 if stksize != uintptr(startingStackSize) {
4384 // non-standard stack size - free it.
4393 if pp.gFree.n >= 64 {
4399 for pp.gFree.n >= 32 {
4400 gp := pp.gFree.pop()
4402 if gp.stack.lo == 0 {
4409 lock(&sched.gFree.lock)
4410 sched.gFree.noStack.pushAll(noStackQ)
4411 sched.gFree.stack.pushAll(stackQ)
4412 sched.gFree.n += inc
4413 unlock(&sched.gFree.lock)
4417 // Get from gfree list.
4418 // If local list is empty, grab a batch from global list.
4419 func gfget(pp *p) *g {
4421 if pp.gFree.empty() && (!sched.gFree.stack.empty() || !sched.gFree.noStack.empty()) {
4422 lock(&sched.gFree.lock)
4423 // Move a batch of free Gs to the P.
4424 for pp.gFree.n < 32 {
4425 // Prefer Gs with stacks.
4426 gp := sched.gFree.stack.pop()
4428 gp = sched.gFree.noStack.pop()
4437 unlock(&sched.gFree.lock)
4440 gp := pp.gFree.pop()
4445 if gp.stack.lo != 0 && gp.stack.hi-gp.stack.lo != uintptr(startingStackSize) {
4446 // Deallocate old stack. We kept it in gfput because it was the
4447 // right size when the goroutine was put on the free list, but
4448 // the right size has changed since then.
4449 systemstack(func() {
4456 if gp.stack.lo == 0 {
4457 // Stack was deallocated in gfput or just above. Allocate a new one.
4458 systemstack(func() {
4459 gp.stack = stackalloc(startingStackSize)
4461 gp.stackguard0 = gp.stack.lo + _StackGuard
4464 racemalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4467 msanmalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4470 asanunpoison(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4476 // Purge all cached G's from gfree list to the global list.
4477 func gfpurge(pp *p) {
4483 for !pp.gFree.empty() {
4484 gp := pp.gFree.pop()
4486 if gp.stack.lo == 0 {
4493 lock(&sched.gFree.lock)
4494 sched.gFree.noStack.pushAll(noStackQ)
4495 sched.gFree.stack.pushAll(stackQ)
4496 sched.gFree.n += inc
4497 unlock(&sched.gFree.lock)
4500 // Breakpoint executes a breakpoint trap.
4505 // dolockOSThread is called by LockOSThread and lockOSThread below
4506 // after they modify m.locked. Do not allow preemption during this call,
4507 // or else the m might be different in this function than in the caller.
4510 func dolockOSThread() {
4511 if GOARCH == "wasm" {
4512 return // no threads on wasm yet
4515 gp.m.lockedg.set(gp)
4516 gp.lockedm.set(gp.m)
4519 // LockOSThread wires the calling goroutine to its current operating system thread.
4520 // The calling goroutine will always execute in that thread,
4521 // and no other goroutine will execute in it,
4522 // until the calling goroutine has made as many calls to
4523 // UnlockOSThread as to LockOSThread.
4524 // If the calling goroutine exits without unlocking the thread,
4525 // the thread will be terminated.
4527 // All init functions are run on the startup thread. Calling LockOSThread
4528 // from an init function will cause the main function to be invoked on
4531 // A goroutine should call LockOSThread before calling OS services or
4532 // non-Go library functions that depend on per-thread state.
4535 func LockOSThread() {
4536 if atomic.Load(&newmHandoff.haveTemplateThread) == 0 && GOOS != "plan9" {
4537 // If we need to start a new thread from the locked
4538 // thread, we need the template thread. Start it now
4539 // while we're in a known-good state.
4540 startTemplateThread()
4544 if gp.m.lockedExt == 0 {
4546 panic("LockOSThread nesting overflow")
4552 func lockOSThread() {
4553 getg().m.lockedInt++
4557 // dounlockOSThread is called by UnlockOSThread and unlockOSThread below
4558 // after they update m->locked. Do not allow preemption during this call,
4559 // or else the m might be in different in this function than in the caller.
4562 func dounlockOSThread() {
4563 if GOARCH == "wasm" {
4564 return // no threads on wasm yet
4567 if gp.m.lockedInt != 0 || gp.m.lockedExt != 0 {
4574 // UnlockOSThread undoes an earlier call to LockOSThread.
4575 // If this drops the number of active LockOSThread calls on the
4576 // calling goroutine to zero, it unwires the calling goroutine from
4577 // its fixed operating system thread.
4578 // If there are no active LockOSThread calls, this is a no-op.
4580 // Before calling UnlockOSThread, the caller must ensure that the OS
4581 // thread is suitable for running other goroutines. If the caller made
4582 // any permanent changes to the state of the thread that would affect
4583 // other goroutines, it should not call this function and thus leave
4584 // the goroutine locked to the OS thread until the goroutine (and
4585 // hence the thread) exits.
4588 func UnlockOSThread() {
4590 if gp.m.lockedExt == 0 {
4598 func unlockOSThread() {
4600 if gp.m.lockedInt == 0 {
4601 systemstack(badunlockosthread)
4607 func badunlockosthread() {
4608 throw("runtime: internal error: misuse of lockOSThread/unlockOSThread")
4611 func gcount() int32 {
4612 n := int32(atomic.Loaduintptr(&allglen)) - sched.gFree.n - sched.ngsys.Load()
4613 for _, pp := range allp {
4617 // All these variables can be changed concurrently, so the result can be inconsistent.
4618 // But at least the current goroutine is running.
4625 func mcount() int32 {
4626 return int32(sched.mnext - sched.nmfreed)
4630 signalLock atomic.Uint32
4632 // Must hold signalLock to write. Reads may be lock-free, but
4633 // signalLock should be taken to synchronize with changes.
4637 func _System() { _System() }
4638 func _ExternalCode() { _ExternalCode() }
4639 func _LostExternalCode() { _LostExternalCode() }
4640 func _GC() { _GC() }
4641 func _LostSIGPROFDuringAtomic64() { _LostSIGPROFDuringAtomic64() }
4642 func _VDSO() { _VDSO() }
4644 // Called if we receive a SIGPROF signal.
4645 // Called by the signal handler, may run during STW.
4647 //go:nowritebarrierrec
4648 func sigprof(pc, sp, lr uintptr, gp *g, mp *m) {
4649 if prof.hz.Load() == 0 {
4653 // If mp.profilehz is 0, then profiling is not enabled for this thread.
4654 // We must check this to avoid a deadlock between setcpuprofilerate
4655 // and the call to cpuprof.add, below.
4656 if mp != nil && mp.profilehz == 0 {
4660 // On mips{,le}/arm, 64bit atomics are emulated with spinlocks, in
4661 // runtime/internal/atomic. If SIGPROF arrives while the program is inside
4662 // the critical section, it creates a deadlock (when writing the sample).
4663 // As a workaround, create a counter of SIGPROFs while in critical section
4664 // to store the count, and pass it to sigprof.add() later when SIGPROF is
4665 // received from somewhere else (with _LostSIGPROFDuringAtomic64 as pc).
