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 tracked in gcController.heapGoal variable). 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 _ConcurrentSweep = true
140 _FinBlockSize = 4 * 1024
142 // debugScanConservative enables debug logging for stack
143 // frames that are scanned conservatively.
144 debugScanConservative = false
146 // sweepMinHeapDistance is a lower bound on the heap distance
147 // (in bytes) reserved for concurrent sweeping between GC
149 sweepMinHeapDistance = 1024 * 1024
153 if unsafe.Sizeof(workbuf{}) != _WorkbufSize {
154 throw("size of Workbuf is suboptimal")
156 // No sweep on the first cycle.
157 sweep.active.state.Store(sweepDrainedMask)
159 // Initialize GC pacer state.
160 // Use the environment variable GOGC for the initial gcPercent value.
161 gcController.init(readGOGC())
164 work.markDoneSema = 1
165 lockInit(&work.sweepWaiters.lock, lockRankSweepWaiters)
166 lockInit(&work.assistQueue.lock, lockRankAssistQueue)
167 lockInit(&work.wbufSpans.lock, lockRankWbufSpans)
170 // gcenable is called after the bulk of the runtime initialization,
171 // just before we're about to start letting user code run.
172 // It kicks off the background sweeper goroutine, the background
173 // scavenger goroutine, and enables GC.
175 // Kick off sweeping and scavenging.
176 c := make(chan int, 2)
181 memstats.enablegc = true // now that runtime is initialized, GC is okay
184 // Garbage collector phase.
185 // Indicates to write barrier and synchronization task to perform.
188 // The compiler knows about this variable.
189 // If you change it, you must change builtin/runtime.go, too.
190 // If you change the first four bytes, you must also change the write
191 // barrier insertion code.
192 var writeBarrier struct {
193 enabled bool // compiler emits a check of this before calling write barrier
194 pad [3]byte // compiler uses 32-bit load for "enabled" field
195 needed bool // whether we need a write barrier for current GC phase
196 cgo bool // whether we need a write barrier for a cgo check
197 alignme uint64 // guarantee alignment so that compiler can use a 32 or 64-bit load
200 // gcBlackenEnabled is 1 if mutator assists and background mark
201 // workers are allowed to blacken objects. This must only be set when
202 // gcphase == _GCmark.
203 var gcBlackenEnabled uint32
206 _GCoff = iota // GC not running; sweeping in background, write barrier disabled
207 _GCmark // GC marking roots and workbufs: allocate black, write barrier ENABLED
208 _GCmarktermination // GC mark termination: allocate black, P's help GC, write barrier ENABLED
212 func setGCPhase(x uint32) {
213 atomic.Store(&gcphase, x)
214 writeBarrier.needed = gcphase == _GCmark || gcphase == _GCmarktermination
215 writeBarrier.enabled = writeBarrier.needed || writeBarrier.cgo
218 // gcMarkWorkerMode represents the mode that a concurrent mark worker
219 // should operate in.
221 // Concurrent marking happens through four different mechanisms. One
222 // is mutator assists, which happen in response to allocations and are
223 // not scheduled. The other three are variations in the per-P mark
224 // workers and are distinguished by gcMarkWorkerMode.
225 type gcMarkWorkerMode int
228 // gcMarkWorkerNotWorker indicates that the next scheduled G is not
229 // starting work and the mode should be ignored.
230 gcMarkWorkerNotWorker gcMarkWorkerMode = iota
232 // gcMarkWorkerDedicatedMode indicates that the P of a mark
233 // worker is dedicated to running that mark worker. The mark
234 // worker should run without preemption.
235 gcMarkWorkerDedicatedMode
237 // gcMarkWorkerFractionalMode indicates that a P is currently
238 // running the "fractional" mark worker. The fractional worker
239 // is necessary when GOMAXPROCS*gcBackgroundUtilization is not
240 // an integer and using only dedicated workers would result in
241 // utilization too far from the target of gcBackgroundUtilization.
242 // The fractional worker should run until it is preempted and
243 // will be scheduled to pick up the fractional part of
244 // GOMAXPROCS*gcBackgroundUtilization.
245 gcMarkWorkerFractionalMode
247 // gcMarkWorkerIdleMode indicates that a P is running the mark
248 // worker because it has nothing else to do. The idle worker
249 // should run until it is preempted and account its time
250 // against gcController.idleMarkTime.
254 // gcMarkWorkerModeStrings are the strings labels of gcMarkWorkerModes
255 // to use in execution traces.
256 var gcMarkWorkerModeStrings = [...]string{
263 // pollFractionalWorkerExit reports whether a fractional mark worker
264 // should self-preempt. It assumes it is called from the fractional
266 func pollFractionalWorkerExit() bool {
267 // This should be kept in sync with the fractional worker
268 // scheduler logic in findRunnableGCWorker.
270 delta := now - gcController.markStartTime
274 p := getg().m.p.ptr()
275 selfTime := p.gcFractionalMarkTime + (now - p.gcMarkWorkerStartTime)
276 // Add some slack to the utilization goal so that the
277 // fractional worker isn't behind again the instant it exits.
278 return float64(selfTime)/float64(delta) > 1.2*gcController.fractionalUtilizationGoal
282 full lfstack // lock-free list of full blocks workbuf
283 empty lfstack // lock-free list of empty blocks workbuf
284 pad0 cpu.CacheLinePad // prevents false-sharing between full/empty and nproc/nwait
288 // free is a list of spans dedicated to workbufs, but
289 // that don't currently contain any workbufs.
291 // busy is a list of all spans containing workbufs on
292 // one of the workbuf lists.
296 // Restore 64-bit alignment on 32-bit.
299 // bytesMarked is the number of bytes marked this cycle. This
300 // includes bytes blackened in scanned objects, noscan objects
301 // that go straight to black, and permagrey objects scanned by
302 // markroot during the concurrent scan phase. This is updated
303 // atomically during the cycle. Updates may be batched
304 // arbitrarily, since the value is only read at the end of the
307 // Because of benign races during marking, this number may not
308 // be the exact number of marked bytes, but it should be very
311 // Put this field here because it needs 64-bit atomic access
312 // (and thus 8-byte alignment even on 32-bit architectures).
