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 or global 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
121 // This slice records the initializing tasks that need to be
122 // done to start up the runtime. It is built by the linker.
123 var runtime_inittasks []*initTask
125 // main_init_done is a signal used by cgocallbackg that initialization
126 // has been completed. It is made before _cgo_notify_runtime_init_done,
127 // so all cgo calls can rely on it existing. When main_init is complete,
128 // it is closed, meaning cgocallbackg can reliably receive from it.
129 var main_init_done chan bool
131 //go:linkname main_main main.main
134 // mainStarted indicates that the main M has started.
137 // runtimeInitTime is the nanotime() at which the runtime started.
138 var runtimeInitTime int64
140 // Value to use for signal mask for newly created M's.
141 var initSigmask sigset
143 // The main goroutine.
147 // Racectx of m0->g0 is used only as the parent of the main goroutine.
148 // It must not be used for anything else.
151 // Max stack size is 1 GB on 64-bit, 250 MB on 32-bit.
152 // Using decimal instead of binary GB and MB because
153 // they look nicer in the stack overflow failure message.
154 if goarch.PtrSize == 8 {
155 maxstacksize = 1000000000
157 maxstacksize = 250000000
160 // An upper limit for max stack size. Used to avoid random crashes
161 // after calling SetMaxStack and trying to allocate a stack that is too big,
162 // since stackalloc works with 32-bit sizes.
163 maxstackceiling = 2 * maxstacksize
165 // Allow newproc to start new Ms.
168 if GOARCH != "wasm" { // no threads on wasm yet, so no sysmon
170 newm(sysmon, nil, -1)
174 // Lock the main goroutine onto this, the main OS thread,
175 // during initialization. Most programs won't care, but a few
176 // do require certain calls to be made by the main thread.
177 // Those can arrange for main.main to run in the main thread
178 // by calling runtime.LockOSThread during initialization
179 // to preserve the lock.
183 throw("runtime.main not on m0")
186 // Record when the world started.
187 // Must be before doInit for tracing init.
188 runtimeInitTime = nanotime()
189 if runtimeInitTime == 0 {
190 throw("nanotime returning zero")
193 if debug.inittrace != 0 {
194 inittrace.id = getg().goid
195 inittrace.active = true
198 doInit(runtime_inittasks) // Must be before defer.
200 // Defer unlock so that runtime.Goexit during init does the unlock too.
210 main_init_done = make(chan bool)
212 if _cgo_pthread_key_created == nil {
213 throw("_cgo_pthread_key_created missing")
216 if _cgo_thread_start == nil {
217 throw("_cgo_thread_start missing")
219 if GOOS != "windows" {
220 if _cgo_setenv == nil {
221 throw("_cgo_setenv missing")
223 if _cgo_unsetenv == nil {
224 throw("_cgo_unsetenv missing")
227 if _cgo_notify_runtime_init_done == nil {
228 throw("_cgo_notify_runtime_init_done missing")
231 // Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
232 if set_crosscall2 == nil {
233 throw("set_crosscall2 missing")
237 // Start the template thread in case we enter Go from
238 // a C-created thread and need to create a new thread.
239 startTemplateThread()
240 cgocall(_cgo_notify_runtime_init_done, nil)
243 // Run the initializing tasks. Depending on build mode this
244 // list can arrive a few different ways, but it will always
245 // contain the init tasks computed by the linker for all the
246 // packages in the program (excluding those added at runtime
247 // by package plugin).
248 for _, m := range activeModules() {
252 // Disable init tracing after main init done to avoid overhead
253 // of collecting statistics in malloc and newproc
254 inittrace.active = false
256 close(main_init_done)
261 if isarchive || islibrary {
262 // A program compiled with -buildmode=c-archive or c-shared
263 // has a main, but it is not executed.
266 fn := main_main // make an indirect call, as the linker doesn't know the address of the main package when laying down the runtime
269 runExitHooks(0) // run hooks now, since racefini does not return
273 // Make racy client program work: if panicking on
274 // another goroutine at the same time as main returns,
275 // let the other goroutine finish printing the panic trace.
276 // Once it does, it will exit. See issues 3934 and 20018.
277 if runningPanicDefers.Load() != 0 {
278 // Running deferred functions should not take long.
279 for c := 0; c < 1000; c++ {
280 if runningPanicDefers.Load() == 0 {
286 if panicking.Load() != 0 {
287 gopark(nil, nil, waitReasonPanicWait, traceBlockForever, 1)
298 // os_beforeExit is called from os.Exit(0).
300 //go:linkname os_beforeExit os.runtime_beforeExit
301 func os_beforeExit(exitCode int) {
302 runExitHooks(exitCode)
303 if exitCode == 0 && raceenabled {
308 // start forcegc helper goroutine
313 func forcegchelper() {
315 lockInit(&forcegc.lock, lockRankForcegc)
318 if forcegc.idle.Load() {
319 throw("forcegc: phase error")
321 forcegc.idle.Store(true)
322 goparkunlock(&forcegc.lock, waitReasonForceGCIdle, traceBlockSystemGoroutine, 1)
323 // this goroutine is explicitly resumed by sysmon
324 if debug.gctrace > 0 {
327 // Time-triggered, fully concurrent.
328 gcStart(gcTrigger{kind: gcTriggerTime, now: nanotime()})
332 // Gosched yields the processor, allowing other goroutines to run. It does not
333 // suspend the current goroutine, so execution resumes automatically.
341 // goschedguarded yields the processor like gosched, but also checks
342 // for forbidden states and opts out of the yield in those cases.
345 func goschedguarded() {
346 mcall(goschedguarded_m)
349 // goschedIfBusy yields the processor like gosched, but only does so if
350 // there are no idle Ps or if we're on the only P and there's nothing in
351 // the run queue. In both cases, there is freely available idle time.
354 func goschedIfBusy() {
356 // Call gosched if gp.preempt is set; we may be in a tight loop that
357 // doesn't otherwise yield.
358 if !gp.preempt && sched.npidle.Load() > 0 {
364 // Puts the current goroutine into a waiting state and calls unlockf on the
367 // If unlockf returns false, the goroutine is resumed.
369 // unlockf must not access this G's stack, as it may be moved between
370 // the call to gopark and the call to unlockf.
372 // Note that because unlockf is called after putting the G into a waiting
373 // state, the G may have already been readied by the time unlockf is called
374 // unless there is external synchronization preventing the G from being
375 // readied. If unlockf returns false, it must guarantee that the G cannot be
376 // externally readied.
378 // Reason explains why the goroutine has been parked. It is displayed in stack
379 // traces and heap dumps. Reasons should be unique and descriptive. Do not
380 // re-use reasons, add new ones.
381 func gopark(unlockf func(*g, unsafe.Pointer) bool, lock unsafe.Pointer, reason waitReason, traceReason traceBlockReason, traceskip int) {
382 if reason != waitReasonSleep {
383 checkTimeouts() // timeouts may expire while two goroutines keep the scheduler busy
387 status := readgstatus(gp)
388 if status != _Grunning && status != _Gscanrunning {
389 throw("gopark: bad g status")
392 mp.waitunlockf = unlockf
393 gp.waitreason = reason
394 mp.waitTraceBlockReason = traceReason
395 mp.waitTraceSkip = traceskip
397 // can't do anything that might move the G between Ms here.
401 // Puts the current goroutine into a waiting state and unlocks the lock.
402 // The goroutine can be made runnable again by calling goready(gp).
403 func goparkunlock(lock *mutex, reason waitReason, traceReason traceBlockReason, traceskip int) {
404 gopark(parkunlock_c, unsafe.Pointer(lock), reason, traceReason, traceskip)
407 func goready(gp *g, traceskip int) {
409 ready(gp, traceskip, true)
414 func acquireSudog() *sudog {
415 // Delicate dance: the semaphore implementation calls
416 // acquireSudog, acquireSudog calls new(sudog),
417 // new calls malloc, malloc can call the garbage collector,
418 // and the garbage collector calls the semaphore implementation
420 // Break the cycle by doing acquirem/releasem around new(sudog).
421 // The acquirem/releasem increments m.locks during new(sudog),
422 // which keeps the garbage collector from being invoked.
425 if len(pp.sudogcache) == 0 {
426 lock(&sched.sudoglock)
427 // First, try to grab a batch from central cache.
428 for len(pp.sudogcache) < cap(pp.sudogcache)/2 && sched.sudogcache != nil {
429 s := sched.sudogcache
430 sched.sudogcache = s.next
432 pp.sudogcache = append(pp.sudogcache, s)
434 unlock(&sched.sudoglock)
435 // If the central cache is empty, allocate a new one.
436 if len(pp.sudogcache) == 0 {
437 pp.sudogcache = append(pp.sudogcache, new(sudog))
440 n := len(pp.sudogcache)
441 s := pp.sudogcache[n-1]
442 pp.sudogcache[n-1] = nil
443 pp.sudogcache = pp.sudogcache[:n-1]
445 throw("acquireSudog: found s.elem != nil in cache")
452 func releaseSudog(s *sudog) {
454 throw("runtime: sudog with non-nil elem")
457 throw("runtime: sudog with non-false isSelect")
460 throw("runtime: sudog with non-nil next")
463 throw("runtime: sudog with non-nil prev")
465 if s.waitlink != nil {
466 throw("runtime: sudog with non-nil waitlink")
469 throw("runtime: sudog with non-nil c")
473 throw("runtime: releaseSudog with non-nil gp.param")
475 mp := acquirem() // avoid rescheduling to another P
477 if len(pp.sudogcache) == cap(pp.sudogcache) {
478 // Transfer half of local cache to the central cache.
479 var first, last *sudog
480 for len(pp.sudogcache) > cap(pp.sudogcache)/2 {
481 n := len(pp.sudogcache)
482 p := pp.sudogcache[n-1]
483 pp.sudogcache[n-1] = nil
484 pp.sudogcache = pp.sudogcache[:n-1]
492 lock(&sched.sudoglock)
493 last.next = sched.sudogcache
494 sched.sudogcache = first
495 unlock(&sched.sudoglock)
497 pp.sudogcache = append(pp.sudogcache, s)
501 // called from assembly.
502 func badmcall(fn func(*g)) {
503 throw("runtime: mcall called on m->g0 stack")
506 func badmcall2(fn func(*g)) {
507 throw("runtime: mcall function returned")
510 func badreflectcall() {
511 panic(plainError("arg size to reflect.call more than 1GB"))
515 //go:nowritebarrierrec
516 func badmorestackg0() {
517 writeErrStr("fatal: morestack on g0\n")
521 //go:nowritebarrierrec
522 func badmorestackgsignal() {
523 writeErrStr("fatal: morestack on gsignal\n")
531 func lockedOSThread() bool {
533 return gp.lockedm != 0 && gp.m.lockedg != 0
537 // allgs contains all Gs ever created (including dead Gs), and thus
540 // Access via the slice is protected by allglock or stop-the-world.
541 // Readers that cannot take the lock may (carefully!) use the atomic
546 // allglen and allgptr are atomic variables that contain len(allgs) and
547 // &allgs[0] respectively. Proper ordering depends on totally-ordered
548 // loads and stores. Writes are protected by allglock.
550 // allgptr is updated before allglen. Readers should read allglen
551 // before allgptr to ensure that allglen is always <= len(allgptr). New
552 // Gs appended during the race can be missed. For a consistent view of
553 // all Gs, allglock must be held.
555 // allgptr copies should always be stored as a concrete type or
556 // unsafe.Pointer, not uintptr, to ensure that GC can still reach it
557 // even if it points to a stale array.
562 func allgadd(gp *g) {
563 if readgstatus(gp) == _Gidle {
564 throw("allgadd: bad status Gidle")
568 allgs = append(allgs, gp)
569 if &allgs[0] != allgptr {
570 atomicstorep(unsafe.Pointer(&allgptr), unsafe.Pointer(&allgs[0]))
572 atomic.Storeuintptr(&allglen, uintptr(len(allgs)))
576 // allGsSnapshot returns a snapshot of the slice of all Gs.
578 // The world must be stopped or allglock must be held.
579 func allGsSnapshot() []*g {
580 assertWorldStoppedOrLockHeld(&allglock)
582 // Because the world is stopped or allglock is held, allgadd
583 // cannot happen concurrently with this. allgs grows
584 // monotonically and existing entries never change, so we can
585 // simply return a copy of the slice header. For added safety,
586 // we trim everything past len because that can still change.
587 return allgs[:len(allgs):len(allgs)]
590 // atomicAllG returns &allgs[0] and len(allgs) for use with atomicAllGIndex.
591 func atomicAllG() (**g, uintptr) {
592 length := atomic.Loaduintptr(&allglen)
593 ptr := (**g)(atomic.Loadp(unsafe.Pointer(&allgptr)))
597 // atomicAllGIndex returns ptr[i] with the allgptr returned from atomicAllG.
598 func atomicAllGIndex(ptr **g, i uintptr) *g {
599 return *(**g)(add(unsafe.Pointer(ptr), i*goarch.PtrSize))
602 // forEachG calls fn on every G from allgs.
604 // forEachG takes a lock to exclude concurrent addition of new Gs.
605 func forEachG(fn func(gp *g)) {
607 for _, gp := range allgs {
613 // forEachGRace calls fn on every G from allgs.
615 // forEachGRace avoids locking, but does not exclude addition of new Gs during
616 // execution, which may be missed.
617 func forEachGRace(fn func(gp *g)) {
618 ptr, length := atomicAllG()
619 for i := uintptr(0); i < length; i++ {
620 gp := atomicAllGIndex(ptr, i)
627 // Number of goroutine ids to grab from sched.goidgen to local per-P cache at once.
628 // 16 seems to provide enough amortization, but other than that it's mostly arbitrary number.
632 // cpuinit sets up CPU feature flags and calls internal/cpu.Initialize. env should be the complete
633 // value of the GODEBUG environment variable.
634 func cpuinit(env string) {
636 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
637 cpu.DebugOptions = true
641 // Support cpu feature variables are used in code generated by the compiler
642 // to guard execution of instructions that can not be assumed to be always supported.
645 x86HasPOPCNT = cpu.X86.HasPOPCNT
646 x86HasSSE41 = cpu.X86.HasSSE41
647 x86HasFMA = cpu.X86.HasFMA
650 armHasVFPv4 = cpu.ARM.HasVFPv4
653 arm64HasATOMICS = cpu.ARM64.HasATOMICS
657 // getGodebugEarly extracts the environment variable GODEBUG from the environment on
658 // Unix-like operating systems and returns it. This function exists to extract GODEBUG
659 // early before much of the runtime is initialized.
660 func getGodebugEarly() string {
661 const prefix = "GODEBUG="
664 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
665 // Similar to goenv_unix but extracts the environment value for
667 // TODO(moehrmann): remove when general goenvs() can be called before cpuinit()
669 for argv_index(argv, argc+1+n) != nil {
673 for i := int32(0); i < n; i++ {
674 p := argv_index(argv, argc+1+i)
675 s := unsafe.String(p, findnull(p))
677 if hasPrefix(s, prefix) {
678 env = gostring(p)[len(prefix):]
686 // The bootstrap sequence is:
690 // make & queue new G
691 // call runtime·mstart
693 // The new G calls runtime·main.
695 lockInit(&sched.lock, lockRankSched)
696 lockInit(&sched.sysmonlock, lockRankSysmon)
697 lockInit(&sched.deferlock, lockRankDefer)
698 lockInit(&sched.sudoglock, lockRankSudog)
699 lockInit(&deadlock, lockRankDeadlock)
700 lockInit(&paniclk, lockRankPanic)
701 lockInit(&allglock, lockRankAllg)
702 lockInit(&allpLock, lockRankAllp)
703 lockInit(&reflectOffs.lock, lockRankReflectOffs)
704 lockInit(&finlock, lockRankFin)
705 lockInit(&cpuprof.lock, lockRankCpuprof)
707 // Enforce that this lock is always a leaf lock.
708 // All of this lock's critical sections should be
710 lockInit(&memstats.heapStats.noPLock, lockRankLeafRank)
712 // raceinit must be the first call to race detector.
713 // In particular, it must be done before mallocinit below calls racemapshadow.
716 gp.racectx, raceprocctx0 = raceinit()
719 sched.maxmcount = 10000
721 // The world starts stopped.
727 godebug := getGodebugEarly()
728 initPageTrace(godebug) // must run after mallocinit but before anything allocates
729 cpuinit(godebug) // must run before alginit
730 alginit() // maps, hash, fastrand must not be used before this call
731 fastrandinit() // must run before mcommoninit
732 mcommoninit(gp.m, -1)
733 modulesinit() // provides activeModules
734 typelinksinit() // uses maps, activeModules
735 itabsinit() // uses activeModules
736 stkobjinit() // must run before GC starts
738 sigsave(&gp.m.sigmask)
739 initSigmask = gp.m.sigmask
748 // if disableMemoryProfiling is set, update MemProfileRate to 0 to turn off memprofile.
749 // Note: parsedebugvars may update MemProfileRate, but when disableMemoryProfiling is
750 // set to true by the linker, it means that nothing is consuming the profile, it is
751 // safe to set MemProfileRate to 0.
752 if disableMemoryProfiling {
757 sched.lastpoll.Store(nanotime())
759 if n, ok := atoi32(gogetenv("GOMAXPROCS")); ok && n > 0 {
762 if procresize(procs) != nil {
763 throw("unknown runnable goroutine during bootstrap")
767 // World is effectively started now, as P's can run.
770 if buildVersion == "" {
771 // Condition should never trigger. This code just serves
772 // to ensure runtime·buildVersion is kept in the resulting binary.
773 buildVersion = "unknown"
775 if len(modinfo) == 1 {
776 // Condition should never trigger. This code just serves
777 // to ensure runtime·modinfo is kept in the resulting binary.
782 func dumpgstatus(gp *g) {
784 print("runtime: gp: gp=", gp, ", goid=", gp.goid, ", gp->atomicstatus=", readgstatus(gp), "\n")
785 print("runtime: getg: g=", thisg, ", goid=", thisg.goid, ", g->atomicstatus=", readgstatus(thisg), "\n")
788 // sched.lock must be held.
790 assertLockHeld(&sched.lock)
792 // Exclude extra M's, which are used for cgocallback from threads
795 // The purpose of the SetMaxThreads limit is to avoid accidental fork
796 // bomb from something like millions of goroutines blocking on system
797 // calls, causing the runtime to create millions of threads. By
798 // definition, this isn't a problem for threads created in C, so we
799 // exclude them from the limit. See https://go.dev/issue/60004.
800 count := mcount() - int32(extraMInUse.Load()) - int32(extraMLength.Load())
801 if count > sched.maxmcount {
802 print("runtime: program exceeds ", sched.maxmcount, "-thread limit\n")
803 throw("thread exhaustion")
807 // mReserveID returns the next ID to use for a new m. This new m is immediately
808 // considered 'running' by checkdead.
810 // sched.lock must be held.
811 func mReserveID() int64 {
812 assertLockHeld(&sched.lock)
814 if sched.mnext+1 < sched.mnext {
815 throw("runtime: thread ID overflow")
823 // Pre-allocated ID may be passed as 'id', or omitted by passing -1.
824 func mcommoninit(mp *m, id int64) {
827 // g0 stack won't make sense for user (and is not necessary unwindable).
829 callers(1, mp.createstack[:])
840 lo := uint32(int64Hash(uint64(mp.id), fastrandseed))
841 hi := uint32(int64Hash(uint64(cputicks()), ^fastrandseed))
845 // Same behavior as for 1.17.
846 // TODO: Simplify this.
847 if goarch.BigEndian {
848 mp.fastrand = uint64(lo)<<32 | uint64(hi)
850 mp.fastrand = uint64(hi)<<32 | uint64(lo)
854 if mp.gsignal != nil {
855 mp.gsignal.stackguard1 = mp.gsignal.stack.lo + stackGuard
858 // Add to allm so garbage collector doesn't free g->m
859 // when it is just in a register or thread-local storage.
862 // NumCgoCall() iterates over allm w/o schedlock,
863 // so we need to publish it safely.
864 atomicstorep(unsafe.Pointer(&allm), unsafe.Pointer(mp))
867 // Allocate memory to hold a cgo traceback if the cgo call crashes.
868 if iscgo || GOOS == "solaris" || GOOS == "illumos" || GOOS == "windows" {
869 mp.cgoCallers = new(cgoCallers)
873 func (mp *m) becomeSpinning() {
875 sched.nmspinning.Add(1)
876 sched.needspinning.Store(0)
879 func (mp *m) hasCgoOnStack() bool {
880 return mp.ncgo > 0 || mp.isextra
883 var fastrandseed uintptr
885 func fastrandinit() {
886 s := (*[unsafe.Sizeof(fastrandseed)]byte)(unsafe.Pointer(&fastrandseed))[:]
890 // Mark gp ready to run.
891 func ready(gp *g, traceskip int, next bool) {
893 traceGoUnpark(gp, traceskip)
896 status := readgstatus(gp)
899 mp := acquirem() // disable preemption because it can be holding p in a local var
900 if status&^_Gscan != _Gwaiting {
902 throw("bad g->status in ready")
905 // status is Gwaiting or Gscanwaiting, make Grunnable and put on runq
906 casgstatus(gp, _Gwaiting, _Grunnable)
907 runqput(mp.p.ptr(), gp, next)
912 // freezeStopWait is a large value that freezetheworld sets
913 // sched.stopwait to in order to request that all Gs permanently stop.
914 const freezeStopWait = 0x7fffffff
916 // freezing is set to non-zero if the runtime is trying to freeze the
918 var freezing atomic.Bool
920 // Similar to stopTheWorld but best-effort and can be called several times.
921 // There is no reverse operation, used during crashing.
922 // This function must not lock any mutexes.
923 func freezetheworld() {
925 if debug.dontfreezetheworld > 0 {
926 // Don't prempt Ps to stop goroutines. That will perturb
927 // scheduler state, making debugging more difficult. Instead,
928 // allow goroutines to continue execution.
930 // fatalpanic will tracebackothers to trace all goroutines. It
931 // is unsafe to trace a running goroutine, so tracebackothers
932 // will skip running goroutines. That is OK and expected, we
933 // expect users of dontfreezetheworld to use core files anyway.