4666 if GOARCH == "mips" || GOARCH == "mipsle" || GOARCH == "arm" {
4667 if f := findfunc(pc); f.valid() {
4668 if hasPrefix(funcname(f), "runtime/internal/atomic") {
4669 cpuprof.lostAtomic++
4673 if GOARCH == "arm" && goarm < 7 && GOOS == "linux" && pc&0xffff0000 == 0xffff0000 {
4674 // runtime/internal/atomic functions call into kernel
4675 // helpers on arm < 7. See
4676 // runtime/internal/atomic/sys_linux_arm.s.
4677 cpuprof.lostAtomic++
4682 // Profiling runs concurrently with GC, so it must not allocate.
4683 // Set a trap in case the code does allocate.
4684 // Note that on windows, one thread takes profiles of all the
4685 // other threads, so mp is usually not getg().m.
4686 // In fact mp may not even be stopped.
4687 // See golang.org/issue/17165.
4688 getg().m.mallocing++
4691 var stk [maxCPUProfStack]uintptr
4693 if mp.ncgo > 0 && mp.curg != nil && mp.curg.syscallpc != 0 && mp.curg.syscallsp != 0 {
4695 // Check cgoCallersUse to make sure that we are not
4696 // interrupting other code that is fiddling with
4697 // cgoCallers. We are running in a signal handler
4698 // with all signals blocked, so we don't have to worry
4699 // about any other code interrupting us.
4700 if mp.cgoCallersUse.Load() == 0 && mp.cgoCallers != nil && mp.cgoCallers[0] != 0 {
4701 for cgoOff < len(mp.cgoCallers) && mp.cgoCallers[cgoOff] != 0 {
4704 n += copy(stk[:], mp.cgoCallers[:cgoOff])
4705 mp.cgoCallers[0] = 0
4708 // Collect Go stack that leads to the cgo call.
4709 u.initAt(mp.curg.syscallpc, mp.curg.syscallsp, 0, mp.curg, unwindSilentErrors)
4710 } else if usesLibcall() && mp.libcallg != 0 && mp.libcallpc != 0 && mp.libcallsp != 0 {
4711 // Libcall, i.e. runtime syscall on windows.
4712 // Collect Go stack that leads to the call.
4713 u.initAt(mp.libcallpc, mp.libcallsp, 0, mp.libcallg.ptr(), unwindSilentErrors)
4714 } else if mp != nil && mp.vdsoSP != 0 {
4715 // VDSO call, e.g. nanotime1 on Linux.
4716 // Collect Go stack that leads to the call.
4717 u.initAt(mp.vdsoPC, mp.vdsoSP, 0, gp, unwindSilentErrors|unwindJumpStack)
4719 u.initAt(pc, sp, lr, gp, unwindSilentErrors|unwindTrap|unwindJumpStack)
4721 n += tracebackPCs(&u, 0, stk[n:])
4724 // Normal traceback is impossible or has failed.
4725 // Account it against abstract "System" or "GC".
4728 pc = abi.FuncPCABIInternal(_VDSO) + sys.PCQuantum
4729 } else if pc > firstmoduledata.etext {
4730 // "ExternalCode" is better than "etext".
4731 pc = abi.FuncPCABIInternal(_ExternalCode) + sys.PCQuantum
4734 if mp.preemptoff != "" {
4735 stk[1] = abi.FuncPCABIInternal(_GC) + sys.PCQuantum
4737 stk[1] = abi.FuncPCABIInternal(_System) + sys.PCQuantum
4741 if prof.hz.Load() != 0 {
4742 // Note: it can happen on Windows that we interrupted a system thread
4743 // with no g, so gp could nil. The other nil checks are done out of
4744 // caution, but not expected to be nil in practice.
4745 var tagPtr *unsafe.Pointer
4746 if gp != nil && gp.m != nil && gp.m.curg != nil {
4747 tagPtr = &gp.m.curg.labels
4749 cpuprof.add(tagPtr, stk[:n])
4753 if gp != nil && gp.m != nil {
4754 if gp.m.curg != nil {
4759 traceCPUSample(gprof, pp, stk[:n])
4761 getg().m.mallocing--
4764 // setcpuprofilerate sets the CPU profiling rate to hz times per second.
4765 // If hz <= 0, setcpuprofilerate turns off CPU profiling.
4766 func setcpuprofilerate(hz int32) {
4767 // Force sane arguments.
4772 // Disable preemption, otherwise we can be rescheduled to another thread
4773 // that has profiling enabled.
4777 // Stop profiler on this thread so that it is safe to lock prof.
4778 // if a profiling signal came in while we had prof locked,
4779 // it would deadlock.
4780 setThreadCPUProfiler(0)
4782 for !prof.signalLock.CompareAndSwap(0, 1) {
4785 if prof.hz.Load() != hz {
4786 setProcessCPUProfiler(hz)
4789 prof.signalLock.Store(0)
4792 sched.profilehz = hz
4796 setThreadCPUProfiler(hz)
4802 // init initializes pp, which may be a freshly allocated p or a
4803 // previously destroyed p, and transitions it to status _Pgcstop.
4804 func (pp *p) init(id int32) {
4806 pp.status = _Pgcstop
4807 pp.sudogcache = pp.sudogbuf[:0]
4808 pp.deferpool = pp.deferpoolbuf[:0]
4810 if pp.mcache == nil {
4813 throw("missing mcache?")
4815 // Use the bootstrap mcache0. Only one P will get
4816 // mcache0: the one with ID 0.
4819 pp.mcache = allocmcache()
4822 if raceenabled && pp.raceprocctx == 0 {
4824 pp.raceprocctx = raceprocctx0
4825 raceprocctx0 = 0 // bootstrap
4827 pp.raceprocctx = raceproccreate()
4830 lockInit(&pp.timersLock, lockRankTimers)
4832 // This P may get timers when it starts running. Set the mask here
4833 // since the P may not go through pidleget (notably P 0 on startup).
4835 // Similarly, we may not go through pidleget before this P starts
4836 // running if it is P 0 on startup.
4840 // destroy releases all of the resources associated with pp and
4841 // transitions it to status _Pdead.
4843 // sched.lock must be held and the world must be stopped.
4844 func (pp *p) destroy() {
4845 assertLockHeld(&sched.lock)
4846 assertWorldStopped()
4848 // Move all runnable goroutines to the global queue
4849 for pp.runqhead != pp.runqtail {
4850 // Pop from tail of local queue
4852 gp := pp.runq[pp.runqtail%uint32(len(pp.runq))].ptr()
4853 // Push onto head of global queue
4856 if pp.runnext != 0 {
4857 globrunqputhead(pp.runnext.ptr())
4860 if len(pp.timers) > 0 {
4861 plocal := getg().m.p.ptr()
4862 // The world is stopped, but we acquire timersLock to
4863 // protect against sysmon calling timeSleepUntil.
4864 // This is the only case where we hold the timersLock of
4865 // more than one P, so there are no deadlock concerns.