315 markrootNext uint32 // next markroot job
316 markrootJobs uint32 // number of markroot jobs
322 // Number of roots of various root types. Set by gcMarkRootPrepare.
324 // nStackRoots == len(stackRoots), but we have nStackRoots for
326 nDataRoots, nBSSRoots, nSpanRoots, nStackRoots int
328 // Base indexes of each root type. Set by gcMarkRootPrepare.
329 baseData, baseBSS, baseSpans, baseStacks, baseEnd uint32
331 // stackRoots is a snapshot of all of the Gs that existed
332 // before the beginning of concurrent marking. The backing
333 // store of this must not be modified because it might be
334 // shared with allgs.
337 // Each type of GC state transition is protected by a lock.
338 // Since multiple threads can simultaneously detect the state
339 // transition condition, any thread that detects a transition
340 // condition must acquire the appropriate transition lock,
341 // re-check the transition condition and return if it no
342 // longer holds or perform the transition if it does.
343 // Likewise, any transition must invalidate the transition
344 // condition before releasing the lock. This ensures that each
345 // transition is performed by exactly one thread and threads
346 // that need the transition to happen block until it has
349 // startSema protects the transition from "off" to mark or
352 // markDoneSema protects transitions from mark to mark termination.
355 bgMarkReady note // signal background mark worker has started
356 bgMarkDone uint32 // cas to 1 when at a background mark completion point
357 // Background mark completion signaling
359 // mode is the concurrency mode of the current GC cycle.
362 // userForced indicates the current GC cycle was forced by an
363 // explicit user call.
366 // totaltime is the CPU nanoseconds spent in GC since the
367 // program started if debug.gctrace > 0.
370 // initialHeapLive is the value of gcController.heapLive at the
371 // beginning of this GC cycle.
372 initialHeapLive uint64
374 // assistQueue is a queue of assists that are blocked because
375 // there was neither enough credit to steal or enough work to
382 // sweepWaiters is a list of blocked goroutines to wake when
383 // we transition from mark termination to sweep.
384 sweepWaiters struct {
389 // cycles is the number of completed GC cycles, where a GC
390 // cycle is sweep termination, mark, mark termination, and
391 // sweep. This differs from memstats.numgc, which is
392 // incremented at mark termination.
395 // Timing/utilization stats for this cycle.
396 stwprocs, maxprocs int32
397 tSweepTerm, tMark, tMarkTerm, tEnd int64 // nanotime() of phase start
399 pauseNS int64 // total STW time this cycle
400 pauseStart int64 // nanotime() of last STW
402 // debug.gctrace heap sizes for this cycle.
403 heap0, heap1, heap2, heapGoal uint64
406 // GC runs a garbage collection and blocks the caller until the
407 // garbage collection is complete. It may also block the entire
410 // We consider a cycle to be: sweep termination, mark, mark
411 // termination, and sweep. This function shouldn't return
412 // until a full cycle has been completed, from beginning to
413 // end. Hence, we always want to finish up the current cycle
414 // and start a new one. That means:
416 // 1. In sweep termination, mark, or mark termination of cycle
417 // N, wait until mark termination N completes and transitions
420 // 2. In sweep N, help with sweep N.
422 // At this point we can begin a full cycle N+1.
424 // 3. Trigger cycle N+1 by starting sweep termination N+1.
426 // 4. Wait for mark termination N+1 to complete.
428 // 5. Help with sweep N+1 until it's done.
430 // This all has to be written to deal with the fact that the
431 // GC may move ahead on its own. For example, when we block
432 // until mark termination N, we may wake up in cycle N+2.
434 // Wait until the current sweep termination, mark, and mark
435 // termination complete.
436 n := atomic.Load(&work.cycles)
439 // We're now in sweep N or later. Trigger GC cycle N+1, which
440 // will first finish sweep N if necessary and then enter sweep
442 gcStart(gcTrigger{kind: gcTriggerCycle, n: n + 1})
444 // Wait for mark termination N+1 to complete.
447 // Finish sweep N+1 before returning. We do this both to
448 // complete the cycle and because runtime.GC() is often used
449 // as part of tests and benchmarks to get the system into a
450 // relatively stable and isolated state.
451 for atomic.Load(&work.cycles) == n+1 && sweepone() != ^uintptr(0) {
456 // Callers may assume that the heap profile reflects the
457 // just-completed cycle when this returns (historically this
458 // happened because this was a STW GC), but right now the
459 // profile still reflects mark termination N, not N+1.
461 // As soon as all of the sweep frees from cycle N+1 are done,
462 // we can go ahead and publish the heap profile.
464 // First, wait for sweeping to finish. (We know there are no
465 // more spans on the sweep queue, but we may be concurrently
466 // sweeping spans, so we have to wait.)
467 for atomic.Load(&work.cycles) == n+1 && !isSweepDone() {
471 // Now we're really done with sweeping, so we can publish the
472 // stable heap profile. Only do this if we haven't already hit
473 // another mark termination.
475 cycle := atomic.Load(&work.cycles)
476 if cycle == n+1 || (gcphase == _GCmark && cycle == n+2) {
482 // gcWaitOnMark blocks until GC finishes the Nth mark phase. If GC has
483 // already completed this mark phase, it returns immediately.
484 func gcWaitOnMark(n uint32) {
486 // Disable phase transitions.
487 lock(&work.sweepWaiters.lock)
488 nMarks := atomic.Load(&work.cycles)
489 if gcphase != _GCmark {
490 // We've already completed this cycle's mark.
495 unlock(&work.sweepWaiters.lock)
499 // Wait until sweep termination, mark, and mark
500 // termination of cycle N complete.
501 work.sweepWaiters.list.push(getg())
502 goparkunlock(&work.sweepWaiters.lock, waitReasonWaitForGCCycle, traceEvGoBlock, 1)
506 // gcMode indicates how concurrent a GC cycle should be.
510 gcBackgroundMode gcMode = iota // concurrent GC and sweep
511 gcForceMode // stop-the-world GC now, concurrent sweep
512 gcForceBlockMode // stop-the-world GC now and STW sweep (forced by user)
515 // A gcTrigger is a predicate for starting a GC cycle. Specifically,
516 // it is an exit condition for the _GCoff phase.
