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). Run through the modules in dependency
248 // order (the order they are initialized by the dynamic
249 // loader, i.e. they are added to the moduledata linked list).
250 for m := &firstmoduledata; m != nil; m = m.next {
254 // Disable init tracing after main init done to avoid overhead
255 // of collecting statistics in malloc and newproc
256 inittrace.active = false
258 close(main_init_done)
263 if isarchive || islibrary {
264 // A program compiled with -buildmode=c-archive or c-shared
265 // has a main, but it is not executed.
268 fn := main_main // make an indirect call, as the linker doesn't know the address of the main package when laying down the runtime
271 runExitHooks(0) // run hooks now, since racefini does not return
275 // Make racy client program work: if panicking on
276 // another goroutine at the same time as main returns,
277 // let the other goroutine finish printing the panic trace.
278 // Once it does, it will exit. See issues 3934 and 20018.
279 if runningPanicDefers.Load() != 0 {
280 // Running deferred functions should not take long.
281 for c := 0; c < 1000; c++ {
282 if runningPanicDefers.Load() == 0 {
288 if panicking.Load() != 0 {
289 gopark(nil, nil, waitReasonPanicWait, traceBlockForever, 1)
300 // os_beforeExit is called from os.Exit(0).
302 //go:linkname os_beforeExit os.runtime_beforeExit
303 func os_beforeExit(exitCode int) {
304 runExitHooks(exitCode)
305 if exitCode == 0 && raceenabled {
310 // start forcegc helper goroutine
315 func forcegchelper() {
317 lockInit(&forcegc.lock, lockRankForcegc)
320 if forcegc.idle.Load() {
321 throw("forcegc: phase error")
323 forcegc.idle.Store(true)
324 goparkunlock(&forcegc.lock, waitReasonForceGCIdle, traceBlockSystemGoroutine, 1)
325 // this goroutine is explicitly resumed by sysmon
326 if debug.gctrace > 0 {
329 // Time-triggered, fully concurrent.
330 gcStart(gcTrigger{kind: gcTriggerTime, now: nanotime()})
334 // Gosched yields the processor, allowing other goroutines to run. It does not
335 // suspend the current goroutine, so execution resumes automatically.
343 // goschedguarded yields the processor like gosched, but also checks
344 // for forbidden states and opts out of the yield in those cases.
347 func goschedguarded() {
348 mcall(goschedguarded_m)
351 // goschedIfBusy yields the processor like gosched, but only does so if
352 // there are no idle Ps or if we're on the only P and there's nothing in
353 // the run queue. In both cases, there is freely available idle time.
356 func goschedIfBusy() {
358 // Call gosched if gp.preempt is set; we may be in a tight loop that
359 // doesn't otherwise yield.
360 if !gp.preempt && sched.npidle.Load() > 0 {
366 // Puts the current goroutine into a waiting state and calls unlockf on the
369 // If unlockf returns false, the goroutine is resumed.
371 // unlockf must not access this G's stack, as it may be moved between
372 // the call to gopark and the call to unlockf.
374 // Note that because unlockf is called after putting the G into a waiting
375 // state, the G may have already been readied by the time unlockf is called
376 // unless there is external synchronization preventing the G from being
377 // readied. If unlockf returns false, it must guarantee that the G cannot be
378 // externally readied.
380 // Reason explains why the goroutine has been parked. It is displayed in stack
381 // traces and heap dumps. Reasons should be unique and descriptive. Do not
382 // re-use reasons, add new ones.
383 func gopark(unlockf func(*g, unsafe.Pointer) bool, lock unsafe.Pointer, reason waitReason, traceReason traceBlockReason, traceskip int) {
384 if reason != waitReasonSleep {
385 checkTimeouts() // timeouts may expire while two goroutines keep the scheduler busy
389 status := readgstatus(gp)
390 if status != _Grunning && status != _Gscanrunning {
391 throw("gopark: bad g status")
394 mp.waitunlockf = unlockf
395 gp.waitreason = reason
396 mp.waitTraceBlockReason = traceReason
397 mp.waitTraceSkip = traceskip
399 // can't do anything that might move the G between Ms here.
403 // Puts the current goroutine into a waiting state and unlocks the lock.
404 // The goroutine can be made runnable again by calling goready(gp).
405 func goparkunlock(lock *mutex, reason waitReason, traceReason traceBlockReason, traceskip int) {
406 gopark(parkunlock_c, unsafe.Pointer(lock), reason, traceReason, traceskip)
409 func goready(gp *g, traceskip int) {
411 ready(gp, traceskip, true)
416 func acquireSudog() *sudog {
417 // Delicate dance: the semaphore implementation calls
418 // acquireSudog, acquireSudog calls new(sudog),
419 // new calls malloc, malloc can call the garbage collector,
420 // and the garbage collector calls the semaphore implementation
422 // Break the cycle by doing acquirem/releasem around new(sudog).
423 // The acquirem/releasem increments m.locks during new(sudog),
424 // which keeps the garbage collector from being invoked.
427 if len(pp.sudogcache) == 0 {
428 lock(&sched.sudoglock)
429 // First, try to grab a batch from central cache.
430 for len(pp.sudogcache) < cap(pp.sudogcache)/2 && sched.sudogcache != nil {
431 s := sched.sudogcache
432 sched.sudogcache = s.next
434 pp.sudogcache = append(pp.sudogcache, s)
436 unlock(&sched.sudoglock)
437 // If the central cache is empty, allocate a new one.
438 if len(pp.sudogcache) == 0 {
439 pp.sudogcache = append(pp.sudogcache, new(sudog))
442 n := len(pp.sudogcache)
443 s := pp.sudogcache[n-1]
444 pp.sudogcache[n-1] = nil
445 pp.sudogcache = pp.sudogcache[:n-1]
447 throw("acquireSudog: found s.elem != nil in cache")
454 func releaseSudog(s *sudog) {
456 throw("runtime: sudog with non-nil elem")
459 throw("runtime: sudog with non-false isSelect")
462 throw("runtime: sudog with non-nil next")
465 throw("runtime: sudog with non-nil prev")
467 if s.waitlink != nil {
468 throw("runtime: sudog with non-nil waitlink")
471 throw("runtime: sudog with non-nil c")
475 throw("runtime: releaseSudog with non-nil gp.param")
477 mp := acquirem() // avoid rescheduling to another P
479 if len(pp.sudogcache) == cap(pp.sudogcache) {
480 // Transfer half of local cache to the central cache.
481 var first, last *sudog
482 for len(pp.sudogcache) > cap(pp.sudogcache)/2 {
483 n := len(pp.sudogcache)
484 p := pp.sudogcache[n-1]
485 pp.sudogcache[n-1] = nil
486 pp.sudogcache = pp.sudogcache[:n-1]
494 lock(&sched.sudoglock)
495 last.next = sched.sudogcache
496 sched.sudogcache = first
497 unlock(&sched.sudoglock)
499 pp.sudogcache = append(pp.sudogcache, s)
503 // called from assembly.
504 func badmcall(fn func(*g)) {
505 throw("runtime: mcall called on m->g0 stack")
508 func badmcall2(fn func(*g)) {
509 throw("runtime: mcall function returned")
512 func badreflectcall() {
513 panic(plainError("arg size to reflect.call more than 1GB"))
517 //go:nowritebarrierrec
518 func badmorestackg0() {
519 if !crashStackImplemented {
520 writeErrStr("fatal: morestack on g0\n")
525 switchToCrashStack(func() {
526 print("runtime: morestack on g0, stack [", hex(g.stack.lo), " ", hex(g.stack.hi), "], sp=", hex(g.sched.sp), ", called from\n")
527 g.m.traceback = 2 // include pc and sp in stack trace
528 traceback1(g.sched.pc, g.sched.sp, g.sched.lr, g, 0)
531 throw("morestack on g0")
536 //go:nowritebarrierrec
537 func badmorestackgsignal() {
538 writeErrStr("fatal: morestack on gsignal\n")
546 // gcrash is a fake g that can be used when crashing due to bad
550 var crashingG atomic.Pointer[g]
552 // Switch to crashstack and call fn, with special handling of
553 // concurrent and recursive cases.
555 // Nosplit as it is called in a bad stack condition (we know
556 // morestack would fail).
559 //go:nowritebarrierrec
560 func switchToCrashStack(fn func()) {
562 if crashingG.CompareAndSwapNoWB(nil, me) {
563 switchToCrashStack0(fn) // should never return
566 if crashingG.Load() == me {
567 // recursive crashing. too bad.
568 writeErrStr("fatal: recursive switchToCrashStack\n")
571 // Another g is crashing. Give it some time, hopefully it will finish traceback.
573 writeErrStr("fatal: concurrent switchToCrashStack\n")
577 const crashStackImplemented = GOARCH == "amd64" || GOARCH == "arm64" || GOARCH == "mips64" || GOARCH == "mips64le" || GOARCH == "riscv64"
580 func switchToCrashStack0(fn func()) // in assembly
582 func lockedOSThread() bool {
584 return gp.lockedm != 0 && gp.m.lockedg != 0
588 // allgs contains all Gs ever created (including dead Gs), and thus
591 // Access via the slice is protected by allglock or stop-the-world.
592 // Readers that cannot take the lock may (carefully!) use the atomic
597 // allglen and allgptr are atomic variables that contain len(allgs) and
598 // &allgs[0] respectively. Proper ordering depends on totally-ordered
599 // loads and stores. Writes are protected by allglock.
601 // allgptr is updated before allglen. Readers should read allglen
602 // before allgptr to ensure that allglen is always <= len(allgptr). New
603 // Gs appended during the race can be missed. For a consistent view of
604 // all Gs, allglock must be held.
606 // allgptr copies should always be stored as a concrete type or
607 // unsafe.Pointer, not uintptr, to ensure that GC can still reach it
608 // even if it points to a stale array.
613 func allgadd(gp *g) {
614 if readgstatus(gp) == _Gidle {
615 throw("allgadd: bad status Gidle")
619 allgs = append(allgs, gp)
620 if &allgs[0] != allgptr {
621 atomicstorep(unsafe.Pointer(&allgptr), unsafe.Pointer(&allgs[0]))
623 atomic.Storeuintptr(&allglen, uintptr(len(allgs)))
627 // allGsSnapshot returns a snapshot of the slice of all Gs.
629 // The world must be stopped or allglock must be held.
630 func allGsSnapshot() []*g {
631 assertWorldStoppedOrLockHeld(&allglock)
633 // Because the world is stopped or allglock is held, allgadd
634 // cannot happen concurrently with this. allgs grows
635 // monotonically and existing entries never change, so we can
636 // simply return a copy of the slice header. For added safety,
637 // we trim everything past len because that can still change.
638 return allgs[:len(allgs):len(allgs)]
641 // atomicAllG returns &allgs[0] and len(allgs) for use with atomicAllGIndex.
642 func atomicAllG() (**g, uintptr) {
643 length := atomic.Loaduintptr(&allglen)
644 ptr := (**g)(atomic.Loadp(unsafe.Pointer(&allgptr)))
648 // atomicAllGIndex returns ptr[i] with the allgptr returned from atomicAllG.
649 func atomicAllGIndex(ptr **g, i uintptr) *g {
650 return *(**g)(add(unsafe.Pointer(ptr), i*goarch.PtrSize))
653 // forEachG calls fn on every G from allgs.
655 // forEachG takes a lock to exclude concurrent addition of new Gs.
656 func forEachG(fn func(gp *g)) {
658 for _, gp := range allgs {
664 // forEachGRace calls fn on every G from allgs.
666 // forEachGRace avoids locking, but does not exclude addition of new Gs during
667 // execution, which may be missed.
668 func forEachGRace(fn func(gp *g)) {
669 ptr, length := atomicAllG()
670 for i := uintptr(0); i < length; i++ {
671 gp := atomicAllGIndex(ptr, i)
678 // Number of goroutine ids to grab from sched.goidgen to local per-P cache at once.
679 // 16 seems to provide enough amortization, but other than that it's mostly arbitrary number.
683 // cpuinit sets up CPU feature flags and calls internal/cpu.Initialize. env should be the complete
684 // value of the GODEBUG environment variable.
685 func cpuinit(env string) {
687 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
688 cpu.DebugOptions = true
692 // Support cpu feature variables are used in code generated by the compiler
693 // to guard execution of instructions that can not be assumed to be always supported.
696 x86HasPOPCNT = cpu.X86.HasPOPCNT
697 x86HasSSE41 = cpu.X86.HasSSE41
698 x86HasFMA = cpu.X86.HasFMA
701 armHasVFPv4 = cpu.ARM.HasVFPv4
704 arm64HasATOMICS = cpu.ARM64.HasATOMICS
708 // getGodebugEarly extracts the environment variable GODEBUG from the environment on
709 // Unix-like operating systems and returns it. This function exists to extract GODEBUG
710 // early before much of the runtime is initialized.
711 func getGodebugEarly() string {
712 const prefix = "GODEBUG="
715 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
716 // Similar to goenv_unix but extracts the environment value for
718 // TODO(moehrmann): remove when general goenvs() can be called before cpuinit()
720 for argv_index(argv, argc+1+n) != nil {
724 for i := int32(0); i < n; i++ {
725 p := argv_index(argv, argc+1+i)
726 s := unsafe.String(p, findnull(p))
728 if hasPrefix(s, prefix) {
729 env = gostring(p)[len(prefix):]
737 // The bootstrap sequence is:
741 // make & queue new G
742 // call runtime·mstart
744 // The new G calls runtime·main.
746 lockInit(&sched.lock, lockRankSched)
747 lockInit(&sched.sysmonlock, lockRankSysmon)
748 lockInit(&sched.deferlock, lockRankDefer)
749 lockInit(&sched.sudoglock, lockRankSudog)
750 lockInit(&deadlock, lockRankDeadlock)
751 lockInit(&paniclk, lockRankPanic)
752 lockInit(&allglock, lockRankAllg)
753 lockInit(&allpLock, lockRankAllp)
754 lockInit(&reflectOffs.lock, lockRankReflectOffs)
755 lockInit(&finlock, lockRankFin)
756 lockInit(&cpuprof.lock, lockRankCpuprof)
758 // Enforce that this lock is always a leaf lock.
759 // All of this lock's critical sections should be
761 lockInit(&memstats.heapStats.noPLock, lockRankLeafRank)
763 // raceinit must be the first call to race detector.
764 // In particular, it must be done before mallocinit below calls racemapshadow.
767 gp.racectx, raceprocctx0 = raceinit()
770 sched.maxmcount = 10000
772 // The world starts stopped.
778 godebug := getGodebugEarly()
779 initPageTrace(godebug) // must run after mallocinit but before anything allocates
780 cpuinit(godebug) // must run before alginit
781 alginit() // maps, hash, fastrand must not be used before this call
782 fastrandinit() // must run before mcommoninit
783 mcommoninit(gp.m, -1)
784 modulesinit() // provides activeModules
785 typelinksinit() // uses maps, activeModules
786 itabsinit() // uses activeModules
787 stkobjinit() // must run before GC starts
789 sigsave(&gp.m.sigmask)
790 initSigmask = gp.m.sigmask
799 // Allocate stack space that can be used when crashing due to bad stack
800 // conditions, e.g. morestack on g0.
801 gcrash.stack = stackalloc(16384)
802 gcrash.stackguard0 = gcrash.stack.lo + 1000
803 gcrash.stackguard1 = gcrash.stack.lo + 1000
805 // if disableMemoryProfiling is set, update MemProfileRate to 0 to turn off memprofile.
806 // Note: parsedebugvars may update MemProfileRate, but when disableMemoryProfiling is
807 // set to true by the linker, it means that nothing is consuming the profile, it is
808 // safe to set MemProfileRate to 0.
809 if disableMemoryProfiling {
814 sched.lastpoll.Store(nanotime())
816 if n, ok := atoi32(gogetenv("GOMAXPROCS")); ok && n > 0 {
819 if procresize(procs) != nil {
820 throw("unknown runnable goroutine during bootstrap")
824 // World is effectively started now, as P's can run.
827 if buildVersion == "" {
828 // Condition should never trigger. This code just serves
829 // to ensure runtime·buildVersion is kept in the resulting binary.
830 buildVersion = "unknown"
832 if len(modinfo) == 1 {
833 // Condition should never trigger. This code just serves
834 // to ensure runtime·modinfo is kept in the resulting binary.
839 func dumpgstatus(gp *g) {
841 print("runtime: gp: gp=", gp, ", goid=", gp.goid, ", gp->atomicstatus=", readgstatus(gp), "\n")
842 print("runtime: getg: g=", thisg, ", goid=", thisg.goid, ", g->atomicstatus=", readgstatus(thisg), "\n")
845 // sched.lock must be held.
847 assertLockHeld(&sched.lock)
849 // Exclude extra M's, which are used for cgocallback from threads
852 // The purpose of the SetMaxThreads limit is to avoid accidental fork
853 // bomb from something like millions of goroutines blocking on system
854 // calls, causing the runtime to create millions of threads. By
855 // definition, this isn't a problem for threads created in C, so we
856 // exclude them from the limit. See https://go.dev/issue/60004.
857 count := mcount() - int32(extraMInUse.Load()) - int32(extraMLength.Load())
858 if count > sched.maxmcount {
859 print("runtime: program exceeds ", sched.maxmcount, "-thread limit\n")
860 throw("thread exhaustion")
864 // mReserveID returns the next ID to use for a new m. This new m is immediately
865 // considered 'running' by checkdead.
867 // sched.lock must be held.
868 func mReserveID() int64 {
869 assertLockHeld(&sched.lock)
871 if sched.mnext+1 < sched.mnext {
872 throw("runtime: thread ID overflow")
880 // Pre-allocated ID may be passed as 'id', or omitted by passing -1.
881 func mcommoninit(mp *m, id int64) {
884 // g0 stack won't make sense for user (and is not necessary unwindable).
886 callers(1, mp.createstack[:])
897 lo := uint32(int64Hash(uint64(mp.id), fastrandseed))
898 hi := uint32(int64Hash(uint64(cputicks()), ^fastrandseed))
902 // Same behavior as for 1.17.
903 // TODO: Simplify this.
904 if goarch.BigEndian {
905 mp.fastrand = uint64(lo)<<32 | uint64(hi)
907 mp.fastrand = uint64(hi)<<32 | uint64(lo)
911 if mp.gsignal != nil {
912 mp.gsignal.stackguard1 = mp.gsignal.stack.lo + stackGuard
915 // Add to allm so garbage collector doesn't free g->m
916 // when it is just in a register or thread-local storage.
919 // NumCgoCall() iterates over allm w/o schedlock,
920 // so we need to publish it safely.
921 atomicstorep(unsafe.Pointer(&allm), unsafe.Pointer(mp))
924 // Allocate memory to hold a cgo traceback if the cgo call crashes.
925 if iscgo || GOOS == "solaris" || GOOS == "illumos" || GOOS == "windows" {
926 mp.cgoCallers = new(cgoCallers)
930 func (mp *m) becomeSpinning() {
932 sched.nmspinning.Add(1)
933 sched.needspinning.Store(0)
936 func (mp *m) hasCgoOnStack() bool {
937 return mp.ncgo > 0 || mp.isextra
940 var fastrandseed uintptr
942 func fastrandinit() {
943 s := (*[unsafe.Sizeof(fastrandseed)]byte)(unsafe.Pointer(&fastrandseed))[:]
947 // Mark gp ready to run.
948 func ready(gp *g, traceskip int, next bool) {
950 traceGoUnpark(gp, traceskip)
953 status := readgstatus(gp)
956 mp := acquirem() // disable preemption because it can be holding p in a local var
957 if status&^_Gscan != _Gwaiting {
959 throw("bad g->status in ready")
962 // status is Gwaiting or Gscanwaiting, make Grunnable and put on runq
963 casgstatus(gp, _Gwaiting, _Grunnable)
964 runqput(mp.p.ptr(), gp, next)
969 // freezeStopWait is a large value that freezetheworld sets
970 // sched.stopwait to in order to request that all Gs permanently stop.
971 const freezeStopWait = 0x7fffffff
973 // freezing is set to non-zero if the runtime is trying to freeze the
975 var freezing atomic.Bool
977 // Similar to stopTheWorld but best-effort and can be called several times.
978 // There is no reverse operation, used during crashing.
979 // This function must not lock any mutexes.
980 func freezetheworld() {
982 if debug.dontfreezetheworld > 0 {
983 // Don't prempt Ps to stop goroutines. That will perturb
984 // scheduler state, making debugging more difficult. Instead,
985 // allow goroutines to continue execution.
987 // fatalpanic will tracebackothers to trace all goroutines. It
988 // is unsafe to trace a running goroutine, so tracebackothers
989 // will skip running goroutines. That is OK and expected, we
990 // expect users of dontfreezetheworld to use core files anyway.
992 // However, allowing the scheduler to continue running free
993 // introduces a race: a goroutine may be stopped when
994 // tracebackothers checks its status, and then start running
995 // later when we are in the middle of traceback, potentially
998 // To mitigate this, when an M naturally enters the scheduler,
999 // schedule checks if freezing is set and if so stops
1000 // execution. This guarantees that while Gs can transition from
1001 // running to stopped, they can never transition from stopped
1004 // The sleep here allows racing Ms that missed freezing and are
1005 // about to run a G to complete the transition to running
1006 // before we start traceback.
1011 // stopwait and preemption requests can be lost
1012 // due to races with concurrently executing threads,
1013 // so try several times
1014 for i := 0; i < 5; i++ {
1015 // this should tell the scheduler to not start any new goroutines
1016 sched.stopwait = freezeStopWait
1017 sched.gcwaiting.Store(true)
1018 // this should stop running goroutines
1020 break // no running goroutines
1030 // All reads and writes of g's status go through readgstatus, casgstatus
1031 // castogscanstatus, casfrom_Gscanstatus.
1034 func readgstatus(gp *g) uint32 {
1035 return gp.atomicstatus.Load()
1038 // The Gscanstatuses are acting like locks and this releases them.
1039 // If it proves to be a performance hit we should be able to make these
1040 // simple atomic stores but for now we are going to throw if
1041 // we see an inconsistent state.
