Source file src/runtime/mgcsweep.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: sweeping
     6  
     7  // The sweeper consists of two different algorithms:
     8  //
     9  // * The object reclaimer finds and frees unmarked slots in spans. It
    10  //   can free a whole span if none of the objects are marked, but that
    11  //   isn't its goal. This can be driven either synchronously by
    12  //   mcentral.cacheSpan for mcentral spans, or asynchronously by
    13  //   sweepone, which looks at all the mcentral lists.
    14  //
    15  // * The span reclaimer looks for spans that contain no marked objects
    16  //   and frees whole spans. This is a separate algorithm because
    17  //   freeing whole spans is the hardest task for the object reclaimer,
    18  //   but is critical when allocating new spans. The entry point for
    19  //   this is mheap_.reclaim and it's driven by a sequential scan of
    20  //   the page marks bitmap in the heap arenas.
    21  //
    22  // Both algorithms ultimately call mspan.sweep, which sweeps a single
    23  // heap span.
    24  
    25  package runtime
    26  
    27  import (
    28  	"internal/runtime/atomic"
    29  	"unsafe"
    30  )
    31  
    32  var sweep sweepdata
    33  
    34  // State of background sweep.
    35  type sweepdata struct {
    36  	lock   mutex
    37  	g      *g
    38  	parked bool
    39  
    40  	// active tracks outstanding sweepers and the sweep
    41  	// termination condition.
    42  	active activeSweep
    43  
    44  	// centralIndex is the current unswept span class.
    45  	// It represents an index into the mcentral span
    46  	// sets. Accessed and updated via its load and
    47  	// update methods. Not protected by a lock.
    48  	//
    49  	// Reset at mark termination.
    50  	// Used by mheap.nextSpanForSweep.
    51  	centralIndex sweepClass
    52  }
    53  
    54  // sweepClass is a spanClass and one bit to represent whether we're currently
    55  // sweeping partial or full spans.
    56  type sweepClass uint32
    57  
    58  const (
    59  	numSweepClasses            = numSpanClasses * 2
    60  	sweepClassDone  sweepClass = sweepClass(^uint32(0))
    61  )
    62  
    63  func (s *sweepClass) load() sweepClass {
    64  	return sweepClass(atomic.Load((*uint32)(s)))
    65  }
    66  
    67  func (s *sweepClass) update(sNew sweepClass) {
    68  	// Only update *s if its current value is less than sNew,
    69  	// since *s increases monotonically.
    70  	sOld := s.load()
    71  	for sOld < sNew && !atomic.Cas((*uint32)(s), uint32(sOld), uint32(sNew)) {
    72  		sOld = s.load()
    73  	}
    74  	// TODO(mknyszek): This isn't the only place we have
    75  	// an atomic monotonically increasing counter. It would
    76  	// be nice to have an "atomic max" which is just implemented
    77  	// as the above on most architectures. Some architectures
    78  	// like RISC-V however have native support for an atomic max.
    79  }
    80  
    81  func (s *sweepClass) clear() {
    82  	atomic.Store((*uint32)(s), 0)
    83  }
    84  
    85  // split returns the underlying span class as well as
    86  // whether we're interested in the full or partial
    87  // unswept lists for that class, indicated as a boolean
    88  // (true means "full").
    89  func (s sweepClass) split() (spc spanClass, full bool) {
    90  	return spanClass(s >> 1), s&1 == 0
    91  }
    92  
    93  // nextSpanForSweep finds and pops the next span for sweeping from the
    94  // central sweep buffers. It returns ownership of the span to the caller.
    95  // Returns nil if no such span exists.
    96  func (h *mheap) nextSpanForSweep() *mspan {
    97  	sg := h.sweepgen
    98  	for sc := sweep.centralIndex.load(); sc < numSweepClasses; sc++ {
    99  		spc, full := sc.split()
   100  		c := &h.central[spc].mcentral
   101  		var s *mspan
   102  		if full {
   103  			s = c.fullUnswept(sg).pop()
   104  		} else {
   105  			s = c.partialUnswept(sg).pop()
   106  		}
   107  		if s != nil {
   108  			// Write down that we found something so future sweepers
   109  			// can start from here.
   110  			sweep.centralIndex.update(sc)
   111  			return s
   112  		}
   113  	}
   114  	// Write down that we found nothing.
   115  	sweep.centralIndex.update(sweepClassDone)
   116  	return nil
   117  }
   118  
   119  const sweepDrainedMask = 1 << 31
   120  
   121  // activeSweep is a type that captures whether sweeping
   122  // is done, and whether there are any outstanding sweepers.
   123  //
   124  // Every potential sweeper must call begin() before they look
   125  // for work, and end() after they've finished sweeping.
   126  type activeSweep struct {
   127  	// state is divided into two parts.
   128  	//
   129  	// The top bit (masked by sweepDrainedMask) is a boolean
   130  	// value indicating whether all the sweep work has been
   131  	// drained from the queue.
