1 // Copyright 2009 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.
5 // Garbage collector (GC).
7 // The GC runs concurrently with mutator threads, is type accurate (aka precise), allows multiple
8 // GC thread to run in parallel. It is a concurrent mark and sweep that uses a write barrier. It is
9 // non-generational and non-compacting. Allocation is done using size segregated per P allocation
10 // areas to minimize fragmentation while eliminating locks in the common case.
12 // The algorithm decomposes into several steps.
13 // This is a high level description of the algorithm being used. For an overview of GC a good
14 // place to start is Richard Jones' gchandbook.org.
16 // The algorithm's intellectual heritage includes Dijkstra's on-the-fly algorithm, see
17 // Edsger W. Dijkstra, Leslie Lamport, A. J. Martin, C. S. Scholten, and E. F. M. Steffens. 1978.
18 // On-the-fly garbage collection: an exercise in cooperation. Commun. ACM 21, 11 (November 1978),
20 // For journal quality proofs that these steps are complete, correct, and terminate see
21 // Hudson, R., and Moss, J.E.B. Copying Garbage Collection without stopping the world.
22 // Concurrency and Computation: Practice and Experience 15(3-5), 2003.
24 // 1. GC performs sweep termination.
26 // a. Stop the world. This causes all Ps to reach a GC safe-point.
28 // b. Sweep any unswept spans. There will only be unswept spans if
29 // this GC cycle was forced before the expected time.
31 // 2. GC performs the mark phase.
33 // a. Prepare for the mark phase by setting gcphase to _GCmark
34 // (from _GCoff), enabling the write barrier, enabling mutator
35 // assists, and enqueueing root mark jobs. No objects may be
36 // scanned until all Ps have enabled the write barrier, which is
37 // accomplished using STW.
39 // b. Start the world. From this point, GC work is done by mark
40 // workers started by the scheduler and by assists performed as
41 // part of allocation. The write barrier shades both the
42 // overwritten pointer and the new pointer value for any pointer
43 // writes (see mbarrier.go for details). Newly allocated objects
44 // are immediately marked black.
46 // c. GC performs root marking jobs. This includes scanning all
47 // stacks, shading all globals, and shading any heap pointers in
48 // off-heap runtime data structures. Scanning a stack stops a
49 // goroutine, shades any pointers found on its stack, and then
50 // resumes the goroutine.
52 // d. GC drains the work queue of grey objects, scanning each grey
53 // object to black and shading all pointers found in the object
54 // (which in turn may add those pointers to the work queue).
56 // e. Because GC work is spread across local caches, GC uses a
57 // distributed termination algorithm to detect when there are no
58 // more root marking jobs or grey objects (see gcMarkDone). At this
59 // point, GC transitions to mark termination.
61 // 3. GC performs mark termination.
65 // b. Set gcphase to _GCmarktermination, and disable workers and
68 // c. Perform housekeeping like flushing mcaches.
70 // 4. GC performs the sweep phase.
72 // a. Prepare for the sweep phase by setting gcphase to _GCoff,
73 // setting up sweep state and disabling the write barrier.
75 // b. Start the world. From this point on, newly allocated objects
76 // are white, and allocating sweeps spans before use if necessary.
78 // c. GC does concurrent sweeping in the background and in response
79 // to allocation. See description below.
81 // 5. When sufficient allocation has taken place, replay the sequence
82 // starting with 1 above. See discussion of GC rate below.
86 // The sweep phase proceeds concurrently with normal program execution.
87 // The heap is swept span-by-span both lazily (when a goroutine needs another span)
88 // and concurrently in a background goroutine (this helps programs that are not CPU bound).
89 // At the end of STW mark termination all spans are marked as "needs sweeping".
91 // The background sweeper goroutine simply sweeps spans one-by-one.
93 // To avoid requesting more OS memory while there are unswept spans, when a
94 // goroutine needs another span, it first attempts to reclaim that much memory
95 // by sweeping. When a goroutine needs to allocate a new small-object span, it
96 // sweeps small-object spans for the same object size until it frees at least
97 // one object. When a goroutine needs to allocate large-object span from heap,
98 // it sweeps spans until it frees at least that many pages into heap. There is
99 // one case where this may not suffice: if a goroutine sweeps and frees two
100 // nonadjacent one-page spans to the heap, it will allocate a new two-page
101 // span, but there can still be other one-page unswept spans which could be
102 // combined into a two-page span.
104 // It's critical to ensure that no operations proceed on unswept spans (that would corrupt
105 // mark bits in GC bitmap). During GC all mcaches are flushed into the central cache,
106 // so they are empty. When a goroutine grabs a new span into mcache, it sweeps it.
107 // When a goroutine explicitly frees an object or sets a finalizer, it ensures that
108 // the span is swept (either by sweeping it, or by waiting for the concurrent sweep to finish).
109 // The finalizer goroutine is kicked off only when all spans are swept.
110 // When the next GC starts, it sweeps all not-yet-swept spans (if any).
113 // Next GC is after we've allocated an extra amount of memory proportional to
114 // the amount already in use. The proportion is controlled by GOGC environment variable
115 // (100 by default). If GOGC=100 and we're using 4M, we'll GC again when we get to 8M
116 // (this mark is computed by the gcController.heapGoal method). This keeps the GC cost in
117 // linear proportion to the allocation cost. Adjusting GOGC just changes the linear constant
118 // (and also the amount of extra memory used).
122 // In order to prevent long pauses while scanning large objects and to
123 // improve parallelism, the garbage collector breaks up scan jobs for
124 // objects larger than maxObletBytes into "oblets" of at most
125 // maxObletBytes. When scanning encounters the beginning of a large
126 // object, it scans only the first oblet and enqueues the remaining
127 // oblets as new scan jobs.
133 "runtime/internal/atomic"
139 _FinBlockSize = 4 * 1024
141 // concurrentSweep is a debug flag. Disabling this flag
142 // ensures all spans are swept while the world is stopped.
143 concurrentSweep = true
145 // debugScanConservative enables debug logging for stack
146 // frames that are scanned conservatively.
147 debugScanConservative = false
149 // sweepMinHeapDistance is a lower bound on the heap distance
150 // (in bytes) reserved for concurrent sweeping between GC
152 sweepMinHeapDistance = 1024 * 1024
155 // heapObjectsCanMove always returns false in the current garbage collector.
156 // It exists for go4.org/unsafe/assume-no-moving-gc, which is an
157 // unfortunate idea that had an even more unfortunate implementation.
158 // Every time a new Go release happened, the package stopped building,
159 // and the authors had to add a new file with a new //go:build line, and
160 // then the entire ecosystem of packages with that as a dependency had to
161 // explicitly update to the new version. Many packages depend on
162 // assume-no-moving-gc transitively, through paths like
163 // inet.af/netaddr -> go4.org/intern -> assume-no-moving-gc.
164 // This was causing a significant amount of friction around each new
165 // release, so we added this bool for the package to //go:linkname
166 // instead. The bool is still unfortunate, but it's not as bad as
167 // breaking the ecosystem on every new release.
169 // If the Go garbage collector ever does move heap objects, we can set
170 // this to true to break all the programs using assume-no-moving-gc.
172 //go:linkname heapObjectsCanMove
173 func heapObjectsCanMove() bool {
178 if unsafe.Sizeof(workbuf{}) != _WorkbufSize {
179 throw("size of Workbuf is suboptimal")
181 // No sweep on the first cycle.