1042 func casfrom_Gscanstatus(gp *g, oldval, newval uint32) {
1045 // Check that transition is valid.
1048 print("runtime: casfrom_Gscanstatus bad oldval gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
1050 throw("casfrom_Gscanstatus:top gp->status is not in scan state")
1051 case _Gscanrunnable,
1056 if newval == oldval&^_Gscan {
1057 success = gp.atomicstatus.CompareAndSwap(oldval, newval)
1061 print("runtime: casfrom_Gscanstatus failed gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
1063 throw("casfrom_Gscanstatus: gp->status is not in scan state")
1065 releaseLockRank(lockRankGscan)
1068 // This will return false if the gp is not in the expected status and the cas fails.
1069 // This acts like a lock acquire while the casfromgstatus acts like a lock release.
1070 func castogscanstatus(gp *g, oldval, newval uint32) bool {
1076 if newval == oldval|_Gscan {
1077 r := gp.atomicstatus.CompareAndSwap(oldval, newval)
1079 acquireLockRank(lockRankGscan)
1085 print("runtime: castogscanstatus oldval=", hex(oldval), " newval=", hex(newval), "\n")
1086 throw("castogscanstatus")
1087 panic("not reached")
1090 // casgstatusAlwaysTrack is a debug flag that causes casgstatus to always track
1091 // various latencies on every transition instead of sampling them.
1092 var casgstatusAlwaysTrack = false
1094 // If asked to move to or from a Gscanstatus this will throw. Use the castogscanstatus
1095 // and casfrom_Gscanstatus instead.
1096 // casgstatus will loop if the g->atomicstatus is in a Gscan status until the routine that
1097 // put it in the Gscan state is finished.
1100 func casgstatus(gp *g, oldval, newval uint32) {
1101 if (oldval&_Gscan != 0) || (newval&_Gscan != 0) || oldval == newval {
1102 systemstack(func() {
1103 print("runtime: casgstatus: oldval=", hex(oldval), " newval=", hex(newval), "\n")
1104 throw("casgstatus: bad incoming values")
1108 acquireLockRank(lockRankGscan)
1109 releaseLockRank(lockRankGscan)
1111 // See https://golang.org/cl/21503 for justification of the yield delay.
1112 const yieldDelay = 5 * 1000
1115 // loop if gp->atomicstatus is in a scan state giving
1116 // GC time to finish and change the state to oldval.
1117 for i := 0; !gp.atomicstatus.CompareAndSwap(oldval, newval); i++ {
1118 if oldval == _Gwaiting && gp.atomicstatus.Load() == _Grunnable {
1119 throw("casgstatus: waiting for Gwaiting but is Grunnable")
1122 nextYield = nanotime() + yieldDelay
1124 if nanotime() < nextYield {
1125 for x := 0; x < 10 && gp.atomicstatus.Load() != oldval; x++ {
1130 nextYield = nanotime() + yieldDelay/2
1134 if oldval == _Grunning {
1135 // Track every gTrackingPeriod time a goroutine transitions out of running.
1136 if casgstatusAlwaysTrack || gp.trackingSeq%gTrackingPeriod == 0 {
1145 // Handle various kinds of tracking.
1148 // - Time spent in runnable.
1149 // - Time spent blocked on a sync.Mutex or sync.RWMutex.
1152 // We transitioned out of runnable, so measure how much
1153 // time we spent in this state and add it to
1156 gp.runnableTime += now - gp.trackingStamp
1157 gp.trackingStamp = 0
1159 if !gp.waitreason.isMutexWait() {
1160 // Not blocking on a lock.
1163 // Blocking on a lock, measure it. Note that because we're
1164 // sampling, we have to multiply by our sampling period to get
1165 // a more representative estimate of the absolute value.
1166 // gTrackingPeriod also represents an accurate sampling period
1167 // because we can only enter this state from _Grunning.
1169 sched.totalMutexWaitTime.Add((now - gp.trackingStamp) * gTrackingPeriod)
1170 gp.trackingStamp = 0
1174 if !gp.waitreason.isMutexWait() {
1175 // Not blocking on a lock.
1178 // Blocking on a lock. Write down the timestamp.
1180 gp.trackingStamp = now
1182 // We just transitioned into runnable, so record what
1183 // time that happened.
1185 gp.trackingStamp = now
1187 // We're transitioning into running, so turn off
1188 // tracking and record how much time we spent in
1191 sched.timeToRun.record(gp.runnableTime)
1196 // casGToWaiting transitions gp from old to _Gwaiting, and sets the wait reason.
1198 // Use this over casgstatus when possible to ensure that a waitreason is set.
1199 func casGToWaiting(gp *g, old uint32, reason waitReason) {
1200 // Set the wait reason before calling casgstatus, because casgstatus will use it.
1201 gp.waitreason = reason
1202 casgstatus(gp, old, _Gwaiting)
1205 // casgstatus(gp, oldstatus, Gcopystack), assuming oldstatus is Gwaiting or Grunnable.
1206 // Returns old status. Cannot call casgstatus directly, because we are racing with an
1207 // async wakeup that might come in from netpoll. If we see Gwaiting from the readgstatus,
1208 // it might have become Grunnable by the time we get to the cas. If we called casgstatus,
1209 // it would loop waiting for the status to go back to Gwaiting, which it never will.
1212 func casgcopystack(gp *g) uint32 {
1214 oldstatus := readgstatus(gp) &^ _Gscan
1215 if oldstatus != _Gwaiting && oldstatus != _Grunnable {
1216 throw("copystack: bad status, not Gwaiting or Grunnable")
1218 if gp.atomicstatus.CompareAndSwap(oldstatus, _Gcopystack) {
1224 // casGToPreemptScan transitions gp from _Grunning to _Gscan|_Gpreempted.
1226 // TODO(austin): This is the only status operation that both changes
1227 // the status and locks the _Gscan bit. Rethink this.
1228 func casGToPreemptScan(gp *g, old, new uint32) {
1229 if old != _Grunning || new != _Gscan|_Gpreempted {
1230 throw("bad g transition")
1232 acquireLockRank(lockRankGscan)
1233 for !gp.atomicstatus.CompareAndSwap(_Grunning, _Gscan|_Gpreempted) {
1237 // casGFromPreempted attempts to transition gp from _Gpreempted to
1238 // _Gwaiting. If successful, the caller is responsible for
1239 // re-scheduling gp.
1240 func casGFromPreempted(gp *g, old, new uint32) bool {
1241 if old != _Gpreempted || new != _Gwaiting {
1242 throw("bad g transition")
1244 gp.waitreason = waitReasonPreempted
1245 return gp.atomicstatus.CompareAndSwap(_Gpreempted, _Gwaiting)
1248 // stwReason is an enumeration of reasons the world is stopping.
1249 type stwReason uint8
1251 // Reasons to stop-the-world.
1253 // Avoid reusing reasons and add new ones instead.
1255 stwUnknown stwReason = iota // "unknown"
1256 stwGCMarkTerm // "GC mark termination"
1257 stwGCSweepTerm // "GC sweep termination"
1258 stwWriteHeapDump // "write heap dump"
1259 stwGoroutineProfile // "goroutine profile"
1260 stwGoroutineProfileCleanup // "goroutine profile cleanup"
1261 stwAllGoroutinesStack // "all goroutines stack trace"
1262 stwReadMemStats // "read mem stats"
1263 stwAllThreadsSyscall // "AllThreadsSyscall"
1264 stwGOMAXPROCS // "GOMAXPROCS"
1265 stwStartTrace // "start trace"
1266 stwStopTrace // "stop trace"
1267 stwForTestCountPagesInUse // "CountPagesInUse (test)"
1268 stwForTestReadMetricsSlow // "ReadMetricsSlow (test)"
1269 stwForTestReadMemStatsSlow // "ReadMemStatsSlow (test)"
1270 stwForTestPageCachePagesLeaked // "PageCachePagesLeaked (test)"
1271 stwForTestResetDebugLog // "ResetDebugLog (test)"
1274 func (r stwReason) String() string {
1275 return stwReasonStrings[r]
1278 // If you add to this list, also add it to src/internal/trace/parser.go.
1279 // If you change the values of any of the stw* constants, bump the trace
1280 // version number and make a copy of this.
1281 var stwReasonStrings = [...]string{
1282 stwUnknown: "unknown",
1283 stwGCMarkTerm: "GC mark termination",
1284 stwGCSweepTerm: "GC sweep termination",
1285 stwWriteHeapDump: "write heap dump",
1286 stwGoroutineProfile: "goroutine profile",
1287 stwGoroutineProfileCleanup: "goroutine profile cleanup",
1288 stwAllGoroutinesStack: "all goroutines stack trace",
1289 stwReadMemStats: "read mem stats",
1290 stwAllThreadsSyscall: "AllThreadsSyscall",
1291 stwGOMAXPROCS: "GOMAXPROCS",
1292 stwStartTrace: "start trace",
1293 stwStopTrace: "stop trace",
1294 stwForTestCountPagesInUse: "CountPagesInUse (test)",
1295 stwForTestReadMetricsSlow: "ReadMetricsSlow (test)",
1296 stwForTestReadMemStatsSlow: "ReadMemStatsSlow (test)",
1297 stwForTestPageCachePagesLeaked: "PageCachePagesLeaked (test)",
1298 stwForTestResetDebugLog: "ResetDebugLog (test)",
1301 // stopTheWorld stops all P's from executing goroutines, interrupting
1302 // all goroutines at GC safe points and records reason as the reason
1303 // for the stop. On return, only the current goroutine's P is running.
1304 // stopTheWorld must not be called from a system stack and the caller
1305 // must not hold worldsema. The caller must call startTheWorld when
1306 // other P's should resume execution.
1308 // stopTheWorld is safe for multiple goroutines to call at the
1309 // same time. Each will execute its own stop, and the stops will
1312 // This is also used by routines that do stack dumps. If the system is
1313 // in panic or being exited, this may not reliably stop all
1315 func stopTheWorld(reason stwReason) {
1316 semacquire(&worldsema)
1318 gp.m.preemptoff = reason.String()
1319 systemstack(func() {
1320 // Mark the goroutine which called stopTheWorld preemptible so its
1321 // stack may be scanned.
1322 // This lets a mark worker scan us while we try to stop the world
1323 // since otherwise we could get in a mutual preemption deadlock.
1324 // We must not modify anything on the G stack because a stack shrink
1325 // may occur. A stack shrink is otherwise OK though because in order
1326 // to return from this function (and to leave the system stack) we
1327 // must have preempted all goroutines, including any attempting
1328 // to scan our stack, in which case, any stack shrinking will
1329 // have already completed by the time we exit.
1330 // Don't provide a wait reason because we're still executing.
1331 casGToWaiting(gp, _Grunning, waitReasonStoppingTheWorld)
1332 stopTheWorldWithSema(reason)
1333 casgstatus(gp, _Gwaiting, _Grunning)
1337 // startTheWorld undoes the effects of stopTheWorld.
1338 func startTheWorld() {
1339 systemstack(func() { startTheWorldWithSema() })
1341 // worldsema must be held over startTheWorldWithSema to ensure
1342 // gomaxprocs cannot change while worldsema is held.
1344 // Release worldsema with direct handoff to the next waiter, but
1345 // acquirem so that semrelease1 doesn't try to yield our time.
1347 // Otherwise if e.g. ReadMemStats is being called in a loop,
1348 // it might stomp on other attempts to stop the world, such as
1349 // for starting or ending GC. The operation this blocks is
1350 // so heavy-weight that we should just try to be as fair as
1353 // We don't want to just allow us to get preempted between now
1354 // and releasing the semaphore because then we keep everyone
1355 // (including, for example, GCs) waiting longer.
1358 semrelease1(&worldsema, true, 0)
1362 // stopTheWorldGC has the same effect as stopTheWorld, but blocks
1363 // until the GC is not running. It also blocks a GC from starting
1364 // until startTheWorldGC is called.
1365 func stopTheWorldGC(reason stwReason) {
1367 stopTheWorld(reason)
1370 // startTheWorldGC undoes the effects of stopTheWorldGC.
1371 func startTheWorldGC() {
1376 // Holding worldsema grants an M the right to try to stop the world.
1377 var worldsema uint32 = 1
1379 // Holding gcsema grants the M the right to block a GC, and blocks
1380 // until the current GC is done. In particular, it prevents gomaxprocs
1381 // from changing concurrently.
1383 // TODO(mknyszek): Once gomaxprocs and the execution tracer can handle
1384 // being changed/enabled during a GC, remove this.
1385 var gcsema uint32 = 1
1387 // stopTheWorldWithSema is the core implementation of stopTheWorld.
1388 // The caller is responsible for acquiring worldsema and disabling
1389 // preemption first and then should stopTheWorldWithSema on the system
1392 // semacquire(&worldsema, 0)
1393 // m.preemptoff = "reason"
1394 // systemstack(stopTheWorldWithSema)
1396 // When finished, the caller must either call startTheWorld or undo
1397 // these three operations separately:
1399 // m.preemptoff = ""
1400 // systemstack(startTheWorldWithSema)
1401 // semrelease(&worldsema)
1403 // It is allowed to acquire worldsema once and then execute multiple
1404 // startTheWorldWithSema/stopTheWorldWithSema pairs.
1405 // Other P's are able to execute between successive calls to
1406 // startTheWorldWithSema and stopTheWorldWithSema.
1407 // Holding worldsema causes any other goroutines invoking
1408 // stopTheWorld to block.
1409 func stopTheWorldWithSema(reason stwReason) {
1411 traceSTWStart(reason)
1415 // If we hold a lock, then we won't be able to stop another M
1416 // that is blocked trying to acquire the lock.
1418 throw("stopTheWorld: holding locks")
1422 sched.stopwait = gomaxprocs
1423 sched.gcwaiting.Store(true)
1426 gp.m.p.ptr().status = _Pgcstop // Pgcstop is only diagnostic.
1428 // try to retake all P's in Psyscall status
1429 for _, pp := range allp {
1431 if s == _Psyscall && atomic.Cas(&pp.status, s, _Pgcstop) {
1443 pp, _ := pidleget(now)
1447 pp.status = _Pgcstop
1450 wait := sched.stopwait > 0
1453 // wait for remaining P's to stop voluntarily
1456 // wait for 100us, then try to re-preempt in case of any races
1457 if notetsleep(&sched.stopnote, 100*1000) {
1458 noteclear(&sched.stopnote)
1467 if sched.stopwait != 0 {
1468 bad = "stopTheWorld: not stopped (stopwait != 0)"
1470 for _, pp := range allp {
1471 if pp.status != _Pgcstop {
1472 bad = "stopTheWorld: not stopped (status != _Pgcstop)"
1476 if freezing.Load() {
1477 // Some other thread is panicking. This can cause the
1478 // sanity checks above to fail if the panic happens in
1479 // the signal handler on a stopped thread. Either way,
1480 // we should halt this thread.
1491 func startTheWorldWithSema() int64 {
1492 assertWorldStopped()
1494 mp := acquirem() // disable preemption because it can be holding p in a local var
1495 if netpollinited() {
1496 list, delta := netpoll(0) // non-blocking
1498 netpollAdjustWaiters(delta)
1507 p1 := procresize(procs)
1508 sched.gcwaiting.Store(false)
1509 if sched.sysmonwait.Load() {
1510 sched.sysmonwait.Store(false)
1511 notewakeup(&sched.sysmonnote)
1524 throw("startTheWorld: inconsistent mp->nextp")
1527 notewakeup(&mp.park)
1529 // Start M to run P. Do not start another M below.
1534 // Capture start-the-world time before doing clean-up tasks.
1535 startTime := nanotime()
1540 // Wakeup an additional proc in case we have excessive runnable goroutines
1541 // in local queues or in the global queue. If we don't, the proc will park itself.
1542 // If we have lots of excessive work, resetspinning will unpark additional procs as necessary.
1550 // usesLibcall indicates whether this runtime performs system calls
1552 func usesLibcall() bool {
1554 case "aix", "darwin", "illumos", "ios", "solaris", "windows":
1557 return GOARCH != "mips64"
1562 // mStackIsSystemAllocated indicates whether this runtime starts on a
1563 // system-allocated stack.
1564 func mStackIsSystemAllocated() bool {
1566 case "aix", "darwin", "plan9", "illumos", "ios", "solaris", "windows":
1569 return GOARCH != "mips64"
1574 // mstart is the entry-point for new Ms.
1575 // It is written in assembly, uses ABI0, is marked TOPFRAME, and calls mstart0.
1578 // mstart0 is the Go entry-point for new Ms.
1579 // This must not split the stack because we may not even have stack
1580 // bounds set up yet.
1582 // May run during STW (because it doesn't have a P yet), so write
1583 // barriers are not allowed.
1586 //go:nowritebarrierrec
1590 osStack := gp.stack.lo == 0
1592 // Initialize stack bounds from system stack.
1593 // Cgo may have left stack size in stack.hi.
1594 // minit may update the stack bounds.
1596 // Note: these bounds may not be very accurate.
1597 // We set hi to &size, but there are things above
1598 // it. The 1024 is supposed to compensate this,
1599 // but is somewhat arbitrary.
1602 size = 16384 * sys.StackGuardMultiplier
1604 gp.stack.hi = uintptr(noescape(unsafe.Pointer(&size)))
1605 gp.stack.lo = gp.stack.hi - size + 1024
1607 // Initialize stack guard so that we can start calling regular
1609 gp.stackguard0 = gp.stack.lo + stackGuard
1610 // This is the g0, so we can also call go:systemstack
1611 // functions, which check stackguard1.
1612 gp.stackguard1 = gp.stackguard0
1615 // Exit this thread.
1616 if mStackIsSystemAllocated() {
1617 // Windows, Solaris, illumos, Darwin, AIX and Plan 9 always system-allocate
1618 // the stack, but put it in gp.stack before mstart,
1619 // so the logic above hasn't set osStack yet.
1625 // The go:noinline is to guarantee the getcallerpc/getcallersp below are safe,
1626 // so that we can set up g0.sched to return to the call of mstart1 above.
1633 throw("bad runtime·mstart")
1636 // Set up m.g0.sched as a label returning to just
1637 // after the mstart1 call in mstart0 above, for use by goexit0 and mcall.
1638 // We're never coming back to mstart1 after we call schedule,
1639 // so other calls can reuse the current frame.
1640 // And goexit0 does a gogo that needs to return from mstart1
1641 // and let mstart0 exit the thread.
1642 gp.sched.g = guintptr(unsafe.Pointer(gp))
1643 gp.sched.pc = getcallerpc()
1644 gp.sched.sp = getcallersp()
1649 // Install signal handlers; after minit so that minit can
1650 // prepare the thread to be able to handle the signals.
1655 if fn := gp.m.mstartfn; fn != nil {
1660 acquirep(gp.m.nextp.ptr())
1666 // mstartm0 implements part of mstart1 that only runs on the m0.
1668 // Write barriers are allowed here because we know the GC can't be
1669 // running yet, so they'll be no-ops.
1671 //go:yeswritebarrierrec
1673 // Create an extra M for callbacks on threads not created by Go.
1674 // An extra M is also needed on Windows for callbacks created by
1675 // syscall.NewCallback. See issue #6751 for details.
1676 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
1683 // mPark causes a thread to park itself, returning once woken.
1688 notesleep(&gp.m.park)
1689 noteclear(&gp.m.park)
1692 // mexit tears down and exits the current thread.
1694 // Don't call this directly to exit the thread, since it must run at
1695 // the top of the thread stack. Instead, use gogo(&gp.m.g0.sched) to
1696 // unwind the stack to the point that exits the thread.
1698 // It is entered with m.p != nil, so write barriers are allowed. It
1699 // will release the P before exiting.
1701 //go:yeswritebarrierrec
1702 func mexit(osStack bool) {
1706 // This is the main thread. Just wedge it.
1708 // On Linux, exiting the main thread puts the process
1709 // into a non-waitable zombie state. On Plan 9,
1710 // exiting the main thread unblocks wait even though
1711 // other threads are still running. On Solaris we can
1712 // neither exitThread nor return from mstart. Other
1713 // bad things probably happen on other platforms.
1715 // We could try to clean up this M more before wedging
1716 // it, but that complicates signal handling.
1717 handoffp(releasep())
1723 throw("locked m0 woke up")
1729 // Free the gsignal stack.
1730 if mp.gsignal != nil {
1731 stackfree(mp.gsignal.stack)
1732 // On some platforms, when calling into VDSO (e.g. nanotime)
1733 // we store our g on the gsignal stack, if there is one.
1734 // Now the stack is freed, unlink it from the m, so we
1735 // won't write to it when calling VDSO code.
1739 // Remove m from allm.
1741 for pprev := &allm; *pprev != nil; pprev = &(*pprev).alllink {
1747 throw("m not found in allm")
1749 // Delay reaping m until it's done with the stack.
1751 // Put mp on the free list, though it will not be reaped while freeWait
1752 // is freeMWait. mp is no longer reachable via allm, so even if it is
1753 // on an OS stack, we must keep a reference to mp alive so that the GC
1754 // doesn't free mp while we are still using it.
1756 // Note that the free list must not be linked through alllink because
1757 // some functions walk allm without locking, so may be using alllink.
1758 mp.freeWait.Store(freeMWait)
1759 mp.freelink = sched.freem
1763 atomic.Xadd64(&ncgocall, int64(mp.ncgocall))
1766 handoffp(releasep())
1767 // After this point we must not have write barriers.
1769 // Invoke the deadlock detector. This must happen after
1770 // handoffp because it may have started a new M to take our
1777 if GOOS == "darwin" || GOOS == "ios" {
1778 // Make sure pendingPreemptSignals is correct when an M exits.
1780 if mp.signalPending.Load() != 0 {
1781 pendingPreemptSignals.Add(-1)
1785 // Destroy all allocated resources. After this is called, we may no
1786 // longer take any locks.
1790 // No more uses of mp, so it is safe to drop the reference.
1791 mp.freeWait.Store(freeMRef)
1793 // Return from mstart and let the system thread
1794 // library free the g0 stack and terminate the thread.