4866 lock(&plocal.timersLock)
4867 lock(&pp.timersLock)
4868 moveTimers(plocal, pp.timers)
4870 pp.numTimers.Store(0)
4871 pp.deletedTimers.Store(0)
4872 pp.timer0When.Store(0)
4873 unlock(&pp.timersLock)
4874 unlock(&plocal.timersLock)
4876 // Flush p's write barrier buffer.
4877 if gcphase != _GCoff {
4881 for i := range pp.sudogbuf {
4882 pp.sudogbuf[i] = nil
4884 pp.sudogcache = pp.sudogbuf[:0]
4885 for j := range pp.deferpoolbuf {
4886 pp.deferpoolbuf[j] = nil
4888 pp.deferpool = pp.deferpoolbuf[:0]
4889 systemstack(func() {
4890 for i := 0; i < pp.mspancache.len; i++ {
4891 // Safe to call since the world is stopped.
4892 mheap_.spanalloc.free(unsafe.Pointer(pp.mspancache.buf[i]))
4894 pp.mspancache.len = 0
4896 pp.pcache.flush(&mheap_.pages)
4897 unlock(&mheap_.lock)
4899 freemcache(pp.mcache)
4904 if pp.timerRaceCtx != 0 {
4905 // The race detector code uses a callback to fetch
4906 // the proc context, so arrange for that callback
4907 // to see the right thing.
4908 // This hack only works because we are the only
4914 racectxend(pp.timerRaceCtx)
4919 raceprocdestroy(pp.raceprocctx)
4926 // Change number of processors.
4928 // sched.lock must be held, and the world must be stopped.
4930 // gcworkbufs must not be being modified by either the GC or the write barrier
4931 // code, so the GC must not be running if the number of Ps actually changes.
4933 // Returns list of Ps with local work, they need to be scheduled by the caller.
4934 func procresize(nprocs int32) *p {
4935 assertLockHeld(&sched.lock)
4936 assertWorldStopped()
4939 if old < 0 || nprocs <= 0 {
4940 throw("procresize: invalid arg")
4943 traceGomaxprocs(nprocs)
4946 // update statistics
4948 if sched.procresizetime != 0 {
4949 sched.totaltime += int64(old) * (now - sched.procresizetime)
4951 sched.procresizetime = now
4953 maskWords := (nprocs + 31) / 32
4955 // Grow allp if necessary.
4956 if nprocs > int32(len(allp)) {
4957 // Synchronize with retake, which could be running
4958 // concurrently since it doesn't run on a P.
4960 if nprocs <= int32(cap(allp)) {
4961 allp = allp[:nprocs]
4963 nallp := make([]*p, nprocs)
4964 // Copy everything up to allp's cap so we
4965 // never lose old allocated Ps.
4966 copy(nallp, allp[:cap(allp)])
4970 if maskWords <= int32(cap(idlepMask)) {
4971 idlepMask = idlepMask[:maskWords]
4972 timerpMask = timerpMask[:maskWords]
4974 nidlepMask := make([]uint32, maskWords)
4975 // No need to copy beyond len, old Ps are irrelevant.
4976 copy(nidlepMask, idlepMask)
4977 idlepMask = nidlepMask
4979 ntimerpMask := make([]uint32, maskWords)
4980 copy(ntimerpMask, timerpMask)
4981 timerpMask = ntimerpMask
4986 // initialize new P's
4987 for i := old; i < nprocs; i++ {
4993 atomicstorep(unsafe.Pointer(&allp[i]), unsafe.Pointer(pp))
4997 if gp.m.p != 0 && gp.m.p.ptr().id < nprocs {
4998 // continue to use the current P
4999 gp.m.p.ptr().status = _Prunning
5000 gp.m.p.ptr().mcache.prepareForSweep()
5002 // release the current P and acquire allp[0].
5004 // We must do this before destroying our current P
5005 // because p.destroy itself has write barriers, so we
5006 // need to do that from a valid P.
5009 // Pretend that we were descheduled
5010 // and then scheduled again to keep
5013 traceProcStop(gp.m.p.ptr())
5027 // g.m.p is now set, so we no longer need mcache0 for bootstrapping.
5030 // release resources from unused P's
5031 for i := nprocs; i < old; i++ {
5034 // can't free P itself because it can be referenced by an M in syscall
5038 if int32(len(allp)) != nprocs {
5040 allp = allp[:nprocs]
5041 idlepMask = idlepMask[:maskWords]
5042 timerpMask = timerpMask[:maskWords]
5047 for i := nprocs - 1; i >= 0; i-- {
5049 if gp.m.p.ptr() == pp {
5057 pp.link.set(runnablePs)
5061 stealOrder.reset(uint32(nprocs))
5062 var int32p *int32 = &gomaxprocs // make compiler check that gomaxprocs is an int32
5063 atomic.Store((*uint32)(unsafe.Pointer(int32p)), uint32(nprocs))
5065 // Notify the limiter that the amount of procs has changed.
5066 gcCPULimiter.resetCapacity(now, nprocs)
5071 // Associate p and the current m.
5073 // This function is allowed to have write barriers even if the caller
5074 // isn't because it immediately acquires pp.
5076 //go:yeswritebarrierrec
5077 func acquirep(pp *p) {
5078 // Do the part that isn't allowed to have write barriers.
5081 // Have p; write barriers now allowed.
5083 // Perform deferred mcache flush before this P can allocate
5084 // from a potentially stale mcache.
5085 pp.mcache.prepareForSweep()
5092 // wirep is the first step of acquirep, which actually associates the
5093 // current M to pp. This is broken out so we can disallow write
5094 // barriers for this part, since we don't yet have a P.
5096 //go:nowritebarrierrec
5102 throw("wirep: already in go")
5104 if pp.m != 0 || pp.status != _Pidle {
5109 print("wirep: p->m=", pp.m, "(", id, ") p->status=", pp.status, "\n")
5110 throw("wirep: invalid p state")
5114 pp.status = _Prunning
5117 // Disassociate p and the current m.
5118 func releasep() *p {
5122 throw("releasep: invalid arg")
5125 if pp.m.ptr() != gp.m || pp.status != _Prunning {
5126 print("releasep: m=", gp.m, " m->p=", gp.m.p.ptr(), " p->m=", hex(pp.m), " p->status=", pp.status, "\n")
5127 throw("releasep: invalid p state")
5130 traceProcStop(gp.m.p.ptr())
5138 func incidlelocked(v int32) {
5140 sched.nmidlelocked += v
5147 // Check for deadlock situation.
5148 // The check is based on number of running M's, if 0 -> deadlock.
5149 // sched.lock must be held.
5151 assertLockHeld(&sched.lock)
5153 // For -buildmode=c-shared or -buildmode=c-archive it's OK if
5154 // there are no running goroutines. The calling program is
5155 // assumed to be running.
5156 if islibrary || isarchive {
5160 // If we are dying because of a signal caught on an already idle thread,
5161 // freezetheworld will cause all running threads to block.
5162 // And runtime will essentially enter into deadlock state,
5163 // except that there is a thread that will call exit soon.