182 sweep.active.state.Store(sweepDrainedMask)
184 // Initialize GC pacer state.
185 // Use the environment variable GOGC for the initial gcPercent value.
186 // Use the environment variable GOMEMLIMIT for the initial memoryLimit value.
187 gcController.init(readGOGC(), readGOMEMLIMIT())
190 work.markDoneSema = 1
191 lockInit(&work.sweepWaiters.lock, lockRankSweepWaiters)
192 lockInit(&work.assistQueue.lock, lockRankAssistQueue)
193 lockInit(&work.wbufSpans.lock, lockRankWbufSpans)
196 // gcenable is called after the bulk of the runtime initialization,
197 // just before we're about to start letting user code run.
198 // It kicks off the background sweeper goroutine, the background
199 // scavenger goroutine, and enables GC.
201 // Kick off sweeping and scavenging.
202 c := make(chan int, 2)
207 memstats.enablegc = true // now that runtime is initialized, GC is okay
210 // Garbage collector phase.
211 // Indicates to write barrier and synchronization task to perform.
214 // The compiler knows about this variable.
215 // If you change it, you must change builtin/runtime.go, too.
216 // If you change the first four bytes, you must also change the write
217 // barrier insertion code.
218 var writeBarrier struct {
219 enabled bool // compiler emits a check of this before calling write barrier
220 pad [3]byte // compiler uses 32-bit load for "enabled" field
221 alignme uint64 // guarantee alignment so that compiler can use a 32 or 64-bit load
224 // gcBlackenEnabled is 1 if mutator assists and background mark
225 // workers are allowed to blacken objects. This must only be set when
226 // gcphase == _GCmark.
227 var gcBlackenEnabled uint32
230 _GCoff = iota // GC not running; sweeping in background, write barrier disabled
231 _GCmark // GC marking roots and workbufs: allocate black, write barrier ENABLED
232 _GCmarktermination // GC mark termination: allocate black, P's help GC, write barrier ENABLED
236 func setGCPhase(x uint32) {
237 atomic.Store(&gcphase, x)
238 writeBarrier.enabled = gcphase == _GCmark || gcphase == _GCmarktermination
241 // gcMarkWorkerMode represents the mode that a concurrent mark worker
242 // should operate in.
244 // Concurrent marking happens through four different mechanisms. One
245 // is mutator assists, which happen in response to allocations and are
246 // not scheduled. The other three are variations in the per-P mark
247 // workers and are distinguished by gcMarkWorkerMode.
248 type gcMarkWorkerMode int
251 // gcMarkWorkerNotWorker indicates that the next scheduled G is not
252 // starting work and the mode should be ignored.
253 gcMarkWorkerNotWorker gcMarkWorkerMode = iota
255 // gcMarkWorkerDedicatedMode indicates that the P of a mark
256 // worker is dedicated to running that mark worker. The mark
257 // worker should run without preemption.
258 gcMarkWorkerDedicatedMode
260 // gcMarkWorkerFractionalMode indicates that a P is currently
261 // running the "fractional" mark worker. The fractional worker
262 // is necessary when GOMAXPROCS*gcBackgroundUtilization is not
263 // an integer and using only dedicated workers would result in
264 // utilization too far from the target of gcBackgroundUtilization.
265 // The fractional worker should run until it is preempted and
266 // will be scheduled to pick up the fractional part of
267 // GOMAXPROCS*gcBackgroundUtilization.
268 gcMarkWorkerFractionalMode
270 // gcMarkWorkerIdleMode indicates that a P is running the mark
271 // worker because it has nothing else to do. The idle worker
272 // should run until it is preempted and account its time
273 // against gcController.idleMarkTime.
277 // gcMarkWorkerModeStrings are the strings labels of gcMarkWorkerModes
278 // to use in execution traces.
279 var gcMarkWorkerModeStrings = [...]string{
286 // pollFractionalWorkerExit reports whether a fractional mark worker
287 // should self-preempt. It assumes it is called from the fractional
289 func pollFractionalWorkerExit() bool {
290 // This should be kept in sync with the fractional worker
291 // scheduler logic in findRunnableGCWorker.
293 delta := now - gcController.markStartTime
297 p := getg().m.p.ptr()
298 selfTime := p.gcFractionalMarkTime + (now - p.gcMarkWorkerStartTime)
299 // Add some slack to the utilization goal so that the
300 // fractional worker isn't behind again the instant it exits.
301 return float64(selfTime)/float64(delta) > 1.2*gcController.fractionalUtilizationGoal
306 type workType struct {
307 full lfstack // lock-free list of full blocks workbuf
308 _ cpu.CacheLinePad // prevents false-sharing between full and empty
309 empty lfstack // lock-free list of empty blocks workbuf
310 _ cpu.CacheLinePad // prevents false-sharing between empty and nproc/nwait
314 // free is a list of spans dedicated to workbufs, but
315 // that don't currently contain any workbufs.
317 // busy is a list of all spans containing workbufs on
318 // one of the workbuf lists.
322 // Restore 64-bit alignment on 32-bit.
325 // bytesMarked is the number of bytes marked this cycle. This
326 // includes bytes blackened in scanned objects, noscan objects
327 // that go straight to black, and permagrey objects scanned by
328 // markroot during the concurrent scan phase. This is updated
329 // atomically during the cycle. Updates may be batched
330 // arbitrarily, since the value is only read at the end of the
333 // Because of benign races during marking, this number may not
334 // be the exact number of marked bytes, but it should be very
337 // Put this field here because it needs 64-bit atomic access
338 // (and thus 8-byte alignment even on 32-bit architectures).
341 markrootNext uint32 // next markroot job
342 markrootJobs uint32 // number of markroot jobs
348 // Number of roots of various root types. Set by gcMarkRootPrepare.
350 // nStackRoots == len(stackRoots), but we have nStackRoots for
352 nDataRoots, nBSSRoots, nSpanRoots, nStackRoots int
354 // Base indexes of each root type. Set by gcMarkRootPrepare.
355 baseData, baseBSS, baseSpans, baseStacks, baseEnd uint32
357 // stackRoots is a snapshot of all of the Gs that existed
358 // before the beginning of concurrent marking. The backing
359 // store of this must not be modified because it might be
360 // shared with allgs.
363 // Each type of GC state transition is protected by a lock.
364 // Since multiple threads can simultaneously detect the state
365 // transition condition, any thread that detects a transition
366 // condition must acquire the appropriate transition lock,
367 // re-check the transition condition and return if it no
368 // longer holds or perform the transition if it does.
369 // Likewise, any transition must invalidate the transition
370 // condition before releasing the lock. This ensures that each
371 // transition is performed by exactly one thread and threads
372 // that need the transition to happen block until it has
375 // startSema protects the transition from "off" to mark or
378 // markDoneSema protects transitions from mark to mark termination.
381 bgMarkReady note // signal background mark worker has started
382 bgMarkDone uint32 // cas to 1 when at a background mark completion point
383 // Background mark completion signaling
385 // mode is the concurrency mode of the current GC cycle.
388 // userForced indicates the current GC cycle was forced by an
389 // explicit user call.
392 // initialHeapLive is the value of gcController.heapLive at the
393 // beginning of this GC cycle.
394 initialHeapLive uint64
396 // assistQueue is a queue of assists that are blocked because
397 // there was neither enough credit to steal or enough work to
404 // sweepWaiters is a list of blocked goroutines to wake when
405 // we transition from mark termination to sweep.
