1 // Copyright 2014 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
11 "runtime/internal/atomic"
12 "runtime/internal/sys"
16 // set using cmd/go/internal/modload.ModInfoProg
19 // Goroutine scheduler
20 // The scheduler's job is to distribute ready-to-run goroutines over worker threads.
22 // The main concepts are:
24 // M - worker thread, or machine.
25 // P - processor, a resource that is required to execute Go code.
26 // M must have an associated P to execute Go code, however it can be
27 // blocked or in a syscall w/o an associated P.
29 // Design doc at https://golang.org/s/go11sched.
31 // Worker thread parking/unparking.
32 // We need to balance between keeping enough running worker threads to utilize
33 // available hardware parallelism and parking excessive running worker threads
34 // to conserve CPU resources and power. This is not simple for two reasons:
35 // (1) scheduler state is intentionally distributed (in particular, per-P work
36 // queues), so it is not possible to compute global predicates on fast paths;
37 // (2) for optimal thread management we would need to know the future (don't park
38 // a worker thread when a new goroutine will be readied in near future).
40 // Three rejected approaches that would work badly:
41 // 1. Centralize all scheduler state (would inhibit scalability).
42 // 2. Direct goroutine handoff. That is, when we ready a new goroutine and there
43 // is a spare P, unpark a thread and handoff it the thread and the goroutine.
44 // This would lead to thread state thrashing, as the thread that readied the
45 // goroutine can be out of work the very next moment, we will need to park it.
46 // Also, it would destroy locality of computation as we want to preserve
47 // dependent goroutines on the same thread; and introduce additional latency.
48 // 3. Unpark an additional thread whenever we ready a goroutine and there is an
49 // idle P, but don't do handoff. This would lead to excessive thread parking/
50 // unparking as the additional threads will instantly park without discovering
53 // The current approach:
55 // This approach applies to three primary sources of potential work: readying a
56 // goroutine, new/modified-earlier timers, and idle-priority GC. See below for
57 // additional details.
59 // We unpark an additional thread when we submit work if (this is wakep()):
60 // 1. There is an idle P, and
61 // 2. There are no "spinning" worker threads.
63 // A worker thread is considered spinning if it is out of local work and did
64 // not find work in the global run queue or netpoller; the spinning state is
65 // denoted in m.spinning and in sched.nmspinning. Threads unparked this way are
66 // also considered spinning; we don't do goroutine handoff so such threads are
67 // out of work initially. Spinning threads spin on looking for work in per-P
68 // run queues and timer heaps or from the GC before parking. If a spinning
69 // thread finds work it takes itself out of the spinning state and proceeds to
70 // execution. If it does not find work it takes itself out of the spinning
71 // state and then parks.
73 // If there is at least one spinning thread (sched.nmspinning>1), we don't
74 // unpark new threads when submitting work. To compensate for that, if the last
75 // spinning thread finds work and stops spinning, it must unpark a new spinning
76 // thread. This approach smooths out unjustified spikes of thread unparking,
77 // but at the same time guarantees eventual maximal CPU parallelism
80 // The main implementation complication is that we need to be very careful
81 // during spinning->non-spinning thread transition. This transition can race
82 // with submission of new work, and either one part or another needs to unpark
83 // another worker thread. If they both fail to do that, we can end up with
84 // semi-persistent CPU underutilization.
86 // The general pattern for submission is:
87 // 1. Submit work to the local run queue, timer heap, or GC state.
88 // 2. #StoreLoad-style memory barrier.
89 // 3. Check sched.nmspinning.
91 // The general pattern for spinning->non-spinning transition is:
92 // 1. Decrement nmspinning.
93 // 2. #StoreLoad-style memory barrier.
94 // 3. Check all per-P work queues and GC for new work.
96 // Note that all this complexity does not apply to global run queue as we are
97 // not sloppy about thread unparking when submitting to global queue. Also see
98 // comments for nmspinning manipulation.
100 // How these different sources of work behave varies, though it doesn't affect
101 // the synchronization approach:
102 // * Ready goroutine: this is an obvious source of work; the goroutine is
103 // immediately ready and must run on some thread eventually.
104 // * New/modified-earlier timer: The current timer implementation (see time.go)
105 // uses netpoll in a thread with no work available to wait for the soonest
106 // timer. If there is no thread waiting, we want a new spinning thread to go
108 // * Idle-priority GC: The GC wakes a stopped idle thread to contribute to
109 // background GC work (note: currently disabled per golang.org/issue/19112).
110 // Also see golang.org/issue/44313, as this should be extended to all GC
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_thread_start == nil {
213 throw("_cgo_thread_start missing")
215 if GOOS != "windows" {
216 if _cgo_setenv == nil {
217 throw("_cgo_setenv missing")
219 if _cgo_unsetenv == nil {
220 throw("_cgo_unsetenv missing")
223 if _cgo_notify_runtime_init_done == nil {
224 throw("_cgo_notify_runtime_init_done missing")
226 // Start the template thread in case we enter Go from
227 // a C-created thread and need to create a new thread.
228 startTemplateThread()
229 cgocall(_cgo_notify_runtime_init_done, nil)
232 // Run the initializing tasks. Depending on build mode this
233 // list can arrive a few different ways, but it will always
234 // contain the init tasks computed by the linker for all the
235 // packages in the program (excluding those added at runtime
236 // by package plugin).
237 for _, m := range activeModules() {
241 // Disable init tracing after main init done to avoid overhead
242 // of collecting statistics in malloc and newproc
243 inittrace.active = false
245 close(main_init_done)
250 if isarchive || islibrary {
251 // A program compiled with -buildmode=c-archive or c-shared
252 // has a main, but it is not executed.
255 fn := main_main // make an indirect call, as the linker doesn't know the address of the main package when laying down the runtime
258 runExitHooks(0) // run hooks now, since racefini does not return
262 // Make racy client program work: if panicking on
263 // another goroutine at the same time as main returns,
264 // let the other goroutine finish printing the panic trace.
265 // Once it does, it will exit. See issues 3934 and 20018.
266 if runningPanicDefers.Load() != 0 {
267 // Running deferred functions should not take long.
268 for c := 0; c < 1000; c++ {
269 if runningPanicDefers.Load() == 0 {
275 if panicking.Load() != 0 {
276 gopark(nil, nil, waitReasonPanicWait, traceEvGoStop, 1)
287 // os_beforeExit is called from os.Exit(0).
289 //go:linkname os_beforeExit os.runtime_beforeExit
290 func os_beforeExit(exitCode int) {
291 runExitHooks(exitCode)
292 if exitCode == 0 && raceenabled {
297 // start forcegc helper goroutine
302 func forcegchelper() {
304 lockInit(&forcegc.lock, lockRankForcegc)
307 if forcegc.idle.Load() {
308 throw("forcegc: phase error")
310 forcegc.idle.Store(true)
311 goparkunlock(&forcegc.lock, waitReasonForceGCIdle, traceEvGoBlock, 1)
312 // this goroutine is explicitly resumed by sysmon
313 if debug.gctrace > 0 {
316 // Time-triggered, fully concurrent.
317 gcStart(gcTrigger{kind: gcTriggerTime, now: nanotime()})
321 // Gosched yields the processor, allowing other goroutines to run. It does not
322 // suspend the current goroutine, so execution resumes automatically.
330 // goschedguarded yields the processor like gosched, but also checks
331 // for forbidden states and opts out of the yield in those cases.
334 func goschedguarded() {
335 mcall(goschedguarded_m)
338 // goschedIfBusy yields the processor like gosched, but only does so if
339 // there are no idle Ps or if we're on the only P and there's nothing in
340 // the run queue. In both cases, there is freely available idle time.
343 func goschedIfBusy() {
345 // Call gosched if gp.preempt is set; we may be in a tight loop that
346 // doesn't otherwise yield.
347 if !gp.preempt && sched.npidle.Load() > 0 {
353 // Puts the current goroutine into a waiting state and calls unlockf on the
356 // If unlockf returns false, the goroutine is resumed.
358 // unlockf must not access this G's stack, as it may be moved between
359 // the call to gopark and the call to unlockf.
361 // Note that because unlockf is called after putting the G into a waiting
362 // state, the G may have already been readied by the time unlockf is called
363 // unless there is external synchronization preventing the G from being
364 // readied. If unlockf returns false, it must guarantee that the G cannot be
365 // externally readied.
367 // Reason explains why the goroutine has been parked. It is displayed in stack
368 // traces and heap dumps. Reasons should be unique and descriptive. Do not
369 // re-use reasons, add new ones.
370 func gopark(unlockf func(*g, unsafe.Pointer) bool, lock unsafe.Pointer, reason waitReason, traceEv byte, traceskip int) {
371 if reason != waitReasonSleep {
372 checkTimeouts() // timeouts may expire while two goroutines keep the scheduler busy
376 status := readgstatus(gp)
377 if status != _Grunning && status != _Gscanrunning {
378 throw("gopark: bad g status")
381 mp.waitunlockf = unlockf
382 gp.waitreason = reason
383 mp.waittraceev = traceEv
384 mp.waittraceskip = traceskip
386 // can't do anything that might move the G between Ms here.
390 // Puts the current goroutine into a waiting state and unlocks the lock.
391 // The goroutine can be made runnable again by calling goready(gp).
392 func goparkunlock(lock *mutex, reason waitReason, traceEv byte, traceskip int) {
393 gopark(parkunlock_c, unsafe.Pointer(lock), reason, traceEv, traceskip)
396 func goready(gp *g, traceskip int) {
398 ready(gp, traceskip, true)
403 func acquireSudog() *sudog {
404 // Delicate dance: the semaphore implementation calls
405 // acquireSudog, acquireSudog calls new(sudog),
406 // new calls malloc, malloc can call the garbage collector,
407 // and the garbage collector calls the semaphore implementation
409 // Break the cycle by doing acquirem/releasem around new(sudog).
410 // The acquirem/releasem increments m.locks during new(sudog),
411 // which keeps the garbage collector from being invoked.
414 if len(pp.sudogcache) == 0 {
415 lock(&sched.sudoglock)
416 // First, try to grab a batch from central cache.
417 for len(pp.sudogcache) < cap(pp.sudogcache)/2 && sched.sudogcache != nil {
418 s := sched.sudogcache
419 sched.sudogcache = s.next
421 pp.sudogcache = append(pp.sudogcache, s)
423 unlock(&sched.sudoglock)
424 // If the central cache is empty, allocate a new one.
425 if len(pp.sudogcache) == 0 {
426 pp.sudogcache = append(pp.sudogcache, new(sudog))
429 n := len(pp.sudogcache)
430 s := pp.sudogcache[n-1]
431 pp.sudogcache[n-1] = nil
432 pp.sudogcache = pp.sudogcache[:n-1]
434 throw("acquireSudog: found s.elem != nil in cache")
441 func releaseSudog(s *sudog) {
443 throw("runtime: sudog with non-nil elem")
446 throw("runtime: sudog with non-false isSelect")
449 throw("runtime: sudog with non-nil next")
452 throw("runtime: sudog with non-nil prev")
454 if s.waitlink != nil {
455 throw("runtime: sudog with non-nil waitlink")
458 throw("runtime: sudog with non-nil c")
462 throw("runtime: releaseSudog with non-nil gp.param")
464 mp := acquirem() // avoid rescheduling to another P
466 if len(pp.sudogcache) == cap(pp.sudogcache) {
467 // Transfer half of local cache to the central cache.
468 var first, last *sudog
469 for len(pp.sudogcache) > cap(pp.sudogcache)/2 {
470 n := len(pp.sudogcache)
471 p := pp.sudogcache[n-1]
472 pp.sudogcache[n-1] = nil
473 pp.sudogcache = pp.sudogcache[:n-1]
481 lock(&sched.sudoglock)
482 last.next = sched.sudogcache
483 sched.sudogcache = first
484 unlock(&sched.sudoglock)
486 pp.sudogcache = append(pp.sudogcache, s)
490 // called from assembly.
491 func badmcall(fn func(*g)) {
492 throw("runtime: mcall called on m->g0 stack")
495 func badmcall2(fn func(*g)) {
496 throw("runtime: mcall function returned")
499 func badreflectcall() {
500 panic(plainError("arg size to reflect.call more than 1GB"))
504 //go:nowritebarrierrec
505 func badmorestackg0() {
506 writeErrStr("fatal: morestack on g0\n")
510 //go:nowritebarrierrec
511 func badmorestackgsignal() {
512 writeErrStr("fatal: morestack on gsignal\n")
520 func lockedOSThread() bool {
522 return gp.lockedm != 0 && gp.m.lockedg != 0
526 // allgs contains all Gs ever created (including dead Gs), and thus
529 // Access via the slice is protected by allglock or stop-the-world.
530 // Readers that cannot take the lock may (carefully!) use the atomic
535 // allglen and allgptr are atomic variables that contain len(allgs) and
536 // &allgs[0] respectively. Proper ordering depends on totally-ordered
537 // loads and stores. Writes are protected by allglock.
539 // allgptr is updated before allglen. Readers should read allglen
540 // before allgptr to ensure that allglen is always <= len(allgptr). New
541 // Gs appended during the race can be missed. For a consistent view of
542 // all Gs, allglock must be held.
544 // allgptr copies should always be stored as a concrete type or
545 // unsafe.Pointer, not uintptr, to ensure that GC can still reach it
546 // even if it points to a stale array.
551 func allgadd(gp *g) {
552 if readgstatus(gp) == _Gidle {
553 throw("allgadd: bad status Gidle")
557 allgs = append(allgs, gp)
558 if &allgs[0] != allgptr {
559 atomicstorep(unsafe.Pointer(&allgptr), unsafe.Pointer(&allgs[0]))
561 atomic.Storeuintptr(&allglen, uintptr(len(allgs)))
565 // allGsSnapshot returns a snapshot of the slice of all Gs.
567 // The world must be stopped or allglock must be held.
568 func allGsSnapshot() []*g {
569 assertWorldStoppedOrLockHeld(&allglock)
571 // Because the world is stopped or allglock is held, allgadd
572 // cannot happen concurrently with this. allgs grows
573 // monotonically and existing entries never change, so we can
574 // simply return a copy of the slice header. For added safety,
575 // we trim everything past len because that can still change.
576 return allgs[:len(allgs):len(allgs)]
579 // atomicAllG returns &allgs[0] and len(allgs) for use with atomicAllGIndex.
580 func atomicAllG() (**g, uintptr) {
581 length := atomic.Loaduintptr(&allglen)
582 ptr := (**g)(atomic.Loadp(unsafe.Pointer(&allgptr)))
586 // atomicAllGIndex returns ptr[i] with the allgptr returned from atomicAllG.
587 func atomicAllGIndex(ptr **g, i uintptr) *g {
588 return *(**g)(add(unsafe.Pointer(ptr), i*goarch.PtrSize))
591 // forEachG calls fn on every G from allgs.
593 // forEachG takes a lock to exclude concurrent addition of new Gs.
594 func forEachG(fn func(gp *g)) {
596 for _, gp := range allgs {
602 // forEachGRace calls fn on every G from allgs.
604 // forEachGRace avoids locking, but does not exclude addition of new Gs during
605 // execution, which may be missed.
606 func forEachGRace(fn func(gp *g)) {
607 ptr, length := atomicAllG()
608 for i := uintptr(0); i < length; i++ {
609 gp := atomicAllGIndex(ptr, i)
616 // Number of goroutine ids to grab from sched.goidgen to local per-P cache at once.
617 // 16 seems to provide enough amortization, but other than that it's mostly arbitrary number.
621 // cpuinit sets up CPU feature flags and calls internal/cpu.Initialize. env should be the complete
622 // value of the GODEBUG environment variable.
623 func cpuinit(env string) {
625 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
626 cpu.DebugOptions = true
630 // Support cpu feature variables are used in code generated by the compiler
631 // to guard execution of instructions that can not be assumed to be always supported.
634 x86HasPOPCNT = cpu.X86.HasPOPCNT
635 x86HasSSE41 = cpu.X86.HasSSE41
636 x86HasFMA = cpu.X86.HasFMA
639 armHasVFPv4 = cpu.ARM.HasVFPv4
642 arm64HasATOMICS = cpu.ARM64.HasATOMICS
646 // getGodebugEarly extracts the environment variable GODEBUG from the environment on
647 // Unix-like operating systems and returns it. This function exists to extract GODEBUG
648 // early before much of the runtime is initialized.
649 func getGodebugEarly() string {
650 const prefix = "GODEBUG="
653 case "aix", "darwin", "ios", "dragonfly", "freebsd", "netbsd", "openbsd", "illumos", "solaris", "linux":
654 // Similar to goenv_unix but extracts the environment value for
656 // TODO(moehrmann): remove when general goenvs() can be called before cpuinit()
658 for argv_index(argv, argc+1+n) != nil {
662 for i := int32(0); i < n; i++ {
663 p := argv_index(argv, argc+1+i)
664 s := unsafe.String(p, findnull(p))
666 if hasPrefix(s, prefix) {
667 env = gostring(p)[len(prefix):]
675 // The bootstrap sequence is:
679 // make & queue new G
680 // call runtime·mstart
682 // The new G calls runtime·main.
684 lockInit(&sched.lock, lockRankSched)
685 lockInit(&sched.sysmonlock, lockRankSysmon)
686 lockInit(&sched.deferlock, lockRankDefer)
687 lockInit(&sched.sudoglock, lockRankSudog)
688 lockInit(&deadlock, lockRankDeadlock)
689 lockInit(&paniclk, lockRankPanic)
690 lockInit(&allglock, lockRankAllg)
691 lockInit(&allpLock, lockRankAllp)
692 lockInit(&reflectOffs.lock, lockRankReflectOffs)
693 lockInit(&finlock, lockRankFin)
694 lockInit(&trace.bufLock, lockRankTraceBuf)
695 lockInit(&trace.stringsLock, lockRankTraceStrings)
696 lockInit(&trace.lock, lockRankTrace)
697 lockInit(&cpuprof.lock, lockRankCpuprof)
698 lockInit(&trace.stackTab.lock, lockRankTraceStackTab)
699 // Enforce that this lock is always a leaf lock.
700 // All of this lock's critical sections should be
702 lockInit(&memstats.heapStats.noPLock, lockRankLeafRank)
704 // raceinit must be the first call to race detector.
705 // In particular, it must be done before mallocinit below calls racemapshadow.
708 gp.racectx, raceprocctx0 = raceinit()
711 sched.maxmcount = 10000
713 // The world starts stopped.
719 godebug := getGodebugEarly()
720 initPageTrace(godebug) // must run after mallocinit but before anything allocates
721 cpuinit(godebug) // must run before alginit
722 alginit() // maps, hash, fastrand must not be used before this call
723 fastrandinit() // must run before mcommoninit
724 mcommoninit(gp.m, -1)
725 modulesinit() // provides activeModules
726 typelinksinit() // uses maps, activeModules
727 itabsinit() // uses activeModules
728 stkobjinit() // must run before GC starts
730 sigsave(&gp.m.sigmask)
731 initSigmask = gp.m.sigmask
738 // if disableMemoryProfiling is set, update MemProfileRate to 0 to turn off memprofile.
739 // Note: parsedebugvars may update MemProfileRate, but when disableMemoryProfiling is
740 // set to true by the linker, it means that nothing is consuming the profile, it is
741 // safe to set MemProfileRate to 0.
742 if disableMemoryProfiling {
747 sched.lastpoll.Store(nanotime())
749 if n, ok := atoi32(gogetenv("GOMAXPROCS")); ok && n > 0 {
752 if procresize(procs) != nil {
753 throw("unknown runnable goroutine during bootstrap")
757 // World is effectively started now, as P's can run.
760 if buildVersion == "" {
761 // Condition should never trigger. This code just serves
762 // to ensure runtime·buildVersion is kept in the resulting binary.
763 buildVersion = "unknown"
765 if len(modinfo) == 1 {
766 // Condition should never trigger. This code just serves
767 // to ensure runtime·modinfo is kept in the resulting binary.
772 func dumpgstatus(gp *g) {
774 print("runtime: gp: gp=", gp, ", goid=", gp.goid, ", gp->atomicstatus=", readgstatus(gp), "\n")
775 print("runtime: getg: g=", thisg, ", goid=", thisg.goid, ", g->atomicstatus=", readgstatus(thisg), "\n")
778 // sched.lock must be held.
780 assertLockHeld(&sched.lock)
782 if mcount() > sched.maxmcount {
783 print("runtime: program exceeds ", sched.maxmcount, "-thread limit\n")
784 throw("thread exhaustion")
788 // mReserveID returns the next ID to use for a new m. This new m is immediately
789 // considered 'running' by checkdead.
791 // sched.lock must be held.
792 func mReserveID() int64 {
793 assertLockHeld(&sched.lock)
795 if sched.mnext+1 < sched.mnext {
796 throw("runtime: thread ID overflow")
804 // Pre-allocated ID may be passed as 'id', or omitted by passing -1.
805 func mcommoninit(mp *m, id int64) {
808 // g0 stack won't make sense for user (and is not necessary unwindable).
810 callers(1, mp.createstack[:])
821 lo := uint32(int64Hash(uint64(mp.id), fastrandseed))
822 hi := uint32(int64Hash(uint64(cputicks()), ^fastrandseed))
826 // Same behavior as for 1.17.
827 // TODO: Simplify this.
828 if goarch.BigEndian {
829 mp.fastrand = uint64(lo)<<32 | uint64(hi)
831 mp.fastrand = uint64(hi)<<32 | uint64(lo)
835 if mp.gsignal != nil {
836 mp.gsignal.stackguard1 = mp.gsignal.stack.lo + stackGuard
839 // Add to allm so garbage collector doesn't free g->m
840 // when it is just in a register or thread-local storage.
843 // NumCgoCall() iterates over allm w/o schedlock,
844 // so we need to publish it safely.