517 type gcTrigger struct {
519 now int64 // gcTriggerTime: current time
520 n uint32 // gcTriggerCycle: cycle number to start
523 type gcTriggerKind int
526 // gcTriggerHeap indicates that a cycle should be started when
527 // the heap size reaches the trigger heap size computed by the
529 gcTriggerHeap gcTriggerKind = iota
531 // gcTriggerTime indicates that a cycle should be started when
532 // it's been more than forcegcperiod nanoseconds since the
533 // previous GC cycle.
536 // gcTriggerCycle indicates that a cycle should be started if
537 // we have not yet started cycle number gcTrigger.n (relative
542 // test reports whether the trigger condition is satisfied, meaning
543 // that the exit condition for the _GCoff phase has been met. The exit
544 // condition should be tested when allocating.
545 func (t gcTrigger) test() bool {
546 if !memstats.enablegc || panicking != 0 || gcphase != _GCoff {
551 // Non-atomic access to gcController.heapLive for performance. If
552 // we are going to trigger on this, this thread just
553 // atomically wrote gcController.heapLive anyway and we'll see our
555 return gcController.heapLive >= gcController.trigger
557 if gcController.gcPercent.Load() < 0 {
560 lastgc := int64(atomic.Load64(&memstats.last_gc_nanotime))
561 return lastgc != 0 && t.now-lastgc > forcegcperiod
563 // t.n > work.cycles, but accounting for wraparound.
564 return int32(t.n-work.cycles) > 0
569 // gcStart starts the GC. It transitions from _GCoff to _GCmark (if
570 // debug.gcstoptheworld == 0) or performs all of GC (if
571 // debug.gcstoptheworld != 0).
573 // This may return without performing this transition in some cases,
574 // such as when called on a system stack or with locks held.
575 func gcStart(trigger gcTrigger) {
576 // Since this is called from malloc and malloc is called in
577 // the guts of a number of libraries that might be holding
578 // locks, don't attempt to start GC in non-preemptible or
579 // potentially unstable situations.
581 if gp := getg(); gp == mp.g0 || mp.locks > 1 || mp.preemptoff != "" {
588 // Pick up the remaining unswept/not being swept spans concurrently
590 // This shouldn't happen if we're being invoked in background
591 // mode since proportional sweep should have just finished
592 // sweeping everything, but rounding errors, etc, may leave a
593 // few spans unswept. In forced mode, this is necessary since
594 // GC can be forced at any point in the sweeping cycle.
596 // We check the transition condition continuously here in case
597 // this G gets delayed in to the next GC cycle.
598 for trigger.test() && sweepone() != ^uintptr(0) {
602 // Perform GC initialization and the sweep termination
604 semacquire(&work.startSema)
605 // Re-check transition condition under transition lock.
607 semrelease(&work.startSema)
611 // For stats, check if this GC was forced by the user.
612 work.userForced = trigger.kind == gcTriggerCycle
614 // In gcstoptheworld debug mode, upgrade the mode accordingly.
615 // We do this after re-checking the transition condition so
616 // that multiple goroutines that detect the heap trigger don't
617 // start multiple STW GCs.
618 mode := gcBackgroundMode
619 if debug.gcstoptheworld == 1 {
621 } else if debug.gcstoptheworld == 2 {
622 mode = gcForceBlockMode
625 // Ok, we're doing it! Stop everybody else
627 semacquire(&worldsema)
633 // Check that all Ps have finished deferred mcache flushes.
634 for _, p := range allp {
635 if fg := atomic.Load(&p.mcache.flushGen); fg != mheap_.sweepgen {
636 println("runtime: p", p.id, "flushGen", fg, "!= sweepgen", mheap_.sweepgen)
637 throw("p mcache not flushed")
641 gcBgMarkStartWorkers()
643 systemstack(gcResetMarkState)
645 work.stwprocs, work.maxprocs = gomaxprocs, gomaxprocs
646 if work.stwprocs > ncpu {
647 // This is used to compute CPU time of the STW phases,
648 // so it can't be more than ncpu, even if GOMAXPROCS is.
651 work.heap0 = atomic.Load64(&gcController.heapLive)
656 work.tSweepTerm = now
657 work.pauseStart = now
661 systemstack(stopTheWorldWithSema)
662 // Finish sweep before we start concurrent scan.
667 // clearpools before we start the GC. If we wait they memory will not be
668 // reclaimed until the next GC cycle.
673 // Assists and workers can start the moment we start
675 gcController.startCycle(now, int(gomaxprocs), trigger)
676 work.heapGoal = gcController.heapGoal
678 // In STW mode, disable scheduling of user Gs. This may also
679 // disable scheduling of this goroutine, so it may block as
680 // soon as we start the world again.
681 if mode != gcBackgroundMode {
682 schedEnableUser(false)
685 // Enter concurrent mark phase and enable
688 // Because the world is stopped, all Ps will
689 // observe that write barriers are enabled by
690 // the time we start the world and begin
693 // Write barriers must be enabled before assists are
694 // enabled because they must be enabled before
695 // any non-leaf heap objects are marked. Since
696 // allocations are blocked until assists can
697 // happen, we want enable assists as early as
701 gcBgMarkPrepare() // Must happen before assist enable.
704 // Mark all active tinyalloc blocks. Since we're
705 // allocating from these, they need to be black like
706 // other allocations. The alternative is to blacken
707 // the tiny block on every allocation from it, which
708 // would slow down the tiny allocator.
711 // At this point all Ps have enabled the write
712 // barrier, thus maintaining the no white to
713 // black invariant. Enable mutator assists to
714 // put back-pressure on fast allocating
716 atomic.Store(&gcBlackenEnabled, 1)
718 // In STW mode, we could block the instant systemstack
719 // returns, so make sure we're not preemptible.
724 now = startTheWorldWithSema(trace.enabled)
725 work.pauseNS += now - work.pauseStart
727 memstats.gcPauseDist.record(now - work.pauseStart)
730 // Release the world sema before Gosched() in STW mode
731 // because we will need to reacquire it later but before
732 // this goroutine becomes runnable again, and we could
733 // self-deadlock otherwise.