   132  	//
   133  	// The rest of the bits are a counter, indicating the
   134  	// number of outstanding concurrent sweepers.
   135  	state atomic.Uint32
   136  }
   137  
   138  // begin registers a new sweeper. Returns a sweepLocker
   139  // for acquiring spans for sweeping. Any outstanding sweeper blocks
   140  // sweep termination.
   141  //
   142  // If the sweepLocker is invalid, the caller can be sure that all
   143  // outstanding sweep work has been drained, so there is nothing left
   144  // to sweep. Note that there may be sweepers currently running, so
   145  // this does not indicate that all sweeping has completed.
   146  //
   147  // Even if the sweepLocker is invalid, its sweepGen is always valid.
   148  func (a *activeSweep) begin() sweepLocker {
   149  	for {
   150  		state := a.state.Load()
   151  		if state&sweepDrainedMask != 0 {
   152  			return sweepLocker{mheap_.sweepgen, false}
   153  		}
   154  		if a.state.CompareAndSwap(state, state+1) {
   155  			return sweepLocker{mheap_.sweepgen, true}
   156  		}
   157  	}
   158  }
   159  
   160  // end deregisters a sweeper. Must be called once for each time
   161  // begin is called if the sweepLocker is valid.
   162  func (a *activeSweep) end(sl sweepLocker) {
   163  	if sl.sweepGen != mheap_.sweepgen {
   164  		throw("sweeper left outstanding across sweep generations")
   165  	}
   166  	for {
   167  		state := a.state.Load()
   168  		if (state&^sweepDrainedMask)-1 >= sweepDrainedMask {
   169  			throw("mismatched begin/end of activeSweep")
   170  		}
   171  		if a.state.CompareAndSwap(state, state-1) {
   172  			if state != sweepDrainedMask {
   173  				return
   174  			}
   175  			if debug.gcpacertrace > 0 {
   176  				live := gcController.heapLive.Load()
   177  				print("pacer: sweep done at heap size ", live>>20, "MB; allocated ", (live-mheap_.sweepHeapLiveBasis)>>20, "MB during sweep; swept ", mheap_.pagesSwept.Load(), " pages at ", mheap_.sweepPagesPerByte, " pages/byte\n")
   178  			}
   179  			return
   180  		}
   181  	}
   182  }
   183  
   184  // markDrained marks the active sweep cycle as having drained
   185  // all remaining work. This is safe to be called concurrently
   186  // with all other methods of activeSweep, though may race.
   187  //
   188  // Returns true if this call was the one that actually performed
   189  // the mark.
   190  func (a *activeSweep) markDrained() bool {
   191  	for {
   192  		state := a.state.Load()
   193  		if state&sweepDrainedMask != 0 {
   194  			return false
   195  		}
   196  		if a.state.CompareAndSwap(state, state|sweepDrainedMask) {
   197  			return true
   198  		}
   199  	}
   200  }
   201  
   202  // sweepers returns the current number of active sweepers.
   203  func (a *activeSweep) sweepers() uint32 {
   204  	return a.state.Load() &^ sweepDrainedMask
   205  }
   206  
   207  // isDone returns true if all sweep work has been drained and no more
   208  // outstanding sweepers exist. That is, when the sweep phase is
   209  // completely done.
   210  func (a *activeSweep) isDone() bool {
   211  	return a.state.Load() == sweepDrainedMask
   212  }
   213  
   214  // reset sets up the activeSweep for the next sweep cycle.
   215  //
   216  // The world must be stopped.
   217  func (a *activeSweep) reset() {
   218  	assertWorldStopped()
   219  	a.state.Store(0)
   220  }
   221  
   222  // finishsweep_m ensures that all spans are swept.
   223  //
   224  // The world must be stopped. This ensures there are no sweeps in
   225  // progress.
   226  //
   227  //go:nowritebarrier
   228  func finishsweep_m() {
   229  	assertWorldStopped()
   230  
   231  	// Sweeping must be complete before marking commences, so
   232  	// sweep any unswept spans. If this is a concurrent GC, there
   233  	// shouldn't be any spans left to sweep, so this should finish
   234  	// instantly. If GC was forced before the concurrent sweep
   235  	// finished, there may be spans to sweep.
   236  	for sweepone() != ^uintptr(0) {
   237  	}
   238  
   239  	// Make sure there aren't any outstanding sweepers left.
   240  	// At this point, with the world stopped, it means one of two
   241  	// things. Either we were able to preempt a sweeper, or that
   242  	// a sweeper didn't call sweep.active.end when it should have.
   243  	// Both cases indicate a bug, so throw.
   244  	if sweep.active.sweepers() != 0 {
   245  		throw("active sweepers found at start of mark phase")
   246  	}
   247  
   248  	// Reset all the unswept buffers, which should be empty.
   249  	// Do this in sweep termination as opposed to mark termination
   250  	// so that we can catch unswept spans and reclaim blocks as
   251  	// soon as possible.