845 atomicstorep(unsafe.Pointer(&allm), unsafe.Pointer(mp))
848 // Allocate memory to hold a cgo traceback if the cgo call crashes.
849 if iscgo || GOOS == "solaris" || GOOS == "illumos" || GOOS == "windows" {
850 mp.cgoCallers = new(cgoCallers)
854 func (mp *m) becomeSpinning() {
856 sched.nmspinning.Add(1)
857 sched.needspinning.Store(0)
860 func (mp *m) hasCgoOnStack() bool {
861 return mp.ncgo > 0 || mp.isextra
864 var fastrandseed uintptr
866 func fastrandinit() {
867 s := (*[unsafe.Sizeof(fastrandseed)]byte)(unsafe.Pointer(&fastrandseed))[:]
871 // Mark gp ready to run.
872 func ready(gp *g, traceskip int, next bool) {
874 traceGoUnpark(gp, traceskip)
877 status := readgstatus(gp)
880 mp := acquirem() // disable preemption because it can be holding p in a local var
881 if status&^_Gscan != _Gwaiting {
883 throw("bad g->status in ready")
886 // status is Gwaiting or Gscanwaiting, make Grunnable and put on runq
887 casgstatus(gp, _Gwaiting, _Grunnable)
888 runqput(mp.p.ptr(), gp, next)
893 // freezeStopWait is a large value that freezetheworld sets
894 // sched.stopwait to in order to request that all Gs permanently stop.
895 const freezeStopWait = 0x7fffffff
897 // freezing is set to non-zero if the runtime is trying to freeze the
899 var freezing atomic.Bool
901 // Similar to stopTheWorld but best-effort and can be called several times.
902 // There is no reverse operation, used during crashing.
903 // This function must not lock any mutexes.
904 func freezetheworld() {
906 // stopwait and preemption requests can be lost
907 // due to races with concurrently executing threads,
908 // so try several times
909 for i := 0; i < 5; i++ {
910 // this should tell the scheduler to not start any new goroutines
911 sched.stopwait = freezeStopWait
912 sched.gcwaiting.Store(true)
913 // this should stop running goroutines
915 break // no running goroutines
925 // All reads and writes of g's status go through readgstatus, casgstatus
926 // castogscanstatus, casfrom_Gscanstatus.
929 func readgstatus(gp *g) uint32 {
930 return gp.atomicstatus.Load()
933 // The Gscanstatuses are acting like locks and this releases them.
934 // If it proves to be a performance hit we should be able to make these
935 // simple atomic stores but for now we are going to throw if
936 // we see an inconsistent state.
937 func casfrom_Gscanstatus(gp *g, oldval, newval uint32) {
940 // Check that transition is valid.
943 print("runtime: casfrom_Gscanstatus bad oldval gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
945 throw("casfrom_Gscanstatus:top gp->status is not in scan state")
951 if newval == oldval&^_Gscan {
952 success = gp.atomicstatus.CompareAndSwap(oldval, newval)
956 print("runtime: casfrom_Gscanstatus failed gp=", gp, ", oldval=", hex(oldval), ", newval=", hex(newval), "\n")
958 throw("casfrom_Gscanstatus: gp->status is not in scan state")
960 releaseLockRank(lockRankGscan)
963 // This will return false if the gp is not in the expected status and the cas fails.
964 // This acts like a lock acquire while the casfromgstatus acts like a lock release.
965 func castogscanstatus(gp *g, oldval, newval uint32) bool {
971 if newval == oldval|_Gscan {
972 r := gp.atomicstatus.CompareAndSwap(oldval, newval)
974 acquireLockRank(lockRankGscan)
980 print("runtime: castogscanstatus oldval=", hex(oldval), " newval=", hex(newval), "\n")
981 throw("castogscanstatus")
985 // casgstatusAlwaysTrack is a debug flag that causes casgstatus to always track
986 // various latencies on every transition instead of sampling them.
987 var casgstatusAlwaysTrack = false
989 // If asked to move to or from a Gscanstatus this will throw. Use the castogscanstatus
990 // and casfrom_Gscanstatus instead.
991 // casgstatus will loop if the g->atomicstatus is in a Gscan status until the routine that
992 // put it in the Gscan state is finished.
995 func casgstatus(gp *g, oldval, newval uint32) {
996 if (oldval&_Gscan != 0) || (newval&_Gscan != 0) || oldval == newval {
998 print("runtime: casgstatus: oldval=", hex(oldval), " newval=", hex(newval), "\n")
999 throw("casgstatus: bad incoming values")
1003 acquireLockRank(lockRankGscan)
1004 releaseLockRank(lockRankGscan)
1006 // See https://golang.org/cl/21503 for justification of the yield delay.
1007 const yieldDelay = 5 * 1000
1010 // loop if gp->atomicstatus is in a scan state giving
1011 // GC time to finish and change the state to oldval.
1012 for i := 0; !gp.atomicstatus.CompareAndSwap(oldval, newval); i++ {
1013 if oldval == _Gwaiting && gp.atomicstatus.Load() == _Grunnable {
1014 throw("casgstatus: waiting for Gwaiting but is Grunnable")
1017 nextYield = nanotime() + yieldDelay
1019 if nanotime() < nextYield {
1020 for x := 0; x < 10 && gp.atomicstatus.Load() != oldval; x++ {
1025 nextYield = nanotime() + yieldDelay/2
1029 if oldval == _Grunning {
1030 // Track every gTrackingPeriod time a goroutine transitions out of running.
1031 if casgstatusAlwaysTrack || gp.trackingSeq%gTrackingPeriod == 0 {
1040 // Handle various kinds of tracking.
1043 // - Time spent in runnable.
1044 // - Time spent blocked on a sync.Mutex or sync.RWMutex.
1047 // We transitioned out of runnable, so measure how much
1048 // time we spent in this state and add it to
1051 gp.runnableTime += now - gp.trackingStamp
1052 gp.trackingStamp = 0
1054 if !gp.waitreason.isMutexWait() {
1055 // Not blocking on a lock.
1058 // Blocking on a lock, measure it. Note that because we're
1059 // sampling, we have to multiply by our sampling period to get
1060 // a more representative estimate of the absolute value.
1061 // gTrackingPeriod also represents an accurate sampling period
1062 // because we can only enter this state from _Grunning.
1064 sched.totalMutexWaitTime.Add((now - gp.trackingStamp) * gTrackingPeriod)
1065 gp.trackingStamp = 0
1069 if !gp.waitreason.isMutexWait() {
1070 // Not blocking on a lock.
1073 // Blocking on a lock. Write down the timestamp.
1075 gp.trackingStamp = now
1077 // We just transitioned into runnable, so record what
1078 // time that happened.
1080 gp.trackingStamp = now
1082 // We're transitioning into running, so turn off
1083 // tracking and record how much time we spent in
1086 sched.timeToRun.record(gp.runnableTime)
1091 // casGToWaiting transitions gp from old to _Gwaiting, and sets the wait reason.
1093 // Use this over casgstatus when possible to ensure that a waitreason is set.
1094 func casGToWaiting(gp *g, old uint32, reason waitReason) {
1095 // Set the wait reason before calling casgstatus, because casgstatus will use it.
1096 gp.waitreason = reason
1097 casgstatus(gp, old, _Gwaiting)
1100 // casgstatus(gp, oldstatus, Gcopystack), assuming oldstatus is Gwaiting or Grunnable.
1101 // Returns old status. Cannot call casgstatus directly, because we are racing with an
1102 // async wakeup that might come in from netpoll. If we see Gwaiting from the readgstatus,
1103 // it might have become Grunnable by the time we get to the cas. If we called casgstatus,
1104 // it would loop waiting for the status to go back to Gwaiting, which it never will.
1107 func casgcopystack(gp *g) uint32 {
1109 oldstatus := readgstatus(gp) &^ _Gscan
1110 if oldstatus != _Gwaiting && oldstatus != _Grunnable {
1111 throw("copystack: bad status, not Gwaiting or Grunnable")
1113 if gp.atomicstatus.CompareAndSwap(oldstatus, _Gcopystack) {
1119 // casGToPreemptScan transitions gp from _Grunning to _Gscan|_Gpreempted.
1121 // TODO(austin): This is the only status operation that both changes
1122 // the status and locks the _Gscan bit. Rethink this.
1123 func casGToPreemptScan(gp *g, old, new uint32) {
1124 if old != _Grunning || new != _Gscan|_Gpreempted {
1125 throw("bad g transition")
1127 acquireLockRank(lockRankGscan)
1128 for !gp.atomicstatus.CompareAndSwap(_Grunning, _Gscan|_Gpreempted) {
1132 // casGFromPreempted attempts to transition gp from _Gpreempted to
1133 // _Gwaiting. If successful, the caller is responsible for
1134 // re-scheduling gp.
1135 func casGFromPreempted(gp *g, old, new uint32) bool {
1136 if old != _Gpreempted || new != _Gwaiting {
1137 throw("bad g transition")
1139 gp.waitreason = waitReasonPreempted
1140 return gp.atomicstatus.CompareAndSwap(_Gpreempted, _Gwaiting)
1143 // stopTheWorld stops all P's from executing goroutines, interrupting
1144 // all goroutines at GC safe points and records reason as the reason
1145 // for the stop. On return, only the current goroutine's P is running.
1146 // stopTheWorld must not be called from a system stack and the caller
1147 // must not hold worldsema. The caller must call startTheWorld when
1148 // other P's should resume execution.
1150 // stopTheWorld is safe for multiple goroutines to call at the
1151 // same time. Each will execute its own stop, and the stops will
1154 // This is also used by routines that do stack dumps. If the system is
1155 // in panic or being exited, this may not reliably stop all
1157 func stopTheWorld(reason string) {
1158 semacquire(&worldsema)
1160 gp.m.preemptoff = reason
1161 systemstack(func() {
1162 // Mark the goroutine which called stopTheWorld preemptible so its
1163 // stack may be scanned.
1164 // This lets a mark worker scan us while we try to stop the world
1165 // since otherwise we could get in a mutual preemption deadlock.
1166 // We must not modify anything on the G stack because a stack shrink
1167 // may occur. A stack shrink is otherwise OK though because in order
1168 // to return from this function (and to leave the system stack) we
1169 // must have preempted all goroutines, including any attempting
1170 // to scan our stack, in which case, any stack shrinking will
1171 // have already completed by the time we exit.
1172 // Don't provide a wait reason because we're still executing.
1173 casGToWaiting(gp, _Grunning, waitReasonStoppingTheWorld)
1174 stopTheWorldWithSema()
1175 casgstatus(gp, _Gwaiting, _Grunning)
1179 // startTheWorld undoes the effects of stopTheWorld.
1180 func startTheWorld() {
1181 systemstack(func() { startTheWorldWithSema(false) })
1183 // worldsema must be held over startTheWorldWithSema to ensure
1184 // gomaxprocs cannot change while worldsema is held.
1186 // Release worldsema with direct handoff to the next waiter, but
1187 // acquirem so that semrelease1 doesn't try to yield our time.
1189 // Otherwise if e.g. ReadMemStats is being called in a loop,
1190 // it might stomp on other attempts to stop the world, such as
1191 // for starting or ending GC. The operation this blocks is
1192 // so heavy-weight that we should just try to be as fair as
1195 // We don't want to just allow us to get preempted between now
1196 // and releasing the semaphore because then we keep everyone
1197 // (including, for example, GCs) waiting longer.
1200 semrelease1(&worldsema, true, 0)
1204 // stopTheWorldGC has the same effect as stopTheWorld, but blocks
1205 // until the GC is not running. It also blocks a GC from starting
1206 // until startTheWorldGC is called.
1207 func stopTheWorldGC(reason string) {
1209 stopTheWorld(reason)
1212 // startTheWorldGC undoes the effects of stopTheWorldGC.
1213 func startTheWorldGC() {
1218 // Holding worldsema grants an M the right to try to stop the world.
1219 var worldsema uint32 = 1
1221 // Holding gcsema grants the M the right to block a GC, and blocks
1222 // until the current GC is done. In particular, it prevents gomaxprocs
1223 // from changing concurrently.
1225 // TODO(mknyszek): Once gomaxprocs and the execution tracer can handle
1226 // being changed/enabled during a GC, remove this.
1227 var gcsema uint32 = 1
1229 // stopTheWorldWithSema is the core implementation of stopTheWorld.
1230 // The caller is responsible for acquiring worldsema and disabling
1231 // preemption first and then should stopTheWorldWithSema on the system
1234 // semacquire(&worldsema, 0)
1235 // m.preemptoff = "reason"
1236 // systemstack(stopTheWorldWithSema)
1238 // When finished, the caller must either call startTheWorld or undo
1239 // these three operations separately:
1241 // m.preemptoff = ""
1242 // systemstack(startTheWorldWithSema)
1243 // semrelease(&worldsema)
1245 // It is allowed to acquire worldsema once and then execute multiple
1246 // startTheWorldWithSema/stopTheWorldWithSema pairs.
1247 // Other P's are able to execute between successive calls to
1248 // startTheWorldWithSema and stopTheWorldWithSema.
1249 // Holding worldsema causes any other goroutines invoking
1250 // stopTheWorld to block.
1251 func stopTheWorldWithSema() {
1254 // If we hold a lock, then we won't be able to stop another M
1255 // that is blocked trying to acquire the lock.
1257 throw("stopTheWorld: holding locks")
1261 sched.stopwait = gomaxprocs
1262 sched.gcwaiting.Store(true)
1265 gp.m.p.ptr().status = _Pgcstop // Pgcstop is only diagnostic.
1267 // try to retake all P's in Psyscall status
1268 for _, pp := range allp {
1270 if s == _Psyscall && atomic.Cas(&pp.status, s, _Pgcstop) {
1282 pp, _ := pidleget(now)
1286 pp.status = _Pgcstop
1289 wait := sched.stopwait > 0
1292 // wait for remaining P's to stop voluntarily
1295 // wait for 100us, then try to re-preempt in case of any races
1296 if notetsleep(&sched.stopnote, 100*1000) {
1297 noteclear(&sched.stopnote)
1306 if sched.stopwait != 0 {
1307 bad = "stopTheWorld: not stopped (stopwait != 0)"
1309 for _, pp := range allp {
1310 if pp.status != _Pgcstop {
1311 bad = "stopTheWorld: not stopped (status != _Pgcstop)"
1315 if freezing.Load() {
1316 // Some other thread is panicking. This can cause the
1317 // sanity checks above to fail if the panic happens in
1318 // the signal handler on a stopped thread. Either way,
1319 // we should halt this thread.
1330 func startTheWorldWithSema(emitTraceEvent bool) int64 {
1331 assertWorldStopped()
1333 mp := acquirem() // disable preemption because it can be holding p in a local var
1334 if netpollinited() {
1335 list := netpoll(0) // non-blocking
1345 p1 := procresize(procs)
1346 sched.gcwaiting.Store(false)
1347 if sched.sysmonwait.Load() {
1348 sched.sysmonwait.Store(false)
1349 notewakeup(&sched.sysmonnote)
1362 throw("startTheWorld: inconsistent mp->nextp")
1365 notewakeup(&mp.park)
1367 // Start M to run P. Do not start another M below.
1372 // Capture start-the-world time before doing clean-up tasks.
1373 startTime := nanotime()
1378 // Wakeup an additional proc in case we have excessive runnable goroutines
1379 // in local queues or in the global queue. If we don't, the proc will park itself.
1380 // If we have lots of excessive work, resetspinning will unpark additional procs as necessary.
1388 // usesLibcall indicates whether this runtime performs system calls
1390 func usesLibcall() bool {
1392 case "aix", "darwin", "illumos", "ios", "solaris", "windows":
1395 return GOARCH == "386" || GOARCH == "amd64" || GOARCH == "arm" || GOARCH == "arm64"
1400 // mStackIsSystemAllocated indicates whether this runtime starts on a
1401 // system-allocated stack.
1402 func mStackIsSystemAllocated() bool {
1404 case "aix", "darwin", "plan9", "illumos", "ios", "solaris", "windows":
1408 case "386", "amd64", "arm", "arm64":
1415 // mstart is the entry-point for new Ms.
1416 // It is written in assembly, uses ABI0, is marked TOPFRAME, and calls mstart0.
1419 // mstart0 is the Go entry-point for new Ms.
1420 // This must not split the stack because we may not even have stack
1421 // bounds set up yet.
1423 // May run during STW (because it doesn't have a P yet), so write
1424 // barriers are not allowed.
1427 //go:nowritebarrierrec
1431 osStack := gp.stack.lo == 0
1433 // Initialize stack bounds from system stack.
1434 // Cgo may have left stack size in stack.hi.
1435 // minit may update the stack bounds.
1437 // Note: these bounds may not be very accurate.
1438 // We set hi to &size, but there are things above
1439 // it. The 1024 is supposed to compensate this,
1440 // but is somewhat arbitrary.
1443 size = 8192 * sys.StackGuardMultiplier
1445 gp.stack.hi = uintptr(noescape(unsafe.Pointer(&size)))
1446 gp.stack.lo = gp.stack.hi - size + 1024
1448 // Initialize stack guard so that we can start calling regular
1450 gp.stackguard0 = gp.stack.lo + stackGuard
1451 // This is the g0, so we can also call go:systemstack
1452 // functions, which check stackguard1.
1453 gp.stackguard1 = gp.stackguard0
1456 // Exit this thread.
1457 if mStackIsSystemAllocated() {
1458 // Windows, Solaris, illumos, Darwin, AIX and Plan 9 always system-allocate
1459 // the stack, but put it in gp.stack before mstart,
1460 // so the logic above hasn't set osStack yet.
1466 // The go:noinline is to guarantee the getcallerpc/getcallersp below are safe,
1467 // so that we can set up g0.sched to return to the call of mstart1 above.
1474 throw("bad runtime·mstart")
1477 // Set up m.g0.sched as a label returning to just
1478 // after the mstart1 call in mstart0 above, for use by goexit0 and mcall.
1479 // We're never coming back to mstart1 after we call schedule,
1480 // so other calls can reuse the current frame.
1481 // And goexit0 does a gogo that needs to return from mstart1
1482 // and let mstart0 exit the thread.
1483 gp.sched.g = guintptr(unsafe.Pointer(gp))
1484 gp.sched.pc = getcallerpc()
1485 gp.sched.sp = getcallersp()
1490 // Install signal handlers; after minit so that minit can
1491 // prepare the thread to be able to handle the signals.
1496 if fn := gp.m.mstartfn; fn != nil {
1501 acquirep(gp.m.nextp.ptr())
1507 // mstartm0 implements part of mstart1 that only runs on the m0.
1509 // Write barriers are allowed here because we know the GC can't be
1510 // running yet, so they'll be no-ops.
1512 //go:yeswritebarrierrec
1514 // Create an extra M for callbacks on threads not created by Go.
1515 // An extra M is also needed on Windows for callbacks created by
1516 // syscall.NewCallback. See issue #6751 for details.
1517 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
1524 // mPark causes a thread to park itself, returning once woken.
1529 notesleep(&gp.m.park)
1530 noteclear(&gp.m.park)
1533 // mexit tears down and exits the current thread.
1535 // Don't call this directly to exit the thread, since it must run at
1536 // the top of the thread stack. Instead, use gogo(&gp.m.g0.sched) to
1537 // unwind the stack to the point that exits the thread.
1539 // It is entered with m.p != nil, so write barriers are allowed. It
1540 // will release the P before exiting.
1542 //go:yeswritebarrierrec
1543 func mexit(osStack bool) {
1547 // This is the main thread. Just wedge it.
1549 // On Linux, exiting the main thread puts the process
1550 // into a non-waitable zombie state. On Plan 9,
1551 // exiting the main thread unblocks wait even though
1552 // other threads are still running. On Solaris we can
1553 // neither exitThread nor return from mstart. Other
1554 // bad things probably happen on other platforms.
1556 // We could try to clean up this M more before wedging
1557 // it, but that complicates signal handling.
1558 handoffp(releasep())
1564 throw("locked m0 woke up")
1570 // Free the gsignal stack.
1571 if mp.gsignal != nil {
1572 stackfree(mp.gsignal.stack)
1573 // On some platforms, when calling into VDSO (e.g. nanotime)
1574 // we store our g on the gsignal stack, if there is one.
1575 // Now the stack is freed, unlink it from the m, so we
1576 // won't write to it when calling VDSO code.
1580 // Remove m from allm.
1582 for pprev := &allm; *pprev != nil; pprev = &(*pprev).alllink {
1588 throw("m not found in allm")
1590 // Delay reaping m until it's done with the stack.
1592 // Put mp on the free list, though it will not be reaped while freeWait
1593 // is freeMWait. mp is no longer reachable via allm, so even if it is
1594 // on an OS stack, we must keep a reference to mp alive so that the GC
1595 // doesn't free mp while we are still using it.
1597 // Note that the free list must not be linked through alllink because
1598 // some functions walk allm without locking, so may be using alllink.
1599 mp.freeWait.Store(freeMWait)
1600 mp.freelink = sched.freem
1604 atomic.Xadd64(&ncgocall, int64(mp.ncgocall))
1607 handoffp(releasep())
1608 // After this point we must not have write barriers.
1610 // Invoke the deadlock detector. This must happen after
1611 // handoffp because it may have started a new M to take our
1618 if GOOS == "darwin" || GOOS == "ios" {
1619 // Make sure pendingPreemptSignals is correct when an M exits.
1621 if mp.signalPending.Load() != 0 {
1622 pendingPreemptSignals.Add(-1)
1626 // Destroy all allocated resources. After this is called, we may no
1627 // longer take any locks.
1631 // No more uses of mp, so it is safe to drop the reference.
1632 mp.freeWait.Store(freeMRef)
1634 // Return from mstart and let the system thread
1635 // library free the g0 stack and terminate the thread.
