Source file src/runtime/mgc.go

     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.
     4  
     5  // Garbage collector (GC).
     6  //
     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.
    11  //
    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.
    15  //
    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),
    19  // 966-975.
    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.
    23  //
    24  // 1. GC performs sweep termination.
    25  //
    26  //    a. Stop the world. This causes all Ps to reach a GC safe-point.
    27  //
    28  //    b. Sweep any unswept spans. There will only be unswept spans if
    29  //    this GC cycle was forced before the expected time.
    30  //
    31  // 2. GC performs the mark phase.
    32  //
    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.
    38  //
    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.
    45  //
    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.
    51  //
    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).
    55  //
    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.
    60  //
    61  // 3. GC performs mark termination.
    62  //
    63  //    a. Stop the world.
    64  //
    65  //    b. Set gcphase to _GCmarktermination, and disable workers and
    66  //    assists.
    67  //
    68  //    c. Perform housekeeping like flushing mcaches.
    69  //
    70  // 4. GC performs the sweep phase.
    71  //
    72  //    a. Prepare for the sweep phase by setting gcphase to _GCoff,
    73  //    setting up sweep state and disabling the write barrier.
    74  //
    75  //    b. Start the world. From this point on, newly allocated objects
    76  //    are white, and allocating sweeps spans before use if necessary.
    77  //
    78  //    c. GC does concurrent sweeping in the background and in response
    79  //    to allocation. See description below.
    80  //
    81  // 5. When sufficient allocation has taken place, replay the sequence
    82  // starting with 1 above. See discussion of GC rate below.
    83  
    84  // Concurrent sweep.
    85  //
    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".
    90  //
    91  // The background sweeper goroutine simply sweeps spans one-by-one.
    92  //
    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.
   103  //
   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).
   111  
   112  // GC rate.
   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).
   119  
   120  // Oblets
   121  //
   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.
   128  
   129  package runtime
   130  
   131  import (
   132  	"internal/cpu"
   133  	"runtime/internal/atomic"
   134  	"unsafe"
   135  )
   136  
   137  const (
   138  	_DebugGC      = 0
   139  	_FinBlockSize = 4 * 1024
   140  
   141  	// concurrentSweep is a debug flag. Disabling this flag
   142  	// ensures all spans are swept while the world is stopped.
   143  	concurrentSweep = true
   144  
   145  	// debugScanConservative enables debug logging for stack
   146  	// frames that are scanned conservatively.
   147  	debugScanConservative = false
   148  
   149  	// sweepMinHeapDistance is a lower bound on the heap distance
   150  	// (in bytes) reserved for concurrent sweeping between GC
   151  	// cycles.
   152  	sweepMinHeapDistance = 1024 * 1024
   153  )
   154  
   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.
   168  //
   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.
   171  //
   172  //go:linkname heapObjectsCanMove
   173  func heapObjectsCanMove() bool {
   174  	return false
   175  }
   176  
   177  func gcinit() {
   178  	if unsafe.Sizeof(workbuf{}) != _WorkbufSize {
   179  		throw("size of Workbuf is suboptimal")
   180  	}
   181  	// No sweep on the first cycle.
   182  	sweep.active.state.Store(sweepDrainedMask)
   183  
   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())
   188  
   189  	work.startSema = 1
   190  	work.markDoneSema = 1
   191  	lockInit(&work.sweepWaiters.lock, lockRankSweepWaiters)
   192  	lockInit(&work.assistQueue.lock, lockRankAssistQueue)
   193  	lockInit(&work.wbufSpans.lock, lockRankWbufSpans)
   194  }
   195  
   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.
   200  func gcenable() {
   201  	// Kick off sweeping and scavenging.
   202  	c := make(chan int, 2)
   203  	go bgsweep(c)
   204  	go bgscavenge(c)
   205  	<-c
   206  	<-c
   207  	memstats.enablegc = true // now that runtime is initialized, GC is okay
   208  }
   209  
   210  // Garbage collector phase.
   211  // Indicates to write barrier and synchronization task to perform.
   212  var gcphase uint32
   213  
   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
   222  }
   223  
   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
   228  
   229  const (
   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
   233  )
   234  
   235  //go:nosplit
   236  func setGCPhase(x uint32) {
   237  	atomic.Store(&gcphase, x)
   238  	writeBarrier.enabled = gcphase == _GCmark || gcphase == _GCmarktermination
   239  }
   240  
   241  // gcMarkWorkerMode represents the mode that a concurrent mark worker
   242  // should operate in.
   243  //
   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
   249  
   250  const (
   251  	// gcMarkWorkerNotWorker indicates that the next scheduled G is not
   252  	// starting work and the mode should be ignored.
   253  	gcMarkWorkerNotWorker gcMarkWorkerMode = iota
   254  
   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
   259  
   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
   269  
   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.
   274  	gcMarkWorkerIdleMode
   275  )
   276  
   277  // gcMarkWorkerModeStrings are the strings labels of gcMarkWorkerModes
   278  // to use in execution traces.
   279  var gcMarkWorkerModeStrings = [...]string{
   280  	"Not worker",
   281  	"GC (dedicated)",
   282  	"GC (fractional)",
   283  	"GC (idle)",
   284  }
   285  
   286  // pollFractionalWorkerExit reports whether a fractional mark worker
   287  // should self-preempt. It assumes it is called from the fractional
   288  // worker.
   289  func pollFractionalWorkerExit() bool {
   290  	// This should be kept in sync with the fractional worker
   291  	// scheduler logic in findRunnableGCWorker.
   292  	now := nanotime()
   293  	delta := now - gcController.markStartTime
   294  	if delta <= 0 {
   295  		return true
   296  	}
   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
   302  }
   303  
   304  var work workType
   305  
   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
   311  
   312  	wbufSpans struct {
   313  		lock mutex
   314  		// free is a list of spans dedicated to workbufs, but
   315  		// that don't currently contain any workbufs.
   316  		free mSpanList
   317  		// busy is a list of all spans containing workbufs on
   318  		// one of the workbuf lists.
   319  		busy mSpanList
   320  	}
   321  
   322  	// Restore 64-bit alignment on 32-bit.
   323  	_ uint32
   324  
   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
   331  	// cycle.
   332  	//
   333  	// Because of benign races during marking, this number may not
   334  	// be the exact number of marked bytes, but it should be very
   335  	// close.
   336  	//
   337  	// Put this field here because it needs 64-bit atomic access
   338  	// (and thus 8-byte alignment even on 32-bit architectures).
   339  	bytesMarked uint64
   340  
   341  	markrootNext uint32 // next markroot job
   342  	markrootJobs uint32 // number of markroot jobs
   343  
   344  	nproc  uint32
   345  	tstart int64
   346  	nwait  uint32
   347  
   348  	// Number of roots of various root types. Set by gcMarkRootPrepare.
   349  	//
   350  	// nStackRoots == len(stackRoots), but we have nStackRoots for
   351  	// consistency.
   352  	nDataRoots, nBSSRoots, nSpanRoots, nStackRoots int
   353  
   354  	// Base indexes of each root type. Set by gcMarkRootPrepare.
   355  	baseData, baseBSS, baseSpans, baseStacks, baseEnd uint32
   356  
   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.
   361  	stackRoots []*g
   362  
   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
   373  	// happened.
   374  	//
   375  	// startSema protects the transition from "off" to mark or
   376  	// mark termination.
   377  	startSema uint32
   378  	// markDoneSema protects transitions from mark to mark termination.
   379  	markDoneSema uint32
   380  
   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
   384  
   385  	// mode is the concurrency mode of the current GC cycle.
   386  	mode gcMode
   387  
   388  	// userForced indicates the current GC cycle was forced by an
   389  	// explicit user call.
   390  	userForced bool
   391  
   392  	// initialHeapLive is the value of gcController.heapLive at the
   393  	// beginning of this GC cycle.
   394  	initialHeapLive uint64
   395  
   396  	// assistQueue is a queue of assists that are blocked because
   397  	// there was neither enough credit to steal or enough work to
   398  	// do.