734 semrelease(&worldsema)
737 // Make sure we block instead of returning to user code
739 if mode != gcBackgroundMode {
743 semrelease(&work.startSema)
746 // gcMarkDoneFlushed counts the number of P's with flushed work.
748 // Ideally this would be a captured local in gcMarkDone, but forEachP
749 // escapes its callback closure, so it can't capture anything.
751 // This is protected by markDoneSema.
752 var gcMarkDoneFlushed uint32
754 // gcMarkDone transitions the GC from mark to mark termination if all
755 // reachable objects have been marked (that is, there are no grey
756 // objects and can be no more in the future). Otherwise, it flushes
757 // all local work to the global queues where it can be discovered by
760 // This should be called when all local mark work has been drained and
761 // there are no remaining workers. Specifically, when
763 // work.nwait == work.nproc && !gcMarkWorkAvailable(p)
765 // The calling context must be preemptible.
767 // Flushing local work is important because idle Ps may have local
768 // work queued. This is the only way to make that work visible and
769 // drive GC to completion.
771 // It is explicitly okay to have write barriers in this function. If
772 // it does transition to mark termination, then all reachable objects
773 // have been marked, so the write barrier cannot shade any more
776 // Ensure only one thread is running the ragged barrier at a
778 semacquire(&work.markDoneSema)
781 // Re-check transition condition under transition lock.
783 // It's critical that this checks the global work queues are
784 // empty before performing the ragged barrier. Otherwise,
785 // there could be global work that a P could take after the P
786 // has passed the ragged barrier.
787 if !(gcphase == _GCmark && work.nwait == work.nproc && !gcMarkWorkAvailable(nil)) {
788 semrelease(&work.markDoneSema)
792 // forEachP needs worldsema to execute, and we'll need it to
793 // stop the world later, so acquire worldsema now.
794 semacquire(&worldsema)
796 // Flush all local buffers and collect flushedWork flags.
797 gcMarkDoneFlushed = 0
800 // Mark the user stack as preemptible so that it may be scanned.
801 // Otherwise, our attempt to force all P's to a safepoint could
802 // result in a deadlock as we attempt to preempt a worker that's
803 // trying to preempt us (e.g. for a stack scan).
804 casgstatus(gp, _Grunning, _Gwaiting)
805 forEachP(func(_p_ *p) {
806 // Flush the write barrier buffer, since this may add
807 // work to the gcWork.
810 // Flush the gcWork, since this may create global work
811 // and set the flushedWork flag.
813 // TODO(austin): Break up these workbufs to
814 // better distribute work.
816 // Collect the flushedWork flag.
817 if _p_.gcw.flushedWork {
818 atomic.Xadd(&gcMarkDoneFlushed, 1)
819 _p_.gcw.flushedWork = false
822 casgstatus(gp, _Gwaiting, _Grunning)
825 if gcMarkDoneFlushed != 0 {
826 // More grey objects were discovered since the
827 // previous termination check, so there may be more
828 // work to do. Keep going. It's possible the
829 // transition condition became true again during the
830 // ragged barrier, so re-check it.
831 semrelease(&worldsema)
835 // There was no global work, no local work, and no Ps
836 // communicated work since we took markDoneSema. Therefore
837 // there are no grey objects and no more objects can be
838 // shaded. Transition to mark termination.
841 work.pauseStart = now
842 getg().m.preemptoff = "gcing"
846 systemstack(stopTheWorldWithSema)
847 // The gcphase is _GCmark, it will transition to _GCmarktermination
848 // below. The important thing is that the wb remains active until
849 // all marking is complete. This includes writes made by the GC.
851 // There is sometimes work left over when we enter mark termination due
852 // to write barriers performed after the completion barrier above.
853 // Detect this and resume concurrent mark. This is obviously
856 // See issue #27993 for details.
858 // Switch to the system stack to call wbBufFlush1, though in this case
859 // it doesn't matter because we're non-preemptible anyway.
862 for _, p := range allp {
871 getg().m.preemptoff = ""
873 now := startTheWorldWithSema(true)
874 work.pauseNS += now - work.pauseStart
875 memstats.gcPauseDist.record(now - work.pauseStart)
877 semrelease(&worldsema)
881 // Disable assists and background workers. We must do
882 // this before waking blocked assists.
883 atomic.Store(&gcBlackenEnabled, 0)
885 // Wake all blocked assists. These will run when we
886 // start the world again.
889 // Likewise, release the transition lock. Blocked
890 // workers and assists will run when we start the
892 semrelease(&work.markDoneSema)
894 // In STW mode, re-enable user goroutines. These will be
895 // queued to run after we start the world.
896 schedEnableUser(true)
898 // endCycle depends on all gcWork cache stats being flushed.
899 // The termination algorithm above ensured that up to
900 // allocations since the ragged barrier.
901 gcController.endCycle(now, int(gomaxprocs), work.userForced)
903 // Perform mark termination. This will restart the world.
907 // World must be stopped and mark assists and background workers must be
909 func gcMarkTermination() {
910 // Start marktermination (write barrier remains enabled for now).
911 setGCPhase(_GCmarktermination)
913 work.heap1 = gcController.heapLive
914 startTime := nanotime()
917 mp.preemptoff = "gcing"
921 casgstatus(gp, _Grunning, _Gwaiting)
922 gp.waitreason = waitReasonGarbageCollection
924 // Run gc on the g0 stack. We do this so that the g stack
925 // we're currently running on will no longer change. Cuts
926 // the root set down a bit (g0 stacks are not scanned, and
927 // we don't need to scan gc's internal state). We also
928 // need to switch to g0 so we can shrink the stack.
931 // Must return immediately.
932 // The outer function's stack may have moved
933 // during gcMark (it shrinks stacks, including the
934 // outer function's stack), so we must not refer
935 // to any of its variables. Return back to the
936 // non-system stack to pick up the new addresses
937 // before continuing.
941 work.heap2 = work.bytesMarked
942 if debug.gccheckmark > 0 {
943 // Run a full non-parallel, stop-the-world
944 // mark using checkmark bits, to check that we
945 // didn't forget to mark anything during the
946 // concurrent mark process.