5164 if panicking.Load() > 0 {
5168 // If we are not running under cgo, but we have an extra M then account
5169 // for it. (It is possible to have an extra M on Windows without cgo to
5170 // accommodate callbacks created by syscall.NewCallback. See issue #6751
5173 if !iscgo && cgoHasExtraM {
5174 mp := lockextra(true)
5175 haveExtraM := extraMCount > 0
5182 run := mcount() - sched.nmidle - sched.nmidlelocked - sched.nmsys
5187 print("runtime: checkdead: nmidle=", sched.nmidle, " nmidlelocked=", sched.nmidlelocked, " mcount=", mcount(), " nmsys=", sched.nmsys, "\n")
5188 throw("checkdead: inconsistent counts")
5192 forEachG(func(gp *g) {
5193 if isSystemGoroutine(gp, false) {
5196 s := readgstatus(gp)
5197 switch s &^ _Gscan {
5204 print("runtime: checkdead: find g ", gp.goid, " in status ", s, "\n")
5205 throw("checkdead: runnable g")
5208 if grunning == 0 { // possible if main goroutine calls runtime·Goexit()
5209 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5210 fatal("no goroutines (main called runtime.Goexit) - deadlock!")
5213 // Maybe jump time forward for playground.
5215 if when := timeSleepUntil(); when < maxWhen {
5218 // Start an M to steal the timer.
5219 pp, _ := pidleget(faketime)
5221 // There should always be a free P since
5222 // nothing is running.
5223 throw("checkdead: no p for timer")
5227 // There should always be a free M since
5228 // nothing is running.
5229 throw("checkdead: no m for timer")
5231 // M must be spinning to steal. We set this to be
5232 // explicit, but since this is the only M it would
5233 // become spinning on its own anyways.
5234 sched.nmspinning.Add(1)
5237 notewakeup(&mp.park)
5242 // There are no goroutines running, so we can look at the P's.
5243 for _, pp := range allp {
5244 if len(pp.timers) > 0 {
5249 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5250 fatal("all goroutines are asleep - deadlock!")
5253 // forcegcperiod is the maximum time in nanoseconds between garbage
5254 // collections. If we go this long without a garbage collection, one
5255 // is forced to run.
5257 // This is a variable for testing purposes. It normally doesn't change.
5258 var forcegcperiod int64 = 2 * 60 * 1e9
5260 // needSysmonWorkaround is true if the workaround for
5261 // golang.org/issue/42515 is needed on NetBSD.
5262 var needSysmonWorkaround bool = false
5264 // Always runs without a P, so write barriers are not allowed.
5266 //go:nowritebarrierrec
5273 lasttrace := int64(0)
5274 idle := 0 // how many cycles in succession we had not wokeup somebody
5278 if idle == 0 { // start with 20us sleep...
5280 } else if idle > 50 { // start doubling the sleep after 1ms...
5283 if delay > 10*1000 { // up to 10ms
5288 // sysmon should not enter deep sleep if schedtrace is enabled so that
5289 // it can print that information at the right time.
5291 // It should also not enter deep sleep if there are any active P's so
5292 // that it can retake P's from syscalls, preempt long running G's, and
5293 // poll the network if all P's are busy for long stretches.
5295 // It should wakeup from deep sleep if any P's become active either due
5296 // to exiting a syscall or waking up due to a timer expiring so that it
5297 // can resume performing those duties. If it wakes from a syscall it
5298 // resets idle and delay as a bet that since it had retaken a P from a
5299 // syscall before, it may need to do it again shortly after the
5300 // application starts work again. It does not reset idle when waking
5301 // from a timer to avoid adding system load to applications that spend
5302 // most of their time sleeping.
5304 if debug.schedtrace <= 0 && (sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs) {
5306 if sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs {
5307 syscallWake := false
5308 next := timeSleepUntil()
5310 sched.sysmonwait.Store(true)
5312 // Make wake-up period small enough
5313 // for the sampling to be correct.
5314 sleep := forcegcperiod / 2
5315 if next-now < sleep {
5318 shouldRelax := sleep >= osRelaxMinNS
5322 syscallWake = notetsleep(&sched.sysmonnote, sleep)
5327 sched.sysmonwait.Store(false)
5328 noteclear(&sched.sysmonnote)
5338 lock(&sched.sysmonlock)
5339 // Update now in case we blocked on sysmonnote or spent a long time
5340 // blocked on schedlock or sysmonlock above.
5343 // trigger libc interceptors if needed
5344 if *cgo_yield != nil {
5345 asmcgocall(*cgo_yield, nil)
5347 // poll network if not polled for more than 10ms
5348 lastpoll := sched.lastpoll.Load()
5349 if netpollinited() && lastpoll != 0 && lastpoll+10*1000*1000 < now {
5350 sched.lastpoll.CompareAndSwap(lastpoll, now)
5351 list := netpoll(0) // non-blocking - returns list of goroutines
5353 // Need to decrement number of idle locked M's
5354 // (pretending that one more is running) before injectglist.
5355 // Otherwise it can lead to the following situation:
5356 // injectglist grabs all P's but before it starts M's to run the P's,
5357 // another M returns from syscall, finishes running its G,
5358 // observes that there is no work to do and no other running M's
5359 // and reports deadlock.
5365 if GOOS == "netbsd" && needSysmonWorkaround {
5366 // netpoll is responsible for waiting for timer
5367 // expiration, so we typically don't have to worry
5368 // about starting an M to service timers. (Note that
5369 // sleep for timeSleepUntil above simply ensures sysmon
5370 // starts running again when that timer expiration may
5371 // cause Go code to run again).
5373 // However, netbsd has a kernel bug that sometimes
5374 // misses netpollBreak wake-ups, which can lead to
5375 // unbounded delays servicing timers. If we detect this
5376 // overrun, then startm to get something to handle the
5379 // See issue 42515 and
5380 // https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=50094.
5381 if next := timeSleepUntil(); next < now {
5385 if scavenger.sysmonWake.Load() != 0 {
5386 // Kick the scavenger awake if someone requested it.
5389 // retake P's blocked in syscalls
5390 // and preempt long running G's
5391 if retake(now) != 0 {
5396 // check if we need to force a GC
5397 if t := (gcTrigger{kind: gcTriggerTime, now: now}); t.test() && forcegc.idle.Load() {
5399 forcegc.idle.Store(false)
5401 list.push(forcegc.g)
5403 unlock(&forcegc.lock)
5405 if debug.schedtrace > 0 && lasttrace+int64(debug.schedtrace)*1000000 <= now {
5407 schedtrace(debug.scheddetail > 0)
5409 unlock(&sched.sysmonlock)
5413 type sysmontick struct {
5420 // forcePreemptNS is the time slice given to a G before it is
5422 const forcePreemptNS = 10 * 1000 * 1000 // 10ms
5424 func retake(now int64) uint32 {
5426 // Prevent allp slice changes. This lock will be completely
5427 // uncontended unless we're already stopping the world.
5429 // We can't use a range loop over allp because we may
5430 // temporarily drop the allpLock. Hence, we need to re-fetch
5431 // allp each time around the loop.