406 sweepWaiters struct {
411 // cycles is the number of completed GC cycles, where a GC
412 // cycle is sweep termination, mark, mark termination, and
413 // sweep. This differs from memstats.numgc, which is
414 // incremented at mark termination.
417 // Timing/utilization stats for this cycle.
418 stwprocs, maxprocs int32
419 tSweepTerm, tMark, tMarkTerm, tEnd int64 // nanotime() of phase start
421 pauseNS int64 // total STW time this cycle
423 // debug.gctrace heap sizes for this cycle.
424 heap0, heap1, heap2 uint64
426 // Cumulative estimated CPU usage.
430 // GC runs a garbage collection and blocks the caller until the
431 // garbage collection is complete. It may also block the entire
434 // We consider a cycle to be: sweep termination, mark, mark
435 // termination, and sweep. This function shouldn't return
436 // until a full cycle has been completed, from beginning to
437 // end. Hence, we always want to finish up the current cycle
438 // and start a new one. That means:
440 // 1. In sweep termination, mark, or mark termination of cycle
441 // N, wait until mark termination N completes and transitions
444 // 2. In sweep N, help with sweep N.
446 // At this point we can begin a full cycle N+1.
448 // 3. Trigger cycle N+1 by starting sweep termination N+1.
450 // 4. Wait for mark termination N+1 to complete.
452 // 5. Help with sweep N+1 until it's done.
454 // This all has to be written to deal with the fact that the
455 // GC may move ahead on its own. For example, when we block
456 // until mark termination N, we may wake up in cycle N+2.
458 // Wait until the current sweep termination, mark, and mark
459 // termination complete.
460 n := work.cycles.Load()
463 // We're now in sweep N or later. Trigger GC cycle N+1, which
464 // will first finish sweep N if necessary and then enter sweep
466 gcStart(gcTrigger{kind: gcTriggerCycle, n: n + 1})
468 // Wait for mark termination N+1 to complete.
471 // Finish sweep N+1 before returning. We do this both to
472 // complete the cycle and because runtime.GC() is often used
473 // as part of tests and benchmarks to get the system into a
474 // relatively stable and isolated state.
475 for work.cycles.Load() == n+1 && sweepone() != ^uintptr(0) {
479 // Callers may assume that the heap profile reflects the
480 // just-completed cycle when this returns (historically this
481 // happened because this was a STW GC), but right now the
482 // profile still reflects mark termination N, not N+1.
484 // As soon as all of the sweep frees from cycle N+1 are done,
485 // we can go ahead and publish the heap profile.
487 // First, wait for sweeping to finish. (We know there are no
488 // more spans on the sweep queue, but we may be concurrently
489 // sweeping spans, so we have to wait.)
490 for work.cycles.Load() == n+1 && !isSweepDone() {
494 // Now we're really done with sweeping, so we can publish the
495 // stable heap profile. Only do this if we haven't already hit
496 // another mark termination.
498 cycle := work.cycles.Load()
499 if cycle == n+1 || (gcphase == _GCmark && cycle == n+2) {
505 // gcWaitOnMark blocks until GC finishes the Nth mark phase. If GC has
506 // already completed this mark phase, it returns immediately.
507 func gcWaitOnMark(n uint32) {
509 // Disable phase transitions.
510 lock(&work.sweepWaiters.lock)
511 nMarks := work.cycles.Load()
512 if gcphase != _GCmark {
513 // We've already completed this cycle's mark.
518 unlock(&work.sweepWaiters.lock)
522 // Wait until sweep termination, mark, and mark
523 // termination of cycle N complete.
524 work.sweepWaiters.list.push(getg())
525 goparkunlock(&work.sweepWaiters.lock, waitReasonWaitForGCCycle, traceBlockUntilGCEnds, 1)
529 // gcMode indicates how concurrent a GC cycle should be.
533 gcBackgroundMode gcMode = iota // concurrent GC and sweep
534 gcForceMode // stop-the-world GC now, concurrent sweep
535 gcForceBlockMode // stop-the-world GC now and STW sweep (forced by user)
538 // A gcTrigger is a predicate for starting a GC cycle. Specifically,
539 // it is an exit condition for the _GCoff phase.
540 type gcTrigger struct {
542 now int64 // gcTriggerTime: current time
543 n uint32 // gcTriggerCycle: cycle number to start
546 type gcTriggerKind int
549 // gcTriggerHeap indicates that a cycle should be started when
550 // the heap size reaches the trigger heap size computed by the
552 gcTriggerHeap gcTriggerKind = iota
554 // gcTriggerTime indicates that a cycle should be started when
555 // it's been more than forcegcperiod nanoseconds since the
556 // previous GC cycle.
559 // gcTriggerCycle indicates that a cycle should be started if
560 // we have not yet started cycle number gcTrigger.n (relative
565 // test reports whether the trigger condition is satisfied, meaning
566 // that the exit condition for the _GCoff phase has been met. The exit
567 // condition should be tested when allocating.
568 func (t gcTrigger) test() bool {
569 if !memstats.enablegc || panicking.Load() != 0 || gcphase != _GCoff {
574 trigger, _ := gcController.trigger()
575 return gcController.heapLive.Load() >= trigger
577 if gcController.gcPercent.Load() < 0 {
580 lastgc := int64(atomic.Load64(&memstats.last_gc_nanotime))
581 return lastgc != 0 && t.now-lastgc > forcegcperiod
583 // t.n > work.cycles, but accounting for wraparound.
584 return int32(t.n-work.cycles.Load()) > 0
589 // gcStart starts the GC. It transitions from _GCoff to _GCmark (if
590 // debug.gcstoptheworld == 0) or performs all of GC (if
591 // debug.gcstoptheworld != 0).
593 // This may return without performing this transition in some cases,
594 // such as when called on a system stack or with locks held.
595 func gcStart(trigger gcTrigger) {
596 // Since this is called from malloc and malloc is called in
597 // the guts of a number of libraries that might be holding
598 // locks, don't attempt to start GC in non-preemptible or
599 // potentially unstable situations.
601 if gp := getg(); gp == mp.g0 || mp.locks > 1 || mp.preemptoff != "" {
608 // Pick up the remaining unswept/not being swept spans concurrently
610 // This shouldn't happen if we're being invoked in background
611 // mode since proportional sweep should have just finished
612 // sweeping everything, but rounding errors, etc, may leave a
613 // few spans unswept. In forced mode, this is necessary since
614 // GC can be forced at any point in the sweeping cycle.
616 // We check the transition condition continuously here in case
617 // this G gets delayed in to the next GC cycle.
618 for trigger.test() && sweepone() != ^uintptr(0) {
621 // Perform GC initialization and the sweep termination
623 semacquire(&work.startSema)
624 // Re-check transition condition under transition lock.
626 semrelease(&work.startSema)
630 // In gcstoptheworld debug mode, upgrade the mode accordingly.
631 // We do this after re-checking the transition condition so
632 // that multiple goroutines that detect the heap trigger don't
633 // start multiple STW GCs.
634 mode := gcBackgroundMode
635 if debug.gcstoptheworld == 1 {
637 } else if debug.gcstoptheworld == 2 {
638 mode = gcForceBlockMode
641 // Ok, we're doing it! Stop everybody else
643 semacquire(&worldsema)
645 // For stats, check if this GC was forced by the user.