1798 // mstart is the thread's entry point, so there's nothing to
1799 // return to. Exit the thread directly. exitThread will clear
1800 // m.freeWait when it's done with the stack and the m can be
1802 exitThread(&mp.freeWait)
1805 // forEachP calls fn(p) for every P p when p reaches a GC safe point.
1806 // If a P is currently executing code, this will bring the P to a GC
1807 // safe point and execute fn on that P. If the P is not executing code
1808 // (it is idle or in a syscall), this will call fn(p) directly while
1809 // preventing the P from exiting its state. This does not ensure that
1810 // fn will run on every CPU executing Go code, but it acts as a global
1811 // memory barrier. GC uses this as a "ragged barrier."
1813 // The caller must hold worldsema.
1816 func forEachP(fn func(*p)) {
1818 pp := getg().m.p.ptr()
1821 if sched.safePointWait != 0 {
1822 throw("forEachP: sched.safePointWait != 0")
1824 sched.safePointWait = gomaxprocs - 1
1825 sched.safePointFn = fn
1827 // Ask all Ps to run the safe point function.
1828 for _, p2 := range allp {
1830 atomic.Store(&p2.runSafePointFn, 1)
1835 // Any P entering _Pidle or _Psyscall from now on will observe
1836 // p.runSafePointFn == 1 and will call runSafePointFn when
1837 // changing its status to _Pidle/_Psyscall.
1839 // Run safe point function for all idle Ps. sched.pidle will
1840 // not change because we hold sched.lock.
1841 for p := sched.pidle.ptr(); p != nil; p = p.link.ptr() {
1842 if atomic.Cas(&p.runSafePointFn, 1, 0) {
1844 sched.safePointWait--
1848 wait := sched.safePointWait > 0
1851 // Run fn for the current P.
1854 // Force Ps currently in _Psyscall into _Pidle and hand them
1855 // off to induce safe point function execution.
1856 for _, p2 := range allp {
1858 if s == _Psyscall && p2.runSafePointFn == 1 && atomic.Cas(&p2.status, s, _Pidle) {
1868 // Wait for remaining Ps to run fn.
1871 // Wait for 100us, then try to re-preempt in
1872 // case of any races.
1874 // Requires system stack.
1875 if notetsleep(&sched.safePointNote, 100*1000) {
1876 noteclear(&sched.safePointNote)
1882 if sched.safePointWait != 0 {
1883 throw("forEachP: not done")
1885 for _, p2 := range allp {
1886 if p2.runSafePointFn != 0 {
1887 throw("forEachP: P did not run fn")
1892 sched.safePointFn = nil
1897 // runSafePointFn runs the safe point function, if any, for this P.
1898 // This should be called like
1900 // if getg().m.p.runSafePointFn != 0 {
1904 // runSafePointFn must be checked on any transition in to _Pidle or
1905 // _Psyscall to avoid a race where forEachP sees that the P is running
1906 // just before the P goes into _Pidle/_Psyscall and neither forEachP
1907 // nor the P run the safe-point function.
1908 func runSafePointFn() {
1909 p := getg().m.p.ptr()
1910 // Resolve the race between forEachP running the safe-point
1911 // function on this P's behalf and this P running the
1912 // safe-point function directly.
1913 if !atomic.Cas(&p.runSafePointFn, 1, 0) {
1916 sched.safePointFn(p)
1918 sched.safePointWait--
1919 if sched.safePointWait == 0 {
1920 notewakeup(&sched.safePointNote)
1925 // When running with cgo, we call _cgo_thread_start
1926 // to start threads for us so that we can play nicely with
1928 var cgoThreadStart unsafe.Pointer
1930 type cgothreadstart struct {
1936 // Allocate a new m unassociated with any thread.
1937 // Can use p for allocation context if needed.
1938 // fn is recorded as the new m's m.mstartfn.
1939 // id is optional pre-allocated m ID. Omit by passing -1.
1941 // This function is allowed to have write barriers even if the caller
1942 // isn't because it borrows pp.
1944 //go:yeswritebarrierrec
1945 func allocm(pp *p, fn func(), id int64) *m {
1948 // The caller owns pp, but we may borrow (i.e., acquirep) it. We must
1949 // disable preemption to ensure it is not stolen, which would make the
1950 // caller lose ownership.
1955 acquirep(pp) // temporarily borrow p for mallocs in this function
1958 // Release the free M list. We need to do this somewhere and
1959 // this may free up a stack we can use.
1960 if sched.freem != nil {
1963 for freem := sched.freem; freem != nil; {
1964 wait := freem.freeWait.Load()
1965 if wait == freeMWait {
1966 next := freem.freelink
1967 freem.freelink = newList
1972 // Free the stack if needed. For freeMRef, there is
1973 // nothing to do except drop freem from the sched.freem
1975 if wait == freeMStack {
1976 // stackfree must be on the system stack, but allocm is
1977 // reachable off the system stack transitively from
1979 systemstack(func() {
1980 stackfree(freem.g0.stack)
1983 freem = freem.freelink
1985 sched.freem = newList
1993 // In case of cgo or Solaris or illumos or Darwin, pthread_create will make us a stack.
1994 // Windows and Plan 9 will layout sched stack on OS stack.
1995 if iscgo || mStackIsSystemAllocated() {
1998 mp.g0 = malg(16384 * sys.StackGuardMultiplier)
2002 if pp == gp.m.p.ptr() {
2007 allocmLock.runlock()
2011 // needm is called when a cgo callback happens on a
2012 // thread without an m (a thread not created by Go).
2013 // In this case, needm is expected to find an m to use
2014 // and return with m, g initialized correctly.
2015 // Since m and g are not set now (likely nil, but see below)
2016 // needm is limited in what routines it can call. In particular
2017 // it can only call nosplit functions (textflag 7) and cannot
2018 // do any scheduling that requires an m.
2020 // In order to avoid needing heavy lifting here, we adopt
2021 // the following strategy: there is a stack of available m's
2022 // that can be stolen. Using compare-and-swap
2023 // to pop from the stack has ABA races, so we simulate
2024 // a lock by doing an exchange (via Casuintptr) to steal the stack
2025 // head and replace the top pointer with MLOCKED (1).
2026 // This serves as a simple spin lock that we can use even
2027 // without an m. The thread that locks the stack in this way
2028 // unlocks the stack by storing a valid stack head pointer.
2030 // In order to make sure that there is always an m structure
2031 // available to be stolen, we maintain the invariant that there
2032 // is always one more than needed. At the beginning of the
2033 // program (if cgo is in use) the list is seeded with a single m.
2034 // If needm finds that it has taken the last m off the list, its job
2035 // is - once it has installed its own m so that it can do things like
2036 // allocate memory - to create a spare m and put it on the list.
2038 // Each of these extra m's also has a g0 and a curg that are
2039 // pressed into service as the scheduling stack and current
2040 // goroutine for the duration of the cgo callback.
2042 // It calls dropm to put the m back on the list,
2043 // 1. when the callback is done with the m in non-pthread platforms,
2044 // 2. or when the C thread exiting on pthread platforms.
2046 // The signal argument indicates whether we're called from a signal
2050 func needm(signal bool) {
2051 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
2052 // Can happen if C/C++ code calls Go from a global ctor.
2053 // Can also happen on Windows if a global ctor uses a
2054 // callback created by syscall.NewCallback. See issue #6751
2057 // Can not throw, because scheduler is not initialized yet.
2058 writeErrStr("fatal error: cgo callback before cgo call\n")
2062 // Save and block signals before getting an M.
2063 // The signal handler may call needm itself,
2064 // and we must avoid a deadlock. Also, once g is installed,
2065 // any incoming signals will try to execute,
2066 // but we won't have the sigaltstack settings and other data
2067 // set up appropriately until the end of minit, which will
2068 // unblock the signals. This is the same dance as when
2069 // starting a new m to run Go code via newosproc.
2074 // getExtraM is safe here because of the invariant above,
2075 // that the extra list always contains or will soon contain
2077 mp, last := getExtraM()
2079 // Set needextram when we've just emptied the list,
2080 // so that the eventual call into cgocallbackg will
2081 // allocate a new m for the extra list. We delay the
2082 // allocation until then so that it can be done
2083 // after exitsyscall makes sure it is okay to be
2084 // running at all (that is, there's no garbage collection
2085 // running right now).
2086 mp.needextram = last
2088 // Store the original signal mask for use by minit.
2089 mp.sigmask = sigmask
2091 // Install TLS on some platforms (previously setg
2092 // would do this if necessary).
2095 // Install g (= m->g0) and set the stack bounds
2096 // to match the current stack.
2099 callbackUpdateSystemStack(mp, sp, signal)
2101 // Should mark we are already in Go now.
2102 // Otherwise, we may call needm again when we get a signal, before cgocallbackg1,
2103 // which means the extram list may be empty, that will cause a deadlock.
2104 mp.isExtraInC = false
2106 // Initialize this thread to use the m.
2110 // mp.curg is now a real goroutine.
2111 casgstatus(mp.curg, _Gdead, _Gsyscall)
2115 // Acquire an extra m and bind it to the C thread when a pthread key has been created.
2118 func needAndBindM() {
2121 if _cgo_pthread_key_created != nil && *(*uintptr)(_cgo_pthread_key_created) != 0 {
2126 // newextram allocates m's and puts them on the extra list.
2127 // It is called with a working local m, so that it can do things
2128 // like call schedlock and allocate.
2130 c := extraMWaiters.Swap(0)
2132 for i := uint32(0); i < c; i++ {
2135 } else if extraMLength.Load() == 0 {
2136 // Make sure there is at least one extra M.
2141 // oneNewExtraM allocates an m and puts it on the extra list.
2142 func oneNewExtraM() {
2143 // Create extra goroutine locked to extra m.
2144 // The goroutine is the context in which the cgo callback will run.
2145 // The sched.pc will never be returned to, but setting it to
2146 // goexit makes clear to the traceback routines where
2147 // the goroutine stack ends.
2148 mp := allocm(nil, nil, -1)
2150 gp.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum
2151 gp.sched.sp = gp.stack.hi
2152 gp.sched.sp -= 4 * goarch.PtrSize // extra space in case of reads slightly beyond frame
2154 gp.sched.g = guintptr(unsafe.Pointer(gp))
2155 gp.syscallpc = gp.sched.pc
2156 gp.syscallsp = gp.sched.sp
2157 gp.stktopsp = gp.sched.sp
2158 // malg returns status as _Gidle. Change to _Gdead before
2159 // adding to allg where GC can see it. We use _Gdead to hide
2160 // this from tracebacks and stack scans since it isn't a
2161 // "real" goroutine until needm grabs it.
2162 casgstatus(gp, _Gidle, _Gdead)
2166 // mark we are in C by default.
2167 mp.isExtraInC = true
2171 gp.goid = sched.goidgen.Add(1)
2173 gp.racectx = racegostart(abi.FuncPCABIInternal(newextram) + sys.PCQuantum)
2176 traceOneNewExtraM(gp)
2178 // put on allg for garbage collector
2181 // gp is now on the allg list, but we don't want it to be
2182 // counted by gcount. It would be more "proper" to increment
2183 // sched.ngfree, but that requires locking. Incrementing ngsys
2184 // has the same effect.
2187 // Add m to the extra list.
2191 // dropm puts the current m back onto the extra list.
2193 // 1. On systems without pthreads, like Windows
2194 // dropm is called when a cgo callback has called needm but is now
2195 // done with the callback and returning back into the non-Go thread.
2197 // The main expense here is the call to signalstack to release the
2198 // m's signal stack, and then the call to needm on the next callback
2199 // from this thread. It is tempting to try to save the m for next time,
2200 // which would eliminate both these costs, but there might not be
2201 // a next time: the current thread (which Go does not control) might exit.
2202 // If we saved the m for that thread, there would be an m leak each time
2203 // such a thread exited. Instead, we acquire and release an m on each
2204 // call. These should typically not be scheduling operations, just a few
2205 // atomics, so the cost should be small.
2207 // 2. On systems with pthreads
2208 // dropm is called while a non-Go thread is exiting.
2209 // We allocate a pthread per-thread variable using pthread_key_create,
2210 // to register a thread-exit-time destructor.
2211 // And store the g into a thread-specific value associated with the pthread key,
2212 // when first return back to C.
2213 // So that the destructor would invoke dropm while the non-Go thread is exiting.
2214 // This is much faster since it avoids expensive signal-related syscalls.
2216 // This always runs without a P, so //go:nowritebarrierrec is required.
2218 // This may run with a different stack than was recorded in g0 (there is no
2219 // call to callbackUpdateSystemStack prior to dropm), so this must be
2220 // //go:nosplit to avoid the stack bounds check.
2222 //go:nowritebarrierrec
2225 // Clear m and g, and return m to the extra list.
2226 // After the call to setg we can only call nosplit functions
2227 // with no pointer manipulation.
2230 // Return mp.curg to dead state.
2231 casgstatus(mp.curg, _Gsyscall, _Gdead)
2232 mp.curg.preemptStop = false
2235 // Block signals before unminit.
2236 // Unminit unregisters the signal handling stack (but needs g on some systems).
2237 // Setg(nil) clears g, which is the signal handler's cue not to run Go handlers.
2238 // It's important not to try to handle a signal between those two steps.
2239 sigmask := mp.sigmask
2245 // Clear g0 stack bounds to ensure that needm always refreshes the
2246 // bounds when reusing this M.
2255 msigrestore(sigmask)
2258 // bindm store the g0 of the current m into a thread-specific value.
2260 // We allocate a pthread per-thread variable using pthread_key_create,
2261 // to register a thread-exit-time destructor.
2262 // We are here setting the thread-specific value of the pthread key, to enable the destructor.
2263 // So that the pthread_key_destructor would dropm while the C thread is exiting.
2265 // And the saved g will be used in pthread_key_destructor,
2266 // since the g stored in the TLS by Go might be cleared in some platforms,
2267 // before the destructor invoked, so, we restore g by the stored g, before dropm.
2269 // We store g0 instead of m, to make the assembly code simpler,
2270 // since we need to restore g0 in runtime.cgocallback.
2272 // On systems without pthreads, like Windows, bindm shouldn't be used.
2274 // NOTE: this always runs without a P, so, nowritebarrierrec required.
2277 //go:nowritebarrierrec
2279 if GOOS == "windows" || GOOS == "plan9" {
2280 fatal("bindm in unexpected GOOS")
2284 fatal("the current g is not g0")
2286 if _cgo_bindm != nil {
2287 asmcgocall(_cgo_bindm, unsafe.Pointer(g))
2291 // A helper function for EnsureDropM.
2292 func getm() uintptr {
2293 return uintptr(unsafe.Pointer(getg().m))
2297 // Locking linked list of extra M's, via mp.schedlink. Must be accessed
2298 // only via lockextra/unlockextra.
2300 // Can't be atomic.Pointer[m] because we use an invalid pointer as a
2301 // "locked" sentinel value. M's on this list remain visible to the GC
2302 // because their mp.curg is on allgs.
2303 extraM atomic.Uintptr
2304 // Number of M's in the extraM list.
2305 extraMLength atomic.Uint32
2306 // Number of waiters in lockextra.
2307 extraMWaiters atomic.Uint32
2309 // Number of extra M's in use by threads.
2310 extraMInUse atomic.Uint32
2313 // lockextra locks the extra list and returns the list head.
2314 // The caller must unlock the list by storing a new list head
2315 // to extram. If nilokay is true, then lockextra will
2316 // return a nil list head if that's what it finds. If nilokay is false,
2317 // lockextra will keep waiting until the list head is no longer nil.
2320 func lockextra(nilokay bool) *m {
2325 old := extraM.Load()
2330 if old == 0 && !nilokay {
2332 // Add 1 to the number of threads
2333 // waiting for an M.
2334 // This is cleared by newextram.
2335 extraMWaiters.Add(1)
2341 if extraM.CompareAndSwap(old, locked) {
2342 return (*m)(unsafe.Pointer(old))
2350 func unlockextra(mp *m, delta int32) {
2351 extraMLength.Add(delta)
2352 extraM.Store(uintptr(unsafe.Pointer(mp)))
2355 // Return an M from the extra M list. Returns last == true if the list becomes
2356 // empty because of this call.
2358 // Spins waiting for an extra M, so caller must ensure that the list always
2359 // contains or will soon contain at least one M.
2362 func getExtraM() (mp *m, last bool) {
2363 mp = lockextra(false)
2365 unlockextra(mp.schedlink.ptr(), -1)
2366 return mp, mp.schedlink.ptr() == nil
2369 // Returns an extra M back to the list. mp must be from getExtraM. Newly
2370 // allocated M's should use addExtraM.
2373 func putExtraM(mp *m) {
2378 // Adds a newly allocated M to the extra M list.
2381 func addExtraM(mp *m) {
2382 mnext := lockextra(true)
2383 mp.schedlink.set(mnext)
2388 // allocmLock is locked for read when creating new Ms in allocm and their
2389 // addition to allm. Thus acquiring this lock for write blocks the
2390 // creation of new Ms.
2393 // execLock serializes exec and clone to avoid bugs or unspecified
2394 // behaviour around exec'ing while creating/destroying threads. See
2399 // These errors are reported (via writeErrStr) by some OS-specific
2400 // versions of newosproc and newosproc0.
2402 failthreadcreate = "runtime: failed to create new OS thread\n"
2403 failallocatestack = "runtime: failed to allocate stack for the new OS thread\n"
2406 // newmHandoff contains a list of m structures that need new OS threads.
2407 // This is used by newm in situations where newm itself can't safely
2408 // start an OS thread.
2409 var newmHandoff struct {
2412 // newm points to a list of M structures that need new OS
2413 // threads. The list is linked through m.schedlink.
2416 // waiting indicates that wake needs to be notified when an m
2417 // is put on the list.
2421 // haveTemplateThread indicates that the templateThread has
2422 // been started. This is not protected by lock. Use cas to set
2424 haveTemplateThread uint32
2427 // Create a new m. It will start off with a call to fn, or else the scheduler.
2428 // fn needs to be static and not a heap allocated closure.
2429 // May run with m.p==nil, so write barriers are not allowed.
2431 // id is optional pre-allocated m ID. Omit by passing -1.
2433 //go:nowritebarrierrec
2434 func newm(fn func(), pp *p, id int64) {
2435 // allocm adds a new M to allm, but they do not start until created by
2436 // the OS in newm1 or the template thread.
2438 // doAllThreadsSyscall requires that every M in allm will eventually
2439 // start and be signal-able, even with a STW.
2441 // Disable preemption here until we start the thread to ensure that
2442 // newm is not preempted between allocm and starting the new thread,
2443 // ensuring that anything added to allm is guaranteed to eventually
2447 mp := allocm(pp, fn, id)
2449 mp.sigmask = initSigmask
2450 if gp := getg(); gp != nil && gp.m != nil && (gp.m.lockedExt != 0 || gp.m.incgo) && GOOS != "plan9" {
2451 // We're on a locked M or a thread that may have been
2452 // started by C. The kernel state of this thread may
2453 // be strange (the user may have locked it for that
2454 // purpose). We don't want to clone that into another
2455 // thread. Instead, ask a known-good thread to create
2456 // the thread for us.
2458 // This is disabled on Plan 9. See golang.org/issue/22227.
2460 // TODO: This may be unnecessary on Windows, which
2461 // doesn't model thread creation off fork.
2462 lock(&newmHandoff.lock)
2463 if newmHandoff.haveTemplateThread == 0 {
2464 throw("on a locked thread with no template thread")
2466 mp.schedlink = newmHandoff.newm
2467 newmHandoff.newm.set(mp)
2468 if newmHandoff.waiting {
2469 newmHandoff.waiting = false
2470 notewakeup(&newmHandoff.wake)
2472 unlock(&newmHandoff.lock)
2473 // The M has not started yet, but the template thread does not
2474 // participate in STW, so it will always process queued Ms and
2475 // it is safe to releasem.
2485 var ts cgothreadstart
2486 if _cgo_thread_start == nil {
2487 throw("_cgo_thread_start missing")
2490 ts.tls = (*uint64)(unsafe.Pointer(&mp.tls[0]))
2491 ts.fn = unsafe.Pointer(abi.FuncPCABI0(mstart))
2493 msanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2496 asanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2498 execLock.rlock() // Prevent process clone.
2499 asmcgocall(_cgo_thread_start, unsafe.Pointer(&ts))
2503 execLock.rlock() // Prevent process clone.
2508 // startTemplateThread starts the template thread if it is not already
2511 // The calling thread must itself be in a known-good state.
2512 func startTemplateThread() {
2513 if GOARCH == "wasm" { // no threads on wasm yet
2517 // Disable preemption to guarantee that the template thread will be
2518 // created before a park once haveTemplateThread is set.
2520 if !atomic.Cas(&newmHandoff.haveTemplateThread, 0, 1) {
2524 newm(templateThread, nil, -1)
2528 // templateThread is a thread in a known-good state that exists solely
2529 // to start new threads in known-good states when the calling thread
2530 // may not be in a good state.
2532 // Many programs never need this, so templateThread is started lazily
2533 // when we first enter a state that might lead to running on a thread
2534 // in an unknown state.
2536 // templateThread runs on an M without a P, so it must not have write
2539 //go:nowritebarrierrec
2540 func templateThread() {
2547 lock(&newmHandoff.lock)
2548 for newmHandoff.newm != 0 {
2549 newm := newmHandoff.newm.ptr()
2550 newmHandoff.newm = 0
2551 unlock(&newmHandoff.lock)
2553 next := newm.schedlink.ptr()
2558 lock(&newmHandoff.lock)
2560 newmHandoff.waiting = true
2561 noteclear(&newmHandoff.wake)
2562 unlock(&newmHandoff.lock)
2563 notesleep(&newmHandoff.wake)
2567 // Stops execution of the current m until new work is available.
2568 // Returns with acquired P.
2572 if gp.m.locks != 0 {
2573 throw("stopm holding locks")
2576 throw("stopm holding p")
2579 throw("stopm spinning")
2586 acquirep(gp.m.nextp.ptr())
2591 // startm's caller incremented nmspinning. Set the new M's spinning.
2592 getg().m.spinning = true
2595 // Schedules some M to run the p (creates an M if necessary).
2596 // If p==nil, tries to get an idle P, if no idle P's does nothing.