5432 for i := 0; i < len(allp); i++ {
5435 // This can happen if procresize has grown
5436 // allp but not yet created new Ps.
5439 pd := &pp.sysmontick
5442 if s == _Prunning || s == _Psyscall {
5443 // Preempt G if it's running for too long.
5444 t := int64(pp.schedtick)
5445 if int64(pd.schedtick) != t {
5446 pd.schedtick = uint32(t)
5448 } else if pd.schedwhen+forcePreemptNS <= now {
5450 // In case of syscall, preemptone() doesn't
5451 // work, because there is no M wired to P.
5456 // Retake P from syscall if it's there for more than 1 sysmon tick (at least 20us).
5457 t := int64(pp.syscalltick)
5458 if !sysretake && int64(pd.syscalltick) != t {
5459 pd.syscalltick = uint32(t)
5460 pd.syscallwhen = now
5463 // On the one hand we don't want to retake Ps if there is no other work to do,
5464 // but on the other hand we want to retake them eventually
5465 // because they can prevent the sysmon thread from deep sleep.
5466 if runqempty(pp) && sched.nmspinning.Load()+sched.npidle.Load() > 0 && pd.syscallwhen+10*1000*1000 > now {
5469 // Drop allpLock so we can take sched.lock.
5471 // Need to decrement number of idle locked M's
5472 // (pretending that one more is running) before the CAS.
5473 // Otherwise the M from which we retake can exit the syscall,
5474 // increment nmidle and report deadlock.
5476 if atomic.Cas(&pp.status, s, _Pidle) {
5493 // Tell all goroutines that they have been preempted and they should stop.
5494 // This function is purely best-effort. It can fail to inform a goroutine if a
5495 // processor just started running it.
5496 // No locks need to be held.
5497 // Returns true if preemption request was issued to at least one goroutine.
5498 func preemptall() bool {
5500 for _, pp := range allp {
5501 if pp.status != _Prunning {
5511 // Tell the goroutine running on processor P to stop.
5512 // This function is purely best-effort. It can incorrectly fail to inform the
5513 // goroutine. It can inform the wrong goroutine. Even if it informs the
5514 // correct goroutine, that goroutine might ignore the request if it is
5515 // simultaneously executing newstack.
5516 // No lock needs to be held.
5517 // Returns true if preemption request was issued.
5518 // The actual preemption will happen at some point in the future
5519 // and will be indicated by the gp->status no longer being
5521 func preemptone(pp *p) bool {
5523 if mp == nil || mp == getg().m {
5527 if gp == nil || gp == mp.g0 {
5533 // Every call in a goroutine checks for stack overflow by
5534 // comparing the current stack pointer to gp->stackguard0.
5535 // Setting gp->stackguard0 to StackPreempt folds
5536 // preemption into the normal stack overflow check.
5537 gp.stackguard0 = stackPreempt
5539 // Request an async preemption of this P.
5540 if preemptMSupported && debug.asyncpreemptoff == 0 {
5550 func schedtrace(detailed bool) {
5557 print("SCHED ", (now-starttime)/1e6, "ms: gomaxprocs=", gomaxprocs, " idleprocs=", sched.npidle.Load(), " threads=", mcount(), " spinningthreads=", sched.nmspinning.Load(), " needspinning=", sched.needspinning.Load(), " idlethreads=", sched.nmidle, " runqueue=", sched.runqsize)
5559 print(" gcwaiting=", sched.gcwaiting.Load(), " nmidlelocked=", sched.nmidlelocked, " stopwait=", sched.stopwait, " sysmonwait=", sched.sysmonwait.Load(), "\n")
5561 // We must be careful while reading data from P's, M's and G's.
5562 // Even if we hold schedlock, most data can be changed concurrently.
5563 // E.g. (p->m ? p->m->id : -1) can crash if p->m changes from non-nil to nil.
5564 for i, pp := range allp {
5566 h := atomic.Load(&pp.runqhead)
5567 t := atomic.Load(&pp.runqtail)
5569 print(" P", i, ": status=", pp.status, " schedtick=", pp.schedtick, " syscalltick=", pp.syscalltick, " m=")
5575 print(" runqsize=", t-h, " gfreecnt=", pp.gFree.n, " timerslen=", len(pp.timers), "\n")
5577 // In non-detailed mode format lengths of per-P run queues as:
5578 // [len1 len2 len3 len4]
5584 if i == len(allp)-1 {
5595 for mp := allm; mp != nil; mp = mp.alllink {
5597 print(" M", mp.id, ": p=")
5609 print(" mallocing=", mp.mallocing, " throwing=", mp.throwing, " preemptoff=", mp.preemptoff, " locks=", mp.locks, " dying=", mp.dying, " spinning=", mp.spinning, " blocked=", mp.blocked, " lockedg=")
5610 if lockedg := mp.lockedg.ptr(); lockedg != nil {
5618 forEachG(func(gp *g) {
5619 print(" G", gp.goid, ": status=", readgstatus(gp), "(", gp.waitreason.String(), ") m=")
5626 if lockedm := gp.lockedm.ptr(); lockedm != nil {
5636 // schedEnableUser enables or disables the scheduling of user
5639 // This does not stop already running user goroutines, so the caller
5640 // should first stop the world when disabling user goroutines.
5641 func schedEnableUser(enable bool) {
5643 if sched.disable.user == !enable {
5647 sched.disable.user = !enable
5649 n := sched.disable.n
5651 globrunqputbatch(&sched.disable.runnable, n)
5653 for ; n != 0 && sched.npidle.Load() != 0; n-- {
5661 // schedEnabled reports whether gp should be scheduled. It returns
5662 // false is scheduling of gp is disabled.
5664 // sched.lock must be held.
5665 func schedEnabled(gp *g) bool {
5666 assertLockHeld(&sched.lock)
5668 if sched.disable.user {
5669 return isSystemGoroutine(gp, true)
5674 // Put mp on midle list.
5675 // sched.lock must be held.
5676 // May run during STW, so write barriers are not allowed.
5678 //go:nowritebarrierrec
5680 assertLockHeld(&sched.lock)
5682 mp.schedlink = sched.midle
5688 // Try to get an m from midle list.
5689 // sched.lock must be held.
5690 // May run during STW, so write barriers are not allowed.
5692 //go:nowritebarrierrec
5694 assertLockHeld(&sched.lock)
5696 mp := sched.midle.ptr()
5698 sched.midle = mp.schedlink
5704 // Put gp on the global runnable queue.
5705 // sched.lock must be held.
5706 // May run during STW, so write barriers are not allowed.
5708 //go:nowritebarrierrec
5709 func globrunqput(gp *g) {
5710 assertLockHeld(&sched.lock)
5712 sched.runq.pushBack(gp)
5716 // Put gp at the head of the global runnable queue.
5717 // sched.lock must be held.
5718 // May run during STW, so write barriers are not allowed.
5720 //go:nowritebarrierrec
5721 func globrunqputhead(gp *g) {
5722 assertLockHeld(&sched.lock)
5728 // Put a batch of runnable goroutines on the global runnable queue.