646 // Update it under gcsema to avoid gctrace getting wrong values.
647 work.userForced = trigger.kind == gcTriggerCycle
649 trace := traceAcquire()
655 // Check that all Ps have finished deferred mcache flushes.
656 for _, p := range allp {
657 if fg := p.mcache.flushGen.Load(); fg != mheap_.sweepgen {
658 println("runtime: p", p.id, "flushGen", fg, "!= sweepgen", mheap_.sweepgen)
659 throw("p mcache not flushed")
663 gcBgMarkStartWorkers()
665 systemstack(gcResetMarkState)
667 work.stwprocs, work.maxprocs = gomaxprocs, gomaxprocs
668 if work.stwprocs > ncpu {
669 // This is used to compute CPU time of the STW phases,
670 // so it can't be more than ncpu, even if GOMAXPROCS is.
673 work.heap0 = gcController.heapLive.Load()
678 work.tSweepTerm = now
680 systemstack(func() { stopTheWorldWithSema(stwGCSweepTerm) })
681 // Finish sweep before we start concurrent scan.
686 // clearpools before we start the GC. If we wait the memory will not be
687 // reclaimed until the next GC cycle.
692 // Assists and workers can start the moment we start
694 gcController.startCycle(now, int(gomaxprocs), trigger)
696 // Notify the CPU limiter that assists may begin.
697 gcCPULimiter.startGCTransition(true, now)
699 // In STW mode, disable scheduling of user Gs. This may also
700 // disable scheduling of this goroutine, so it may block as
701 // soon as we start the world again.
702 if mode != gcBackgroundMode {
703 schedEnableUser(false)
706 // Enter concurrent mark phase and enable
709 // Because the world is stopped, all Ps will
710 // observe that write barriers are enabled by
711 // the time we start the world and begin
714 // Write barriers must be enabled before assists are
715 // enabled because they must be enabled before
716 // any non-leaf heap objects are marked. Since
717 // allocations are blocked until assists can
718 // happen, we want to enable assists as early as
722 gcBgMarkPrepare() // Must happen before assists are enabled.
725 // Mark all active tinyalloc blocks. Since we're
726 // allocating from these, they need to be black like
727 // other allocations. The alternative is to blacken
728 // the tiny block on every allocation from it, which
729 // would slow down the tiny allocator.
732 // At this point all Ps have enabled the write
733 // barrier, thus maintaining the no white to
734 // black invariant. Enable mutator assists to
735 // put back-pressure on fast allocating
737 atomic.Store(&gcBlackenEnabled, 1)
739 // In STW mode, we could block the instant systemstack
740 // returns, so make sure we're not preemptible.
745 now = startTheWorldWithSema()
746 work.pauseNS += now - pauseStart
748 memstats.gcPauseDist.record(now - pauseStart)
750 sweepTermCpu := int64(work.stwprocs) * (work.tMark - work.tSweepTerm)
751 work.cpuStats.gcPauseTime += sweepTermCpu
752 work.cpuStats.gcTotalTime += sweepTermCpu
754 // Release the CPU limiter.
755 gcCPULimiter.finishGCTransition(now)
758 // Release the world sema before Gosched() in STW mode
759 // because we will need to reacquire it later but before
760 // this goroutine becomes runnable again, and we could
761 // self-deadlock otherwise.
762 semrelease(&worldsema)
765 // Make sure we block instead of returning to user code
767 if mode != gcBackgroundMode {
771 semrelease(&work.startSema)
774 // gcMarkDoneFlushed counts the number of P's with flushed work.
776 // Ideally this would be a captured local in gcMarkDone, but forEachP
777 // escapes its callback closure, so it can't capture anything.
779 // This is protected by markDoneSema.
780 var gcMarkDoneFlushed uint32
782 // gcMarkDone transitions the GC from mark to mark termination if all
783 // reachable objects have been marked (that is, there are no grey
784 // objects and can be no more in the future). Otherwise, it flushes
785 // all local work to the global queues where it can be discovered by
788 // This should be called when all local mark work has been drained and
789 // there are no remaining workers. Specifically, when
791 // work.nwait == work.nproc && !gcMarkWorkAvailable(p)
793 // The calling context must be preemptible.
795 // Flushing local work is important because idle Ps may have local
796 // work queued. This is the only way to make that work visible and
797 // drive GC to completion.
799 // It is explicitly okay to have write barriers in this function. If
800 // it does transition to mark termination, then all reachable objects
801 // have been marked, so the write barrier cannot shade any more
804 // Ensure only one thread is running the ragged barrier at a
806 semacquire(&work.markDoneSema)
809 // Re-check transition condition under transition lock.
811 // It's critical that this checks the global work queues are
812 // empty before performing the ragged barrier. Otherwise,
813 // there could be global work that a P could take after the P
814 // has passed the ragged barrier.
815 if !(gcphase == _GCmark && work.nwait == work.nproc && !gcMarkWorkAvailable(nil)) {
816 semrelease(&work.markDoneSema)
820 // forEachP needs worldsema to execute, and we'll need it to
821 // stop the world later, so acquire worldsema now.
822 semacquire(&worldsema)
824 // Flush all local buffers and collect flushedWork flags.
825 gcMarkDoneFlushed = 0
826 forEachP(waitReasonGCMarkTermination, func(pp *p) {
827 // Flush the write barrier buffer, since this may add
828 // work to the gcWork.
831 // Flush the gcWork, since this may create global work
832 // and set the flushedWork flag.
834 // TODO(austin): Break up these workbufs to
835 // better distribute work.
837 // Collect the flushedWork flag.
838 if pp.gcw.flushedWork {
839 atomic.Xadd(&gcMarkDoneFlushed, 1)
840 pp.gcw.flushedWork = false
844 if gcMarkDoneFlushed != 0 {
845 // More grey objects were discovered since the
846 // previous termination check, so there may be more
847 // work to do. Keep going. It's possible the
848 // transition condition became true again during the
849 // ragged barrier, so re-check it.
850 semrelease(&worldsema)
854 // There was no global work, no local work, and no Ps
855 // communicated work since we took markDoneSema. Therefore
856 // there are no grey objects and no more objects can be
857 // shaded. Transition to mark termination.
861 getg().m.preemptoff = "gcing"
862 systemstack(func() { stopTheWorldWithSema(stwGCMarkTerm) })
863 // The gcphase is _GCmark, it will transition to _GCmarktermination
864 // below. The important thing is that the wb remains active until
865 // all marking is complete. This includes writes made by the GC.
867 // There is sometimes work left over when we enter mark termination due
868 // to write barriers performed after the completion barrier above.
869 // Detect this and resume concurrent mark. This is obviously
872 // See issue #27993 for details.
874 // Switch to the system stack to call wbBufFlush1, though in this case
875 // it doesn't matter because we're non-preemptible anyway.
878 for _, p := range allp {
887 getg().m.preemptoff = ""
889 now := startTheWorldWithSema()
890 work.pauseNS += now - pauseStart
891 memstats.gcPauseDist.record(now - pauseStart)
893 semrelease(&worldsema)
897 gcComputeStartingStackSize()
899 // Disable assists and background workers. We must do
900 // this before waking blocked assists.
901 atomic.Store(&gcBlackenEnabled, 0)
903 // Notify the CPU limiter that GC assists will now cease.
904 gcCPULimiter.startGCTransition(false, now)
906 // Wake all blocked assists. These will run when we
907 // start the world again.