935 // However, allowing the scheduler to continue running free
936 // introduces a race: a goroutine may be stopped when
937 // tracebackothers checks its status, and then start running
938 // later when we are in the middle of traceback, potentially
941 // To mitigate this, when an M naturally enters the scheduler,
942 // schedule checks if freezing is set and if so stops
943 // execution. This guarantees that while Gs can transition from
944 // running to stopped, they can never transition from stopped
947 // The sleep here allows racing Ms that missed freezing and are
948 // about to run a G to complete the transition to running
949 // before we start traceback.
954 // stopwait and preemption requests can be lost
955 // due to races with concurrently executing threads,
956 // so try several times
957 for i := 0; i < 5; i++ {
958 // this should tell the scheduler to not start any new goroutines
959 sched.stopwait = freezeStopWait
960 sched.gcwaiting.Store(true)
961 // this should stop running goroutines
963 break // no running goroutines
973 // All reads and writes of g's status go through readgstatus, casgstatus
974 // castogscanstatus, casfrom_Gscanstatus.
977 func readgstatus(gp *g) uint32 {
978 return gp.atomicstatus.Load()
981 // The Gscanstatuses are acting like locks and this releases them.
982 // If it proves to be a performance hit we should be able to make these
983 // simple atomic stores but for now we are going to throw if
984 // we see an inconsistent state.
985 func casfrom_Gscanstatus(gp *g, oldval, newval uint32) {
988 // Check that transition is valid.
991 print("runtime: casfrom_Gscanstatus bad oldval gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
993 throw("casfrom_Gscanstatus:top gp->status is not in scan state")
999 if newval == oldval&^_Gscan {
1000 success = gp.atomicstatus.CompareAndSwap(oldval, newval)
1004 print("runtime: casfrom_Gscanstatus failed gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
1006 throw("casfrom_Gscanstatus: gp->status is not in scan state")
1008 releaseLockRank(lockRankGscan)
1011 // This will return false if the gp is not in the expected status and the cas fails.
1012 // This acts like a lock acquire while the casfromgstatus acts like a lock release.
1013 func castogscanstatus(gp *g, oldval, newval uint32) bool {
1019 if newval == oldval|_Gscan {
1020 r := gp.atomicstatus.CompareAndSwap(oldval, newval)
1022 acquireLockRank(lockRankGscan)
1028 print("runtime: castogscanstatus oldval=", hex(oldval), " newval=", hex(newval), "\n")
1029 throw("castogscanstatus")
1030 panic("not reached")
1033 // casgstatusAlwaysTrack is a debug flag that causes casgstatus to always track
1034 // various latencies on every transition instead of sampling them.
1035 var casgstatusAlwaysTrack = false
1037 // If asked to move to or from a Gscanstatus this will throw. Use the castogscanstatus
1038 // and casfrom_Gscanstatus instead.
1039 // casgstatus will loop if the g->atomicstatus is in a Gscan status until the routine that
1040 // put it in the Gscan state is finished.
1043 func casgstatus(gp *g, oldval, newval uint32) {
1044 if (oldval&_Gscan != 0) || (newval&_Gscan != 0) || oldval == newval {
1045 systemstack(func() {
1046 print("runtime: casgstatus: oldval=", hex(oldval), " newval=", hex(newval), "\n")
1047 throw("casgstatus: bad incoming values")
1051 acquireLockRank(lockRankGscan)
1052 releaseLockRank(lockRankGscan)
1054 // See https://golang.org/cl/21503 for justification of the yield delay.
1055 const yieldDelay = 5 * 1000
1058 // loop if gp->atomicstatus is in a scan state giving
1059 // GC time to finish and change the state to oldval.
1060 for i := 0; !gp.atomicstatus.CompareAndSwap(oldval, newval); i++ {
1061 if oldval == _Gwaiting && gp.atomicstatus.Load() == _Grunnable {
1062 throw("casgstatus: waiting for Gwaiting but is Grunnable")
1065 nextYield = nanotime() + yieldDelay
1067 if nanotime() < nextYield {
1068 for x := 0; x < 10 && gp.atomicstatus.Load() != oldval; x++ {
1073 nextYield = nanotime() + yieldDelay/2
1077 if oldval == _Grunning {
1078 // Track every gTrackingPeriod time a goroutine transitions out of running.
1079 if casgstatusAlwaysTrack || gp.trackingSeq%gTrackingPeriod == 0 {
1088 // Handle various kinds of tracking.
1091 // - Time spent in runnable.
1092 // - Time spent blocked on a sync.Mutex or sync.RWMutex.
1095 // We transitioned out of runnable, so measure how much
1096 // time we spent in this state and add it to
1099 gp.runnableTime += now - gp.trackingStamp
1100 gp.trackingStamp = 0
1102 if !gp.waitreason.isMutexWait() {
1103 // Not blocking on a lock.
1106 // Blocking on a lock, measure it. Note that because we're
1107 // sampling, we have to multiply by our sampling period to get
1108 // a more representative estimate of the absolute value.
1109 // gTrackingPeriod also represents an accurate sampling period
1110 // because we can only enter this state from _Grunning.
1112 sched.totalMutexWaitTime.Add((now - gp.trackingStamp) * gTrackingPeriod)
1113 gp.trackingStamp = 0
1117 if !gp.waitreason.isMutexWait() {
1118 // Not blocking on a lock.
1121 // Blocking on a lock. Write down the timestamp.
1123 gp.trackingStamp = now
1125 // We just transitioned into runnable, so record what
1126 // time that happened.
1128 gp.trackingStamp = now
1130 // We're transitioning into running, so turn off
1131 // tracking and record how much time we spent in
1134 sched.timeToRun.record(gp.runnableTime)
1139 // casGToWaiting transitions gp from old to _Gwaiting, and sets the wait reason.
1141 // Use this over casgstatus when possible to ensure that a waitreason is set.
1142 func casGToWaiting(gp *g, old uint32, reason waitReason) {
1143 // Set the wait reason before calling casgstatus, because casgstatus will use it.
1144 gp.waitreason = reason
1145 casgstatus(gp, old, _Gwaiting)
1148 // casgstatus(gp, oldstatus, Gcopystack), assuming oldstatus is Gwaiting or Grunnable.
1149 // Returns old status. Cannot call casgstatus directly, because we are racing with an
1150 // async wakeup that might come in from netpoll. If we see Gwaiting from the readgstatus,
1151 // it might have become Grunnable by the time we get to the cas. If we called casgstatus,
1152 // it would loop waiting for the status to go back to Gwaiting, which it never will.
1155 func casgcopystack(gp *g) uint32 {
1157 oldstatus := readgstatus(gp) &^ _Gscan
1158 if oldstatus != _Gwaiting && oldstatus != _Grunnable {
1159 throw("copystack: bad status, not Gwaiting or Grunnable")
1161 if gp.atomicstatus.CompareAndSwap(oldstatus, _Gcopystack) {
1167 // casGToPreemptScan transitions gp from _Grunning to _Gscan|_Gpreempted.
1169 // TODO(austin): This is the only status operation that both changes
1170 // the status and locks the _Gscan bit. Rethink this.
1171 func casGToPreemptScan(gp *g, old, new uint32) {
1172 if old != _Grunning || new != _Gscan|_Gpreempted {
1173 throw("bad g transition")
1175 acquireLockRank(lockRankGscan)
1176 for !gp.atomicstatus.CompareAndSwap(_Grunning, _Gscan|_Gpreempted) {
1180 // casGFromPreempted attempts to transition gp from _Gpreempted to
1181 // _Gwaiting. If successful, the caller is responsible for
1182 // re-scheduling gp.
1183 func casGFromPreempted(gp *g, old, new uint32) bool {
1184 if old != _Gpreempted || new != _Gwaiting {
1185 throw("bad g transition")
1187 gp.waitreason = waitReasonPreempted
1188 return gp.atomicstatus.CompareAndSwap(_Gpreempted, _Gwaiting)
1191 // stwReason is an enumeration of reasons the world is stopping.
1192 type stwReason uint8
1194 // Reasons to stop-the-world.
1196 // Avoid reusing reasons and add new ones instead.
1198 stwUnknown stwReason = iota // "unknown"
1199 stwGCMarkTerm // "GC mark termination"
1200 stwGCSweepTerm // "GC sweep termination"
1201 stwWriteHeapDump // "write heap dump"
1202 stwGoroutineProfile // "goroutine profile"
1203 stwGoroutineProfileCleanup // "goroutine profile cleanup"
1204 stwAllGoroutinesStack // "all goroutines stack trace"
1205 stwReadMemStats // "read mem stats"
1206 stwAllThreadsSyscall // "AllThreadsSyscall"
1207 stwGOMAXPROCS // "GOMAXPROCS"
1208 stwStartTrace // "start trace"
1209 stwStopTrace // "stop trace"
1210 stwForTestCountPagesInUse // "CountPagesInUse (test)"
1211 stwForTestReadMetricsSlow // "ReadMetricsSlow (test)"
1212 stwForTestReadMemStatsSlow // "ReadMemStatsSlow (test)"
1213 stwForTestPageCachePagesLeaked // "PageCachePagesLeaked (test)"
1214 stwForTestResetDebugLog // "ResetDebugLog (test)"
1217 func (r stwReason) String() string {
1218 return stwReasonStrings[r]
1221 // If you add to this list, also add it to src/internal/trace/parser.go.
1222 // If you change the values of any of the stw* constants, bump the trace
1223 // version number and make a copy of this.
1224 var stwReasonStrings = [...]string{
1225 stwUnknown: "unknown",
1226 stwGCMarkTerm: "GC mark termination",
1227 stwGCSweepTerm: "GC sweep termination",
1228 stwWriteHeapDump: "write heap dump",
1229 stwGoroutineProfile: "goroutine profile",
1230 stwGoroutineProfileCleanup: "goroutine profile cleanup",
1231 stwAllGoroutinesStack: "all goroutines stack trace",
1232 stwReadMemStats: "read mem stats",
1233 stwAllThreadsSyscall: "AllThreadsSyscall",
1234 stwGOMAXPROCS: "GOMAXPROCS",
1235 stwStartTrace: "start trace",
1236 stwStopTrace: "stop trace",
1237 stwForTestCountPagesInUse: "CountPagesInUse (test)",
1238 stwForTestReadMetricsSlow: "ReadMetricsSlow (test)",
1239 stwForTestReadMemStatsSlow: "ReadMemStatsSlow (test)",
1240 stwForTestPageCachePagesLeaked: "PageCachePagesLeaked (test)",
1241 stwForTestResetDebugLog: "ResetDebugLog (test)",
1244 // stopTheWorld stops all P's from executing goroutines, interrupting
1245 // all goroutines at GC safe points and records reason as the reason
1246 // for the stop. On return, only the current goroutine's P is running.
1247 // stopTheWorld must not be called from a system stack and the caller
1248 // must not hold worldsema. The caller must call startTheWorld when
1249 // other P's should resume execution.
1251 // stopTheWorld is safe for multiple goroutines to call at the
1252 // same time. Each will execute its own stop, and the stops will
1255 // This is also used by routines that do stack dumps. If the system is
1256 // in panic or being exited, this may not reliably stop all
1258 func stopTheWorld(reason stwReason) {
1259 semacquire(&worldsema)
1261 gp.m.preemptoff = reason.String()
1262 systemstack(func() {
1263 // Mark the goroutine which called stopTheWorld preemptible so its
1264 // stack may be scanned.
1265 // This lets a mark worker scan us while we try to stop the world
1266 // since otherwise we could get in a mutual preemption deadlock.
1267 // We must not modify anything on the G stack because a stack shrink
1268 // may occur. A stack shrink is otherwise OK though because in order
1269 // to return from this function (and to leave the system stack) we
1270 // must have preempted all goroutines, including any attempting
1271 // to scan our stack, in which case, any stack shrinking will
1272 // have already completed by the time we exit.
1273 // Don't provide a wait reason because we're still executing.
1274 casGToWaiting(gp, _Grunning, waitReasonStoppingTheWorld)
1275 stopTheWorldWithSema(reason)
1276 casgstatus(gp, _Gwaiting, _Grunning)
1280 // startTheWorld undoes the effects of stopTheWorld.
1281 func startTheWorld() {
1282 systemstack(func() { startTheWorldWithSema() })
1284 // worldsema must be held over startTheWorldWithSema to ensure
1285 // gomaxprocs cannot change while worldsema is held.
1287 // Release worldsema with direct handoff to the next waiter, but
1288 // acquirem so that semrelease1 doesn't try to yield our time.
1290 // Otherwise if e.g. ReadMemStats is being called in a loop,
1291 // it might stomp on other attempts to stop the world, such as
1292 // for starting or ending GC. The operation this blocks is
1293 // so heavy-weight that we should just try to be as fair as
1296 // We don't want to just allow us to get preempted between now
1297 // and releasing the semaphore because then we keep everyone
1298 // (including, for example, GCs) waiting longer.
1301 semrelease1(&worldsema, true, 0)
1305 // stopTheWorldGC has the same effect as stopTheWorld, but blocks
1306 // until the GC is not running. It also blocks a GC from starting
1307 // until startTheWorldGC is called.
1308 func stopTheWorldGC(reason stwReason) {
1310 stopTheWorld(reason)
1313 // startTheWorldGC undoes the effects of stopTheWorldGC.
1314 func startTheWorldGC() {
1319 // Holding worldsema grants an M the right to try to stop the world.
1320 var worldsema uint32 = 1
1322 // Holding gcsema grants the M the right to block a GC, and blocks
1323 // until the current GC is done. In particular, it prevents gomaxprocs
1324 // from changing concurrently.
1326 // TODO(mknyszek): Once gomaxprocs and the execution tracer can handle
1327 // being changed/enabled during a GC, remove this.
1328 var gcsema uint32 = 1
1330 // stopTheWorldWithSema is the core implementation of stopTheWorld.
1331 // The caller is responsible for acquiring worldsema and disabling
1332 // preemption first and then should stopTheWorldWithSema on the system
1335 // semacquire(&worldsema, 0)
1336 // m.preemptoff = "reason"
1337 // systemstack(stopTheWorldWithSema)
1339 // When finished, the caller must either call startTheWorld or undo
1340 // these three operations separately:
1342 // m.preemptoff = ""
1343 // systemstack(startTheWorldWithSema)
1344 // semrelease(&worldsema)
1346 // It is allowed to acquire worldsema once and then execute multiple
1347 // startTheWorldWithSema/stopTheWorldWithSema pairs.
1348 // Other P's are able to execute between successive calls to
1349 // startTheWorldWithSema and stopTheWorldWithSema.
1350 // Holding worldsema causes any other goroutines invoking
1351 // stopTheWorld to block.
1352 func stopTheWorldWithSema(reason stwReason) {
1354 traceSTWStart(reason)
1358 // If we hold a lock, then we won't be able to stop another M
1359 // that is blocked trying to acquire the lock.
1361 throw("stopTheWorld: holding locks")
1365 sched.stopwait = gomaxprocs
1366 sched.gcwaiting.Store(true)
1369 gp.m.p.ptr().status = _Pgcstop // Pgcstop is only diagnostic.
1371 // try to retake all P's in Psyscall status
1372 for _, pp := range allp {
1374 if s == _Psyscall && atomic.Cas(&pp.status, s, _Pgcstop) {
1386 pp, _ := pidleget(now)
1390 pp.status = _Pgcstop
1393 wait := sched.stopwait > 0
1396 // wait for remaining P's to stop voluntarily
1399 // wait for 100us, then try to re-preempt in case of any races
1400 if notetsleep(&sched.stopnote, 100*1000) {
1401 noteclear(&sched.stopnote)
1410 if sched.stopwait != 0 {
1411 bad = "stopTheWorld: not stopped (stopwait != 0)"
1413 for _, pp := range allp {
1414 if pp.status != _Pgcstop {
1415 bad = "stopTheWorld: not stopped (status != _Pgcstop)"
1419 if freezing.Load() {
1420 // Some other thread is panicking. This can cause the
1421 // sanity checks above to fail if the panic happens in
1422 // the signal handler on a stopped thread. Either way,
1423 // we should halt this thread.
1434 func startTheWorldWithSema() int64 {
1435 assertWorldStopped()
1437 mp := acquirem() // disable preemption because it can be holding p in a local var
1438 if netpollinited() {
1439 list, delta := netpoll(0) // non-blocking
1441 netpollAdjustWaiters(delta)
1450 p1 := procresize(procs)
1451 sched.gcwaiting.Store(false)
1452 if sched.sysmonwait.Load() {
1453 sched.sysmonwait.Store(false)
1454 notewakeup(&sched.sysmonnote)
1467 throw("startTheWorld: inconsistent mp->nextp")
1470 notewakeup(&mp.park)
1472 // Start M to run P. Do not start another M below.
1477 // Capture start-the-world time before doing clean-up tasks.
1478 startTime := nanotime()
1483 // Wakeup an additional proc in case we have excessive runnable goroutines
1484 // in local queues or in the global queue. If we don't, the proc will park itself.
1485 // If we have lots of excessive work, resetspinning will unpark additional procs as necessary.
1493 // usesLibcall indicates whether this runtime performs system calls
1495 func usesLibcall() bool {
1497 case "aix", "darwin", "illumos", "ios", "solaris", "windows":
1500 return GOARCH == "386" || GOARCH == "amd64" || GOARCH == "arm" || GOARCH == "arm64"
1505 // mStackIsSystemAllocated indicates whether this runtime starts on a
1506 // system-allocated stack.
1507 func mStackIsSystemAllocated() bool {
1509 case "aix", "darwin", "plan9", "illumos", "ios", "solaris", "windows":
1513 case "386", "amd64", "arm", "arm64":
1520 // mstart is the entry-point for new Ms.
1521 // It is written in assembly, uses ABI0, is marked TOPFRAME, and calls mstart0.
1524 // mstart0 is the Go entry-point for new Ms.
1525 // This must not split the stack because we may not even have stack
1526 // bounds set up yet.
1528 // May run during STW (because it doesn't have a P yet), so write
1529 // barriers are not allowed.
1532 //go:nowritebarrierrec
1536 osStack := gp.stack.lo == 0
1538 // Initialize stack bounds from system stack.
1539 // Cgo may have left stack size in stack.hi.
1540 // minit may update the stack bounds.
1542 // Note: these bounds may not be very accurate.
1543 // We set hi to &size, but there are things above
1544 // it. The 1024 is supposed to compensate this,
1545 // but is somewhat arbitrary.
1548 size = 8192 * sys.StackGuardMultiplier
1550 gp.stack.hi = uintptr(noescape(unsafe.Pointer(&size)))
1551 gp.stack.lo = gp.stack.hi - size + 1024
1553 // Initialize stack guard so that we can start calling regular
1555 gp.stackguard0 = gp.stack.lo + stackGuard
1556 // This is the g0, so we can also call go:systemstack
1557 // functions, which check stackguard1.
1558 gp.stackguard1 = gp.stackguard0
1561 // Exit this thread.
1562 if mStackIsSystemAllocated() {
1563 // Windows, Solaris, illumos, Darwin, AIX and Plan 9 always system-allocate
1564 // the stack, but put it in gp.stack before mstart,
1565 // so the logic above hasn't set osStack yet.
1571 // The go:noinline is to guarantee the getcallerpc/getcallersp below are safe,
1572 // so that we can set up g0.sched to return to the call of mstart1 above.
1579 throw("bad runtime·mstart")
1582 // Set up m.g0.sched as a label returning to just
1583 // after the mstart1 call in mstart0 above, for use by goexit0 and mcall.
1584 // We're never coming back to mstart1 after we call schedule,
1585 // so other calls can reuse the current frame.
1586 // And goexit0 does a gogo that needs to return from mstart1
1587 // and let mstart0 exit the thread.
1588 gp.sched.g = guintptr(unsafe.Pointer(gp))
1589 gp.sched.pc = getcallerpc()
1590 gp.sched.sp = getcallersp()
1595 // Install signal handlers; after minit so that minit can
1596 // prepare the thread to be able to handle the signals.
1601 if fn := gp.m.mstartfn; fn != nil {
1606 acquirep(gp.m.nextp.ptr())
1612 // mstartm0 implements part of mstart1 that only runs on the m0.
1614 // Write barriers are allowed here because we know the GC can't be
1615 // running yet, so they'll be no-ops.
1617 //go:yeswritebarrierrec
1619 // Create an extra M for callbacks on threads not created by Go.
1620 // An extra M is also needed on Windows for callbacks created by
1621 // syscall.NewCallback. See issue #6751 for details.
1622 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
1629 // mPark causes a thread to park itself, returning once woken.
1634 notesleep(&gp.m.park)
1635 noteclear(&gp.m.park)
1638 // mexit tears down and exits the current thread.
1640 // Don't call this directly to exit the thread, since it must run at
1641 // the top of the thread stack. Instead, use gogo(&gp.m.g0.sched) to
1642 // unwind the stack to the point that exits the thread.
1644 // It is entered with m.p != nil, so write barriers are allowed. It
1645 // will release the P before exiting.
1647 //go:yeswritebarrierrec
1648 func mexit(osStack bool) {
1652 // This is the main thread. Just wedge it.
1654 // On Linux, exiting the main thread puts the process
1655 // into a non-waitable zombie state. On Plan 9,
1656 // exiting the main thread unblocks wait even though
1657 // other threads are still running. On Solaris we can
1658 // neither exitThread nor return from mstart. Other
1659 // bad things probably happen on other platforms.
1661 // We could try to clean up this M more before wedging
1662 // it, but that complicates signal handling.
1663 handoffp(releasep())
1669 throw("locked m0 woke up")
1675 // Free the gsignal stack.
1676 if mp.gsignal != nil {
1677 stackfree(mp.gsignal.stack)
1678 // On some platforms, when calling into VDSO (e.g. nanotime)
1679 // we store our g on the gsignal stack, if there is one.
1680 // Now the stack is freed, unlink it from the m, so we
1681 // won't write to it when calling VDSO code.
1685 // Remove m from allm.
1687 for pprev := &allm; *pprev != nil; pprev = &(*pprev).alllink {
1693 throw("m not found in allm")
1695 // Delay reaping m until it's done with the stack.
1697 // Put mp on the free list, though it will not be reaped while freeWait
1698 // is freeMWait. mp is no longer reachable via allm, so even if it is
1699 // on an OS stack, we must keep a reference to mp alive so that the GC
1700 // doesn't free mp while we are still using it.
1702 // Note that the free list must not be linked through alllink because
1703 // some functions walk allm without locking, so may be using alllink.