5729 // This clears *batch.
5730 // sched.lock must be held.
5731 // May run during STW, so write barriers are not allowed.
5733 //go:nowritebarrierrec
5734 func globrunqputbatch(batch *gQueue, n int32) {
5735 assertLockHeld(&sched.lock)
5737 sched.runq.pushBackAll(*batch)
5742 // Try get a batch of G's from the global runnable queue.
5743 // sched.lock must be held.
5744 func globrunqget(pp *p, max int32) *g {
5745 assertLockHeld(&sched.lock)
5747 if sched.runqsize == 0 {
5751 n := sched.runqsize/gomaxprocs + 1
5752 if n > sched.runqsize {
5755 if max > 0 && n > max {
5758 if n > int32(len(pp.runq))/2 {
5759 n = int32(len(pp.runq)) / 2
5764 gp := sched.runq.pop()
5767 gp1 := sched.runq.pop()
5768 runqput(pp, gp1, false)
5773 // pMask is an atomic bitstring with one bit per P.
5776 // read returns true if P id's bit is set.
5777 func (p pMask) read(id uint32) bool {
5779 mask := uint32(1) << (id % 32)
5780 return (atomic.Load(&p[word]) & mask) != 0
5783 // set sets P id's bit.
5784 func (p pMask) set(id int32) {
5786 mask := uint32(1) << (id % 32)
5787 atomic.Or(&p[word], mask)
5790 // clear clears P id's bit.
5791 func (p pMask) clear(id int32) {
5793 mask := uint32(1) << (id % 32)
5794 atomic.And(&p[word], ^mask)
5797 // updateTimerPMask clears pp's timer mask if it has no timers on its heap.
5799 // Ideally, the timer mask would be kept immediately consistent on any timer
5800 // operations. Unfortunately, updating a shared global data structure in the
5801 // timer hot path adds too much overhead in applications frequently switching
5802 // between no timers and some timers.
5804 // As a compromise, the timer mask is updated only on pidleget / pidleput. A
5805 // running P (returned by pidleget) may add a timer at any time, so its mask
5806 // must be set. An idle P (passed to pidleput) cannot add new timers while
5807 // idle, so if it has no timers at that time, its mask may be cleared.
5809 // Thus, we get the following effects on timer-stealing in findrunnable:
5811 // - Idle Ps with no timers when they go idle are never checked in findrunnable
5812 // (for work- or timer-stealing; this is the ideal case).
5813 // - Running Ps must always be checked.
5814 // - Idle Ps whose timers are stolen must continue to be checked until they run
5815 // again, even after timer expiration.
5817 // When the P starts running again, the mask should be set, as a timer may be
5818 // added at any time.
5820 // TODO(prattmic): Additional targeted updates may improve the above cases.
5821 // e.g., updating the mask when stealing a timer.
5822 func updateTimerPMask(pp *p) {
5823 if pp.numTimers.Load() > 0 {
5827 // Looks like there are no timers, however another P may transiently
5828 // decrement numTimers when handling a timerModified timer in
5829 // checkTimers. We must take timersLock to serialize with these changes.
5830 lock(&pp.timersLock)
5831 if pp.numTimers.Load() == 0 {
5832 timerpMask.clear(pp.id)
5834 unlock(&pp.timersLock)
5837 // pidleput puts p on the _Pidle list. now must be a relatively recent call
5838 // to nanotime or zero. Returns now or the current time if now was zero.
5840 // This releases ownership of p. Once sched.lock is released it is no longer
5843 // sched.lock must be held.
5845 // May run during STW, so write barriers are not allowed.
5847 //go:nowritebarrierrec
5848 func pidleput(pp *p, now int64) int64 {
5849 assertLockHeld(&sched.lock)
5852 throw("pidleput: P has non-empty run queue")
5857 updateTimerPMask(pp) // clear if there are no timers.
5858 idlepMask.set(pp.id)
5859 pp.link = sched.pidle
5862 if !pp.limiterEvent.start(limiterEventIdle, now) {
5863 throw("must be able to track idle limiter event")
5868 // pidleget tries to get a p from the _Pidle list, acquiring ownership.
5870 // sched.lock must be held.
5872 // May run during STW, so write barriers are not allowed.
5874 //go:nowritebarrierrec
5875 func pidleget(now int64) (*p, int64) {
5876 assertLockHeld(&sched.lock)
5878 pp := sched.pidle.ptr()
5880 // Timer may get added at any time now.
5884 timerpMask.set(pp.id)
5885 idlepMask.clear(pp.id)
5886 sched.pidle = pp.link
5887 sched.npidle.Add(-1)
5888 pp.limiterEvent.stop(limiterEventIdle, now)
5893 // pidlegetSpinning tries to get a p from the _Pidle list, acquiring ownership.
5894 // This is called by spinning Ms (or callers than need a spinning M) that have
5895 // found work. If no P is available, this must synchronized with non-spinning
5896 // Ms that may be preparing to drop their P without discovering this work.
5898 // sched.lock must be held.
5900 // May run during STW, so write barriers are not allowed.
5902 //go:nowritebarrierrec
5903 func pidlegetSpinning(now int64) (*p, int64) {
5904 assertLockHeld(&sched.lock)
5906 pp, now := pidleget(now)
5908 // See "Delicate dance" comment in findrunnable. We found work
5909 // that we cannot take, we must synchronize with non-spinning
5910 // Ms that may be preparing to drop their P.
5911 sched.needspinning.Store(1)
5918 // runqempty reports whether pp has no Gs on its local run queue.
5919 // It never returns true spuriously.
5920 func runqempty(pp *p) bool {
5921 // Defend against a race where 1) pp has G1 in runqnext but runqhead == runqtail,
5922 // 2) runqput on pp kicks G1 to the runq, 3) runqget on pp empties runqnext.
5923 // Simply observing that runqhead == runqtail and then observing that runqnext == nil
5924 // does not mean the queue is empty.
5926 head := atomic.Load(&pp.runqhead)
5927 tail := atomic.Load(&pp.runqtail)
5928 runnext := atomic.Loaduintptr((*uintptr)(unsafe.Pointer(&pp.runnext)))
5929 if tail == atomic.Load(&pp.runqtail) {
5930 return head == tail && runnext == 0
5935 // To shake out latent assumptions about scheduling order,
5936 // we introduce some randomness into scheduling decisions
5937 // when running with the race detector.
5938 // The need for this was made obvious by changing the
5939 // (deterministic) scheduling order in Go 1.5 and breaking
5940 // many poorly-written tests.
5941 // With the randomness here, as long as the tests pass
5942 // consistently with -race, they shouldn't have latent scheduling
5944 const randomizeScheduler = raceenabled
5946 // runqput tries to put g on the local runnable queue.
5947 // If next is false, runqput adds g to the tail of the runnable queue.
5948 // If next is true, runqput puts g in the pp.runnext slot.
5949 // If the run queue is full, runnext puts g on the global queue.
5950 // Executed only by the owner P.