   252  	sg := mheap_.sweepgen
   253  	for i := range mheap_.central {
   254  		c := &mheap_.central[i].mcentral
   255  		c.partialUnswept(sg).reset()
   256  		c.fullUnswept(sg).reset()
   257  	}
   258  
   259  	// Sweeping is done, so there won't be any new memory to
   260  	// scavenge for a bit.
   261  	//
   262  	// If the scavenger isn't already awake, wake it up. There's
   263  	// definitely work for it to do at this point.
   264  	scavenger.wake()
   265  
   266  	nextMarkBitArenaEpoch()
   267  }
   268  
   269  func bgsweep(c chan int) {
   270  	sweep.g = getg()
   271  
   272  	lockInit(&sweep.lock, lockRankSweep)
   273  	lock(&sweep.lock)
   274  	sweep.parked = true
   275  	c <- 1
   276  	goparkunlock(&sweep.lock, waitReasonGCSweepWait, traceBlockGCSweep, 1)
   277  
   278  	for {
   279  		// bgsweep attempts to be a "low priority" goroutine by intentionally
   280  		// yielding time. It's OK if it doesn't run, because goroutines allocating
   281  		// memory will sweep and ensure that all spans are swept before the next
   282  		// GC cycle. We really only want to run when we're idle.
   283  		//
   284  		// However, calling Gosched after each span swept produces a tremendous
   285  		// amount of tracing events, sometimes up to 50% of events in a trace. It's
   286  		// also inefficient to call into the scheduler so much because sweeping a
   287  		// single span is in general a very fast operation, taking as little as 30 ns
   288  		// on modern hardware. (See #54767.)
   289  		//
   290  		// As a result, bgsweep sweeps in batches, and only calls into the scheduler
   291  		// at the end of every batch. Furthermore, it only yields its time if there
   292  		// isn't spare idle time available on other cores. If there's available idle
   293  		// time, helping to sweep can reduce allocation latencies by getting ahead of
   294  		// the proportional sweeper and having spans ready to go for allocation.
   295  		const sweepBatchSize = 10
   296  		nSwept := 0
   297  		for sweepone() != ^uintptr(0) {
   298  			nSwept++
   299  			if nSwept%sweepBatchSize == 0 {
   300  				goschedIfBusy()
   301  			}
   302  		}
   303  		for freeSomeWbufs(true) {
   304  			// N.B. freeSomeWbufs is already batched internally.
   305  			goschedIfBusy()
   306  		}
   307  		lock(&sweep.lock)
   308  		if !isSweepDone() {
   309  			// This can happen if a GC runs between
   310  			// gosweepone returning ^0 above
   311  			// and the lock being acquired.
   312  			unlock(&sweep.lock)
   313  			continue
   314  		}
   315  		sweep.parked = true
   316  		goparkunlock(&sweep.lock, waitReasonGCSweepWait, traceBlockGCSweep, 1)
   317  	}
   318  }
   319  
   320  // sweepLocker acquires sweep ownership of spans.
   321  type sweepLocker struct {
   322  	// sweepGen is the sweep generation of the heap.
   323  	sweepGen uint32
   324  	valid    bool
   325  }
   326  
   327  // sweepLocked represents sweep ownership of a span.
   328  type sweepLocked struct {
   329  	*mspan
   330  }
   331  
   332  // tryAcquire attempts to acquire sweep ownership of span s. If it
   333  // successfully acquires ownership, it blocks sweep completion.
   334  func (l *sweepLocker) tryAcquire(s *mspan) (sweepLocked, bool) {
   335  	if !l.valid {
   336  		throw("use of invalid sweepLocker")
   337  	}
   338  	// Check before attempting to CAS.
   339  	if atomic.Load(&s.sweepgen) != l.sweepGen-2 {
   340  		return sweepLocked{}, false
   341  	}
   342  	// Attempt to acquire sweep ownership of s.
   343  	if !atomic.Cas(&s.sweepgen, l.sweepGen-2, l.sweepGen-1) {
   344  		return sweepLocked{}, false
   345  	}
   346  	return sweepLocked{s}, true
   347  }
   348  
   349  // sweepone sweeps some unswept heap span and returns the number of pages returned
   350  // to the heap, or ^uintptr(0) if there was nothing to sweep.
   351  func sweepone() uintptr {
   352  	gp := getg()
   353  
   354  	// Increment locks to ensure that the goroutine is not preempted
   355  	// in the middle of sweep thus leaving the span in an inconsistent state for next GC
   356  	gp.m.locks++
   357  
   358  	// TODO(austin): sweepone is almost always called in a loop;
   359  	// lift the sweepLocker into its callers.
   360  	sl := sweep.active.begin()
   361  	if !sl.valid {
   362  		gp.m.locks--
   363  		return ^uintptr(0)
   364  	}
   365  
   366  	// Find a span to sweep.