2597 // May run with m.p==nil, so write barriers are not allowed.
2598 // If spinning is set, the caller has incremented nmspinning and must provide a
2599 // P. startm will set m.spinning in the newly started M.
2601 // Callers passing a non-nil P must call from a non-preemptible context. See
2602 // comment on acquirem below.
2604 // Argument lockheld indicates whether the caller already acquired the
2605 // scheduler lock. Callers holding the lock when making the call must pass
2606 // true. The lock might be temporarily dropped, but will be reacquired before
2609 // Must not have write barriers because this may be called without a P.
2611 //go:nowritebarrierrec
2612 func startm(pp *p, spinning, lockheld bool) {
2613 // Disable preemption.
2615 // Every owned P must have an owner that will eventually stop it in the
2616 // event of a GC stop request. startm takes transient ownership of a P
2617 // (either from argument or pidleget below) and transfers ownership to
2618 // a started M, which will be responsible for performing the stop.
2620 // Preemption must be disabled during this transient ownership,
2621 // otherwise the P this is running on may enter GC stop while still
2622 // holding the transient P, leaving that P in limbo and deadlocking the
2625 // Callers passing a non-nil P must already be in non-preemptible
2626 // context, otherwise such preemption could occur on function entry to
2627 // startm. Callers passing a nil P may be preemptible, so we must
2628 // disable preemption before acquiring a P from pidleget below.
2635 // TODO(prattmic): All remaining calls to this function
2636 // with _p_ == nil could be cleaned up to find a P
2637 // before calling startm.
2638 throw("startm: P required for spinning=true")
2651 // No M is available, we must drop sched.lock and call newm.
2652 // However, we already own a P to assign to the M.
2654 // Once sched.lock is released, another G (e.g., in a syscall),
2655 // could find no idle P while checkdead finds a runnable G but
2656 // no running M's because this new M hasn't started yet, thus
2657 // throwing in an apparent deadlock.
2658 // This apparent deadlock is possible when startm is called
2659 // from sysmon, which doesn't count as a running M.
2661 // Avoid this situation by pre-allocating the ID for the new M,
2662 // thus marking it as 'running' before we drop sched.lock. This
2663 // new M will eventually run the scheduler to execute any
2670 // The caller incremented nmspinning, so set m.spinning in the new M.
2678 // Ownership transfer of pp committed by start in newm.
2679 // Preemption is now safe.
2687 throw("startm: m is spinning")
2690 throw("startm: m has p")
2692 if spinning && !runqempty(pp) {
2693 throw("startm: p has runnable gs")
2695 // The caller incremented nmspinning, so set m.spinning in the new M.
2696 nmp.spinning = spinning
2698 notewakeup(&nmp.park)
2699 // Ownership transfer of pp committed by wakeup. Preemption is now
2704 // Hands off P from syscall or locked M.
2705 // Always runs without a P, so write barriers are not allowed.
2707 //go:nowritebarrierrec
2708 func handoffp(pp *p) {
2709 // handoffp must start an M in any situation where
2710 // findrunnable would return a G to run on pp.
2712 // if it has local work, start it straight away
2713 if !runqempty(pp) || sched.runqsize != 0 {
2714 startm(pp, false, false)
2717 // if there's trace work to do, start it straight away
2718 if (traceEnabled() || traceShuttingDown()) && traceReaderAvailable() != nil {
2719 startm(pp, false, false)
2722 // if it has GC work, start it straight away
2723 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) {
2724 startm(pp, false, false)
2727 // no local work, check that there are no spinning/idle M's,
2728 // otherwise our help is not required
2729 if sched.nmspinning.Load()+sched.npidle.Load() == 0 && sched.nmspinning.CompareAndSwap(0, 1) { // TODO: fast atomic
2730 sched.needspinning.Store(0)
2731 startm(pp, true, false)
2735 if sched.gcwaiting.Load() {
2736 pp.status = _Pgcstop
2738 if sched.stopwait == 0 {
2739 notewakeup(&sched.stopnote)
2744 if pp.runSafePointFn != 0 && atomic.Cas(&pp.runSafePointFn, 1, 0) {
2745 sched.safePointFn(pp)
2746 sched.safePointWait--
2747 if sched.safePointWait == 0 {
2748 notewakeup(&sched.safePointNote)
2751 if sched.runqsize != 0 {
2753 startm(pp, false, false)
2756 // If this is the last running P and nobody is polling network,
2757 // need to wakeup another M to poll network.
2758 if sched.npidle.Load() == gomaxprocs-1 && sched.lastpoll.Load() != 0 {
2760 startm(pp, false, false)
2764 // The scheduler lock cannot be held when calling wakeNetPoller below
2765 // because wakeNetPoller may call wakep which may call startm.
2766 when := nobarrierWakeTime(pp)
2775 // Tries to add one more P to execute G's.
2776 // Called when a G is made runnable (newproc, ready).
2777 // Must be called with a P.
2779 // Be conservative about spinning threads, only start one if none exist
2781 if sched.nmspinning.Load() != 0 || !sched.nmspinning.CompareAndSwap(0, 1) {
2785 // Disable preemption until ownership of pp transfers to the next M in
2786 // startm. Otherwise preemption here would leave pp stuck waiting to
2789 // See preemption comment on acquirem in startm for more details.
2794 pp, _ = pidlegetSpinning(0)
2796 if sched.nmspinning.Add(-1) < 0 {
2797 throw("wakep: negative nmspinning")
2803 // Since we always have a P, the race in the "No M is available"
2804 // comment in startm doesn't apply during the small window between the
2805 // unlock here and lock in startm. A checkdead in between will always
2806 // see at least one running M (ours).
2809 startm(pp, true, false)
2814 // Stops execution of the current m that is locked to a g until the g is runnable again.
2815 // Returns with acquired P.
2816 func stoplockedm() {
2819 if gp.m.lockedg == 0 || gp.m.lockedg.ptr().lockedm.ptr() != gp.m {
2820 throw("stoplockedm: inconsistent locking")
2823 // Schedule another M to run this p.
2828 // Wait until another thread schedules lockedg again.
2830 status := readgstatus(gp.m.lockedg.ptr())
2831 if status&^_Gscan != _Grunnable {
2832 print("runtime:stoplockedm: lockedg (atomicstatus=", status, ") is not Grunnable or Gscanrunnable\n")
2833 dumpgstatus(gp.m.lockedg.ptr())
2834 throw("stoplockedm: not runnable")
2836 acquirep(gp.m.nextp.ptr())
2840 // Schedules the locked m to run the locked gp.
2841 // May run during STW, so write barriers are not allowed.
2843 //go:nowritebarrierrec
2844 func startlockedm(gp *g) {
2845 mp := gp.lockedm.ptr()
2847 throw("startlockedm: locked to me")
2850 throw("startlockedm: m has p")
2852 // directly handoff current P to the locked m
2856 notewakeup(&mp.park)
2860 // Stops the current m for stopTheWorld.
2861 // Returns when the world is restarted.
2865 if !sched.gcwaiting.Load() {
2866 throw("gcstopm: not waiting for gc")
2869 gp.m.spinning = false
2870 // OK to just drop nmspinning here,
2871 // startTheWorld will unpark threads as necessary.
2872 if sched.nmspinning.Add(-1) < 0 {
2873 throw("gcstopm: negative nmspinning")
2878 pp.status = _Pgcstop
2880 if sched.stopwait == 0 {
2881 notewakeup(&sched.stopnote)
2887 // Schedules gp to run on the current M.
2888 // If inheritTime is true, gp inherits the remaining time in the
2889 // current time slice. Otherwise, it starts a new time slice.
2892 // Write barriers are allowed because this is called immediately after
2893 // acquiring a P in several places.
2895 //go:yeswritebarrierrec
2896 func execute(gp *g, inheritTime bool) {
2899 if goroutineProfile.active {
2900 // Make sure that gp has had its stack written out to the goroutine
2901 // profile, exactly as it was when the goroutine profiler first stopped
2903 tryRecordGoroutineProfile(gp, osyield)
2906 // Assign gp.m before entering _Grunning so running Gs have an
2910 casgstatus(gp, _Grunnable, _Grunning)
2913 gp.stackguard0 = gp.stack.lo + stackGuard
2915 mp.p.ptr().schedtick++
2918 // Check whether the profiler needs to be turned on or off.
2919 hz := sched.profilehz
2920 if mp.profilehz != hz {
2921 setThreadCPUProfiler(hz)
2925 // GoSysExit has to happen when we have a P, but before GoStart.
2926 // So we emit it here.
2927 if gp.syscallsp != 0 {
2936 // Finds a runnable goroutine to execute.
2937 // Tries to steal from other P's, get g from local or global queue, poll network.
2938 // tryWakeP indicates that the returned goroutine is not normal (GC worker, trace
2939 // reader) so the caller should try to wake a P.
2940 func findRunnable() (gp *g, inheritTime, tryWakeP bool) {
2943 // The conditions here and in handoffp must agree: if
2944 // findrunnable would return a G to run, handoffp must start
2949 if sched.gcwaiting.Load() {
2953 if pp.runSafePointFn != 0 {
2957 // now and pollUntil are saved for work stealing later,
2958 // which may steal timers. It's important that between now
2959 // and then, nothing blocks, so these numbers remain mostly
2961 now, pollUntil, _ := checkTimers(pp, 0)
2963 // Try to schedule the trace reader.
2964 if traceEnabled() || traceShuttingDown() {
2967 casgstatus(gp, _Gwaiting, _Grunnable)
2968 traceGoUnpark(gp, 0)
2969 return gp, false, true
2973 // Try to schedule a GC worker.
2974 if gcBlackenEnabled != 0 {
2975 gp, tnow := gcController.findRunnableGCWorker(pp, now)
2977 return gp, false, true
2982 // Check the global runnable queue once in a while to ensure fairness.
2983 // Otherwise two goroutines can completely occupy the local runqueue
2984 // by constantly respawning each other.
2985 if pp.schedtick%61 == 0 && sched.runqsize > 0 {
2987 gp := globrunqget(pp, 1)
2990 return gp, false, false
2994 // Wake up the finalizer G.
2995 if fingStatus.Load()&(fingWait|fingWake) == fingWait|fingWake {
2996 if gp := wakefing(); gp != nil {
3000 if *cgo_yield != nil {
3001 asmcgocall(*cgo_yield, nil)
3005 if gp, inheritTime := runqget(pp); gp != nil {
3006 return gp, inheritTime, false
3010 if sched.runqsize != 0 {
3012 gp := globrunqget(pp, 0)
3015 return gp, false, false
3020 // This netpoll is only an optimization before we resort to stealing.
3021 // We can safely skip it if there are no waiters or a thread is blocked
3022 // in netpoll already. If there is any kind of logical race with that
3023 // blocked thread (e.g. it has already returned from netpoll, but does
3024 // not set lastpoll yet), this thread will do blocking netpoll below
3026 if netpollinited() && netpollAnyWaiters() && sched.lastpoll.Load() != 0 {
3027 if list, delta := netpoll(0); !list.empty() { // non-blocking
3030 netpollAdjustWaiters(delta)
3031 casgstatus(gp, _Gwaiting, _Grunnable)
3033 traceGoUnpark(gp, 0)
3035 return gp, false, false
3039 // Spinning Ms: steal work from other Ps.
3041 // Limit the number of spinning Ms to half the number of busy Ps.
3042 // This is necessary to prevent excessive CPU consumption when
3043 // GOMAXPROCS>>1 but the program parallelism is low.
3044 if mp.spinning || 2*sched.nmspinning.Load() < gomaxprocs-sched.npidle.Load() {
3049 gp, inheritTime, tnow, w, newWork := stealWork(now)
3051 // Successfully stole.
3052 return gp, inheritTime, false
3055 // There may be new timer or GC work; restart to
3061 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3062 // Earlier timer to wait for.
3067 // We have nothing to do.
3069 // If we're in the GC mark phase, can safely scan and blacken objects,
3070 // and have work to do, run idle-time marking rather than give up the P.
3071 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) && gcController.addIdleMarkWorker() {
3072 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
3074 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
3076 casgstatus(gp, _Gwaiting, _Grunnable)
3078 traceGoUnpark(gp, 0)
3080 return gp, false, false
3082 gcController.removeIdleMarkWorker()
3086 // If a callback returned and no other goroutine is awake,
3087 // then wake event handler goroutine which pauses execution
3088 // until a callback was triggered.
3089 gp, otherReady := beforeIdle(now, pollUntil)
3091 casgstatus(gp, _Gwaiting, _Grunnable)
3093 traceGoUnpark(gp, 0)
3095 return gp, false, false
3101 // Before we drop our P, make a snapshot of the allp slice,
3102 // which can change underfoot once we no longer block
3103 // safe-points. We don't need to snapshot the contents because
3104 // everything up to cap(allp) is immutable.
3105 allpSnapshot := allp
3106 // Also snapshot masks. Value changes are OK, but we can't allow
3107 // len to change out from under us.
3108 idlepMaskSnapshot := idlepMask
3109 timerpMaskSnapshot := timerpMask
3111 // return P and block
3113 if sched.gcwaiting.Load() || pp.runSafePointFn != 0 {
3117 if sched.runqsize != 0 {
3118 gp := globrunqget(pp, 0)
3120 return gp, false, false
3122 if !mp.spinning && sched.needspinning.Load() == 1 {
3123 // See "Delicate dance" comment below.
3128 if releasep() != pp {
3129 throw("findrunnable: wrong p")
3131 now = pidleput(pp, now)
3134 // Delicate dance: thread transitions from spinning to non-spinning
3135 // state, potentially concurrently with submission of new work. We must
3136 // drop nmspinning first and then check all sources again (with
3137 // #StoreLoad memory barrier in between). If we do it the other way
3138 // around, another thread can submit work after we've checked all
3139 // sources but before we drop nmspinning; as a result nobody will
3140 // unpark a thread to run the work.
3142 // This applies to the following sources of work:
3144 // * Goroutines added to the global or a per-P run queue.
3145 // * New/modified-earlier timers on a per-P timer heap.
3146 // * Idle-priority GC work (barring golang.org/issue/19112).
3148 // If we discover new work below, we need to restore m.spinning as a
3149 // signal for resetspinning to unpark a new worker thread (because
3150 // there can be more than one starving goroutine).
3152 // However, if after discovering new work we also observe no idle Ps
3153 // (either here or in resetspinning), we have a problem. We may be
3154 // racing with a non-spinning M in the block above, having found no
3155 // work and preparing to release its P and park. Allowing that P to go
3156 // idle will result in loss of work conservation (idle P while there is
3157 // runnable work). This could result in complete deadlock in the
3158 // unlikely event that we discover new work (from netpoll) right as we
3159 // are racing with _all_ other Ps going idle.
3161 // We use sched.needspinning to synchronize with non-spinning Ms going
3162 // idle. If needspinning is set when they are about to drop their P,
3163 // they abort the drop and instead become a new spinning M on our
3164 // behalf. If we are not racing and the system is truly fully loaded
3165 // then no spinning threads are required, and the next thread to
3166 // naturally become spinning will clear the flag.
3168 // Also see "Worker thread parking/unparking" comment at the top of the
3170 wasSpinning := mp.spinning
3173 if sched.nmspinning.Add(-1) < 0 {
3174 throw("findrunnable: negative nmspinning")
3177 // Note the for correctness, only the last M transitioning from
3178 // spinning to non-spinning must perform these rechecks to
3179 // ensure no missed work. However, the runtime has some cases
3180 // of transient increments of nmspinning that are decremented
3181 // without going through this path, so we must be conservative
3182 // and perform the check on all spinning Ms.
3184 // See https://go.dev/issue/43997.
3186 // Check global and P runqueues again.
3189 if sched.runqsize != 0 {
3190 pp, _ := pidlegetSpinning(0)
3192 gp := globrunqget(pp, 0)
3194 throw("global runq empty with non-zero runqsize")
3199 return gp, false, false
3204 pp := checkRunqsNoP(allpSnapshot, idlepMaskSnapshot)
3211 // Check for idle-priority GC work again.
3212 pp, gp := checkIdleGCNoP()
3217 // Run the idle worker.
3218 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
3219 casgstatus(gp, _Gwaiting, _Grunnable)
3221 traceGoUnpark(gp, 0)
3223 return gp, false, false
3226 // Finally, check for timer creation or expiry concurrently with
3227 // transitioning from spinning to non-spinning.
3229 // Note that we cannot use checkTimers here because it calls
3230 // adjusttimers which may need to allocate memory, and that isn't
3231 // allowed when we don't have an active P.
3232 pollUntil = checkTimersNoP(allpSnapshot, timerpMaskSnapshot, pollUntil)
3235 // Poll network until next timer.
3236 if netpollinited() && (netpollAnyWaiters() || pollUntil != 0) && sched.lastpoll.Swap(0) != 0 {
3237 sched.pollUntil.Store(pollUntil)
3239 throw("findrunnable: netpoll with p")
3242 throw("findrunnable: netpoll with spinning")
3249 delay = pollUntil - now
3255 // When using fake time, just poll.
3258 list, delta := netpoll(delay) // block until new work is available
3259 // Refresh now again, after potentially blocking.
3261 sched.pollUntil.Store(0)
3262 sched.lastpoll.Store(now)
3263 if faketime != 0 && list.empty() {
3264 // Using fake time and nothing is ready; stop M.
3265 // When all M's stop, checkdead will call timejump.
3270 pp, _ := pidleget(now)
3274 netpollAdjustWaiters(delta)
3280 netpollAdjustWaiters(delta)
3281 casgstatus(gp, _Gwaiting, _Grunnable)
3283 traceGoUnpark(gp, 0)
3285 return gp, false, false
3292 } else if pollUntil != 0 && netpollinited() {
3293 pollerPollUntil := sched.pollUntil.Load()
3294 if pollerPollUntil == 0 || pollerPollUntil > pollUntil {
3302 // pollWork reports whether there is non-background work this P could
3303 // be doing. This is a fairly lightweight check to be used for
3304 // background work loops, like idle GC. It checks a subset of the
3305 // conditions checked by the actual scheduler.
3306 func pollWork() bool {
3307 if sched.runqsize != 0 {
3310 p := getg().m.p.ptr()
3314 if netpollinited() && netpollAnyWaiters() && sched.lastpoll.Load() != 0 {
3315 if list, delta := netpoll(0); !list.empty() {
3317 netpollAdjustWaiters(delta)
3324 // stealWork attempts to steal a runnable goroutine or timer from any P.
3326 // If newWork is true, new work may have been readied.
3328 // If now is not 0 it is the current time. stealWork returns the passed time or
3329 // the current time if now was passed as 0.
3330 func stealWork(now int64) (gp *g, inheritTime bool, rnow, pollUntil int64, newWork bool) {
3331 pp := getg().m.p.ptr()
3335 const stealTries = 4
3336 for i := 0; i < stealTries; i++ {
3337 stealTimersOrRunNextG := i == stealTries-1
3339 for enum := stealOrder.start(fastrand()); !enum.done(); enum.next() {
3340 if sched.gcwaiting.Load() {
3341 // GC work may be available.
3342 return nil, false, now, pollUntil, true
3344 p2 := allp[enum.position()]
3349 // Steal timers from p2. This call to checkTimers is the only place
3350 // where we might hold a lock on a different P's timers. We do this
3351 // once on the last pass before checking runnext because stealing
3352 // from the other P's runnext should be the last resort, so if there
3353 // are timers to steal do that first.
3355 // We only check timers on one of the stealing iterations because
3356 // the time stored in now doesn't change in this loop and checking
3357 // the timers for each P more than once with the same value of now
3358 // is probably a waste of time.
3360 // timerpMask tells us whether the P may have timers at all. If it
3361 // can't, no need to check at all.
3362 if stealTimersOrRunNextG && timerpMask.read(enum.position()) {
3363 tnow, w, ran := checkTimers(p2, now)
3365 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3369 // Running the timers may have
3370 // made an arbitrary number of G's
3371 // ready and added them to this P's
3372 // local run queue. That invalidates
3373 // the assumption of runqsteal
3374 // that it always has room to add
3375 // stolen G's. So check now if there
3376 // is a local G to run.
3377 if gp, inheritTime := runqget(pp); gp != nil {
3378 return gp, inheritTime, now, pollUntil, ranTimer
3384 // Don't bother to attempt to steal if p2 is idle.
3385 if !idlepMask.read(enum.position()) {
3386 if gp := runqsteal(pp, p2, stealTimersOrRunNextG); gp != nil {
3387 return gp, false, now, pollUntil, ranTimer
3393 // No goroutines found to steal. Regardless, running a timer may have
3394 // made some goroutine ready that we missed. Indicate the next timer to
3396 return nil, false, now, pollUntil, ranTimer
3399 // Check all Ps for a runnable G to steal.
3401 // On entry we have no P. If a G is available to steal and a P is available,
3402 // the P is returned which the caller should acquire and attempt to steal the
3404 func checkRunqsNoP(allpSnapshot []*p, idlepMaskSnapshot pMask) *p {
3405 for id, p2 := range allpSnapshot {
3406 if !idlepMaskSnapshot.read(uint32(id)) && !runqempty(p2) {
3408 pp, _ := pidlegetSpinning(0)
3410 // Can't get a P, don't bother checking remaining Ps.
3419 // No work available.
3423 // Check all Ps for a timer expiring sooner than pollUntil.
3425 // Returns updated pollUntil value.
3426 func checkTimersNoP(allpSnapshot []*p, timerpMaskSnapshot pMask, pollUntil int64) int64 {
3427 for id, p2 := range allpSnapshot {
3428 if timerpMaskSnapshot.read(uint32(id)) {
3429 w := nobarrierWakeTime(p2)
3430 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3439 // Check for idle-priority GC, without a P on entry.
3441 // If some GC work, a P, and a worker G are all available, the P and G will be
3442 // returned. The returned P has not been wired yet.
3443 func checkIdleGCNoP() (*p, *g) {
3444 // N.B. Since we have no P, gcBlackenEnabled may change at any time; we
3445 // must check again after acquiring a P. As an optimization, we also check
3446 // if an idle mark worker is needed at all. This is OK here, because if we
3447 // observe that one isn't needed, at least one is currently running. Even if
3448 // it stops running, its own journey into the scheduler should schedule it
3449 // again, if need be (at which point, this check will pass, if relevant).