5951 func runqput(pp *p, gp *g, next bool) {
5952 if randomizeScheduler && next && fastrandn(2) == 0 {
5958 oldnext := pp.runnext
5959 if !pp.runnext.cas(oldnext, guintptr(unsafe.Pointer(gp))) {
5965 // Kick the old runnext out to the regular run queue.
5970 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
5972 if t-h < uint32(len(pp.runq)) {
5973 pp.runq[t%uint32(len(pp.runq))].set(gp)
5974 atomic.StoreRel(&pp.runqtail, t+1) // store-release, makes the item available for consumption
5977 if runqputslow(pp, gp, h, t) {
5980 // the queue is not full, now the put above must succeed
5984 // Put g and a batch of work from local runnable queue on global queue.
5985 // Executed only by the owner P.
5986 func runqputslow(pp *p, gp *g, h, t uint32) bool {
5987 var batch [len(pp.runq)/2 + 1]*g
5989 // First, grab a batch from local queue.
5992 if n != uint32(len(pp.runq)/2) {
5993 throw("runqputslow: queue is not full")
5995 for i := uint32(0); i < n; i++ {
5996 batch[i] = pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
5998 if !atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6003 if randomizeScheduler {
6004 for i := uint32(1); i <= n; i++ {
6005 j := fastrandn(i + 1)
6006 batch[i], batch[j] = batch[j], batch[i]
6010 // Link the goroutines.
6011 for i := uint32(0); i < n; i++ {
6012 batch[i].schedlink.set(batch[i+1])
6015 q.head.set(batch[0])
6016 q.tail.set(batch[n])
6018 // Now put the batch on global queue.
6020 globrunqputbatch(&q, int32(n+1))
6025 // runqputbatch tries to put all the G's on q on the local runnable queue.
6026 // If the queue is full, they are put on the global queue; in that case
6027 // this will temporarily acquire the scheduler lock.
6028 // Executed only by the owner P.
6029 func runqputbatch(pp *p, q *gQueue, qsize int) {
6030 h := atomic.LoadAcq(&pp.runqhead)
6033 for !q.empty() && t-h < uint32(len(pp.runq)) {
6035 pp.runq[t%uint32(len(pp.runq))].set(gp)
6041 if randomizeScheduler {
6042 off := func(o uint32) uint32 {
6043 return (pp.runqtail + o) % uint32(len(pp.runq))
6045 for i := uint32(1); i < n; i++ {
6046 j := fastrandn(i + 1)
6047 pp.runq[off(i)], pp.runq[off(j)] = pp.runq[off(j)], pp.runq[off(i)]
6051 atomic.StoreRel(&pp.runqtail, t)
6054 globrunqputbatch(q, int32(qsize))
6059 // Get g from local runnable queue.
6060 // If inheritTime is true, gp should inherit the remaining time in the
6061 // current time slice. Otherwise, it should start a new time slice.
6062 // Executed only by the owner P.
6063 func runqget(pp *p) (gp *g, inheritTime bool) {
6064 // If there's a runnext, it's the next G to run.
6066 // If the runnext is non-0 and the CAS fails, it could only have been stolen by another P,
6067 // because other Ps can race to set runnext to 0, but only the current P can set it to non-0.
6068 // Hence, there's no need to retry this CAS if it fails.
6069 if next != 0 && pp.runnext.cas(next, 0) {
6070 return next.ptr(), true
6074 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6079 gp := pp.runq[h%uint32(len(pp.runq))].ptr()
6080 if atomic.CasRel(&pp.runqhead, h, h+1) { // cas-release, commits consume
6086 // runqdrain drains the local runnable queue of pp and returns all goroutines in it.
6087 // Executed only by the owner P.
6088 func runqdrain(pp *p) (drainQ gQueue, n uint32) {
6089 oldNext := pp.runnext
6090 if oldNext != 0 && pp.runnext.cas(oldNext, 0) {
6091 drainQ.pushBack(oldNext.ptr())
6096 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6102 if qn > uint32(len(pp.runq)) { // read inconsistent h and t
6106 if !atomic.CasRel(&pp.runqhead, h, h+qn) { // cas-release, commits consume
6110 // We've inverted the order in which it gets G's from the local P's runnable queue
6111 // and then advances the head pointer because we don't want to mess up the statuses of G's
6112 // while runqdrain() and runqsteal() are running in parallel.
6113 // Thus we should advance the head pointer before draining the local P into a gQueue,
6114 // so that we can update any gp.schedlink only after we take the full ownership of G,
6115 // meanwhile, other P's can't access to all G's in local P's runnable queue and steal them.
6116 // See https://groups.google.com/g/golang-dev/c/0pTKxEKhHSc/m/6Q85QjdVBQAJ for more details.
6117 for i := uint32(0); i < qn; i++ {
6118 gp := pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
6125 // Grabs a batch of goroutines from pp's runnable queue into batch.
6126 // Batch is a ring buffer starting at batchHead.
6127 // Returns number of grabbed goroutines.
6128 // Can be executed by any P.
6129 func runqgrab(pp *p, batch *[256]guintptr, batchHead uint32, stealRunNextG bool) uint32 {
6131 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6132 t := atomic.LoadAcq(&pp.runqtail) // load-acquire, synchronize with the producer
6137 // Try to steal from pp.runnext.
6138 if next := pp.runnext; next != 0 {
6139 if pp.status == _Prunning {
6140 // Sleep to ensure that pp isn't about to run the g
6141 // we are about to steal.
6142 // The important use case here is when the g running
6143 // on pp ready()s another g and then almost
6144 // immediately blocks. Instead of stealing runnext
6145 // in this window, back off to give pp a chance to
6146 // schedule runnext. This will avoid thrashing gs
6147 // between different Ps.
6148 // A sync chan send/recv takes ~50ns as of time of
6149 // writing, so 3us gives ~50x overshoot.
6150 if GOOS != "windows" && GOOS != "openbsd" && GOOS != "netbsd" {
6153 // On some platforms system timer granularity is
6154 // 1-15ms, which is way too much for this
6155 // optimization. So just yield.
6159 if !pp.runnext.cas(next, 0) {
6162 batch[batchHead%uint32(len(batch))] = next
6168 if n > uint32(len(pp.runq)/2) { // read inconsistent h and t
6171 for i := uint32(0); i < n; i++ {
6172 g := pp.runq[(h+i)%uint32(len(pp.runq))]
6173 batch[(batchHead+i)%uint32(len(batch))] = g
6175 if atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6181 // Steal half of elements from local runnable queue of p2
6182 // and put onto local runnable queue of p.
6183 // Returns one of the stolen elements (or nil if failed).
6184 func runqsteal(pp, p2 *p, stealRunNextG bool) *g {
6186 n := runqgrab(p2, &pp.runq, t, stealRunNextG)
6191 gp := pp.runq[(t+n)%uint32(len(pp.runq))].ptr()
6195 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
6196 if t-h+n >= uint32(len(pp.runq)) {
6197 throw("runqsteal: runq overflow")
6199 atomic.StoreRel(&pp.runqtail, t+n) // store-release, makes the item available for consumption
6203 // A gQueue is a dequeue of Gs linked through g.schedlink. A G can only
6204 // be on one gQueue or gList at a time.