   367  	npages := ^uintptr(0)
   368  	var noMoreWork bool
   369  	for {
   370  		s := mheap_.nextSpanForSweep()
   371  		if s == nil {
   372  			noMoreWork = sweep.active.markDrained()
   373  			break
   374  		}
   375  		if state := s.state.get(); state != mSpanInUse {
   376  			// This can happen if direct sweeping already
   377  			// swept this span, but in that case the sweep
   378  			// generation should always be up-to-date.
   379  			if !(s.sweepgen == sl.sweepGen || s.sweepgen == sl.sweepGen+3) {
   380  				print("runtime: bad span s.state=", state, " s.sweepgen=", s.sweepgen, " sweepgen=", sl.sweepGen, "\n")
   381  				throw("non in-use span in unswept list")
   382  			}
   383  			continue
   384  		}
   385  		if s, ok := sl.tryAcquire(s); ok {
   386  			// Sweep the span we found.
   387  			npages = s.npages
   388  			if s.sweep(false) {
   389  				// Whole span was freed. Count it toward the
   390  				// page reclaimer credit since these pages can
   391  				// now be used for span allocation.
   392  				mheap_.reclaimCredit.Add(npages)
   393  			} else {
   394  				// Span is still in-use, so this returned no
   395  				// pages to the heap and the span needs to
   396  				// move to the swept in-use list.
   397  				npages = 0
   398  			}
   399  			break
   400  		}
   401  	}
   402  	sweep.active.end(sl)
   403  
   404  	if noMoreWork {
   405  		// The sweep list is empty. There may still be
   406  		// concurrent sweeps running, but we're at least very
   407  		// close to done sweeping.
   408  
   409  		// Move the scavenge gen forward (signaling
   410  		// that there's new work to do) and wake the scavenger.
   411  		//
   412  		// The scavenger is signaled by the last sweeper because once
   413  		// sweeping is done, we will definitely have useful work for
   414  		// the scavenger to do, since the scavenger only runs over the
   415  		// heap once per GC cycle. This update is not done during sweep
   416  		// termination because in some cases there may be a long delay
   417  		// between sweep done and sweep termination (e.g. not enough
   418  		// allocations to trigger a GC) which would be nice to fill in
   419  		// with scavenging work.
   420  		if debug.scavtrace > 0 {
   421  			systemstack(func() {
   422  				lock(&mheap_.lock)
   423  
   424  				// Get released stats.
   425  				releasedBg := mheap_.pages.scav.releasedBg.Load()
   426  				releasedEager := mheap_.pages.scav.releasedEager.Load()
   427  
   428  				// Print the line.
   429  				printScavTrace(releasedBg, releasedEager, false)
   430  
   431  				// Update the stats.
   432  				mheap_.pages.scav.releasedBg.Add(-releasedBg)
   433  				mheap_.pages.scav.releasedEager.Add(-releasedEager)
   434  				unlock(&mheap_.lock)
   435  			})
   436  		}
   437  		scavenger.ready()
   438  	}
   439  
   440  	gp.m.locks--
   441  	return npages
   442  }
   443  
   444  // isSweepDone reports whether all spans are swept.
   445  //
   446  // Note that this condition may transition from false to true at any
   447  // time as the sweeper runs. It may transition from true to false if a
   448  // GC runs; to prevent that the caller must be non-preemptible or must
   449  // somehow block GC progress.
   450  func isSweepDone() bool {
   451  	return sweep.active.isDone()
   452  }
   453  
   454  // Returns only when span s has been swept.
   455  //
   456  //go:nowritebarrier
   457  func (s *mspan) ensureSwept() {
   458  	// Caller must disable preemption.
   459  	// Otherwise when this function returns the span can become unswept again
   460  	// (if GC is triggered on another goroutine).
   461  	gp := getg()
   462  	if gp.m.locks == 0 && gp.m.mallocing == 0 && gp != gp.m.g0 {
   463  		throw("mspan.ensureSwept: m is not locked")
   464  	}
   465  
   466  	// If this operation fails, then that means that there are
   467  	// no more spans to be swept. In this case, either s has already
   468  	// been swept, or is about to be acquired for sweeping and swept.
   469  	sl := sweep.active.begin()
   470  	if sl.valid {
   471  		// The caller must be sure that the span is a mSpanInUse span.
   472  		if s, ok := sl.tryAcquire(s); ok {
   473  			s.sweep(false)
   474  			sweep.active.end(sl)
   475  			return
   476  		}
   477  		sweep.active.end(sl)
   478  	}
   479  
   480  	// Unfortunately we can't sweep the span ourselves. Somebody else
   481  	// got to it first. We don't have efficient means to wait, but that's
   482  	// OK, it will be swept fairly soon.
   483  	for {
   484  		spangen := atomic.Load(&s.sweepgen)
   485  		if spangen == sl.sweepGen || spangen == sl.sweepGen+3 {
   486  			break
   487  		}
   488  		osyield()
   489  	}
   490  }
   491  
   492  // sweep frees or collects finalizers for blocks not marked in the mark phase.
   493  // It clears the mark bits in preparation for the next GC round.
   494  // Returns true if the span was returned to heap.