1704 mp.freeWait.Store(freeMWait)
1705 mp.freelink = sched.freem
1709 atomic.Xadd64(&ncgocall, int64(mp.ncgocall))
1712 handoffp(releasep())
1713 // After this point we must not have write barriers.
1715 // Invoke the deadlock detector. This must happen after
1716 // handoffp because it may have started a new M to take our
1723 if GOOS == "darwin" || GOOS == "ios" {
1724 // Make sure pendingPreemptSignals is correct when an M exits.
1726 if mp.signalPending.Load() != 0 {
1727 pendingPreemptSignals.Add(-1)
1731 // Destroy all allocated resources. After this is called, we may no
1732 // longer take any locks.
1736 // No more uses of mp, so it is safe to drop the reference.
1737 mp.freeWait.Store(freeMRef)
1739 // Return from mstart and let the system thread
1740 // library free the g0 stack and terminate the thread.
1744 // mstart is the thread's entry point, so there's nothing to
1745 // return to. Exit the thread directly. exitThread will clear
1746 // m.freeWait when it's done with the stack and the m can be
1748 exitThread(&mp.freeWait)
1751 // forEachP calls fn(p) for every P p when p reaches a GC safe point.
1752 // If a P is currently executing code, this will bring the P to a GC
1753 // safe point and execute fn on that P. If the P is not executing code
1754 // (it is idle or in a syscall), this will call fn(p) directly while
1755 // preventing the P from exiting its state. This does not ensure that
1756 // fn will run on every CPU executing Go code, but it acts as a global
1757 // memory barrier. GC uses this as a "ragged barrier."
1759 // The caller must hold worldsema.
1762 func forEachP(fn func(*p)) {
1764 pp := getg().m.p.ptr()
1767 if sched.safePointWait != 0 {
1768 throw("forEachP: sched.safePointWait != 0")
1770 sched.safePointWait = gomaxprocs - 1
1771 sched.safePointFn = fn
1773 // Ask all Ps to run the safe point function.
1774 for _, p2 := range allp {
1776 atomic.Store(&p2.runSafePointFn, 1)
1781 // Any P entering _Pidle or _Psyscall from now on will observe
1782 // p.runSafePointFn == 1 and will call runSafePointFn when
1783 // changing its status to _Pidle/_Psyscall.
1785 // Run safe point function for all idle Ps. sched.pidle will
1786 // not change because we hold sched.lock.
1787 for p := sched.pidle.ptr(); p != nil; p = p.link.ptr() {
1788 if atomic.Cas(&p.runSafePointFn, 1, 0) {
1790 sched.safePointWait--
1794 wait := sched.safePointWait > 0
1797 // Run fn for the current P.
1800 // Force Ps currently in _Psyscall into _Pidle and hand them
1801 // off to induce safe point function execution.
1802 for _, p2 := range allp {
1804 if s == _Psyscall && p2.runSafePointFn == 1 && atomic.Cas(&p2.status, s, _Pidle) {
1814 // Wait for remaining Ps to run fn.
1817 // Wait for 100us, then try to re-preempt in
1818 // case of any races.
1820 // Requires system stack.
1821 if notetsleep(&sched.safePointNote, 100*1000) {
1822 noteclear(&sched.safePointNote)
1828 if sched.safePointWait != 0 {
1829 throw("forEachP: not done")
1831 for _, p2 := range allp {
1832 if p2.runSafePointFn != 0 {
1833 throw("forEachP: P did not run fn")
1838 sched.safePointFn = nil
1843 // runSafePointFn runs the safe point function, if any, for this P.
1844 // This should be called like
1846 // if getg().m.p.runSafePointFn != 0 {
1850 // runSafePointFn must be checked on any transition in to _Pidle or
1851 // _Psyscall to avoid a race where forEachP sees that the P is running
1852 // just before the P goes into _Pidle/_Psyscall and neither forEachP
1853 // nor the P run the safe-point function.
1854 func runSafePointFn() {
1855 p := getg().m.p.ptr()
1856 // Resolve the race between forEachP running the safe-point
1857 // function on this P's behalf and this P running the
1858 // safe-point function directly.
1859 if !atomic.Cas(&p.runSafePointFn, 1, 0) {
1862 sched.safePointFn(p)
1864 sched.safePointWait--
1865 if sched.safePointWait == 0 {
1866 notewakeup(&sched.safePointNote)
1871 // When running with cgo, we call _cgo_thread_start
1872 // to start threads for us so that we can play nicely with
1874 var cgoThreadStart unsafe.Pointer
1876 type cgothreadstart struct {
1882 // Allocate a new m unassociated with any thread.
1883 // Can use p for allocation context if needed.
1884 // fn is recorded as the new m's m.mstartfn.
1885 // id is optional pre-allocated m ID. Omit by passing -1.
1887 // This function is allowed to have write barriers even if the caller
1888 // isn't because it borrows pp.
1890 //go:yeswritebarrierrec
1891 func allocm(pp *p, fn func(), id int64) *m {
1894 // The caller owns pp, but we may borrow (i.e., acquirep) it. We must
1895 // disable preemption to ensure it is not stolen, which would make the
1896 // caller lose ownership.
1901 acquirep(pp) // temporarily borrow p for mallocs in this function
1904 // Release the free M list. We need to do this somewhere and
1905 // this may free up a stack we can use.
1906 if sched.freem != nil {
1909 for freem := sched.freem; freem != nil; {
1910 wait := freem.freeWait.Load()
1911 if wait == freeMWait {
1912 next := freem.freelink
1913 freem.freelink = newList
1918 // Free the stack if needed. For freeMRef, there is
1919 // nothing to do except drop freem from the sched.freem
1921 if wait == freeMStack {
1922 // stackfree must be on the system stack, but allocm is
1923 // reachable off the system stack transitively from
1925 systemstack(func() {
1926 stackfree(freem.g0.stack)
1929 freem = freem.freelink
1931 sched.freem = newList
1939 // In case of cgo or Solaris or illumos or Darwin, pthread_create will make us a stack.
1940 // Windows and Plan 9 will layout sched stack on OS stack.
1941 if iscgo || mStackIsSystemAllocated() {
1944 mp.g0 = malg(8192 * sys.StackGuardMultiplier)
1948 if pp == gp.m.p.ptr() {
1953 allocmLock.runlock()
1957 // needm is called when a cgo callback happens on a
1958 // thread without an m (a thread not created by Go).
1959 // In this case, needm is expected to find an m to use
1960 // and return with m, g initialized correctly.
1961 // Since m and g are not set now (likely nil, but see below)
1962 // needm is limited in what routines it can call. In particular
1963 // it can only call nosplit functions (textflag 7) and cannot
1964 // do any scheduling that requires an m.
1966 // In order to avoid needing heavy lifting here, we adopt
1967 // the following strategy: there is a stack of available m's
1968 // that can be stolen. Using compare-and-swap
1969 // to pop from the stack has ABA races, so we simulate
1970 // a lock by doing an exchange (via Casuintptr) to steal the stack
1971 // head and replace the top pointer with MLOCKED (1).
1972 // This serves as a simple spin lock that we can use even
1973 // without an m. The thread that locks the stack in this way
1974 // unlocks the stack by storing a valid stack head pointer.
1976 // In order to make sure that there is always an m structure
1977 // available to be stolen, we maintain the invariant that there
1978 // is always one more than needed. At the beginning of the
1979 // program (if cgo is in use) the list is seeded with a single m.
1980 // If needm finds that it has taken the last m off the list, its job
1981 // is - once it has installed its own m so that it can do things like
1982 // allocate memory - to create a spare m and put it on the list.
1984 // Each of these extra m's also has a g0 and a curg that are
1985 // pressed into service as the scheduling stack and current
1986 // goroutine for the duration of the cgo callback.
1988 // It calls dropm to put the m back on the list,
1989 // 1. when the callback is done with the m in non-pthread platforms,
1990 // 2. or when the C thread exiting on pthread platforms.
1992 // The signal argument indicates whether we're called from a signal
1996 func needm(signal bool) {
1997 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
1998 // Can happen if C/C++ code calls Go from a global ctor.
1999 // Can also happen on Windows if a global ctor uses a
2000 // callback created by syscall.NewCallback. See issue #6751
2003 // Can not throw, because scheduler is not initialized yet.
2004 writeErrStr("fatal error: cgo callback before cgo call\n")
2008 // Save and block signals before getting an M.
2009 // The signal handler may call needm itself,
2010 // and we must avoid a deadlock. Also, once g is installed,
2011 // any incoming signals will try to execute,
2012 // but we won't have the sigaltstack settings and other data
2013 // set up appropriately until the end of minit, which will
2014 // unblock the signals. This is the same dance as when
2015 // starting a new m to run Go code via newosproc.
2020 // getExtraM is safe here because of the invariant above,
2021 // that the extra list always contains or will soon contain
2023 mp, last := getExtraM()
2025 // Set needextram when we've just emptied the list,
2026 // so that the eventual call into cgocallbackg will
2027 // allocate a new m for the extra list. We delay the
2028 // allocation until then so that it can be done
2029 // after exitsyscall makes sure it is okay to be
2030 // running at all (that is, there's no garbage collection
2031 // running right now).
2032 mp.needextram = last
2034 // Store the original signal mask for use by minit.
2035 mp.sigmask = sigmask
2037 // Install TLS on some platforms (previously setg
2038 // would do this if necessary).
2041 // Install g (= m->g0) and set the stack bounds
2042 // to match the current stack. If we don't actually know
2043 // how big the stack is, like we don't know how big any
2044 // scheduling stack is, but we assume there's at least 32 kB.
2045 // If we can get a more accurate stack bound from pthread,
2049 gp.stack.hi = getcallersp() + 1024
2050 gp.stack.lo = getcallersp() - 32*1024
2051 if !signal && _cgo_getstackbound != nil {
2052 // Don't adjust if called from the signal handler.
2053 // We are on the signal stack, not the pthread stack.
2054 // (We could get the stack bounds from sigaltstack, but
2055 // we're getting out of the signal handler very soon
2056 // anyway. Not worth it.)
2057 var bounds [2]uintptr
2058 asmcgocall(_cgo_getstackbound, unsafe.Pointer(&bounds))
2059 // getstackbound is an unsupported no-op on Windows.
2061 gp.stack.lo = bounds[0]
2062 gp.stack.hi = bounds[1]
2065 gp.stackguard0 = gp.stack.lo + stackGuard
2067 // Should mark we are already in Go now.
2068 // Otherwise, we may call needm again when we get a signal, before cgocallbackg1,
2069 // which means the extram list may be empty, that will cause a deadlock.
2070 mp.isExtraInC = false
2072 // Initialize this thread to use the m.
2076 // mp.curg is now a real goroutine.
2077 casgstatus(mp.curg, _Gdead, _Gsyscall)
2081 // Acquire an extra m and bind it to the C thread when a pthread key has been created.
2084 func needAndBindM() {
2087 if _cgo_pthread_key_created != nil && *(*uintptr)(_cgo_pthread_key_created) != 0 {
2092 // newextram allocates m's and puts them on the extra list.
2093 // It is called with a working local m, so that it can do things
2094 // like call schedlock and allocate.
2096 c := extraMWaiters.Swap(0)
2098 for i := uint32(0); i < c; i++ {
2101 } else if extraMLength.Load() == 0 {
2102 // Make sure there is at least one extra M.
2107 // oneNewExtraM allocates an m and puts it on the extra list.
2108 func oneNewExtraM() {
2109 // Create extra goroutine locked to extra m.
2110 // The goroutine is the context in which the cgo callback will run.
2111 // The sched.pc will never be returned to, but setting it to
2112 // goexit makes clear to the traceback routines where
2113 // the goroutine stack ends.
2114 mp := allocm(nil, nil, -1)
2116 gp.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum
2117 gp.sched.sp = gp.stack.hi
2118 gp.sched.sp -= 4 * goarch.PtrSize // extra space in case of reads slightly beyond frame
2120 gp.sched.g = guintptr(unsafe.Pointer(gp))
2121 gp.syscallpc = gp.sched.pc
2122 gp.syscallsp = gp.sched.sp
2123 gp.stktopsp = gp.sched.sp
2124 // malg returns status as _Gidle. Change to _Gdead before
2125 // adding to allg where GC can see it. We use _Gdead to hide
2126 // this from tracebacks and stack scans since it isn't a
2127 // "real" goroutine until needm grabs it.
2128 casgstatus(gp, _Gidle, _Gdead)
2132 // mark we are in C by default.
2133 mp.isExtraInC = true
2137 gp.goid = sched.goidgen.Add(1)
2139 gp.racectx = racegostart(abi.FuncPCABIInternal(newextram) + sys.PCQuantum)
2142 traceOneNewExtraM(gp)
2144 // put on allg for garbage collector
2147 // gp is now on the allg list, but we don't want it to be
2148 // counted by gcount. It would be more "proper" to increment
2149 // sched.ngfree, but that requires locking. Incrementing ngsys
2150 // has the same effect.
2153 // Add m to the extra list.
2157 // dropm puts the current m back onto the extra list.
2159 // 1. On systems without pthreads, like Windows
2160 // dropm is called when a cgo callback has called needm but is now
2161 // done with the callback and returning back into the non-Go thread.
2163 // The main expense here is the call to signalstack to release the
2164 // m's signal stack, and then the call to needm on the next callback
2165 // from this thread. It is tempting to try to save the m for next time,
2166 // which would eliminate both these costs, but there might not be
2167 // a next time: the current thread (which Go does not control) might exit.
2168 // If we saved the m for that thread, there would be an m leak each time
2169 // such a thread exited. Instead, we acquire and release an m on each
2170 // call. These should typically not be scheduling operations, just a few
2171 // atomics, so the cost should be small.
2173 // 2. On systems with pthreads
2174 // dropm is called while a non-Go thread is exiting.
2175 // We allocate a pthread per-thread variable using pthread_key_create,
2176 // to register a thread-exit-time destructor.
2177 // And store the g into a thread-specific value associated with the pthread key,
2178 // when first return back to C.
2179 // So that the destructor would invoke dropm while the non-Go thread is exiting.
2180 // This is much faster since it avoids expensive signal-related syscalls.
2182 // NOTE: this always runs without a P, so, nowritebarrierrec required.
2184 //go:nowritebarrierrec
2186 // Clear m and g, and return m to the extra list.
2187 // After the call to setg we can only call nosplit functions
2188 // with no pointer manipulation.
2191 // Return mp.curg to dead state.
2192 casgstatus(mp.curg, _Gsyscall, _Gdead)
2193 mp.curg.preemptStop = false
2196 // Block signals before unminit.
2197 // Unminit unregisters the signal handling stack (but needs g on some systems).
2198 // Setg(nil) clears g, which is the signal handler's cue not to run Go handlers.
2199 // It's important not to try to handle a signal between those two steps.
2200 sigmask := mp.sigmask
2208 msigrestore(sigmask)
2211 // bindm store the g0 of the current m into a thread-specific value.
2213 // We allocate a pthread per-thread variable using pthread_key_create,
2214 // to register a thread-exit-time destructor.
2215 // We are here setting the thread-specific value of the pthread key, to enable the destructor.
2216 // So that the pthread_key_destructor would dropm while the C thread is exiting.
2218 // And the saved g will be used in pthread_key_destructor,
2219 // since the g stored in the TLS by Go might be cleared in some platforms,
2220 // before the destructor invoked, so, we restore g by the stored g, before dropm.
2222 // We store g0 instead of m, to make the assembly code simpler,
2223 // since we need to restore g0 in runtime.cgocallback.
2225 // On systems without pthreads, like Windows, bindm shouldn't be used.
2227 // NOTE: this always runs without a P, so, nowritebarrierrec required.
2230 //go:nowritebarrierrec
2232 if GOOS == "windows" || GOOS == "plan9" {
2233 fatal("bindm in unexpected GOOS")
2237 fatal("the current g is not g0")
2239 if _cgo_bindm != nil {
2240 asmcgocall(_cgo_bindm, unsafe.Pointer(g))
2244 // A helper function for EnsureDropM.
2245 func getm() uintptr {
2246 return uintptr(unsafe.Pointer(getg().m))
2250 // Locking linked list of extra M's, via mp.schedlink. Must be accessed
2251 // only via lockextra/unlockextra.
2253 // Can't be atomic.Pointer[m] because we use an invalid pointer as a
2254 // "locked" sentinel value. M's on this list remain visible to the GC
2255 // because their mp.curg is on allgs.
2256 extraM atomic.Uintptr
2257 // Number of M's in the extraM list.
2258 extraMLength atomic.Uint32
2259 // Number of waiters in lockextra.
2260 extraMWaiters atomic.Uint32
2262 // Number of extra M's in use by threads.
2263 extraMInUse atomic.Uint32
2266 // lockextra locks the extra list and returns the list head.
2267 // The caller must unlock the list by storing a new list head
2268 // to extram. If nilokay is true, then lockextra will
2269 // return a nil list head if that's what it finds. If nilokay is false,
2270 // lockextra will keep waiting until the list head is no longer nil.
2273 func lockextra(nilokay bool) *m {
2278 old := extraM.Load()
2283 if old == 0 && !nilokay {
2285 // Add 1 to the number of threads
2286 // waiting for an M.
2287 // This is cleared by newextram.
2288 extraMWaiters.Add(1)
2294 if extraM.CompareAndSwap(old, locked) {
2295 return (*m)(unsafe.Pointer(old))
2303 func unlockextra(mp *m, delta int32) {
2304 extraMLength.Add(delta)
2305 extraM.Store(uintptr(unsafe.Pointer(mp)))
2308 // Return an M from the extra M list. Returns last == true if the list becomes
2309 // empty because of this call.
2311 // Spins waiting for an extra M, so caller must ensure that the list always
2312 // contains or will soon contain at least one M.
2315 func getExtraM() (mp *m, last bool) {
2316 mp = lockextra(false)
2318 unlockextra(mp.schedlink.ptr(), -1)
2319 return mp, mp.schedlink.ptr() == nil
2322 // Returns an extra M back to the list. mp must be from getExtraM. Newly
2323 // allocated M's should use addExtraM.
2326 func putExtraM(mp *m) {
2331 // Adds a newly allocated M to the extra M list.
2334 func addExtraM(mp *m) {
2335 mnext := lockextra(true)
2336 mp.schedlink.set(mnext)
2341 // allocmLock is locked for read when creating new Ms in allocm and their
2342 // addition to allm. Thus acquiring this lock for write blocks the
2343 // creation of new Ms.
2346 // execLock serializes exec and clone to avoid bugs or unspecified
2347 // behaviour around exec'ing while creating/destroying threads. See
2352 // These errors are reported (via writeErrStr) by some OS-specific
2353 // versions of newosproc and newosproc0.
2355 failthreadcreate = "runtime: failed to create new OS thread\n"
2356 failallocatestack = "runtime: failed to allocate stack for the new OS thread\n"
2359 // newmHandoff contains a list of m structures that need new OS threads.
2360 // This is used by newm in situations where newm itself can't safely
2361 // start an OS thread.
2362 var newmHandoff struct {
2365 // newm points to a list of M structures that need new OS
2366 // threads. The list is linked through m.schedlink.
2369 // waiting indicates that wake needs to be notified when an m
2370 // is put on the list.
2374 // haveTemplateThread indicates that the templateThread has
2375 // been started. This is not protected by lock. Use cas to set
2377 haveTemplateThread uint32
2380 // Create a new m. It will start off with a call to fn, or else the scheduler.
2381 // fn needs to be static and not a heap allocated closure.
2382 // May run with m.p==nil, so write barriers are not allowed.
2384 // id is optional pre-allocated m ID. Omit by passing -1.
2386 //go:nowritebarrierrec
2387 func newm(fn func(), pp *p, id int64) {
2388 // allocm adds a new M to allm, but they do not start until created by
2389 // the OS in newm1 or the template thread.
2391 // doAllThreadsSyscall requires that every M in allm will eventually
2392 // start and be signal-able, even with a STW.
2394 // Disable preemption here until we start the thread to ensure that
2395 // newm is not preempted between allocm and starting the new thread,
2396 // ensuring that anything added to allm is guaranteed to eventually
2400 mp := allocm(pp, fn, id)
2402 mp.sigmask = initSigmask
2403 if gp := getg(); gp != nil && gp.m != nil && (gp.m.lockedExt != 0 || gp.m.incgo) && GOOS != "plan9" {
2404 // We're on a locked M or a thread that may have been
2405 // started by C. The kernel state of this thread may
2406 // be strange (the user may have locked it for that
2407 // purpose). We don't want to clone that into another
2408 // thread. Instead, ask a known-good thread to create
2409 // the thread for us.
2411 // This is disabled on Plan 9. See golang.org/issue/22227.
2413 // TODO: This may be unnecessary on Windows, which
2414 // doesn't model thread creation off fork.
2415 lock(&newmHandoff.lock)
2416 if newmHandoff.haveTemplateThread == 0 {
2417 throw("on a locked thread with no template thread")
2419 mp.schedlink = newmHandoff.newm
2420 newmHandoff.newm.set(mp)
2421 if newmHandoff.waiting {
2422 newmHandoff.waiting = false
2423 notewakeup(&newmHandoff.wake)
2425 unlock(&newmHandoff.lock)
2426 // The M has not started yet, but the template thread does not
2427 // participate in STW, so it will always process queued Ms and
2428 // it is safe to releasem.
2438 var ts cgothreadstart
2439 if _cgo_thread_start == nil {
2440 throw("_cgo_thread_start missing")
2443 ts.tls = (*uint64)(unsafe.Pointer(&mp.tls[0]))
2444 ts.fn = unsafe.Pointer(abi.FuncPCABI0(mstart))
2446 msanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2449 asanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2451 execLock.rlock() // Prevent process clone.
2452 asmcgocall(_cgo_thread_start, unsafe.Pointer(&ts))
2456 execLock.rlock() // Prevent process clone.
2461 // startTemplateThread starts the template thread if it is not already
2464 // The calling thread must itself be in a known-good state.
2465 func startTemplateThread() {
2466 if GOARCH == "wasm" { // no threads on wasm yet
2470 // Disable preemption to guarantee that the template thread will be
2471 // created before a park once haveTemplateThread is set.
2473 if !atomic.Cas(&newmHandoff.haveTemplateThread, 0, 1) {
2477 newm(templateThread, nil, -1)
2481 // templateThread is a thread in a known-good state that exists solely
2482 // to start new threads in known-good states when the calling thread
2483 // may not be in a good state.