   399  	assistQueue struct {
   400  		lock mutex
   401  		q    gQueue
   402  	}
   403  
   404  	// sweepWaiters is a list of blocked goroutines to wake when
   405  	// we transition from mark termination to sweep.
   406  	sweepWaiters struct {
   407  		lock mutex
   408  		list gList
   409  	}
   410  
   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.
   415  	cycles atomic.Uint32
   416  
   417  	// Timing/utilization stats for this cycle.
   418  	stwprocs, maxprocs                 int32
   419  	tSweepTerm, tMark, tMarkTerm, tEnd int64 // nanotime() of phase start
   420  
   421  	pauseNS int64 // total STW time this cycle
   422  
   423  	// debug.gctrace heap sizes for this cycle.
   424  	heap0, heap1, heap2 uint64
   425  
   426  	// Cumulative estimated CPU usage.
   427  	cpuStats
   428  }
   429  
   430  // GC runs a garbage collection and blocks the caller until the
   431  // garbage collection is complete. It may also block the entire
   432  // program.
   433  func GC() {
   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:
   439  	//
   440  	// 1. In sweep termination, mark, or mark termination of cycle
   441  	// N, wait until mark termination N completes and transitions
   442  	// to sweep N.
   443  	//
   444  	// 2. In sweep N, help with sweep N.
   445  	//
   446  	// At this point we can begin a full cycle N+1.
   447  	//
   448  	// 3. Trigger cycle N+1 by starting sweep termination N+1.
   449  	//
   450  	// 4. Wait for mark termination N+1 to complete.
   451  	//
   452  	// 5. Help with sweep N+1 until it's done.
   453  	//
   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.
   457  
   458  	// Wait until the current sweep termination, mark, and mark
   459  	// termination complete.
   460  	n := work.cycles.Load()
   461  	gcWaitOnMark(n)
   462  
   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
   465  	// termination N+1.
   466  	gcStart(gcTrigger{kind: gcTriggerCycle, n: n + 1})
   467  
   468  	// Wait for mark termination N+1 to complete.
   469  	gcWaitOnMark(n + 1)
   470  
   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) {
   476  		Gosched()
   477  	}
   478  
   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.
   483  	//
   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.
   486  	//
   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() {
   491  		Gosched()
   492  	}
   493  
   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.
   497  	mp := acquirem()
   498  	cycle := work.cycles.Load()
   499  	if cycle == n+1 || (gcphase == _GCmark && cycle == n+2) {
   500  		mProf_PostSweep()
   501  	}
   502  	releasem(mp)
   503  }
   504  
   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) {
   508  	for {
   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.
   514  			nMarks++
   515  		}
   516  		if nMarks > n {
   517  			// We're done.
   518  			unlock(&work.sweepWaiters.lock)
   519  			return
   520  		}
   521  
   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)
   526  	}
   527  }
   528  
   529  // gcMode indicates how concurrent a GC cycle should be.
   530  type gcMode int
   531  
   532  const (
   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)
   536  )
   537  
   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 {
   541  	kind gcTriggerKind
   542  	now  int64  // gcTriggerTime: current time
   543  	n    uint32 // gcTriggerCycle: cycle number to start
   544  }
   545  
   546  type gcTriggerKind int
   547  
   548  const (
   549  	// gcTriggerHeap indicates that a cycle should be started when
   550  	// the heap size reaches the trigger heap size computed by the
   551  	// controller.
   552  	gcTriggerHeap gcTriggerKind = iota
   553  
   554  	// gcTriggerTime indicates that a cycle should be started when
   555  	// it's been more than forcegcperiod nanoseconds since the
   556  	// previous GC cycle.
   557  	gcTriggerTime
   558  
   559  	// gcTriggerCycle indicates that a cycle should be started if
   560  	// we have not yet started cycle number gcTrigger.n (relative
   561  	// to work.cycles).
   562  	gcTriggerCycle
   563  )
   564  
   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 {
   570  		return false
   571  	}
   572  	switch t.kind {
   573  	case gcTriggerHeap:
   574  		trigger, _ := gcController.trigger()
   575  		return gcController.heapLive.Load() >= trigger
   576  	case gcTriggerTime:
   577  		if gcController.gcPercent.Load() < 0 {
   578  			return false
   579  		}
   580  		lastgc := int64(atomic.Load64(&memstats.last_gc_nanotime))
   581  		return lastgc != 0 && t.now-lastgc > forcegcperiod
   582  	case gcTriggerCycle:
   583  		// t.n > work.cycles, but accounting for wraparound.
   584  		return int32(t.n-work.cycles.Load()) > 0
   585  	}
   586  	return true
   587  }
   588  
   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).
   592  //
   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.
   600  	mp := acquirem()
   601  	if gp := getg(); gp == mp.g0 || mp.locks > 1 || mp.preemptoff != "" {
   602  		releasem(mp)
   603  		return
   604  	}
   605  	releasem(mp)
   606  	mp = nil
   607  
   608  	// Pick up the remaining unswept/not being swept spans concurrently
   609  	//
   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.
   615  	//
   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) {
   619  	}
   620  
   621  	// Perform GC initialization and the sweep termination
   622  	// transition.
   623  	semacquire(&work.startSema)
   624  	// Re-check transition condition under transition lock.
   625  	if !trigger.test() {
   626  		semrelease(&work.startSema)
   627  		return
   628  	}
   629  
   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 {
   636  		mode = gcForceMode
   637  	} else if debug.gcstoptheworld == 2 {
   638  		mode = gcForceBlockMode
   639  	}
   640  
   641  	// Ok, we're doing it! Stop everybody else
   642  	semacquire(&gcsema)
   643  	semacquire(&worldsema)
   644  
   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
   648  
   649  	trace := traceAcquire()
   650  	if trace.ok() {
   651  		trace.GCStart()
   652  		traceRelease(trace)
   653  	}
   654  
   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")
   660  		}
   661  	}
   662  
   663  	gcBgMarkStartWorkers()
   664  
   665  	systemstack(gcResetMarkState)
   666  
   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.
   671  		work.stwprocs = ncpu
   672  	}
   673  	work.heap0 = gcController.heapLive.Load()
   674  	work.pauseNS = 0
   675  	work.mode = mode
   676  
   677  	now := nanotime()
   678  	work.tSweepTerm = now
   679  	var stw worldStop
   680  	systemstack(func() {
   681  		stw = stopTheWorldWithSema(stwGCSweepTerm)
   682  	})
   683  	// Finish sweep before we start concurrent scan.
   684  	systemstack(func() {
   685  		finishsweep_m()
   686  	})
   687  
   688  	// clearpools before we start the GC. If we wait the memory will not be
   689  	// reclaimed until the next GC cycle.
   690  	clearpools()
   691  
   692  	work.cycles.Add(1)
   693  
   694  	// Assists and workers can start the moment we start
   695  	// the world.
   696  	gcController.startCycle(now, int(gomaxprocs), trigger)
   697  
   698  	// Notify the CPU limiter that assists may begin.
   699  	gcCPULimiter.startGCTransition(true, now)
   700  
   701  	// In STW mode, disable scheduling of user Gs. This may also
   702  	// disable scheduling of this goroutine, so it may block as
   703  	// soon as we start the world again.
   704  	if mode != gcBackgroundMode {
   705  		schedEnableUser(false)
   706  	}
   707  
   708  	// Enter concurrent mark phase and enable
   709  	// write barriers.
   710  	//
   711  	// Because the world is stopped, all Ps will
   712  	// observe that write barriers are enabled by
   713  	// the time we start the world and begin
   714  	// scanning.
   715  	//
   716  	// Write barriers must be enabled before assists are
   717  	// enabled because they must be enabled before
   718  	// any non-leaf heap objects are marked. Since
   719  	// allocations are blocked until assists can
   720  	// happen, we want to enable assists as early as
   721  	// possible.