1639 // mstart is the thread's entry point, so there's nothing to
1640 // return to. Exit the thread directly. exitThread will clear
1641 // m.freeWait when it's done with the stack and the m can be
1643 exitThread(&mp.freeWait)
1646 // forEachP calls fn(p) for every P p when p reaches a GC safe point.
1647 // If a P is currently executing code, this will bring the P to a GC
1648 // safe point and execute fn on that P. If the P is not executing code
1649 // (it is idle or in a syscall), this will call fn(p) directly while
1650 // preventing the P from exiting its state. This does not ensure that
1651 // fn will run on every CPU executing Go code, but it acts as a global
1652 // memory barrier. GC uses this as a "ragged barrier."
1654 // The caller must hold worldsema.
1657 func forEachP(fn func(*p)) {
1659 pp := getg().m.p.ptr()
1662 if sched.safePointWait != 0 {
1663 throw("forEachP: sched.safePointWait != 0")
1665 sched.safePointWait = gomaxprocs - 1
1666 sched.safePointFn = fn
1668 // Ask all Ps to run the safe point function.
1669 for _, p2 := range allp {
1671 atomic.Store(&p2.runSafePointFn, 1)
1676 // Any P entering _Pidle or _Psyscall from now on will observe
1677 // p.runSafePointFn == 1 and will call runSafePointFn when
1678 // changing its status to _Pidle/_Psyscall.
1680 // Run safe point function for all idle Ps. sched.pidle will
1681 // not change because we hold sched.lock.
1682 for p := sched.pidle.ptr(); p != nil; p = p.link.ptr() {
1683 if atomic.Cas(&p.runSafePointFn, 1, 0) {
1685 sched.safePointWait--
1689 wait := sched.safePointWait > 0
1692 // Run fn for the current P.
1695 // Force Ps currently in _Psyscall into _Pidle and hand them
1696 // off to induce safe point function execution.
1697 for _, p2 := range allp {
1699 if s == _Psyscall && p2.runSafePointFn == 1 && atomic.Cas(&p2.status, s, _Pidle) {
1709 // Wait for remaining Ps to run fn.
1712 // Wait for 100us, then try to re-preempt in
1713 // case of any races.
1715 // Requires system stack.
1716 if notetsleep(&sched.safePointNote, 100*1000) {
1717 noteclear(&sched.safePointNote)
1723 if sched.safePointWait != 0 {
1724 throw("forEachP: not done")
1726 for _, p2 := range allp {
1727 if p2.runSafePointFn != 0 {
1728 throw("forEachP: P did not run fn")
1733 sched.safePointFn = nil
1738 // runSafePointFn runs the safe point function, if any, for this P.
1739 // This should be called like
1741 // if getg().m.p.runSafePointFn != 0 {
1745 // runSafePointFn must be checked on any transition in to _Pidle or
1746 // _Psyscall to avoid a race where forEachP sees that the P is running
1747 // just before the P goes into _Pidle/_Psyscall and neither forEachP
1748 // nor the P run the safe-point function.
1749 func runSafePointFn() {
1750 p := getg().m.p.ptr()
1751 // Resolve the race between forEachP running the safe-point
1752 // function on this P's behalf and this P running the
1753 // safe-point function directly.
1754 if !atomic.Cas(&p.runSafePointFn, 1, 0) {
1757 sched.safePointFn(p)
1759 sched.safePointWait--
1760 if sched.safePointWait == 0 {
1761 notewakeup(&sched.safePointNote)
1766 // When running with cgo, we call _cgo_thread_start
1767 // to start threads for us so that we can play nicely with
1769 var cgoThreadStart unsafe.Pointer
1771 type cgothreadstart struct {
1777 // Allocate a new m unassociated with any thread.
1778 // Can use p for allocation context if needed.
1779 // fn is recorded as the new m's m.mstartfn.
1780 // id is optional pre-allocated m ID. Omit by passing -1.
1782 // This function is allowed to have write barriers even if the caller
1783 // isn't because it borrows pp.
1785 //go:yeswritebarrierrec
1786 func allocm(pp *p, fn func(), id int64) *m {
1789 // The caller owns pp, but we may borrow (i.e., acquirep) it. We must
1790 // disable preemption to ensure it is not stolen, which would make the
1791 // caller lose ownership.
1796 acquirep(pp) // temporarily borrow p for mallocs in this function
1799 // Release the free M list. We need to do this somewhere and
1800 // this may free up a stack we can use.
1801 if sched.freem != nil {
1804 for freem := sched.freem; freem != nil; {
1805 wait := freem.freeWait.Load()
1806 if wait == freeMWait {
1807 next := freem.freelink
1808 freem.freelink = newList
1813 // Free the stack if needed. For freeMRef, there is
1814 // nothing to do except drop freem from the sched.freem
1816 if wait == freeMStack {
1817 // stackfree must be on the system stack, but allocm is
1818 // reachable off the system stack transitively from
1820 systemstack(func() {
1821 stackfree(freem.g0.stack)
1824 freem = freem.freelink
1826 sched.freem = newList
1834 // In case of cgo or Solaris or illumos or Darwin, pthread_create will make us a stack.
1835 // Windows and Plan 9 will layout sched stack on OS stack.
1836 if iscgo || mStackIsSystemAllocated() {
1839 mp.g0 = malg(8192 * sys.StackGuardMultiplier)
1843 if pp == gp.m.p.ptr() {
1848 allocmLock.runlock()
1852 // needm is called when a cgo callback happens on a
1853 // thread without an m (a thread not created by Go).
1854 // In this case, needm is expected to find an m to use
1855 // and return with m, g initialized correctly.
1856 // Since m and g are not set now (likely nil, but see below)
1857 // needm is limited in what routines it can call. In particular
1858 // it can only call nosplit functions (textflag 7) and cannot
1859 // do any scheduling that requires an m.
1861 // In order to avoid needing heavy lifting here, we adopt
1862 // the following strategy: there is a stack of available m's
1863 // that can be stolen. Using compare-and-swap
1864 // to pop from the stack has ABA races, so we simulate
1865 // a lock by doing an exchange (via Casuintptr) to steal the stack
1866 // head and replace the top pointer with MLOCKED (1).
1867 // This serves as a simple spin lock that we can use even
1868 // without an m. The thread that locks the stack in this way
1869 // unlocks the stack by storing a valid stack head pointer.
1871 // In order to make sure that there is always an m structure
1872 // available to be stolen, we maintain the invariant that there
1873 // is always one more than needed. At the beginning of the
1874 // program (if cgo is in use) the list is seeded with a single m.
1875 // If needm finds that it has taken the last m off the list, its job
1876 // is - once it has installed its own m so that it can do things like
1877 // allocate memory - to create a spare m and put it on the list.
1879 // Each of these extra m's also has a g0 and a curg that are
1880 // pressed into service as the scheduling stack and current
1881 // goroutine for the duration of the cgo callback.
1883 // When the callback is done with the m, it calls dropm to
1884 // put the m back on the list.
1888 if (iscgo || GOOS == "windows") && !cgoHasExtraM {
1889 // Can happen if C/C++ code calls Go from a global ctor.
1890 // Can also happen on Windows if a global ctor uses a
1891 // callback created by syscall.NewCallback. See issue #6751
1894 // Can not throw, because scheduler is not initialized yet.
1895 writeErrStr("fatal error: cgo callback before cgo call\n")
1899 // Save and block signals before getting an M.
1900 // The signal handler may call needm itself,
1901 // and we must avoid a deadlock. Also, once g is installed,
1902 // any incoming signals will try to execute,
1903 // but we won't have the sigaltstack settings and other data
1904 // set up appropriately until the end of minit, which will
1905 // unblock the signals. This is the same dance as when
1906 // starting a new m to run Go code via newosproc.
1911 // nilokay=false is safe here because of the invariant above,
1912 // that the extra list always contains or will soon contain
1914 mp, last := getExtraM(false)
1916 // Set needextram when we've just emptied the list,
1917 // so that the eventual call into cgocallbackg will
1918 // allocate a new m for the extra list. We delay the
1919 // allocation until then so that it can be done
1920 // after exitsyscall makes sure it is okay to be
1921 // running at all (that is, there's no garbage collection
1922 // running right now).
1923 mp.needextram = last
1925 // Store the original signal mask for use by minit.
1926 mp.sigmask = sigmask
1928 // Install TLS on some platforms (previously setg
1929 // would do this if necessary).
1932 // Install g (= m->g0) and set the stack bounds
1933 // to match the current stack. We don't actually know
1934 // how big the stack is, like we don't know how big any
1935 // scheduling stack is, but we assume there's at least 32 kB,
1936 // which is more than enough for us.
1939 gp.stack.hi = getcallersp() + 1024
1940 gp.stack.lo = getcallersp() - 32*1024
1941 gp.stackguard0 = gp.stack.lo + stackGuard
1943 // Initialize this thread to use the m.
1947 // mp.curg is now a real goroutine.
1948 casgstatus(mp.curg, _Gdead, _Gsyscall)
1952 // newextram allocates m's and puts them on the extra list.
1953 // It is called with a working local m, so that it can do things
1954 // like call schedlock and allocate.
1956 c := extraMWaiters.Swap(0)
1958 for i := uint32(0); i < c; i++ {
1961 } else if extraMLength.Load() == 0 {
1962 // Make sure there is at least one extra M.
1967 // oneNewExtraM allocates an m and puts it on the extra list.
1968 func oneNewExtraM() {
1969 // Create extra goroutine locked to extra m.
1970 // The goroutine is the context in which the cgo callback will run.
1971 // The sched.pc will never be returned to, but setting it to
1972 // goexit makes clear to the traceback routines where
1973 // the goroutine stack ends.
1974 mp := allocm(nil, nil, -1)
1976 gp.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum
1977 gp.sched.sp = gp.stack.hi
1978 gp.sched.sp -= 4 * goarch.PtrSize // extra space in case of reads slightly beyond frame
1980 gp.sched.g = guintptr(unsafe.Pointer(gp))
1981 gp.syscallpc = gp.sched.pc
1982 gp.syscallsp = gp.sched.sp
1983 gp.stktopsp = gp.sched.sp
1984 // malg returns status as _Gidle. Change to _Gdead before
1985 // adding to allg where GC can see it. We use _Gdead to hide
1986 // this from tracebacks and stack scans since it isn't a
1987 // "real" goroutine until needm grabs it.
1988 casgstatus(gp, _Gidle, _Gdead)
1995 gp.goid = sched.goidgen.Add(1)
1996 gp.sysblocktraced = true
1998 gp.racectx = racegostart(abi.FuncPCABIInternal(newextram) + sys.PCQuantum)
2001 // Trigger two trace events for the locked g in the extra m,
2002 // since the next event of the g will be traceEvGoSysExit in exitsyscall,
2003 // while calling from C thread to Go.
2004 traceGoCreate(gp, 0) // no start pc
2006 traceEvent(traceEvGoInSyscall, -1, gp.goid)
2008 // put on allg for garbage collector
2011 // gp is now on the allg list, but we don't want it to be
2012 // counted by gcount. It would be more "proper" to increment
2013 // sched.ngfree, but that requires locking. Incrementing ngsys
2014 // has the same effect.
2017 // Add m to the extra list.
2021 // dropm is called when a cgo callback has called needm but is now
2022 // done with the callback and returning back into the non-Go thread.
2023 // It puts the current m back onto the extra list.
2025 // The main expense here is the call to signalstack to release the
2026 // m's signal stack, and then the call to needm on the next callback
2027 // from this thread. It is tempting to try to save the m for next time,
2028 // which would eliminate both these costs, but there might not be
2029 // a next time: the current thread (which Go does not control) might exit.
2030 // If we saved the m for that thread, there would be an m leak each time
2031 // such a thread exited. Instead, we acquire and release an m on each
2032 // call. These should typically not be scheduling operations, just a few
2033 // atomics, so the cost should be small.
2035 // TODO(rsc): An alternative would be to allocate a dummy pthread per-thread
2036 // variable using pthread_key_create. Unlike the pthread keys we already use
2037 // on OS X, this dummy key would never be read by Go code. It would exist
2038 // only so that we could register at thread-exit-time destructor.
2039 // That destructor would put the m back onto the extra list.
2040 // This is purely a performance optimization. The current version,
2041 // in which dropm happens on each cgo call, is still correct too.
2042 // We may have to keep the current version on systems with cgo
2043 // but without pthreads, like Windows.
2045 // Clear m and g, and return m to the extra list.
2046 // After the call to setg we can only call nosplit functions
2047 // with no pointer manipulation.
2050 // Return mp.curg to dead state.
2051 casgstatus(mp.curg, _Gsyscall, _Gdead)
2052 mp.curg.preemptStop = false
2055 // Block signals before unminit.
2056 // Unminit unregisters the signal handling stack (but needs g on some systems).
2057 // Setg(nil) clears g, which is the signal handler's cue not to run Go handlers.
2058 // It's important not to try to handle a signal between those two steps.
2059 sigmask := mp.sigmask
2067 msigrestore(sigmask)
2070 // A helper function for EnsureDropM.
2071 func getm() uintptr {
2072 return uintptr(unsafe.Pointer(getg().m))
2076 // Locking linked list of extra M's, via mp.schedlink. Must be accessed
2077 // only via lockextra/unlockextra.
2079 // Can't be atomic.Pointer[m] because we use an invalid pointer as a
2080 // "locked" sentinel value. M's on this list remain visible to the GC
2081 // because their mp.curg is on allgs.
2082 extraM atomic.Uintptr
2083 // Number of M's in the extraM list.
2084 extraMLength atomic.Uint32
2085 // Number of waiters in lockextra.
2086 extraMWaiters atomic.Uint32
2089 // lockextra locks the extra list and returns the list head.
2090 // The caller must unlock the list by storing a new list head
2091 // to extram. If nilokay is true, then lockextra will
2092 // return a nil list head if that's what it finds. If nilokay is false,
2093 // lockextra will keep waiting until the list head is no longer nil.
2096 func lockextra(nilokay bool) *m {
2101 old := extraM.Load()
2106 if old == 0 && !nilokay {
2108 // Add 1 to the number of threads
2109 // waiting for an M.
2110 // This is cleared by newextram.
2111 extraMWaiters.Add(1)
2117 if extraM.CompareAndSwap(old, locked) {
2118 return (*m)(unsafe.Pointer(old))
2126 func unlockextra(mp *m, delta int32) {
2127 extraMLength.Add(delta)
2128 extraM.Store(uintptr(unsafe.Pointer(mp)))
2132 // Return an M from the extra M list. Returns last == true if the list becomes
2133 // empty because of this call.
2136 func getExtraM(nilokay bool) (mp *m, last bool) {
2137 mp = lockextra(nilokay)
2142 unlockextra(mp.schedlink.ptr(), -1)
2143 return mp, mp.schedlink.ptr() == nil
2146 // Put an extra M on the list.
2149 func putExtraM(mp *m) {
2150 mnext := lockextra(true)
2151 mp.schedlink.set(mnext)
2156 // allocmLock is locked for read when creating new Ms in allocm and their
2157 // addition to allm. Thus acquiring this lock for write blocks the
2158 // creation of new Ms.
2161 // execLock serializes exec and clone to avoid bugs or unspecified
2162 // behaviour around exec'ing while creating/destroying threads. See
2167 // These errors are reported (via writeErrStr) by some OS-specific
2168 // versions of newosproc and newosproc0.
2170 failthreadcreate = "runtime: failed to create new OS thread\n"
2171 failallocatestack = "runtime: failed to allocate stack for the new OS thread\n"
2174 // newmHandoff contains a list of m structures that need new OS threads.
2175 // This is used by newm in situations where newm itself can't safely
2176 // start an OS thread.
2177 var newmHandoff struct {
2180 // newm points to a list of M structures that need new OS
2181 // threads. The list is linked through m.schedlink.
2184 // waiting indicates that wake needs to be notified when an m
2185 // is put on the list.
2189 // haveTemplateThread indicates that the templateThread has
2190 // been started. This is not protected by lock. Use cas to set
2192 haveTemplateThread uint32
2195 // Create a new m. It will start off with a call to fn, or else the scheduler.
2196 // fn needs to be static and not a heap allocated closure.
2197 // May run with m.p==nil, so write barriers are not allowed.
2199 // id is optional pre-allocated m ID. Omit by passing -1.
2201 //go:nowritebarrierrec
2202 func newm(fn func(), pp *p, id int64) {
2203 // allocm adds a new M to allm, but they do not start until created by
2204 // the OS in newm1 or the template thread.
2206 // doAllThreadsSyscall requires that every M in allm will eventually
2207 // start and be signal-able, even with a STW.
2209 // Disable preemption here until we start the thread to ensure that
2210 // newm is not preempted between allocm and starting the new thread,
2211 // ensuring that anything added to allm is guaranteed to eventually
2215 mp := allocm(pp, fn, id)
2217 mp.sigmask = initSigmask
2218 if gp := getg(); gp != nil && gp.m != nil && (gp.m.lockedExt != 0 || gp.m.incgo) && GOOS != "plan9" {
2219 // We're on a locked M or a thread that may have been
2220 // started by C. The kernel state of this thread may
2221 // be strange (the user may have locked it for that
2222 // purpose). We don't want to clone that into another
2223 // thread. Instead, ask a known-good thread to create
2224 // the thread for us.
2226 // This is disabled on Plan 9. See golang.org/issue/22227.
2228 // TODO: This may be unnecessary on Windows, which
2229 // doesn't model thread creation off fork.
2230 lock(&newmHandoff.lock)
2231 if newmHandoff.haveTemplateThread == 0 {
2232 throw("on a locked thread with no template thread")
2234 mp.schedlink = newmHandoff.newm
2235 newmHandoff.newm.set(mp)
2236 if newmHandoff.waiting {
2237 newmHandoff.waiting = false
2238 notewakeup(&newmHandoff.wake)
2240 unlock(&newmHandoff.lock)
2241 // The M has not started yet, but the template thread does not
2242 // participate in STW, so it will always process queued Ms and
2243 // it is safe to releasem.
2253 var ts cgothreadstart
2254 if _cgo_thread_start == nil {
2255 throw("_cgo_thread_start missing")
2258 ts.tls = (*uint64)(unsafe.Pointer(&mp.tls[0]))
2259 ts.fn = unsafe.Pointer(abi.FuncPCABI0(mstart))
2261 msanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2264 asanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
2266 execLock.rlock() // Prevent process clone.
2267 asmcgocall(_cgo_thread_start, unsafe.Pointer(&ts))
2271 execLock.rlock() // Prevent process clone.
2276 // startTemplateThread starts the template thread if it is not already
2279 // The calling thread must itself be in a known-good state.
2280 func startTemplateThread() {
2281 if GOARCH == "wasm" { // no threads on wasm yet
2285 // Disable preemption to guarantee that the template thread will be
2286 // created before a park once haveTemplateThread is set.
2288 if !atomic.Cas(&newmHandoff.haveTemplateThread, 0, 1) {
2292 newm(templateThread, nil, -1)
2296 // templateThread is a thread in a known-good state that exists solely
2297 // to start new threads in known-good states when the calling thread
2298 // may not be in a good state.
2300 // Many programs never need this, so templateThread is started lazily
2301 // when we first enter a state that might lead to running on a thread
2302 // in an unknown state.
2304 // templateThread runs on an M without a P, so it must not have write
2307 //go:nowritebarrierrec
2308 func templateThread() {
2315 lock(&newmHandoff.lock)
2316 for newmHandoff.newm != 0 {
2317 newm := newmHandoff.newm.ptr()
2318 newmHandoff.newm = 0
2319 unlock(&newmHandoff.lock)
2321 next := newm.schedlink.ptr()
2326 lock(&newmHandoff.lock)
2328 newmHandoff.waiting = true
2329 noteclear(&newmHandoff.wake)
2330 unlock(&newmHandoff.lock)
2331 notesleep(&newmHandoff.wake)
2335 // Stops execution of the current m until new work is available.
2336 // Returns with acquired P.
2340 if gp.m.locks != 0 {
2341 throw("stopm holding locks")
2344 throw("stopm holding p")
2347 throw("stopm spinning")
2354 acquirep(gp.m.nextp.ptr())
2359 // startm's caller incremented nmspinning. Set the new M's spinning.
2360 getg().m.spinning = true
2363 // Schedules some M to run the p (creates an M if necessary).
2364 // If p==nil, tries to get an idle P, if no idle P's does nothing.
2365 // May run with m.p==nil, so write barriers are not allowed.
2366 // If spinning is set, the caller has incremented nmspinning and must provide a
2367 // P. startm will set m.spinning in the newly started M.
2369 // Callers passing a non-nil P must call from a non-preemptible context. See
2370 // comment on acquirem below.
2372 // Argument lockheld indicates whether the caller already acquired the
2373 // scheduler lock. Callers holding the lock when making the call must pass
2374 // true. The lock might be temporarily dropped, but will be reacquired before
2377 // Must not have write barriers because this may be called without a P.
2379 //go:nowritebarrierrec
2380 func startm(pp *p, spinning, lockheld bool) {
2381 // Disable preemption.
2383 // Every owned P must have an owner that will eventually stop it in the
2384 // event of a GC stop request. startm takes transient ownership of a P
2385 // (either from argument or pidleget below) and transfers ownership to
2386 // a started M, which will be responsible for performing the stop.
2388 // Preemption must be disabled during this transient ownership,
2389 // otherwise the P this is running on may enter GC stop while still
2390 // holding the transient P, leaving that P in limbo and deadlocking the
2393 // Callers passing a non-nil P must already be in non-preemptible
2394 // context, otherwise such preemption could occur on function entry to
2395 // startm. Callers passing a nil P may be preemptible, so we must
2396 // disable preemption before acquiring a P from pidleget below.
2403 // TODO(prattmic): All remaining calls to this function
2404 // with _p_ == nil could be cleaned up to find a P
2405 // before calling startm.
2406 throw("startm: P required for spinning=true")
2419 // No M is available, we must drop sched.lock and call newm.
2420 // However, we already own a P to assign to the M.