910 // Likewise, release the transition lock. Blocked
911 // workers and assists will run when we start the
913 semrelease(&work.markDoneSema)
915 // In STW mode, re-enable user goroutines. These will be
916 // queued to run after we start the world.
917 schedEnableUser(true)
919 // endCycle depends on all gcWork cache stats being flushed.
920 // The termination algorithm above ensured that up to
921 // allocations since the ragged barrier.
922 gcController.endCycle(now, int(gomaxprocs), work.userForced)
924 // Perform mark termination. This will restart the world.
925 gcMarkTermination(pauseStart)
928 // World must be stopped and mark assists and background workers must be
930 func gcMarkTermination(pauseStart int64) {
931 // Start marktermination (write barrier remains enabled for now).
932 setGCPhase(_GCmarktermination)
934 work.heap1 = gcController.heapLive.Load()
935 startTime := nanotime()
938 mp.preemptoff = "gcing"
941 // N.B. The execution tracer is not aware of this status
942 // transition and handles it specially based on the
944 casGToWaiting(curgp, _Grunning, waitReasonGarbageCollection)
946 // Run gc on the g0 stack. We do this so that the g stack
947 // we're currently running on will no longer change. Cuts
948 // the root set down a bit (g0 stacks are not scanned, and
949 // we don't need to scan gc's internal state). We also
950 // need to switch to g0 so we can shrink the stack.
953 // Must return immediately.
954 // The outer function's stack may have moved
955 // during gcMark (it shrinks stacks, including the
956 // outer function's stack), so we must not refer
957 // to any of its variables. Return back to the
958 // non-system stack to pick up the new addresses
959 // before continuing.
964 work.heap2 = work.bytesMarked
965 if debug.gccheckmark > 0 {
966 // Run a full non-parallel, stop-the-world
967 // mark using checkmark bits, to check that we
968 // didn't forget to mark anything during the
969 // concurrent mark process.
972 gcw := &getg().m.p.ptr().gcw
974 wbBufFlush1(getg().m.p.ptr())
979 // marking is complete so we can turn the write barrier off
981 stwSwept = gcSweep(work.mode)
985 casgstatus(curgp, _Gwaiting, _Grunning)
987 trace := traceAcquire()
996 if gcphase != _GCoff {
997 throw("gc done but gcphase != _GCoff")
1000 // Record heapInUse for scavenger.
1001 memstats.lastHeapInUse = gcController.heapInUse.load()
1003 // Update GC trigger and pacing, as well as downstream consumers
1004 // of this pacing information, for the next cycle.
1005 systemstack(gcControllerCommit)
1007 // Update timing memstats
1009 sec, nsec, _ := time_now()
1010 unixNow := sec*1e9 + int64(nsec)
1011 work.pauseNS += now - pauseStart
1013 memstats.gcPauseDist.record(now - pauseStart)
1014 atomic.Store64(&memstats.last_gc_unix, uint64(unixNow)) // must be Unix time to make sense to user
1015 atomic.Store64(&memstats.last_gc_nanotime, uint64(now)) // monotonic time for us
1016 memstats.pause_ns[memstats.numgc%uint32(len(memstats.pause_ns))] = uint64(work.pauseNS)
1017 memstats.pause_end[memstats.numgc%uint32(len(memstats.pause_end))] = uint64(unixNow)
1018 memstats.pause_total_ns += uint64(work.pauseNS)
1020 markTermCpu := int64(work.stwprocs) * (work.tEnd - work.tMarkTerm)
1021 work.cpuStats.gcPauseTime += markTermCpu
1022 work.cpuStats.gcTotalTime += markTermCpu
1024 // Accumulate CPU stats.
1026 // Pass gcMarkPhase=true so we can get all the latest GC CPU stats in there too.
1027 work.cpuStats.accumulate(now, true)
1029 // Compute overall GC CPU utilization.
1030 // Omit idle marking time from the overall utilization here since it's "free".
1031 memstats.gc_cpu_fraction = float64(work.cpuStats.gcTotalTime-work.cpuStats.gcIdleTime) / float64(work.cpuStats.totalTime)
1033 // Reset assist time and background time stats.
1035 // Do this now, instead of at the start of the next GC cycle, because
1036 // these two may keep accumulating even if the GC is not active.
1037 scavenge.assistTime.Store(0)
1038 scavenge.backgroundTime.Store(0)
1040 // Reset idle time stat.
1041 sched.idleTime.Store(0)
1043 if work.userForced {
1044 memstats.numforcedgc++
1047 // Bump GC cycle count and wake goroutines waiting on sweep.
1048 lock(&work.sweepWaiters.lock)
1050 injectglist(&work.sweepWaiters.list)
1051 unlock(&work.sweepWaiters.lock)
1053 // Increment the scavenge generation now.
1055 // This moment represents peak heap in use because we're
1056 // about to start sweeping.
1057 mheap_.pages.scav.index.nextGen()
1059 // Release the CPU limiter.
1060 gcCPULimiter.finishGCTransition(now)
1062 // Finish the current heap profiling cycle and start a new
1063 // heap profiling cycle. We do this before starting the world
1064 // so events don't leak into the wrong cycle.
1067 // There may be stale spans in mcaches that need to be swept.
1068 // Those aren't tracked in any sweep lists, so we need to
1069 // count them against sweep completion until we ensure all
1070 // those spans have been forced out.
1072 // If gcSweep fully swept the heap (for example if the sweep
1073 // is not concurrent due to a GODEBUG setting), then we expect
1074 // the sweepLocker to be invalid, since sweeping is done.
1076 // N.B. Below we might duplicate some work from gcSweep; this is
1077 // fine as all that work is idempotent within a GC cycle, and
1078 // we're still holding worldsema so a new cycle can't start.
1079 sl := sweep.active.begin()
1080 if !stwSwept && !sl.valid {
1081 throw("failed to set sweep barrier")
1082 } else if stwSwept && sl.valid {
1083 throw("non-concurrent sweep failed to drain all sweep queues")
1086 systemstack(func() { startTheWorldWithSema() })
1088 // Flush the heap profile so we can start a new cycle next GC.
1089 // This is relatively expensive, so we don't do it with the
1093 // Prepare workbufs for freeing by the sweeper. We do this
1094 // asynchronously because it can take non-trivial time.
1095 prepareFreeWorkbufs()
1097 // Free stack spans. This must be done between GC cycles.
1098 systemstack(freeStackSpans)
1100 // Ensure all mcaches are flushed. Each P will flush its own
1101 // mcache before allocating, but idle Ps may not. Since this
1102 // is necessary to sweep all spans, we need to ensure all
1103 // mcaches are flushed before we start the next GC cycle.
1105 // While we're here, flush the page cache for idle Ps to avoid
1106 // having pages get stuck on them. These pages are hidden from
1107 // the scavenger, so in small idle heaps a significant amount
1108 // of additional memory might be held onto.
1110 // Also, flush the pinner cache, to avoid leaking that memory
1112 forEachP(waitReasonFlushProcCaches, func(pp *p) {
1113 pp.mcache.prepareForSweep()
1114 if pp.status == _Pidle {
1115 systemstack(func() {
1117 pp.pcache.flush(&mheap_.pages)
1118 unlock(&mheap_.lock)
1121 pp.pinnerCache = nil
1124 // Now that we've swept stale spans in mcaches, they don't
1125 // count against unswept spans.