   722  	setGCPhase(_GCmark)
   723  
   724  	gcBgMarkPrepare() // Must happen before assists are enabled.
   725  	gcMarkRootPrepare()
   726  
   727  	// Mark all active tinyalloc blocks. Since we're
   728  	// allocating from these, they need to be black like
   729  	// other allocations. The alternative is to blacken
   730  	// the tiny block on every allocation from it, which
   731  	// would slow down the tiny allocator.
   732  	gcMarkTinyAllocs()
   733  
   734  	// At this point all Ps have enabled the write
   735  	// barrier, thus maintaining the no white to
   736  	// black invariant. Enable mutator assists to
   737  	// put back-pressure on fast allocating
   738  	// mutators.
   739  	atomic.Store(&gcBlackenEnabled, 1)
   740  
   741  	// In STW mode, we could block the instant systemstack
   742  	// returns, so make sure we're not preemptible.
   743  	mp = acquirem()
   744  
   745  	// Concurrent mark.
   746  	systemstack(func() {
   747  		now = startTheWorldWithSema(0, stw)
   748  		work.pauseNS += now - stw.start
   749  		work.tMark = now
   750  
   751  		sweepTermCpu := int64(work.stwprocs) * (work.tMark - work.tSweepTerm)
   752  		work.cpuStats.gcPauseTime += sweepTermCpu
   753  		work.cpuStats.gcTotalTime += sweepTermCpu
   754  
   755  		// Release the CPU limiter.
   756  		gcCPULimiter.finishGCTransition(now)
   757  	})
   758  
   759  	// Release the world sema before Gosched() in STW mode
   760  	// because we will need to reacquire it later but before
   761  	// this goroutine becomes runnable again, and we could
   762  	// self-deadlock otherwise.
   763  	semrelease(&worldsema)
   764  	releasem(mp)
   765  
   766  	// Make sure we block instead of returning to user code
   767  	// in STW mode.
   768  	if mode != gcBackgroundMode {
   769  		Gosched()
   770  	}
   771  
   772  	semrelease(&work.startSema)
   773  }
   774  
   775  // gcMarkDoneFlushed counts the number of P's with flushed work.
   776  //
   777  // Ideally this would be a captured local in gcMarkDone, but forEachP
   778  // escapes its callback closure, so it can't capture anything.
   779  //
   780  // This is protected by markDoneSema.
   781  var gcMarkDoneFlushed uint32
   782  
   783  // gcMarkDone transitions the GC from mark to mark termination if all
   784  // reachable objects have been marked (that is, there are no grey
   785  // objects and can be no more in the future). Otherwise, it flushes
   786  // all local work to the global queues where it can be discovered by
   787  // other workers.
   788  //
   789  // This should be called when all local mark work has been drained and
   790  // there are no remaining workers. Specifically, when
   791  //
   792  //	work.nwait == work.nproc && !gcMarkWorkAvailable(p)
   793  //
   794  // The calling context must be preemptible.
   795  //
   796  // Flushing local work is important because idle Ps may have local
   797  // work queued. This is the only way to make that work visible and
   798  // drive GC to completion.
   799  //
   800  // It is explicitly okay to have write barriers in this function. If
   801  // it does transition to mark termination, then all reachable objects
   802  // have been marked, so the write barrier cannot shade any more
   803  // objects.
   804  func gcMarkDone() {
   805  	// Ensure only one thread is running the ragged barrier at a
   806  	// time.
   807  	semacquire(&work.markDoneSema)
   808  
   809  top:
   810  	// Re-check transition condition under transition lock.
   811  	//
   812  	// It's critical that this checks the global work queues are
   813  	// empty before performing the ragged barrier. Otherwise,
   814  	// there could be global work that a P could take after the P
   815  	// has passed the ragged barrier.
   816  	if !(gcphase == _GCmark && work.nwait == work.nproc && !gcMarkWorkAvailable(nil)) {
   817  		semrelease(&work.markDoneSema)
   818  		return
   819  	}
   820  
   821  	// forEachP needs worldsema to execute, and we'll need it to
   822  	// stop the world later, so acquire worldsema now.
   823  	semacquire(&worldsema)
   824  
   825  	// Flush all local buffers and collect flushedWork flags.
   826  	gcMarkDoneFlushed = 0
   827  	forEachP(waitReasonGCMarkTermination, func(pp *p) {
   828  		// Flush the write barrier buffer, since this may add
   829  		// work to the gcWork.
   830  		wbBufFlush1(pp)
   831  
   832  		// Flush the gcWork, since this may create global work
   833  		// and set the flushedWork flag.
   834  		//
   835  		// TODO(austin): Break up these workbufs to
   836  		// better distribute work.
   837  		pp.gcw.dispose()
   838  		// Collect the flushedWork flag.
   839  		if pp.gcw.flushedWork {
   840  			atomic.Xadd(&gcMarkDoneFlushed, 1)
   841  			pp.gcw.flushedWork = false
   842  		}
   843  	})
   844  
   845  	if gcMarkDoneFlushed != 0 {
   846  		// More grey objects were discovered since the
   847  		// previous termination check, so there may be more
   848  		// work to do. Keep going. It's possible the
   849  		// transition condition became true again during the
   850  		// ragged barrier, so re-check it.
   851  		semrelease(&worldsema)
   852  		goto top
   853  	}
   854  
   855  	// There was no global work, no local work, and no Ps
   856  	// communicated work since we took markDoneSema. Therefore
   857  	// there are no grey objects and no more objects can be
   858  	// shaded. Transition to mark termination.
   859  	now := nanotime()
   860  	work.tMarkTerm = now
   861  	getg().m.preemptoff = "gcing"
   862  	var stw worldStop
   863  	systemstack(func() {
   864  		stw = stopTheWorldWithSema(stwGCMarkTerm)
   865  	})
   866  	// The gcphase is _GCmark, it will transition to _GCmarktermination
   867  	// below. The important thing is that the wb remains active until
   868  	// all marking is complete. This includes writes made by the GC.
   869  
   870  	// There is sometimes work left over when we enter mark termination due
   871  	// to write barriers performed after the completion barrier above.
   872  	// Detect this and resume concurrent mark. This is obviously
   873  	// unfortunate.
   874  	//
   875  	// See issue #27993 for details.
   876  	//
   877  	// Switch to the system stack to call wbBufFlush1, though in this case
   878  	// it doesn't matter because we're non-preemptible anyway.
   879  	restart := false
   880  	systemstack(func() {
   881  		for _, p := range allp {
   882  			wbBufFlush1(p)
   883  			if !p.gcw.empty() {
   884  				restart = true
   885  				break
   886  			}
   887  		}
   888  	})
   889  	if restart {
   890  		getg().m.preemptoff = ""
   891  		systemstack(func() {
   892  			now := startTheWorldWithSema(0, stw)
   893  			work.pauseNS += now - stw.start
   894  		})
   895  		semrelease(&worldsema)
   896  		goto top
   897  	}
   898  
   899  	gcComputeStartingStackSize()
   900  
   901  	// Disable assists and background workers. We must do
   902  	// this before waking blocked assists.
   903  	atomic.Store(&gcBlackenEnabled, 0)
   904  
   905  	// Notify the CPU limiter that GC assists will now cease.
   906  	gcCPULimiter.startGCTransition(false, now)
   907  
   908  	// Wake all blocked assists. These will run when we
   909  	// start the world again.
   910  	gcWakeAllAssists()
   911  
   912  	// Likewise, release the transition lock. Blocked
   913  	// workers and assists will run when we start the
   914  	// world again.
   915  	semrelease(&work.markDoneSema)
   916  
   917  	// In STW mode, re-enable user goroutines. These will be
   918  	// queued to run after we start the world.
   919  	schedEnableUser(true)
   920  
   921  	// endCycle depends on all gcWork cache stats being flushed.
   922  	// The termination algorithm above ensured that up to
   923  	// allocations since the ragged barrier.
   924  	gcController.endCycle(now, int(gomaxprocs), work.userForced)
   925  
   926  	// Perform mark termination. This will restart the world.