2485 // Many programs never need this, so templateThread is started lazily
2486 // when we first enter a state that might lead to running on a thread
2487 // in an unknown state.
2489 // templateThread runs on an M without a P, so it must not have write
2492 //go:nowritebarrierrec
2493 func templateThread() {
2500 lock(&newmHandoff.lock)
2501 for newmHandoff.newm != 0 {
2502 newm := newmHandoff.newm.ptr()
2503 newmHandoff.newm = 0
2504 unlock(&newmHandoff.lock)
2506 next := newm.schedlink.ptr()
2511 lock(&newmHandoff.lock)
2513 newmHandoff.waiting = true
2514 noteclear(&newmHandoff.wake)
2515 unlock(&newmHandoff.lock)
2516 notesleep(&newmHandoff.wake)
2520 // Stops execution of the current m until new work is available.
2521 // Returns with acquired P.
2525 if gp.m.locks != 0 {
2526 throw("stopm holding locks")
2529 throw("stopm holding p")
2532 throw("stopm spinning")
2539 acquirep(gp.m.nextp.ptr())
2544 // startm's caller incremented nmspinning. Set the new M's spinning.
2545 getg().m.spinning = true
2548 // Schedules some M to run the p (creates an M if necessary).
2549 // If p==nil, tries to get an idle P, if no idle P's does nothing.
2550 // May run with m.p==nil, so write barriers are not allowed.
2551 // If spinning is set, the caller has incremented nmspinning and must provide a
2552 // P. startm will set m.spinning in the newly started M.
2554 // Callers passing a non-nil P must call from a non-preemptible context. See
2555 // comment on acquirem below.
2557 // Argument lockheld indicates whether the caller already acquired the
2558 // scheduler lock. Callers holding the lock when making the call must pass
2559 // true. The lock might be temporarily dropped, but will be reacquired before
2562 // Must not have write barriers because this may be called without a P.
2564 //go:nowritebarrierrec
2565 func startm(pp *p, spinning, lockheld bool) {
2566 // Disable preemption.
2568 // Every owned P must have an owner that will eventually stop it in the
2569 // event of a GC stop request. startm takes transient ownership of a P
2570 // (either from argument or pidleget below) and transfers ownership to
2571 // a started M, which will be responsible for performing the stop.
2573 // Preemption must be disabled during this transient ownership,
2574 // otherwise the P this is running on may enter GC stop while still
2575 // holding the transient P, leaving that P in limbo and deadlocking the
2578 // Callers passing a non-nil P must already be in non-preemptible
2579 // context, otherwise such preemption could occur on function entry to
2580 // startm. Callers passing a nil P may be preemptible, so we must
2581 // disable preemption before acquiring a P from pidleget below.
2588 // TODO(prattmic): All remaining calls to this function
2589 // with _p_ == nil could be cleaned up to find a P
2590 // before calling startm.
2591 throw("startm: P required for spinning=true")
2604 // No M is available, we must drop sched.lock and call newm.
2605 // However, we already own a P to assign to the M.
2607 // Once sched.lock is released, another G (e.g., in a syscall),
2608 // could find no idle P while checkdead finds a runnable G but
2609 // no running M's because this new M hasn't started yet, thus
2610 // throwing in an apparent deadlock.
2611 // This apparent deadlock is possible when startm is called
2612 // from sysmon, which doesn't count as a running M.
2614 // Avoid this situation by pre-allocating the ID for the new M,
2615 // thus marking it as 'running' before we drop sched.lock. This
2616 // new M will eventually run the scheduler to execute any
2623 // The caller incremented nmspinning, so set m.spinning in the new M.
2631 // Ownership transfer of pp committed by start in newm.
2632 // Preemption is now safe.
2640 throw("startm: m is spinning")
2643 throw("startm: m has p")
2645 if spinning && !runqempty(pp) {
2646 throw("startm: p has runnable gs")
2648 // The caller incremented nmspinning, so set m.spinning in the new M.
2649 nmp.spinning = spinning
2651 notewakeup(&nmp.park)
2652 // Ownership transfer of pp committed by wakeup. Preemption is now
2657 // Hands off P from syscall or locked M.
2658 // Always runs without a P, so write barriers are not allowed.
2660 //go:nowritebarrierrec
2661 func handoffp(pp *p) {
2662 // handoffp must start an M in any situation where
2663 // findrunnable would return a G to run on pp.
2665 // if it has local work, start it straight away
2666 if !runqempty(pp) || sched.runqsize != 0 {
2667 startm(pp, false, false)
2670 // if there's trace work to do, start it straight away
2671 if (traceEnabled() || traceShuttingDown()) && traceReaderAvailable() != nil {
2672 startm(pp, false, false)
2675 // if it has GC work, start it straight away
2676 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) {
2677 startm(pp, false, false)
2680 // no local work, check that there are no spinning/idle M's,
2681 // otherwise our help is not required
2682 if sched.nmspinning.Load()+sched.npidle.Load() == 0 && sched.nmspinning.CompareAndSwap(0, 1) { // TODO: fast atomic
2683 sched.needspinning.Store(0)
2684 startm(pp, true, false)
2688 if sched.gcwaiting.Load() {
2689 pp.status = _Pgcstop
2691 if sched.stopwait == 0 {
2692 notewakeup(&sched.stopnote)
2697 if pp.runSafePointFn != 0 && atomic.Cas(&pp.runSafePointFn, 1, 0) {
2698 sched.safePointFn(pp)
2699 sched.safePointWait--
2700 if sched.safePointWait == 0 {
2701 notewakeup(&sched.safePointNote)
2704 if sched.runqsize != 0 {
2706 startm(pp, false, false)
2709 // If this is the last running P and nobody is polling network,
2710 // need to wakeup another M to poll network.
2711 if sched.npidle.Load() == gomaxprocs-1 && sched.lastpoll.Load() != 0 {
2713 startm(pp, false, false)
2717 // The scheduler lock cannot be held when calling wakeNetPoller below
2718 // because wakeNetPoller may call wakep which may call startm.
2719 when := nobarrierWakeTime(pp)
2728 // Tries to add one more P to execute G's.
2729 // Called when a G is made runnable (newproc, ready).
2730 // Must be called with a P.
2732 // Be conservative about spinning threads, only start one if none exist
2734 if sched.nmspinning.Load() != 0 || !sched.nmspinning.CompareAndSwap(0, 1) {
2738 // Disable preemption until ownership of pp transfers to the next M in
2739 // startm. Otherwise preemption here would leave pp stuck waiting to
2742 // See preemption comment on acquirem in startm for more details.
2747 pp, _ = pidlegetSpinning(0)
2749 if sched.nmspinning.Add(-1) < 0 {
2750 throw("wakep: negative nmspinning")
2756 // Since we always have a P, the race in the "No M is available"
2757 // comment in startm doesn't apply during the small window between the
2758 // unlock here and lock in startm. A checkdead in between will always
2759 // see at least one running M (ours).
2762 startm(pp, true, false)
2767 // Stops execution of the current m that is locked to a g until the g is runnable again.
2768 // Returns with acquired P.
2769 func stoplockedm() {
2772 if gp.m.lockedg == 0 || gp.m.lockedg.ptr().lockedm.ptr() != gp.m {
2773 throw("stoplockedm: inconsistent locking")
2776 // Schedule another M to run this p.
2781 // Wait until another thread schedules lockedg again.
2783 status := readgstatus(gp.m.lockedg.ptr())
2784 if status&^_Gscan != _Grunnable {
2785 print("runtime:stoplockedm: lockedg (atomicstatus=", status, ") is not Grunnable or Gscanrunnable\n")
2786 dumpgstatus(gp.m.lockedg.ptr())
2787 throw("stoplockedm: not runnable")
2789 acquirep(gp.m.nextp.ptr())
2793 // Schedules the locked m to run the locked gp.
2794 // May run during STW, so write barriers are not allowed.
2796 //go:nowritebarrierrec
2797 func startlockedm(gp *g) {
2798 mp := gp.lockedm.ptr()
2800 throw("startlockedm: locked to me")
2803 throw("startlockedm: m has p")
2805 // directly handoff current P to the locked m
2809 notewakeup(&mp.park)
2813 // Stops the current m for stopTheWorld.
2814 // Returns when the world is restarted.
2818 if !sched.gcwaiting.Load() {
2819 throw("gcstopm: not waiting for gc")
2822 gp.m.spinning = false
2823 // OK to just drop nmspinning here,
2824 // startTheWorld will unpark threads as necessary.
2825 if sched.nmspinning.Add(-1) < 0 {
2826 throw("gcstopm: negative nmspinning")
2831 pp.status = _Pgcstop
2833 if sched.stopwait == 0 {
2834 notewakeup(&sched.stopnote)
2840 // Schedules gp to run on the current M.
2841 // If inheritTime is true, gp inherits the remaining time in the
2842 // current time slice. Otherwise, it starts a new time slice.
2845 // Write barriers are allowed because this is called immediately after
2846 // acquiring a P in several places.
2848 //go:yeswritebarrierrec
2849 func execute(gp *g, inheritTime bool) {
2852 if goroutineProfile.active {
2853 // Make sure that gp has had its stack written out to the goroutine
2854 // profile, exactly as it was when the goroutine profiler first stopped
2856 tryRecordGoroutineProfile(gp, osyield)
2859 // Assign gp.m before entering _Grunning so running Gs have an
2863 casgstatus(gp, _Grunnable, _Grunning)
2866 gp.stackguard0 = gp.stack.lo + stackGuard
2868 mp.p.ptr().schedtick++
2871 // Check whether the profiler needs to be turned on or off.
2872 hz := sched.profilehz
2873 if mp.profilehz != hz {
2874 setThreadCPUProfiler(hz)
2878 // GoSysExit has to happen when we have a P, but before GoStart.
2879 // So we emit it here.
2880 if gp.syscallsp != 0 {
2889 // Finds a runnable goroutine to execute.
2890 // Tries to steal from other P's, get g from local or global queue, poll network.
2891 // tryWakeP indicates that the returned goroutine is not normal (GC worker, trace
2892 // reader) so the caller should try to wake a P.
2893 func findRunnable() (gp *g, inheritTime, tryWakeP bool) {
2896 // The conditions here and in handoffp must agree: if
2897 // findrunnable would return a G to run, handoffp must start
2902 if sched.gcwaiting.Load() {
2906 if pp.runSafePointFn != 0 {
2910 // now and pollUntil are saved for work stealing later,
2911 // which may steal timers. It's important that between now
2912 // and then, nothing blocks, so these numbers remain mostly
2914 now, pollUntil, _ := checkTimers(pp, 0)
2916 // Try to schedule the trace reader.
2917 if traceEnabled() || traceShuttingDown() {
2920 casgstatus(gp, _Gwaiting, _Grunnable)
2921 traceGoUnpark(gp, 0)
2922 return gp, false, true
2926 // Try to schedule a GC worker.
2927 if gcBlackenEnabled != 0 {
2928 gp, tnow := gcController.findRunnableGCWorker(pp, now)
2930 return gp, false, true
2935 // Check the global runnable queue once in a while to ensure fairness.
2936 // Otherwise two goroutines can completely occupy the local runqueue
2937 // by constantly respawning each other.
2938 if pp.schedtick%61 == 0 && sched.runqsize > 0 {
2940 gp := globrunqget(pp, 1)
2943 return gp, false, false
2947 // Wake up the finalizer G.
2948 if fingStatus.Load()&(fingWait|fingWake) == fingWait|fingWake {
2949 if gp := wakefing(); gp != nil {
2953 if *cgo_yield != nil {
2954 asmcgocall(*cgo_yield, nil)
2958 if gp, inheritTime := runqget(pp); gp != nil {
2959 return gp, inheritTime, false
2963 if sched.runqsize != 0 {
2965 gp := globrunqget(pp, 0)
2968 return gp, false, false
2973 // This netpoll is only an optimization before we resort to stealing.
2974 // We can safely skip it if there are no waiters or a thread is blocked
2975 // in netpoll already. If there is any kind of logical race with that
2976 // blocked thread (e.g. it has already returned from netpoll, but does
2977 // not set lastpoll yet), this thread will do blocking netpoll below
2979 if netpollinited() && netpollAnyWaiters() && sched.lastpoll.Load() != 0 {
2980 if list, delta := netpoll(0); !list.empty() { // non-blocking
2983 netpollAdjustWaiters(delta)
2984 casgstatus(gp, _Gwaiting, _Grunnable)
2986 traceGoUnpark(gp, 0)
2988 return gp, false, false
2992 // Spinning Ms: steal work from other Ps.
2994 // Limit the number of spinning Ms to half the number of busy Ps.
2995 // This is necessary to prevent excessive CPU consumption when
2996 // GOMAXPROCS>>1 but the program parallelism is low.
2997 if mp.spinning || 2*sched.nmspinning.Load() < gomaxprocs-sched.npidle.Load() {
3002 gp, inheritTime, tnow, w, newWork := stealWork(now)
3004 // Successfully stole.
3005 return gp, inheritTime, false
3008 // There may be new timer or GC work; restart to
3014 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3015 // Earlier timer to wait for.
3020 // We have nothing to do.
3022 // If we're in the GC mark phase, can safely scan and blacken objects,
3023 // and have work to do, run idle-time marking rather than give up the P.
3024 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) && gcController.addIdleMarkWorker() {
3025 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
3027 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
3029 casgstatus(gp, _Gwaiting, _Grunnable)
3031 traceGoUnpark(gp, 0)
3033 return gp, false, false
3035 gcController.removeIdleMarkWorker()
3039 // If a callback returned and no other goroutine is awake,
3040 // then wake event handler goroutine which pauses execution
3041 // until a callback was triggered.
3042 gp, otherReady := beforeIdle(now, pollUntil)
3044 casgstatus(gp, _Gwaiting, _Grunnable)
3046 traceGoUnpark(gp, 0)
3048 return gp, false, false
3054 // Before we drop our P, make a snapshot of the allp slice,
3055 // which can change underfoot once we no longer block
3056 // safe-points. We don't need to snapshot the contents because
3057 // everything up to cap(allp) is immutable.
3058 allpSnapshot := allp
3059 // Also snapshot masks. Value changes are OK, but we can't allow
3060 // len to change out from under us.
3061 idlepMaskSnapshot := idlepMask
3062 timerpMaskSnapshot := timerpMask
3064 // return P and block
3066 if sched.gcwaiting.Load() || pp.runSafePointFn != 0 {
3070 if sched.runqsize != 0 {
3071 gp := globrunqget(pp, 0)
3073 return gp, false, false
3075 if !mp.spinning && sched.needspinning.Load() == 1 {
3076 // See "Delicate dance" comment below.
3081 if releasep() != pp {
3082 throw("findrunnable: wrong p")
3084 now = pidleput(pp, now)
3087 // Delicate dance: thread transitions from spinning to non-spinning
3088 // state, potentially concurrently with submission of new work. We must
3089 // drop nmspinning first and then check all sources again (with
3090 // #StoreLoad memory barrier in between). If we do it the other way
3091 // around, another thread can submit work after we've checked all
3092 // sources but before we drop nmspinning; as a result nobody will
3093 // unpark a thread to run the work.
3095 // This applies to the following sources of work:
3097 // * Goroutines added to the global or a per-P run queue.
3098 // * New/modified-earlier timers on a per-P timer heap.
3099 // * Idle-priority GC work (barring golang.org/issue/19112).
3101 // If we discover new work below, we need to restore m.spinning as a
3102 // signal for resetspinning to unpark a new worker thread (because
3103 // there can be more than one starving goroutine).
3105 // However, if after discovering new work we also observe no idle Ps
3106 // (either here or in resetspinning), we have a problem. We may be
3107 // racing with a non-spinning M in the block above, having found no
3108 // work and preparing to release its P and park. Allowing that P to go
3109 // idle will result in loss of work conservation (idle P while there is
3110 // runnable work). This could result in complete deadlock in the
3111 // unlikely event that we discover new work (from netpoll) right as we
3112 // are racing with _all_ other Ps going idle.
3114 // We use sched.needspinning to synchronize with non-spinning Ms going
3115 // idle. If needspinning is set when they are about to drop their P,
3116 // they abort the drop and instead become a new spinning M on our
3117 // behalf. If we are not racing and the system is truly fully loaded
3118 // then no spinning threads are required, and the next thread to
3119 // naturally become spinning will clear the flag.
3121 // Also see "Worker thread parking/unparking" comment at the top of the
3123 wasSpinning := mp.spinning
3126 if sched.nmspinning.Add(-1) < 0 {
3127 throw("findrunnable: negative nmspinning")
3130 // Note the for correctness, only the last M transitioning from
3131 // spinning to non-spinning must perform these rechecks to
3132 // ensure no missed work. However, the runtime has some cases
3133 // of transient increments of nmspinning that are decremented
3134 // without going through this path, so we must be conservative
3135 // and perform the check on all spinning Ms.
3137 // See https://go.dev/issue/43997.
3139 // Check global and P runqueues again.
3142 if sched.runqsize != 0 {
3143 pp, _ := pidlegetSpinning(0)
3145 gp := globrunqget(pp, 0)
3147 throw("global runq empty with non-zero runqsize")
3152 return gp, false, false
3157 pp := checkRunqsNoP(allpSnapshot, idlepMaskSnapshot)
3164 // Check for idle-priority GC work again.
3165 pp, gp := checkIdleGCNoP()
3170 // Run the idle worker.
3171 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
3172 casgstatus(gp, _Gwaiting, _Grunnable)
3174 traceGoUnpark(gp, 0)
3176 return gp, false, false
3179 // Finally, check for timer creation or expiry concurrently with
3180 // transitioning from spinning to non-spinning.
3182 // Note that we cannot use checkTimers here because it calls
3183 // adjusttimers which may need to allocate memory, and that isn't
3184 // allowed when we don't have an active P.
3185 pollUntil = checkTimersNoP(allpSnapshot, timerpMaskSnapshot, pollUntil)
3188 // Poll network until next timer.
3189 if netpollinited() && (netpollAnyWaiters() || pollUntil != 0) && sched.lastpoll.Swap(0) != 0 {
3190 sched.pollUntil.Store(pollUntil)
3192 throw("findrunnable: netpoll with p")
3195 throw("findrunnable: netpoll with spinning")
3202 delay = pollUntil - now
3208 // When using fake time, just poll.
3211 list, delta := netpoll(delay) // block until new work is available
3212 // Refresh now again, after potentially blocking.
3214 sched.pollUntil.Store(0)
3215 sched.lastpoll.Store(now)
3216 if faketime != 0 && list.empty() {
3217 // Using fake time and nothing is ready; stop M.
3218 // When all M's stop, checkdead will call timejump.
3223 pp, _ := pidleget(now)
3227 netpollAdjustWaiters(delta)
3233 netpollAdjustWaiters(delta)
3234 casgstatus(gp, _Gwaiting, _Grunnable)
3236 traceGoUnpark(gp, 0)
3238 return gp, false, false
3245 } else if pollUntil != 0 && netpollinited() {
3246 pollerPollUntil := sched.pollUntil.Load()
3247 if pollerPollUntil == 0 || pollerPollUntil > pollUntil {
3255 // pollWork reports whether there is non-background work this P could
3256 // be doing. This is a fairly lightweight check to be used for
3257 // background work loops, like idle GC. It checks a subset of the
3258 // conditions checked by the actual scheduler.
3259 func pollWork() bool {
3260 if sched.runqsize != 0 {
3263 p := getg().m.p.ptr()
3267 if netpollinited() && netpollAnyWaiters() && sched.lastpoll.Load() != 0 {
3268 if list, delta := netpoll(0); !list.empty() {
3270 netpollAdjustWaiters(delta)
3277 // stealWork attempts to steal a runnable goroutine or timer from any P.
3279 // If newWork is true, new work may have been readied.
3281 // If now is not 0 it is the current time. stealWork returns the passed time or
3282 // the current time if now was passed as 0.
3283 func stealWork(now int64) (gp *g, inheritTime bool, rnow, pollUntil int64, newWork bool) {
3284 pp := getg().m.p.ptr()
3288 const stealTries = 4
3289 for i := 0; i < stealTries; i++ {
3290 stealTimersOrRunNextG := i == stealTries-1
3292 for enum := stealOrder.start(fastrand()); !enum.done(); enum.next() {
3293 if sched.gcwaiting.Load() {
3294 // GC work may be available.
3295 return nil, false, now, pollUntil, true
3297 p2 := allp[enum.position()]
3302 // Steal timers from p2. This call to checkTimers is the only place
3303 // where we might hold a lock on a different P's timers. We do this
3304 // once on the last pass before checking runnext because stealing
3305 // from the other P's runnext should be the last resort, so if there
3306 // are timers to steal do that first.
3308 // We only check timers on one of the stealing iterations because
3309 // the time stored in now doesn't change in this loop and checking
3310 // the timers for each P more than once with the same value of now
3311 // is probably a waste of time.
3313 // timerpMask tells us whether the P may have timers at all. If it
3314 // can't, no need to check at all.
3315 if stealTimersOrRunNextG && timerpMask.read(enum.position()) {
3316 tnow, w, ran := checkTimers(p2, now)
3318 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3322 // Running the timers may have
3323 // made an arbitrary number of G's
3324 // ready and added them to this P's
3325 // local run queue. That invalidates
3326 // the assumption of runqsteal
3327 // that it always has room to add
3328 // stolen G's. So check now if there
3329 // is a local G to run.
3330 if gp, inheritTime := runqget(pp); gp != nil {
3331 return gp, inheritTime, now, pollUntil, ranTimer
3337 // Don't bother to attempt to steal if p2 is idle.
3338 if !idlepMask.read(enum.position()) {
3339 if gp := runqsteal(pp, p2, stealTimersOrRunNextG); gp != nil {
3340 return gp, false, now, pollUntil, ranTimer
3346 // No goroutines found to steal. Regardless, running a timer may have
3347 // made some goroutine ready that we missed. Indicate the next timer to
3349 return nil, false, now, pollUntil, ranTimer
3352 // Check all Ps for a runnable G to steal.
3354 // On entry we have no P. If a G is available to steal and a P is available,
3355 // the P is returned which the caller should acquire and attempt to steal the
3357 func checkRunqsNoP(allpSnapshot []*p, idlepMaskSnapshot pMask) *p {
3358 for id, p2 := range allpSnapshot {
3359 if !idlepMaskSnapshot.read(uint32(id)) && !runqempty(p2) {
3361 pp, _ := pidlegetSpinning(0)
3363 // Can't get a P, don't bother checking remaining Ps.