949 gcw := &getg().m.p.ptr().gcw
951 wbBufFlush1(getg().m.p.ptr())
956 // marking is complete so we can turn the write barrier off
962 casgstatus(gp, _Gwaiting, _Grunning)
971 if gcphase != _GCoff {
972 throw("gc done but gcphase != _GCoff")
975 // Record heap_inuse for scavenger.
976 memstats.last_heap_inuse = memstats.heap_inuse
978 // Update GC trigger and pacing for the next cycle.
979 gcController.commit()
980 gcPaceSweeper(gcController.trigger)
981 gcPaceScavenger(gcController.heapGoal, gcController.lastHeapGoal)
983 // Update timing memstats
985 sec, nsec, _ := time_now()
986 unixNow := sec*1e9 + int64(nsec)
987 work.pauseNS += now - work.pauseStart
989 memstats.gcPauseDist.record(now - work.pauseStart)
990 atomic.Store64(&memstats.last_gc_unix, uint64(unixNow)) // must be Unix time to make sense to user
991 atomic.Store64(&memstats.last_gc_nanotime, uint64(now)) // monotonic time for us
992 memstats.pause_ns[memstats.numgc%uint32(len(memstats.pause_ns))] = uint64(work.pauseNS)
993 memstats.pause_end[memstats.numgc%uint32(len(memstats.pause_end))] = uint64(unixNow)
994 memstats.pause_total_ns += uint64(work.pauseNS)
996 // Update work.totaltime.
997 sweepTermCpu := int64(work.stwprocs) * (work.tMark - work.tSweepTerm)
998 // We report idle marking time below, but omit it from the
999 // overall utilization here since it's "free".
1000 markCpu := gcController.assistTime + gcController.dedicatedMarkTime + gcController.fractionalMarkTime
1001 markTermCpu := int64(work.stwprocs) * (work.tEnd - work.tMarkTerm)
1002 cycleCpu := sweepTermCpu + markCpu + markTermCpu
1003 work.totaltime += cycleCpu
1005 // Compute overall GC CPU utilization.
1006 totalCpu := sched.totaltime + (now-sched.procresizetime)*int64(gomaxprocs)
1007 memstats.gc_cpu_fraction = float64(work.totaltime) / float64(totalCpu)
1009 // Reset sweep state.
1011 sweep.npausesweep = 0
1013 if work.userForced {
1014 memstats.numforcedgc++
1017 // Bump GC cycle count and wake goroutines waiting on sweep.
1018 lock(&work.sweepWaiters.lock)
1020 injectglist(&work.sweepWaiters.list)
1021 unlock(&work.sweepWaiters.lock)
1023 // Finish the current heap profiling cycle and start a new
1024 // heap profiling cycle. We do this before starting the world
1025 // so events don't leak into the wrong cycle.
1028 // There may be stale spans in mcaches that need to be swept.
1029 // Those aren't tracked in any sweep lists, so we need to
1030 // count them against sweep completion until we ensure all
1031 // those spans have been forced out.
1032 sl := sweep.active.begin()
1034 throw("failed to set sweep barrier")
1037 systemstack(func() { startTheWorldWithSema(true) })
1039 // Flush the heap profile so we can start a new cycle next GC.
1040 // This is relatively expensive, so we don't do it with the
1044 // Prepare workbufs for freeing by the sweeper. We do this
1045 // asynchronously because it can take non-trivial time.
1046 prepareFreeWorkbufs()
1048 // Free stack spans. This must be done between GC cycles.
1049 systemstack(freeStackSpans)
1051 // Ensure all mcaches are flushed. Each P will flush its own
1052 // mcache before allocating, but idle Ps may not. Since this
1053 // is necessary to sweep all spans, we need to ensure all
1054 // mcaches are flushed before we start the next GC cycle.
1055 systemstack(func() {
1056 forEachP(func(_p_ *p) {
1057 _p_.mcache.prepareForSweep()
1060 // Now that we've swept stale spans in mcaches, they don't
1061 // count against unswept spans.
1062 sweep.active.end(sl)
1064 // Print gctrace before dropping worldsema. As soon as we drop
1065 // worldsema another cycle could start and smash the stats
1066 // we're trying to print.
1067 if debug.gctrace > 0 {
1068 util := int(memstats.gc_cpu_fraction * 100)
1072 print("gc ", memstats.numgc,
1073 " @", string(itoaDiv(sbuf[:], uint64(work.tSweepTerm-runtimeInitTime)/1e6, 3)), "s ",
1075 prev := work.tSweepTerm
1076 for i, ns := range []int64{work.tMark, work.tMarkTerm, work.tEnd} {
1080 print(string(fmtNSAsMS(sbuf[:], uint64(ns-prev))))
1083 print(" ms clock, ")
1084 for i, ns := range []int64{sweepTermCpu, gcController.assistTime, gcController.dedicatedMarkTime + gcController.fractionalMarkTime, gcController.idleMarkTime, markTermCpu} {
1085 if i == 2 || i == 3 {
1086 // Separate mark time components with /.
1091 print(string(fmtNSAsMS(sbuf[:], uint64(ns))))
1094 work.heap0>>20, "->", work.heap1>>20, "->", work.heap2>>20, " MB, ",
1095 work.heapGoal>>20, " MB goal, ",
1096 gcController.stackScan>>20, " MB stacks, ",
1097 gcController.globalsScan>>20, " MB globals, ",
1098 work.maxprocs, " P")
1099 if work.userForced {
1106 semrelease(&worldsema)
1108 // Careful: another GC cycle may start now.
1113 // now that gc is done, kick off finalizer thread if needed
1114 if !concurrentSweep {
1115 // give the queued finalizers, if any, a chance to run
1120 // gcBgMarkStartWorkers prepares background mark worker goroutines. These
1121 // goroutines will not run until the mark phase, but they must be started while
1122 // the work is not stopped and from a regular G stack. The caller must hold
1124 func gcBgMarkStartWorkers() {
1125 // Background marking is performed by per-P G's. Ensure that each P has
1126 // a background GC G.
1128 // Worker Gs don't exit if gomaxprocs is reduced. If it is raised
1129 // again, we can reuse the old workers; no need to create new workers.