   927  	gcMarkTermination(stw)
   928  }
   929  
   930  // World must be stopped and mark assists and background workers must be
   931  // disabled.
   932  func gcMarkTermination(stw worldStop) {
   933  	// Start marktermination (write barrier remains enabled for now).
   934  	setGCPhase(_GCmarktermination)
   935  
   936  	work.heap1 = gcController.heapLive.Load()
   937  	startTime := nanotime()
   938  
   939  	mp := acquirem()
   940  	mp.preemptoff = "gcing"
   941  	mp.traceback = 2
   942  	curgp := mp.curg
   943  	// N.B. The execution tracer is not aware of this status
   944  	// transition and handles it specially based on the
   945  	// wait reason.
   946  	casGToWaiting(curgp, _Grunning, waitReasonGarbageCollection)
   947  
   948  	// Run gc on the g0 stack. We do this so that the g stack
   949  	// we're currently running on will no longer change. Cuts
   950  	// the root set down a bit (g0 stacks are not scanned, and
   951  	// we don't need to scan gc's internal state).  We also
   952  	// need to switch to g0 so we can shrink the stack.
   953  	systemstack(func() {
   954  		gcMark(startTime)
   955  		// Must return immediately.
   956  		// The outer function's stack may have moved
   957  		// during gcMark (it shrinks stacks, including the
   958  		// outer function's stack), so we must not refer
   959  		// to any of its variables. Return back to the
   960  		// non-system stack to pick up the new addresses
   961  		// before continuing.
   962  	})
   963  
   964  	var stwSwept bool
   965  	systemstack(func() {
   966  		work.heap2 = work.bytesMarked
   967  		if debug.gccheckmark > 0 {
   968  			// Run a full non-parallel, stop-the-world
   969  			// mark using checkmark bits, to check that we
   970  			// didn't forget to mark anything during the
   971  			// concurrent mark process.
   972  			startCheckmarks()
   973  			gcResetMarkState()
   974  			gcw := &getg().m.p.ptr().gcw
   975  			gcDrain(gcw, 0)
   976  			wbBufFlush1(getg().m.p.ptr())
   977  			gcw.dispose()
   978  			endCheckmarks()
   979  		}
   980  
   981  		// marking is complete so we can turn the write barrier off
   982  		setGCPhase(_GCoff)
   983  		stwSwept = gcSweep(work.mode)
   984  	})
   985  
   986  	mp.traceback = 0
   987  	casgstatus(curgp, _Gwaiting, _Grunning)
   988  
   989  	trace := traceAcquire()
   990  	if trace.ok() {
   991  		trace.GCDone()
   992  		traceRelease(trace)
   993  	}
   994  
   995  	// all done
   996  	mp.preemptoff = ""
   997  
   998  	if gcphase != _GCoff {
   999  		throw("gc done but gcphase != _GCoff")
  1000  	}
  1001  
  1002  	// Record heapInUse for scavenger.
  1003  	memstats.lastHeapInUse = gcController.heapInUse.load()
  1004  
  1005  	// Update GC trigger and pacing, as well as downstream consumers
  1006  	// of this pacing information, for the next cycle.
  1007  	systemstack(gcControllerCommit)
  1008  
  1009  	// Update timing memstats
  1010  	now := nanotime()
  1011  	sec, nsec, _ := time_now()
  1012  	unixNow := sec*1e9 + int64(nsec)
  1013  	work.pauseNS += now - stw.start
  1014  	work.tEnd = now
  1015  	atomic.Store64(&memstats.last_gc_unix, uint64(unixNow)) // must be Unix time to make sense to user
  1016  	atomic.Store64(&memstats.last_gc_nanotime, uint64(now)) // monotonic time for us
  1017  	memstats.pause_ns[memstats.numgc%uint32(len(memstats.pause_ns))] = uint64(work.pauseNS)
  1018  	memstats.pause_end[memstats.numgc%uint32(len(memstats.pause_end))] = uint64(unixNow)
  1019  	memstats.pause_total_ns += uint64(work.pauseNS)
  1020  
  1021  	markTermCpu := int64(work.stwprocs) * (work.tEnd - work.tMarkTerm)
  1022  	work.cpuStats.gcPauseTime += markTermCpu
  1023  	work.cpuStats.gcTotalTime += markTermCpu
  1024  
  1025  	// Accumulate CPU stats.
  1026  	//
  1027  	// Pass gcMarkPhase=true so we can get all the latest GC CPU stats in there too.
  1028  	work.cpuStats.accumulate(now, true)
  1029  
  1030  	// Compute overall GC CPU utilization.
  1031  	// Omit idle marking time from the overall utilization here since it's "free".
  1032  	memstats.gc_cpu_fraction = float64(work.cpuStats.gcTotalTime-work.cpuStats.gcIdleTime) / float64(work.cpuStats.totalTime)
  1033  
  1034  	// Reset assist time and background time stats.
  1035  	//
  1036  	// Do this now, instead of at the start of the next GC cycle, because
  1037  	// these two may keep accumulating even if the GC is not active.
  1038  	scavenge.assistTime.Store(0)
  1039  	scavenge.backgroundTime.Store(0)
  1040  
  1041  	// Reset idle time stat.
  1042  	sched.idleTime.Store(0)
  1043  
  1044  	if work.userForced {
  1045  		memstats.numforcedgc++
  1046  	}
  1047  
  1048  	// Bump GC cycle count and wake goroutines waiting on sweep.
  1049  	lock(&work.sweepWaiters.lock)
  1050  	memstats.numgc++
  1051  	injectglist(&work.sweepWaiters.list)
  1052  	unlock(&work.sweepWaiters.lock)
  1053  
  1054  	// Increment the scavenge generation now.
  1055  	//
  1056  	// This moment represents peak heap in use because we're
  1057  	// about to start sweeping.
  1058  	mheap_.pages.scav.index.nextGen()
  1059  
  1060  	// Release the CPU limiter.
  1061  	gcCPULimiter.finishGCTransition(now)
  1062  
  1063  	// Finish the current heap profiling cycle and start a new
  1064  	// heap profiling cycle. We do this before starting the world
  1065  	// so events don't leak into the wrong cycle.
  1066  	mProf_NextCycle()
  1067  
  1068  	// There may be stale spans in mcaches that need to be swept.
  1069  	// Those aren't tracked in any sweep lists, so we need to
  1070  	// count them against sweep completion until we ensure all
  1071  	// those spans have been forced out.
  1072  	//
  1073  	// If gcSweep fully swept the heap (for example if the sweep
  1074  	// is not concurrent due to a GODEBUG setting), then we expect
  1075  	// the sweepLocker to be invalid, since sweeping is done.
  1076  	//
  1077  	// N.B. Below we might duplicate some work from gcSweep; this is
  1078  	// fine as all that work is idempotent within a GC cycle, and
  1079  	// we're still holding worldsema so a new cycle can't start.
  1080  	sl := sweep.active.begin()
  1081  	if !stwSwept && !sl.valid {
  1082  		throw("failed to set sweep barrier")
  1083  	} else if stwSwept && sl.valid {
  1084  		throw("non-concurrent sweep failed to drain all sweep queues")
  1085  	}
  1086  
  1087  	systemstack(func() {
  1088  		// The memstats updated above must be updated with the world
  1089  		// stopped to ensure consistency of some values, such as
  1090  		// sched.idleTime and sched.totaltime. memstats also include
  1091  		// the pause time (work,pauseNS), forcing computation of the
  1092  		// total pause time before the pause actually ends.
  1093  		//
  1094  		// Here we reuse the same now for start the world so that the
  1095  		// time added to /sched/pauses/total/gc:seconds will be
  1096  		// consistent with the value in memstats.
  1097  		startTheWorldWithSema(now, stw)
  1098  	})
  1099  
  1100  	// Flush the heap profile so we can start a new cycle next GC.
  1101  	// This is relatively expensive, so we don't do it with the
  1102  	// world stopped.