2422 // Once sched.lock is released, another G (e.g., in a syscall),
2423 // could find no idle P while checkdead finds a runnable G but
2424 // no running M's because this new M hasn't started yet, thus
2425 // throwing in an apparent deadlock.
2426 // This apparent deadlock is possible when startm is called
2427 // from sysmon, which doesn't count as a running M.
2429 // Avoid this situation by pre-allocating the ID for the new M,
2430 // thus marking it as 'running' before we drop sched.lock. This
2431 // new M will eventually run the scheduler to execute any
2438 // The caller incremented nmspinning, so set m.spinning in the new M.
2446 // Ownership transfer of pp committed by start in newm.
2447 // Preemption is now safe.
2455 throw("startm: m is spinning")
2458 throw("startm: m has p")
2460 if spinning && !runqempty(pp) {
2461 throw("startm: p has runnable gs")
2463 // The caller incremented nmspinning, so set m.spinning in the new M.
2464 nmp.spinning = spinning
2466 notewakeup(&nmp.park)
2467 // Ownership transfer of pp committed by wakeup. Preemption is now
2472 // Hands off P from syscall or locked M.
2473 // Always runs without a P, so write barriers are not allowed.
2475 //go:nowritebarrierrec
2476 func handoffp(pp *p) {
2477 // handoffp must start an M in any situation where
2478 // findrunnable would return a G to run on pp.
2480 // if it has local work, start it straight away
2481 if !runqempty(pp) || sched.runqsize != 0 {
2482 startm(pp, false, false)
2485 // if there's trace work to do, start it straight away
2486 if (trace.enabled || trace.shutdown) && traceReaderAvailable() != nil {
2487 startm(pp, false, false)
2490 // if it has GC work, start it straight away
2491 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) {
2492 startm(pp, false, false)
2495 // no local work, check that there are no spinning/idle M's,
2496 // otherwise our help is not required
2497 if sched.nmspinning.Load()+sched.npidle.Load() == 0 && sched.nmspinning.CompareAndSwap(0, 1) { // TODO: fast atomic
2498 sched.needspinning.Store(0)
2499 startm(pp, true, false)
2503 if sched.gcwaiting.Load() {
2504 pp.status = _Pgcstop
2506 if sched.stopwait == 0 {
2507 notewakeup(&sched.stopnote)
2512 if pp.runSafePointFn != 0 && atomic.Cas(&pp.runSafePointFn, 1, 0) {
2513 sched.safePointFn(pp)
2514 sched.safePointWait--
2515 if sched.safePointWait == 0 {
2516 notewakeup(&sched.safePointNote)
2519 if sched.runqsize != 0 {
2521 startm(pp, false, false)
2524 // If this is the last running P and nobody is polling network,
2525 // need to wakeup another M to poll network.
2526 if sched.npidle.Load() == gomaxprocs-1 && sched.lastpoll.Load() != 0 {
2528 startm(pp, false, false)
2532 // The scheduler lock cannot be held when calling wakeNetPoller below
2533 // because wakeNetPoller may call wakep which may call startm.
2534 when := nobarrierWakeTime(pp)
2543 // Tries to add one more P to execute G's.
2544 // Called when a G is made runnable (newproc, ready).
2545 // Must be called with a P.
2547 // Be conservative about spinning threads, only start one if none exist
2549 if sched.nmspinning.Load() != 0 || !sched.nmspinning.CompareAndSwap(0, 1) {
2553 // Disable preemption until ownership of pp transfers to the next M in
2554 // startm. Otherwise preemption here would leave pp stuck waiting to
2557 // See preemption comment on acquirem in startm for more details.
2562 pp, _ = pidlegetSpinning(0)
2564 if sched.nmspinning.Add(-1) < 0 {
2565 throw("wakep: negative nmspinning")
2571 // Since we always have a P, the race in the "No M is available"
2572 // comment in startm doesn't apply during the small window between the
2573 // unlock here and lock in startm. A checkdead in between will always
2574 // see at least one running M (ours).
2577 startm(pp, true, false)
2582 // Stops execution of the current m that is locked to a g until the g is runnable again.
2583 // Returns with acquired P.
2584 func stoplockedm() {
2587 if gp.m.lockedg == 0 || gp.m.lockedg.ptr().lockedm.ptr() != gp.m {
2588 throw("stoplockedm: inconsistent locking")
2591 // Schedule another M to run this p.
2596 // Wait until another thread schedules lockedg again.
2598 status := readgstatus(gp.m.lockedg.ptr())
2599 if status&^_Gscan != _Grunnable {
2600 print("runtime:stoplockedm: lockedg (atomicstatus=", status, ") is not Grunnable or Gscanrunnable\n")
2601 dumpgstatus(gp.m.lockedg.ptr())
2602 throw("stoplockedm: not runnable")
2604 acquirep(gp.m.nextp.ptr())
2608 // Schedules the locked m to run the locked gp.
2609 // May run during STW, so write barriers are not allowed.
2611 //go:nowritebarrierrec
2612 func startlockedm(gp *g) {
2613 mp := gp.lockedm.ptr()
2615 throw("startlockedm: locked to me")
2618 throw("startlockedm: m has p")
2620 // directly handoff current P to the locked m
2624 notewakeup(&mp.park)
2628 // Stops the current m for stopTheWorld.
2629 // Returns when the world is restarted.
2633 if !sched.gcwaiting.Load() {
2634 throw("gcstopm: not waiting for gc")
2637 gp.m.spinning = false
2638 // OK to just drop nmspinning here,
2639 // startTheWorld will unpark threads as necessary.
2640 if sched.nmspinning.Add(-1) < 0 {
2641 throw("gcstopm: negative nmspinning")
2646 pp.status = _Pgcstop
2648 if sched.stopwait == 0 {
2649 notewakeup(&sched.stopnote)
2655 // Schedules gp to run on the current M.
2656 // If inheritTime is true, gp inherits the remaining time in the
2657 // current time slice. Otherwise, it starts a new time slice.
2660 // Write barriers are allowed because this is called immediately after
2661 // acquiring a P in several places.
2663 //go:yeswritebarrierrec
2664 func execute(gp *g, inheritTime bool) {
2667 if goroutineProfile.active {
2668 // Make sure that gp has had its stack written out to the goroutine
2669 // profile, exactly as it was when the goroutine profiler first stopped
2671 tryRecordGoroutineProfile(gp, osyield)
2674 // Assign gp.m before entering _Grunning so running Gs have an
2678 casgstatus(gp, _Grunnable, _Grunning)
2681 gp.stackguard0 = gp.stack.lo + stackGuard
2683 mp.p.ptr().schedtick++
2686 // Check whether the profiler needs to be turned on or off.
2687 hz := sched.profilehz
2688 if mp.profilehz != hz {
2689 setThreadCPUProfiler(hz)
2693 // GoSysExit has to happen when we have a P, but before GoStart.
2694 // So we emit it here.
2695 if gp.syscallsp != 0 && gp.sysblocktraced {
2696 traceGoSysExit(gp.sysexitticks)
2704 // Finds a runnable goroutine to execute.
2705 // Tries to steal from other P's, get g from local or global queue, poll network.
2706 // tryWakeP indicates that the returned goroutine is not normal (GC worker, trace
2707 // reader) so the caller should try to wake a P.
2708 func findRunnable() (gp *g, inheritTime, tryWakeP bool) {
2711 // The conditions here and in handoffp must agree: if
2712 // findrunnable would return a G to run, handoffp must start
2717 if sched.gcwaiting.Load() {
2721 if pp.runSafePointFn != 0 {
2725 // now and pollUntil are saved for work stealing later,
2726 // which may steal timers. It's important that between now
2727 // and then, nothing blocks, so these numbers remain mostly
2729 now, pollUntil, _ := checkTimers(pp, 0)
2731 // Try to schedule the trace reader.
2732 if trace.enabled || trace.shutdown {
2735 casgstatus(gp, _Gwaiting, _Grunnable)
2736 traceGoUnpark(gp, 0)
2737 return gp, false, true
2741 // Try to schedule a GC worker.
2742 if gcBlackenEnabled != 0 {
2743 gp, tnow := gcController.findRunnableGCWorker(pp, now)
2745 return gp, false, true
2750 // Check the global runnable queue once in a while to ensure fairness.
2751 // Otherwise two goroutines can completely occupy the local runqueue
2752 // by constantly respawning each other.
2753 if pp.schedtick%61 == 0 && sched.runqsize > 0 {
2755 gp := globrunqget(pp, 1)
2758 return gp, false, false
2762 // Wake up the finalizer G.
2763 if fingStatus.Load()&(fingWait|fingWake) == fingWait|fingWake {
2764 if gp := wakefing(); gp != nil {
2768 if *cgo_yield != nil {
2769 asmcgocall(*cgo_yield, nil)
2773 if gp, inheritTime := runqget(pp); gp != nil {
2774 return gp, inheritTime, false
2778 if sched.runqsize != 0 {
2780 gp := globrunqget(pp, 0)
2783 return gp, false, false
2788 // This netpoll is only an optimization before we resort to stealing.
2789 // We can safely skip it if there are no waiters or a thread is blocked
2790 // in netpoll already. If there is any kind of logical race with that
2791 // blocked thread (e.g. it has already returned from netpoll, but does
2792 // not set lastpoll yet), this thread will do blocking netpoll below
2794 if netpollinited() && netpollWaiters.Load() > 0 && sched.lastpoll.Load() != 0 {
2795 if list := netpoll(0); !list.empty() { // non-blocking
2798 casgstatus(gp, _Gwaiting, _Grunnable)
2800 traceGoUnpark(gp, 0)
2802 return gp, false, false
2806 // Spinning Ms: steal work from other Ps.
2808 // Limit the number of spinning Ms to half the number of busy Ps.
2809 // This is necessary to prevent excessive CPU consumption when
2810 // GOMAXPROCS>>1 but the program parallelism is low.
2811 if mp.spinning || 2*sched.nmspinning.Load() < gomaxprocs-sched.npidle.Load() {
2816 gp, inheritTime, tnow, w, newWork := stealWork(now)
2818 // Successfully stole.
2819 return gp, inheritTime, false
2822 // There may be new timer or GC work; restart to
2828 if w != 0 && (pollUntil == 0 || w < pollUntil) {
2829 // Earlier timer to wait for.
2834 // We have nothing to do.
2836 // If we're in the GC mark phase, can safely scan and blacken objects,
2837 // and have work to do, run idle-time marking rather than give up the P.
2838 if gcBlackenEnabled != 0 && gcMarkWorkAvailable(pp) && gcController.addIdleMarkWorker() {
2839 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
2841 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
2843 casgstatus(gp, _Gwaiting, _Grunnable)
2845 traceGoUnpark(gp, 0)
2847 return gp, false, false
2849 gcController.removeIdleMarkWorker()
2853 // If a callback returned and no other goroutine is awake,
2854 // then wake event handler goroutine which pauses execution
2855 // until a callback was triggered.
2856 gp, otherReady := beforeIdle(now, pollUntil)
2858 casgstatus(gp, _Gwaiting, _Grunnable)
2860 traceGoUnpark(gp, 0)
2862 return gp, false, false
2868 // Before we drop our P, make a snapshot of the allp slice,
2869 // which can change underfoot once we no longer block
2870 // safe-points. We don't need to snapshot the contents because
2871 // everything up to cap(allp) is immutable.
2872 allpSnapshot := allp
2873 // Also snapshot masks. Value changes are OK, but we can't allow
2874 // len to change out from under us.
2875 idlepMaskSnapshot := idlepMask
2876 timerpMaskSnapshot := timerpMask
2878 // return P and block
2880 if sched.gcwaiting.Load() || pp.runSafePointFn != 0 {
2884 if sched.runqsize != 0 {
2885 gp := globrunqget(pp, 0)
2887 return gp, false, false
2889 if !mp.spinning && sched.needspinning.Load() == 1 {
2890 // See "Delicate dance" comment below.
2895 if releasep() != pp {
2896 throw("findrunnable: wrong p")
2898 now = pidleput(pp, now)
2901 // Delicate dance: thread transitions from spinning to non-spinning
2902 // state, potentially concurrently with submission of new work. We must
2903 // drop nmspinning first and then check all sources again (with
2904 // #StoreLoad memory barrier in between). If we do it the other way
2905 // around, another thread can submit work after we've checked all
2906 // sources but before we drop nmspinning; as a result nobody will
2907 // unpark a thread to run the work.
2909 // This applies to the following sources of work:
2911 // * Goroutines added to a per-P run queue.
2912 // * New/modified-earlier timers on a per-P timer heap.
2913 // * Idle-priority GC work (barring golang.org/issue/19112).
2915 // If we discover new work below, we need to restore m.spinning as a
2916 // signal for resetspinning to unpark a new worker thread (because
2917 // there can be more than one starving goroutine).
2919 // However, if after discovering new work we also observe no idle Ps
2920 // (either here or in resetspinning), we have a problem. We may be
2921 // racing with a non-spinning M in the block above, having found no
2922 // work and preparing to release its P and park. Allowing that P to go
2923 // idle will result in loss of work conservation (idle P while there is
2924 // runnable work). This could result in complete deadlock in the
2925 // unlikely event that we discover new work (from netpoll) right as we
2926 // are racing with _all_ other Ps going idle.
2928 // We use sched.needspinning to synchronize with non-spinning Ms going
2929 // idle. If needspinning is set when they are about to drop their P,
2930 // they abort the drop and instead become a new spinning M on our
2931 // behalf. If we are not racing and the system is truly fully loaded
2932 // then no spinning threads are required, and the next thread to
2933 // naturally become spinning will clear the flag.
2935 // Also see "Worker thread parking/unparking" comment at the top of the
2937 wasSpinning := mp.spinning
2940 if sched.nmspinning.Add(-1) < 0 {
2941 throw("findrunnable: negative nmspinning")
2944 // Note the for correctness, only the last M transitioning from
2945 // spinning to non-spinning must perform these rechecks to
2946 // ensure no missed work. However, the runtime has some cases
2947 // of transient increments of nmspinning that are decremented
2948 // without going through this path, so we must be conservative
2949 // and perform the check on all spinning Ms.
2951 // See https://go.dev/issue/43997.
2953 // Check all runqueues once again.
2954 pp := checkRunqsNoP(allpSnapshot, idlepMaskSnapshot)
2961 // Check for idle-priority GC work again.
2962 pp, gp := checkIdleGCNoP()
2967 // Run the idle worker.
2968 pp.gcMarkWorkerMode = gcMarkWorkerIdleMode
2969 casgstatus(gp, _Gwaiting, _Grunnable)
2971 traceGoUnpark(gp, 0)
2973 return gp, false, false
2976 // Finally, check for timer creation or expiry concurrently with
2977 // transitioning from spinning to non-spinning.
2979 // Note that we cannot use checkTimers here because it calls
2980 // adjusttimers which may need to allocate memory, and that isn't
2981 // allowed when we don't have an active P.
2982 pollUntil = checkTimersNoP(allpSnapshot, timerpMaskSnapshot, pollUntil)
2985 // Poll network until next timer.
2986 if netpollinited() && (netpollWaiters.Load() > 0 || pollUntil != 0) && sched.lastpoll.Swap(0) != 0 {
2987 sched.pollUntil.Store(pollUntil)
2989 throw("findrunnable: netpoll with p")
2992 throw("findrunnable: netpoll with spinning")
2998 delay = pollUntil - now
3004 // When using fake time, just poll.
3007 list := netpoll(delay) // block until new work is available
3008 sched.pollUntil.Store(0)
3009 sched.lastpoll.Store(now)
3010 if faketime != 0 && list.empty() {
3011 // Using fake time and nothing is ready; stop M.
3012 // When all M's stop, checkdead will call timejump.
3017 pp, _ := pidleget(now)
3026 casgstatus(gp, _Gwaiting, _Grunnable)
3028 traceGoUnpark(gp, 0)
3030 return gp, false, false
3037 } else if pollUntil != 0 && netpollinited() {
3038 pollerPollUntil := sched.pollUntil.Load()
3039 if pollerPollUntil == 0 || pollerPollUntil > pollUntil {
3047 // pollWork reports whether there is non-background work this P could
3048 // be doing. This is a fairly lightweight check to be used for
3049 // background work loops, like idle GC. It checks a subset of the
3050 // conditions checked by the actual scheduler.
3051 func pollWork() bool {
3052 if sched.runqsize != 0 {
3055 p := getg().m.p.ptr()
3059 if netpollinited() && netpollWaiters.Load() > 0 && sched.lastpoll.Load() != 0 {
3060 if list := netpoll(0); !list.empty() {
3068 // stealWork attempts to steal a runnable goroutine or timer from any P.
3070 // If newWork is true, new work may have been readied.
3072 // If now is not 0 it is the current time. stealWork returns the passed time or
3073 // the current time if now was passed as 0.
3074 func stealWork(now int64) (gp *g, inheritTime bool, rnow, pollUntil int64, newWork bool) {
3075 pp := getg().m.p.ptr()
3079 const stealTries = 4
3080 for i := 0; i < stealTries; i++ {
3081 stealTimersOrRunNextG := i == stealTries-1
3083 for enum := stealOrder.start(fastrand()); !enum.done(); enum.next() {
3084 if sched.gcwaiting.Load() {
3085 // GC work may be available.
3086 return nil, false, now, pollUntil, true
3088 p2 := allp[enum.position()]
3093 // Steal timers from p2. This call to checkTimers is the only place
3094 // where we might hold a lock on a different P's timers. We do this
3095 // once on the last pass before checking runnext because stealing
3096 // from the other P's runnext should be the last resort, so if there
3097 // are timers to steal do that first.
3099 // We only check timers on one of the stealing iterations because
3100 // the time stored in now doesn't change in this loop and checking
3101 // the timers for each P more than once with the same value of now
3102 // is probably a waste of time.
3104 // timerpMask tells us whether the P may have timers at all. If it
3105 // can't, no need to check at all.
3106 if stealTimersOrRunNextG && timerpMask.read(enum.position()) {
3107 tnow, w, ran := checkTimers(p2, now)
3109 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3113 // Running the timers may have
3114 // made an arbitrary number of G's
3115 // ready and added them to this P's
3116 // local run queue. That invalidates
3117 // the assumption of runqsteal
3118 // that it always has room to add
3119 // stolen G's. So check now if there
3120 // is a local G to run.
3121 if gp, inheritTime := runqget(pp); gp != nil {
3122 return gp, inheritTime, now, pollUntil, ranTimer
3128 // Don't bother to attempt to steal if p2 is idle.
3129 if !idlepMask.read(enum.position()) {
3130 if gp := runqsteal(pp, p2, stealTimersOrRunNextG); gp != nil {
3131 return gp, false, now, pollUntil, ranTimer
3137 // No goroutines found to steal. Regardless, running a timer may have
3138 // made some goroutine ready that we missed. Indicate the next timer to
3140 return nil, false, now, pollUntil, ranTimer
3143 // Check all Ps for a runnable G to steal.
3145 // On entry we have no P. If a G is available to steal and a P is available,
3146 // the P is returned which the caller should acquire and attempt to steal the
3148 func checkRunqsNoP(allpSnapshot []*p, idlepMaskSnapshot pMask) *p {
3149 for id, p2 := range allpSnapshot {
3150 if !idlepMaskSnapshot.read(uint32(id)) && !runqempty(p2) {
3152 pp, _ := pidlegetSpinning(0)
3154 // Can't get a P, don't bother checking remaining Ps.
3163 // No work available.
3167 // Check all Ps for a timer expiring sooner than pollUntil.
3169 // Returns updated pollUntil value.
3170 func checkTimersNoP(allpSnapshot []*p, timerpMaskSnapshot pMask, pollUntil int64) int64 {
3171 for id, p2 := range allpSnapshot {
3172 if timerpMaskSnapshot.read(uint32(id)) {
3173 w := nobarrierWakeTime(p2)
3174 if w != 0 && (pollUntil == 0 || w < pollUntil) {
3183 // Check for idle-priority GC, without a P on entry.
3185 // If some GC work, a P, and a worker G are all available, the P and G will be
3186 // returned. The returned P has not been wired yet.
3187 func checkIdleGCNoP() (*p, *g) {
3188 // N.B. Since we have no P, gcBlackenEnabled may change at any time; we
3189 // must check again after acquiring a P. As an optimization, we also check
3190 // if an idle mark worker is needed at all. This is OK here, because if we
3191 // observe that one isn't needed, at least one is currently running. Even if
3192 // it stops running, its own journey into the scheduler should schedule it
3193 // again, if need be (at which point, this check will pass, if relevant).
3194 if atomic.Load(&gcBlackenEnabled) == 0 || !gcController.needIdleMarkWorker() {
3197 if !gcMarkWorkAvailable(nil) {
3201 // Work is available; we can start an idle GC worker only if there is
3202 // an available P and available worker G.
3204 // We can attempt to acquire these in either order, though both have
3205 // synchronization concerns (see below). Workers are almost always
3206 // available (see comment in findRunnableGCWorker for the one case
3207 // there may be none). Since we're slightly less likely to find a P,
3208 // check for that first.
3210 // Synchronization: note that we must hold sched.lock until we are
3211 // committed to keeping it. Otherwise we cannot put the unnecessary P
3212 // back in sched.pidle without performing the full set of idle
3213 // transition checks.
3215 // If we were to check gcBgMarkWorkerPool first, we must somehow handle
3216 // the assumption in gcControllerState.findRunnableGCWorker that an
3217 // empty gcBgMarkWorkerPool is only possible if gcMarkDone is running.
3219 pp, now := pidlegetSpinning(0)
3225 // Now that we own a P, gcBlackenEnabled can't change (as it requires STW).