1127 // Note: this sweepLocker may not be valid if sweeping had
1128 // already completed during the STW. See the corresponding
1129 // begin() call that produced sl.
1130 sweep.active.end(sl)
1133 // Print gctrace before dropping worldsema. As soon as we drop
1134 // worldsema another cycle could start and smash the stats
1135 // we're trying to print.
1136 if debug.gctrace > 0 {
1137 util := int(memstats.gc_cpu_fraction * 100)
1141 print("gc ", memstats.numgc,
1142 " @", string(itoaDiv(sbuf[:], uint64(work.tSweepTerm-runtimeInitTime)/1e6, 3)), "s ",
1144 prev := work.tSweepTerm
1145 for i, ns := range []int64{work.tMark, work.tMarkTerm, work.tEnd} {
1149 print(string(fmtNSAsMS(sbuf[:], uint64(ns-prev))))
1152 print(" ms clock, ")
1153 for i, ns := range []int64{
1154 int64(work.stwprocs) * (work.tMark - work.tSweepTerm),
1155 gcController.assistTime.Load(),
1156 gcController.dedicatedMarkTime.Load() + gcController.fractionalMarkTime.Load(),
1157 gcController.idleMarkTime.Load(),
1160 if i == 2 || i == 3 {
1161 // Separate mark time components with /.
1166 print(string(fmtNSAsMS(sbuf[:], uint64(ns))))
1169 work.heap0>>20, "->", work.heap1>>20, "->", work.heap2>>20, " MB, ",
1170 gcController.lastHeapGoal>>20, " MB goal, ",
1171 gcController.lastStackScan.Load()>>20, " MB stacks, ",
1172 gcController.globalsScan.Load()>>20, " MB globals, ",
1173 work.maxprocs, " P")
1174 if work.userForced {
1181 // Set any arena chunks that were deferred to fault.
1182 lock(&userArenaState.lock)
1183 faultList := userArenaState.fault
1184 userArenaState.fault = nil
1185 unlock(&userArenaState.lock)
1186 for _, lc := range faultList {
1187 lc.mspan.setUserArenaChunkToFault()
1190 // Enable huge pages on some metadata if we cross a heap threshold.
1191 if gcController.heapGoal() > minHeapForMetadataHugePages {
1192 systemstack(func() {
1193 mheap_.enableMetadataHugePages()
1197 semrelease(&worldsema)
1199 // Careful: another GC cycle may start now.
1204 // now that gc is done, kick off finalizer thread if needed
1205 if !concurrentSweep {
1206 // give the queued finalizers, if any, a chance to run
1211 // gcBgMarkStartWorkers prepares background mark worker goroutines. These
1212 // goroutines will not run until the mark phase, but they must be started while
1213 // the work is not stopped and from a regular G stack. The caller must hold
1215 func gcBgMarkStartWorkers() {
1216 // Background marking is performed by per-P G's. Ensure that each P has
1217 // a background GC G.
1219 // Worker Gs don't exit if gomaxprocs is reduced. If it is raised
1220 // again, we can reuse the old workers; no need to create new workers.
1221 for gcBgMarkWorkerCount < gomaxprocs {
1224 notetsleepg(&work.bgMarkReady, -1)
1225 noteclear(&work.bgMarkReady)
1226 // The worker is now guaranteed to be added to the pool before
1227 // its P's next findRunnableGCWorker.
1229 gcBgMarkWorkerCount++
1233 // gcBgMarkPrepare sets up state for background marking.
1234 // Mutator assists must not yet be enabled.
1235 func gcBgMarkPrepare() {
1236 // Background marking will stop when the work queues are empty
1237 // and there are no more workers (note that, since this is
1238 // concurrent, this may be a transient state, but mark
1239 // termination will clean it up). Between background workers
1240 // and assists, we don't really know how many workers there
1241 // will be, so we pretend to have an arbitrarily large number
1242 // of workers, almost all of which are "waiting". While a
1243 // worker is working it decrements nwait. If nproc == nwait,
1244 // there are no workers.
1245 work.nproc = ^uint32(0)
1246 work.nwait = ^uint32(0)
1249 // gcBgMarkWorkerNode is an entry in the gcBgMarkWorkerPool. It points to a single
1250 // gcBgMarkWorker goroutine.
1251 type gcBgMarkWorkerNode struct {
1252 // Unused workers are managed in a lock-free stack. This field must be first.
1255 // The g of this worker.
1258 // Release this m on park. This is used to communicate with the unlock
1259 // function, which cannot access the G's stack. It is unused outside of
1260 // gcBgMarkWorker().
1264 func gcBgMarkWorker() {
1267 // We pass node to a gopark unlock function, so it can't be on
1268 // the stack (see gopark). Prevent deadlock from recursively
1269 // starting GC by disabling preemption.
1270 gp.m.preemptoff = "GC worker init"
1271 node := new(gcBgMarkWorkerNode)
1272 gp.m.preemptoff = ""
1276 node.m.set(acquirem())
1277 notewakeup(&work.bgMarkReady)
1278 // After this point, the background mark worker is generally scheduled
1279 // cooperatively by gcController.findRunnableGCWorker. While performing
1280 // work on the P, preemption is disabled because we are working on
1281 // P-local work buffers. When the preempt flag is set, this puts itself
1282 // into _Gwaiting to be woken up by gcController.findRunnableGCWorker
1283 // at the appropriate time.
1285 // When preemption is enabled (e.g., while in gcMarkDone), this worker
1286 // may be preempted and schedule as a _Grunnable G from a runq. That is
1287 // fine; it will eventually gopark again for further scheduling via
1288 // findRunnableGCWorker.
1290 // Since we disable preemption before notifying bgMarkReady, we
1291 // guarantee that this G will be in the worker pool for the next
1292 // findRunnableGCWorker. This isn't strictly necessary, but it reduces
1293 // latency between _GCmark starting and the workers starting.
1296 // Go to sleep until woken by
1297 // gcController.findRunnableGCWorker.
1298 gopark(func(g *g, nodep unsafe.Pointer) bool {
1299 node := (*gcBgMarkWorkerNode)(nodep)
1301 if mp := node.m.ptr(); mp != nil {
1302 // The worker G is no longer running; release
1305 // N.B. it is _safe_ to release the M as soon
1306 // as we are no longer performing P-local mark
1309 // However, since we cooperatively stop work
1310 // when gp.preempt is set, if we releasem in
1311 // the loop then the following call to gopark
1312 // would immediately preempt the G. This is
1313 // also safe, but inefficient: the G must
1314 // schedule again only to enter gopark and park
1315 // again. Thus, we defer the release until
1316 // after parking the G.
1320 // Release this G to the pool.
1321 gcBgMarkWorkerPool.push(&node.node)
1322 // Note that at this point, the G may immediately be
1323 // rescheduled and may be running.
1325 }, unsafe.Pointer(node), waitReasonGCWorkerIdle, traceBlockSystemGoroutine, 0)
1327 // Preemption must not occur here, or another G might see
1328 // p.gcMarkWorkerMode.
1330 // Disable preemption so we can use the gcw. If the
1331 // scheduler wants to preempt us, we'll stop draining,
1332 // dispose the gcw, and then preempt.
1333 node.m.set(acquirem())
1334 pp := gp.m.p.ptr() // P can't change with preemption disabled.