  1103  	mProf_Flush()
  1104  
  1105  	// Prepare workbufs for freeing by the sweeper. We do this
  1106  	// asynchronously because it can take non-trivial time.
  1107  	prepareFreeWorkbufs()
  1108  
  1109  	// Free stack spans. This must be done between GC cycles.
  1110  	systemstack(freeStackSpans)
  1111  
  1112  	// Ensure all mcaches are flushed. Each P will flush its own
  1113  	// mcache before allocating, but idle Ps may not. Since this
  1114  	// is necessary to sweep all spans, we need to ensure all
  1115  	// mcaches are flushed before we start the next GC cycle.
  1116  	//
  1117  	// While we're here, flush the page cache for idle Ps to avoid
  1118  	// having pages get stuck on them. These pages are hidden from
  1119  	// the scavenger, so in small idle heaps a significant amount
  1120  	// of additional memory might be held onto.
  1121  	//
  1122  	// Also, flush the pinner cache, to avoid leaking that memory
  1123  	// indefinitely.
  1124  	forEachP(waitReasonFlushProcCaches, func(pp *p) {
  1125  		pp.mcache.prepareForSweep()
  1126  		if pp.status == _Pidle {
  1127  			systemstack(func() {
  1128  				lock(&mheap_.lock)
  1129  				pp.pcache.flush(&mheap_.pages)
  1130  				unlock(&mheap_.lock)
  1131  			})
  1132  		}
  1133  		pp.pinnerCache = nil
  1134  	})
  1135  	if sl.valid {
  1136  		// Now that we've swept stale spans in mcaches, they don't
  1137  		// count against unswept spans.
  1138  		//
  1139  		// Note: this sweepLocker may not be valid if sweeping had
  1140  		// already completed during the STW. See the corresponding
  1141  		// begin() call that produced sl.
  1142  		sweep.active.end(sl)
  1143  	}
  1144  
  1145  	// Print gctrace before dropping worldsema. As soon as we drop
  1146  	// worldsema another cycle could start and smash the stats
  1147  	// we're trying to print.
  1148  	if debug.gctrace > 0 {
  1149  		util := int(memstats.gc_cpu_fraction * 100)
  1150  
  1151  		var sbuf [24]byte
  1152  		printlock()
  1153  		print("gc ", memstats.numgc,
  1154  			" @", string(itoaDiv(sbuf[:], uint64(work.tSweepTerm-runtimeInitTime)/1e6, 3)), "s ",
  1155  			util, "%: ")
  1156  		prev := work.tSweepTerm
  1157  		for i, ns := range []int64{work.tMark, work.tMarkTerm, work.tEnd} {
  1158  			if i != 0 {
  1159  				print("+")
  1160  			}
  1161  			print(string(fmtNSAsMS(sbuf[:], uint64(ns-prev))))
  1162  			prev = ns
  1163  		}
  1164  		print(" ms clock, ")
  1165  		for i, ns := range []int64{
  1166  			int64(work.stwprocs) * (work.tMark - work.tSweepTerm),
  1167  			gcController.assistTime.Load(),
  1168  			gcController.dedicatedMarkTime.Load() + gcController.fractionalMarkTime.Load(),
  1169  			gcController.idleMarkTime.Load(),
  1170  			markTermCpu,
  1171  		} {
  1172  			if i == 2 || i == 3 {
  1173  				// Separate mark time components with /.
  1174  				print("/")
  1175  			} else if i != 0 {
  1176  				print("+")
  1177  			}
  1178  			print(string(fmtNSAsMS(sbuf[:], uint64(ns))))
  1179  		}
  1180  		print(" ms cpu, ",
  1181  			work.heap0>>20, "->", work.heap1>>20, "->", work.heap2>>20, " MB, ",
  1182  			gcController.lastHeapGoal>>20, " MB goal, ",
  1183  			gcController.lastStackScan.Load()>>20, " MB stacks, ",
  1184  			gcController.globalsScan.Load()>>20, " MB globals, ",
  1185  			work.maxprocs, " P")
  1186  		if work.userForced {
  1187  			print(" (forced)")
  1188  		}
  1189  		print("\n")
  1190  		printunlock()
  1191  	}
  1192  
  1193  	// Set any arena chunks that were deferred to fault.
  1194  	lock(&userArenaState.lock)
  1195  	faultList := userArenaState.fault
  1196  	userArenaState.fault = nil
  1197  	unlock(&userArenaState.lock)
  1198  	for _, lc := range faultList {
  1199  		lc.mspan.setUserArenaChunkToFault()
  1200  	}
  1201  
  1202  	// Enable huge pages on some metadata if we cross a heap threshold.
  1203  	if gcController.heapGoal() > minHeapForMetadataHugePages {
  1204  		systemstack(func() {
  1205  			mheap_.enableMetadataHugePages()
  1206  		})
  1207  	}
  1208  
  1209  	semrelease(&worldsema)
  1210  	semrelease(&gcsema)
  1211  	// Careful: another GC cycle may start now.
  1212  
  1213  	releasem(mp)
  1214  	mp = nil
  1215  
  1216  	// now that gc is done, kick off finalizer thread if needed
  1217  	if !concurrentSweep {
  1218  		// give the queued finalizers, if any, a chance to run
  1219  		Gosched()
  1220  	}
  1221  }
  1222  
  1223  // gcBgMarkStartWorkers prepares background mark worker goroutines. These
  1224  // goroutines will not run until the mark phase, but they must be started while
  1225  // the work is not stopped and from a regular G stack. The caller must hold
  1226  // worldsema.
  1227  func gcBgMarkStartWorkers() {
  1228  	// Background marking is performed by per-P G's. Ensure that each P has
  1229  	// a background GC G.
  1230  	//
  1231  	// Worker Gs don't exit if gomaxprocs is reduced. If it is raised
  1232  	// again, we can reuse the old workers; no need to create new workers.
  1233  	for gcBgMarkWorkerCount < gomaxprocs {
  1234  		go gcBgMarkWorker()
  1235  
  1236  		notetsleepg(&work.bgMarkReady, -1)
  1237  		noteclear(&work.bgMarkReady)
  1238  		// The worker is now guaranteed to be added to the pool before
  1239  		// its P's next findRunnableGCWorker.
  1240  
  1241  		gcBgMarkWorkerCount++
  1242  	}
  1243  }
  1244  
  1245  // gcBgMarkPrepare sets up state for background marking.
  1246  // Mutator assists must not yet be enabled.
  1247  func gcBgMarkPrepare() {
  1248  	// Background marking will stop when the work queues are empty
  1249  	// and there are no more workers (note that, since this is
  1250  	// concurrent, this may be a transient state, but mark
  1251  	// termination will clean it up). Between background workers
  1252  	// and assists, we don't really know how many workers there
  1253  	// will be, so we pretend to have an arbitrarily large number
  1254  	// of workers, almost all of which are "waiting". While a
  1255  	// worker is working it decrements nwait. If nproc == nwait,
  1256  	// there are no workers.
  1257  	work.nproc = ^uint32(0)
  1258  	work.nwait = ^uint32(0)
  1259  }
  1260  
  1261  // gcBgMarkWorkerNode is an entry in the gcBgMarkWorkerPool. It points to a single
  1262  // gcBgMarkWorker goroutine.
  1263  type gcBgMarkWorkerNode struct {
  1264  	// Unused workers are managed in a lock-free stack. This field must be first.
  1265  	node lfnode
  1266  
  1267  	// The g of this worker.
  1268  	gp guintptr
  1269  
  1270  	// Release this m on park. This is used to communicate with the unlock
  1271  	// function, which cannot access the G's stack. It is unused outside of
  1272  	// gcBgMarkWorker().
  1273  	m muintptr
  1274  }
  1275  
  1276  func gcBgMarkWorker() {
  1277  	gp := getg()
  1278  
  1279  	// We pass node to a gopark unlock function, so it can't be on
  1280  	// the stack (see gopark). Prevent deadlock from recursively
  1281  	// starting GC by disabling preemption.