3226 if gcBlackenEnabled == 0 || !gcController.addIdleMarkWorker() {
3232 node := (*gcBgMarkWorkerNode)(gcBgMarkWorkerPool.pop())
3236 gcController.removeIdleMarkWorker()
3242 return pp, node.gp.ptr()
3245 // wakeNetPoller wakes up the thread sleeping in the network poller if it isn't
3246 // going to wake up before the when argument; or it wakes an idle P to service
3247 // timers and the network poller if there isn't one already.
3248 func wakeNetPoller(when int64) {
3249 if sched.lastpoll.Load() == 0 {
3250 // In findrunnable we ensure that when polling the pollUntil
3251 // field is either zero or the time to which the current
3252 // poll is expected to run. This can have a spurious wakeup
3253 // but should never miss a wakeup.
3254 pollerPollUntil := sched.pollUntil.Load()
3255 if pollerPollUntil == 0 || pollerPollUntil > when {
3259 // There are no threads in the network poller, try to get
3260 // one there so it can handle new timers.
3261 if GOOS != "plan9" { // Temporary workaround - see issue #42303.
3267 func resetspinning() {
3270 throw("resetspinning: not a spinning m")
3272 gp.m.spinning = false
3273 nmspinning := sched.nmspinning.Add(-1)
3275 throw("findrunnable: negative nmspinning")
3277 // M wakeup policy is deliberately somewhat conservative, so check if we
3278 // need to wakeup another P here. See "Worker thread parking/unparking"
3279 // comment at the top of the file for details.
3283 // injectglist adds each runnable G on the list to some run queue,
3284 // and clears glist. If there is no current P, they are added to the
3285 // global queue, and up to npidle M's are started to run them.
3286 // Otherwise, for each idle P, this adds a G to the global queue
3287 // and starts an M. Any remaining G's are added to the current P's
3289 // This may temporarily acquire sched.lock.
3290 // Can run concurrently with GC.
3291 func injectglist(glist *gList) {
3296 for gp := glist.head.ptr(); gp != nil; gp = gp.schedlink.ptr() {
3297 traceGoUnpark(gp, 0)
3301 // Mark all the goroutines as runnable before we put them
3302 // on the run queues.
3303 head := glist.head.ptr()
3306 for gp := head; gp != nil; gp = gp.schedlink.ptr() {
3309 casgstatus(gp, _Gwaiting, _Grunnable)
3312 // Turn the gList into a gQueue.
3318 startIdle := func(n int) {
3319 for i := 0; i < n; i++ {
3320 mp := acquirem() // See comment in startm.
3323 pp, _ := pidlegetSpinning(0)
3330 startm(pp, false, true)
3336 pp := getg().m.p.ptr()
3339 globrunqputbatch(&q, int32(qsize))
3345 npidle := int(sched.npidle.Load())
3348 for n = 0; n < npidle && !q.empty(); n++ {
3354 globrunqputbatch(&globq, int32(n))
3361 runqputbatch(pp, &q, qsize)
3365 // One round of scheduler: find a runnable goroutine and execute it.
3371 throw("schedule: holding locks")
3374 if mp.lockedg != 0 {
3376 execute(mp.lockedg.ptr(), false) // Never returns.
3379 // We should not schedule away from a g that is executing a cgo call,
3380 // since the cgo call is using the m's g0 stack.
3382 throw("schedule: in cgo")
3389 // Safety check: if we are spinning, the run queue should be empty.
3390 // Check this before calling checkTimers, as that might call
3391 // goready to put a ready goroutine on the local run queue.
3392 if mp.spinning && (pp.runnext != 0 || pp.runqhead != pp.runqtail) {
3393 throw("schedule: spinning with local work")
3396 gp, inheritTime, tryWakeP := findRunnable() // blocks until work is available
3398 // This thread is going to run a goroutine and is not spinning anymore,
3399 // so if it was marked as spinning we need to reset it now and potentially
3400 // start a new spinning M.
3405 if sched.disable.user && !schedEnabled(gp) {
3406 // Scheduling of this goroutine is disabled. Put it on
3407 // the list of pending runnable goroutines for when we
3408 // re-enable user scheduling and look again.
3410 if schedEnabled(gp) {
3411 // Something re-enabled scheduling while we
3412 // were acquiring the lock.
3415 sched.disable.runnable.pushBack(gp)
3422 // If about to schedule a not-normal goroutine (a GCworker or tracereader),
3423 // wake a P if there is one.
3427 if gp.lockedm != 0 {
3428 // Hands off own p to the locked m,
3429 // then blocks waiting for a new p.
3434 execute(gp, inheritTime)
3437 // dropg removes the association between m and the current goroutine m->curg (gp for short).
3438 // Typically a caller sets gp's status away from Grunning and then
3439 // immediately calls dropg to finish the job. The caller is also responsible
3440 // for arranging that gp will be restarted using ready at an
3441 // appropriate time. After calling dropg and arranging for gp to be
3442 // readied later, the caller can do other work but eventually should
3443 // call schedule to restart the scheduling of goroutines on this m.
3447 setMNoWB(&gp.m.curg.m, nil)
3448 setGNoWB(&gp.m.curg, nil)
3451 // checkTimers runs any timers for the P that are ready.
3452 // If now is not 0 it is the current time.
3453 // It returns the passed time or the current time if now was passed as 0.
3454 // and the time when the next timer should run or 0 if there is no next timer,
3455 // and reports whether it ran any timers.
3456 // If the time when the next timer should run is not 0,
3457 // it is always larger than the returned time.
3458 // We pass now in and out to avoid extra calls of nanotime.
3460 //go:yeswritebarrierrec
3461 func checkTimers(pp *p, now int64) (rnow, pollUntil int64, ran bool) {
3462 // If it's not yet time for the first timer, or the first adjusted
3463 // timer, then there is nothing to do.
3464 next := pp.timer0When.Load()
3465 nextAdj := pp.timerModifiedEarliest.Load()
3466 if next == 0 || (nextAdj != 0 && nextAdj < next) {
3471 // No timers to run or adjust.
3472 return now, 0, false
3479 // Next timer is not ready to run, but keep going
3480 // if we would clear deleted timers.
3481 // This corresponds to the condition below where
3482 // we decide whether to call clearDeletedTimers.
3483 if pp != getg().m.p.ptr() || int(pp.deletedTimers.Load()) <= int(pp.numTimers.Load()/4) {
3484 return now, next, false
3488 lock(&pp.timersLock)
3490 if len(pp.timers) > 0 {
3491 adjusttimers(pp, now)
3492 for len(pp.timers) > 0 {
3493 // Note that runtimer may temporarily unlock
3495 if tw := runtimer(pp, now); tw != 0 {
3505 // If this is the local P, and there are a lot of deleted timers,
3506 // clear them out. We only do this for the local P to reduce
3507 // lock contention on timersLock.
3508 if pp == getg().m.p.ptr() && int(pp.deletedTimers.Load()) > len(pp.timers)/4 {
3509 clearDeletedTimers(pp)
3512 unlock(&pp.timersLock)
3514 return now, pollUntil, ran
3517 func parkunlock_c(gp *g, lock unsafe.Pointer) bool {
3518 unlock((*mutex)(lock))
3522 // park continuation on g0.
3523 func park_m(gp *g) {
3527 traceGoPark(mp.waittraceev, mp.waittraceskip)
3530 // N.B. Not using casGToWaiting here because the waitreason is
3531 // set by park_m's caller.
3532 casgstatus(gp, _Grunning, _Gwaiting)
3535 if fn := mp.waitunlockf; fn != nil {
3536 ok := fn(gp, mp.waitlock)
3537 mp.waitunlockf = nil
3541 traceGoUnpark(gp, 2)
3543 casgstatus(gp, _Gwaiting, _Grunnable)
3544 execute(gp, true) // Schedule it back, never returns.
3550 func goschedImpl(gp *g) {
3551 status := readgstatus(gp)
3552 if status&^_Gscan != _Grunning {
3554 throw("bad g status")
3556 casgstatus(gp, _Grunning, _Grunnable)
3565 // Gosched continuation on g0.
3566 func gosched_m(gp *g) {
3573 // goschedguarded is a forbidden-states-avoided version of gosched_m.
3574 func goschedguarded_m(gp *g) {
3576 if !canPreemptM(gp.m) {
3577 gogo(&gp.sched) // never return
3586 func gopreempt_m(gp *g) {
3593 // preemptPark parks gp and puts it in _Gpreempted.
3596 func preemptPark(gp *g) {
3598 traceGoPark(traceEvGoBlock, 0)
3600 status := readgstatus(gp)
3601 if status&^_Gscan != _Grunning {
3603 throw("bad g status")
3606 if gp.asyncSafePoint {
3607 // Double-check that async preemption does not
3608 // happen in SPWRITE assembly functions.
3609 // isAsyncSafePoint must exclude this case.
3610 f := findfunc(gp.sched.pc)
3612 throw("preempt at unknown pc")
3614 if f.flag&abi.FuncFlagSPWrite != 0 {
3615 println("runtime: unexpected SPWRITE function", funcname(f), "in async preempt")
3616 throw("preempt SPWRITE")
3620 // Transition from _Grunning to _Gscan|_Gpreempted. We can't
3621 // be in _Grunning when we dropg because then we'd be running
3622 // without an M, but the moment we're in _Gpreempted,
3623 // something could claim this G before we've fully cleaned it
3624 // up. Hence, we set the scan bit to lock down further
3625 // transitions until we can dropg.
3626 casGToPreemptScan(gp, _Grunning, _Gscan|_Gpreempted)
3628 casfrom_Gscanstatus(gp, _Gscan|_Gpreempted, _Gpreempted)
3632 // goyield is like Gosched, but it:
3633 // - emits a GoPreempt trace event instead of a GoSched trace event
3634 // - puts the current G on the runq of the current P instead of the globrunq
3640 func goyield_m(gp *g) {
3645 casgstatus(gp, _Grunning, _Grunnable)
3647 runqput(pp, gp, false)
3651 // Finishes execution of the current goroutine.
3662 // goexit continuation on g0.
3663 func goexit0(gp *g) {
3667 casgstatus(gp, _Grunning, _Gdead)
3668 gcController.addScannableStack(pp, -int64(gp.stack.hi-gp.stack.lo))
3669 if isSystemGoroutine(gp, false) {
3673 locked := gp.lockedm != 0
3676 gp.preemptStop = false
3677 gp.paniconfault = false
3678 gp._defer = nil // should be true already but just in case.
3679 gp._panic = nil // non-nil for Goexit during panic. points at stack-allocated data.
3681 gp.waitreason = waitReasonZero
3686 if gcBlackenEnabled != 0 && gp.gcAssistBytes > 0 {
3687 // Flush assist credit to the global pool. This gives
3688 // better information to pacing if the application is
3689 // rapidly creating an exiting goroutines.
3690 assistWorkPerByte := gcController.assistWorkPerByte.Load()
3691 scanCredit := int64(assistWorkPerByte * float64(gp.gcAssistBytes))
3692 gcController.bgScanCredit.Add(scanCredit)
3693 gp.gcAssistBytes = 0
3698 if GOARCH == "wasm" { // no threads yet on wasm
3700 schedule() // never returns
3703 if mp.lockedInt != 0 {
3704 print("invalid m->lockedInt = ", mp.lockedInt, "\n")
3705 throw("internal lockOSThread error")
3709 // The goroutine may have locked this thread because
3710 // it put it in an unusual kernel state. Kill it
3711 // rather than returning it to the thread pool.
3713 // Return to mstart, which will release the P and exit
3715 if GOOS != "plan9" { // See golang.org/issue/22227.
3718 // Clear lockedExt on plan9 since we may end up re-using
3726 // save updates getg().sched to refer to pc and sp so that a following
3727 // gogo will restore pc and sp.
3729 // save must not have write barriers because invoking a write barrier
3730 // can clobber getg().sched.
3733 //go:nowritebarrierrec
3734 func save(pc, sp uintptr) {
3737 if gp == gp.m.g0 || gp == gp.m.gsignal {
3738 // m.g0.sched is special and must describe the context
3739 // for exiting the thread. mstart1 writes to it directly.
3740 // m.gsignal.sched should not be used at all.
3741 // This check makes sure save calls do not accidentally
3742 // run in contexts where they'd write to system g's.
3743 throw("save on system g not allowed")
3750 // We need to ensure ctxt is zero, but can't have a write
3751 // barrier here. However, it should always already be zero.
3753 if gp.sched.ctxt != nil {
3758 // The goroutine g is about to enter a system call.
3759 // Record that it's not using the cpu anymore.
3760 // This is called only from the go syscall library and cgocall,
3761 // not from the low-level system calls used by the runtime.
3763 // Entersyscall cannot split the stack: the save must
3764 // make g->sched refer to the caller's stack segment, because
3765 // entersyscall is going to return immediately after.
3767 // Nothing entersyscall calls can split the stack either.
3768 // We cannot safely move the stack during an active call to syscall,
3769 // because we do not know which of the uintptr arguments are
3770 // really pointers (back into the stack).
3771 // In practice, this means that we make the fast path run through
3772 // entersyscall doing no-split things, and the slow path has to use systemstack
3773 // to run bigger things on the system stack.
3775 // reentersyscall is the entry point used by cgo callbacks, where explicitly
3776 // saved SP and PC are restored. This is needed when exitsyscall will be called
3777 // from a function further up in the call stack than the parent, as g->syscallsp
3778 // must always point to a valid stack frame. entersyscall below is the normal
3779 // entry point for syscalls, which obtains the SP and PC from the caller.
3782 // At the start of a syscall we emit traceGoSysCall to capture the stack trace.
3783 // If the syscall does not block, that is it, we do not emit any other events.
3784 // If the syscall blocks (that is, P is retaken), retaker emits traceGoSysBlock;
3785 // when syscall returns we emit traceGoSysExit and when the goroutine starts running
3786 // (potentially instantly, if exitsyscallfast returns true) we emit traceGoStart.
3787 // To ensure that traceGoSysExit is emitted strictly after traceGoSysBlock,
3788 // we remember current value of syscalltick in m (gp.m.syscalltick = gp.m.p.ptr().syscalltick),
3789 // whoever emits traceGoSysBlock increments p.syscalltick afterwards;
3790 // and we wait for the increment before emitting traceGoSysExit.
3791 // Note that the increment is done even if tracing is not enabled,
3792 // because tracing can be enabled in the middle of syscall. We don't want the wait to hang.
3795 func reentersyscall(pc, sp uintptr) {
3798 // Disable preemption because during this function g is in Gsyscall status,
3799 // but can have inconsistent g->sched, do not let GC observe it.
3802 // Entersyscall must not call any function that might split/grow the stack.
3803 // (See details in comment above.)
3804 // Catch calls that might, by replacing the stack guard with something that
3805 // will trip any stack check and leaving a flag to tell newstack to die.
3806 gp.stackguard0 = stackPreempt
3807 gp.throwsplit = true
3809 // Leave SP around for GC and traceback.
3813 casgstatus(gp, _Grunning, _Gsyscall)
3814 if staticLockRanking {
3815 // When doing static lock ranking casgstatus can call
3816 // systemstack which clobbers g.sched.
3819 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
3820 systemstack(func() {
3821 print("entersyscall inconsistent ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
3822 throw("entersyscall")
3827 systemstack(traceGoSysCall)
3828 // systemstack itself clobbers g.sched.{pc,sp} and we might
3829 // need them later when the G is genuinely blocked in a
3834 if sched.sysmonwait.Load() {
3835 systemstack(entersyscall_sysmon)
3839 if gp.m.p.ptr().runSafePointFn != 0 {
3840 // runSafePointFn may stack split if run on this stack
3841 systemstack(runSafePointFn)
3845 gp.m.syscalltick = gp.m.p.ptr().syscalltick
3846 gp.sysblocktraced = true
3851 atomic.Store(&pp.status, _Psyscall)
3852 if sched.gcwaiting.Load() {
3853 systemstack(entersyscall_gcwait)
3860 // Standard syscall entry used by the go syscall library and normal cgo calls.
3862 // This is exported via linkname to assembly in the syscall package and x/sys.
3865 //go:linkname entersyscall
3866 func entersyscall() {
3867 reentersyscall(getcallerpc(), getcallersp())
3870 func entersyscall_sysmon() {
3872 if sched.sysmonwait.Load() {
3873 sched.sysmonwait.Store(false)
3874 notewakeup(&sched.sysmonnote)
3879 func entersyscall_gcwait() {
3881 pp := gp.m.oldp.ptr()
3884 if sched.stopwait > 0 && atomic.Cas(&pp.status, _Psyscall, _Pgcstop) {
3890 if sched.stopwait--; sched.stopwait == 0 {
3891 notewakeup(&sched.stopnote)
3897 // The same as entersyscall(), but with a hint that the syscall is blocking.
3900 func entersyscallblock() {
3903 gp.m.locks++ // see comment in entersyscall
3904 gp.throwsplit = true
3905 gp.stackguard0 = stackPreempt // see comment in entersyscall
3906 gp.m.syscalltick = gp.m.p.ptr().syscalltick
3907 gp.sysblocktraced = true
3908 gp.m.p.ptr().syscalltick++
3910 // Leave SP around for GC and traceback.
3914 gp.syscallsp = gp.sched.sp
3915 gp.syscallpc = gp.sched.pc
3916 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
3920 systemstack(func() {
3921 print("entersyscallblock inconsistent ", hex(sp1), " ", hex(sp2), " ", hex(sp3), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
3922 throw("entersyscallblock")
3925 casgstatus(gp, _Grunning, _Gsyscall)
3926 if gp.syscallsp < gp.stack.lo || gp.stack.hi < gp.syscallsp {
3927 systemstack(func() {
3928 print("entersyscallblock inconsistent ", hex(sp), " ", hex(gp.sched.sp), " ", hex(gp.syscallsp), " [", hex(gp.stack.lo), ",", hex(gp.stack.hi), "]\n")
3929 throw("entersyscallblock")
3933 systemstack(entersyscallblock_handoff)
3935 // Resave for traceback during blocked call.
3936 save(getcallerpc(), getcallersp())
3941 func entersyscallblock_handoff() {
3944 traceGoSysBlock(getg().m.p.ptr())
3946 handoffp(releasep())
3949 // The goroutine g exited its system call.
3950 // Arrange for it to run on a cpu again.
3951 // This is called only from the go syscall library, not
3952 // from the low-level system calls used by the runtime.
3954 // Write barriers are not allowed because our P may have been stolen.
3956 // This is exported via linkname to assembly in the syscall package.
3959 //go:nowritebarrierrec
3960 //go:linkname exitsyscall
3961 func exitsyscall() {
3964 gp.m.locks++ // see comment in entersyscall
3965 if getcallersp() > gp.syscallsp {
3966 throw("exitsyscall: syscall frame is no longer valid")
3970 oldp := gp.m.oldp.ptr()
3972 if exitsyscallfast(oldp) {
3973 // When exitsyscallfast returns success, we have a P so can now use
3975 if goroutineProfile.active {
3976 // Make sure that gp has had its stack written out to the goroutine
3977 // profile, exactly as it was when the goroutine profiler first
3978 // stopped the world.
3979 systemstack(func() {
3980 tryRecordGoroutineProfileWB(gp)
3984 if oldp != gp.m.p.ptr() || gp.m.syscalltick != gp.m.p.ptr().syscalltick {
3985 systemstack(traceGoStart)
3988 // There's a cpu for us, so we can run.
3989 gp.m.p.ptr().syscalltick++
3990 // We need to cas the status and scan before resuming...
3991 casgstatus(gp, _Gsyscall, _Grunning)
3993 // Garbage collector isn't running (since we are),
3994 // so okay to clear syscallsp.
3998 // restore the preemption request in case we've cleared it in newstack
3999 gp.stackguard0 = stackPreempt
4001 // otherwise restore the real stackGuard, we've spoiled it in entersyscall/entersyscallblock
4002 gp.stackguard0 = gp.stack.lo + stackGuard
4004 gp.throwsplit = false
4006 if sched.disable.user && !schedEnabled(gp) {
4007 // Scheduling of this goroutine is disabled.
4016 // Wait till traceGoSysBlock event is emitted.
4017 // This ensures consistency of the trace (the goroutine is started after it is blocked).
4018 for oldp != nil && oldp.syscalltick == gp.m.syscalltick {
4021 // We can't trace syscall exit right now because we don't have a P.
4022 // Tracing code can invoke write barriers that cannot run without a P.
4023 // So instead we remember the syscall exit time and emit the event
4024 // in execute when we have a P.
4025 gp.sysexitticks = cputicks()
4030 // Call the scheduler.
4033 // Scheduler returned, so we're allowed to run now.
4034 // Delete the syscallsp information that we left for
4035 // the garbage collector during the system call.
4036 // Must wait until now because until gosched returns
4037 // we don't know for sure that the garbage collector
4040 gp.m.p.ptr().syscalltick++
4041 gp.throwsplit = false
4045 func exitsyscallfast(oldp *p) bool {
4048 // Freezetheworld sets stopwait but does not retake P's.
4049 if sched.stopwait == freezeStopWait {
4053 // Try to re-acquire the last P.
4054 if oldp != nil && oldp.status == _Psyscall && atomic.Cas(&oldp.status, _Psyscall, _Pidle) {
4055 // There's a cpu for us, so we can run.
4057 exitsyscallfast_reacquired()
4061 // Try to get any other idle P.
4062 if sched.pidle != 0 {
4064 systemstack(func() {
4065 ok = exitsyscallfast_pidle()
4066 if ok && trace.enabled {
4068 // Wait till traceGoSysBlock event is emitted.
4069 // This ensures consistency of the trace (the goroutine is started after it is blocked).
4070 for oldp.syscalltick == gp.m.syscalltick {
4084 // exitsyscallfast_reacquired is the exitsyscall path on which this G
4085 // has successfully reacquired the P it was running on before the
4089 func exitsyscallfast_reacquired() {
4091 if gp.m.syscalltick != gp.m.p.ptr().syscalltick {
4093 // The p was retaken and then enter into syscall again (since gp.m.syscalltick has changed).