1336 if gcBlackenEnabled == 0 {
1337 println("worker mode", pp.gcMarkWorkerMode)
1338 throw("gcBgMarkWorker: blackening not enabled")
1341 if pp.gcMarkWorkerMode == gcMarkWorkerNotWorker {
1342 throw("gcBgMarkWorker: mode not set")
1345 startTime := nanotime()
1346 pp.gcMarkWorkerStartTime = startTime
1347 var trackLimiterEvent bool
1348 if pp.gcMarkWorkerMode == gcMarkWorkerIdleMode {
1349 trackLimiterEvent = pp.limiterEvent.start(limiterEventIdleMarkWork, startTime)
1352 decnwait := atomic.Xadd(&work.nwait, -1)
1353 if decnwait == work.nproc {
1354 println("runtime: work.nwait=", decnwait, "work.nproc=", work.nproc)
1355 throw("work.nwait was > work.nproc")
1358 systemstack(func() {
1359 // Mark our goroutine preemptible so its stack
1360 // can be scanned. This lets two mark workers
1361 // scan each other (otherwise, they would
1362 // deadlock). We must not modify anything on
1363 // the G stack. However, stack shrinking is
1364 // disabled for mark workers, so it is safe to
1365 // read from the G stack.
1367 // N.B. The execution tracer is not aware of this status
1368 // transition and handles it specially based on the
1370 casGToWaiting(gp, _Grunning, waitReasonGCWorkerActive)
1371 switch pp.gcMarkWorkerMode {
1373 throw("gcBgMarkWorker: unexpected gcMarkWorkerMode")
1374 case gcMarkWorkerDedicatedMode:
1375 gcDrainMarkWorkerDedicated(&pp.gcw, true)
1377 // We were preempted. This is
1378 // a useful signal to kick
1379 // everything out of the run
1380 // queue so it can run
1382 if drainQ, n := runqdrain(pp); n > 0 {
1384 globrunqputbatch(&drainQ, int32(n))
1388 // Go back to draining, this time
1389 // without preemption.
1390 gcDrainMarkWorkerDedicated(&pp.gcw, false)
1391 case gcMarkWorkerFractionalMode:
1392 gcDrainMarkWorkerFractional(&pp.gcw)
1393 case gcMarkWorkerIdleMode:
1394 gcDrainMarkWorkerIdle(&pp.gcw)
1396 casgstatus(gp, _Gwaiting, _Grunning)
1399 // Account for time and mark us as stopped.
1401 duration := now - startTime
1402 gcController.markWorkerStop(pp.gcMarkWorkerMode, duration)
1403 if trackLimiterEvent {
1404 pp.limiterEvent.stop(limiterEventIdleMarkWork, now)
1406 if pp.gcMarkWorkerMode == gcMarkWorkerFractionalMode {
1407 atomic.Xaddint64(&pp.gcFractionalMarkTime, duration)
1410 // Was this the last worker and did we run out
1412 incnwait := atomic.Xadd(&work.nwait, +1)
1413 if incnwait > work.nproc {
1414 println("runtime: p.gcMarkWorkerMode=", pp.gcMarkWorkerMode,
1415 "work.nwait=", incnwait, "work.nproc=", work.nproc)
1416 throw("work.nwait > work.nproc")
1419 // We'll releasem after this point and thus this P may run
1420 // something else. We must clear the worker mode to avoid
1421 // attributing the mode to a different (non-worker) G in
1423 pp.gcMarkWorkerMode = gcMarkWorkerNotWorker
1425 // If this worker reached a background mark completion
1426 // point, signal the main GC goroutine.
1427 if incnwait == work.nproc && !gcMarkWorkAvailable(nil) {
1428 // We don't need the P-local buffers here, allow
1429 // preemption because we may schedule like a regular
1430 // goroutine in gcMarkDone (block on locks, etc).
1431 releasem(node.m.ptr())
1439 // gcMarkWorkAvailable reports whether executing a mark worker
1440 // on p is potentially useful. p may be nil, in which case it only
1441 // checks the global sources of work.
1442 func gcMarkWorkAvailable(p *p) bool {
1443 if p != nil && !p.gcw.empty() {
1446 if !work.full.empty() {
1447 return true // global work available
1449 if work.markrootNext < work.markrootJobs {
1450 return true // root scan work available
1455 // gcMark runs the mark (or, for concurrent GC, mark termination)
1456 // All gcWork caches must be empty.
1457 // STW is in effect at this point.
1458 func gcMark(startTime int64) {
1459 if debug.allocfreetrace > 0 {
1463 if gcphase != _GCmarktermination {
1464 throw("in gcMark expecting to see gcphase as _GCmarktermination")
1466 work.tstart = startTime
1468 // Check that there's no marking work remaining.
1469 if work.full != 0 || work.markrootNext < work.markrootJobs {
1470 print("runtime: full=", hex(work.full), " next=", work.markrootNext, " jobs=", work.markrootJobs, " nDataRoots=", work.nDataRoots, " nBSSRoots=", work.nBSSRoots, " nSpanRoots=", work.nSpanRoots, " nStackRoots=", work.nStackRoots, "\n")
1471 panic("non-empty mark queue after concurrent mark")
1474 if debug.gccheckmark > 0 {
1475 // This is expensive when there's a large number of
1476 // Gs, so only do it if checkmark is also enabled.
1480 // Drop allg snapshot. allgs may have grown, in which case
1481 // this is the only reference to the old backing store and
1482 // there's no need to keep it around.
1483 work.stackRoots = nil
1485 // Clear out buffers and double-check that all gcWork caches
1486 // are empty. This should be ensured by gcMarkDone before we
1487 // enter mark termination.
1489 // TODO: We could clear out buffers just before mark if this
1490 // has a non-negligible impact on STW time.
1491 for _, p := range allp {
1492 // The write barrier may have buffered pointers since
1493 // the gcMarkDone barrier. However, since the barrier
1494 // ensured all reachable objects were marked, all of
1495 // these must be pointers to black objects. Hence we
1496 // can just discard the write barrier buffer.
1497 if debug.gccheckmark > 0 {
1498 // For debugging, flush the buffer and make
1499 // sure it really was all marked.
1508 print("runtime: P ", p.id, " flushedWork ", gcw.flushedWork)
1509 if gcw.wbuf1 == nil {
1510 print(" wbuf1=<nil>")
1512 print(" wbuf1.n=", gcw.wbuf1.nobj)
1514 if gcw.wbuf2 == nil {
1515 print(" wbuf2=<nil>")
1517 print(" wbuf2.n=", gcw.wbuf2.nobj)
1520 throw("P has cached GC work at end of mark termination")
1522 // There may still be cached empty buffers, which we
1523 // need to flush since we're going to free them. Also,
1524 // there may be non-zero stats because we allocated
1525 // black after the gcMarkDone barrier.
1529 // Flush scanAlloc from each mcache since we're about to modify
1530 // heapScan directly. If we were to flush this later, then scanAlloc
1531 // might have incorrect information.
1533 // Note that it's not important to retain this information; we know
1534 // exactly what heapScan is at this point via scanWork.
1535 for _, p := range allp {
1543 // Reset controller state.
1544 gcController.resetLive(work.bytesMarked)
1547 // gcSweep must be called on the system stack because it acquires the heap
1548 // lock. See mheap for details.
1550 // Returns true if the heap was fully swept by this function.
1552 // The world must be stopped.