  1282  	gp.m.preemptoff = "GC worker init"
  1283  	node := new(gcBgMarkWorkerNode)
  1284  	gp.m.preemptoff = ""
  1285  
  1286  	node.gp.set(gp)
  1287  
  1288  	node.m.set(acquirem())
  1289  	notewakeup(&work.bgMarkReady)
  1290  	// After this point, the background mark worker is generally scheduled
  1291  	// cooperatively by gcController.findRunnableGCWorker. While performing
  1292  	// work on the P, preemption is disabled because we are working on
  1293  	// P-local work buffers. When the preempt flag is set, this puts itself
  1294  	// into _Gwaiting to be woken up by gcController.findRunnableGCWorker
  1295  	// at the appropriate time.
  1296  	//
  1297  	// When preemption is enabled (e.g., while in gcMarkDone), this worker
  1298  	// may be preempted and schedule as a _Grunnable G from a runq. That is
  1299  	// fine; it will eventually gopark again for further scheduling via
  1300  	// findRunnableGCWorker.
  1301  	//
  1302  	// Since we disable preemption before notifying bgMarkReady, we
  1303  	// guarantee that this G will be in the worker pool for the next
  1304  	// findRunnableGCWorker. This isn't strictly necessary, but it reduces
  1305  	// latency between _GCmark starting and the workers starting.
  1306  
  1307  	for {
  1308  		// Go to sleep until woken by
  1309  		// gcController.findRunnableGCWorker.
  1310  		gopark(func(g *g, nodep unsafe.Pointer) bool {
  1311  			node := (*gcBgMarkWorkerNode)(nodep)
  1312  
  1313  			if mp := node.m.ptr(); mp != nil {
  1314  				// The worker G is no longer running; release
  1315  				// the M.
  1316  				//
  1317  				// N.B. it is _safe_ to release the M as soon
  1318  				// as we are no longer performing P-local mark
  1319  				// work.
  1320  				//
  1321  				// However, since we cooperatively stop work
  1322  				// when gp.preempt is set, if we releasem in
  1323  				// the loop then the following call to gopark
  1324  				// would immediately preempt the G. This is
  1325  				// also safe, but inefficient: the G must
  1326  				// schedule again only to enter gopark and park
  1327  				// again. Thus, we defer the release until
  1328  				// after parking the G.
  1329  				releasem(mp)
  1330  			}
  1331  
  1332  			// Release this G to the pool.
  1333  			gcBgMarkWorkerPool.push(&node.node)
  1334  			// Note that at this point, the G may immediately be
  1335  			// rescheduled and may be running.
  1336  			return true
  1337  		}, unsafe.Pointer(node), waitReasonGCWorkerIdle, traceBlockSystemGoroutine, 0)
  1338  
  1339  		// Preemption must not occur here, or another G might see
  1340  		// p.gcMarkWorkerMode.
  1341  
  1342  		// Disable preemption so we can use the gcw. If the
  1343  		// scheduler wants to preempt us, we'll stop draining,
  1344  		// dispose the gcw, and then preempt.
  1345  		node.m.set(acquirem())
  1346  		pp := gp.m.p.ptr() // P can't change with preemption disabled.
  1347  
  1348  		if gcBlackenEnabled == 0 {
  1349  			println("worker mode", pp.gcMarkWorkerMode)
  1350  			throw("gcBgMarkWorker: blackening not enabled")
  1351  		}
  1352  
  1353  		if pp.gcMarkWorkerMode == gcMarkWorkerNotWorker {
  1354  			throw("gcBgMarkWorker: mode not set")
  1355  		}
  1356  
  1357  		startTime := nanotime()
  1358  		pp.gcMarkWorkerStartTime = startTime
  1359  		var trackLimiterEvent bool
  1360  		if pp.gcMarkWorkerMode == gcMarkWorkerIdleMode {
  1361  			trackLimiterEvent = pp.limiterEvent.start(limiterEventIdleMarkWork, startTime)
  1362  		}
  1363  
  1364  		decnwait := atomic.Xadd(&work.nwait, -1)
  1365  		if decnwait == work.nproc {
  1366  			println("runtime: work.nwait=", decnwait, "work.nproc=", work.nproc)
  1367  			throw("work.nwait was > work.nproc")
  1368  		}
  1369  
  1370  		systemstack(func() {
  1371  			// Mark our goroutine preemptible so its stack
  1372  			// can be scanned. This lets two mark workers
  1373  			// scan each other (otherwise, they would
  1374  			// deadlock). We must not modify anything on
  1375  			// the G stack. However, stack shrinking is
  1376  			// disabled for mark workers, so it is safe to
  1377  			// read from the G stack.
  1378  			//
  1379  			// N.B. The execution tracer is not aware of this status
  1380  			// transition and handles it specially based on the
  1381  			// wait reason.
  1382  			casGToWaiting(gp, _Grunning, waitReasonGCWorkerActive)
  1383  			switch pp.gcMarkWorkerMode {
  1384  			default:
  1385  				throw("gcBgMarkWorker: unexpected gcMarkWorkerMode")
  1386  			case gcMarkWorkerDedicatedMode:
  1387  				gcDrainMarkWorkerDedicated(&pp.gcw, true)
  1388  				if gp.preempt {
  1389  					// We were preempted. This is
  1390  					// a useful signal to kick
  1391  					// everything out of the run
  1392  					// queue so it can run
  1393  					// somewhere else.
  1394  					if drainQ, n := runqdrain(pp); n > 0 {
  1395  						lock(&sched.lock)
  1396  						globrunqputbatch(&drainQ, int32(n))
  1397  						unlock(&sched.lock)
  1398  					}
  1399  				}
  1400  				// Go back to draining, this time
  1401  				// without preemption.
  1402  				gcDrainMarkWorkerDedicated(&pp.gcw, false)
  1403  			case gcMarkWorkerFractionalMode:
  1404  				gcDrainMarkWorkerFractional(&pp.gcw)
  1405  			case gcMarkWorkerIdleMode:
  1406  				gcDrainMarkWorkerIdle(&pp.gcw)
  1407  			}
  1408  			casgstatus(gp, _Gwaiting, _Grunning)
  1409  		})
  1410  
  1411  		// Account for time and mark us as stopped.
  1412  		now := nanotime()
  1413  		duration := now - startTime
  1414  		gcController.markWorkerStop(pp.gcMarkWorkerMode, duration)
  1415  		if trackLimiterEvent {
  1416  			pp.limiterEvent.stop(limiterEventIdleMarkWork, now)
  1417  		}
  1418  		if pp.gcMarkWorkerMode == gcMarkWorkerFractionalMode {
  1419  			atomic.Xaddint64(&pp.gcFractionalMarkTime, duration)
  1420  		}
  1421  
  1422  		// Was this the last worker and did we run out
  1423  		// of work?
  1424  		incnwait := atomic.Xadd(&work.nwait, +1)
  1425  		if incnwait > work.nproc {
  1426  			println("runtime: p.gcMarkWorkerMode=", pp.gcMarkWorkerMode,
  1427  				"work.nwait=", incnwait, "work.nproc=", work.nproc)
  1428  			throw("work.nwait > work.nproc")
  1429  		}
  1430  
  1431  		// We'll releasem after this point and thus this P may run
  1432  		// something else. We must clear the worker mode to avoid
  1433  		// attributing the mode to a different (non-worker) G in
  1434  		// traceGoStart.
  1435  		pp.gcMarkWorkerMode = gcMarkWorkerNotWorker
  1436  
  1437  		// If this worker reached a background mark completion
  1438  		// point, signal the main GC goroutine.
  1439  		if incnwait == work.nproc && !gcMarkWorkAvailable(nil) {
  1440  			// We don't need the P-local buffers here, allow
  1441  			// preemption because we may schedule like a regular
  1442  			// goroutine in gcMarkDone (block on locks, etc).