   495  // If preserve=true, don't return it to heap nor relink in mcentral lists;
   496  // caller takes care of it.
   497  func (sl *sweepLocked) sweep(preserve bool) bool {
   498  	// It's critical that we enter this function with preemption disabled,
   499  	// GC must not start while we are in the middle of this function.
   500  	gp := getg()
   501  	if gp.m.locks == 0 && gp.m.mallocing == 0 && gp != gp.m.g0 {
   502  		throw("mspan.sweep: m is not locked")
   503  	}
   504  
   505  	s := sl.mspan
   506  	if !preserve {
   507  		// We'll release ownership of this span. Nil it out to
   508  		// prevent the caller from accidentally using it.
   509  		sl.mspan = nil
   510  	}
   511  
   512  	sweepgen := mheap_.sweepgen
   513  	if state := s.state.get(); state != mSpanInUse || s.sweepgen != sweepgen-1 {
   514  		print("mspan.sweep: state=", state, " sweepgen=", s.sweepgen, " mheap.sweepgen=", sweepgen, "\n")
   515  		throw("mspan.sweep: bad span state")
   516  	}
   517  
   518  	trace := traceAcquire()
   519  	if trace.ok() {
   520  		trace.GCSweepSpan(s.npages * _PageSize)
   521  		traceRelease(trace)
   522  	}
   523  
   524  	mheap_.pagesSwept.Add(int64(s.npages))
   525  
   526  	spc := s.spanclass
   527  	size := s.elemsize
   528  
   529  	// The allocBits indicate which unmarked objects don't need to be
   530  	// processed since they were free at the end of the last GC cycle
   531  	// and were not allocated since then.
   532  	// If the allocBits index is >= s.freeindex and the bit
   533  	// is not marked then the object remains unallocated
   534  	// since the last GC.
   535  	// This situation is analogous to being on a freelist.
   536  
   537  	// Unlink & free special records for any objects we're about to free.
   538  	// Two complications here:
   539  	// 1. An object can have both finalizer and profile special records.
   540  	//    In such case we need to queue finalizer for execution,
   541  	//    mark the object as live and preserve the profile special.
   542  	// 2. A tiny object can have several finalizers setup for different offsets.
   543  	//    If such object is not marked, we need to queue all finalizers at once.
   544  	// Both 1 and 2 are possible at the same time.
   545  	hadSpecials := s.specials != nil
   546  	siter := newSpecialsIter(s)
   547  	for siter.valid() {
   548  		// A finalizer can be set for an inner byte of an object, find object beginning.
   549  		objIndex := uintptr(siter.s.offset) / size
   550  		p := s.base() + objIndex*size
   551  		mbits := s.markBitsForIndex(objIndex)
   552  		if !mbits.isMarked() {
   553  			// This object is not marked and has at least one special record.
   554  			// Pass 1: see if it has a finalizer.
   555  			hasFinAndRevived := false
   556  			endOffset := p - s.base() + size
   557  			for tmp := siter.s; tmp != nil && uintptr(tmp.offset) < endOffset; tmp = tmp.next {
   558  				if tmp.kind == _KindSpecialFinalizer {
   559  					// Stop freeing of object if it has a finalizer.
   560  					mbits.setMarkedNonAtomic()
   561  					hasFinAndRevived = true
   562  					break
   563  				}
   564  			}
   565  			if hasFinAndRevived {
   566  				// Pass 2: queue all finalizers and clear any weak handles. Weak handles are cleared
   567  				// before finalization as specified by the weak package. See the documentation
   568  				// for that package for more details.
   569  				for siter.valid() && uintptr(siter.s.offset) < endOffset {
   570  					// Find the exact byte for which the special was setup
   571  					// (as opposed to object beginning).
   572  					special := siter.s
   573  					p := s.base() + uintptr(special.offset)
   574  					if special.kind == _KindSpecialFinalizer || special.kind == _KindSpecialWeakHandle {
   575  						siter.unlinkAndNext()
   576  						freeSpecial(special, unsafe.Pointer(p), size)
   577  					} else {
   578  						// All other specials only apply when an object is freed,
   579  						// so just keep the special record.
   580  						siter.next()
   581  					}
   582  				}
   583  			} else {
   584  				// Pass 2: the object is truly dead, free (and handle) all specials.
   585  				for siter.valid() && uintptr(siter.s.offset) < endOffset {
   586  					// Find the exact byte for which the special was setup
   587  					// (as opposed to object beginning).
   588  					special := siter.s
   589  					p := s.base() + uintptr(special.offset)
   590  					siter.unlinkAndNext()
   591  					freeSpecial(special, unsafe.Pointer(p), size)
   592  				}
   593  			}
   594  		} else {
   595  			// object is still live
   596  			if siter.s.kind == _KindSpecialReachable {
   597  				special := siter.unlinkAndNext()
   598  				(*specialReachable)(unsafe.Pointer(special)).reachable = true
   599  				freeSpecial(special, unsafe.Pointer(p), size)
   600  			} else {
   601  				// keep special record
   602  				siter.next()
   603  			}
   604  		}
   605  	}
   606  	if hadSpecials && s.specials == nil {
   607  		spanHasNoSpecials(s)
   608  	}
   609  
   610  	if traceAllocFreeEnabled() || debug.clobberfree != 0 || raceenabled || msanenabled || asanenabled {
   611  		// Find all newly freed objects.