4094 // traceGoSysBlock for this syscall was already emitted,
4095 // but here we effectively retake the p from the new syscall running on the same p.
4096 systemstack(func() {
4097 // Denote blocking of the new syscall.
4098 traceGoSysBlock(gp.m.p.ptr())
4099 // Denote completion of the current syscall.
4103 gp.m.p.ptr().syscalltick++
4107 func exitsyscallfast_pidle() bool {
4109 pp, _ := pidleget(0)
4110 if pp != nil && sched.sysmonwait.Load() {
4111 sched.sysmonwait.Store(false)
4112 notewakeup(&sched.sysmonnote)
4122 // exitsyscall slow path on g0.
4123 // Failed to acquire P, enqueue gp as runnable.
4125 // Called via mcall, so gp is the calling g from this M.
4127 //go:nowritebarrierrec
4128 func exitsyscall0(gp *g) {
4129 casgstatus(gp, _Gsyscall, _Grunnable)
4133 if schedEnabled(gp) {
4140 // Below, we stoplockedm if gp is locked. globrunqput releases
4141 // ownership of gp, so we must check if gp is locked prior to
4142 // committing the release by unlocking sched.lock, otherwise we
4143 // could race with another M transitioning gp from unlocked to
4145 locked = gp.lockedm != 0
4146 } else if sched.sysmonwait.Load() {
4147 sched.sysmonwait.Store(false)
4148 notewakeup(&sched.sysmonnote)
4153 execute(gp, false) // Never returns.
4156 // Wait until another thread schedules gp and so m again.
4158 // N.B. lockedm must be this M, as this g was running on this M
4159 // before entersyscall.
4161 execute(gp, false) // Never returns.
4164 schedule() // Never returns.
4167 // Called from syscall package before fork.
4169 //go:linkname syscall_runtime_BeforeFork syscall.runtime_BeforeFork
4171 func syscall_runtime_BeforeFork() {
4174 // Block signals during a fork, so that the child does not run
4175 // a signal handler before exec if a signal is sent to the process
4176 // group. See issue #18600.
4178 sigsave(&gp.m.sigmask)
4181 // This function is called before fork in syscall package.
4182 // Code between fork and exec must not allocate memory nor even try to grow stack.
4183 // Here we spoil g.stackguard0 to reliably detect any attempts to grow stack.
4184 // runtime_AfterFork will undo this in parent process, but not in child.
4185 gp.stackguard0 = stackFork
4188 // Called from syscall package after fork in parent.
4190 //go:linkname syscall_runtime_AfterFork syscall.runtime_AfterFork
4192 func syscall_runtime_AfterFork() {
4195 // See the comments in beforefork.
4196 gp.stackguard0 = gp.stack.lo + stackGuard
4198 msigrestore(gp.m.sigmask)
4203 // inForkedChild is true while manipulating signals in the child process.
4204 // This is used to avoid calling libc functions in case we are using vfork.
4205 var inForkedChild bool
4207 // Called from syscall package after fork in child.
4208 // It resets non-sigignored signals to the default handler, and
4209 // restores the signal mask in preparation for the exec.
4211 // Because this might be called during a vfork, and therefore may be
4212 // temporarily sharing address space with the parent process, this must
4213 // not change any global variables or calling into C code that may do so.
4215 //go:linkname syscall_runtime_AfterForkInChild syscall.runtime_AfterForkInChild
4217 //go:nowritebarrierrec
4218 func syscall_runtime_AfterForkInChild() {
4219 // It's OK to change the global variable inForkedChild here
4220 // because we are going to change it back. There is no race here,
4221 // because if we are sharing address space with the parent process,
4222 // then the parent process can not be running concurrently.
4223 inForkedChild = true
4225 clearSignalHandlers()
4227 // When we are the child we are the only thread running,
4228 // so we know that nothing else has changed gp.m.sigmask.
4229 msigrestore(getg().m.sigmask)
4231 inForkedChild = false
4234 // pendingPreemptSignals is the number of preemption signals
4235 // that have been sent but not received. This is only used on Darwin.
4237 var pendingPreemptSignals atomic.Int32
4239 // Called from syscall package before Exec.
4241 //go:linkname syscall_runtime_BeforeExec syscall.runtime_BeforeExec
4242 func syscall_runtime_BeforeExec() {
4243 // Prevent thread creation during exec.
4246 // On Darwin, wait for all pending preemption signals to
4247 // be received. See issue #41702.
4248 if GOOS == "darwin" || GOOS == "ios" {
4249 for pendingPreemptSignals.Load() > 0 {
4255 // Called from syscall package after Exec.
4257 //go:linkname syscall_runtime_AfterExec syscall.runtime_AfterExec
4258 func syscall_runtime_AfterExec() {
4262 // Allocate a new g, with a stack big enough for stacksize bytes.
4263 func malg(stacksize int32) *g {
4266 stacksize = round2(stackSystem + stacksize)
4267 systemstack(func() {
4268 newg.stack = stackalloc(uint32(stacksize))
4270 newg.stackguard0 = newg.stack.lo + stackGuard
4271 newg.stackguard1 = ^uintptr(0)
4272 // Clear the bottom word of the stack. We record g
4273 // there on gsignal stack during VDSO on ARM and ARM64.
4274 *(*uintptr)(unsafe.Pointer(newg.stack.lo)) = 0
4279 // Create a new g running fn.
4280 // Put it on the queue of g's waiting to run.
4281 // The compiler turns a go statement into a call to this.
4282 func newproc(fn *funcval) {
4285 systemstack(func() {
4286 newg := newproc1(fn, gp, pc)
4288 pp := getg().m.p.ptr()
4289 runqput(pp, newg, true)
4297 // Create a new g in state _Grunnable, starting at fn. callerpc is the
4298 // address of the go statement that created this. The caller is responsible
4299 // for adding the new g to the scheduler.
4300 func newproc1(fn *funcval, callergp *g, callerpc uintptr) *g {
4302 fatal("go of nil func value")
4305 mp := acquirem() // disable preemption because we hold M and P in local vars.
4309 newg = malg(stackMin)
4310 casgstatus(newg, _Gidle, _Gdead)
4311 allgadd(newg) // publishes with a g->status of Gdead so GC scanner doesn't look at uninitialized stack.
4313 if newg.stack.hi == 0 {
4314 throw("newproc1: newg missing stack")
4317 if readgstatus(newg) != _Gdead {
4318 throw("newproc1: new g is not Gdead")
4321 totalSize := uintptr(4*goarch.PtrSize + sys.MinFrameSize) // extra space in case of reads slightly beyond frame
4322 totalSize = alignUp(totalSize, sys.StackAlign)
4323 sp := newg.stack.hi - totalSize
4327 *(*uintptr)(unsafe.Pointer(sp)) = 0
4329 spArg += sys.MinFrameSize
4332 memclrNoHeapPointers(unsafe.Pointer(&newg.sched), unsafe.Sizeof(newg.sched))
4335 newg.sched.pc = abi.FuncPCABI0(goexit) + sys.PCQuantum // +PCQuantum so that previous instruction is in same function
4336 newg.sched.g = guintptr(unsafe.Pointer(newg))
4337 gostartcallfn(&newg.sched, fn)
4338 newg.parentGoid = callergp.goid
4339 newg.gopc = callerpc
4340 newg.ancestors = saveAncestors(callergp)
4341 newg.startpc = fn.fn
4342 if isSystemGoroutine(newg, false) {
4345 // Only user goroutines inherit pprof labels.
4347 newg.labels = mp.curg.labels
4349 if goroutineProfile.active {
4350 // A concurrent goroutine profile is running. It should include
4351 // exactly the set of goroutines that were alive when the goroutine
4352 // profiler first stopped the world. That does not include newg, so
4353 // mark it as not needing a profile before transitioning it from
4355 newg.goroutineProfiled.Store(goroutineProfileSatisfied)
4358 // Track initial transition?
4359 newg.trackingSeq = uint8(fastrand())
4360 if newg.trackingSeq%gTrackingPeriod == 0 {
4361 newg.tracking = true
4363 casgstatus(newg, _Gdead, _Grunnable)
4364 gcController.addScannableStack(pp, int64(newg.stack.hi-newg.stack.lo))
4366 if pp.goidcache == pp.goidcacheend {
4367 // Sched.goidgen is the last allocated id,
4368 // this batch must be [sched.goidgen+1, sched.goidgen+GoidCacheBatch].
4369 // At startup sched.goidgen=0, so main goroutine receives goid=1.
4370 pp.goidcache = sched.goidgen.Add(_GoidCacheBatch)
4371 pp.goidcache -= _GoidCacheBatch - 1
4372 pp.goidcacheend = pp.goidcache + _GoidCacheBatch
4374 newg.goid = pp.goidcache
4377 newg.racectx = racegostart(callerpc)
4378 if newg.labels != nil {
4379 // See note in proflabel.go on labelSync's role in synchronizing
4380 // with the reads in the signal handler.
4381 racereleasemergeg(newg, unsafe.Pointer(&labelSync))
4385 traceGoCreate(newg, newg.startpc)
4392 // saveAncestors copies previous ancestors of the given caller g and
4393 // includes info for the current caller into a new set of tracebacks for
4394 // a g being created.
4395 func saveAncestors(callergp *g) *[]ancestorInfo {
4396 // Copy all prior info, except for the root goroutine (goid 0).
4397 if debug.tracebackancestors <= 0 || callergp.goid == 0 {
4400 var callerAncestors []ancestorInfo
4401 if callergp.ancestors != nil {
4402 callerAncestors = *callergp.ancestors
4404 n := int32(len(callerAncestors)) + 1
4405 if n > debug.tracebackancestors {
4406 n = debug.tracebackancestors
4408 ancestors := make([]ancestorInfo, n)
4409 copy(ancestors[1:], callerAncestors)
4411 var pcs [tracebackInnerFrames]uintptr
4412 npcs := gcallers(callergp, 0, pcs[:])
4413 ipcs := make([]uintptr, npcs)
4415 ancestors[0] = ancestorInfo{
4417 goid: callergp.goid,
4418 gopc: callergp.gopc,
4421 ancestorsp := new([]ancestorInfo)
4422 *ancestorsp = ancestors
4426 // Put on gfree list.
4427 // If local list is too long, transfer a batch to the global list.
4428 func gfput(pp *p, gp *g) {
4429 if readgstatus(gp) != _Gdead {
4430 throw("gfput: bad status (not Gdead)")
4433 stksize := gp.stack.hi - gp.stack.lo
4435 if stksize != uintptr(startingStackSize) {
4436 // non-standard stack size - free it.
4445 if pp.gFree.n >= 64 {
4451 for pp.gFree.n >= 32 {
4452 gp := pp.gFree.pop()
4454 if gp.stack.lo == 0 {
4461 lock(&sched.gFree.lock)
4462 sched.gFree.noStack.pushAll(noStackQ)
4463 sched.gFree.stack.pushAll(stackQ)
4464 sched.gFree.n += inc
4465 unlock(&sched.gFree.lock)
4469 // Get from gfree list.
4470 // If local list is empty, grab a batch from global list.
4471 func gfget(pp *p) *g {
4473 if pp.gFree.empty() && (!sched.gFree.stack.empty() || !sched.gFree.noStack.empty()) {
4474 lock(&sched.gFree.lock)
4475 // Move a batch of free Gs to the P.
4476 for pp.gFree.n < 32 {
4477 // Prefer Gs with stacks.
4478 gp := sched.gFree.stack.pop()
4480 gp = sched.gFree.noStack.pop()
4489 unlock(&sched.gFree.lock)
4492 gp := pp.gFree.pop()
4497 if gp.stack.lo != 0 && gp.stack.hi-gp.stack.lo != uintptr(startingStackSize) {
4498 // Deallocate old stack. We kept it in gfput because it was the
4499 // right size when the goroutine was put on the free list, but
4500 // the right size has changed since then.
4501 systemstack(func() {
4508 if gp.stack.lo == 0 {
4509 // Stack was deallocated in gfput or just above. Allocate a new one.
4510 systemstack(func() {
4511 gp.stack = stackalloc(startingStackSize)
4513 gp.stackguard0 = gp.stack.lo + stackGuard
4516 racemalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4519 msanmalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4522 asanunpoison(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
4528 // Purge all cached G's from gfree list to the global list.
4529 func gfpurge(pp *p) {
4535 for !pp.gFree.empty() {
4536 gp := pp.gFree.pop()
4538 if gp.stack.lo == 0 {
4545 lock(&sched.gFree.lock)
4546 sched.gFree.noStack.pushAll(noStackQ)
4547 sched.gFree.stack.pushAll(stackQ)
4548 sched.gFree.n += inc
4549 unlock(&sched.gFree.lock)
4552 // Breakpoint executes a breakpoint trap.
4557 // dolockOSThread is called by LockOSThread and lockOSThread below
4558 // after they modify m.locked. Do not allow preemption during this call,
4559 // or else the m might be different in this function than in the caller.
4562 func dolockOSThread() {
4563 if GOARCH == "wasm" {
4564 return // no threads on wasm yet
4567 gp.m.lockedg.set(gp)
4568 gp.lockedm.set(gp.m)
4571 // LockOSThread wires the calling goroutine to its current operating system thread.
4572 // The calling goroutine will always execute in that thread,
4573 // and no other goroutine will execute in it,
4574 // until the calling goroutine has made as many calls to
4575 // UnlockOSThread as to LockOSThread.
4576 // If the calling goroutine exits without unlocking the thread,
4577 // the thread will be terminated.
4579 // All init functions are run on the startup thread. Calling LockOSThread
4580 // from an init function will cause the main function to be invoked on
4583 // A goroutine should call LockOSThread before calling OS services or
4584 // non-Go library functions that depend on per-thread state.
4587 func LockOSThread() {
4588 if atomic.Load(&newmHandoff.haveTemplateThread) == 0 && GOOS != "plan9" {
4589 // If we need to start a new thread from the locked
4590 // thread, we need the template thread. Start it now
4591 // while we're in a known-good state.
4592 startTemplateThread()
4596 if gp.m.lockedExt == 0 {
4598 panic("LockOSThread nesting overflow")
4604 func lockOSThread() {
4605 getg().m.lockedInt++
4609 // dounlockOSThread is called by UnlockOSThread and unlockOSThread below
4610 // after they update m->locked. Do not allow preemption during this call,
4611 // or else the m might be in different in this function than in the caller.
4614 func dounlockOSThread() {
4615 if GOARCH == "wasm" {
4616 return // no threads on wasm yet
4619 if gp.m.lockedInt != 0 || gp.m.lockedExt != 0 {
4626 // UnlockOSThread undoes an earlier call to LockOSThread.
4627 // If this drops the number of active LockOSThread calls on the
4628 // calling goroutine to zero, it unwires the calling goroutine from
4629 // its fixed operating system thread.
4630 // If there are no active LockOSThread calls, this is a no-op.
4632 // Before calling UnlockOSThread, the caller must ensure that the OS
4633 // thread is suitable for running other goroutines. If the caller made
4634 // any permanent changes to the state of the thread that would affect
4635 // other goroutines, it should not call this function and thus leave
4636 // the goroutine locked to the OS thread until the goroutine (and
4637 // hence the thread) exits.
4640 func UnlockOSThread() {
4642 if gp.m.lockedExt == 0 {
4650 func unlockOSThread() {
4652 if gp.m.lockedInt == 0 {
4653 systemstack(badunlockosthread)
4659 func badunlockosthread() {
4660 throw("runtime: internal error: misuse of lockOSThread/unlockOSThread")
4663 func gcount() int32 {
4664 n := int32(atomic.Loaduintptr(&allglen)) - sched.gFree.n - sched.ngsys.Load()
4665 for _, pp := range allp {
4669 // All these variables can be changed concurrently, so the result can be inconsistent.
4670 // But at least the current goroutine is running.
4677 func mcount() int32 {
4678 return int32(sched.mnext - sched.nmfreed)
4682 signalLock atomic.Uint32
4684 // Must hold signalLock to write. Reads may be lock-free, but
4685 // signalLock should be taken to synchronize with changes.
4689 func _System() { _System() }
4690 func _ExternalCode() { _ExternalCode() }
4691 func _LostExternalCode() { _LostExternalCode() }
4692 func _GC() { _GC() }
4693 func _LostSIGPROFDuringAtomic64() { _LostSIGPROFDuringAtomic64() }
4694 func _VDSO() { _VDSO() }
4696 // Called if we receive a SIGPROF signal.
4697 // Called by the signal handler, may run during STW.
4699 //go:nowritebarrierrec
4700 func sigprof(pc, sp, lr uintptr, gp *g, mp *m) {
4701 if prof.hz.Load() == 0 {
4705 // If mp.profilehz is 0, then profiling is not enabled for this thread.
4706 // We must check this to avoid a deadlock between setcpuprofilerate
4707 // and the call to cpuprof.add, below.
4708 if mp != nil && mp.profilehz == 0 {
4712 // On mips{,le}/arm, 64bit atomics are emulated with spinlocks, in
4713 // runtime/internal/atomic. If SIGPROF arrives while the program is inside
4714 // the critical section, it creates a deadlock (when writing the sample).
4715 // As a workaround, create a counter of SIGPROFs while in critical section
4716 // to store the count, and pass it to sigprof.add() later when SIGPROF is
4717 // received from somewhere else (with _LostSIGPROFDuringAtomic64 as pc).
4718 if GOARCH == "mips" || GOARCH == "mipsle" || GOARCH == "arm" {
4719 if f := findfunc(pc); f.valid() {
4720 if hasPrefix(funcname(f), "runtime/internal/atomic") {
4721 cpuprof.lostAtomic++
4725 if GOARCH == "arm" && goarm < 7 && GOOS == "linux" && pc&0xffff0000 == 0xffff0000 {
4726 // runtime/internal/atomic functions call into kernel
4727 // helpers on arm < 7. See
4728 // runtime/internal/atomic/sys_linux_arm.s.
4729 cpuprof.lostAtomic++
4734 // Profiling runs concurrently with GC, so it must not allocate.
4735 // Set a trap in case the code does allocate.
4736 // Note that on windows, one thread takes profiles of all the
4737 // other threads, so mp is usually not getg().m.
4738 // In fact mp may not even be stopped.
4739 // See golang.org/issue/17165.
4740 getg().m.mallocing++
4743 var stk [maxCPUProfStack]uintptr
4745 if mp.ncgo > 0 && mp.curg != nil && mp.curg.syscallpc != 0 && mp.curg.syscallsp != 0 {
4747 // Check cgoCallersUse to make sure that we are not
4748 // interrupting other code that is fiddling with
4749 // cgoCallers. We are running in a signal handler
4750 // with all signals blocked, so we don't have to worry
4751 // about any other code interrupting us.
4752 if mp.cgoCallersUse.Load() == 0 && mp.cgoCallers != nil && mp.cgoCallers[0] != 0 {
4753 for cgoOff < len(mp.cgoCallers) && mp.cgoCallers[cgoOff] != 0 {
4756 n += copy(stk[:], mp.cgoCallers[:cgoOff])
4757 mp.cgoCallers[0] = 0
4760 // Collect Go stack that leads to the cgo call.
4761 u.initAt(mp.curg.syscallpc, mp.curg.syscallsp, 0, mp.curg, unwindSilentErrors)
4762 } else if usesLibcall() && mp.libcallg != 0 && mp.libcallpc != 0 && mp.libcallsp != 0 {
4763 // Libcall, i.e. runtime syscall on windows.
4764 // Collect Go stack that leads to the call.
4765 u.initAt(mp.libcallpc, mp.libcallsp, 0, mp.libcallg.ptr(), unwindSilentErrors)
4766 } else if mp != nil && mp.vdsoSP != 0 {
4767 // VDSO call, e.g. nanotime1 on Linux.
4768 // Collect Go stack that leads to the call.
4769 u.initAt(mp.vdsoPC, mp.vdsoSP, 0, gp, unwindSilentErrors|unwindJumpStack)
4771 u.initAt(pc, sp, lr, gp, unwindSilentErrors|unwindTrap|unwindJumpStack)
4773 n += tracebackPCs(&u, 0, stk[n:])
4776 // Normal traceback is impossible or has failed.
4777 // Account it against abstract "System" or "GC".
4780 pc = abi.FuncPCABIInternal(_VDSO) + sys.PCQuantum
4781 } else if pc > firstmoduledata.etext {
4782 // "ExternalCode" is better than "etext".
4783 pc = abi.FuncPCABIInternal(_ExternalCode) + sys.PCQuantum
4786 if mp.preemptoff != "" {
4787 stk[1] = abi.FuncPCABIInternal(_GC) + sys.PCQuantum
4789 stk[1] = abi.FuncPCABIInternal(_System) + sys.PCQuantum
4793 if prof.hz.Load() != 0 {
4794 // Note: it can happen on Windows that we interrupted a system thread
4795 // with no g, so gp could nil. The other nil checks are done out of
4796 // caution, but not expected to be nil in practice.
4797 var tagPtr *unsafe.Pointer
4798 if gp != nil && gp.m != nil && gp.m.curg != nil {
4799 tagPtr = &gp.m.curg.labels
4801 cpuprof.add(tagPtr, stk[:n])
4805 if gp != nil && gp.m != nil {
4806 if gp.m.curg != nil {
4811 traceCPUSample(gprof, pp, stk[:n])
4813 getg().m.mallocing--
4816 // setcpuprofilerate sets the CPU profiling rate to hz times per second.
4817 // If hz <= 0, setcpuprofilerate turns off CPU profiling.
4818 func setcpuprofilerate(hz int32) {
4819 // Force sane arguments.
4824 // Disable preemption, otherwise we can be rescheduled to another thread
4825 // that has profiling enabled.
4829 // Stop profiler on this thread so that it is safe to lock prof.
4830 // if a profiling signal came in while we had prof locked,
4831 // it would deadlock.