  1443  			releasem(node.m.ptr())
  1444  			node.m.set(nil)
  1445  
  1446  			gcMarkDone()
  1447  		}
  1448  	}
  1449  }
  1450  
  1451  // gcMarkWorkAvailable reports whether executing a mark worker
  1452  // on p is potentially useful. p may be nil, in which case it only
  1453  // checks the global sources of work.
  1454  func gcMarkWorkAvailable(p *p) bool {
  1455  	if p != nil && !p.gcw.empty() {
  1456  		return true
  1457  	}
  1458  	if !work.full.empty() {
  1459  		return true // global work available
  1460  	}
  1461  	if work.markrootNext < work.markrootJobs {
  1462  		return true // root scan work available
  1463  	}
  1464  	return false
  1465  }
  1466  
  1467  // gcMark runs the mark (or, for concurrent GC, mark termination)
  1468  // All gcWork caches must be empty.
  1469  // STW is in effect at this point.
  1470  func gcMark(startTime int64) {
  1471  	if debug.allocfreetrace > 0 {
  1472  		tracegc()
  1473  	}
  1474  
  1475  	if gcphase != _GCmarktermination {
  1476  		throw("in gcMark expecting to see gcphase as _GCmarktermination")
  1477  	}
  1478  	work.tstart = startTime
  1479  
  1480  	// Check that there's no marking work remaining.
  1481  	if work.full != 0 || work.markrootNext < work.markrootJobs {
  1482  		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")
  1483  		panic("non-empty mark queue after concurrent mark")
  1484  	}
  1485  
  1486  	if debug.gccheckmark > 0 {
  1487  		// This is expensive when there's a large number of
  1488  		// Gs, so only do it if checkmark is also enabled.
  1489  		gcMarkRootCheck()
  1490  	}
  1491  
  1492  	// Drop allg snapshot. allgs may have grown, in which case
  1493  	// this is the only reference to the old backing store and
  1494  	// there's no need to keep it around.
  1495  	work.stackRoots = nil
  1496  
  1497  	// Clear out buffers and double-check that all gcWork caches
  1498  	// are empty. This should be ensured by gcMarkDone before we
  1499  	// enter mark termination.
  1500  	//
  1501  	// TODO: We could clear out buffers just before mark if this
  1502  	// has a non-negligible impact on STW time.
  1503  	for _, p := range allp {
  1504  		// The write barrier may have buffered pointers since
  1505  		// the gcMarkDone barrier. However, since the barrier
  1506  		// ensured all reachable objects were marked, all of
  1507  		// these must be pointers to black objects. Hence we
  1508  		// can just discard the write barrier buffer.
  1509  		if debug.gccheckmark > 0 {
  1510  			// For debugging, flush the buffer and make
  1511  			// sure it really was all marked.
  1512  			wbBufFlush1(p)
  1513  		} else {
  1514  			p.wbBuf.reset()
  1515  		}
  1516  
  1517  		gcw := &p.gcw
  1518  		if !gcw.empty() {
  1519  			printlock()
  1520  			print("runtime: P ", p.id, " flushedWork ", gcw.flushedWork)
  1521  			if gcw.wbuf1 == nil {
  1522  				print(" wbuf1=<nil>")
  1523  			} else {
  1524  				print(" wbuf1.n=", gcw.wbuf1.nobj)
  1525  			}
  1526  			if gcw.wbuf2 == nil {
  1527  				print(" wbuf2=<nil>")
  1528  			} else {
  1529  				print(" wbuf2.n=", gcw.wbuf2.nobj)
  1530  			}
  1531  			print("\n")
  1532  			throw("P has cached GC work at end of mark termination")
  1533  		}
  1534  		// There may still be cached empty buffers, which we
  1535  		// need to flush since we're going to free them. Also,
  1536  		// there may be non-zero stats because we allocated
  1537  		// black after the gcMarkDone barrier.
  1538  		gcw.dispose()
  1539  	}
  1540  
  1541  	// Flush scanAlloc from each mcache since we're about to modify
  1542  	// heapScan directly. If we were to flush this later, then scanAlloc
  1543  	// might have incorrect information.
  1544  	//
  1545  	// Note that it's not important to retain this information; we know
  1546  	// exactly what heapScan is at this point via scanWork.
  1547  	for _, p := range allp {
  1548  		c := p.mcache
  1549  		if c == nil {
  1550  			continue
  1551  		}
  1552  		c.scanAlloc = 0
  1553  	}
  1554  
  1555  	// Reset controller state.
  1556  	gcController.resetLive(work.bytesMarked)
  1557  }
  1558  
  1559  // gcSweep must be called on the system stack because it acquires the heap
  1560  // lock. See mheap for details.
  1561  //
  1562  // Returns true if the heap was fully swept by this function.
  1563  //
  1564  // The world must be stopped.
  1565  //
  1566  //go:systemstack
  1567  func gcSweep(mode gcMode) bool {
  1568  	assertWorldStopped()
  1569  
  1570  	if gcphase != _GCoff {
  1571  		throw("gcSweep being done but phase is not GCoff")
  1572  	}
  1573  
  1574  	lock(&mheap_.lock)
  1575  	mheap_.sweepgen += 2
  1576  	sweep.active.reset()
  1577  	mheap_.pagesSwept.Store(0)
  1578  	mheap_.sweepArenas = mheap_.allArenas
  1579  	mheap_.reclaimIndex.Store(0)
  1580  	mheap_.reclaimCredit.Store(0)
  1581  	unlock(&mheap_.lock)
  1582  
  1583  	sweep.centralIndex.clear()
  1584  
  1585  	if !concurrentSweep || mode == gcForceBlockMode {
  1586  		// Special case synchronous sweep.
  1587  		// Record that no proportional sweeping has to happen.
  1588  		lock(&mheap_.lock)
  1589  		mheap_.sweepPagesPerByte = 0
  1590  		unlock(&mheap_.lock)
  1591  		// Flush all mcaches.
  1592  		for _, pp := range allp {
  1593  			pp.mcache.prepareForSweep()
  1594  		}
  1595  		// Sweep all spans eagerly.
  1596  		for sweepone() != ^uintptr(0) {
  1597  		}
  1598  		// Free workbufs eagerly.
  1599  		prepareFreeWorkbufs()
  1600  		for freeSomeWbufs(false) {
  1601  		}
  1602  		// All "free" events for this mark/sweep cycle have
  1603  		// now happened, so we can make this profile cycle
  1604  		// available immediately.
  1605  		mProf_NextCycle()
  1606  		mProf_Flush()
  1607  		return true
  1608  	}
  1609  
  1610  	// Background sweep.
  1611  	lock(&sweep.lock)
  1612  	if sweep.parked {
  1613  		sweep.parked = false
  1614  		ready(sweep.g, 0, true)
  1615  	}
  1616  	unlock(&sweep.lock)
  1617  	return false
  1618  }
  1619  
  1620  // gcResetMarkState resets global state prior to marking (concurrent
  1621  // or STW) and resets the stack scan state of all Gs.
  1622  //
  1623  // This is safe to do without the world stopped because any Gs created
  1624  // during or after this will start out in the reset state.
  1625  //
  1626  // gcResetMarkState must be called on the system stack because it acquires
  1627  // the heap lock. See mheap for details.
  1628  //
  1629  //go:systemstack
  1630  func gcResetMarkState() {
  1631  	// This may be called during a concurrent phase, so lock to make sure
  1632  	// allgs doesn't change.
  1633  	forEachG(func(gp *g) {
  1634  		gp.gcscandone = false // set to true in gcphasework
  1635  		gp.gcAssistBytes = 0
  1636  	})
  1637  
  1638  	// Clear page marks. This is just 1MB per 64GB of heap, so the
  1639  	// time here is pretty trivial.