   612  		mbits := s.markBitsForBase()
   613  		abits := s.allocBitsForIndex(0)
   614  		for i := uintptr(0); i < uintptr(s.nelems); i++ {
   615  			if !mbits.isMarked() && (abits.index < uintptr(s.freeindex) || abits.isMarked()) {
   616  				x := s.base() + i*s.elemsize
   617  				if traceAllocFreeEnabled() {
   618  					trace := traceAcquire()
   619  					if trace.ok() {
   620  						trace.HeapObjectFree(x)
   621  						traceRelease(trace)
   622  					}
   623  				}
   624  				if debug.clobberfree != 0 {
   625  					clobberfree(unsafe.Pointer(x), size)
   626  				}
   627  				// User arenas are handled on explicit free.
   628  				if raceenabled && !s.isUserArenaChunk {
   629  					racefree(unsafe.Pointer(x), size)
   630  				}
   631  				if msanenabled && !s.isUserArenaChunk {
   632  					msanfree(unsafe.Pointer(x), size)
   633  				}
   634  				if asanenabled && !s.isUserArenaChunk {
   635  					asanpoison(unsafe.Pointer(x), size)
   636  				}
   637  			}
   638  			mbits.advance()
   639  			abits.advance()
   640  		}
   641  	}
   642  
   643  	// Check for zombie objects.
   644  	if s.freeindex < s.nelems {
   645  		// Everything < freeindex is allocated and hence
   646  		// cannot be zombies.
   647  		//
   648  		// Check the first bitmap byte, where we have to be
   649  		// careful with freeindex.
   650  		obj := uintptr(s.freeindex)
   651  		if (*s.gcmarkBits.bytep(obj / 8)&^*s.allocBits.bytep(obj / 8))>>(obj%8) != 0 {
   652  			s.reportZombies()
   653  		}
   654  		// Check remaining bytes.
   655  		for i := obj/8 + 1; i < divRoundUp(uintptr(s.nelems), 8); i++ {
   656  			if *s.gcmarkBits.bytep(i)&^*s.allocBits.bytep(i) != 0 {
   657  				s.reportZombies()
   658  			}
   659  		}
   660  	}
   661  
   662  	// Count the number of free objects in this span.
   663  	nalloc := uint16(s.countAlloc())
   664  	nfreed := s.allocCount - nalloc
   665  	if nalloc > s.allocCount {
   666  		// The zombie check above should have caught this in
   667  		// more detail.
   668  		print("runtime: nelems=", s.nelems, " nalloc=", nalloc, " previous allocCount=", s.allocCount, " nfreed=", nfreed, "\n")
   669  		throw("sweep increased allocation count")
   670  	}
   671  
   672  	s.allocCount = nalloc
   673  	s.freeindex = 0 // reset allocation index to start of span.
   674  	s.freeIndexForScan = 0
   675  	if traceEnabled() {
   676  		getg().m.p.ptr().trace.reclaimed += uintptr(nfreed) * s.elemsize
   677  	}
   678  
   679  	// gcmarkBits becomes the allocBits.
   680  	// get a fresh cleared gcmarkBits in preparation for next GC
   681  	s.allocBits = s.gcmarkBits
   682  	s.gcmarkBits = newMarkBits(uintptr(s.nelems))
   683  
   684  	// refresh pinnerBits if they exists
   685  	if s.pinnerBits != nil {
   686  		s.refreshPinnerBits()
   687  	}
   688  
   689  	// Initialize alloc bits cache.
   690  	s.refillAllocCache(0)
   691  
   692  	// The span must be in our exclusive ownership until we update sweepgen,
   693  	// check for potential races.
   694  	if state := s.state.get(); state != mSpanInUse || s.sweepgen != sweepgen-1 {
   695  		print("mspan.sweep: state=", state, " sweepgen=", s.sweepgen, " mheap.sweepgen=", sweepgen, "\n")
   696  		throw("mspan.sweep: bad span state after sweep")
   697  	}
   698  	if s.sweepgen == sweepgen+1 || s.sweepgen == sweepgen+3 {
   699  		throw("swept cached span")
   700  	}
   701  
   702  	// We need to set s.sweepgen = h.sweepgen only when all blocks are swept,
   703  	// because of the potential for a concurrent free/SetFinalizer.
   704  	//
   705  	// But we need to set it before we make the span available for allocation
   706  	// (return it to heap or mcentral), because allocation code assumes that a
   707  	// span is already swept if available for allocation.
   708  	//
   709  	// Serialization point.
   710  	// At this point the mark bits are cleared and allocation ready
   711  	// to go so release the span.