3450 if atomic.Load(&gcBlackenEnabled) == 0 || !gcController.needIdleMarkWorker() {
3453 if !gcMarkWorkAvailable(nil) {
3457 // Work is available; we can start an idle GC worker only if there is
3458 // an available P and available worker G.
3460 // We can attempt to acquire these in either order, though both have
3461 // synchronization concerns (see below). Workers are almost always
3462 // available (see comment in findRunnableGCWorker for the one case
3463 // there may be none). Since we're slightly less likely to find a P,
3464 // check for that first.
3466 // Synchronization: note that we must hold sched.lock until we are
3467 // committed to keeping it. Otherwise we cannot put the unnecessary P
3468 // back in sched.pidle without performing the full set of idle
3469 // transition checks.
3471 // If we were to check gcBgMarkWorkerPool first, we must somehow handle
3472 // the assumption in gcControllerState.findRunnableGCWorker that an
3473 // empty gcBgMarkWorkerPool is only possible if gcMarkDone is running.
3475 pp, now := pidlegetSpinning(0)
3481 // Now that we own a P, gcBlackenEnabled can't change (as it requires STW).
3482 if gcBlackenEnabled == 0 || !gcController.addIdleMarkWorker() {
3488 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
3492 gcController.removeIdleMarkWorker()
3498 return pp, node.gp.ptr()
3501 // wakeNetPoller wakes up the thread sleeping in the network poller if it isn't
3502 // going to wake up before the when argument; or it wakes an idle P to service
3503 // timers and the network poller if there isn't one already.
3504 func wakeNetPoller(when int64) {
3505 if sched.lastpoll.Load() == 0 {
3506 // In findrunnable we ensure that when polling the pollUntil
3507 // field is either zero or the time to which the current
3508 // poll is expected to run. This can have a spurious wakeup
3509 // but should never miss a wakeup.
3510 pollerPollUntil := sched.pollUntil.Load()
3511 if pollerPollUntil == 0 || pollerPollUntil > when {
3515 // There are no threads in the network poller, try to get
3516 // one there so it can handle new timers.
3517 if GOOS != "plan9" { // Temporary workaround - see issue #42303.
3523 func resetspinning() {
3526 throw("resetspinning: not a spinning m")
3528 gp.m.spinning = false
3529 nmspinning := sched.nmspinning.Add(-1)
3531 throw("findrunnable: negative nmspinning")
3533 // M wakeup policy is deliberately somewhat conservative, so check if we
3534 // need to wakeup another P here. See "Worker thread parking/unparking"
3535 // comment at the top of the file for details.
3539 // injectglist adds each runnable G on the list to some run queue,
3540 // and clears glist. If there is no current P, they are added to the
3541 // global queue, and up to npidle M's are started to run them.
3542 // Otherwise, for each idle P, this adds a G to the global queue
3543 // and starts an M. Any remaining G's are added to the current P's
3545 // This may temporarily acquire sched.lock.
3546 // Can run concurrently with GC.
3547 func injectglist(glist *gList) {
3552 for gp := glist.head.ptr(); gp != nil; gp = gp.schedlink.ptr() {
3553 traceGoUnpark(gp, 0)
3557 // Mark all the goroutines as runnable before we put them
3558 // on the run queues.
3559 head := glist.head.ptr()
3562 for gp := head; gp != nil; gp = gp.schedlink.ptr() {
3565 casgstatus(gp, _Gwaiting, _Grunnable)
3568 // Turn the gList into a gQueue.
3574 startIdle := func(n int) {
3575 for i := 0; i < n; i++ {
3576 mp := acquirem() // See comment in startm.
3579 pp, _ := pidlegetSpinning(0)
3586 startm(pp, false, true)
3592 pp := getg().m.p.ptr()
3595 globrunqputbatch(&q, int32(qsize))
3601 npidle := int(sched.npidle.Load())
3604 for n = 0; n < npidle && !q.empty(); n++ {
3610 globrunqputbatch(&globq, int32(n))
3617 runqputbatch(pp, &q, qsize)
3621 // One round of scheduler: find a runnable goroutine and execute it.
3627 throw("schedule: holding locks")
3630 if mp.lockedg != 0 {
3632 execute(mp.lockedg.ptr(), false) // Never returns.
3635 // We should not schedule away from a g that is executing a cgo call,
3636 // since the cgo call is using the m's g0 stack.
3638 throw("schedule: in cgo")
3645 // Safety check: if we are spinning, the run queue should be empty.
3646 // Check this before calling checkTimers, as that might call
3647 // goready to put a ready goroutine on the local run queue.
3648 if mp.spinning && (pp.runnext != 0 || pp.runqhead != pp.runqtail) {
3649 throw("schedule: spinning with local work")
3652 gp, inheritTime, tryWakeP := findRunnable() // blocks until work is available
3654 if debug.dontfreezetheworld > 0 && freezing.Load() {
3655 // See comment in freezetheworld. We don't want to perturb
3656 // scheduler state, so we didn't gcstopm in findRunnable, but
3657 // also don't want to allow new goroutines to run.
3659 // Deadlock here rather than in the findRunnable loop so if
3660 // findRunnable is stuck in a loop we don't perturb that
3666 // This thread is going to run a goroutine and is not spinning anymore,
3667 // so if it was marked as spinning we need to reset it now and potentially
3668 // start a new spinning M.
3673 if sched.disable.user && !schedEnabled(gp) {
3674 // Scheduling of this goroutine is disabled. Put it on
3675 // the list of pending runnable goroutines for when we
3676 // re-enable user scheduling and look again.
3678 if schedEnabled(gp) {
3679 // Something re-enabled scheduling while we
3680 // were acquiring the lock.
3683 sched.disable.runnable.pushBack(gp)
3690 // If about to schedule a not-normal goroutine (a GCworker or tracereader),
3691 // wake a P if there is one.
3695 if gp.lockedm != 0 {
3696 // Hands off own p to the locked m,
3697 // then blocks waiting for a new p.
3702 execute(gp, inheritTime)
3705 // dropg removes the association between m and the current goroutine m->curg (gp for short).
3706 // Typically a caller sets gp's status away from Grunning and then
3707 // immediately calls dropg to finish the job. The caller is also responsible
3708 // for arranging that gp will be restarted using ready at an
3709 // appropriate time. After calling dropg and arranging for gp to be
3710 // readied later, the caller can do other work but eventually should
3711 // call schedule to restart the scheduling of goroutines on this m.
3715 setMNoWB(&gp.m.curg.m, nil)
3716 setGNoWB(&gp.m.curg, nil)
3719 // checkTimers runs any timers for the P that are ready.
3720 // If now is not 0 it is the current time.
3721 // It returns the passed time or the current time if now was passed as 0.
3722 // and the time when the next timer should run or 0 if there is no next timer,
3723 // and reports whether it ran any timers.
3724 // If the time when the next timer should run is not 0,
3725 // it is always larger than the returned time.
3726 // We pass now in and out to avoid extra calls of nanotime.
3728 //go:yeswritebarrierrec
3729 func checkTimers(pp *p, now int64) (rnow, pollUntil int64, ran bool) {
3730 // If it's not yet time for the first timer, or the first adjusted
3731 // timer, then there is nothing to do.
3732 next := pp.timer0When.Load()
3733 nextAdj := pp.timerModifiedEarliest.Load()
3734 if next == 0 || (nextAdj != 0 && nextAdj < next) {
3739 // No timers to run or adjust.
3740 return now, 0, false
3747 // Next timer is not ready to run, but keep going
3748 // if we would clear deleted timers.
3749 // This corresponds to the condition below where
3750 // we decide whether to call clearDeletedTimers.
3751 if pp != getg().m.p.ptr() || int(pp.deletedTimers.Load()) <= int(pp.numTimers.Load()/4) {
3752 return now, next, false
3756 lock(&pp.timersLock)
3758 if len(pp.timers) > 0 {
3759 adjusttimers(pp, now)
3760 for len(pp.timers) > 0 {
3761 // Note that runtimer may temporarily unlock
3763 if tw := runtimer(pp, now); tw != 0 {
3773 // If this is the local P, and there are a lot of deleted timers,
3774 // clear them out. We only do this for the local P to reduce
3775 // lock contention on timersLock.
3776 if pp == getg().m.p.ptr() && int(pp.deletedTimers.Load()) > len(pp.timers)/4 {
3777 clearDeletedTimers(pp)
3780 unlock(&pp.timersLock)
3782 return now, pollUntil, ran
3785 func parkunlock_c(gp *g, lock unsafe.Pointer) bool {
3786 unlock((*mutex)(lock))
3790 // park continuation on g0.
3791 func park_m(gp *g) {
3795 traceGoPark(mp.waitTraceBlockReason, mp.waitTraceSkip)
3798 // N.B. Not using casGToWaiting here because the waitreason is
3799 // set by park_m's caller.
3800 casgstatus(gp, _Grunning, _Gwaiting)
3803 if fn := mp.waitunlockf; fn != nil {
3804 ok := fn(gp, mp.waitlock)
3805 mp.waitunlockf = nil
3809 traceGoUnpark(gp, 2)
3811 casgstatus(gp, _Gwaiting, _Grunnable)
3812 execute(gp, true) // Schedule it back, never returns.
3818 func goschedImpl(gp *g) {
3819 status := readgstatus(gp)
3820 if status&^_Gscan != _Grunning {
3822 throw("bad g status")
3824 casgstatus(gp, _Grunning, _Grunnable)
3837 // Gosched continuation on g0.
3838 func gosched_m(gp *g) {
3845 // goschedguarded is a forbidden-states-avoided version of gosched_m.
3846 func goschedguarded_m(gp *g) {
3848 if !canPreemptM(gp.m) {
3849 gogo(&gp.sched) // never return
3858 func gopreempt_m(gp *g) {
3865 // preemptPark parks gp and puts it in _Gpreempted.
3868 func preemptPark(gp *g) {
3870 traceGoPark(traceBlockPreempted, 0)
3872 status := readgstatus(gp)
3873 if status&^_Gscan != _Grunning {
3875 throw("bad g status")
3878 if gp.asyncSafePoint {
3879 // Double-check that async preemption does not
3880 // happen in SPWRITE assembly functions.
3881 // isAsyncSafePoint must exclude this case.
3882 f := findfunc(gp.sched.pc)
3884 throw("preempt at unknown pc")
3886 if f.flag&abi.FuncFlagSPWrite != 0 {
3887 println("runtime: unexpected SPWRITE function", funcname(f), "in async preempt")
3888 throw("preempt SPWRITE")
3892 // Transition from _Grunning to _Gscan|_Gpreempted. We can't
3893 // be in _Grunning when we dropg because then we'd be running
3894 // without an M, but the moment we're in _Gpreempted,
3895 // something could claim this G before we've fully cleaned it
3896 // up. Hence, we set the scan bit to lock down further
3897 // transitions until we can dropg.
3898 casGToPreemptScan(gp, _Grunning, _Gscan|_Gpreempted)
3900 casfrom_Gscanstatus(gp, _Gscan|_Gpreempted, _Gpreempted)
3904 // goyield is like Gosched, but it:
3905 // - emits a GoPreempt trace event instead of a GoSched trace event
3906 // - puts the current G on the runq of the current P instead of the globrunq
3912 func goyield_m(gp *g) {
3917 casgstatus(gp, _Grunning, _Grunnable)
3919 runqput(pp, gp, false)
3923 // Finishes execution of the current goroutine.
3934 // goexit continuation on g0.
3935 func goexit0(gp *g) {
3939 casgstatus(gp, _Grunning, _Gdead)
3940 gcController.addScannableStack(pp, -int64(gp.stack.hi-gp.stack.lo))
3941 if isSystemGoroutine(gp, false) {
3945 locked := gp.lockedm != 0
3948 gp.preemptStop = false
3949 gp.paniconfault = false
3950 gp._defer = nil // should be true already but just in case.
3951 gp._panic = nil // non-nil for Goexit during panic. points at stack-allocated data.
3953 gp.waitreason = waitReasonZero
3958 if gcBlackenEnabled != 0 && gp.gcAssistBytes > 0 {
3959 // Flush assist credit to the global pool. This gives
3960 // better information to pacing if the application is
3961 // rapidly creating an exiting goroutines.
3962 assistWorkPerByte := gcController.assistWorkPerByte.Load()
3963 scanCredit := int64(assistWorkPerByte * float64(gp.gcAssistBytes))
3964 gcController.bgScanCredit.Add(scanCredit)
3965 gp.gcAssistBytes = 0
3970 if GOARCH == "wasm" { // no threads yet on wasm
3972 schedule() // never returns
3975 if mp.lockedInt != 0 {
3976 print("invalid m->lockedInt = ", mp.lockedInt, "\n")
3977 throw("internal lockOSThread error")
3981 // The goroutine may have locked this thread because
3982 // it put it in an unusual kernel state. Kill it
3983 // rather than returning it to the thread pool.
3985 // Return to mstart, which will release the P and exit
3987 if GOOS != "plan9" { // See golang.org/issue/22227.
3990 // Clear lockedExt on plan9 since we may end up re-using
3998 // save updates getg().sched to refer to pc and sp so that a following
3999 // gogo will restore pc and sp.
4001 // save must not have write barriers because invoking a write barrier
4002 // can clobber getg().sched.
4005 //go:nowritebarrierrec
4006 func save(pc, sp uintptr) {
4009 if gp == gp.m.g0 || gp == gp.m.gsignal {
4010 // m.g0.sched is special and must describe the context
4011 // for exiting the thread. mstart1 writes to it directly.
4012 // m.gsignal.sched should not be used at all.
4013 // This check makes sure save calls do not accidentally
4014 // run in contexts where they'd write to system g's.
4015 throw("save on system g not allowed")
4022 // We need to ensure ctxt is zero, but can't have a write
4023 // barrier here. However, it should always already be zero.
4025 if gp.sched.ctxt != nil {
4030 // The goroutine g is about to enter a system call.
4031 // Record that it's not using the cpu anymore.
4032 // This is called only from the go syscall library and cgocall,
4033 // not from the low-level system calls used by the runtime.
4035 // Entersyscall cannot split the stack: the save must
4036 // make g->sched refer to the caller's stack segment, because
4037 // entersyscall is going to return immediately after.
4039 // Nothing entersyscall calls can split the stack either.
4040 // We cannot safely move the stack during an active call to syscall,
4041 // because we do not know which of the uintptr arguments are
4042 // really pointers (back into the stack).
4043 // In practice, this means that we make the fast path run through
4044 // entersyscall doing no-split things, and the slow path has to use systemstack
4045 // to run bigger things on the system stack.
4047 // reentersyscall is the entry point used by cgo callbacks, where explicitly
4048 // saved SP and PC are restored. This is needed when exitsyscall will be called
4049 // from a function further up in the call stack than the parent, as g->syscallsp
4050 // must always point to a valid stack frame. entersyscall below is the normal
4051 // entry point for syscalls, which obtains the SP and PC from the caller.
4054 // At the start of a syscall we emit traceGoSysCall to capture the stack trace.
4055 // If the syscall does not block, that is it, we do not emit any other events.
4056 // If the syscall blocks (that is, P is retaken), retaker emits traceGoSysBlock;
4057 // when syscall returns we emit traceGoSysExit and when the goroutine starts running
4058 // (potentially instantly, if exitsyscallfast returns true) we emit traceGoStart.
4059 // To ensure that traceGoSysExit is emitted strictly after traceGoSysBlock,
4060 // we remember current value of syscalltick in m (gp.m.syscalltick = gp.m.p.ptr().syscalltick),
4061 // whoever emits traceGoSysBlock increments p.syscalltick afterwards;
4062 // and we wait for the increment before emitting traceGoSysExit.
4063 // Note that the increment is done even if tracing is not enabled,
4064 // because tracing can be enabled in the middle of syscall. We don't want the wait to hang.
4067 func reentersyscall(pc, sp uintptr) {
4070 // Disable preemption because during this function g is in Gsyscall status,
4071 // but can have inconsistent g->sched, do not let GC observe it.
4074 // Entersyscall must not call any function that might split/grow the stack.
4075 // (See details in comment above.)
4076 // Catch calls that might, by replacing the stack guard with something that
4077 // will trip any stack check and leaving a flag to tell newstack to die.
4078 gp.stackguard0 = stackPreempt
4079 gp.throwsplit = true
4081 // Leave SP around for GC and traceback.
4085 casgstatus(gp, _Grunning, _Gsyscall)
4086 if staticLockRanking {
4087 // When doing static lock ranking casgstatus can call
4088 // systemstack which clobbers g.sched.
4091 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
4092 systemstack(func() {
4093 print("entersyscall inconsistent ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
4094 throw("entersyscall")
4099 systemstack(traceGoSysCall)
4100 // systemstack itself clobbers g.sched.{pc,sp} and we might
4101 // need them later when the G is genuinely blocked in a
4106 if sched.sysmonwait.Load() {
4107 systemstack(entersyscall_sysmon)
4111 if gp.m.p.ptr().runSafePointFn != 0 {
4112 // runSafePointFn may stack split if run on this stack
4113 systemstack(runSafePointFn)
4117 gp.m.syscalltick = gp.m.p.ptr().syscalltick
4122 atomic.Store(&pp.status, _Psyscall)
4123 if sched.gcwaiting.Load() {
4124 systemstack(entersyscall_gcwait)
4131 // Standard syscall entry used by the go syscall library and normal cgo calls.
4133 // This is exported via linkname to assembly in the syscall package and x/sys.
4136 //go:linkname entersyscall
4137 func entersyscall() {
4138 reentersyscall(getcallerpc(), getcallersp())
4141 func entersyscall_sysmon() {
4143 if sched.sysmonwait.Load() {
4144 sched.sysmonwait.Store(false)
4145 notewakeup(&sched.sysmonnote)
4150 func entersyscall_gcwait() {
4152 pp := gp.m.oldp.ptr()
4155 if sched.stopwait > 0 && atomic.Cas(&pp.status, _Psyscall, _Pgcstop) {
4161 if sched.stopwait--; sched.stopwait == 0 {
4162 notewakeup(&sched.stopnote)
4168 // The same as entersyscall(), but with a hint that the syscall is blocking.
4171 func entersyscallblock() {
4174 gp.m.locks++ // see comment in entersyscall
4175 gp.throwsplit = true
4176 gp.stackguard0 = stackPreempt // see comment in entersyscall
4177 gp.m.syscalltick = gp.m.p.ptr().syscalltick
4178 gp.m.p.ptr().syscalltick++
4180 // Leave SP around for GC and traceback.
4184 gp.syscallsp = gp.sched.sp
4185 gp.syscallpc = gp.sched.pc
4186 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
4190 systemstack(func() {
4191 print("entersyscallblock inconsistent ", hex(sp1), " ", hex(sp2), " ", hex(sp3), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
4192 throw("entersyscallblock")
4195 casgstatus(gp, _Grunning, _Gsyscall)
4196 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
4197 systemstack(func() {
4198 print("entersyscallblock inconsistent ", hex(sp), " ", hex(gp.sched.sp), " ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
4199 throw("entersyscallblock")
4203 systemstack(entersyscallblock_handoff)
4205 // Resave for traceback during blocked call.
4206 save(getcallerpc(), getcallersp())
4211 func entersyscallblock_handoff() {
4214 traceGoSysBlock(getg().m.p.ptr())
4216 handoffp(releasep())
4219 // The goroutine g exited its system call.
4220 // Arrange for it to run on a cpu again.
4221 // This is called only from the go syscall library, not
4222 // from the low-level system calls used by the runtime.
4224 // Write barriers are not allowed because our P may have been stolen.
4226 // This is exported via linkname to assembly in the syscall package.
4229 //go:nowritebarrierrec
4230 //go:linkname exitsyscall
4231 func exitsyscall() {
4234 gp.m.locks++ // see comment in entersyscall
4235 if getcallersp() > gp.syscallsp {
4236 throw("exitsyscall: syscall frame is no longer valid")
4240 oldp := gp.m.oldp.ptr()
4242 if exitsyscallfast(oldp) {
4243 // When exitsyscallfast returns success, we have a P so can now use
4245 if goroutineProfile.active {
4246 // Make sure that gp has had its stack written out to the goroutine
4247 // profile, exactly as it was when the goroutine profiler first
4248 // stopped the world.
4249 systemstack(func() {
4250 tryRecordGoroutineProfileWB(gp)
4254 if oldp != gp.m.p.ptr() || gp.m.syscalltick != gp.m.p.ptr().syscalltick {
4255 systemstack(traceGoStart)
4258 // There's a cpu for us, so we can run.
4259 gp.m.p.ptr().syscalltick++
4260 // We need to cas the status and scan before resuming...
4261 casgstatus(gp, _Gsyscall, _Grunning)
4263 // Garbage collector isn't running (since we are),
4264 // so okay to clear syscallsp.
4268 // restore the preemption request in case we've cleared it in newstack
4269 gp.stackguard0 = stackPreempt
4271 // otherwise restore the real stackGuard, we've spoiled it in entersyscall/entersyscallblock
4272 gp.stackguard0 = gp.stack.lo + stackGuard
4274 gp.throwsplit = false
4276 if sched.disable.user && !schedEnabled(gp) {
4277 // Scheduling of this goroutine is disabled.
4285 // Wait till traceGoSysBlock event is emitted.
4286 // This ensures consistency of the trace (the goroutine is started after it is blocked).
4287 for oldp != nil && oldp.syscalltick == gp.m.syscalltick {
4290 // We can't trace syscall exit right now because we don't have a P.
4291 // Tracing code can invoke write barriers that cannot run without a P.
4292 // So instead we remember the syscall exit time and emit the event
4293 // in execute when we have a P.
4294 gp.trace.sysExitTime = traceClockNow()
4299 // Call the scheduler.
4302 // Scheduler returned, so we're allowed to run now.
4303 // Delete the syscallsp information that we left for
4304 // the garbage collector during the system call.