3372 // No work available.
3376 // Check all Ps for a timer expiring sooner than pollUntil.
3378 // Returns updated pollUntil value.
3379 func checkTimersNoP(allpSnapshot []*p, timerpMaskSnapshot pMask, pollUntil int64) int64 {
3380 for id, p2 := range allpSnapshot {
3381 if timerpMaskSnapshot.read(uint32(id)) {
3382 w := nobarrierWakeTime(p2)
3383 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3392 // Check for idle-priority GC, without a P on entry.
3394 // If some GC work, a P, and a worker G are all available, the P and G will be
3395 // returned. The returned P has not been wired yet.
3396 func checkIdleGCNoP() (*p, *g) {
3397 // N.B. Since we have no P, gcBlackenEnabled may change at any time; we
3398 // must check again after acquiring a P. As an optimization, we also check
3399 // if an idle mark worker is needed at all. This is OK here, because if we
3400 // observe that one isn't needed, at least one is currently running. Even if
3401 // it stops running, its own journey into the scheduler should schedule it
3402 // again, if need be (at which point, this check will pass, if relevant).
3403 if atomic.Load(&gcBlackenEnabled) == 0 || !gcController.needIdleMarkWorker() {
3406 if !gcMarkWorkAvailable(nil) {
3410 // Work is available; we can start an idle GC worker only if there is
3411 // an available P and available worker G.
3413 // We can attempt to acquire these in either order, though both have
3414 // synchronization concerns (see below). Workers are almost always
3415 // available (see comment in findRunnableGCWorker for the one case
3416 // there may be none). Since we're slightly less likely to find a P,
3417 // check for that first.
3419 // Synchronization: note that we must hold sched.lock until we are
3420 // committed to keeping it. Otherwise we cannot put the unnecessary P
3421 // back in sched.pidle without performing the full set of idle
3422 // transition checks.
3424 // If we were to check gcBgMarkWorkerPool first, we must somehow handle
3425 // the assumption in gcControllerState.findRunnableGCWorker that an
3426 // empty gcBgMarkWorkerPool is only possible if gcMarkDone is running.
3428 pp, now := pidlegetSpinning(0)
3434 // Now that we own a P, gcBlackenEnabled can't change (as it requires STW).
3435 if gcBlackenEnabled == 0 || !gcController.addIdleMarkWorker() {
3441 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
3445 gcController.removeIdleMarkWorker()
3451 return pp, node.gp.ptr()
3454 // wakeNetPoller wakes up the thread sleeping in the network poller if it isn't
3455 // going to wake up before the when argument; or it wakes an idle P to service
3456 // timers and the network poller if there isn't one already.
3457 func wakeNetPoller(when int64) {
3458 if sched.lastpoll.Load() == 0 {
3459 // In findrunnable we ensure that when polling the pollUntil
3460 // field is either zero or the time to which the current
3461 // poll is expected to run. This can have a spurious wakeup
3462 // but should never miss a wakeup.
3463 pollerPollUntil := sched.pollUntil.Load()
3464 if pollerPollUntil == 0 || pollerPollUntil > when {
3468 // There are no threads in the network poller, try to get
3469 // one there so it can handle new timers.
3470 if GOOS != "plan9" { // Temporary workaround - see issue #42303.
3476 func resetspinning() {
3479 throw("resetspinning: not a spinning m")
3481 gp.m.spinning = false
3482 nmspinning := sched.nmspinning.Add(-1)
3484 throw("findrunnable: negative nmspinning")
3486 // M wakeup policy is deliberately somewhat conservative, so check if we
3487 // need to wakeup another P here. See "Worker thread parking/unparking"
3488 // comment at the top of the file for details.
3492 // injectglist adds each runnable G on the list to some run queue,
3493 // and clears glist. If there is no current P, they are added to the
3494 // global queue, and up to npidle M's are started to run them.
3495 // Otherwise, for each idle P, this adds a G to the global queue
3496 // and starts an M. Any remaining G's are added to the current P's
3498 // This may temporarily acquire sched.lock.
3499 // Can run concurrently with GC.
3500 func injectglist(glist *gList) {
3505 for gp := glist.head.ptr(); gp != nil; gp = gp.schedlink.ptr() {
3506 traceGoUnpark(gp, 0)
3510 // Mark all the goroutines as runnable before we put them
3511 // on the run queues.
3512 head := glist.head.ptr()
3515 for gp := head; gp != nil; gp = gp.schedlink.ptr() {
3518 casgstatus(gp, _Gwaiting, _Grunnable)
3521 // Turn the gList into a gQueue.
3527 startIdle := func(n int) {
3528 for i := 0; i < n; i++ {
3529 mp := acquirem() // See comment in startm.
3532 pp, _ := pidlegetSpinning(0)
3539 startm(pp, false, true)
3545 pp := getg().m.p.ptr()
3548 globrunqputbatch(&q, int32(qsize))
3554 npidle := int(sched.npidle.Load())
3557 for n = 0; n < npidle && !q.empty(); n++ {
3563 globrunqputbatch(&globq, int32(n))
3570 runqputbatch(pp, &q, qsize)
3574 // One round of scheduler: find a runnable goroutine and execute it.
3580 throw("schedule: holding locks")
3583 if mp.lockedg != 0 {
3585 execute(mp.lockedg.ptr(), false) // Never returns.
3588 // We should not schedule away from a g that is executing a cgo call,
3589 // since the cgo call is using the m's g0 stack.
3591 throw("schedule: in cgo")
3598 // Safety check: if we are spinning, the run queue should be empty.
3599 // Check this before calling checkTimers, as that might call
3600 // goready to put a ready goroutine on the local run queue.
3601 if mp.spinning && (pp.runnext != 0 || pp.runqhead != pp.runqtail) {
3602 throw("schedule: spinning with local work")
3605 gp, inheritTime, tryWakeP := findRunnable() // blocks until work is available
3607 if debug.dontfreezetheworld > 0 && freezing.Load() {
3608 // See comment in freezetheworld. We don't want to perturb
3609 // scheduler state, so we didn't gcstopm in findRunnable, but
3610 // also don't want to allow new goroutines to run.
3612 // Deadlock here rather than in the findRunnable loop so if
3613 // findRunnable is stuck in a loop we don't perturb that
3619 // This thread is going to run a goroutine and is not spinning anymore,
3620 // so if it was marked as spinning we need to reset it now and potentially
3621 // start a new spinning M.
3626 if sched.disable.user && !schedEnabled(gp) {
3627 // Scheduling of this goroutine is disabled. Put it on
3628 // the list of pending runnable goroutines for when we
3629 // re-enable user scheduling and look again.
3631 if schedEnabled(gp) {
3632 // Something re-enabled scheduling while we
3633 // were acquiring the lock.
3636 sched.disable.runnable.pushBack(gp)
3643 // If about to schedule a not-normal goroutine (a GCworker or tracereader),
3644 // wake a P if there is one.
3648 if gp.lockedm != 0 {
3649 // Hands off own p to the locked m,
3650 // then blocks waiting for a new p.
3655 execute(gp, inheritTime)
3658 // dropg removes the association between m and the current goroutine m->curg (gp for short).
3659 // Typically a caller sets gp's status away from Grunning and then
3660 // immediately calls dropg to finish the job. The caller is also responsible
3661 // for arranging that gp will be restarted using ready at an
3662 // appropriate time. After calling dropg and arranging for gp to be
3663 // readied later, the caller can do other work but eventually should
3664 // call schedule to restart the scheduling of goroutines on this m.
3668 setMNoWB(&gp.m.curg.m, nil)
3669 setGNoWB(&gp.m.curg, nil)
3672 // checkTimers runs any timers for the P that are ready.
3673 // If now is not 0 it is the current time.
3674 // It returns the passed time or the current time if now was passed as 0.
3675 // and the time when the next timer should run or 0 if there is no next timer,
3676 // and reports whether it ran any timers.
3677 // If the time when the next timer should run is not 0,
3678 // it is always larger than the returned time.
3679 // We pass now in and out to avoid extra calls of nanotime.
3681 //go:yeswritebarrierrec
3682 func checkTimers(pp *p, now int64) (rnow, pollUntil int64, ran bool) {
3683 // If it's not yet time for the first timer, or the first adjusted
3684 // timer, then there is nothing to do.
3685 next := pp.timer0When.Load()
3686 nextAdj := pp.timerModifiedEarliest.Load()
3687 if next == 0 || (nextAdj != 0 && nextAdj < next) {
3692 // No timers to run or adjust.
3693 return now, 0, false
3700 // Next timer is not ready to run, but keep going
3701 // if we would clear deleted timers.
3702 // This corresponds to the condition below where
3703 // we decide whether to call clearDeletedTimers.
3704 if pp != getg().m.p.ptr() || int(pp.deletedTimers.Load()) <= int(pp.numTimers.Load()/4) {
3705 return now, next, false
3709 lock(&pp.timersLock)
3711 if len(pp.timers) > 0 {
3712 adjusttimers(pp, now)
3713 for len(pp.timers) > 0 {
3714 // Note that runtimer may temporarily unlock
3716 if tw := runtimer(pp, now); tw != 0 {
3726 // If this is the local P, and there are a lot of deleted timers,
3727 // clear them out. We only do this for the local P to reduce
3728 // lock contention on timersLock.
3729 if pp == getg().m.p.ptr() && int(pp.deletedTimers.Load()) > len(pp.timers)/4 {
3730 clearDeletedTimers(pp)
3733 unlock(&pp.timersLock)
3735 return now, pollUntil, ran
3738 func parkunlock_c(gp *g, lock unsafe.Pointer) bool {
3739 unlock((*mutex)(lock))
3743 // park continuation on g0.
3744 func park_m(gp *g) {
3748 traceGoPark(mp.waitTraceBlockReason, mp.waitTraceSkip)
3751 // N.B. Not using casGToWaiting here because the waitreason is
3752 // set by park_m's caller.
3753 casgstatus(gp, _Grunning, _Gwaiting)
3756 if fn := mp.waitunlockf; fn != nil {
3757 ok := fn(gp, mp.waitlock)
3758 mp.waitunlockf = nil
3762 traceGoUnpark(gp, 2)
3764 casgstatus(gp, _Gwaiting, _Grunnable)
3765 execute(gp, true) // Schedule it back, never returns.
3771 func goschedImpl(gp *g) {
3772 status := readgstatus(gp)
3773 if status&^_Gscan != _Grunning {
3775 throw("bad g status")
3777 casgstatus(gp, _Grunning, _Grunnable)
3790 // Gosched continuation on g0.
3791 func gosched_m(gp *g) {
3798 // goschedguarded is a forbidden-states-avoided version of gosched_m.
3799 func goschedguarded_m(gp *g) {
3801 if !canPreemptM(gp.m) {
3802 gogo(&gp.sched) // never return
3811 func gopreempt_m(gp *g) {
3818 // preemptPark parks gp and puts it in _Gpreempted.
3821 func preemptPark(gp *g) {
3823 traceGoPark(traceBlockPreempted, 0)
3825 status := readgstatus(gp)
3826 if status&^_Gscan != _Grunning {
3828 throw("bad g status")
3831 if gp.asyncSafePoint {
3832 // Double-check that async preemption does not
3833 // happen in SPWRITE assembly functions.
3834 // isAsyncSafePoint must exclude this case.
3835 f := findfunc(gp.sched.pc)
3837 throw("preempt at unknown pc")
3839 if f.flag&abi.FuncFlagSPWrite != 0 {
3840 println("runtime: unexpected SPWRITE function", funcname(f), "in async preempt")
3841 throw("preempt SPWRITE")
3845 // Transition from _Grunning to _Gscan|_Gpreempted. We can't
3846 // be in _Grunning when we dropg because then we'd be running
3847 // without an M, but the moment we're in _Gpreempted,
3848 // something could claim this G before we've fully cleaned it
3849 // up. Hence, we set the scan bit to lock down further
3850 // transitions until we can dropg.
3851 casGToPreemptScan(gp, _Grunning, _Gscan|_Gpreempted)
3853 casfrom_Gscanstatus(gp, _Gscan|_Gpreempted, _Gpreempted)
3857 // goyield is like Gosched, but it:
3858 // - emits a GoPreempt trace event instead of a GoSched trace event
3859 // - puts the current G on the runq of the current P instead of the globrunq
3865 func goyield_m(gp *g) {
3870 casgstatus(gp, _Grunning, _Grunnable)
3872 runqput(pp, gp, false)
3876 // Finishes execution of the current goroutine.
3887 // goexit continuation on g0.
3888 func goexit0(gp *g) {
3892 casgstatus(gp, _Grunning, _Gdead)
3893 gcController.addScannableStack(pp, -int64(gp.stack.hi-gp.stack.lo))
3894 if isSystemGoroutine(gp, false) {
3898 locked := gp.lockedm != 0
3901 gp.preemptStop = false
3902 gp.paniconfault = false
3903 gp._defer = nil // should be true already but just in case.
3904 gp._panic = nil // non-nil for Goexit during panic. points at stack-allocated data.
3906 gp.waitreason = waitReasonZero
3911 if gcBlackenEnabled != 0 && gp.gcAssistBytes > 0 {
3912 // Flush assist credit to the global pool. This gives
3913 // better information to pacing if the application is
3914 // rapidly creating an exiting goroutines.
3915 assistWorkPerByte := gcController.assistWorkPerByte.Load()
3916 scanCredit := int64(assistWorkPerByte * float64(gp.gcAssistBytes))
3917 gcController.bgScanCredit.Add(scanCredit)
3918 gp.gcAssistBytes = 0
3923 if GOARCH == "wasm" { // no threads yet on wasm
3925 schedule() // never returns
3928 if mp.lockedInt != 0 {
3929 print("invalid m->lockedInt = ", mp.lockedInt, "\n")
3930 throw("internal lockOSThread error")
3934 // The goroutine may have locked this thread because
3935 // it put it in an unusual kernel state. Kill it
3936 // rather than returning it to the thread pool.
3938 // Return to mstart, which will release the P and exit
3940 if GOOS != "plan9" { // See golang.org/issue/22227.
3943 // Clear lockedExt on plan9 since we may end up re-using
3951 // save updates getg().sched to refer to pc and sp so that a following
3952 // gogo will restore pc and sp.
3954 // save must not have write barriers because invoking a write barrier
3955 // can clobber getg().sched.
3958 //go:nowritebarrierrec
3959 func save(pc, sp uintptr) {
3962 if gp == gp.m.g0 || gp == gp.m.gsignal {
3963 // m.g0.sched is special and must describe the context
3964 // for exiting the thread. mstart1 writes to it directly.
3965 // m.gsignal.sched should not be used at all.
3966 // This check makes sure save calls do not accidentally
3967 // run in contexts where they'd write to system g's.
3968 throw("save on system g not allowed")
3975 // We need to ensure ctxt is zero, but can't have a write
3976 // barrier here. However, it should always already be zero.
3978 if gp.sched.ctxt != nil {
3983 // The goroutine g is about to enter a system call.
3984 // Record that it's not using the cpu anymore.
3985 // This is called only from the go syscall library and cgocall,
3986 // not from the low-level system calls used by the runtime.
3988 // Entersyscall cannot split the stack: the save must
3989 // make g->sched refer to the caller's stack segment, because
3990 // entersyscall is going to return immediately after.
3992 // Nothing entersyscall calls can split the stack either.
3993 // We cannot safely move the stack during an active call to syscall,
3994 // because we do not know which of the uintptr arguments are
3995 // really pointers (back into the stack).
3996 // In practice, this means that we make the fast path run through
3997 // entersyscall doing no-split things, and the slow path has to use systemstack
3998 // to run bigger things on the system stack.
4000 // reentersyscall is the entry point used by cgo callbacks, where explicitly
4001 // saved SP and PC are restored. This is needed when exitsyscall will be called
4002 // from a function further up in the call stack than the parent, as g->syscallsp
4003 // must always point to a valid stack frame. entersyscall below is the normal
4004 // entry point for syscalls, which obtains the SP and PC from the caller.
4007 // At the start of a syscall we emit traceGoSysCall to capture the stack trace.
4008 // If the syscall does not block, that is it, we do not emit any other events.
4009 // If the syscall blocks (that is, P is retaken), retaker emits traceGoSysBlock;
4010 // when syscall returns we emit traceGoSysExit and when the goroutine starts running
4011 // (potentially instantly, if exitsyscallfast returns true) we emit traceGoStart.
4012 // To ensure that traceGoSysExit is emitted strictly after traceGoSysBlock,
4013 // we remember current value of syscalltick in m (gp.m.syscalltick = gp.m.p.ptr().syscalltick),
4014 // whoever emits traceGoSysBlock increments p.syscalltick afterwards;
4015 // and we wait for the increment before emitting traceGoSysExit.
4016 // Note that the increment is done even if tracing is not enabled,
4017 // because tracing can be enabled in the middle of syscall. We don't want the wait to hang.
4020 func reentersyscall(pc, sp uintptr) {
4023 // Disable preemption because during this function g is in Gsyscall status,
4024 // but can have inconsistent g->sched, do not let GC observe it.
4027 // Entersyscall must not call any function that might split/grow the stack.
4028 // (See details in comment above.)
4029 // Catch calls that might, by replacing the stack guard with something that
4030 // will trip any stack check and leaving a flag to tell newstack to die.
4031 gp.stackguard0 = stackPreempt
4032 gp.throwsplit = true
4034 // Leave SP around for GC and traceback.
4038 casgstatus(gp, _Grunning, _Gsyscall)
4039 if staticLockRanking {
4040 // When doing static lock ranking casgstatus can call
4041 // systemstack which clobbers g.sched.
4044 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
4045 systemstack(func() {
4046 print("entersyscall inconsistent ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
4047 throw("entersyscall")
4052 systemstack(traceGoSysCall)
4053 // systemstack itself clobbers g.sched.{pc,sp} and we might
4054 // need them later when the G is genuinely blocked in a
4059 if sched.sysmonwait.Load() {
4060 systemstack(entersyscall_sysmon)
4064 if gp.m.p.ptr().runSafePointFn != 0 {
4065 // runSafePointFn may stack split if run on this stack
4066 systemstack(runSafePointFn)
4070 gp.m.syscalltick = gp.m.p.ptr().syscalltick
4075 atomic.Store(&pp.status, _Psyscall)
4076 if sched.gcwaiting.Load() {
4077 systemstack(entersyscall_gcwait)
4084 // Standard syscall entry used by the go syscall library and normal cgo calls.
4086 // This is exported via linkname to assembly in the syscall package and x/sys.
4089 //go:linkname entersyscall
4090 func entersyscall() {
4091 reentersyscall(getcallerpc(), getcallersp())
4094 func entersyscall_sysmon() {
4096 if sched.sysmonwait.Load() {
4097 sched.sysmonwait.Store(false)
4098 notewakeup(&sched.sysmonnote)
4103 func entersyscall_gcwait() {
4105 pp := gp.m.oldp.ptr()
4108 if sched.stopwait > 0 && atomic.Cas(&pp.status, _Psyscall, _Pgcstop) {
4114 if sched.stopwait--; sched.stopwait == 0 {
4115 notewakeup(&sched.stopnote)
4121 // The same as entersyscall(), but with a hint that the syscall is blocking.
4124 func entersyscallblock() {
4127 gp.m.locks++ // see comment in entersyscall
4128 gp.throwsplit = true
4129 gp.stackguard0 = stackPreempt // see comment in entersyscall
4130 gp.m.syscalltick = gp.m.p.ptr().syscalltick
4131 gp.m.p.ptr().syscalltick++
4133 // Leave SP around for GC and traceback.
4137 gp.syscallsp = gp.sched.sp
4138 gp.syscallpc = gp.sched.pc
4139 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
4143 systemstack(func() {
4144 print("entersyscallblock inconsistent ", hex(sp1), " ", hex(sp2), " ", hex(sp3), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
4145 throw("entersyscallblock")
4148 casgstatus(gp, _Grunning, _Gsyscall)
4149 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
4150 systemstack(func() {
4151 print("entersyscallblock inconsistent ", hex(sp), " ", hex(gp.sched.sp), " ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
4152 throw("entersyscallblock")
4156 systemstack(entersyscallblock_handoff)
4158 // Resave for traceback during blocked call.
4159 save(getcallerpc(), getcallersp())
4164 func entersyscallblock_handoff() {
4167 traceGoSysBlock(getg().m.p.ptr())
4169 handoffp(releasep())
4172 // The goroutine g exited its system call.
4173 // Arrange for it to run on a cpu again.
4174 // This is called only from the go syscall library, not
4175 // from the low-level system calls used by the runtime.
4177 // Write barriers are not allowed because our P may have been stolen.
4179 // This is exported via linkname to assembly in the syscall package.
4182 //go:nowritebarrierrec
4183 //go:linkname exitsyscall
4184 func exitsyscall() {
4187 gp.m.locks++ // see comment in entersyscall
4188 if getcallersp() > gp.syscallsp {
4189 throw("exitsyscall: syscall frame is no longer valid")
4193 oldp := gp.m.oldp.ptr()
4195 if exitsyscallfast(oldp) {
4196 // When exitsyscallfast returns success, we have a P so can now use
4198 if goroutineProfile.active {
4199 // Make sure that gp has had its stack written out to the goroutine
4200 // profile, exactly as it was when the goroutine profiler first
4201 // stopped the world.
4202 systemstack(func() {
4203 tryRecordGoroutineProfileWB(gp)
4207 if oldp != gp.m.p.ptr() || gp.m.syscalltick != gp.m.p.ptr().syscalltick {
4208 systemstack(traceGoStart)
4211 // There's a cpu for us, so we can run.
4212 gp.m.p.ptr().syscalltick++
4213 // We need to cas the status and scan before resuming...
4214 casgstatus(gp, _Gsyscall, _Grunning)
4216 // Garbage collector isn't running (since we are),
4217 // so okay to clear syscallsp.
4221 // restore the preemption request in case we've cleared it in newstack
4222 gp.stackguard0 = stackPreempt
4224 // otherwise restore the real stackGuard, we've spoiled it in entersyscall/entersyscallblock
4225 gp.stackguard0 = gp.stack.lo + stackGuard
4227 gp.throwsplit = false
4229 if sched.disable.user && !schedEnabled(gp) {
4230 // Scheduling of this goroutine is disabled.