4832 setThreadCPUProfiler(0)
4834 for !prof.signalLock.CompareAndSwap(0, 1) {
4837 if prof.hz.Load() != hz {
4838 setProcessCPUProfiler(hz)
4841 prof.signalLock.Store(0)
4844 sched.profilehz = hz
4848 setThreadCPUProfiler(hz)
4854 // init initializes pp, which may be a freshly allocated p or a
4855 // previously destroyed p, and transitions it to status _Pgcstop.
4856 func (pp *p) init(id int32) {
4858 pp.status = _Pgcstop
4859 pp.sudogcache = pp.sudogbuf[:0]
4860 pp.deferpool = pp.deferpoolbuf[:0]
4862 if pp.mcache == nil {
4865 throw("missing mcache?")
4867 // Use the bootstrap mcache0. Only one P will get
4868 // mcache0: the one with ID 0.
4871 pp.mcache = allocmcache()
4874 if raceenabled && pp.raceprocctx == 0 {
4876 pp.raceprocctx = raceprocctx0
4877 raceprocctx0 = 0 // bootstrap
4879 pp.raceprocctx = raceproccreate()
4882 lockInit(&pp.timersLock, lockRankTimers)
4884 // This P may get timers when it starts running. Set the mask here
4885 // since the P may not go through pidleget (notably P 0 on startup).
4887 // Similarly, we may not go through pidleget before this P starts
4888 // running if it is P 0 on startup.
4892 // destroy releases all of the resources associated with pp and
4893 // transitions it to status _Pdead.
4895 // sched.lock must be held and the world must be stopped.
4896 func (pp *p) destroy() {
4897 assertLockHeld(&sched.lock)
4898 assertWorldStopped()
4900 // Move all runnable goroutines to the global queue
4901 for pp.runqhead != pp.runqtail {
4902 // Pop from tail of local queue
4904 gp := pp.runq[pp.runqtail%uint32(len(pp.runq))].ptr()
4905 // Push onto head of global queue
4908 if pp.runnext != 0 {
4909 globrunqputhead(pp.runnext.ptr())
4912 if len(pp.timers) > 0 {
4913 plocal := getg().m.p.ptr()
4914 // The world is stopped, but we acquire timersLock to
4915 // protect against sysmon calling timeSleepUntil.
4916 // This is the only case where we hold the timersLock of
4917 // more than one P, so there are no deadlock concerns.
4918 lock(&plocal.timersLock)
4919 lock(&pp.timersLock)
4920 moveTimers(plocal, pp.timers)
4922 pp.numTimers.Store(0)
4923 pp.deletedTimers.Store(0)
4924 pp.timer0When.Store(0)
4925 unlock(&pp.timersLock)
4926 unlock(&plocal.timersLock)
4928 // Flush p's write barrier buffer.
4929 if gcphase != _GCoff {
4933 for i := range pp.sudogbuf {
4934 pp.sudogbuf[i] = nil
4936 pp.sudogcache = pp.sudogbuf[:0]
4937 for j := range pp.deferpoolbuf {
4938 pp.deferpoolbuf[j] = nil
4940 pp.deferpool = pp.deferpoolbuf[:0]
4941 systemstack(func() {
4942 for i := 0; i < pp.mspancache.len; i++ {
4943 // Safe to call since the world is stopped.
4944 mheap_.spanalloc.free(unsafe.Pointer(pp.mspancache.buf[i]))
4946 pp.mspancache.len = 0
4948 pp.pcache.flush(&mheap_.pages)
4949 unlock(&mheap_.lock)
4951 freemcache(pp.mcache)
4956 if pp.timerRaceCtx != 0 {
4957 // The race detector code uses a callback to fetch
4958 // the proc context, so arrange for that callback
4959 // to see the right thing.
4960 // This hack only works because we are the only
4966 racectxend(pp.timerRaceCtx)
4971 raceprocdestroy(pp.raceprocctx)
4978 // Change number of processors.
4980 // sched.lock must be held, and the world must be stopped.
4982 // gcworkbufs must not be being modified by either the GC or the write barrier
4983 // code, so the GC must not be running if the number of Ps actually changes.
4985 // Returns list of Ps with local work, they need to be scheduled by the caller.
4986 func procresize(nprocs int32) *p {
4987 assertLockHeld(&sched.lock)
4988 assertWorldStopped()
4991 if old < 0 || nprocs <= 0 {
4992 throw("procresize: invalid arg")
4995 traceGomaxprocs(nprocs)
4998 // update statistics
5000 if sched.procresizetime != 0 {
5001 sched.totaltime += int64(old) * (now - sched.procresizetime)
5003 sched.procresizetime = now
5005 maskWords := (nprocs + 31) / 32
5007 // Grow allp if necessary.
5008 if nprocs > int32(len(allp)) {
5009 // Synchronize with retake, which could be running
5010 // concurrently since it doesn't run on a P.
5012 if nprocs <= int32(cap(allp)) {
5013 allp = allp[:nprocs]
5015 nallp := make([]*p, nprocs)
5016 // Copy everything up to allp's cap so we
5017 // never lose old allocated Ps.
5018 copy(nallp, allp[:cap(allp)])
5022 if maskWords <= int32(cap(idlepMask)) {
5023 idlepMask = idlepMask[:maskWords]
5024 timerpMask = timerpMask[:maskWords]
5026 nidlepMask := make([]uint32, maskWords)
5027 // No need to copy beyond len, old Ps are irrelevant.
5028 copy(nidlepMask, idlepMask)
5029 idlepMask = nidlepMask
5031 ntimerpMask := make([]uint32, maskWords)
5032 copy(ntimerpMask, timerpMask)
5033 timerpMask = ntimerpMask
5038 // initialize new P's
5039 for i := old; i < nprocs; i++ {
5045 atomicstorep(unsafe.Pointer(&allp[i]), unsafe.Pointer(pp))
5049 if gp.m.p != 0 && gp.m.p.ptr().id < nprocs {
5050 // continue to use the current P
5051 gp.m.p.ptr().status = _Prunning
5052 gp.m.p.ptr().mcache.prepareForSweep()
5054 // release the current P and acquire allp[0].
5056 // We must do this before destroying our current P
5057 // because p.destroy itself has write barriers, so we
5058 // need to do that from a valid P.
5061 // Pretend that we were descheduled
5062 // and then scheduled again to keep
5065 traceProcStop(gp.m.p.ptr())
5079 // g.m.p is now set, so we no longer need mcache0 for bootstrapping.
5082 // release resources from unused P's
5083 for i := nprocs; i < old; i++ {
5086 // can't free P itself because it can be referenced by an M in syscall
5090 if int32(len(allp)) != nprocs {
5092 allp = allp[:nprocs]
5093 idlepMask = idlepMask[:maskWords]
5094 timerpMask = timerpMask[:maskWords]
5099 for i := nprocs - 1; i >= 0; i-- {
5101 if gp.m.p.ptr() == pp {
5109 pp.link.set(runnablePs)
5113 stealOrder.reset(uint32(nprocs))
5114 var int32p *int32 = &gomaxprocs // make compiler check that gomaxprocs is an int32
5115 atomic.Store((*uint32)(unsafe.Pointer(int32p)), uint32(nprocs))
5117 // Notify the limiter that the amount of procs has changed.
5118 gcCPULimiter.resetCapacity(now, nprocs)
5123 // Associate p and the current m.
5125 // This function is allowed to have write barriers even if the caller
5126 // isn't because it immediately acquires pp.
5128 //go:yeswritebarrierrec
5129 func acquirep(pp *p) {
5130 // Do the part that isn't allowed to have write barriers.
5133 // Have p; write barriers now allowed.
5135 // Perform deferred mcache flush before this P can allocate
5136 // from a potentially stale mcache.
5137 pp.mcache.prepareForSweep()
5144 // wirep is the first step of acquirep, which actually associates the
5145 // current M to pp. This is broken out so we can disallow write
5146 // barriers for this part, since we don't yet have a P.
5148 //go:nowritebarrierrec
5154 throw("wirep: already in go")
5156 if pp.m != 0 || pp.status != _Pidle {
5161 print("wirep: p->m=", pp.m, "(", id, ") p->status=", pp.status, "\n")
5162 throw("wirep: invalid p state")
5166 pp.status = _Prunning
5169 // Disassociate p and the current m.
5170 func releasep() *p {
5174 throw("releasep: invalid arg")
5177 if pp.m.ptr() != gp.m || pp.status != _Prunning {
5178 print("releasep: m=", gp.m, " m->p=", gp.m.p.ptr(), " p->m=", hex(pp.m), " p->status=", pp.status, "\n")
5179 throw("releasep: invalid p state")
5182 traceProcStop(gp.m.p.ptr())
5190 func incidlelocked(v int32) {
5192 sched.nmidlelocked += v
5199 // Check for deadlock situation.
5200 // The check is based on number of running M's, if 0 -> deadlock.
5201 // sched.lock must be held.
5203 assertLockHeld(&sched.lock)
5205 // For -buildmode=c-shared or -buildmode=c-archive it's OK if
5206 // there are no running goroutines. The calling program is
5207 // assumed to be running.
5208 if islibrary || isarchive {
5212 // If we are dying because of a signal caught on an already idle thread,
5213 // freezetheworld will cause all running threads to block.
5214 // And runtime will essentially enter into deadlock state,
5215 // except that there is a thread that will call exit soon.
5216 if panicking.Load() > 0 {
5220 // If we are not running under cgo, but we have an extra M then account
5221 // for it. (It is possible to have an extra M on Windows without cgo to
5222 // accommodate callbacks created by syscall.NewCallback. See issue #6751
5225 if !iscgo && cgoHasExtraM && extraMLength.Load() > 0 {
5229 run := mcount() - sched.nmidle - sched.nmidlelocked - sched.nmsys
5234 print("runtime: checkdead: nmidle=", sched.nmidle, " nmidlelocked=", sched.nmidlelocked, " mcount=", mcount(), " nmsys=", sched.nmsys, "\n")
5236 throw("checkdead: inconsistent counts")
5240 forEachG(func(gp *g) {
5241 if isSystemGoroutine(gp, false) {
5244 s := readgstatus(gp)
5245 switch s &^ _Gscan {
5252 print("runtime: checkdead: find g ", gp.goid, " in status ", s, "\n")
5254 throw("checkdead: runnable g")
5257 if grunning == 0 { // possible if main goroutine calls runtime·Goexit()
5258 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5259 fatal("no goroutines (main called runtime.Goexit) - deadlock!")
5262 // Maybe jump time forward for playground.
5264 if when := timeSleepUntil(); when < maxWhen {
5267 // Start an M to steal the timer.
5268 pp, _ := pidleget(faketime)
5270 // There should always be a free P since
5271 // nothing is running.
5273 throw("checkdead: no p for timer")
5277 // There should always be a free M since
5278 // nothing is running.
5280 throw("checkdead: no m for timer")
5282 // M must be spinning to steal. We set this to be
5283 // explicit, but since this is the only M it would
5284 // become spinning on its own anyways.
5285 sched.nmspinning.Add(1)
5288 notewakeup(&mp.park)
5293 // There are no goroutines running, so we can look at the P's.
5294 for _, pp := range allp {
5295 if len(pp.timers) > 0 {
5300 unlock(&sched.lock) // unlock so that GODEBUG=scheddetail=1 doesn't hang
5301 fatal("all goroutines are asleep - deadlock!")
5304 // forcegcperiod is the maximum time in nanoseconds between garbage
5305 // collections. If we go this long without a garbage collection, one
5306 // is forced to run.
5308 // This is a variable for testing purposes. It normally doesn't change.
5309 var forcegcperiod int64 = 2 * 60 * 1e9
5311 // needSysmonWorkaround is true if the workaround for
5312 // golang.org/issue/42515 is needed on NetBSD.
5313 var needSysmonWorkaround bool = false
5315 // Always runs without a P, so write barriers are not allowed.
5317 //go:nowritebarrierrec
5324 lasttrace := int64(0)
5325 idle := 0 // how many cycles in succession we had not wokeup somebody
5329 if idle == 0 { // start with 20us sleep...
5331 } else if idle > 50 { // start doubling the sleep after 1ms...
5334 if delay > 10*1000 { // up to 10ms
5339 // sysmon should not enter deep sleep if schedtrace is enabled so that
5340 // it can print that information at the right time.
5342 // It should also not enter deep sleep if there are any active P's so
5343 // that it can retake P's from syscalls, preempt long running G's, and
5344 // poll the network if all P's are busy for long stretches.
5346 // It should wakeup from deep sleep if any P's become active either due
5347 // to exiting a syscall or waking up due to a timer expiring so that it
5348 // can resume performing those duties. If it wakes from a syscall it
5349 // resets idle and delay as a bet that since it had retaken a P from a
5350 // syscall before, it may need to do it again shortly after the
5351 // application starts work again. It does not reset idle when waking
5352 // from a timer to avoid adding system load to applications that spend
5353 // most of their time sleeping.
5355 if debug.schedtrace <= 0 && (sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs) {
5357 if sched.gcwaiting.Load() || sched.npidle.Load() == gomaxprocs {
5358 syscallWake := false
5359 next := timeSleepUntil()
5361 sched.sysmonwait.Store(true)
5363 // Make wake-up period small enough
5364 // for the sampling to be correct.
5365 sleep := forcegcperiod / 2
5366 if next-now < sleep {
5369 shouldRelax := sleep >= osRelaxMinNS
5373 syscallWake = notetsleep(&sched.sysmonnote, sleep)
5378 sched.sysmonwait.Store(false)
5379 noteclear(&sched.sysmonnote)
5389 lock(&sched.sysmonlock)
5390 // Update now in case we blocked on sysmonnote or spent a long time
5391 // blocked on schedlock or sysmonlock above.
5394 // trigger libc interceptors if needed
5395 if *cgo_yield != nil {
5396 asmcgocall(*cgo_yield, nil)
5398 // poll network if not polled for more than 10ms
5399 lastpoll := sched.lastpoll.Load()
5400 if netpollinited() && lastpoll != 0 && lastpoll+10*1000*1000 < now {
5401 sched.lastpoll.CompareAndSwap(lastpoll, now)
5402 list := netpoll(0) // non-blocking - returns list of goroutines
5404 // Need to decrement number of idle locked M's
5405 // (pretending that one more is running) before injectglist.
5406 // Otherwise it can lead to the following situation:
5407 // injectglist grabs all P's but before it starts M's to run the P's,
5408 // another M returns from syscall, finishes running its G,
5409 // observes that there is no work to do and no other running M's
5410 // and reports deadlock.
5416 if GOOS == "netbsd" && needSysmonWorkaround {
5417 // netpoll is responsible for waiting for timer
5418 // expiration, so we typically don't have to worry
5419 // about starting an M to service timers. (Note that
5420 // sleep for timeSleepUntil above simply ensures sysmon
5421 // starts running again when that timer expiration may
5422 // cause Go code to run again).
5424 // However, netbsd has a kernel bug that sometimes
5425 // misses netpollBreak wake-ups, which can lead to
5426 // unbounded delays servicing timers. If we detect this
5427 // overrun, then startm to get something to handle the
5430 // See issue 42515 and
5431 // https://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=50094.
5432 if next := timeSleepUntil(); next < now {
5433 startm(nil, false, false)
5436 if scavenger.sysmonWake.Load() != 0 {
5437 // Kick the scavenger awake if someone requested it.
5440 // retake P's blocked in syscalls
5441 // and preempt long running G's
5442 if retake(now) != 0 {
5447 // check if we need to force a GC
5448 if t := (gcTrigger{kind: gcTriggerTime, now: now}); t.test() && forcegc.idle.Load() {
5450 forcegc.idle.Store(false)
5452 list.push(forcegc.g)
5454 unlock(&forcegc.lock)
5456 if debug.schedtrace > 0 && lasttrace+int64(debug.schedtrace)*1000000 <= now {
5458 schedtrace(debug.scheddetail > 0)
5460 unlock(&sched.sysmonlock)
5464 type sysmontick struct {
5471 // forcePreemptNS is the time slice given to a G before it is
5473 const forcePreemptNS = 10 * 1000 * 1000 // 10ms
5475 func retake(now int64) uint32 {
5477 // Prevent allp slice changes. This lock will be completely
5478 // uncontended unless we're already stopping the world.
5480 // We can't use a range loop over allp because we may
5481 // temporarily drop the allpLock. Hence, we need to re-fetch
5482 // allp each time around the loop.
5483 for i := 0; i < len(allp); i++ {
5486 // This can happen if procresize has grown
5487 // allp but not yet created new Ps.
5490 pd := &pp.sysmontick
5493 if s == _Prunning || s == _Psyscall {
5494 // Preempt G if it's running for too long.
5495 t := int64(pp.schedtick)
5496 if int64(pd.schedtick) != t {
5497 pd.schedtick = uint32(t)
5499 } else if pd.schedwhen+forcePreemptNS <= now {
5501 // In case of syscall, preemptone() doesn't
5502 // work, because there is no M wired to P.
5507 // Retake P from syscall if it's there for more than 1 sysmon tick (at least 20us).
5508 t := int64(pp.syscalltick)
5509 if !sysretake && int64(pd.syscalltick) != t {
5510 pd.syscalltick = uint32(t)
5511 pd.syscallwhen = now
5514 // On the one hand we don't want to retake Ps if there is no other work to do,
5515 // but on the other hand we want to retake them eventually
5516 // because they can prevent the sysmon thread from deep sleep.
5517 if runqempty(pp) && sched.nmspinning.Load()+sched.npidle.Load() > 0 && pd.syscallwhen+10*1000*1000 > now {
5520 // Drop allpLock so we can take sched.lock.
5522 // Need to decrement number of idle locked M's
5523 // (pretending that one more is running) before the CAS.
5524 // Otherwise the M from which we retake can exit the syscall,
5525 // increment nmidle and report deadlock.
5527 if atomic.Cas(&pp.status, s, _Pidle) {
5544 // Tell all goroutines that they have been preempted and they should stop.
5545 // This function is purely best-effort. It can fail to inform a goroutine if a
5546 // processor just started running it.
5547 // No locks need to be held.
5548 // Returns true if preemption request was issued to at least one goroutine.
5549 func preemptall() bool {
5551 for _, pp := range allp {
5552 if pp.status != _Prunning {
5562 // Tell the goroutine running on processor P to stop.
5563 // This function is purely best-effort. It can incorrectly fail to inform the
5564 // goroutine. It can inform the wrong goroutine. Even if it informs the
5565 // correct goroutine, that goroutine might ignore the request if it is
5566 // simultaneously executing newstack.
5567 // No lock needs to be held.
5568 // Returns true if preemption request was issued.
5569 // The actual preemption will happen at some point in the future
5570 // and will be indicated by the gp->status no longer being
5572 func preemptone(pp *p) bool {
5574 if mp == nil || mp == getg().m {
5578 if gp == nil || gp == mp.g0 {
5584 // Every call in a goroutine checks for stack overflow by
5585 // comparing the current stack pointer to gp->stackguard0.
5586 // Setting gp->stackguard0 to StackPreempt folds
5587 // preemption into the normal stack overflow check.
5588 gp.stackguard0 = stackPreempt
5590 // Request an async preemption of this P.
5591 if preemptMSupported && debug.asyncpreemptoff == 0 {
5601 func schedtrace(detailed bool) {
5608 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)
5610 print(" gcwaiting=", sched.gcwaiting.Load(), " nmidlelocked=", sched.nmidlelocked, " stopwait=", sched.stopwait, " sysmonwait=", sched.sysmonwait.Load(), "\n")
5612 // We must be careful while reading data from P's, M's and G's.
5613 // Even if we hold schedlock, most data can be changed concurrently.
5614 // E.g. (p->m ? p->m->id : -1) can crash if p->m changes from non-nil to nil.
5615 for i, pp := range allp {
5617 h := atomic.Load(&pp.runqhead)
5618 t := atomic.Load(&pp.runqtail)
5620 print(" P", i, ": status=", pp.status, " schedtick=", pp.schedtick, " syscalltick=", pp.syscalltick, " m=")
5626 print(" runqsize=", t-h, " gfreecnt=", pp.gFree.n, " timerslen=", len(pp.timers), "\n")
5628 // In non-detailed mode format lengths of per-P run queues as:
5629 // [len1 len2 len3 len4]
5635 if i == len(allp)-1 {
5646 for mp := allm; mp != nil; mp = mp.alllink {
5648 print(" M", mp.id, ": p=")
5660 print(" mallocing=", mp.mallocing, " throwing=", mp.throwing, " preemptoff=", mp.preemptoff, " locks=", mp.locks, " dying=", mp.dying, " spinning=", mp.spinning, " blocked=", mp.blocked, " lockedg=")
5661 if lockedg := mp.lockedg.ptr(); lockedg != nil {
5669 forEachG(func(gp *g) {
5670 print(" G", gp.goid, ": status=", readgstatus(gp), "(", gp.waitreason.String(), ") m=")
5677 if lockedm := gp.lockedm.ptr(); lockedm != nil {
5687 // schedEnableUser enables or disables the scheduling of user
5690 // This does not stop already running user goroutines, so the caller
5691 // should first stop the world when disabling user goroutines.
5692 func schedEnableUser(enable bool) {
5694 if sched.disable.user == !enable {
5698 sched.disable.user = !enable
5700 n := sched.disable.n
5702 globrunqputbatch(&sched.disable.runnable, n)
5704 for ; n != 0 && sched.npidle.Load() != 0; n-- {
5705 startm(nil, false, false)
5712 // schedEnabled reports whether gp should be scheduled. It returns
5713 // false is scheduling of gp is disabled.
5715 // sched.lock must be held.
5716 func schedEnabled(gp *g) bool {
5717 assertLockHeld(&sched.lock)
5719 if sched.disable.user {
5720 return isSystemGoroutine(gp, true)
5725 // Put mp on midle list.
5726 // sched.lock must be held.
5727 // May run during STW, so write barriers are not allowed.