4305 // Must wait until now because until gosched returns
4306 // we don't know for sure that the garbage collector
4309 gp.m.p.ptr().syscalltick++
4310 gp.throwsplit = false
4314 func exitsyscallfast(oldp *p) bool {
4317 // Freezetheworld sets stopwait but does not retake P's.
4318 if sched.stopwait == freezeStopWait {
4322 // Try to re-acquire the last P.
4323 if oldp != nil && oldp.status == _Psyscall && atomic.Cas(&oldp.status, _Psyscall, _Pidle) {
4324 // There's a cpu for us, so we can run.
4326 exitsyscallfast_reacquired()
4330 // Try to get any other idle P.
4331 if sched.pidle != 0 {
4333 systemstack(func() {
4334 ok = exitsyscallfast_pidle()
4335 if ok && traceEnabled() {
4337 // Wait till traceGoSysBlock event is emitted.
4338 // This ensures consistency of the trace (the goroutine is started after it is blocked).
4339 for oldp.syscalltick == gp.m.syscalltick {
4353 // exitsyscallfast_reacquired is the exitsyscall path on which this G
4354 // has successfully reacquired the P it was running on before the
4358 func exitsyscallfast_reacquired() {
4360 if gp.m.syscalltick != gp.m.p.ptr().syscalltick {
4362 // The p was retaken and then enter into syscall again (since gp.m.syscalltick has changed).
4363 // traceGoSysBlock for this syscall was already emitted,
4364 // but here we effectively retake the p from the new syscall running on the same p.
4365 systemstack(func() {
4366 // Denote blocking of the new syscall.
4367 traceGoSysBlock(gp.m.p.ptr())
4368 // Denote completion of the current syscall.
4372 gp.m.p.ptr().syscalltick++
4376 func exitsyscallfast_pidle() bool {
4378 pp, _ := pidleget(0)
4379 if pp != nil && sched.sysmonwait.Load() {
4380 sched.sysmonwait.Store(false)
4381 notewakeup(&sched.sysmonnote)
4391 // exitsyscall slow path on g0.
4392 // Failed to acquire P, enqueue gp as runnable.
4394 // Called via mcall, so gp is the calling g from this M.
4396 //go:nowritebarrierrec
4397 func exitsyscall0(gp *g) {
4398 casgstatus(gp, _Gsyscall, _Grunnable)
4402 if schedEnabled(gp) {
4409 // Below, we stoplockedm if gp is locked. globrunqput releases
4410 // ownership of gp, so we must check if gp is locked prior to
4411 // committing the release by unlocking sched.lock, otherwise we
4412 // could race with another M transitioning gp from unlocked to
4414 locked = gp.lockedm != 0
4415 } else if sched.sysmonwait.Load() {
4416 sched.sysmonwait.Store(false)
4417 notewakeup(&sched.sysmonnote)
4422 execute(gp, false) // Never returns.
4425 // Wait until another thread schedules gp and so m again.
4427 // N.B. lockedm must be this M, as this g was running on this M
4428 // before entersyscall.
4430 execute(gp, false) // Never returns.
4433 schedule() // Never returns.
4436 // Called from syscall package before fork.
4438 //go:linkname syscall_runtime_BeforeFork syscall.runtime_BeforeFork
4440 func syscall_runtime_BeforeFork() {
4443 // Block signals during a fork, so that the child does not run
4444 // a signal handler before exec if a signal is sent to the process
4445 // group. See issue #18600.
4447 sigsave(&gp.m.sigmask)
4450 // This function is called before fork in syscall package.
4451 // Code between fork and exec must not allocate memory nor even try to grow stack.
4452 // Here we spoil g.stackguard0 to reliably detect any attempts to grow stack.
4453 // runtime_AfterFork will undo this in parent process, but not in child.
4454 gp.stackguard0 = stackFork
4457 // Called from syscall package after fork in parent.
4459 //go:linkname syscall_runtime_AfterFork syscall.runtime_AfterFork
4461 func syscall_runtime_AfterFork() {
4464 // See the comments in beforefork.
4465 gp.stackguard0 = gp.stack.lo + stackGuard
4467 msigrestore(gp.m.sigmask)
4472 // inForkedChild is true while manipulating signals in the child process.
4473 // This is used to avoid calling libc functions in case we are using vfork.
4474 var inForkedChild bool
4476 // Called from syscall package after fork in child.
4477 // It resets non-sigignored signals to the default handler, and
4478 // restores the signal mask in preparation for the exec.
4480 // Because this might be called during a vfork, and therefore may be
4481 // temporarily sharing address space with the parent process, this must
4482 // not change any global variables or calling into C code that may do so.
4484 //go:linkname syscall_runtime_AfterForkInChild syscall.runtime_AfterForkInChild
4486 //go:nowritebarrierrec
4487 func syscall_runtime_AfterForkInChild() {
4488 // It's OK to change the global variable inForkedChild here
4489 // because we are going to change it back. There is no race here,
4490 // because if we are sharing address space with the parent process,
4491 // then the parent process can not be running concurrently.
4492 inForkedChild = true
4494 clearSignalHandlers()
4496 // When we are the child we are the only thread running,
4497 // so we know that nothing else has changed gp.m.sigmask.
4498 msigrestore(getg().m.sigmask)
4500 inForkedChild = false
4503 // pendingPreemptSignals is the number of preemption signals
4504 // that have been sent but not received. This is only used on Darwin.
4506 var pendingPreemptSignals atomic.Int32
4508 // Called from syscall package before Exec.
4510 //go:linkname syscall_runtime_BeforeExec syscall.runtime_BeforeExec
4511 func syscall_runtime_BeforeExec() {
4512 // Prevent thread creation during exec.
4515 // On Darwin, wait for all pending preemption signals to
4516 // be received. See issue #41702.
4517 if GOOS == "darwin" || GOOS == "ios" {
4518 for pendingPreemptSignals.Load() > 0 {
4524 // Called from syscall package after Exec.
4526 //go:linkname syscall_runtime_AfterExec syscall.runtime_AfterExec
4527 func syscall_runtime_AfterExec() {
4531 // Allocate a new g, with a stack big enough for stacksize bytes.
4532 func malg(stacksize int32) *g {
4535 stacksize = round2(stackSystem + stacksize)
4536 systemstack(func() {
4537 newg.stack = stackalloc(uint32(stacksize))
4539 newg.stackguard0 = newg.stack.lo + stackGuard
4540 newg.stackguard1 = ^uintptr(0)
4541 // Clear the bottom word of the stack. We record g
4542 // there on gsignal stack during VDSO on ARM and ARM64.
4543 *(*uintptr)(unsafe.Pointer(newg.stack.lo)) = 0
4548 // Create a new g running fn.
4549 // Put it on the queue of g's waiting to run.
4550 // The compiler turns a go statement into a call to this.
4551 func newproc(fn *funcval) {
4554 systemstack(func() {
4555 newg := newproc1(fn, gp, pc)
4557 pp := getg().m.p.ptr()
4558 runqput(pp, newg, true)
4566 // Create a new g in state _Grunnable, starting at fn. callerpc is the
4567 // address of the go statement that created this. The caller is responsible
4568 // for adding the new g to the scheduler.
4569 func newproc1(fn *funcval, callergp *g, callerpc uintptr) *g {
4571 fatal("go of nil func value")
4574 mp := acquirem() // disable preemption because we hold M and P in local vars.
4578 newg = malg(stackMin)
4579 casgstatus(newg, _Gidle, _Gdead)
4580 allgadd(newg) // publishes with a g->status of Gdead so GC scanner doesn't look at uninitialized stack.
4582 if newg.stack.hi == 0 {
4583 throw("newproc1: newg missing stack")
4586 if readgstatus(newg) != _Gdead {
4587 throw("newproc1: new g is not Gdead")
4590 totalSize := uintptr(4*goarch.PtrSize + sys.MinFrameSize) // extra space in case of reads slightly beyond frame
4591 totalSize = alignUp(totalSize, sys.StackAlign)
4592 sp := newg.stack.hi - totalSize
4595 *(*uintptr)(unsafe.Pointer(sp)) = 0
4598 if GOARCH == "arm64" {
4600 *(*uintptr)(unsafe.Pointer(sp - goarch.PtrSize)) = 0
4603 memclrNoHeapPointers(unsafe.Pointer(&newg.sched), unsafe.Sizeof(newg.sched))
4606 newg.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum // +PCQuantum so that previous instruction is in same function
4607 newg.sched.g = guintptr(unsafe.Pointer(newg))
4608 gostartcallfn(&newg.sched, fn)
4609 newg.parentGoid = callergp.goid
4610 newg.gopc = callerpc
4611 newg.ancestors = saveAncestors(callergp)
4612 newg.startpc = fn.fn
4613 if isSystemGoroutine(newg, false) {
4616 // Only user goroutines inherit pprof labels.
4618 newg.labels = mp.curg.labels
4620 if goroutineProfile.active {
4621 // A concurrent goroutine profile is running. It should include
4622 // exactly the set of goroutines that were alive when the goroutine
4623 // profiler first stopped the world. That does not include newg, so
4624 // mark it as not needing a profile before transitioning it from
4626 newg.goroutineProfiled.Store(goroutineProfileSatisfied)
4629 // Track initial transition?
4630 newg.trackingSeq = uint8(fastrand())
4631 if newg.trackingSeq%gTrackingPeriod == 0 {
4632 newg.tracking = true
4634 casgstatus(newg, _Gdead, _Grunnable)
4635 gcController.addScannableStack(pp, int64(newg.stack.hi-newg.stack.lo))
4637 if pp.goidcache == pp.goidcacheend {
4638 // Sched.goidgen is the last allocated id,
4639 // this batch must be [sched.goidgen+1, sched.goidgen+GoidCacheBatch].
4640 // At startup sched.goidgen=0, so main goroutine receives goid=1.
4641 pp.goidcache = sched.goidgen.Add(_GoidCacheBatch)
4642 pp.goidcache -= _GoidCacheBatch - 1
4643 pp.goidcacheend = pp.goidcache + _GoidCacheBatch
4645 newg.goid = pp.goidcache
4648 newg.racectx = racegostart(callerpc)
4650 if newg.labels != nil {
4651 // See note in proflabel.go on labelSync's role in synchronizing
4652 // with the reads in the signal handler.
4653 racereleasemergeg(newg, unsafe.Pointer(&labelSync))
4657 traceGoCreate(newg, newg.startpc)
4664 // saveAncestors copies previous ancestors of the given caller g and
4665 // includes info for the current caller into a new set of tracebacks for
4666 // a g being created.
4667 func saveAncestors(callergp *g) *[]ancestorInfo {
4668 // Copy all prior info, except for the root goroutine (goid 0).
4669 if debug.tracebackancestors <= 0 || callergp.goid == 0 {
4672 var callerAncestors []ancestorInfo
4673 if callergp.ancestors != nil {
4674 callerAncestors = *callergp.ancestors
4676 n := int32(len(callerAncestors)) + 1
4677 if n > debug.tracebackancestors {
4678 n = debug.tracebackancestors
4680 ancestors := make([]ancestorInfo, n)
4681 copy(ancestors[1:], callerAncestors)
4683 var pcs [tracebackInnerFrames]uintptr
4684 npcs := gcallers(callergp, 0, pcs[:])
4685 ipcs := make([]uintptr, npcs)
4687 ancestors[0] = ancestorInfo{
4689 goid: callergp.goid,
4690 gopc: callergp.gopc,
4693 ancestorsp := new([]ancestorInfo)
4694 *ancestorsp = ancestors
4698 // Put on gfree list.
4699 // If local list is too long, transfer a batch to the global list.
4700 func gfput(pp *p, gp *g) {
4701 if readgstatus(gp) != _Gdead {
4702 throw("gfput: bad status (not Gdead)")
4705 stksize := gp.stack.hi - gp.stack.lo
4707 if stksize != uintptr(startingStackSize) {
4708 // non-standard stack size - free it.
4717 if pp.gFree.n >= 64 {
4723 for pp.gFree.n >= 32 {
4724 gp := pp.gFree.pop()
4726 if gp.stack.lo == 0 {
4733 lock(&sched.gFree.lock)
4734 sched.gFree.noStack.pushAll(noStackQ)
4735 sched.gFree.stack.pushAll(stackQ)
4736 sched.gFree.n += inc
4737 unlock(&sched.gFree.lock)
4741 // Get from gfree list.
4742 // If local list is empty, grab a batch from global list.
4743 func gfget(pp *p) *g {
4745 if pp.gFree.empty() && (!sched.gFree.stack.empty() || !sched.gFree.noStack.empty()) {
4746 lock(&sched.gFree.lock)
4747 // Move a batch of free Gs to the P.
4748 for pp.gFree.n < 32 {
4749 // Prefer Gs with stacks.
4750 gp := sched.gFree.stack.pop()
4752 gp = sched.gFree.noStack.pop()
4761 unlock(&sched.gFree.lock)
4764 gp := pp.gFree.pop()
4769 if gp.stack.lo != 0 && gp.stack.hi-gp.stack.lo != uintptr(startingStackSize) {
4770 // Deallocate old stack. We kept it in gfput because it was the
4771 // right size when the goroutine was put on the free list, but
4772 // the right size has changed since then.
4773 systemstack(func() {
4780 if gp.stack.lo == 0 {
4781 // Stack was deallocated in gfput or just above. Allocate a new one.
4782 systemstack(func() {
4783 gp.stack = stackalloc(startingStackSize)
4785 gp.stackguard0 = gp.stack.lo + stackGuard
4788 racemalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4791 msanmalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4794 asanunpoison(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4800 // Purge all cached G's from gfree list to the global list.
4801 func gfpurge(pp *p) {
4807 for !pp.gFree.empty() {
4808 gp := pp.gFree.pop()
4810 if gp.stack.lo == 0 {
4817 lock(&sched.gFree.lock)
4818 sched.gFree.noStack.pushAll(noStackQ)
4819 sched.gFree.stack.pushAll(stackQ)
4820 sched.gFree.n += inc
4821 unlock(&sched.gFree.lock)
4824 // Breakpoint executes a breakpoint trap.
4829 // dolockOSThread is called by LockOSThread and lockOSThread below
4830 // after they modify m.locked. Do not allow preemption during this call,
4831 // or else the m might be different in this function than in the caller.
4834 func dolockOSThread() {
4835 if GOARCH == "wasm" {
4836 return // no threads on wasm yet
4839 gp.m.lockedg.set(gp)
4840 gp.lockedm.set(gp.m)
4843 // LockOSThread wires the calling goroutine to its current operating system thread.
4844 // The calling goroutine will always execute in that thread,
4845 // and no other goroutine will execute in it,
4846 // until the calling goroutine has made as many calls to
4847 // UnlockOSThread as to LockOSThread.
4848 // If the calling goroutine exits without unlocking the thread,
4849 // the thread will be terminated.
4851 // All init functions are run on the startup thread. Calling LockOSThread
4852 // from an init function will cause the main function to be invoked on
4855 // A goroutine should call LockOSThread before calling OS services or
4856 // non-Go library functions that depend on per-thread state.
4859 func LockOSThread() {
4860 if atomic.Load(&newmHandoff.haveTemplateThread) == 0 && GOOS != "plan9" {
4861 // If we need to start a new thread from the locked
4862 // thread, we need the template thread. Start it now
4863 // while we're in a known-good state.
4864 startTemplateThread()
4868 if gp.m.lockedExt == 0 {
4870 panic("LockOSThread nesting overflow")
4876 func lockOSThread() {
4877 getg().m.lockedInt++
4881 // dounlockOSThread is called by UnlockOSThread and unlockOSThread below
4882 // after they update m->locked. Do not allow preemption during this call,
4883 // or else the m might be in different in this function than in the caller.
4886 func dounlockOSThread() {
4887 if GOARCH == "wasm" {
4888 return // no threads on wasm yet
4891 if gp.m.lockedInt != 0 || gp.m.lockedExt != 0 {
4898 // UnlockOSThread undoes an earlier call to LockOSThread.
4899 // If this drops the number of active LockOSThread calls on the
4900 // calling goroutine to zero, it unwires the calling goroutine from
4901 // its fixed operating system thread.
4902 // If there are no active LockOSThread calls, this is a no-op.
4904 // Before calling UnlockOSThread, the caller must ensure that the OS
4905 // thread is suitable for running other goroutines. If the caller made
4906 // any permanent changes to the state of the thread that would affect
4907 // other goroutines, it should not call this function and thus leave
4908 // the goroutine locked to the OS thread until the goroutine (and
4909 // hence the thread) exits.
4912 func UnlockOSThread() {
4914 if gp.m.lockedExt == 0 {
4922 func unlockOSThread() {
4924 if gp.m.lockedInt == 0 {
4925 systemstack(badunlockosthread)
4931 func badunlockosthread() {
4932 throw("runtime: internal error: misuse of lockOSThread/unlockOSThread")
4935 func gcount() int32 {
4936 n := int32(atomic.Loaduintptr(&allglen)) - sched.gFree.n - sched.ngsys.Load()
4937 for _, pp := range allp {
4941 // All these variables can be changed concurrently, so the result can be inconsistent.
4942 // But at least the current goroutine is running.
4949 func mcount() int32 {
4950 return int32(sched.mnext - sched.nmfreed)
4954 signalLock atomic.Uint32
4956 // Must hold signalLock to write. Reads may be lock-free, but
4957 // signalLock should be taken to synchronize with changes.
4961 func _System() { _System() }
4962 func _ExternalCode() { _ExternalCode() }
4963 func _LostExternalCode() { _LostExternalCode() }
4964 func _GC() { _GC() }
4965 func _LostSIGPROFDuringAtomic64() { _LostSIGPROFDuringAtomic64() }
4966 func _VDSO() { _VDSO() }
4968 // Called if we receive a SIGPROF signal.
4969 // Called by the signal handler, may run during STW.
4971 //go:nowritebarrierrec
4972 func sigprof(pc, sp, lr uintptr, gp *g, mp *m) {
4973 if prof.hz.Load() == 0 {
4977 // If mp.profilehz is 0, then profiling is not enabled for this thread.
4978 // We must check this to avoid a deadlock between setcpuprofilerate
4979 // and the call to cpuprof.add, below.
4980 if mp != nil && mp.profilehz == 0 {
4984 // On mips{,le}/arm, 64bit atomics are emulated with spinlocks, in
4985 // runtime/internal/atomic. If SIGPROF arrives while the program is inside
4986 // the critical section, it creates a deadlock (when writing the sample).
4987 // As a workaround, create a counter of SIGPROFs while in critical section
4988 // to store the count, and pass it to sigprof.add() later when SIGPROF is
4989 // received from somewhere else (with _LostSIGPROFDuringAtomic64 as pc).
4990 if GOARCH == "mips" || GOARCH == "mipsle" || GOARCH == "arm" {
4991 if f := findfunc(pc); f.valid() {
4992 if hasPrefix(funcname(f), "runtime/internal/atomic") {
4993 cpuprof.lostAtomic++
4997 if GOARCH == "arm" && goarm < 7 && GOOS == "linux" && pc&0xffff0000 == 0xffff0000 {
4998 // runtime/internal/atomic functions call into kernel
4999 // helpers on arm < 7. See
5000 // runtime/internal/atomic/sys_linux_arm.s.
5001 cpuprof.lostAtomic++
5006 // Profiling runs concurrently with GC, so it must not allocate.
5007 // Set a trap in case the code does allocate.
5008 // Note that on windows, one thread takes profiles of all the
5009 // other threads, so mp is usually not getg().m.
5010 // In fact mp may not even be stopped.
5011 // See golang.org/issue/17165.
5012 getg().m.mallocing++
5015 var stk [maxCPUProfStack]uintptr
5017 if mp.ncgo > 0 && mp.curg != nil && mp.curg.syscallpc != 0 && mp.curg.syscallsp != 0 {
5019 // Check cgoCallersUse to make sure that we are not
5020 // interrupting other code that is fiddling with
5021 // cgoCallers. We are running in a signal handler
5022 // with all signals blocked, so we don't have to worry
5023 // about any other code interrupting us.
5024 if mp.cgoCallersUse.Load() == 0 && mp.cgoCallers != nil && mp.cgoCallers[0] != 0 {
5025 for cgoOff < len(mp.cgoCallers) && mp.cgoCallers[cgoOff] != 0 {
5028 n += copy(stk[:], mp.cgoCallers[:cgoOff])
5029 mp.cgoCallers[0] = 0
5032 // Collect Go stack that leads to the cgo call.
5033 u.initAt(mp.curg.syscallpc, mp.curg.syscallsp, 0, mp.curg, unwindSilentErrors)
5034 } else if usesLibcall() && mp.libcallg != 0 && mp.libcallpc != 0 && mp.libcallsp != 0 {
5035 // Libcall, i.e. runtime syscall on windows.
5036 // Collect Go stack that leads to the call.
5037 u.initAt(mp.libcallpc, mp.libcallsp, 0, mp.libcallg.ptr(), unwindSilentErrors)
5038 } else if mp != nil && mp.vdsoSP != 0 {
5039 // VDSO call, e.g. nanotime1 on Linux.
5040 // Collect Go stack that leads to the call.
5041 u.initAt(mp.vdsoPC, mp.vdsoSP, 0, gp, unwindSilentErrors|unwindJumpStack)
5043 u.initAt(pc, sp, lr, gp, unwindSilentErrors|unwindTrap|unwindJumpStack)
5045 n += tracebackPCs(&u, 0, stk[n:])
5048 // Normal traceback is impossible or has failed.
5049 // Account it against abstract "System" or "GC".
5052 pc = abi.FuncPCABIInternal(_VDSO) + sys.PCQuantum
5053 } else if pc > firstmoduledata.etext {
5054 // "ExternalCode" is better than "etext".
5055 pc = abi.FuncPCABIInternal(_ExternalCode) + sys.PCQuantum
5058 if mp.preemptoff != "" {
5059 stk[1] = abi.FuncPCABIInternal(_GC) + sys.PCQuantum
5061 stk[1] = abi.FuncPCABIInternal(_System) + sys.PCQuantum
5065 if prof.hz.Load() != 0 {
5066 // Note: it can happen on Windows that we interrupted a system thread
5067 // with no g, so gp could nil. The other nil checks are done out of
5068 // caution, but not expected to be nil in practice.