1130 for gcBgMarkWorkerCount < gomaxprocs {
1133 notetsleepg(&work.bgMarkReady, -1)
1134 noteclear(&work.bgMarkReady)
1135 // The worker is now guaranteed to be added to the pool before
1136 // its P's next findRunnableGCWorker.
1138 gcBgMarkWorkerCount++
1142 // gcBgMarkPrepare sets up state for background marking.
1143 // Mutator assists must not yet be enabled.
1144 func gcBgMarkPrepare() {
1145 // Background marking will stop when the work queues are empty
1146 // and there are no more workers (note that, since this is
1147 // concurrent, this may be a transient state, but mark
1148 // termination will clean it up). Between background workers
1149 // and assists, we don't really know how many workers there
1150 // will be, so we pretend to have an arbitrarily large number
1151 // of workers, almost all of which are "waiting". While a
1152 // worker is working it decrements nwait. If nproc == nwait,
1153 // there are no workers.
1154 work.nproc = ^uint32(0)
1155 work.nwait = ^uint32(0)
1158 // gcBgMarkWorker is an entry in the gcBgMarkWorkerPool. It points to a single
1159 // gcBgMarkWorker goroutine.
1160 type gcBgMarkWorkerNode struct {
1161 // Unused workers are managed in a lock-free stack. This field must be first.
1164 // The g of this worker.
1167 // Release this m on park. This is used to communicate with the unlock
1168 // function, which cannot access the G's stack. It is unused outside of
1169 // gcBgMarkWorker().
1173 func gcBgMarkWorker() {
1176 // We pass node to a gopark unlock function, so it can't be on
1177 // the stack (see gopark). Prevent deadlock from recursively
1178 // starting GC by disabling preemption.
1179 gp.m.preemptoff = "GC worker init"
1180 node := new(gcBgMarkWorkerNode)
1181 gp.m.preemptoff = ""
1185 node.m.set(acquirem())
1186 notewakeup(&work.bgMarkReady)
1187 // After this point, the background mark worker is generally scheduled
1188 // cooperatively by gcController.findRunnableGCWorker. While performing
1189 // work on the P, preemption is disabled because we are working on
1190 // P-local work buffers. When the preempt flag is set, this puts itself
1191 // into _Gwaiting to be woken up by gcController.findRunnableGCWorker
1192 // at the appropriate time.
1194 // When preemption is enabled (e.g., while in gcMarkDone), this worker
1195 // may be preempted and schedule as a _Grunnable G from a runq. That is
1196 // fine; it will eventually gopark again for further scheduling via
1197 // findRunnableGCWorker.
1199 // Since we disable preemption before notifying bgMarkReady, we
1200 // guarantee that this G will be in the worker pool for the next
1201 // findRunnableGCWorker. This isn't strictly necessary, but it reduces
1202 // latency between _GCmark starting and the workers starting.
1205 // Go to sleep until woken by
1206 // gcController.findRunnableGCWorker.
1207 gopark(func(g *g, nodep unsafe.Pointer) bool {
1208 node := (*gcBgMarkWorkerNode)(nodep)
1210 if mp := node.m.ptr(); mp != nil {
1211 // The worker G is no longer running; release
1214 // N.B. it is _safe_ to release the M as soon
1215 // as we are no longer performing P-local mark
1218 // However, since we cooperatively stop work
1219 // when gp.preempt is set, if we releasem in
1220 // the loop then the following call to gopark
1221 // would immediately preempt the G. This is
1222 // also safe, but inefficient: the G must
1223 // schedule again only to enter gopark and park
1224 // again. Thus, we defer the release until
1225 // after parking the G.
1229 // Release this G to the pool.
1230 gcBgMarkWorkerPool.push(&node.node)
1231 // Note that at this point, the G may immediately be
1232 // rescheduled and may be running.
1234 }, unsafe.Pointer(node), waitReasonGCWorkerIdle, traceEvGoBlock, 0)
1236 // Preemption must not occur here, or another G might see
1237 // p.gcMarkWorkerMode.
1239 // Disable preemption so we can use the gcw. If the
1240 // scheduler wants to preempt us, we'll stop draining,
1241 // dispose the gcw, and then preempt.
1242 node.m.set(acquirem())
1243 pp := gp.m.p.ptr() // P can't change with preemption disabled.
1245 if gcBlackenEnabled == 0 {
1246 println("worker mode", pp.gcMarkWorkerMode)
1247 throw("gcBgMarkWorker: blackening not enabled")
1250 if pp.gcMarkWorkerMode == gcMarkWorkerNotWorker {
1251 throw("gcBgMarkWorker: mode not set")
1254 startTime := nanotime()
1255 pp.gcMarkWorkerStartTime = startTime
1257 decnwait := atomic.Xadd(&work.nwait, -1)
1258 if decnwait == work.nproc {
1259 println("runtime: work.nwait=", decnwait, "work.nproc=", work.nproc)
1260 throw("work.nwait was > work.nproc")
1263 systemstack(func() {
1264 // Mark our goroutine preemptible so its stack
1265 // can be scanned. This lets two mark workers
1266 // scan each other (otherwise, they would
1267 // deadlock). We must not modify anything on
1268 // the G stack. However, stack shrinking is
1269 // disabled for mark workers, so it is safe to
1270 // read from the G stack.
1271 casgstatus(gp, _Grunning, _Gwaiting)
1272 switch pp.gcMarkWorkerMode {
1274 throw("gcBgMarkWorker: unexpected gcMarkWorkerMode")
1275 case gcMarkWorkerDedicatedMode:
1276 gcDrain(&pp.gcw, gcDrainUntilPreempt|gcDrainFlushBgCredit)
1278 // We were preempted. This is
1279 // a useful signal to kick
1280 // everything out of the run
1281 // queue so it can run
1283 if drainQ, n := runqdrain(pp); n > 0 {
1285 globrunqputbatch(&drainQ, int32(n))
1289 // Go back to draining, this time
1290 // without preemption.