   712  	atomic.Store(&s.sweepgen, sweepgen)
   713  
   714  	if s.isUserArenaChunk {
   715  		if preserve {
   716  			// This is a case that should never be handled by a sweeper that
   717  			// preserves the span for reuse.
   718  			throw("sweep: tried to preserve a user arena span")
   719  		}
   720  		if nalloc > 0 {
   721  			// There still exist pointers into the span or the span hasn't been
   722  			// freed yet. It's not ready to be reused. Put it back on the
   723  			// full swept list for the next cycle.
   724  			mheap_.central[spc].mcentral.fullSwept(sweepgen).push(s)
   725  			return false
   726  		}
   727  
   728  		// It's only at this point that the sweeper doesn't actually need to look
   729  		// at this arena anymore, so subtract from pagesInUse now.
   730  		mheap_.pagesInUse.Add(-s.npages)
   731  		s.state.set(mSpanDead)
   732  
   733  		// The arena is ready to be recycled. Remove it from the quarantine list
   734  		// and place it on the ready list. Don't add it back to any sweep lists.
   735  		systemstack(func() {
   736  			// It's the arena code's responsibility to get the chunk on the quarantine
   737  			// list by the time all references to the chunk are gone.
   738  			if s.list != &mheap_.userArena.quarantineList {
   739  				throw("user arena span is on the wrong list")
   740  			}
   741  			lock(&mheap_.lock)
   742  			mheap_.userArena.quarantineList.remove(s)
   743  			mheap_.userArena.readyList.insert(s)
   744  			unlock(&mheap_.lock)
   745  		})
   746  		return false
   747  	}
   748  
   749  	if spc.sizeclass() != 0 {
   750  		// Handle spans for small objects.
   751  		if nfreed > 0 {
   752  			// Only mark the span as needing zeroing if we've freed any
   753  			// objects, because a fresh span that had been allocated into,
   754  			// wasn't totally filled, but then swept, still has all of its
   755  			// free slots zeroed.
   756  			s.needzero = 1
   757  			stats := memstats.heapStats.acquire()
   758  			atomic.Xadd64(&stats.smallFreeCount[spc.sizeclass()], int64(nfreed))
   759  			memstats.heapStats.release()
   760  
   761  			// Count the frees in the inconsistent, internal stats.
   762  			gcController.totalFree.Add(int64(nfreed) * int64(s.elemsize))
   763  		}
   764  		if !preserve {
   765  			// The caller may not have removed this span from whatever
   766  			// unswept set its on but taken ownership of the span for
   767  			// sweeping by updating sweepgen. If this span still is in
   768  			// an unswept set, then the mcentral will pop it off the
   769  			// set, check its sweepgen, and ignore it.
   770  			if nalloc == 0 {
   771  				// Free totally free span directly back to the heap.
   772  				mheap_.freeSpan(s)
   773  				return true
   774  			}
   775  			// Return span back to the right mcentral list.
   776  			if nalloc == s.nelems {
   777  				mheap_.central[spc].mcentral.fullSwept(sweepgen).push(s)
   778  			} else {
   779  				mheap_.central[spc].mcentral.partialSwept(sweepgen).push(s)
   780  			}
   781  		}
   782  	} else if !preserve {
   783  		// Handle spans for large objects.
   784  		if nfreed != 0 {
   785  			// Free large object span to heap.
   786  
   787  			// Count the free in the consistent, external stats.
   788  			//
   789  			// Do this before freeSpan, which might update heapStats' inHeap
   790  			// value. If it does so, then metrics that subtract object footprint
   791  			// from inHeap might overflow. See #67019.
   792  			stats := memstats.heapStats.acquire()
   793  			atomic.Xadd64(&stats.largeFreeCount, 1)
   794  			atomic.Xadd64(&stats.largeFree, int64(size))
   795  			memstats.heapStats.release()
   796  
   797  			// Count the free in the inconsistent, internal stats.
   798  			gcController.totalFree.Add(int64(size))
   799  
   800  			// NOTE(rsc,dvyukov): The original implementation of efence
   801  			// in CL 22060046 used sysFree instead of sysFault, so that
   802  			// the operating system would eventually give the memory
   803  			// back to us again, so that an efence program could run
   804  			// longer without running out of memory. Unfortunately,
   805  			// calling sysFree here without any kind of adjustment of the
   806  			// heap data structures means that when the memory does
   807  			// come back to us, we have the wrong metadata for it, either in
   808  			// the mspan structures or in the garbage collection bitmap.
   809  			// Using sysFault here means that the program will run out of
   810  			// memory fairly quickly in efence mode, but at least it won't
   811  			// have mysterious crashes due to confused memory reuse.
   812  			// It should be possible to switch back to sysFree if we also
   813  			// implement and then call some kind of mheap.deleteSpan.
   814  			if debug.efence > 0 {
   815  				s.limit = 0 // prevent mlookup from finding this span
   816  				sysFault(unsafe.Pointer(s.base()), size)
   817  			} else {
   818  				mheap_.freeSpan(s)
   819  			}
   820  			return true
   821  		}
   822  
   823  		// Add a large span directly onto the full+swept list.