4238 // Wait till traceGoSysBlock event is emitted.
4239 // This ensures consistency of the trace (the goroutine is started after it is blocked).
4240 for oldp != nil && oldp.syscalltick == gp.m.syscalltick {
4243 // We can't trace syscall exit right now because we don't have a P.
4244 // Tracing code can invoke write barriers that cannot run without a P.
4245 // So instead we remember the syscall exit time and emit the event
4246 // in execute when we have a P.
4247 gp.trace.sysExitTime = traceClockNow()
4252 // Call the scheduler.
4255 // Scheduler returned, so we're allowed to run now.
4256 // Delete the syscallsp information that we left for
4257 // the garbage collector during the system call.
4258 // Must wait until now because until gosched returns
4259 // we don't know for sure that the garbage collector
4262 gp.m.p.ptr().syscalltick++
4263 gp.throwsplit = false
4267 func exitsyscallfast(oldp *p) bool {
4270 // Freezetheworld sets stopwait but does not retake P's.
4271 if sched.stopwait == freezeStopWait {
4275 // Try to re-acquire the last P.
4276 if oldp != nil && oldp.status == _Psyscall && atomic.Cas(&oldp.status, _Psyscall, _Pidle) {
4277 // There's a cpu for us, so we can run.
4279 exitsyscallfast_reacquired()
4283 // Try to get any other idle P.
4284 if sched.pidle != 0 {
4286 systemstack(func() {
4287 ok = exitsyscallfast_pidle()
4288 if ok && traceEnabled() {
4290 // Wait till traceGoSysBlock event is emitted.
4291 // This ensures consistency of the trace (the goroutine is started after it is blocked).
4292 for oldp.syscalltick == gp.m.syscalltick {
4306 // exitsyscallfast_reacquired is the exitsyscall path on which this G
4307 // has successfully reacquired the P it was running on before the
4311 func exitsyscallfast_reacquired() {
4313 if gp.m.syscalltick != gp.m.p.ptr().syscalltick {
4315 // The p was retaken and then enter into syscall again (since gp.m.syscalltick has changed).
4316 // traceGoSysBlock for this syscall was already emitted,
4317 // but here we effectively retake the p from the new syscall running on the same p.
4318 systemstack(func() {
4319 // Denote blocking of the new syscall.
4320 traceGoSysBlock(gp.m.p.ptr())
4321 // Denote completion of the current syscall.
4325 gp.m.p.ptr().syscalltick++
4329 func exitsyscallfast_pidle() bool {
4331 pp, _ := pidleget(0)
4332 if pp != nil && sched.sysmonwait.Load() {
4333 sched.sysmonwait.Store(false)
4334 notewakeup(&sched.sysmonnote)
4344 // exitsyscall slow path on g0.
4345 // Failed to acquire P, enqueue gp as runnable.
4347 // Called via mcall, so gp is the calling g from this M.
4349 //go:nowritebarrierrec
4350 func exitsyscall0(gp *g) {
4351 casgstatus(gp, _Gsyscall, _Grunnable)
4355 if schedEnabled(gp) {
4362 // Below, we stoplockedm if gp is locked. globrunqput releases
4363 // ownership of gp, so we must check if gp is locked prior to
4364 // committing the release by unlocking sched.lock, otherwise we
4365 // could race with another M transitioning gp from unlocked to
4367 locked = gp.lockedm != 0
4368 } else if sched.sysmonwait.Load() {
4369 sched.sysmonwait.Store(false)
4370 notewakeup(&sched.sysmonnote)
4375 execute(gp, false) // Never returns.
4378 // Wait until another thread schedules gp and so m again.
4380 // N.B. lockedm must be this M, as this g was running on this M
4381 // before entersyscall.
4383 execute(gp, false) // Never returns.
4386 schedule() // Never returns.
4389 // Called from syscall package before fork.
4391 //go:linkname syscall_runtime_BeforeFork syscall.runtime_BeforeFork
4393 func syscall_runtime_BeforeFork() {
4396 // Block signals during a fork, so that the child does not run
4397 // a signal handler before exec if a signal is sent to the process
4398 // group. See issue #18600.
4400 sigsave(&gp.m.sigmask)
4403 // This function is called before fork in syscall package.
4404 // Code between fork and exec must not allocate memory nor even try to grow stack.
4405 // Here we spoil g.stackguard0 to reliably detect any attempts to grow stack.
4406 // runtime_AfterFork will undo this in parent process, but not in child.
4407 gp.stackguard0 = stackFork
4410 // Called from syscall package after fork in parent.
4412 //go:linkname syscall_runtime_AfterFork syscall.runtime_AfterFork
4414 func syscall_runtime_AfterFork() {
4417 // See the comments in beforefork.
4418 gp.stackguard0 = gp.stack.lo + stackGuard
4420 msigrestore(gp.m.sigmask)
4425 // inForkedChild is true while manipulating signals in the child process.
4426 // This is used to avoid calling libc functions in case we are using vfork.
4427 var inForkedChild bool
4429 // Called from syscall package after fork in child.
4430 // It resets non-sigignored signals to the default handler, and
4431 // restores the signal mask in preparation for the exec.
4433 // Because this might be called during a vfork, and therefore may be
4434 // temporarily sharing address space with the parent process, this must
4435 // not change any global variables or calling into C code that may do so.
4437 //go:linkname syscall_runtime_AfterForkInChild syscall.runtime_AfterForkInChild
4439 //go:nowritebarrierrec
4440 func syscall_runtime_AfterForkInChild() {
4441 // It's OK to change the global variable inForkedChild here
4442 // because we are going to change it back. There is no race here,
4443 // because if we are sharing address space with the parent process,
4444 // then the parent process can not be running concurrently.
4445 inForkedChild = true
4447 clearSignalHandlers()
4449 // When we are the child we are the only thread running,
4450 // so we know that nothing else has changed gp.m.sigmask.
4451 msigrestore(getg().m.sigmask)
4453 inForkedChild = false
4456 // pendingPreemptSignals is the number of preemption signals
4457 // that have been sent but not received. This is only used on Darwin.
4459 var pendingPreemptSignals atomic.Int32
4461 // Called from syscall package before Exec.
4463 //go:linkname syscall_runtime_BeforeExec syscall.runtime_BeforeExec
4464 func syscall_runtime_BeforeExec() {
4465 // Prevent thread creation during exec.
4468 // On Darwin, wait for all pending preemption signals to
4469 // be received. See issue #41702.
4470 if GOOS == "darwin" || GOOS == "ios" {
4471 for pendingPreemptSignals.Load() > 0 {
4477 // Called from syscall package after Exec.
4479 //go:linkname syscall_runtime_AfterExec syscall.runtime_AfterExec
4480 func syscall_runtime_AfterExec() {
4484 // Allocate a new g, with a stack big enough for stacksize bytes.
4485 func malg(stacksize int32) *g {
4488 stacksize = round2(stackSystem + stacksize)
4489 systemstack(func() {
4490 newg.stack = stackalloc(uint32(stacksize))
4492 newg.stackguard0 = newg.stack.lo + stackGuard
4493 newg.stackguard1 = ^uintptr(0)
4494 // Clear the bottom word of the stack. We record g
4495 // there on gsignal stack during VDSO on ARM and ARM64.
4496 *(*uintptr)(unsafe.Pointer(newg.stack.lo)) = 0
4501 // Create a new g running fn.
4502 // Put it on the queue of g's waiting to run.
4503 // The compiler turns a go statement into a call to this.
4504 func newproc(fn *funcval) {
4507 systemstack(func() {
4508 newg := newproc1(fn, gp, pc)
4510 pp := getg().m.p.ptr()
4511 runqput(pp, newg, true)
4519 // Create a new g in state _Grunnable, starting at fn. callerpc is the
4520 // address of the go statement that created this. The caller is responsible
4521 // for adding the new g to the scheduler.
4522 func newproc1(fn *funcval, callergp *g, callerpc uintptr) *g {
4524 fatal("go of nil func value")
4527 mp := acquirem() // disable preemption because we hold M and P in local vars.
4531 newg = malg(stackMin)
4532 casgstatus(newg, _Gidle, _Gdead)
4533 allgadd(newg) // publishes with a g->status of Gdead so GC scanner doesn't look at uninitialized stack.
4535 if newg.stack.hi == 0 {
4536 throw("newproc1: newg missing stack")
4539 if readgstatus(newg) != _Gdead {
4540 throw("newproc1: new g is not Gdead")
4543 totalSize := uintptr(4*goarch.PtrSize + sys.MinFrameSize) // extra space in case of reads slightly beyond frame
4544 totalSize = alignUp(totalSize, sys.StackAlign)
4545 sp := newg.stack.hi - totalSize
4549 *(*uintptr)(unsafe.Pointer(sp)) = 0
4551 spArg += sys.MinFrameSize
4554 memclrNoHeapPointers(unsafe.Pointer(&newg.sched), unsafe.Sizeof(newg.sched))
4557 newg.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum // +PCQuantum so that previous instruction is in same function
4558 newg.sched.g = guintptr(unsafe.Pointer(newg))
4559 gostartcallfn(&newg.sched, fn)
4560 newg.parentGoid = callergp.goid
4561 newg.gopc = callerpc
4562 newg.ancestors = saveAncestors(callergp)
4563 newg.startpc = fn.fn
4564 if isSystemGoroutine(newg, false) {
4567 // Only user goroutines inherit pprof labels.
4569 newg.labels = mp.curg.labels
4571 if goroutineProfile.active {
4572 // A concurrent goroutine profile is running. It should include
4573 // exactly the set of goroutines that were alive when the goroutine
4574 // profiler first stopped the world. That does not include newg, so
4575 // mark it as not needing a profile before transitioning it from
4577 newg.goroutineProfiled.Store(goroutineProfileSatisfied)
4580 // Track initial transition?
4581 newg.trackingSeq = uint8(fastrand())
4582 if newg.trackingSeq%gTrackingPeriod == 0 {
4583 newg.tracking = true
4585 casgstatus(newg, _Gdead, _Grunnable)
4586 gcController.addScannableStack(pp, int64(newg.stack.hi-newg.stack.lo))
4588 if pp.goidcache == pp.goidcacheend {
4589 // Sched.goidgen is the last allocated id,
4590 // this batch must be [sched.goidgen+1, sched.goidgen+GoidCacheBatch].
4591 // At startup sched.goidgen=0, so main goroutine receives goid=1.
4592 pp.goidcache = sched.goidgen.Add(_GoidCacheBatch)
4593 pp.goidcache -= _GoidCacheBatch - 1
4594 pp.goidcacheend = pp.goidcache + _GoidCacheBatch
4596 newg.goid = pp.goidcache
4599 newg.racectx = racegostart(callerpc)
4601 if newg.labels != nil {
4602 // See note in proflabel.go on labelSync's role in synchronizing
4603 // with the reads in the signal handler.
4604 racereleasemergeg(newg, unsafe.Pointer(&labelSync))
4608 traceGoCreate(newg, newg.startpc)
4615 // saveAncestors copies previous ancestors of the given caller g and
4616 // includes info for the current caller into a new set of tracebacks for
4617 // a g being created.
4618 func saveAncestors(callergp *g) *[]ancestorInfo {
4619 // Copy all prior info, except for the root goroutine (goid 0).
4620 if debug.tracebackancestors <= 0 || callergp.goid == 0 {
4623 var callerAncestors []ancestorInfo
4624 if callergp.ancestors != nil {
4625 callerAncestors = *callergp.ancestors
4627 n := int32(len(callerAncestors)) + 1
4628 if n > debug.tracebackancestors {
4629 n = debug.tracebackancestors
4631 ancestors := make([]ancestorInfo, n)
4632 copy(ancestors[1:], callerAncestors)
4634 var pcs [tracebackInnerFrames]uintptr
4635 npcs := gcallers(callergp, 0, pcs[:])
4636 ipcs := make([]uintptr, npcs)
4638 ancestors[0] = ancestorInfo{
4640 goid: callergp.goid,
4641 gopc: callergp.gopc,
4644 ancestorsp := new([]ancestorInfo)
4645 *ancestorsp = ancestors
4649 // Put on gfree list.
4650 // If local list is too long, transfer a batch to the global list.
4651 func gfput(pp *p, gp *g) {
4652 if readgstatus(gp) != _Gdead {
4653 throw("gfput: bad status (not Gdead)")
4656 stksize := gp.stack.hi - gp.stack.lo
4658 if stksize != uintptr(startingStackSize) {
4659 // non-standard stack size - free it.
4668 if pp.gFree.n >= 64 {
4674 for pp.gFree.n >= 32 {
4675 gp := pp.gFree.pop()
4677 if gp.stack.lo == 0 {
4684 lock(&sched.gFree.lock)
4685 sched.gFree.noStack.pushAll(noStackQ)
4686 sched.gFree.stack.pushAll(stackQ)
4687 sched.gFree.n += inc
4688 unlock(&sched.gFree.lock)
4692 // Get from gfree list.
4693 // If local list is empty, grab a batch from global list.
4694 func gfget(pp *p) *g {
4696 if pp.gFree.empty() && (!sched.gFree.stack.empty() || !sched.gFree.noStack.empty()) {
4697 lock(&sched.gFree.lock)
4698 // Move a batch of free Gs to the P.
4699 for pp.gFree.n < 32 {
4700 // Prefer Gs with stacks.
4701 gp := sched.gFree.stack.pop()
4703 gp = sched.gFree.noStack.pop()
4712 unlock(&sched.gFree.lock)
4715 gp := pp.gFree.pop()
4720 if gp.stack.lo != 0 && gp.stack.hi-gp.stack.lo != uintptr(startingStackSize) {
4721 // Deallocate old stack. We kept it in gfput because it was the
4722 // right size when the goroutine was put on the free list, but
4723 // the right size has changed since then.
4724 systemstack(func() {
4731 if gp.stack.lo == 0 {
4732 // Stack was deallocated in gfput or just above. Allocate a new one.
4733 systemstack(func() {
4734 gp.stack = stackalloc(startingStackSize)
4736 gp.stackguard0 = gp.stack.lo + stackGuard
4739 racemalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4742 msanmalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4745 asanunpoison(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4751 // Purge all cached G's from gfree list to the global list.
4752 func gfpurge(pp *p) {
4758 for !pp.gFree.empty() {
4759 gp := pp.gFree.pop()
4761 if gp.stack.lo == 0 {
4768 lock(&sched.gFree.lock)
4769 sched.gFree.noStack.pushAll(noStackQ)
4770 sched.gFree.stack.pushAll(stackQ)
4771 sched.gFree.n += inc
4772 unlock(&sched.gFree.lock)
4775 // Breakpoint executes a breakpoint trap.
4780 // dolockOSThread is called by LockOSThread and lockOSThread below
4781 // after they modify m.locked. Do not allow preemption during this call,
4782 // or else the m might be different in this function than in the caller.
4785 func dolockOSThread() {
4786 if GOARCH == "wasm" {
4787 return // no threads on wasm yet
4790 gp.m.lockedg.set(gp)
4791 gp.lockedm.set(gp.m)
4794 // LockOSThread wires the calling goroutine to its current operating system thread.
4795 // The calling goroutine will always execute in that thread,
4796 // and no other goroutine will execute in it,
4797 // until the calling goroutine has made as many calls to
4798 // UnlockOSThread as to LockOSThread.
4799 // If the calling goroutine exits without unlocking the thread,
4800 // the thread will be terminated.
4802 // All init functions are run on the startup thread. Calling LockOSThread
4803 // from an init function will cause the main function to be invoked on
4806 // A goroutine should call LockOSThread before calling OS services or
4807 // non-Go library functions that depend on per-thread state.
4810 func LockOSThread() {
4811 if atomic.Load(&newmHandoff.haveTemplateThread) == 0 && GOOS != "plan9" {
4812 // If we need to start a new thread from the locked
4813 // thread, we need the template thread. Start it now
4814 // while we're in a known-good state.
4815 startTemplateThread()
4819 if gp.m.lockedExt == 0 {
4821 panic("LockOSThread nesting overflow")
4827 func lockOSThread() {
4828 getg().m.lockedInt++
4832 // dounlockOSThread is called by UnlockOSThread and unlockOSThread below
4833 // after they update m->locked. Do not allow preemption during this call,
4834 // or else the m might be in different in this function than in the caller.
4837 func dounlockOSThread() {
4838 if GOARCH == "wasm" {
4839 return // no threads on wasm yet
4842 if gp.m.lockedInt != 0 || gp.m.lockedExt != 0 {
4849 // UnlockOSThread undoes an earlier call to LockOSThread.
4850 // If this drops the number of active LockOSThread calls on the
4851 // calling goroutine to zero, it unwires the calling goroutine from
4852 // its fixed operating system thread.
4853 // If there are no active LockOSThread calls, this is a no-op.
4855 // Before calling UnlockOSThread, the caller must ensure that the OS
4856 // thread is suitable for running other goroutines. If the caller made
4857 // any permanent changes to the state of the thread that would affect
4858 // other goroutines, it should not call this function and thus leave
4859 // the goroutine locked to the OS thread until the goroutine (and
4860 // hence the thread) exits.
4863 func UnlockOSThread() {
4865 if gp.m.lockedExt == 0 {
4873 func unlockOSThread() {
4875 if gp.m.lockedInt == 0 {
4876 systemstack(badunlockosthread)
4882 func badunlockosthread() {
4883 throw("runtime: internal error: misuse of lockOSThread/unlockOSThread")
4886 func gcount() int32 {
4887 n := int32(atomic.Loaduintptr(&allglen)) - sched.gFree.n - sched.ngsys.Load()
4888 for _, pp := range allp {
4892 // All these variables can be changed concurrently, so the result can be inconsistent.
4893 // But at least the current goroutine is running.
4900 func mcount() int32 {
4901 return int32(sched.mnext - sched.nmfreed)
4905 signalLock atomic.Uint32
4907 // Must hold signalLock to write. Reads may be lock-free, but
4908 // signalLock should be taken to synchronize with changes.
4912 func _System() { _System() }
4913 func _ExternalCode() { _ExternalCode() }
4914 func _LostExternalCode() { _LostExternalCode() }
4915 func _GC() { _GC() }
4916 func _LostSIGPROFDuringAtomic64() { _LostSIGPROFDuringAtomic64() }
4917 func _VDSO() { _VDSO() }
4919 // Called if we receive a SIGPROF signal.
4920 // Called by the signal handler, may run during STW.
4922 //go:nowritebarrierrec
4923 func sigprof(pc, sp, lr uintptr, gp *g, mp *m) {
4924 if prof.hz.Load() == 0 {
4928 // If mp.profilehz is 0, then profiling is not enabled for this thread.
4929 // We must check this to avoid a deadlock between setcpuprofilerate
4930 // and the call to cpuprof.add, below.
4931 if mp != nil && mp.profilehz == 0 {
4935 // On mips{,le}/arm, 64bit atomics are emulated with spinlocks, in
4936 // runtime/internal/atomic. If SIGPROF arrives while the program is inside
4937 // the critical section, it creates a deadlock (when writing the sample).
4938 // As a workaround, create a counter of SIGPROFs while in critical section
4939 // to store the count, and pass it to sigprof.add() later when SIGPROF is
4940 // received from somewhere else (with _LostSIGPROFDuringAtomic64 as pc).
4941 if GOARCH == "mips" || GOARCH == "mipsle" || GOARCH == "arm" {
4942 if f := findfunc(pc); f.valid() {
4943 if hasPrefix(funcname(f), "runtime/internal/atomic") {
4944 cpuprof.lostAtomic++
4948 if GOARCH == "arm" && goarm < 7 && GOOS == "linux" && pc&0xffff0000 == 0xffff0000 {
4949 // runtime/internal/atomic functions call into kernel
4950 // helpers on arm < 7. See
4951 // runtime/internal/atomic/sys_linux_arm.s.
4952 cpuprof.lostAtomic++
4957 // Profiling runs concurrently with GC, so it must not allocate.
4958 // Set a trap in case the code does allocate.
4959 // Note that on windows, one thread takes profiles of all the
4960 // other threads, so mp is usually not getg().m.
4961 // In fact mp may not even be stopped.
4962 // See golang.org/issue/17165.
4963 getg().m.mallocing++
4966 var stk [maxCPUProfStack]uintptr
4968 if mp.ncgo > 0 && mp.curg != nil && mp.curg.syscallpc != 0 && mp.curg.syscallsp != 0 {
4970 // Check cgoCallersUse to make sure that we are not
4971 // interrupting other code that is fiddling with
4972 // cgoCallers. We are running in a signal handler
4973 // with all signals blocked, so we don't have to worry
4974 // about any other code interrupting us.
4975 if mp.cgoCallersUse.Load() == 0 && mp.cgoCallers != nil && mp.cgoCallers[0] != 0 {
4976 for cgoOff < len(mp.cgoCallers) && mp.cgoCallers[cgoOff] != 0 {
4979 n += copy(stk[:], mp.cgoCallers[:cgoOff])
4980 mp.cgoCallers[0] = 0
4983 // Collect Go stack that leads to the cgo call.
4984 u.initAt(mp.curg.syscallpc, mp.curg.syscallsp, 0, mp.curg, unwindSilentErrors)
4985 } else if usesLibcall() && mp.libcallg != 0 && mp.libcallpc != 0 && mp.libcallsp != 0 {
4986 // Libcall, i.e. runtime syscall on windows.
4987 // Collect Go stack that leads to the call.
4988 u.initAt(mp.libcallpc, mp.libcallsp, 0, mp.libcallg.ptr(), unwindSilentErrors)
4989 } else if mp != nil && mp.vdsoSP != 0 {
4990 // VDSO call, e.g. nanotime1 on Linux.
4991 // Collect Go stack that leads to the call.
4992 u.initAt(mp.vdsoPC, mp.vdsoSP, 0, gp, unwindSilentErrors|unwindJumpStack)
4994 u.initAt(pc, sp, lr, gp, unwindSilentErrors|unwindTrap|unwindJumpStack)
4996 n += tracebackPCs(&u, 0, stk[n:])
4999 // Normal traceback is impossible or has failed.
5000 // Account it against abstract "System" or "GC".
5003 pc = abi.FuncPCABIInternal(_VDSO) + sys.PCQuantum
5004 } else if pc > firstmoduledata.etext {
5005 // "ExternalCode" is better than "etext".