1555 func gcSweep(mode gcMode) bool {
1556 assertWorldStopped()
1558 if gcphase != _GCoff {
1559 throw("gcSweep being done but phase is not GCoff")
1563 mheap_.sweepgen += 2
1564 sweep.active.reset()
1565 mheap_.pagesSwept.Store(0)
1566 mheap_.sweepArenas = mheap_.allArenas
1567 mheap_.reclaimIndex.Store(0)
1568 mheap_.reclaimCredit.Store(0)
1569 unlock(&mheap_.lock)
1571 sweep.centralIndex.clear()
1573 if !concurrentSweep || mode == gcForceBlockMode {
1574 // Special case synchronous sweep.
1575 // Record that no proportional sweeping has to happen.
1577 mheap_.sweepPagesPerByte = 0
1578 unlock(&mheap_.lock)
1579 // Flush all mcaches.
1580 for _, pp := range allp {
1581 pp.mcache.prepareForSweep()
1583 // Sweep all spans eagerly.
1584 for sweepone() != ^uintptr(0) {
1586 // Free workbufs eagerly.
1587 prepareFreeWorkbufs()
1588 for freeSomeWbufs(false) {
1590 // All "free" events for this mark/sweep cycle have
1591 // now happened, so we can make this profile cycle
1592 // available immediately.
1598 // Background sweep.
1601 sweep.parked = false
1602 ready(sweep.g, 0, true)
1608 // gcResetMarkState resets global state prior to marking (concurrent
1609 // or STW) and resets the stack scan state of all Gs.
1611 // This is safe to do without the world stopped because any Gs created
1612 // during or after this will start out in the reset state.
1614 // gcResetMarkState must be called on the system stack because it acquires
1615 // the heap lock. See mheap for details.
1618 func gcResetMarkState() {
1619 // This may be called during a concurrent phase, so lock to make sure
1620 // allgs doesn't change.
1621 forEachG(func(gp *g) {
1622 gp.gcscandone = false // set to true in gcphasework
1623 gp.gcAssistBytes = 0
1626 // Clear page marks. This is just 1MB per 64GB of heap, so the
1627 // time here is pretty trivial.
1629 arenas := mheap_.allArenas
1630 unlock(&mheap_.lock)
1631 for _, ai := range arenas {
1632 ha := mheap_.arenas[ai.l1()][ai.l2()]
1633 for i := range ha.pageMarks {
1638 work.bytesMarked = 0
1639 work.initialHeapLive = gcController.heapLive.Load()
1642 // Hooks for other packages
1644 var poolcleanup func()
1645 var boringCaches []unsafe.Pointer // for crypto/internal/boring
1647 //go:linkname sync_runtime_registerPoolCleanup sync.runtime_registerPoolCleanup
1648 func sync_runtime_registerPoolCleanup(f func()) {
1652 //go:linkname boring_registerCache crypto/internal/boring/bcache.registerCache
1653 func boring_registerCache(p unsafe.Pointer) {
1654 boringCaches = append(boringCaches, p)
1659 if poolcleanup != nil {
1663 // clear boringcrypto caches
1664 for _, p := range boringCaches {
1665 atomicstorep(p, nil)
1668 // Clear central sudog cache.
1669 // Leave per-P caches alone, they have strictly bounded size.
1670 // Disconnect cached list before dropping it on the floor,
1671 // so that a dangling ref to one entry does not pin all of them.
1672 lock(&sched.sudoglock)
1673 var sg, sgnext *sudog
1674 for sg = sched.sudogcache; sg != nil; sg = sgnext {
1678 sched.sudogcache = nil
1679 unlock(&sched.sudoglock)
1681 // Clear central defer pool.
1682 // Leave per-P pools alone, they have strictly bounded size.
1683 lock(&sched.deferlock)
1684 // disconnect cached list before dropping it on the floor,
1685 // so that a dangling ref to one entry does not pin all of them.
1686 var d, dlink *_defer
1687 for d = sched.deferpool; d != nil; d = dlink {
1691 sched.deferpool = nil
1692 unlock(&sched.deferlock)
1697 // itoaDiv formats val/(10**dec) into buf.
1698 func itoaDiv(buf []byte, val uint64, dec int) []byte {
1701 for val >= 10 || i >= idec {
1702 buf[i] = byte(val%10 + '0')
1710 buf[i] = byte(val + '0')
1714 // fmtNSAsMS nicely formats ns nanoseconds as milliseconds.
1715 func fmtNSAsMS(buf []byte, ns uint64) []byte {
1717 // Format as whole milliseconds.
1718 return itoaDiv(buf, ns/1e6, 0)
1720 // Format two digits of precision, with at most three decimal places.
1731 return itoaDiv(buf, x, dec)
1734 // Helpers for testing GC.
1736 // gcTestMoveStackOnNextCall causes the stack to be moved on a call
1737 // immediately following the call to this. It may not work correctly
1738 // if any other work appears after this call (such as returning).
1739 // Typically the following call should be marked go:noinline so it
1740 // performs a stack check.
1742 // In rare cases this may not cause the stack to move, specifically if
1743 // there's a preemption between this call and the next.
1744 func gcTestMoveStackOnNextCall() {
1746 gp.stackguard0 = stackForceMove
1749 // gcTestIsReachable performs a GC and returns a bit set where bit i
1750 // is set if ptrs[i] is reachable.
1751 func gcTestIsReachable(ptrs ...unsafe.Pointer) (mask uint64) {
1752 // This takes the pointers as unsafe.Pointers in order to keep
1753 // them live long enough for us to attach specials. After
1754 // that, we drop our references to them.
1757 panic("too many pointers for uint64 mask")
1760 // Block GC while we attach specials and drop our references
1761 // to ptrs. Otherwise, if a GC is in progress, it could mark
1762 // them reachable via this function before we have a chance to
1766 // Create reachability specials for ptrs.
1767 specials := make([]*specialReachable, len(ptrs))
1768 for i, p := range ptrs {
1769 lock(&mheap_.speciallock)
1770 s := (*specialReachable)(mheap_.specialReachableAlloc.alloc())
1771 unlock(&mheap_.speciallock)
1772 s.special.kind = _KindSpecialReachable
1773 if !addspecial(p, &s.special) {
1774 throw("already have a reachable special (duplicate pointer?)")
1777 // Make sure we don't retain ptrs.
1783 // Force a full GC and sweep.
1786 // Process specials.
1787 for i, s := range specials {
1790 println("runtime: object", i, "was not swept")
1791 throw("IsReachable failed")
1796 lock(&mheap_.speciallock)
1797 mheap_.specialReachableAlloc.free(unsafe.Pointer(s))
1798 unlock(&mheap_.speciallock)
1804 // gcTestPointerClass returns the category of what p points to, one of:
1805 // "heap", "stack", "data", "bss", "other". This is useful for checking
1806 // that a test is doing what it's intended to do.
1808 // This is nosplit simply to avoid extra pointer shuffling that may
1809 // complicate a test.
1812 func gcTestPointerClass(p unsafe.Pointer) string {
1813 p2 := uintptr(noescape(p))
1815 if gp.stack.lo <= p2 && p2 < gp.stack.hi {
1818 if base, _, _ := findObject(p2, 0, 0); base != 0 {
1821 for _, datap := range activeModules() {
1822 if datap.data <= p2 && p2 < datap.edata || datap.noptrdata <= p2 && p2 < datap.enoptrdata {
1825 if datap.bss <= p2 && p2 < datap.ebss || datap.noptrbss <= p2 && p2 <= datap.enoptrbss {