5069 var tagPtr *unsafe.Pointer
5070 if gp != nil && gp.m != nil && gp.m.curg != nil {
5071 tagPtr = &gp.m.curg.labels
5073 cpuprof.add(tagPtr, stk[:n])
5077 if gp != nil && gp.m != nil {
5078 if gp.m.curg != nil {
5083 traceCPUSample(gprof, pp, stk[:n])
5085 getg().m.mallocing--
5088 // setcpuprofilerate sets the CPU profiling rate to hz times per second.
5089 // If hz <= 0, setcpuprofilerate turns off CPU profiling.
5090 func setcpuprofilerate(hz int32) {
5091 // Force sane arguments.
5096 // Disable preemption, otherwise we can be rescheduled to another thread
5097 // that has profiling enabled.
5101 // Stop profiler on this thread so that it is safe to lock prof.
5102 // if a profiling signal came in while we had prof locked,
5103 // it would deadlock.
5104 setThreadCPUProfiler(0)
5106 for !prof.signalLock.CompareAndSwap(0, 1) {
5109 if prof.hz.Load() != hz {
5110 setProcessCPUProfiler(hz)
5113 prof.signalLock.Store(0)
5116 sched.profilehz = hz
5120 setThreadCPUProfiler(hz)
5126 // init initializes pp, which may be a freshly allocated p or a
5127 // previously destroyed p, and transitions it to status _Pgcstop.
5128 func (pp *p) init(id int32) {
5130 pp.status = _Pgcstop
5131 pp.sudogcache = pp.sudogbuf[:0]
5132 pp.deferpool = pp.deferpoolbuf[:0]
5134 if pp.mcache == nil {
5137 throw("missing mcache?")
5139 // Use the bootstrap mcache0. Only one P will get
5140 // mcache0: the one with ID 0.
5143 pp.mcache = allocmcache()
5146 if raceenabled && pp.raceprocctx == 0 {
5148 pp.raceprocctx = raceprocctx0
5149 raceprocctx0 = 0 // bootstrap
5151 pp.raceprocctx = raceproccreate()
5154 lockInit(&pp.timersLock, lockRankTimers)
5156 // This P may get timers when it starts running. Set the mask here
5157 // since the P may not go through pidleget (notably P 0 on startup).
5159 // Similarly, we may not go through pidleget before this P starts
5160 // running if it is P 0 on startup.
5164 // destroy releases all of the resources associated with pp and
5165 // transitions it to status _Pdead.
5167 // sched.lock must be held and the world must be stopped.
5168 func (pp *p) destroy() {
5169 assertLockHeld(&sched.lock)
5170 assertWorldStopped()
5172 // Move all runnable goroutines to the global queue
5173 for pp.runqhead != pp.runqtail {
5174 // Pop from tail of local queue
5176 gp := pp.runq[pp.runqtail%uint32(len(pp.runq))].ptr()
5177 // Push onto head of global queue
5180 if pp.runnext != 0 {
5181 globrunqputhead(pp.runnext.ptr())
5184 if len(pp.timers) > 0 {
5185 plocal := getg().m.p.ptr()
5186 // The world is stopped, but we acquire timersLock to
5187 // protect against sysmon calling timeSleepUntil.
5188 // This is the only case where we hold the timersLock of
5189 // more than one P, so there are no deadlock concerns.
5190 lock(&plocal.timersLock)
5191 lock(&pp.timersLock)
5192 moveTimers(plocal, pp.timers)
5194 pp.numTimers.Store(0)
5195 pp.deletedTimers.Store(0)
5196 pp.timer0When.Store(0)
5197 unlock(&pp.timersLock)
5198 unlock(&plocal.timersLock)
5200 // Flush p's write barrier buffer.
5201 if gcphase != _GCoff {
5205 for i := range pp.sudogbuf {
5206 pp.sudogbuf[i] = nil
5208 pp.sudogcache = pp.sudogbuf[:0]
5209 pp.pinnerCache = nil
5210 for j := range pp.deferpoolbuf {
5211 pp.deferpoolbuf[j] = nil
5213 pp.deferpool = pp.deferpoolbuf[:0]
5214 systemstack(func() {
5215 for i := 0; i < pp.mspancache.len; i++ {
5216 // Safe to call since the world is stopped.
5217 mheap_.spanalloc.free(unsafe.Pointer(pp.mspancache.buf[i]))
5219 pp.mspancache.len = 0
5221 pp.pcache.flush(&mheap_.pages)
5222 unlock(&mheap_.lock)
5224 freemcache(pp.mcache)
5229 if pp.timerRaceCtx != 0 {
5230 // The race detector code uses a callback to fetch
5231 // the proc context, so arrange for that callback
5232 // to see the right thing.
5233 // This hack only works because we are the only
5239 racectxend(pp.timerRaceCtx)
5244 raceprocdestroy(pp.raceprocctx)
5251 // Change number of processors.
5253 // sched.lock must be held, and the world must be stopped.
5255 // gcworkbufs must not be being modified by either the GC or the write barrier
5256 // code, so the GC must not be running if the number of Ps actually changes.
5258 // Returns list of Ps with local work, they need to be scheduled by the caller.
5259 func procresize(nprocs int32) *p {
5260 assertLockHeld(&sched.lock)
5261 assertWorldStopped()
5264 if old < 0 || nprocs <= 0 {
5265 throw("procresize: invalid arg")
5268 traceGomaxprocs(nprocs)
5271 // update statistics
5273 if sched.procresizetime != 0 {
5274 sched.totaltime += int64(old) * (now - sched.procresizetime)
5276 sched.procresizetime = now
5278 maskWords := (nprocs + 31) / 32
5280 // Grow allp if necessary.
5281 if nprocs > int32(len(allp)) {
5282 // Synchronize with retake, which could be running
5283 // concurrently since it doesn't run on a P.
5285 if nprocs <= int32(cap(allp)) {
5286 allp = allp[:nprocs]
5288 nallp := make([]*p, nprocs)
5289 // Copy everything up to allp's cap so we
5290 // never lose old allocated Ps.
5291 copy(nallp, allp[:cap(allp)])
5295 if maskWords <= int32(cap(idlepMask)) {
5296 idlepMask = idlepMask[:maskWords]
5297 timerpMask = timerpMask[:maskWords]
5299 nidlepMask := make([]uint32, maskWords)
5300 // No need to copy beyond len, old Ps are irrelevant.
5301 copy(nidlepMask, idlepMask)
5302 idlepMask = nidlepMask
5304 ntimerpMask := make([]uint32, maskWords)
5305 copy(ntimerpMask, timerpMask)
5306 timerpMask = ntimerpMask
5311 // initialize new P's
5312 for i := old; i < nprocs; i++ {
5318 atomicstorep(unsafe.Pointer(&allp[i]), unsafe.Pointer(pp))
5322 if gp.m.p != 0 && gp.m.p.ptr().id < nprocs {
5323 // continue to use the current P
5324 gp.m.p.ptr().status = _Prunning
5325 gp.m.p.ptr().mcache.prepareForSweep()
5327 // release the current P and acquire allp[0].
5329 // We must do this before destroying our current P
5330 // because p.destroy itself has write barriers, so we
5331 // need to do that from a valid P.
5334 // Pretend that we were descheduled
5335 // and then scheduled again to keep
5338 traceProcStop(gp.m.p.ptr())
5352 // g.m.p is now set, so we no longer need mcache0 for bootstrapping.
5355 // release resources from unused P's
5356 for i := nprocs; i < old; i++ {
5359 // can't free P itself because it can be referenced by an M in syscall
5363 if int32(len(allp)) != nprocs {
5365 allp = allp[:nprocs]
5366 idlepMask = idlepMask[:maskWords]
5367 timerpMask = timerpMask[:maskWords]
5372 for i := nprocs - 1; i >= 0; i-- {
5374 if gp.m.p.ptr() == pp {
5382 pp.link.set(runnablePs)
5386 stealOrder.reset(uint32(nprocs))
5387 var int32p *int32 = &gomaxprocs // make compiler check that gomaxprocs is an int32
5388 atomic.Store((*uint32)(unsafe.Pointer(int32p)), uint32(nprocs))
5390 // Notify the limiter that the amount of procs has changed.
5391 gcCPULimiter.resetCapacity(now, nprocs)
5396 // Associate p and the current m.
5398 // This function is allowed to have write barriers even if the caller
5399 // isn't because it immediately acquires pp.
5401 //go:yeswritebarrierrec
5402 func acquirep(pp *p) {
5403 // Do the part that isn't allowed to have write barriers.
5406 // Have p; write barriers now allowed.
5408 // Perform deferred mcache flush before this P can allocate
5409 // from a potentially stale mcache.
5410 pp.mcache.prepareForSweep()
5417 // wirep is the first step of acquirep, which actually associates the
5418 // current M to pp. This is broken out so we can disallow write
5419 // barriers for this part, since we don't yet have a P.
5421 //go:nowritebarrierrec
5427 throw("wirep: already in go")
5429 if pp.m != 0 || pp.status != _Pidle {
5434 print("wirep: p->m=", pp.m, "(", id, ") p->status=", pp.status, "\n")
5435 throw("wirep: invalid p state")
5439 pp.status = _Prunning
5442 // Disassociate p and the current m.
5443 func releasep() *p {
5447 throw("releasep: invalid arg")
5450 if pp.m.ptr() != gp.m || pp.status != _Prunning {
5451 print("releasep: m=", gp.m, " m->p=", gp.m.p.ptr(), " p->m=", hex(pp.m), " p->status=", pp.status, "\n")
5452 throw("releasep: invalid p state")
5455 traceProcStop(gp.m.p.ptr())
5463 func incidlelocked(v int32) {
5465 sched.nmidlelocked += v
5472 // Check for deadlock situation.
5473 // The check is based on number of running M's, if 0 -> deadlock.
5474 // sched.lock must be held.
5476 assertLockHeld(&sched.lock)
5478 // For -buildmode=c-shared or -buildmode=c-archive it's OK if
5479 // there are no running goroutines. The calling program is
5480 // assumed to be running.
5481 if islibrary || isarchive {
5485 // If we are dying because of a signal caught on an already idle thread,
5486 // freezetheworld will cause all running threads to block.
5487 // And runtime will essentially enter into deadlock state,
5488 // except that there is a thread that will call exit soon.
5489 if panicking.Load() > 0 {
5493 // If we are not running under cgo, but we have an extra M then account
5494 // for it. (It is possible to have an extra M on Windows without cgo to
5495 // accommodate callbacks created by syscall.NewCallback. See issue #6751
5498 if !iscgo && cgoHasExtraM && extraMLength.Load() > 0 {
5502 run := mcount() - sched.nmidle - sched.nmidlelocked - sched.nmsys
5507 print("runtime: checkdead: nmidle=", sched.nmidle, " nmidlelocked=", sched.nmidlelocked, " mcount=", mcount(), " nmsys=", sched.nmsys, "\n")
5509 throw("checkdead: inconsistent counts")
5513 forEachG(func(gp *g) {
5514 if isSystemGoroutine(gp, false) {
5517 s := readgstatus(gp)
5518 switch s &^ _Gscan {
5525 print("runtime: checkdead: find g ", gp.goid, " in status ", s, "\n")
5527 throw("checkdead: runnable g")
5530 if grunning == 0 { // possible if main goroutine calls runtime·Goexit()
5531 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5532 fatal("no goroutines (main called runtime.Goexit) - deadlock!")
5535 // Maybe jump time forward for playground.
5537 if when := timeSleepUntil(); when < maxWhen {
5540 // Start an M to steal the timer.
5541 pp, _ := pidleget(faketime)
5543 // There should always be a free P since
5544 // nothing is running.
5546 throw("checkdead: no p for timer")
5550 // There should always be a free M since
5551 // nothing is running.
5553 throw("checkdead: no m for timer")
5555 // M must be spinning to steal. We set this to be
5556 // explicit, but since this is the only M it would
5557 // become spinning on its own anyways.
5558 sched.nmspinning.Add(1)
5561 notewakeup(&mp.park)
5566 // There are no goroutines running, so we can look at the P's.
5567 for _, pp := range allp {
5568 if len(pp.timers) > 0 {
5573 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5574 fatal("all goroutines are asleep - deadlock!")
5577 // forcegcperiod is the maximum time in nanoseconds between garbage
5578 // collections. If we go this long without a garbage collection, one
5579 // is forced to run.
5581 // This is a variable for testing purposes. It normally doesn't change.
5582 var forcegcperiod int64 = 2 * 60 * 1e9
5584 // needSysmonWorkaround is true if the workaround for
5585 // golang.org/issue/42515 is needed on NetBSD.
5586 var needSysmonWorkaround bool = false
5588 // Always runs without a P, so write barriers are not allowed.
5590 //go:nowritebarrierrec
5597 lasttrace := int64(0)
5598 idle := 0 // how many cycles in succession we had not wokeup somebody
5602 if idle == 0 { // start with 20us sleep...
5604 } else if idle > 50 { // start doubling the sleep after 1ms...
5607 if delay > 10*1000 { // up to 10ms
5612 // sysmon should not enter deep sleep if schedtrace is enabled so that
5613 // it can print that information at the right time.
5615 // It should also not enter deep sleep if there are any active P's so
5616 // that it can retake P's from syscalls, preempt long running G's, and
5617 // poll the network if all P's are busy for long stretches.
5619 // It should wakeup from deep sleep if any P's become active either due
5620 // to exiting a syscall or waking up due to a timer expiring so that it
5621 // can resume performing those duties. If it wakes from a syscall it
5622 // resets idle and delay as a bet that since it had retaken a P from a
5623 // syscall before, it may need to do it again shortly after the
5624 // application starts work again. It does not reset idle when waking
5625 // from a timer to avoid adding system load to applications that spend
5626 // most of their time sleeping.
5628 if debug.schedtrace <= 0 && (sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs) {
5630 if sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs {
5631 syscallWake := false
5632 next := timeSleepUntil()
5634 sched.sysmonwait.Store(true)
5636 // Make wake-up period small enough
5637 // for the sampling to be correct.
5638 sleep := forcegcperiod / 2
5639 if next-now < sleep {
5642 shouldRelax := sleep >= osRelaxMinNS
5646 syscallWake = notetsleep(&sched.sysmonnote, sleep)
5651 sched.sysmonwait.Store(false)
5652 noteclear(&sched.sysmonnote)
5662 lock(&sched.sysmonlock)
5663 // Update now in case we blocked on sysmonnote or spent a long time
5664 // blocked on schedlock or sysmonlock above.
5667 // trigger libc interceptors if needed
5668 if *cgo_yield != nil {
5669 asmcgocall(*cgo_yield, nil)
5671 // poll network if not polled for more than 10ms
5672 lastpoll := sched.lastpoll.Load()
5673 if netpollinited() && lastpoll != 0 && lastpoll+10*1000*1000 < now {
5674 sched.lastpoll.CompareAndSwap(lastpoll, now)
5675 list, delta := netpoll(0) // non-blocking - returns list of goroutines
5677 // Need to decrement number of idle locked M's
5678 // (pretending that one more is running) before injectglist.
5679 // Otherwise it can lead to the following situation:
5680 // injectglist grabs all P's but before it starts M's to run the P's,
5681 // another M returns from syscall, finishes running its G,
5682 // observes that there is no work to do and no other running M's
5683 // and reports deadlock.
5687 netpollAdjustWaiters(delta)
5690 if GOOS == "netbsd" && needSysmonWorkaround {
5691 // netpoll is responsible for waiting for timer
5692 // expiration, so we typically don't have to worry
5693 // about starting an M to service timers. (Note that
5694 // sleep for timeSleepUntil above simply ensures sysmon
5695 // starts running again when that timer expiration may
5696 // cause Go code to run again).
5698 // However, netbsd has a kernel bug that sometimes
5699 // misses netpollBreak wake-ups, which can lead to
5700 // unbounded delays servicing timers. If we detect this
5701 // overrun, then startm to get something to handle the
5704 // See issue 42515 and
5705 // https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=50094.
5706 if next := timeSleepUntil(); next < now {
5707 startm(nil, false, false)
5710 if scavenger.sysmonWake.Load() != 0 {
5711 // Kick the scavenger awake if someone requested it.
5714 // retake P's blocked in syscalls
5715 // and preempt long running G's
5716 if retake(now) != 0 {
5721 // check if we need to force a GC
5722 if t := (gcTrigger{kind: gcTriggerTime, now: now}); t.test() && forcegc.idle.Load() {
5724 forcegc.idle.Store(false)
5726 list.push(forcegc.g)
5728 unlock(&forcegc.lock)
5730 if debug.schedtrace > 0 && lasttrace+int64(debug.schedtrace)*1000000 <= now {
5732 schedtrace(debug.scheddetail > 0)
5734 unlock(&sched.sysmonlock)
5738 type sysmontick struct {
5745 // forcePreemptNS is the time slice given to a G before it is
5747 const forcePreemptNS = 10 * 1000 * 1000 // 10ms
5749 func retake(now int64) uint32 {
5751 // Prevent allp slice changes. This lock will be completely
5752 // uncontended unless we're already stopping the world.
5754 // We can't use a range loop over allp because we may
5755 // temporarily drop the allpLock. Hence, we need to re-fetch
5756 // allp each time around the loop.
5757 for i := 0; i < len(allp); i++ {
5760 // This can happen if procresize has grown
5761 // allp but not yet created new Ps.
5764 pd := &pp.sysmontick
5767 if s == _Prunning || s == _Psyscall {
5768 // Preempt G if it's running for too long.
5769 t := int64(pp.schedtick)
5770 if int64(pd.schedtick) != t {
5771 pd.schedtick = uint32(t)
5773 } else if pd.schedwhen+forcePreemptNS <= now {
5775 // In case of syscall, preemptone() doesn't
5776 // work, because there is no M wired to P.
5781 // Retake P from syscall if it's there for more than 1 sysmon tick (at least 20us).
5782 t := int64(pp.syscalltick)
5783 if !sysretake && int64(pd.syscalltick) != t {
5784 pd.syscalltick = uint32(t)
5785 pd.syscallwhen = now
5788 // On the one hand we don't want to retake Ps if there is no other work to do,
5789 // but on the other hand we want to retake them eventually
5790 // because they can prevent the sysmon thread from deep sleep.
5791 if runqempty(pp) && sched.nmspinning.Load()+sched.npidle.Load() > 0 && pd.syscallwhen+10*1000*1000 > now {
5794 // Drop allpLock so we can take sched.lock.
5796 // Need to decrement number of idle locked M's
5797 // (pretending that one more is running) before the CAS.
5798 // Otherwise the M from which we retake can exit the syscall,
5799 // increment nmidle and report deadlock.
5801 if atomic.Cas(&pp.status, s, _Pidle) {
5818 // Tell all goroutines that they have been preempted and they should stop.
5819 // This function is purely best-effort. It can fail to inform a goroutine if a
5820 // processor just started running it.
5821 // No locks need to be held.
5822 // Returns true if preemption request was issued to at least one goroutine.
5823 func preemptall() bool {
5825 for _, pp := range allp {
5826 if pp.status != _Prunning {
5836 // Tell the goroutine running on processor P to stop.
5837 // This function is purely best-effort. It can incorrectly fail to inform the
5838 // goroutine. It can inform the wrong goroutine. Even if it informs the
5839 // correct goroutine, that goroutine might ignore the request if it is
5840 // simultaneously executing newstack.
5841 // No lock needs to be held.
5842 // Returns true if preemption request was issued.
5843 // The actual preemption will happen at some point in the future
5844 // and will be indicated by the gp->status no longer being
5846 func preemptone(pp *p) bool {
5848 if mp == nil || mp == getg().m {
5852 if gp == nil || gp == mp.g0 {
5858 // Every call in a goroutine checks for stack overflow by
5859 // comparing the current stack pointer to gp->stackguard0.
5860 // Setting gp->stackguard0 to StackPreempt folds
5861 // preemption into the normal stack overflow check.
5862 gp.stackguard0 = stackPreempt
5864 // Request an async preemption of this P.
5865 if preemptMSupported && debug.asyncpreemptoff == 0 {
5875 func schedtrace(detailed bool) {
5882 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)
5884 print(" gcwaiting=", sched.gcwaiting.Load(), " nmidlelocked=", sched.nmidlelocked, " stopwait=", sched.stopwait, " sysmonwait=", sched.sysmonwait.Load(), "\n")
5886 // We must be careful while reading data from P's, M's and G's.
5887 // Even if we hold schedlock, most data can be changed concurrently.
5888 // E.g. (p->m ? p->m->id : -1) can crash if p->m changes from non-nil to nil.
5889 for i, pp := range allp {
5891 h := atomic.Load(&pp.runqhead)
5892 t := atomic.Load(&pp.runqtail)
5894 print(" P", i, ": status=", pp.status, " schedtick=", pp.schedtick, " syscalltick=", pp.syscalltick, " m=")
5900 print(" runqsize=", t-h, " gfreecnt=", pp.gFree.n, " timerslen=", len(pp.timers), "\n")
5902 // In non-detailed mode format lengths of per-P run queues as:
5903 // [len1 len2 len3 len4]
5909 if i == len(allp)-1 {
5920 for mp := allm; mp != nil; mp = mp.alllink {
5922 print(" M", mp.id, ": p=")
5934 print(" mallocing=", mp.mallocing, " throwing=", mp.throwing, " preemptoff=", mp.preemptoff, " locks=", mp.locks, " dying=", mp.dying, " spinning=", mp.spinning, " blocked=", mp.blocked, " lockedg=")
5935 if lockedg := mp.lockedg.ptr(); lockedg != nil {
5943 forEachG(func(gp *g) {
5944 print(" G", gp.goid, ": status=", readgstatus(gp), "(", gp.waitreason.String(), ") m=")
5951 if lockedm := gp.lockedm.ptr(); lockedm != nil {
5961 // schedEnableUser enables or disables the scheduling of user
5964 // This does not stop already running user goroutines, so the caller
5965 // should first stop the world when disabling user goroutines.