5006 pc = abi.FuncPCABIInternal(_ExternalCode) + sys.PCQuantum
5009 if mp.preemptoff != "" {
5010 stk[1] = abi.FuncPCABIInternal(_GC) + sys.PCQuantum
5012 stk[1] = abi.FuncPCABIInternal(_System) + sys.PCQuantum
5016 if prof.hz.Load() != 0 {
5017 // Note: it can happen on Windows that we interrupted a system thread
5018 // with no g, so gp could nil. The other nil checks are done out of
5019 // caution, but not expected to be nil in practice.
5020 var tagPtr *unsafe.Pointer
5021 if gp != nil && gp.m != nil && gp.m.curg != nil {
5022 tagPtr = &gp.m.curg.labels
5024 cpuprof.add(tagPtr, stk[:n])
5028 if gp != nil && gp.m != nil {
5029 if gp.m.curg != nil {
5034 traceCPUSample(gprof, pp, stk[:n])
5036 getg().m.mallocing--
5039 // setcpuprofilerate sets the CPU profiling rate to hz times per second.
5040 // If hz <= 0, setcpuprofilerate turns off CPU profiling.
5041 func setcpuprofilerate(hz int32) {
5042 // Force sane arguments.
5047 // Disable preemption, otherwise we can be rescheduled to another thread
5048 // that has profiling enabled.
5052 // Stop profiler on this thread so that it is safe to lock prof.
5053 // if a profiling signal came in while we had prof locked,
5054 // it would deadlock.
5055 setThreadCPUProfiler(0)
5057 for !prof.signalLock.CompareAndSwap(0, 1) {
5060 if prof.hz.Load() != hz {
5061 setProcessCPUProfiler(hz)
5064 prof.signalLock.Store(0)
5067 sched.profilehz = hz
5071 setThreadCPUProfiler(hz)
5077 // init initializes pp, which may be a freshly allocated p or a
5078 // previously destroyed p, and transitions it to status _Pgcstop.
5079 func (pp *p) init(id int32) {
5081 pp.status = _Pgcstop
5082 pp.sudogcache = pp.sudogbuf[:0]
5083 pp.deferpool = pp.deferpoolbuf[:0]
5085 if pp.mcache == nil {
5088 throw("missing mcache?")
5090 // Use the bootstrap mcache0. Only one P will get
5091 // mcache0: the one with ID 0.
5094 pp.mcache = allocmcache()
5097 if raceenabled && pp.raceprocctx == 0 {
5099 pp.raceprocctx = raceprocctx0
5100 raceprocctx0 = 0 // bootstrap
5102 pp.raceprocctx = raceproccreate()
5105 lockInit(&pp.timersLock, lockRankTimers)
5107 // This P may get timers when it starts running. Set the mask here
5108 // since the P may not go through pidleget (notably P 0 on startup).
5110 // Similarly, we may not go through pidleget before this P starts
5111 // running if it is P 0 on startup.
5115 // destroy releases all of the resources associated with pp and
5116 // transitions it to status _Pdead.
5118 // sched.lock must be held and the world must be stopped.
5119 func (pp *p) destroy() {
5120 assertLockHeld(&sched.lock)
5121 assertWorldStopped()
5123 // Move all runnable goroutines to the global queue
5124 for pp.runqhead != pp.runqtail {
5125 // Pop from tail of local queue
5127 gp := pp.runq[pp.runqtail%uint32(len(pp.runq))].ptr()
5128 // Push onto head of global queue
5131 if pp.runnext != 0 {
5132 globrunqputhead(pp.runnext.ptr())
5135 if len(pp.timers) > 0 {
5136 plocal := getg().m.p.ptr()
5137 // The world is stopped, but we acquire timersLock to
5138 // protect against sysmon calling timeSleepUntil.
5139 // This is the only case where we hold the timersLock of
5140 // more than one P, so there are no deadlock concerns.
5141 lock(&plocal.timersLock)
5142 lock(&pp.timersLock)
5143 moveTimers(plocal, pp.timers)
5145 pp.numTimers.Store(0)
5146 pp.deletedTimers.Store(0)
5147 pp.timer0When.Store(0)
5148 unlock(&pp.timersLock)
5149 unlock(&plocal.timersLock)
5151 // Flush p's write barrier buffer.
5152 if gcphase != _GCoff {
5156 for i := range pp.sudogbuf {
5157 pp.sudogbuf[i] = nil
5159 pp.sudogcache = pp.sudogbuf[:0]
5160 pp.pinnerCache = nil
5161 for j := range pp.deferpoolbuf {
5162 pp.deferpoolbuf[j] = nil
5164 pp.deferpool = pp.deferpoolbuf[:0]
5165 systemstack(func() {
5166 for i := 0; i < pp.mspancache.len; i++ {
5167 // Safe to call since the world is stopped.
5168 mheap_.spanalloc.free(unsafe.Pointer(pp.mspancache.buf[i]))
5170 pp.mspancache.len = 0
5172 pp.pcache.flush(&mheap_.pages)
5173 unlock(&mheap_.lock)
5175 freemcache(pp.mcache)
5180 if pp.timerRaceCtx != 0 {
5181 // The race detector code uses a callback to fetch
5182 // the proc context, so arrange for that callback
5183 // to see the right thing.
5184 // This hack only works because we are the only
5190 racectxend(pp.timerRaceCtx)
5195 raceprocdestroy(pp.raceprocctx)
5202 // Change number of processors.
5204 // sched.lock must be held, and the world must be stopped.
5206 // gcworkbufs must not be being modified by either the GC or the write barrier
5207 // code, so the GC must not be running if the number of Ps actually changes.
5209 // Returns list of Ps with local work, they need to be scheduled by the caller.
5210 func procresize(nprocs int32) *p {
5211 assertLockHeld(&sched.lock)
5212 assertWorldStopped()
5215 if old < 0 || nprocs <= 0 {
5216 throw("procresize: invalid arg")
5219 traceGomaxprocs(nprocs)
5222 // update statistics
5224 if sched.procresizetime != 0 {
5225 sched.totaltime += int64(old) * (now - sched.procresizetime)
5227 sched.procresizetime = now
5229 maskWords := (nprocs + 31) / 32
5231 // Grow allp if necessary.
5232 if nprocs > int32(len(allp)) {
5233 // Synchronize with retake, which could be running
5234 // concurrently since it doesn't run on a P.
5236 if nprocs <= int32(cap(allp)) {
5237 allp = allp[:nprocs]
5239 nallp := make([]*p, nprocs)
5240 // Copy everything up to allp's cap so we
5241 // never lose old allocated Ps.
5242 copy(nallp, allp[:cap(allp)])
5246 if maskWords <= int32(cap(idlepMask)) {
5247 idlepMask = idlepMask[:maskWords]
5248 timerpMask = timerpMask[:maskWords]
5250 nidlepMask := make([]uint32, maskWords)
5251 // No need to copy beyond len, old Ps are irrelevant.
5252 copy(nidlepMask, idlepMask)
5253 idlepMask = nidlepMask
5255 ntimerpMask := make([]uint32, maskWords)
5256 copy(ntimerpMask, timerpMask)
5257 timerpMask = ntimerpMask
5262 // initialize new P's
5263 for i := old; i < nprocs; i++ {
5269 atomicstorep(unsafe.Pointer(&allp[i]), unsafe.Pointer(pp))
5273 if gp.m.p != 0 && gp.m.p.ptr().id < nprocs {
5274 // continue to use the current P
5275 gp.m.p.ptr().status = _Prunning
5276 gp.m.p.ptr().mcache.prepareForSweep()
5278 // release the current P and acquire allp[0].
5280 // We must do this before destroying our current P
5281 // because p.destroy itself has write barriers, so we
5282 // need to do that from a valid P.
5285 // Pretend that we were descheduled
5286 // and then scheduled again to keep
5289 traceProcStop(gp.m.p.ptr())
5303 // g.m.p is now set, so we no longer need mcache0 for bootstrapping.
5306 // release resources from unused P's
5307 for i := nprocs; i < old; i++ {
5310 // can't free P itself because it can be referenced by an M in syscall
5314 if int32(len(allp)) != nprocs {
5316 allp = allp[:nprocs]
5317 idlepMask = idlepMask[:maskWords]
5318 timerpMask = timerpMask[:maskWords]
5323 for i := nprocs - 1; i >= 0; i-- {
5325 if gp.m.p.ptr() == pp {
5333 pp.link.set(runnablePs)
5337 stealOrder.reset(uint32(nprocs))
5338 var int32p *int32 = &gomaxprocs // make compiler check that gomaxprocs is an int32
5339 atomic.Store((*uint32)(unsafe.Pointer(int32p)), uint32(nprocs))
5341 // Notify the limiter that the amount of procs has changed.
5342 gcCPULimiter.resetCapacity(now, nprocs)
5347 // Associate p and the current m.
5349 // This function is allowed to have write barriers even if the caller
5350 // isn't because it immediately acquires pp.
5352 //go:yeswritebarrierrec
5353 func acquirep(pp *p) {
5354 // Do the part that isn't allowed to have write barriers.
5357 // Have p; write barriers now allowed.
5359 // Perform deferred mcache flush before this P can allocate
5360 // from a potentially stale mcache.
5361 pp.mcache.prepareForSweep()
5368 // wirep is the first step of acquirep, which actually associates the
5369 // current M to pp. This is broken out so we can disallow write
5370 // barriers for this part, since we don't yet have a P.
5372 //go:nowritebarrierrec
5378 throw("wirep: already in go")
5380 if pp.m != 0 || pp.status != _Pidle {
5385 print("wirep: p->m=", pp.m, "(", id, ") p->status=", pp.status, "\n")
5386 throw("wirep: invalid p state")
5390 pp.status = _Prunning
5393 // Disassociate p and the current m.
5394 func releasep() *p {
5398 throw("releasep: invalid arg")
5401 if pp.m.ptr() != gp.m || pp.status != _Prunning {
5402 print("releasep: m=", gp.m, " m->p=", gp.m.p.ptr(), " p->m=", hex(pp.m), " p->status=", pp.status, "\n")
5403 throw("releasep: invalid p state")
5406 traceProcStop(gp.m.p.ptr())
5414 func incidlelocked(v int32) {
5416 sched.nmidlelocked += v
5423 // Check for deadlock situation.
5424 // The check is based on number of running M's, if 0 -> deadlock.
5425 // sched.lock must be held.
5427 assertLockHeld(&sched.lock)
5429 // For -buildmode=c-shared or -buildmode=c-archive it's OK if
5430 // there are no running goroutines. The calling program is
5431 // assumed to be running.
5432 if islibrary || isarchive {
5436 // If we are dying because of a signal caught on an already idle thread,
5437 // freezetheworld will cause all running threads to block.
5438 // And runtime will essentially enter into deadlock state,
5439 // except that there is a thread that will call exit soon.
5440 if panicking.Load() > 0 {
5444 // If we are not running under cgo, but we have an extra M then account
5445 // for it. (It is possible to have an extra M on Windows without cgo to
5446 // accommodate callbacks created by syscall.NewCallback. See issue #6751
5449 if !iscgo && cgoHasExtraM && extraMLength.Load() > 0 {
5453 run := mcount() - sched.nmidle - sched.nmidlelocked - sched.nmsys
5458 print("runtime: checkdead: nmidle=", sched.nmidle, " nmidlelocked=", sched.nmidlelocked, " mcount=", mcount(), " nmsys=", sched.nmsys, "\n")
5460 throw("checkdead: inconsistent counts")
5464 forEachG(func(gp *g) {
5465 if isSystemGoroutine(gp, false) {
5468 s := readgstatus(gp)
5469 switch s &^ _Gscan {
5476 print("runtime: checkdead: find g ", gp.goid, " in status ", s, "\n")
5478 throw("checkdead: runnable g")
5481 if grunning == 0 { // possible if main goroutine calls runtime·Goexit()
5482 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5483 fatal("no goroutines (main called runtime.Goexit) - deadlock!")
5486 // Maybe jump time forward for playground.
5488 if when := timeSleepUntil(); when < maxWhen {
5491 // Start an M to steal the timer.
5492 pp, _ := pidleget(faketime)
5494 // There should always be a free P since
5495 // nothing is running.
5497 throw("checkdead: no p for timer")
5501 // There should always be a free M since
5502 // nothing is running.
5504 throw("checkdead: no m for timer")
5506 // M must be spinning to steal. We set this to be
5507 // explicit, but since this is the only M it would
5508 // become spinning on its own anyways.
5509 sched.nmspinning.Add(1)
5512 notewakeup(&mp.park)
5517 // There are no goroutines running, so we can look at the P's.
5518 for _, pp := range allp {
5519 if len(pp.timers) > 0 {
5524 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5525 fatal("all goroutines are asleep - deadlock!")
5528 // forcegcperiod is the maximum time in nanoseconds between garbage
5529 // collections. If we go this long without a garbage collection, one
5530 // is forced to run.
5532 // This is a variable for testing purposes. It normally doesn't change.
5533 var forcegcperiod int64 = 2 * 60 * 1e9
5535 // needSysmonWorkaround is true if the workaround for
5536 // golang.org/issue/42515 is needed on NetBSD.
5537 var needSysmonWorkaround bool = false
5539 // Always runs without a P, so write barriers are not allowed.
5541 //go:nowritebarrierrec
5548 lasttrace := int64(0)
5549 idle := 0 // how many cycles in succession we had not wokeup somebody
5553 if idle == 0 { // start with 20us sleep...
5555 } else if idle > 50 { // start doubling the sleep after 1ms...
5558 if delay > 10*1000 { // up to 10ms
5563 // sysmon should not enter deep sleep if schedtrace is enabled so that
5564 // it can print that information at the right time.
5566 // It should also not enter deep sleep if there are any active P's so
5567 // that it can retake P's from syscalls, preempt long running G's, and
5568 // poll the network if all P's are busy for long stretches.
5570 // It should wakeup from deep sleep if any P's become active either due
5571 // to exiting a syscall or waking up due to a timer expiring so that it
5572 // can resume performing those duties. If it wakes from a syscall it
5573 // resets idle and delay as a bet that since it had retaken a P from a
5574 // syscall before, it may need to do it again shortly after the
5575 // application starts work again. It does not reset idle when waking
5576 // from a timer to avoid adding system load to applications that spend
5577 // most of their time sleeping.
5579 if debug.schedtrace <= 0 && (sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs) {
5581 if sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs {
5582 syscallWake := false
5583 next := timeSleepUntil()
5585 sched.sysmonwait.Store(true)
5587 // Make wake-up period small enough
5588 // for the sampling to be correct.
5589 sleep := forcegcperiod / 2
5590 if next-now < sleep {
5593 shouldRelax := sleep >= osRelaxMinNS
5597 syscallWake = notetsleep(&sched.sysmonnote, sleep)
5602 sched.sysmonwait.Store(false)
5603 noteclear(&sched.sysmonnote)
5613 lock(&sched.sysmonlock)
5614 // Update now in case we blocked on sysmonnote or spent a long time
5615 // blocked on schedlock or sysmonlock above.
5618 // trigger libc interceptors if needed
5619 if *cgo_yield != nil {
5620 asmcgocall(*cgo_yield, nil)
5622 // poll network if not polled for more than 10ms
5623 lastpoll := sched.lastpoll.Load()
5624 if netpollinited() && lastpoll != 0 && lastpoll+10*1000*1000 < now {
5625 sched.lastpoll.CompareAndSwap(lastpoll, now)
5626 list, delta := netpoll(0) // non-blocking - returns list of goroutines
5628 // Need to decrement number of idle locked M's
5629 // (pretending that one more is running) before injectglist.
5630 // Otherwise it can lead to the following situation:
5631 // injectglist grabs all P's but before it starts M's to run the P's,
5632 // another M returns from syscall, finishes running its G,
5633 // observes that there is no work to do and no other running M's
5634 // and reports deadlock.
5638 netpollAdjustWaiters(delta)
5641 if GOOS == "netbsd" && needSysmonWorkaround {
5642 // netpoll is responsible for waiting for timer
5643 // expiration, so we typically don't have to worry
5644 // about starting an M to service timers. (Note that
5645 // sleep for timeSleepUntil above simply ensures sysmon
5646 // starts running again when that timer expiration may
5647 // cause Go code to run again).
5649 // However, netbsd has a kernel bug that sometimes
5650 // misses netpollBreak wake-ups, which can lead to
5651 // unbounded delays servicing timers. If we detect this
5652 // overrun, then startm to get something to handle the
5655 // See issue 42515 and
5656 // https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=50094.
5657 if next := timeSleepUntil(); next < now {
5658 startm(nil, false, false)
5661 if scavenger.sysmonWake.Load() != 0 {
5662 // Kick the scavenger awake if someone requested it.
5665 // retake P's blocked in syscalls
5666 // and preempt long running G's
5667 if retake(now) != 0 {
5672 // check if we need to force a GC
5673 if t := (gcTrigger{kind: gcTriggerTime, now: now}); t.test() && forcegc.idle.Load() {
5675 forcegc.idle.Store(false)
5677 list.push(forcegc.g)
5679 unlock(&forcegc.lock)
5681 if debug.schedtrace > 0 && lasttrace+int64(debug.schedtrace)*1000000 <= now {
5683 schedtrace(debug.scheddetail > 0)
5685 unlock(&sched.sysmonlock)
5689 type sysmontick struct {
5696 // forcePreemptNS is the time slice given to a G before it is
5698 const forcePreemptNS = 10 * 1000 * 1000 // 10ms
5700 func retake(now int64) uint32 {
5702 // Prevent allp slice changes. This lock will be completely
5703 // uncontended unless we're already stopping the world.
5705 // We can't use a range loop over allp because we may
5706 // temporarily drop the allpLock. Hence, we need to re-fetch
5707 // allp each time around the loop.
5708 for i := 0; i < len(allp); i++ {
5711 // This can happen if procresize has grown
5712 // allp but not yet created new Ps.
5715 pd := &pp.sysmontick
5718 if s == _Prunning || s == _Psyscall {
5719 // Preempt G if it's running for too long.
5720 t := int64(pp.schedtick)
5721 if int64(pd.schedtick) != t {
5722 pd.schedtick = uint32(t)
5724 } else if pd.schedwhen+forcePreemptNS <= now {
5726 // In case of syscall, preemptone() doesn't
5727 // work, because there is no M wired to P.
5732 // Retake P from syscall if it's there for more than 1 sysmon tick (at least 20us).
5733 t := int64(pp.syscalltick)
5734 if !sysretake && int64(pd.syscalltick) != t {
5735 pd.syscalltick = uint32(t)
5736 pd.syscallwhen = now
5739 // On the one hand we don't want to retake Ps if there is no other work to do,
5740 // but on the other hand we want to retake them eventually
5741 // because they can prevent the sysmon thread from deep sleep.
5742 if runqempty(pp) && sched.nmspinning.Load()+sched.npidle.Load() > 0 && pd.syscallwhen+10*1000*1000 > now {
5745 // Drop allpLock so we can take sched.lock.
5747 // Need to decrement number of idle locked M's
5748 // (pretending that one more is running) before the CAS.
5749 // Otherwise the M from which we retake can exit the syscall,
5750 // increment nmidle and report deadlock.
5752 if atomic.Cas(&pp.status, s, _Pidle) {
5769 // Tell all goroutines that they have been preempted and they should stop.
5770 // This function is purely best-effort. It can fail to inform a goroutine if a
5771 // processor just started running it.
5772 // No locks need to be held.
5773 // Returns true if preemption request was issued to at least one goroutine.
5774 func preemptall() bool {
5776 for _, pp := range allp {
5777 if pp.status != _Prunning {
5787 // Tell the goroutine running on processor P to stop.
5788 // This function is purely best-effort. It can incorrectly fail to inform the
5789 // goroutine. It can inform the wrong goroutine. Even if it informs the
5790 // correct goroutine, that goroutine might ignore the request if it is
5791 // simultaneously executing newstack.
5792 // No lock needs to be held.
5793 // Returns true if preemption request was issued.
5794 // The actual preemption will happen at some point in the future
5795 // and will be indicated by the gp->status no longer being
5797 func preemptone(pp *p) bool {
5799 if mp == nil || mp == getg().m {
5803 if gp == nil || gp == mp.g0 {
5809 // Every call in a goroutine checks for stack overflow by
5810 // comparing the current stack pointer to gp->stackguard0.
5811 // Setting gp->stackguard0 to StackPreempt folds
5812 // preemption into the normal stack overflow check.
5813 gp.stackguard0 = stackPreempt
5815 // Request an async preemption of this P.
5816 if preemptMSupported && debug.asyncpreemptoff == 0 {
5826 func schedtrace(detailed bool) {
5833 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)
5835 print(" gcwaiting=", sched.gcwaiting.Load(), " nmidlelocked=", sched.nmidlelocked, " stopwait=", sched.stopwait, " sysmonwait=", sched.sysmonwait.Load(), "\n")
5837 // We must be careful while reading data from P's, M's and G's.
5838 // Even if we hold schedlock, most data can be changed concurrently.
5839 // E.g. (p->m ? p->m->id : -1) can crash if p->m changes from non-nil to nil.
5840 for i, pp := range allp {
5842 h := atomic.Load(&pp.runqhead)
5843 t := atomic.Load(&pp.runqtail)
5845 print(" P", i, ": status=", pp.status, " schedtick=", pp.schedtick, " syscalltick=", pp.syscalltick, " m=")
5851 print(" runqsize=", t-h, " gfreecnt=", pp.gFree.n, " timerslen=", len(pp.timers), "\n")
5853 // In non-detailed mode format lengths of per-P run queues as:
5854 // [len1 len2 len3 len4]
5860 if i == len(allp)-1 {
5871 for mp := allm; mp != nil; mp = mp.alllink {
5873 print(" M", mp.id, ": p=")
5885 print(" mallocing=", mp.mallocing, " throwing=", mp.throwing, " preemptoff=", mp.preemptoff, " locks=", mp.locks, " dying=", mp.dying, " spinning=", mp.spinning, " blocked=", mp.blocked, " lockedg=")
5886 if lockedg := mp.lockedg.ptr(); lockedg != nil {
5894 forEachG(func(gp *g) {
5895 print(" G", gp.goid, ": status=", readgstatus(gp), "(", gp.waitreason.String(), ") m=")
5902 if lockedm := gp.lockedm.ptr(); lockedm != nil {
5912 // schedEnableUser enables or disables the scheduling of user
5915 // This does not stop already running user goroutines, so the caller
5916 // should first stop the world when disabling user goroutines.