1291 gcDrain(&pp.gcw, gcDrainFlushBgCredit)
1292 case gcMarkWorkerFractionalMode:
1293 gcDrain(&pp.gcw, gcDrainFractional|gcDrainUntilPreempt|gcDrainFlushBgCredit)
1294 case gcMarkWorkerIdleMode:
1295 gcDrain(&pp.gcw, gcDrainIdle|gcDrainUntilPreempt|gcDrainFlushBgCredit)
1297 casgstatus(gp, _Gwaiting, _Grunning)
1300 // Account for time and mark us as stopped.
1301 duration := nanotime() - startTime
1302 gcController.markWorkerStop(pp.gcMarkWorkerMode, duration)
1303 if pp.gcMarkWorkerMode == gcMarkWorkerFractionalMode {
1304 atomic.Xaddint64(&pp.gcFractionalMarkTime, duration)
1307 // Was this the last worker and did we run out
1309 incnwait := atomic.Xadd(&work.nwait, +1)
1310 if incnwait > work.nproc {
1311 println("runtime: p.gcMarkWorkerMode=", pp.gcMarkWorkerMode,
1312 "work.nwait=", incnwait, "work.nproc=", work.nproc)
1313 throw("work.nwait > work.nproc")
1316 // We'll releasem after this point and thus this P may run
1317 // something else. We must clear the worker mode to avoid
1318 // attributing the mode to a different (non-worker) G in
1320 pp.gcMarkWorkerMode = gcMarkWorkerNotWorker
1322 // If this worker reached a background mark completion
1323 // point, signal the main GC goroutine.
1324 if incnwait == work.nproc && !gcMarkWorkAvailable(nil) {
1325 // We don't need the P-local buffers here, allow
1326 // preemption because we may schedule like a regular
1327 // goroutine in gcMarkDone (block on locks, etc).
1328 releasem(node.m.ptr())
1336 // gcMarkWorkAvailable reports whether executing a mark worker
1337 // on p is potentially useful. p may be nil, in which case it only
1338 // checks the global sources of work.
1339 func gcMarkWorkAvailable(p *p) bool {
1340 if p != nil && !p.gcw.empty() {
1343 if !work.full.empty() {
1344 return true // global work available
1346 if work.markrootNext < work.markrootJobs {
1347 return true // root scan work available
1352 // gcMark runs the mark (or, for concurrent GC, mark termination)
1353 // All gcWork caches must be empty.
1354 // STW is in effect at this point.
1355 func gcMark(startTime int64) {
1356 if debug.allocfreetrace > 0 {
1360 if gcphase != _GCmarktermination {
1361 throw("in gcMark expecting to see gcphase as _GCmarktermination")
1363 work.tstart = startTime
1365 // Check that there's no marking work remaining.
1366 if work.full != 0 || work.markrootNext < work.markrootJobs {
1367 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")
1368 panic("non-empty mark queue after concurrent mark")
1371 if debug.gccheckmark > 0 {
1372 // This is expensive when there's a large number of
1373 // Gs, so only do it if checkmark is also enabled.
1377 throw("work.full != 0")
1380 // Drop allg snapshot. allgs may have grown, in which case
1381 // this is the only reference to the old backing store and
1382 // there's no need to keep it around.
1383 work.stackRoots = nil
1385 // Clear out buffers and double-check that all gcWork caches
1386 // are empty. This should be ensured by gcMarkDone before we
1387 // enter mark termination.
1389 // TODO: We could clear out buffers just before mark if this
1390 // has a non-negligible impact on STW time.
1391 for _, p := range allp {
1392 // The write barrier may have buffered pointers since
1393 // the gcMarkDone barrier. However, since the barrier
1394 // ensured all reachable objects were marked, all of
1395 // these must be pointers to black objects. Hence we
1396 // can just discard the write barrier buffer.
1397 if debug.gccheckmark > 0 {
1398 // For debugging, flush the buffer and make
1399 // sure it really was all marked.
1408 print("runtime: P ", p.id, " flushedWork ", gcw.flushedWork)
1409 if gcw.wbuf1 == nil {
1410 print(" wbuf1=<nil>")
1412 print(" wbuf1.n=", gcw.wbuf1.nobj)
1414 if gcw.wbuf2 == nil {
1415 print(" wbuf2=<nil>")
1417 print(" wbuf2.n=", gcw.wbuf2.nobj)
1420 throw("P has cached GC work at end of mark termination")
1422 // There may still be cached empty buffers, which we
1423 // need to flush since we're going to free them. Also,
1424 // there may be non-zero stats because we allocated
1425 // black after the gcMarkDone barrier.
1429 // Flush scanAlloc from each mcache since we're about to modify
1430 // heapScan directly. If we were to flush this later, then scanAlloc
1431 // might have incorrect information.
1433 // Note that it's not important to retain this information; we know
1434 // exactly what heapScan is at this point via scanWork.
1435 for _, p := range allp {
1443 // Reset controller state.
1444 gcController.resetLive(work.bytesMarked)
1447 // gcSweep must be called on the system stack because it acquires the heap
1448 // lock. See mheap for details.
1450 // The world must be stopped.
1453 func gcSweep(mode gcMode) {
1454 assertWorldStopped()
1456 if gcphase != _GCoff {
1457 throw("gcSweep being done but phase is not GCoff")
1461 mheap_.sweepgen += 2
1462 sweep.active.reset()
1463 mheap_.pagesSwept.Store(0)
1464 mheap_.sweepArenas = mheap_.allArenas
1465 mheap_.reclaimIndex.Store(0)
1466 mheap_.reclaimCredit.Store(0)
1467 unlock(&mheap_.lock)
1469 sweep.centralIndex.clear()
1471 if !_ConcurrentSweep || mode == gcForceBlockMode {
1472 // Special case synchronous sweep.
1473 // Record that no proportional sweeping has to happen.
1475 mheap_.sweepPagesPerByte = 0
1476 unlock(&mheap_.lock)
1477 // Sweep all spans eagerly.
1478 for sweepone() != ^uintptr(0) {
1481 // Free workbufs eagerly.
1482 prepareFreeWorkbufs()
1483 for freeSomeWbufs(false) {
1485 // All "free" events for this mark/sweep cycle have
1486 // now happened, so we can make this profile cycle
1487 // available immediately.
1493 // Background sweep.