6205 type gQueue struct {
6210 // empty reports whether q is empty.
6211 func (q *gQueue) empty() bool {
6215 // push adds gp to the head of q.
6216 func (q *gQueue) push(gp *g) {
6217 gp.schedlink = q.head
6224 // pushBack adds gp to the tail of q.
6225 func (q *gQueue) pushBack(gp *g) {
6228 q.tail.ptr().schedlink.set(gp)
6235 // pushBackAll adds all Gs in q2 to the tail of q. After this q2 must
6237 func (q *gQueue) pushBackAll(q2 gQueue) {
6241 q2.tail.ptr().schedlink = 0
6243 q.tail.ptr().schedlink = q2.head
6250 // pop removes and returns the head of queue q. It returns nil if
6252 func (q *gQueue) pop() *g {
6255 q.head = gp.schedlink
6263 // popList takes all Gs in q and returns them as a gList.
6264 func (q *gQueue) popList() gList {
6265 stack := gList{q.head}
6270 // A gList is a list of Gs linked through g.schedlink. A G can only be
6271 // on one gQueue or gList at a time.
6276 // empty reports whether l is empty.
6277 func (l *gList) empty() bool {
6281 // push adds gp to the head of l.
6282 func (l *gList) push(gp *g) {
6283 gp.schedlink = l.head
6287 // pushAll prepends all Gs in q to l.
6288 func (l *gList) pushAll(q gQueue) {
6290 q.tail.ptr().schedlink = l.head
6295 // pop removes and returns the head of l. If l is empty, it returns nil.
6296 func (l *gList) pop() *g {
6299 l.head = gp.schedlink
6304 //go:linkname setMaxThreads runtime/debug.setMaxThreads
6305 func setMaxThreads(in int) (out int) {
6307 out = int(sched.maxmcount)
6308 if in > 0x7fffffff { // MaxInt32
6309 sched.maxmcount = 0x7fffffff
6311 sched.maxmcount = int32(in)
6319 func procPin() int {
6324 return int(mp.p.ptr().id)
6333 //go:linkname sync_runtime_procPin sync.runtime_procPin
6335 func sync_runtime_procPin() int {
6339 //go:linkname sync_runtime_procUnpin sync.runtime_procUnpin
6341 func sync_runtime_procUnpin() {
6345 //go:linkname sync_atomic_runtime_procPin sync/atomic.runtime_procPin
6347 func sync_atomic_runtime_procPin() int {
6351 //go:linkname sync_atomic_runtime_procUnpin sync/atomic.runtime_procUnpin
6353 func sync_atomic_runtime_procUnpin() {
6357 // Active spinning for sync.Mutex.
6359 //go:linkname sync_runtime_canSpin sync.runtime_canSpin
6361 func sync_runtime_canSpin(i int) bool {
6362 // sync.Mutex is cooperative, so we are conservative with spinning.
6363 // Spin only few times and only if running on a multicore machine and
6364 // GOMAXPROCS>1 and there is at least one other running P and local runq is empty.
6365 // As opposed to runtime mutex we don't do passive spinning here,
6366 // because there can be work on global runq or on other Ps.
6367 if i >= active_spin || ncpu <= 1 || gomaxprocs <= sched.npidle.Load()+sched.nmspinning.Load()+1 {
6370 if p := getg().m.p.ptr(); !runqempty(p) {
6376 //go:linkname sync_runtime_doSpin sync.runtime_doSpin
6378 func sync_runtime_doSpin() {
6379 procyield(active_spin_cnt)
6382 var stealOrder randomOrder
6384 // randomOrder/randomEnum are helper types for randomized work stealing.
6385 // They allow to enumerate all Ps in different pseudo-random orders without repetitions.
6386 // The algorithm is based on the fact that if we have X such that X and GOMAXPROCS
6387 // are coprime, then a sequences of (i + X) % GOMAXPROCS gives the required enumeration.
6388 type randomOrder struct {
6393 type randomEnum struct {
6400 func (ord *randomOrder) reset(count uint32) {
6402 ord.coprimes = ord.coprimes[:0]
6403 for i := uint32(1); i <= count; i++ {
6404 if gcd(i, count) == 1 {
6405 ord.coprimes = append(ord.coprimes, i)
6410 func (ord *randomOrder) start(i uint32) randomEnum {
6414 inc: ord.coprimes[i/ord.count%uint32(len(ord.coprimes))],
6418 func (enum *randomEnum) done() bool {
6419 return enum.i == enum.count
6422 func (enum *randomEnum) next() {
6424 enum.pos = (enum.pos + enum.inc) % enum.count
6427 func (enum *randomEnum) position() uint32 {
6431 func gcd(a, b uint32) uint32 {
6438 // An initTask represents the set of initializations that need to be done for a package.
6439 // Keep in sync with ../../test/initempty.go:initTask
6440 type initTask struct {
6441 // TODO: pack the first 3 fields more tightly?
6442 state uintptr // 0 = uninitialized, 1 = in progress, 2 = done
6445 // followed by ndeps instances of an *initTask, one per package depended on
6446 // followed by nfns pcs, one per init function to run
6449 // inittrace stores statistics for init functions which are
6450 // updated by malloc and newproc when active is true.
6451 var inittrace tracestat
6453 type tracestat struct {
6454 active bool // init tracing activation status
6455 id uint64 // init goroutine id
6456 allocs uint64 // heap allocations
6457 bytes uint64 // heap allocated bytes
6460 func doInit(t *initTask) {
6462 case 2: // fully initialized
6464 case 1: // initialization in progress
6465 throw("recursive call during initialization - linker skew")
6466 default: // not initialized yet
6467 t.state = 1 // initialization in progress
6469 for i := uintptr(0); i < t.ndeps; i++ {
6470 p := add(unsafe.Pointer(t), (3+i)*goarch.PtrSize)
6471 t2 := *(**initTask)(p)
6476 t.state = 2 // initialization done
6485 if inittrace.active {
6487 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6491 firstFunc := add(unsafe.Pointer(t), (3+t.ndeps)*goarch.PtrSize)
6492 for i := uintptr(0); i < t.nfns; i++ {
6493 p := add(firstFunc, i*goarch.PtrSize)
6494 f := *(*func())(unsafe.Pointer(&p))
6498 if inittrace.active {
6500 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6503 f := *(*func())(unsafe.Pointer(&firstFunc))
6504 pkg := funcpkgpath(findfunc(abi.FuncPCABIInternal(f)))
6507 print("init ", pkg, " @")
6508 print(string(fmtNSAsMS(sbuf[:], uint64(start-runtimeInitTime))), " ms, ")
6509 print(string(fmtNSAsMS(sbuf[:], uint64(end-start))), " ms clock, ")
6510 print(string(itoa(sbuf[:], after.bytes-before.bytes)), " bytes, ")
6511 print(string(itoa(sbuf[:], after.allocs-before.allocs)), " allocs")
6515 t.state = 2 // initialization done