5966 func schedEnableUser(enable bool) {
5968 if sched.disable.user == !enable {
5972 sched.disable.user = !enable
5974 n := sched.disable.n
5976 globrunqputbatch(&sched.disable.runnable, n)
5978 for ; n != 0 && sched.npidle.Load() != 0; n-- {
5979 startm(nil, false, false)
5986 // schedEnabled reports whether gp should be scheduled. It returns
5987 // false is scheduling of gp is disabled.
5989 // sched.lock must be held.
5990 func schedEnabled(gp *g) bool {
5991 assertLockHeld(&sched.lock)
5993 if sched.disable.user {
5994 return isSystemGoroutine(gp, true)
5999 // Put mp on midle list.
6000 // sched.lock must be held.
6001 // May run during STW, so write barriers are not allowed.
6003 //go:nowritebarrierrec
6005 assertLockHeld(&sched.lock)
6007 mp.schedlink = sched.midle
6013 // Try to get an m from midle list.
6014 // sched.lock must be held.
6015 // May run during STW, so write barriers are not allowed.
6017 //go:nowritebarrierrec
6019 assertLockHeld(&sched.lock)
6021 mp := sched.midle.ptr()
6023 sched.midle = mp.schedlink
6029 // Put gp on the global runnable queue.
6030 // sched.lock must be held.
6031 // May run during STW, so write barriers are not allowed.
6033 //go:nowritebarrierrec
6034 func globrunqput(gp *g) {
6035 assertLockHeld(&sched.lock)
6037 sched.runq.pushBack(gp)
6041 // Put gp at the head of the global runnable queue.
6042 // sched.lock must be held.
6043 // May run during STW, so write barriers are not allowed.
6045 //go:nowritebarrierrec
6046 func globrunqputhead(gp *g) {
6047 assertLockHeld(&sched.lock)
6053 // Put a batch of runnable goroutines on the global runnable queue.
6054 // This clears *batch.
6055 // sched.lock must be held.
6056 // May run during STW, so write barriers are not allowed.
6058 //go:nowritebarrierrec
6059 func globrunqputbatch(batch *gQueue, n int32) {
6060 assertLockHeld(&sched.lock)
6062 sched.runq.pushBackAll(*batch)
6067 // Try get a batch of G's from the global runnable queue.
6068 // sched.lock must be held.
6069 func globrunqget(pp *p, max int32) *g {
6070 assertLockHeld(&sched.lock)
6072 if sched.runqsize == 0 {
6076 n := sched.runqsize/gomaxprocs + 1
6077 if n > sched.runqsize {
6080 if max > 0 && n > max {
6083 if n > int32(len(pp.runq))/2 {
6084 n = int32(len(pp.runq)) / 2
6089 gp := sched.runq.pop()
6092 gp1 := sched.runq.pop()
6093 runqput(pp, gp1, false)
6098 // pMask is an atomic bitstring with one bit per P.
6101 // read returns true if P id's bit is set.
6102 func (p pMask) read(id uint32) bool {
6104 mask := uint32(1) << (id % 32)
6105 return (atomic.Load(&p[word]) & mask) != 0
6108 // set sets P id's bit.
6109 func (p pMask) set(id int32) {
6111 mask := uint32(1) << (id % 32)
6112 atomic.Or(&p[word], mask)
6115 // clear clears P id's bit.
6116 func (p pMask) clear(id int32) {
6118 mask := uint32(1) << (id % 32)
6119 atomic.And(&p[word], ^mask)
6122 // updateTimerPMask clears pp's timer mask if it has no timers on its heap.
6124 // Ideally, the timer mask would be kept immediately consistent on any timer
6125 // operations. Unfortunately, updating a shared global data structure in the
6126 // timer hot path adds too much overhead in applications frequently switching
6127 // between no timers and some timers.
6129 // As a compromise, the timer mask is updated only on pidleget / pidleput. A
6130 // running P (returned by pidleget) may add a timer at any time, so its mask
6131 // must be set. An idle P (passed to pidleput) cannot add new timers while
6132 // idle, so if it has no timers at that time, its mask may be cleared.
6134 // Thus, we get the following effects on timer-stealing in findrunnable:
6136 // - Idle Ps with no timers when they go idle are never checked in findrunnable
6137 // (for work- or timer-stealing; this is the ideal case).
6138 // - Running Ps must always be checked.
6139 // - Idle Ps whose timers are stolen must continue to be checked until they run
6140 // again, even after timer expiration.
6142 // When the P starts running again, the mask should be set, as a timer may be
6143 // added at any time.
6145 // TODO(prattmic): Additional targeted updates may improve the above cases.
6146 // e.g., updating the mask when stealing a timer.
6147 func updateTimerPMask(pp *p) {
6148 if pp.numTimers.Load() > 0 {
6152 // Looks like there are no timers, however another P may transiently
6153 // decrement numTimers when handling a timerModified timer in
6154 // checkTimers. We must take timersLock to serialize with these changes.
6155 lock(&pp.timersLock)
6156 if pp.numTimers.Load() == 0 {
6157 timerpMask.clear(pp.id)
6159 unlock(&pp.timersLock)
6162 // pidleput puts p on the _Pidle list. now must be a relatively recent call
6163 // to nanotime or zero. Returns now or the current time if now was zero.
6165 // This releases ownership of p. Once sched.lock is released it is no longer
6168 // sched.lock must be held.
6170 // May run during STW, so write barriers are not allowed.
6172 //go:nowritebarrierrec
6173 func pidleput(pp *p, now int64) int64 {
6174 assertLockHeld(&sched.lock)
6177 throw("pidleput: P has non-empty run queue")
6182 updateTimerPMask(pp) // clear if there are no timers.
6183 idlepMask.set(pp.id)
6184 pp.link = sched.pidle
6187 if !pp.limiterEvent.start(limiterEventIdle, now) {
6188 throw("must be able to track idle limiter event")
6193 // pidleget tries to get a p from the _Pidle list, acquiring ownership.
6195 // sched.lock must be held.
6197 // May run during STW, so write barriers are not allowed.
6199 //go:nowritebarrierrec
6200 func pidleget(now int64) (*p, int64) {
6201 assertLockHeld(&sched.lock)
6203 pp := sched.pidle.ptr()
6205 // Timer may get added at any time now.
6209 timerpMask.set(pp.id)
6210 idlepMask.clear(pp.id)
6211 sched.pidle = pp.link
6212 sched.npidle.Add(-1)
6213 pp.limiterEvent.stop(limiterEventIdle, now)
6218 // pidlegetSpinning tries to get a p from the _Pidle list, acquiring ownership.
6219 // This is called by spinning Ms (or callers than need a spinning M) that have
6220 // found work. If no P is available, this must synchronized with non-spinning
6221 // Ms that may be preparing to drop their P without discovering this work.
6223 // sched.lock must be held.
6225 // May run during STW, so write barriers are not allowed.
6227 //go:nowritebarrierrec
6228 func pidlegetSpinning(now int64) (*p, int64) {
6229 assertLockHeld(&sched.lock)
6231 pp, now := pidleget(now)
6233 // See "Delicate dance" comment in findrunnable. We found work
6234 // that we cannot take, we must synchronize with non-spinning
6235 // Ms that may be preparing to drop their P.
6236 sched.needspinning.Store(1)
6243 // runqempty reports whether pp has no Gs on its local run queue.
6244 // It never returns true spuriously.
6245 func runqempty(pp *p) bool {
6246 // Defend against a race where 1) pp has G1 in runqnext but runqhead == runqtail,
6247 // 2) runqput on pp kicks G1 to the runq, 3) runqget on pp empties runqnext.
6248 // Simply observing that runqhead == runqtail and then observing that runqnext == nil
6249 // does not mean the queue is empty.
6251 head := atomic.Load(&pp.runqhead)
6252 tail := atomic.Load(&pp.runqtail)
6253 runnext := atomic.Loaduintptr((*uintptr)(unsafe.Pointer(&pp.runnext)))
6254 if tail == atomic.Load(&pp.runqtail) {
6255 return head == tail && runnext == 0
6260 // To shake out latent assumptions about scheduling order,
6261 // we introduce some randomness into scheduling decisions
6262 // when running with the race detector.
6263 // The need for this was made obvious by changing the
6264 // (deterministic) scheduling order in Go 1.5 and breaking
6265 // many poorly-written tests.
6266 // With the randomness here, as long as the tests pass
6267 // consistently with -race, they shouldn't have latent scheduling
6269 const randomizeScheduler = raceenabled
6271 // runqput tries to put g on the local runnable queue.
6272 // If next is false, runqput adds g to the tail of the runnable queue.
6273 // If next is true, runqput puts g in the pp.runnext slot.
6274 // If the run queue is full, runnext puts g on the global queue.
6275 // Executed only by the owner P.
6276 func runqput(pp *p, gp *g, next bool) {
6277 if randomizeScheduler && next && fastrandn(2) == 0 {
6283 oldnext := pp.runnext
6284 if !pp.runnext.cas(oldnext, guintptr(unsafe.Pointer(gp))) {
6290 // Kick the old runnext out to the regular run queue.
6295 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
6297 if t-h < uint32(len(pp.runq)) {
6298 pp.runq[t%uint32(len(pp.runq))].set(gp)
6299 atomic.StoreRel(&pp.runqtail, t+1) // store-release, makes the item available for consumption
6302 if runqputslow(pp, gp, h, t) {
6305 // the queue is not full, now the put above must succeed
6309 // Put g and a batch of work from local runnable queue on global queue.
6310 // Executed only by the owner P.
6311 func runqputslow(pp *p, gp *g, h, t uint32) bool {
6312 var batch [len(pp.runq)/2 + 1]*g
6314 // First, grab a batch from local queue.
6317 if n != uint32(len(pp.runq)/2) {
6318 throw("runqputslow: queue is not full")
6320 for i := uint32(0); i < n; i++ {
6321 batch[i] = pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
6323 if !atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6328 if randomizeScheduler {
6329 for i := uint32(1); i <= n; i++ {
6330 j := fastrandn(i + 1)
6331 batch[i], batch[j] = batch[j], batch[i]
6335 // Link the goroutines.
6336 for i := uint32(0); i < n; i++ {
6337 batch[i].schedlink.set(batch[i+1])
6340 q.head.set(batch[0])
6341 q.tail.set(batch[n])
6343 // Now put the batch on global queue.
6345 globrunqputbatch(&q, int32(n+1))
6350 // runqputbatch tries to put all the G's on q on the local runnable queue.
6351 // If the queue is full, they are put on the global queue; in that case
6352 // this will temporarily acquire the scheduler lock.
6353 // Executed only by the owner P.
6354 func runqputbatch(pp *p, q *gQueue, qsize int) {
6355 h := atomic.LoadAcq(&pp.runqhead)
6358 for !q.empty() && t-h < uint32(len(pp.runq)) {
6360 pp.runq[t%uint32(len(pp.runq))].set(gp)
6366 if randomizeScheduler {
6367 off := func(o uint32) uint32 {
6368 return (pp.runqtail + o) % uint32(len(pp.runq))
6370 for i := uint32(1); i < n; i++ {
6371 j := fastrandn(i + 1)
6372 pp.runq[off(i)], pp.runq[off(j)] = pp.runq[off(j)], pp.runq[off(i)]
6376 atomic.StoreRel(&pp.runqtail, t)
6379 globrunqputbatch(q, int32(qsize))
6384 // Get g from local runnable queue.
6385 // If inheritTime is true, gp should inherit the remaining time in the
6386 // current time slice. Otherwise, it should start a new time slice.
6387 // Executed only by the owner P.
6388 func runqget(pp *p) (gp *g, inheritTime bool) {
6389 // If there's a runnext, it's the next G to run.
6391 // If the runnext is non-0 and the CAS fails, it could only have been stolen by another P,
6392 // because other Ps can race to set runnext to 0, but only the current P can set it to non-0.
6393 // Hence, there's no need to retry this CAS if it fails.
6394 if next != 0 && pp.runnext.cas(next, 0) {
6395 return next.ptr(), true
6399 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6404 gp := pp.runq[h%uint32(len(pp.runq))].ptr()
6405 if atomic.CasRel(&pp.runqhead, h, h+1) { // cas-release, commits consume
6411 // runqdrain drains the local runnable queue of pp and returns all goroutines in it.
6412 // Executed only by the owner P.
6413 func runqdrain(pp *p) (drainQ gQueue, n uint32) {
6414 oldNext := pp.runnext
6415 if oldNext != 0 && pp.runnext.cas(oldNext, 0) {
6416 drainQ.pushBack(oldNext.ptr())
6421 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6427 if qn > uint32(len(pp.runq)) { // read inconsistent h and t
6431 if !atomic.CasRel(&pp.runqhead, h, h+qn) { // cas-release, commits consume
6435 // We've inverted the order in which it gets G's from the local P's runnable queue
6436 // and then advances the head pointer because we don't want to mess up the statuses of G's
6437 // while runqdrain() and runqsteal() are running in parallel.
6438 // Thus we should advance the head pointer before draining the local P into a gQueue,
6439 // so that we can update any gp.schedlink only after we take the full ownership of G,
6440 // meanwhile, other P's can't access to all G's in local P's runnable queue and steal them.
6441 // See https://groups.google.com/g/golang-dev/c/0pTKxEKhHSc/m/6Q85QjdVBQAJ for more details.
6442 for i := uint32(0); i < qn; i++ {
6443 gp := pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
6450 // Grabs a batch of goroutines from pp's runnable queue into batch.
6451 // Batch is a ring buffer starting at batchHead.
6452 // Returns number of grabbed goroutines.
6453 // Can be executed by any P.
6454 func runqgrab(pp *p, batch *[256]guintptr, batchHead uint32, stealRunNextG bool) uint32 {
6456 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6457 t := atomic.LoadAcq(&pp.runqtail) // load-acquire, synchronize with the producer
6462 // Try to steal from pp.runnext.
6463 if next := pp.runnext; next != 0 {
6464 if pp.status == _Prunning {
6465 // Sleep to ensure that pp isn't about to run the g
6466 // we are about to steal.
6467 // The important use case here is when the g running
6468 // on pp ready()s another g and then almost
6469 // immediately blocks. Instead of stealing runnext
6470 // in this window, back off to give pp a chance to
6471 // schedule runnext. This will avoid thrashing gs
6472 // between different Ps.
6473 // A sync chan send/recv takes ~50ns as of time of
6474 // writing, so 3us gives ~50x overshoot.
6475 if GOOS != "windows" && GOOS != "openbsd" && GOOS != "netbsd" {
6478 // On some platforms system timer granularity is
6479 // 1-15ms, which is way too much for this
6480 // optimization. So just yield.
6484 if !pp.runnext.cas(next, 0) {
6487 batch[batchHead%uint32(len(batch))] = next
6493 if n > uint32(len(pp.runq)/2) { // read inconsistent h and t
6496 for i := uint32(0); i < n; i++ {
6497 g := pp.runq[(h+i)%uint32(len(pp.runq))]
6498 batch[(batchHead+i)%uint32(len(batch))] = g
6500 if atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6506 // Steal half of elements from local runnable queue of p2
6507 // and put onto local runnable queue of p.
6508 // Returns one of the stolen elements (or nil if failed).
6509 func runqsteal(pp, p2 *p, stealRunNextG bool) *g {
6511 n := runqgrab(p2, &pp.runq, t, stealRunNextG)
6516 gp := pp.runq[(t+n)%uint32(len(pp.runq))].ptr()
6520 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
6521 if t-h+n >= uint32(len(pp.runq)) {
6522 throw("runqsteal: runq overflow")
6524 atomic.StoreRel(&pp.runqtail, t+n) // store-release, makes the item available for consumption
6528 // A gQueue is a dequeue of Gs linked through g.schedlink. A G can only
6529 // be on one gQueue or gList at a time.
6530 type gQueue struct {
6535 // empty reports whether q is empty.
6536 func (q *gQueue) empty() bool {
6540 // push adds gp to the head of q.
6541 func (q *gQueue) push(gp *g) {
6542 gp.schedlink = q.head
6549 // pushBack adds gp to the tail of q.
6550 func (q *gQueue) pushBack(gp *g) {
6553 q.tail.ptr().schedlink.set(gp)
6560 // pushBackAll adds all Gs in q2 to the tail of q. After this q2 must
6562 func (q *gQueue) pushBackAll(q2 gQueue) {
6566 q2.tail.ptr().schedlink = 0
6568 q.tail.ptr().schedlink = q2.head
6575 // pop removes and returns the head of queue q. It returns nil if
6577 func (q *gQueue) pop() *g {
6580 q.head = gp.schedlink
6588 // popList takes all Gs in q and returns them as a gList.
6589 func (q *gQueue) popList() gList {
6590 stack := gList{q.head}
6595 // A gList is a list of Gs linked through g.schedlink. A G can only be
6596 // on one gQueue or gList at a time.
6601 // empty reports whether l is empty.
6602 func (l *gList) empty() bool {
6606 // push adds gp to the head of l.
6607 func (l *gList) push(gp *g) {
6608 gp.schedlink = l.head
6612 // pushAll prepends all Gs in q to l.
6613 func (l *gList) pushAll(q gQueue) {
6615 q.tail.ptr().schedlink = l.head
6620 // pop removes and returns the head of l. If l is empty, it returns nil.
6621 func (l *gList) pop() *g {
6624 l.head = gp.schedlink
6629 //go:linkname setMaxThreads runtime/debug.setMaxThreads
6630 func setMaxThreads(in int) (out int) {
6632 out = int(sched.maxmcount)
6633 if in > 0x7fffffff { // MaxInt32
6634 sched.maxmcount = 0x7fffffff
6636 sched.maxmcount = int32(in)
6644 func procPin() int {
6649 return int(mp.p.ptr().id)
6658 //go:linkname sync_runtime_procPin sync.runtime_procPin
6660 func sync_runtime_procPin() int {
6664 //go:linkname sync_runtime_procUnpin sync.runtime_procUnpin
6666 func sync_runtime_procUnpin() {
6670 //go:linkname sync_atomic_runtime_procPin sync/atomic.runtime_procPin
6672 func sync_atomic_runtime_procPin() int {
6676 //go:linkname sync_atomic_runtime_procUnpin sync/atomic.runtime_procUnpin
6678 func sync_atomic_runtime_procUnpin() {
6682 // Active spinning for sync.Mutex.
6684 //go:linkname sync_runtime_canSpin sync.runtime_canSpin
6686 func sync_runtime_canSpin(i int) bool {
6687 // sync.Mutex is cooperative, so we are conservative with spinning.
6688 // Spin only few times and only if running on a multicore machine and
6689 // GOMAXPROCS>1 and there is at least one other running P and local runq is empty.
6690 // As opposed to runtime mutex we don't do passive spinning here,
6691 // because there can be work on global runq or on other Ps.
6692 if i >= active_spin || ncpu <= 1 || gomaxprocs <= sched.npidle.Load()+sched.nmspinning.Load()+1 {
6695 if p := getg().m.p.ptr(); !runqempty(p) {
6701 //go:linkname sync_runtime_doSpin sync.runtime_doSpin
6703 func sync_runtime_doSpin() {
6704 procyield(active_spin_cnt)
6707 var stealOrder randomOrder
6709 // randomOrder/randomEnum are helper types for randomized work stealing.
6710 // They allow to enumerate all Ps in different pseudo-random orders without repetitions.
6711 // The algorithm is based on the fact that if we have X such that X and GOMAXPROCS
6712 // are coprime, then a sequences of (i + X) % GOMAXPROCS gives the required enumeration.
6713 type randomOrder struct {
6718 type randomEnum struct {
6725 func (ord *randomOrder) reset(count uint32) {
6727 ord.coprimes = ord.coprimes[:0]
6728 for i := uint32(1); i <= count; i++ {
6729 if gcd(i, count) == 1 {
6730 ord.coprimes = append(ord.coprimes, i)
6735 func (ord *randomOrder) start(i uint32) randomEnum {
6739 inc: ord.coprimes[i/ord.count%uint32(len(ord.coprimes))],
6743 func (enum *randomEnum) done() bool {
6744 return enum.i == enum.count
6747 func (enum *randomEnum) next() {
6749 enum.pos = (enum.pos + enum.inc) % enum.count
6752 func (enum *randomEnum) position() uint32 {
6756 func gcd(a, b uint32) uint32 {
6763 // An initTask represents the set of initializations that need to be done for a package.
6764 // Keep in sync with ../../test/noinit.go:initTask
6765 type initTask struct {
6766 state uint32 // 0 = uninitialized, 1 = in progress, 2 = done
6768 // followed by nfns pcs, uintptr sized, one per init function to run
6771 // inittrace stores statistics for init functions which are
6772 // updated by malloc and newproc when active is true.
6773 var inittrace tracestat
6775 type tracestat struct {
6776 active bool // init tracing activation status
6777 id uint64 // init goroutine id
6778 allocs uint64 // heap allocations
6779 bytes uint64 // heap allocated bytes
6782 func doInit(ts []*initTask) {
6783 for _, t := range ts {
6788 func doInit1(t *initTask) {
6790 case 2: // fully initialized
6792 case 1: // initialization in progress
6793 throw("recursive call during initialization - linker skew")
6794 default: // not initialized yet
6795 t.state = 1 // initialization in progress
6802 if inittrace.active {
6804 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6809 // We should have pruned all of these in the linker.
6810 throw("inittask with no functions")
6813 firstFunc := add(unsafe.Pointer(t), 8)
6814 for i := uint32(0); i < t.nfns; i++ {
6815 p := add(firstFunc, uintptr(i)*goarch.PtrSize)
6816 f := *(*func())(unsafe.Pointer(&p))
6820 if inittrace.active {
6822 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6825 f := *(*func())(unsafe.Pointer(&firstFunc))
6826 pkg := funcpkgpath(findfunc(abi.FuncPCABIInternal(f)))
6829 print("init ", pkg, " @")
6830 print(string(fmtNSAsMS(sbuf[:], uint64(start-runtimeInitTime))), " ms, ")
6831 print(string(fmtNSAsMS(sbuf[:], uint64(end-start))), " ms clock, ")
6832 print(string(itoa(sbuf[:], after.bytes-before.bytes)), " bytes, ")
6833 print(string(itoa(sbuf[:], after.allocs-before.allocs)), " allocs")
6837 t.state = 2 // initialization done