  1640  	lock(&mheap_.lock)
  1641  	arenas := mheap_.allArenas
  1642  	unlock(&mheap_.lock)
  1643  	for _, ai := range arenas {
  1644  		ha := mheap_.arenas[ai.l1()][ai.l2()]
  1645  		for i := range ha.pageMarks {
  1646  			ha.pageMarks[i] = 0
  1647  		}
  1648  	}
  1649  
  1650  	work.bytesMarked = 0
  1651  	work.initialHeapLive = gcController.heapLive.Load()
  1652  }
  1653  
  1654  // Hooks for other packages
  1655  
  1656  var poolcleanup func()
  1657  var boringCaches []unsafe.Pointer // for crypto/internal/boring
  1658  
  1659  //go:linkname sync_runtime_registerPoolCleanup sync.runtime_registerPoolCleanup
  1660  func sync_runtime_registerPoolCleanup(f func()) {
  1661  	poolcleanup = f
  1662  }
  1663  
  1664  //go:linkname boring_registerCache crypto/internal/boring/bcache.registerCache
  1665  func boring_registerCache(p unsafe.Pointer) {
  1666  	boringCaches = append(boringCaches, p)
  1667  }
  1668  
  1669  func clearpools() {
  1670  	// clear sync.Pools
  1671  	if poolcleanup != nil {
  1672  		poolcleanup()
  1673  	}
  1674  
  1675  	// clear boringcrypto caches
  1676  	for _, p := range boringCaches {
  1677  		atomicstorep(p, nil)
  1678  	}
  1679  
  1680  	// Clear central sudog cache.
  1681  	// Leave per-P caches alone, they have strictly bounded size.
  1682  	// Disconnect cached list before dropping it on the floor,
  1683  	// so that a dangling ref to one entry does not pin all of them.
  1684  	lock(&sched.sudoglock)
  1685  	var sg, sgnext *sudog
  1686  	for sg = sched.sudogcache; sg != nil; sg = sgnext {
  1687  		sgnext = sg.next
  1688  		sg.next = nil
  1689  	}
  1690  	sched.sudogcache = nil
  1691  	unlock(&sched.sudoglock)
  1692  
  1693  	// Clear central defer pool.
  1694  	// Leave per-P pools alone, they have strictly bounded size.
  1695  	lock(&sched.deferlock)
  1696  	// disconnect cached list before dropping it on the floor,
  1697  	// so that a dangling ref to one entry does not pin all of them.
  1698  	var d, dlink *_defer
  1699  	for d = sched.deferpool; d != nil; d = dlink {
  1700  		dlink = d.link
  1701  		d.link = nil
  1702  	}
  1703  	sched.deferpool = nil
  1704  	unlock(&sched.deferlock)
  1705  }
  1706  
  1707  // Timing
  1708  
  1709  // itoaDiv formats val/(10**dec) into buf.
  1710  func itoaDiv(buf []byte, val uint64, dec int) []byte {
  1711  	i := len(buf) - 1
  1712  	idec := i - dec
  1713  	for val >= 10 || i >= idec {
  1714  		buf[i] = byte(val%10 + '0')
  1715  		i--
  1716  		if i == idec {
  1717  			buf[i] = '.'
  1718  			i--
  1719  		}
  1720  		val /= 10
  1721  	}
  1722  	buf[i] = byte(val + '0')
  1723  	return buf[i:]
  1724  }
  1725  
  1726  // fmtNSAsMS nicely formats ns nanoseconds as milliseconds.
  1727  func fmtNSAsMS(buf []byte, ns uint64) []byte {
  1728  	if ns >= 10e6 {
  1729  		// Format as whole milliseconds.
  1730  		return itoaDiv(buf, ns/1e6, 0)
  1731  	}
  1732  	// Format two digits of precision, with at most three decimal places.
  1733  	x := ns / 1e3
  1734  	if x == 0 {
  1735  		buf[0] = '0'
  1736  		return buf[:1]
  1737  	}
  1738  	dec := 3
  1739  	for x >= 100 {
  1740  		x /= 10
  1741  		dec--
  1742  	}
  1743  	return itoaDiv(buf, x, dec)
  1744  }
  1745  
  1746  // Helpers for testing GC.
  1747  
  1748  // gcTestMoveStackOnNextCall causes the stack to be moved on a call
  1749  // immediately following the call to this. It may not work correctly
  1750  // if any other work appears after this call (such as returning).
  1751  // Typically the following call should be marked go:noinline so it
  1752  // performs a stack check.
  1753  //
  1754  // In rare cases this may not cause the stack to move, specifically if
  1755  // there's a preemption between this call and the next.
  1756  func gcTestMoveStackOnNextCall() {
  1757  	gp := getg()
  1758  	gp.stackguard0 = stackForceMove
  1759  }
  1760  
  1761  // gcTestIsReachable performs a GC and returns a bit set where bit i
  1762  // is set if ptrs[i] is reachable.
  1763  func gcTestIsReachable(ptrs ...unsafe.Pointer) (mask uint64) {
  1764  	// This takes the pointers as unsafe.Pointers in order to keep
  1765  	// them live long enough for us to attach specials. After
  1766  	// that, we drop our references to them.
  1767  
  1768  	if len(ptrs) > 64 {
  1769  		panic("too many pointers for uint64 mask")
  1770  	}
  1771  
  1772  	// Block GC while we attach specials and drop our references
  1773  	// to ptrs. Otherwise, if a GC is in progress, it could mark
  1774  	// them reachable via this function before we have a chance to
  1775  	// drop them.
  1776  	semacquire(&gcsema)
  1777  
  1778  	// Create reachability specials for ptrs.
  1779  	specials := make([]*specialReachable, len(ptrs))
  1780  	for i, p := range ptrs {
  1781  		lock(&mheap_.speciallock)
  1782  		s := (*specialReachable)(mheap_.specialReachableAlloc.alloc())
  1783  		unlock(&mheap_.speciallock)
  1784  		s.special.kind = _KindSpecialReachable
  1785  		if !addspecial(p, &s.special) {
  1786  			throw("already have a reachable special (duplicate pointer?)")
  1787  		}
  1788  		specials[i] = s
  1789  		// Make sure we don't retain ptrs.
  1790  		ptrs[i] = nil
  1791  	}
  1792  
  1793  	semrelease(&gcsema)
  1794  
  1795  	// Force a full GC and sweep.
  1796  	GC()
  1797  
  1798  	// Process specials.
  1799  	for i, s := range specials {
  1800  		if !s.done {
  1801  			printlock()
  1802  			println("runtime: object", i, "was not swept")
  1803  			throw("IsReachable failed")
  1804  		}
  1805  		if s.reachable {
  1806  			mask |= 1 << i
  1807  		}
  1808  		lock(&mheap_.speciallock)
  1809  		mheap_.specialReachableAlloc.free(unsafe.Pointer(s))
  1810  		unlock(&mheap_.speciallock)
  1811  	}
  1812  
  1813  	return mask
  1814  }
  1815  
  1816  // gcTestPointerClass returns the category of what p points to, one of:
  1817  // "heap", "stack", "data", "bss", "other". This is useful for checking
  1818  // that a test is doing what it's intended to do.
  1819  //
  1820  // This is nosplit simply to avoid extra pointer shuffling that may
  1821  // complicate a test.
  1822  //
  1823  //go:nosplit
  1824  func gcTestPointerClass(p unsafe.Pointer) string {
  1825  	p2 := uintptr(noescape(p))
  1826  	gp := getg()
  1827  	if gp.stack.lo <= p2 && p2 < gp.stack.hi {
  1828  		return "stack"
  1829  	}
  1830  	if base, _, _ := findObject(p2, 0, 0); base != 0 {
  1831  		return "heap"
  1832  	}
  1833  	for _, datap := range activeModules() {
  1834  		if datap.data <= p2 && p2 < datap.edata || datap.noptrdata <= p2 && p2 < datap.enoptrdata {
  1835  			return "data"
  1836  		}
  1837  		if datap.bss <= p2 && p2 < datap.ebss || datap.noptrbss <= p2 && p2 <= datap.enoptrbss {
  1838  			return "bss"
  1839  		}
  1840  	}
  1841  	KeepAlive(p)
  1842  	return "other"
  1843  }
  1844  

View as plain text