1496 sweep.parked = false
1497 ready(sweep.g, 0, true)
1502 // gcResetMarkState resets global state prior to marking (concurrent
1503 // or STW) and resets the stack scan state of all Gs.
1505 // This is safe to do without the world stopped because any Gs created
1506 // during or after this will start out in the reset state.
1508 // gcResetMarkState must be called on the system stack because it acquires
1509 // the heap lock. See mheap for details.
1512 func gcResetMarkState() {
1513 // This may be called during a concurrent phase, so lock to make sure
1514 // allgs doesn't change.
1515 forEachG(func(gp *g) {
1516 gp.gcscandone = false // set to true in gcphasework
1517 gp.gcAssistBytes = 0
1520 // Clear page marks. This is just 1MB per 64GB of heap, so the
1521 // time here is pretty trivial.
1523 arenas := mheap_.allArenas
1524 unlock(&mheap_.lock)
1525 for _, ai := range arenas {
1526 ha := mheap_.arenas[ai.l1()][ai.l2()]
1527 for i := range ha.pageMarks {
1532 work.bytesMarked = 0
1533 work.initialHeapLive = atomic.Load64(&gcController.heapLive)
1536 // Hooks for other packages
1538 var poolcleanup func()
1539 var boringCaches []unsafe.Pointer // for crypto/internal/boring
1541 //go:linkname sync_runtime_registerPoolCleanup sync.runtime_registerPoolCleanup
1542 func sync_runtime_registerPoolCleanup(f func()) {
1546 //go:linkname boring_registerCache crypto/internal/boring.registerCache
1547 func boring_registerCache(p unsafe.Pointer) {
1548 boringCaches = append(boringCaches, p)
1553 if poolcleanup != nil {
1557 // clear boringcrypto caches
1558 for _, p := range boringCaches {
1559 atomicstorep(p, nil)
1562 // Clear central sudog cache.
1563 // Leave per-P caches alone, they have strictly bounded size.
1564 // Disconnect cached list before dropping it on the floor,
1565 // so that a dangling ref to one entry does not pin all of them.
1566 lock(&sched.sudoglock)
1567 var sg, sgnext *sudog
1568 for sg = sched.sudogcache; sg != nil; sg = sgnext {
1572 sched.sudogcache = nil
1573 unlock(&sched.sudoglock)
1575 // Clear central defer pool.
1576 // Leave per-P pools alone, they have strictly bounded size.
1577 lock(&sched.deferlock)
1578 // disconnect cached list before dropping it on the floor,
1579 // so that a dangling ref to one entry does not pin all of them.
1580 var d, dlink *_defer
1581 for d = sched.deferpool; d != nil; d = dlink {
1585 sched.deferpool = nil
1586 unlock(&sched.deferlock)
1591 // itoaDiv formats val/(10**dec) into buf.
1592 func itoaDiv(buf []byte, val uint64, dec int) []byte {
1595 for val >= 10 || i >= idec {
1596 buf[i] = byte(val%10 + '0')
1604 buf[i] = byte(val + '0')
1608 // fmtNSAsMS nicely formats ns nanoseconds as milliseconds.
1609 func fmtNSAsMS(buf []byte, ns uint64) []byte {
1611 // Format as whole milliseconds.
1612 return itoaDiv(buf, ns/1e6, 0)
1614 // Format two digits of precision, with at most three decimal places.
1625 return itoaDiv(buf, x, dec)
1628 // Helpers for testing GC.
1630 // gcTestMoveStackOnNextCall causes the stack to be moved on a call
1631 // immediately following the call to this. It may not work correctly
1632 // if any other work appears after this call (such as returning).
1633 // Typically the following call should be marked go:noinline so it
1634 // performs a stack check.
1636 // In rare cases this may not cause the stack to move, specifically if
1637 // there's a preemption between this call and the next.
1638 func gcTestMoveStackOnNextCall() {
1640 gp.stackguard0 = stackForceMove
1643 // gcTestIsReachable performs a GC and returns a bit set where bit i
1644 // is set if ptrs[i] is reachable.
1645 func gcTestIsReachable(ptrs ...unsafe.Pointer) (mask uint64) {
1646 // This takes the pointers as unsafe.Pointers in order to keep
1647 // them live long enough for us to attach specials. After
1648 // that, we drop our references to them.
1651 panic("too many pointers for uint64 mask")
1654 // Block GC while we attach specials and drop our references
1655 // to ptrs. Otherwise, if a GC is in progress, it could mark
1656 // them reachable via this function before we have a chance to
1660 // Create reachability specials for ptrs.
1661 specials := make([]*specialReachable, len(ptrs))
1662 for i, p := range ptrs {
1663 lock(&mheap_.speciallock)
1664 s := (*specialReachable)(mheap_.specialReachableAlloc.alloc())
1665 unlock(&mheap_.speciallock)
1666 s.special.kind = _KindSpecialReachable
1667 if !addspecial(p, &s.special) {
1668 throw("already have a reachable special (duplicate pointer?)")
1671 // Make sure we don't retain ptrs.
1677 // Force a full GC and sweep.
1680 // Process specials.
1681 for i, s := range specials {
1684 println("runtime: object", i, "was not swept")
1685 throw("IsReachable failed")
1690 lock(&mheap_.speciallock)
1691 mheap_.specialReachableAlloc.free(unsafe.Pointer(s))
1692 unlock(&mheap_.speciallock)
1698 // gcTestPointerClass returns the category of what p points to, one of:
1699 // "heap", "stack", "data", "bss", "other". This is useful for checking
1700 // that a test is doing what it's intended to do.
1702 // This is nosplit simply to avoid extra pointer shuffling that may
1703 // complicate a test.
1706 func gcTestPointerClass(p unsafe.Pointer) string {
1707 p2 := uintptr(noescape(p))
1709 if gp.stack.lo <= p2 && p2 < gp.stack.hi {
1712 if base, _, _ := findObject(p2, 0, 0); base != 0 {
1715 for _, datap := range activeModules() {
1716 if datap.data <= p2 && p2 < datap.edata || datap.noptrdata <= p2 && p2 < datap.enoptrdata {
1719 if datap.bss <= p2 && p2 < datap.ebss || datap.noptrbss <= p2 && p2 <= datap.enoptrbss {