5917 func schedEnableUser(enable bool) {
5919 if sched.disable.user == !enable {
5923 sched.disable.user = !enable
5925 n := sched.disable.n
5927 globrunqputbatch(&sched.disable.runnable, n)
5929 for ; n != 0 && sched.npidle.Load() != 0; n-- {
5930 startm(nil, false, false)
5937 // schedEnabled reports whether gp should be scheduled. It returns
5938 // false is scheduling of gp is disabled.
5940 // sched.lock must be held.
5941 func schedEnabled(gp *g) bool {
5942 assertLockHeld(&sched.lock)
5944 if sched.disable.user {
5945 return isSystemGoroutine(gp, true)
5950 // Put mp on midle list.
5951 // sched.lock must be held.
5952 // May run during STW, so write barriers are not allowed.
5954 //go:nowritebarrierrec
5956 assertLockHeld(&sched.lock)
5958 mp.schedlink = sched.midle
5964 // Try to get an m from midle list.
5965 // sched.lock must be held.
5966 // May run during STW, so write barriers are not allowed.
5968 //go:nowritebarrierrec
5970 assertLockHeld(&sched.lock)
5972 mp := sched.midle.ptr()
5974 sched.midle = mp.schedlink
5980 // Put gp on the global runnable queue.
5981 // sched.lock must be held.
5982 // May run during STW, so write barriers are not allowed.
5984 //go:nowritebarrierrec
5985 func globrunqput(gp *g) {
5986 assertLockHeld(&sched.lock)
5988 sched.runq.pushBack(gp)
5992 // Put gp at the head of the global runnable queue.
5993 // sched.lock must be held.
5994 // May run during STW, so write barriers are not allowed.
5996 //go:nowritebarrierrec
5997 func globrunqputhead(gp *g) {
5998 assertLockHeld(&sched.lock)
6004 // Put a batch of runnable goroutines on the global runnable queue.
6005 // This clears *batch.
6006 // sched.lock must be held.
6007 // May run during STW, so write barriers are not allowed.
6009 //go:nowritebarrierrec
6010 func globrunqputbatch(batch *gQueue, n int32) {
6011 assertLockHeld(&sched.lock)
6013 sched.runq.pushBackAll(*batch)
6018 // Try get a batch of G's from the global runnable queue.
6019 // sched.lock must be held.
6020 func globrunqget(pp *p, max int32) *g {
6021 assertLockHeld(&sched.lock)
6023 if sched.runqsize == 0 {
6027 n := sched.runqsize/gomaxprocs + 1
6028 if n > sched.runqsize {
6031 if max > 0 && n > max {
6034 if n > int32(len(pp.runq))/2 {
6035 n = int32(len(pp.runq)) / 2
6040 gp := sched.runq.pop()
6043 gp1 := sched.runq.pop()
6044 runqput(pp, gp1, false)
6049 // pMask is an atomic bitstring with one bit per P.
6052 // read returns true if P id's bit is set.
6053 func (p pMask) read(id uint32) bool {
6055 mask := uint32(1) << (id % 32)
6056 return (atomic.Load(&p[word]) & mask) != 0
6059 // set sets P id's bit.
6060 func (p pMask) set(id int32) {
6062 mask := uint32(1) << (id % 32)
6063 atomic.Or(&p[word], mask)
6066 // clear clears P id's bit.
6067 func (p pMask) clear(id int32) {
6069 mask := uint32(1) << (id % 32)
6070 atomic.And(&p[word], ^mask)
6073 // updateTimerPMask clears pp's timer mask if it has no timers on its heap.
6075 // Ideally, the timer mask would be kept immediately consistent on any timer
6076 // operations. Unfortunately, updating a shared global data structure in the
6077 // timer hot path adds too much overhead in applications frequently switching
6078 // between no timers and some timers.
6080 // As a compromise, the timer mask is updated only on pidleget / pidleput. A
6081 // running P (returned by pidleget) may add a timer at any time, so its mask
6082 // must be set. An idle P (passed to pidleput) cannot add new timers while
6083 // idle, so if it has no timers at that time, its mask may be cleared.
6085 // Thus, we get the following effects on timer-stealing in findrunnable:
6087 // - Idle Ps with no timers when they go idle are never checked in findrunnable
6088 // (for work- or timer-stealing; this is the ideal case).
6089 // - Running Ps must always be checked.
6090 // - Idle Ps whose timers are stolen must continue to be checked until they run
6091 // again, even after timer expiration.
6093 // When the P starts running again, the mask should be set, as a timer may be
6094 // added at any time.
6096 // TODO(prattmic): Additional targeted updates may improve the above cases.
6097 // e.g., updating the mask when stealing a timer.
6098 func updateTimerPMask(pp *p) {
6099 if pp.numTimers.Load() > 0 {
6103 // Looks like there are no timers, however another P may transiently
6104 // decrement numTimers when handling a timerModified timer in
6105 // checkTimers. We must take timersLock to serialize with these changes.
6106 lock(&pp.timersLock)
6107 if pp.numTimers.Load() == 0 {
6108 timerpMask.clear(pp.id)
6110 unlock(&pp.timersLock)
6113 // pidleput puts p on the _Pidle list. now must be a relatively recent call
6114 // to nanotime or zero. Returns now or the current time if now was zero.
6116 // This releases ownership of p. Once sched.lock is released it is no longer
6119 // sched.lock must be held.
6121 // May run during STW, so write barriers are not allowed.
6123 //go:nowritebarrierrec
6124 func pidleput(pp *p, now int64) int64 {
6125 assertLockHeld(&sched.lock)
6128 throw("pidleput: P has non-empty run queue")
6133 updateTimerPMask(pp) // clear if there are no timers.
6134 idlepMask.set(pp.id)
6135 pp.link = sched.pidle
6138 if !pp.limiterEvent.start(limiterEventIdle, now) {
6139 throw("must be able to track idle limiter event")
6144 // pidleget tries to get a p from the _Pidle list, acquiring ownership.
6146 // sched.lock must be held.
6148 // May run during STW, so write barriers are not allowed.
6150 //go:nowritebarrierrec
6151 func pidleget(now int64) (*p, int64) {
6152 assertLockHeld(&sched.lock)
6154 pp := sched.pidle.ptr()
6156 // Timer may get added at any time now.
6160 timerpMask.set(pp.id)
6161 idlepMask.clear(pp.id)
6162 sched.pidle = pp.link
6163 sched.npidle.Add(-1)
6164 pp.limiterEvent.stop(limiterEventIdle, now)
6169 // pidlegetSpinning tries to get a p from the _Pidle list, acquiring ownership.
6170 // This is called by spinning Ms (or callers than need a spinning M) that have
6171 // found work. If no P is available, this must synchronized with non-spinning
6172 // Ms that may be preparing to drop their P without discovering this work.
6174 // sched.lock must be held.
6176 // May run during STW, so write barriers are not allowed.
6178 //go:nowritebarrierrec
6179 func pidlegetSpinning(now int64) (*p, int64) {
6180 assertLockHeld(&sched.lock)
6182 pp, now := pidleget(now)
6184 // See "Delicate dance" comment in findrunnable. We found work
6185 // that we cannot take, we must synchronize with non-spinning
6186 // Ms that may be preparing to drop their P.
6187 sched.needspinning.Store(1)
6194 // runqempty reports whether pp has no Gs on its local run queue.
6195 // It never returns true spuriously.
6196 func runqempty(pp *p) bool {
6197 // Defend against a race where 1) pp has G1 in runqnext but runqhead == runqtail,
6198 // 2) runqput on pp kicks G1 to the runq, 3) runqget on pp empties runqnext.
6199 // Simply observing that runqhead == runqtail and then observing that runqnext == nil
6200 // does not mean the queue is empty.
6202 head := atomic.Load(&pp.runqhead)
6203 tail := atomic.Load(&pp.runqtail)
6204 runnext := atomic.Loaduintptr((*uintptr)(unsafe.Pointer(&pp.runnext)))
6205 if tail == atomic.Load(&pp.runqtail) {
6206 return head == tail && runnext == 0
6211 // To shake out latent assumptions about scheduling order,
6212 // we introduce some randomness into scheduling decisions
6213 // when running with the race detector.
6214 // The need for this was made obvious by changing the
6215 // (deterministic) scheduling order in Go 1.5 and breaking
6216 // many poorly-written tests.
6217 // With the randomness here, as long as the tests pass
6218 // consistently with -race, they shouldn't have latent scheduling
6220 const randomizeScheduler = raceenabled
6222 // runqput tries to put g on the local runnable queue.
6223 // If next is false, runqput adds g to the tail of the runnable queue.
6224 // If next is true, runqput puts g in the pp.runnext slot.
6225 // If the run queue is full, runnext puts g on the global queue.
6226 // Executed only by the owner P.
6227 func runqput(pp *p, gp *g, next bool) {
6228 if randomizeScheduler && next && fastrandn(2) == 0 {
6234 oldnext := pp.runnext
6235 if !pp.runnext.cas(oldnext, guintptr(unsafe.Pointer(gp))) {
6241 // Kick the old runnext out to the regular run queue.
6246 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
6248 if t-h < uint32(len(pp.runq)) {
6249 pp.runq[t%uint32(len(pp.runq))].set(gp)
6250 atomic.StoreRel(&pp.runqtail, t+1) // store-release, makes the item available for consumption
6253 if runqputslow(pp, gp, h, t) {
6256 // the queue is not full, now the put above must succeed
6260 // Put g and a batch of work from local runnable queue on global queue.
6261 // Executed only by the owner P.
6262 func runqputslow(pp *p, gp *g, h, t uint32) bool {
6263 var batch [len(pp.runq)/2 + 1]*g
6265 // First, grab a batch from local queue.
6268 if n != uint32(len(pp.runq)/2) {
6269 throw("runqputslow: queue is not full")
6271 for i := uint32(0); i < n; i++ {
6272 batch[i] = pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
6274 if !atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6279 if randomizeScheduler {
6280 for i := uint32(1); i <= n; i++ {
6281 j := fastrandn(i + 1)
6282 batch[i], batch[j] = batch[j], batch[i]
6286 // Link the goroutines.
6287 for i := uint32(0); i < n; i++ {
6288 batch[i].schedlink.set(batch[i+1])
6291 q.head.set(batch[0])
6292 q.tail.set(batch[n])
6294 // Now put the batch on global queue.
6296 globrunqputbatch(&q, int32(n+1))
6301 // runqputbatch tries to put all the G's on q on the local runnable queue.
6302 // If the queue is full, they are put on the global queue; in that case
6303 // this will temporarily acquire the scheduler lock.
6304 // Executed only by the owner P.
6305 func runqputbatch(pp *p, q *gQueue, qsize int) {
6306 h := atomic.LoadAcq(&pp.runqhead)
6309 for !q.empty() && t-h < uint32(len(pp.runq)) {
6311 pp.runq[t%uint32(len(pp.runq))].set(gp)
6317 if randomizeScheduler {
6318 off := func(o uint32) uint32 {
6319 return (pp.runqtail + o) % uint32(len(pp.runq))
6321 for i := uint32(1); i < n; i++ {
6322 j := fastrandn(i + 1)
6323 pp.runq[off(i)], pp.runq[off(j)] = pp.runq[off(j)], pp.runq[off(i)]
6327 atomic.StoreRel(&pp.runqtail, t)
6330 globrunqputbatch(q, int32(qsize))
6335 // Get g from local runnable queue.
6336 // If inheritTime is true, gp should inherit the remaining time in the
6337 // current time slice. Otherwise, it should start a new time slice.
6338 // Executed only by the owner P.
6339 func runqget(pp *p) (gp *g, inheritTime bool) {
6340 // If there's a runnext, it's the next G to run.
6342 // If the runnext is non-0 and the CAS fails, it could only have been stolen by another P,
6343 // because other Ps can race to set runnext to 0, but only the current P can set it to non-0.
6344 // Hence, there's no need to retry this CAS if it fails.
6345 if next != 0 && pp.runnext.cas(next, 0) {
6346 return next.ptr(), true
6350 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6355 gp := pp.runq[h%uint32(len(pp.runq))].ptr()
6356 if atomic.CasRel(&pp.runqhead, h, h+1) { // cas-release, commits consume
6362 // runqdrain drains the local runnable queue of pp and returns all goroutines in it.
6363 // Executed only by the owner P.
6364 func runqdrain(pp *p) (drainQ gQueue, n uint32) {
6365 oldNext := pp.runnext
6366 if oldNext != 0 && pp.runnext.cas(oldNext, 0) {
6367 drainQ.pushBack(oldNext.ptr())
6372 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6378 if qn > uint32(len(pp.runq)) { // read inconsistent h and t
6382 if !atomic.CasRel(&pp.runqhead, h, h+qn) { // cas-release, commits consume
6386 // We've inverted the order in which it gets G's from the local P's runnable queue
6387 // and then advances the head pointer because we don't want to mess up the statuses of G's
6388 // while runqdrain() and runqsteal() are running in parallel.
6389 // Thus we should advance the head pointer before draining the local P into a gQueue,
6390 // so that we can update any gp.schedlink only after we take the full ownership of G,
6391 // meanwhile, other P's can't access to all G's in local P's runnable queue and steal them.
6392 // See https://groups.google.com/g/golang-dev/c/0pTKxEKhHSc/m/6Q85QjdVBQAJ for more details.
6393 for i := uint32(0); i < qn; i++ {
6394 gp := pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
6401 // Grabs a batch of goroutines from pp's runnable queue into batch.
6402 // Batch is a ring buffer starting at batchHead.
6403 // Returns number of grabbed goroutines.
6404 // Can be executed by any P.
6405 func runqgrab(pp *p, batch *[256]guintptr, batchHead uint32, stealRunNextG bool) uint32 {
6407 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6408 t := atomic.LoadAcq(&pp.runqtail) // load-acquire, synchronize with the producer
6413 // Try to steal from pp.runnext.
6414 if next := pp.runnext; next != 0 {
6415 if pp.status == _Prunning {
6416 // Sleep to ensure that pp isn't about to run the g
6417 // we are about to steal.
6418 // The important use case here is when the g running
6419 // on pp ready()s another g and then almost
6420 // immediately blocks. Instead of stealing runnext
6421 // in this window, back off to give pp a chance to
6422 // schedule runnext. This will avoid thrashing gs
6423 // between different Ps.
6424 // A sync chan send/recv takes ~50ns as of time of
6425 // writing, so 3us gives ~50x overshoot.
6426 if GOOS != "windows" && GOOS != "openbsd" && GOOS != "netbsd" {
6429 // On some platforms system timer granularity is
6430 // 1-15ms, which is way too much for this
6431 // optimization. So just yield.
6435 if !pp.runnext.cas(next, 0) {
6438 batch[batchHead%uint32(len(batch))] = next
6444 if n > uint32(len(pp.runq)/2) { // read inconsistent h and t
6447 for i := uint32(0); i < n; i++ {
6448 g := pp.runq[(h+i)%uint32(len(pp.runq))]
6449 batch[(batchHead+i)%uint32(len(batch))] = g
6451 if atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6457 // Steal half of elements from local runnable queue of p2
6458 // and put onto local runnable queue of p.
6459 // Returns one of the stolen elements (or nil if failed).
6460 func runqsteal(pp, p2 *p, stealRunNextG bool) *g {
6462 n := runqgrab(p2, &pp.runq, t, stealRunNextG)
6467 gp := pp.runq[(t+n)%uint32(len(pp.runq))].ptr()
6471 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
6472 if t-h+n >= uint32(len(pp.runq)) {
6473 throw("runqsteal: runq overflow")
6475 atomic.StoreRel(&pp.runqtail, t+n) // store-release, makes the item available for consumption
6479 // A gQueue is a dequeue of Gs linked through g.schedlink. A G can only
6480 // be on one gQueue or gList at a time.
6481 type gQueue struct {
6486 // empty reports whether q is empty.
6487 func (q *gQueue) empty() bool {
6491 // push adds gp to the head of q.
6492 func (q *gQueue) push(gp *g) {
6493 gp.schedlink = q.head
6500 // pushBack adds gp to the tail of q.
6501 func (q *gQueue) pushBack(gp *g) {
6504 q.tail.ptr().schedlink.set(gp)
6511 // pushBackAll adds all Gs in q2 to the tail of q. After this q2 must
6513 func (q *gQueue) pushBackAll(q2 gQueue) {
6517 q2.tail.ptr().schedlink = 0
6519 q.tail.ptr().schedlink = q2.head
6526 // pop removes and returns the head of queue q. It returns nil if
6528 func (q *gQueue) pop() *g {
6531 q.head = gp.schedlink
6539 // popList takes all Gs in q and returns them as a gList.
6540 func (q *gQueue) popList() gList {
6541 stack := gList{q.head}
6546 // A gList is a list of Gs linked through g.schedlink. A G can only be
6547 // on one gQueue or gList at a time.
6552 // empty reports whether l is empty.
6553 func (l *gList) empty() bool {
6557 // push adds gp to the head of l.
6558 func (l *gList) push(gp *g) {
6559 gp.schedlink = l.head
6563 // pushAll prepends all Gs in q to l.
6564 func (l *gList) pushAll(q gQueue) {
6566 q.tail.ptr().schedlink = l.head
6571 // pop removes and returns the head of l. If l is empty, it returns nil.
6572 func (l *gList) pop() *g {
6575 l.head = gp.schedlink
6580 //go:linkname setMaxThreads runtime/debug.setMaxThreads
6581 func setMaxThreads(in int) (out int) {
6583 out = int(sched.maxmcount)
6584 if in > 0x7fffffff { // MaxInt32
6585 sched.maxmcount = 0x7fffffff
6587 sched.maxmcount = int32(in)
6595 func procPin() int {
6600 return int(mp.p.ptr().id)
6609 //go:linkname sync_runtime_procPin sync.runtime_procPin
6611 func sync_runtime_procPin() int {
6615 //go:linkname sync_runtime_procUnpin sync.runtime_procUnpin
6617 func sync_runtime_procUnpin() {
6621 //go:linkname sync_atomic_runtime_procPin sync/atomic.runtime_procPin
6623 func sync_atomic_runtime_procPin() int {
6627 //go:linkname sync_atomic_runtime_procUnpin sync/atomic.runtime_procUnpin
6629 func sync_atomic_runtime_procUnpin() {
6633 // Active spinning for sync.Mutex.
6635 //go:linkname sync_runtime_canSpin sync.runtime_canSpin
6637 func sync_runtime_canSpin(i int) bool {
6638 // sync.Mutex is cooperative, so we are conservative with spinning.
6639 // Spin only few times and only if running on a multicore machine and
6640 // GOMAXPROCS>1 and there is at least one other running P and local runq is empty.
6641 // As opposed to runtime mutex we don't do passive spinning here,
6642 // because there can be work on global runq or on other Ps.
6643 if i >= active_spin || ncpu <= 1 || gomaxprocs <= sched.npidle.Load()+sched.nmspinning.Load()+1 {
6646 if p := getg().m.p.ptr(); !runqempty(p) {
6652 //go:linkname sync_runtime_doSpin sync.runtime_doSpin
6654 func sync_runtime_doSpin() {
6655 procyield(active_spin_cnt)
6658 var stealOrder randomOrder
6660 // randomOrder/randomEnum are helper types for randomized work stealing.
6661 // They allow to enumerate all Ps in different pseudo-random orders without repetitions.
6662 // The algorithm is based on the fact that if we have X such that X and GOMAXPROCS
6663 // are coprime, then a sequences of (i + X) % GOMAXPROCS gives the required enumeration.
6664 type randomOrder struct {
6669 type randomEnum struct {
6676 func (ord *randomOrder) reset(count uint32) {
6678 ord.coprimes = ord.coprimes[:0]
6679 for i := uint32(1); i <= count; i++ {
6680 if gcd(i, count) == 1 {
6681 ord.coprimes = append(ord.coprimes, i)
6686 func (ord *randomOrder) start(i uint32) randomEnum {
6690 inc: ord.coprimes[i/ord.count%uint32(len(ord.coprimes))],
6694 func (enum *randomEnum) done() bool {
6695 return enum.i == enum.count
6698 func (enum *randomEnum) next() {
6700 enum.pos = (enum.pos + enum.inc) % enum.count
6703 func (enum *randomEnum) position() uint32 {
6707 func gcd(a, b uint32) uint32 {
6714 // An initTask represents the set of initializations that need to be done for a package.
6715 // Keep in sync with ../../test/noinit.go:initTask
6716 type initTask struct {
6717 state uint32 // 0 = uninitialized, 1 = in progress, 2 = done
6719 // followed by nfns pcs, uintptr sized, one per init function to run
6722 // inittrace stores statistics for init functions which are
6723 // updated by malloc and newproc when active is true.
6724 var inittrace tracestat
6726 type tracestat struct {
6727 active bool // init tracing activation status
6728 id uint64 // init goroutine id
6729 allocs uint64 // heap allocations
6730 bytes uint64 // heap allocated bytes
6733 func doInit(ts []*initTask) {
6734 for _, t := range ts {
6739 func doInit1(t *initTask) {
6741 case 2: // fully initialized
6743 case 1: // initialization in progress
6744 throw("recursive call during initialization - linker skew")
6745 default: // not initialized yet
6746 t.state = 1 // initialization in progress
6753 if inittrace.active {
6755 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6760 // We should have pruned all of these in the linker.
6761 throw("inittask with no functions")
6764 firstFunc := add(unsafe.Pointer(t), 8)
6765 for i := uint32(0); i < t.nfns; i++ {
6766 p := add(firstFunc, uintptr(i)*goarch.PtrSize)
6767 f := *(*func())(unsafe.Pointer(&p))
6771 if inittrace.active {
6773 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6776 f := *(*func())(unsafe.Pointer(&firstFunc))
6777 pkg := funcpkgpath(findfunc(abi.FuncPCABIInternal(f)))
6780 print("init ", pkg, " @")
6781 print(string(fmtNSAsMS(sbuf[:], uint64(start-runtimeInitTime))), " ms, ")
6782 print(string(fmtNSAsMS(sbuf[:], uint64(end-start))), " ms clock, ")
6783 print(string(itoa(sbuf[:], after.bytes-before.bytes)), " bytes, ")
6784 print(string(itoa(sbuf[:], after.allocs-before.allocs)), " allocs")
6788 t.state = 2 // initialization done