5729 //go:nowritebarrierrec
5731 assertLockHeld(&sched.lock)
5733 mp.schedlink = sched.midle
5739 // Try to get an m from midle list.
5740 // sched.lock must be held.
5741 // May run during STW, so write barriers are not allowed.
5743 //go:nowritebarrierrec
5745 assertLockHeld(&sched.lock)
5747 mp := sched.midle.ptr()
5749 sched.midle = mp.schedlink
5755 // Put gp on the global runnable queue.
5756 // sched.lock must be held.
5757 // May run during STW, so write barriers are not allowed.
5759 //go:nowritebarrierrec
5760 func globrunqput(gp *g) {
5761 assertLockHeld(&sched.lock)
5763 sched.runq.pushBack(gp)
5767 // Put gp at the head of the global runnable queue.
5768 // sched.lock must be held.
5769 // May run during STW, so write barriers are not allowed.
5771 //go:nowritebarrierrec
5772 func globrunqputhead(gp *g) {
5773 assertLockHeld(&sched.lock)
5779 // Put a batch of runnable goroutines on the global runnable queue.
5780 // This clears *batch.
5781 // sched.lock must be held.
5782 // May run during STW, so write barriers are not allowed.
5784 //go:nowritebarrierrec
5785 func globrunqputbatch(batch *gQueue, n int32) {
5786 assertLockHeld(&sched.lock)
5788 sched.runq.pushBackAll(*batch)
5793 // Try get a batch of G's from the global runnable queue.
5794 // sched.lock must be held.
5795 func globrunqget(pp *p, max int32) *g {
5796 assertLockHeld(&sched.lock)
5798 if sched.runqsize == 0 {
5802 n := sched.runqsize/gomaxprocs + 1
5803 if n > sched.runqsize {
5806 if max > 0 && n > max {
5809 if n > int32(len(pp.runq))/2 {
5810 n = int32(len(pp.runq)) / 2
5815 gp := sched.runq.pop()
5818 gp1 := sched.runq.pop()
5819 runqput(pp, gp1, false)
5824 // pMask is an atomic bitstring with one bit per P.
5827 // read returns true if P id's bit is set.
5828 func (p pMask) read(id uint32) bool {
5830 mask := uint32(1) << (id % 32)
5831 return (atomic.Load(&p[word]) & mask) != 0
5834 // set sets P id's bit.
5835 func (p pMask) set(id int32) {
5837 mask := uint32(1) << (id % 32)
5838 atomic.Or(&p[word], mask)
5841 // clear clears P id's bit.
5842 func (p pMask) clear(id int32) {
5844 mask := uint32(1) << (id % 32)
5845 atomic.And(&p[word], ^mask)
5848 // updateTimerPMask clears pp's timer mask if it has no timers on its heap.
5850 // Ideally, the timer mask would be kept immediately consistent on any timer
5851 // operations. Unfortunately, updating a shared global data structure in the
5852 // timer hot path adds too much overhead in applications frequently switching
5853 // between no timers and some timers.
5855 // As a compromise, the timer mask is updated only on pidleget / pidleput. A
5856 // running P (returned by pidleget) may add a timer at any time, so its mask
5857 // must be set. An idle P (passed to pidleput) cannot add new timers while
5858 // idle, so if it has no timers at that time, its mask may be cleared.
5860 // Thus, we get the following effects on timer-stealing in findrunnable:
5862 // - Idle Ps with no timers when they go idle are never checked in findrunnable
5863 // (for work- or timer-stealing; this is the ideal case).
5864 // - Running Ps must always be checked.
5865 // - Idle Ps whose timers are stolen must continue to be checked until they run
5866 // again, even after timer expiration.
5868 // When the P starts running again, the mask should be set, as a timer may be
5869 // added at any time.
5871 // TODO(prattmic): Additional targeted updates may improve the above cases.
5872 // e.g., updating the mask when stealing a timer.
5873 func updateTimerPMask(pp *p) {
5874 if pp.numTimers.Load() > 0 {
5878 // Looks like there are no timers, however another P may transiently
5879 // decrement numTimers when handling a timerModified timer in
5880 // checkTimers. We must take timersLock to serialize with these changes.
5881 lock(&pp.timersLock)
5882 if pp.numTimers.Load() == 0 {
5883 timerpMask.clear(pp.id)
5885 unlock(&pp.timersLock)
5888 // pidleput puts p on the _Pidle list. now must be a relatively recent call
5889 // to nanotime or zero. Returns now or the current time if now was zero.
5891 // This releases ownership of p. Once sched.lock is released it is no longer
5894 // sched.lock must be held.
5896 // May run during STW, so write barriers are not allowed.
5898 //go:nowritebarrierrec
5899 func pidleput(pp *p, now int64) int64 {
5900 assertLockHeld(&sched.lock)
5903 throw("pidleput: P has non-empty run queue")
5908 updateTimerPMask(pp) // clear if there are no timers.
5909 idlepMask.set(pp.id)
5910 pp.link = sched.pidle
5913 if !pp.limiterEvent.start(limiterEventIdle, now) {
5914 throw("must be able to track idle limiter event")
5919 // pidleget tries to get a p from the _Pidle list, acquiring ownership.
5921 // sched.lock must be held.
5923 // May run during STW, so write barriers are not allowed.
5925 //go:nowritebarrierrec
5926 func pidleget(now int64) (*p, int64) {
5927 assertLockHeld(&sched.lock)
5929 pp := sched.pidle.ptr()
5931 // Timer may get added at any time now.
5935 timerpMask.set(pp.id)
5936 idlepMask.clear(pp.id)
5937 sched.pidle = pp.link
5938 sched.npidle.Add(-1)
5939 pp.limiterEvent.stop(limiterEventIdle, now)
5944 // pidlegetSpinning tries to get a p from the _Pidle list, acquiring ownership.
5945 // This is called by spinning Ms (or callers than need a spinning M) that have
5946 // found work. If no P is available, this must synchronized with non-spinning
5947 // Ms that may be preparing to drop their P without discovering this work.
5949 // sched.lock must be held.
5951 // May run during STW, so write barriers are not allowed.
5953 //go:nowritebarrierrec
5954 func pidlegetSpinning(now int64) (*p, int64) {
5955 assertLockHeld(&sched.lock)
5957 pp, now := pidleget(now)
5959 // See "Delicate dance" comment in findrunnable. We found work
5960 // that we cannot take, we must synchronize with non-spinning
5961 // Ms that may be preparing to drop their P.
5962 sched.needspinning.Store(1)
5969 // runqempty reports whether pp has no Gs on its local run queue.
5970 // It never returns true spuriously.
5971 func runqempty(pp *p) bool {
5972 // Defend against a race where 1) pp has G1 in runqnext but runqhead == runqtail,
5973 // 2) runqput on pp kicks G1 to the runq, 3) runqget on pp empties runqnext.
5974 // Simply observing that runqhead == runqtail and then observing that runqnext == nil
5975 // does not mean the queue is empty.
5977 head := atomic.Load(&pp.runqhead)
5978 tail := atomic.Load(&pp.runqtail)
5979 runnext := atomic.Loaduintptr((*uintptr)(unsafe.Pointer(&pp.runnext)))
5980 if tail == atomic.Load(&pp.runqtail) {
5981 return head == tail && runnext == 0
5986 // To shake out latent assumptions about scheduling order,
5987 // we introduce some randomness into scheduling decisions
5988 // when running with the race detector.
5989 // The need for this was made obvious by changing the
5990 // (deterministic) scheduling order in Go 1.5 and breaking
5991 // many poorly-written tests.
5992 // With the randomness here, as long as the tests pass
5993 // consistently with -race, they shouldn't have latent scheduling
5995 const randomizeScheduler = raceenabled
5997 // runqput tries to put g on the local runnable queue.
5998 // If next is false, runqput adds g to the tail of the runnable queue.
5999 // If next is true, runqput puts g in the pp.runnext slot.
6000 // If the run queue is full, runnext puts g on the global queue.
6001 // Executed only by the owner P.
6002 func runqput(pp *p, gp *g, next bool) {
6003 if randomizeScheduler && next && fastrandn(2) == 0 {
6009 oldnext := pp.runnext
6010 if !pp.runnext.cas(oldnext, guintptr(unsafe.Pointer(gp))) {
6016 // Kick the old runnext out to the regular run queue.
6021 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
6023 if t-h < uint32(len(pp.runq)) {
6024 pp.runq[t%uint32(len(pp.runq))].set(gp)
6025 atomic.StoreRel(&pp.runqtail, t+1) // store-release, makes the item available for consumption
6028 if runqputslow(pp, gp, h, t) {
6031 // the queue is not full, now the put above must succeed
6035 // Put g and a batch of work from local runnable queue on global queue.
6036 // Executed only by the owner P.
6037 func runqputslow(pp *p, gp *g, h, t uint32) bool {
6038 var batch [len(pp.runq)/2 + 1]*g
6040 // First, grab a batch from local queue.
6043 if n != uint32(len(pp.runq)/2) {
6044 throw("runqputslow: queue is not full")
6046 for i := uint32(0); i < n; i++ {
6047 batch[i] = pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
6049 if !atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6054 if randomizeScheduler {
6055 for i := uint32(1); i <= n; i++ {
6056 j := fastrandn(i + 1)
6057 batch[i], batch[j] = batch[j], batch[i]
6061 // Link the goroutines.
6062 for i := uint32(0); i < n; i++ {
6063 batch[i].schedlink.set(batch[i+1])
6066 q.head.set(batch[0])
6067 q.tail.set(batch[n])
6069 // Now put the batch on global queue.
6071 globrunqputbatch(&q, int32(n+1))
6076 // runqputbatch tries to put all the G's on q on the local runnable queue.
6077 // If the queue is full, they are put on the global queue; in that case
6078 // this will temporarily acquire the scheduler lock.
6079 // Executed only by the owner P.
6080 func runqputbatch(pp *p, q *gQueue, qsize int) {
6081 h := atomic.LoadAcq(&pp.runqhead)
6084 for !q.empty() && t-h < uint32(len(pp.runq)) {
6086 pp.runq[t%uint32(len(pp.runq))].set(gp)
6092 if randomizeScheduler {
6093 off := func(o uint32) uint32 {
6094 return (pp.runqtail + o) % uint32(len(pp.runq))
6096 for i := uint32(1); i < n; i++ {
6097 j := fastrandn(i + 1)
6098 pp.runq[off(i)], pp.runq[off(j)] = pp.runq[off(j)], pp.runq[off(i)]
6102 atomic.StoreRel(&pp.runqtail, t)
6105 globrunqputbatch(q, int32(qsize))
6110 // Get g from local runnable queue.
6111 // If inheritTime is true, gp should inherit the remaining time in the
6112 // current time slice. Otherwise, it should start a new time slice.
6113 // Executed only by the owner P.
6114 func runqget(pp *p) (gp *g, inheritTime bool) {
6115 // If there's a runnext, it's the next G to run.
6117 // If the runnext is non-0 and the CAS fails, it could only have been stolen by another P,
6118 // because other Ps can race to set runnext to 0, but only the current P can set it to non-0.
6119 // Hence, there's no need to retry this CAS if it fails.
6120 if next != 0 && pp.runnext.cas(next, 0) {
6121 return next.ptr(), true
6125 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6130 gp := pp.runq[h%uint32(len(pp.runq))].ptr()
6131 if atomic.CasRel(&pp.runqhead, h, h+1) { // cas-release, commits consume
6137 // runqdrain drains the local runnable queue of pp and returns all goroutines in it.
6138 // Executed only by the owner P.
6139 func runqdrain(pp *p) (drainQ gQueue, n uint32) {
6140 oldNext := pp.runnext
6141 if oldNext != 0 && pp.runnext.cas(oldNext, 0) {
6142 drainQ.pushBack(oldNext.ptr())
6147 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6153 if qn > uint32(len(pp.runq)) { // read inconsistent h and t
6157 if !atomic.CasRel(&pp.runqhead, h, h+qn) { // cas-release, commits consume
6161 // We've inverted the order in which it gets G's from the local P's runnable queue
6162 // and then advances the head pointer because we don't want to mess up the statuses of G's
6163 // while runqdrain() and runqsteal() are running in parallel.
6164 // Thus we should advance the head pointer before draining the local P into a gQueue,
6165 // so that we can update any gp.schedlink only after we take the full ownership of G,
6166 // meanwhile, other P's can't access to all G's in local P's runnable queue and steal them.
6167 // See https://groups.google.com/g/golang-dev/c/0pTKxEKhHSc/m/6Q85QjdVBQAJ for more details.
6168 for i := uint32(0); i < qn; i++ {
6169 gp := pp.runq[(h+i)%uint32(len(pp.runq))].ptr()
6176 // Grabs a batch of goroutines from pp's runnable queue into batch.
6177 // Batch is a ring buffer starting at batchHead.
6178 // Returns number of grabbed goroutines.
6179 // Can be executed by any P.
6180 func runqgrab(pp *p, batch *[256]guintptr, batchHead uint32, stealRunNextG bool) uint32 {
6182 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with other consumers
6183 t := atomic.LoadAcq(&pp.runqtail) // load-acquire, synchronize with the producer
6188 // Try to steal from pp.runnext.
6189 if next := pp.runnext; next != 0 {
6190 if pp.status == _Prunning {
6191 // Sleep to ensure that pp isn't about to run the g
6192 // we are about to steal.
6193 // The important use case here is when the g running
6194 // on pp ready()s another g and then almost
6195 // immediately blocks. Instead of stealing runnext
6196 // in this window, back off to give pp a chance to
6197 // schedule runnext. This will avoid thrashing gs
6198 // between different Ps.
6199 // A sync chan send/recv takes ~50ns as of time of
6200 // writing, so 3us gives ~50x overshoot.
6201 if GOOS != "windows" && GOOS != "openbsd" && GOOS != "netbsd" {
6204 // On some platforms system timer granularity is
6205 // 1-15ms, which is way too much for this
6206 // optimization. So just yield.
6210 if !pp.runnext.cas(next, 0) {
6213 batch[batchHead%uint32(len(batch))] = next
6219 if n > uint32(len(pp.runq)/2) { // read inconsistent h and t
6222 for i := uint32(0); i < n; i++ {
6223 g := pp.runq[(h+i)%uint32(len(pp.runq))]
6224 batch[(batchHead+i)%uint32(len(batch))] = g
6226 if atomic.CasRel(&pp.runqhead, h, h+n) { // cas-release, commits consume
6232 // Steal half of elements from local runnable queue of p2
6233 // and put onto local runnable queue of p.
6234 // Returns one of the stolen elements (or nil if failed).
6235 func runqsteal(pp, p2 *p, stealRunNextG bool) *g {
6237 n := runqgrab(p2, &pp.runq, t, stealRunNextG)
6242 gp := pp.runq[(t+n)%uint32(len(pp.runq))].ptr()
6246 h := atomic.LoadAcq(&pp.runqhead) // load-acquire, synchronize with consumers
6247 if t-h+n >= uint32(len(pp.runq)) {
6248 throw("runqsteal: runq overflow")
6250 atomic.StoreRel(&pp.runqtail, t+n) // store-release, makes the item available for consumption
6254 // A gQueue is a dequeue of Gs linked through g.schedlink. A G can only
6255 // be on one gQueue or gList at a time.
6256 type gQueue struct {
6261 // empty reports whether q is empty.
6262 func (q *gQueue) empty() bool {
6266 // push adds gp to the head of q.
6267 func (q *gQueue) push(gp *g) {
6268 gp.schedlink = q.head
6275 // pushBack adds gp to the tail of q.
6276 func (q *gQueue) pushBack(gp *g) {
6279 q.tail.ptr().schedlink.set(gp)
6286 // pushBackAll adds all Gs in q2 to the tail of q. After this q2 must
6288 func (q *gQueue) pushBackAll(q2 gQueue) {
6292 q2.tail.ptr().schedlink = 0
6294 q.tail.ptr().schedlink = q2.head
6301 // pop removes and returns the head of queue q. It returns nil if
6303 func (q *gQueue) pop() *g {
6306 q.head = gp.schedlink
6314 // popList takes all Gs in q and returns them as a gList.
6315 func (q *gQueue) popList() gList {
6316 stack := gList{q.head}
6321 // A gList is a list of Gs linked through g.schedlink. A G can only be
6322 // on one gQueue or gList at a time.
6327 // empty reports whether l is empty.
6328 func (l *gList) empty() bool {
6332 // push adds gp to the head of l.
6333 func (l *gList) push(gp *g) {
6334 gp.schedlink = l.head
6338 // pushAll prepends all Gs in q to l.
6339 func (l *gList) pushAll(q gQueue) {
6341 q.tail.ptr().schedlink = l.head
6346 // pop removes and returns the head of l. If l is empty, it returns nil.
6347 func (l *gList) pop() *g {
6350 l.head = gp.schedlink
6355 //go:linkname setMaxThreads runtime/debug.setMaxThreads
6356 func setMaxThreads(in int) (out int) {
6358 out = int(sched.maxmcount)
6359 if in > 0x7fffffff { // MaxInt32
6360 sched.maxmcount = 0x7fffffff
6362 sched.maxmcount = int32(in)
6370 func procPin() int {
6375 return int(mp.p.ptr().id)
6384 //go:linkname sync_runtime_procPin sync.runtime_procPin
6386 func sync_runtime_procPin() int {
6390 //go:linkname sync_runtime_procUnpin sync.runtime_procUnpin
6392 func sync_runtime_procUnpin() {
6396 //go:linkname sync_atomic_runtime_procPin sync/atomic.runtime_procPin
6398 func sync_atomic_runtime_procPin() int {
6402 //go:linkname sync_atomic_runtime_procUnpin sync/atomic.runtime_procUnpin
6404 func sync_atomic_runtime_procUnpin() {
6408 // Active spinning for sync.Mutex.
6410 //go:linkname sync_runtime_canSpin sync.runtime_canSpin
6412 func sync_runtime_canSpin(i int) bool {
6413 // sync.Mutex is cooperative, so we are conservative with spinning.
6414 // Spin only few times and only if running on a multicore machine and
6415 // GOMAXPROCS>1 and there is at least one other running P and local runq is empty.
6416 // As opposed to runtime mutex we don't do passive spinning here,
6417 // because there can be work on global runq or on other Ps.
6418 if i >= active_spin || ncpu <= 1 || gomaxprocs <= sched.npidle.Load()+sched.nmspinning.Load()+1 {
6421 if p := getg().m.p.ptr(); !runqempty(p) {
6427 //go:linkname sync_runtime_doSpin sync.runtime_doSpin
6429 func sync_runtime_doSpin() {
6430 procyield(active_spin_cnt)
6433 var stealOrder randomOrder
6435 // randomOrder/randomEnum are helper types for randomized work stealing.
6436 // They allow to enumerate all Ps in different pseudo-random orders without repetitions.
6437 // The algorithm is based on the fact that if we have X such that X and GOMAXPROCS
6438 // are coprime, then a sequences of (i + X) % GOMAXPROCS gives the required enumeration.
6439 type randomOrder struct {
6444 type randomEnum struct {
6451 func (ord *randomOrder) reset(count uint32) {
6453 ord.coprimes = ord.coprimes[:0]
6454 for i := uint32(1); i <= count; i++ {
6455 if gcd(i, count) == 1 {
6456 ord.coprimes = append(ord.coprimes, i)
6461 func (ord *randomOrder) start(i uint32) randomEnum {
6465 inc: ord.coprimes[i/ord.count%uint32(len(ord.coprimes))],
6469 func (enum *randomEnum) done() bool {
6470 return enum.i == enum.count
6473 func (enum *randomEnum) next() {
6475 enum.pos = (enum.pos + enum.inc) % enum.count
6478 func (enum *randomEnum) position() uint32 {
6482 func gcd(a, b uint32) uint32 {
6489 // An initTask represents the set of initializations that need to be done for a package.
6490 // Keep in sync with ../../test/noinit.go:initTask
6491 type initTask struct {
6492 state uint32 // 0 = uninitialized, 1 = in progress, 2 = done
6494 // followed by nfns pcs, uintptr sized, one per init function to run
6497 // inittrace stores statistics for init functions which are
6498 // updated by malloc and newproc when active is true.
6499 var inittrace tracestat
6501 type tracestat struct {
6502 active bool // init tracing activation status
6503 id uint64 // init goroutine id
6504 allocs uint64 // heap allocations
6505 bytes uint64 // heap allocated bytes
6508 func doInit(ts []*initTask) {
6509 for _, t := range ts {
6514 func doInit1(t *initTask) {
6516 case 2: // fully initialized
6518 case 1: // initialization in progress
6519 throw("recursive call during initialization - linker skew")
6520 default: // not initialized yet
6521 t.state = 1 // initialization in progress
6528 if inittrace.active {
6530 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6535 // We should have pruned all of these in the linker.
6536 throw("inittask with no functions")
6539 firstFunc := add(unsafe.Pointer(t), 8)
6540 for i := uint32(0); i < t.nfns; i++ {
6541 p := add(firstFunc, uintptr(i)*goarch.PtrSize)
6542 f := *(*func())(unsafe.Pointer(&p))
6546 if inittrace.active {
6548 // Load stats non-atomically since tracinit is updated only by this init goroutine.
6551 f := *(*func())(unsafe.Pointer(&firstFunc))
6552 pkg := funcpkgpath(findfunc(abi.FuncPCABIInternal(f)))
6555 print("init ", pkg, " @")
6556 print(string(fmtNSAsMS(sbuf[:], uint64(start-runtimeInitTime))), " ms, ")
6557 print(string(fmtNSAsMS(sbuf[:], uint64(end-start))), " ms clock, ")
6558 print(string(itoa(sbuf[:], after.bytes-before.bytes)), " bytes, ")
6559 print(string(itoa(sbuf[:], after.allocs-before.allocs)), " allocs")
6563 t.state = 2 // initialization done