   824  		mheap_.central[spc].mcentral.fullSwept(sweepgen).push(s)
   825  	}
   826  	return false
   827  }
   828  
   829  // reportZombies reports any marked but free objects in s and throws.
   830  //
   831  // This generally means one of the following:
   832  //
   833  // 1. User code converted a pointer to a uintptr and then back
   834  // unsafely, and a GC ran while the uintptr was the only reference to
   835  // an object.
   836  //
   837  // 2. User code (or a compiler bug) constructed a bad pointer that
   838  // points to a free slot, often a past-the-end pointer.
   839  //
   840  // 3. The GC two cycles ago missed a pointer and freed a live object,
   841  // but it was still live in the last cycle, so this GC cycle found a
   842  // pointer to that object and marked it.
   843  func (s *mspan) reportZombies() {
   844  	printlock()
   845  	print("runtime: marked free object in span ", s, ", elemsize=", s.elemsize, " freeindex=", s.freeindex, " (bad use of unsafe.Pointer or having race conditions? try -d=checkptr or -race)\n")
   846  	mbits := s.markBitsForBase()
   847  	abits := s.allocBitsForIndex(0)
   848  	for i := uintptr(0); i < uintptr(s.nelems); i++ {
   849  		addr := s.base() + i*s.elemsize
   850  		print(hex(addr))
   851  		alloc := i < uintptr(s.freeindex) || abits.isMarked()
   852  		if alloc {
   853  			print(" alloc")
   854  		} else {
   855  			print(" free ")
   856  		}
   857  		if mbits.isMarked() {
   858  			print(" marked  ")
   859  		} else {
   860  			print(" unmarked")
   861  		}
   862  		zombie := mbits.isMarked() && !alloc
   863  		if zombie {
   864  			print(" zombie")
   865  		}
   866  		print("\n")
   867  		if zombie {
   868  			length := s.elemsize
   869  			if length > 1024 {
   870  				length = 1024
   871  			}
   872  			hexdumpWords(addr, addr+length, nil)
   873  		}
   874  		mbits.advance()
   875  		abits.advance()
   876  	}
   877  	throw("found pointer to free object")
   878  }
   879  
   880  // deductSweepCredit deducts sweep credit for allocating a span of
   881  // size spanBytes. This must be performed *before* the span is
   882  // allocated to ensure the system has enough credit. If necessary, it
   883  // performs sweeping to prevent going in to debt. If the caller will
   884  // also sweep pages (e.g., for a large allocation), it can pass a
   885  // non-zero callerSweepPages to leave that many pages unswept.
   886  //
   887  // deductSweepCredit makes a worst-case assumption that all spanBytes
   888  // bytes of the ultimately allocated span will be available for object
   889  // allocation.
   890  //
   891  // deductSweepCredit is the core of the "proportional sweep" system.
   892  // It uses statistics gathered by the garbage collector to perform
   893  // enough sweeping so that all pages are swept during the concurrent
   894  // sweep phase between GC cycles.
   895  //
   896  // mheap_ must NOT be locked.
   897  func deductSweepCredit(spanBytes uintptr, callerSweepPages uintptr) {
   898  	if mheap_.sweepPagesPerByte == 0 {
   899  		// Proportional sweep is done or disabled.
   900  		return
   901  	}
   902  
   903  	trace := traceAcquire()
   904  	if trace.ok() {
   905  		trace.GCSweepStart()
   906  		traceRelease(trace)
   907  	}
   908  
   909  	// Fix debt if necessary.
   910  retry:
   911  	sweptBasis := mheap_.pagesSweptBasis.Load()
   912  	live := gcController.heapLive.Load()
   913  	liveBasis := mheap_.sweepHeapLiveBasis
   914  	newHeapLive := spanBytes
   915  	if liveBasis < live {
   916  		// Only do this subtraction when we don't overflow. Otherwise, pagesTarget
   917  		// might be computed as something really huge, causing us to get stuck
   918  		// sweeping here until the next mark phase.
   919  		//
   920  		// Overflow can happen here if gcPaceSweeper is called concurrently with
   921  		// sweeping (i.e. not during a STW, like it usually is) because this code
   922  		// is intentionally racy. A concurrent call to gcPaceSweeper can happen
   923  		// if a GC tuning parameter is modified and we read an older value of
   924  		// heapLive than what was used to set the basis.
   925  		//
   926  		// This state should be transient, so it's fine to just let newHeapLive
   927  		// be a relatively small number. We'll probably just skip this attempt to
   928  		// sweep.
   929  		//
   930  		// See issue #57523.
   931  		newHeapLive += uintptr(live - liveBasis)
   932  	}
   933  	pagesTarget := int64(mheap_.sweepPagesPerByte*float64(newHeapLive)) - int64(callerSweepPages)
   934  	for pagesTarget > int64(mheap_.pagesSwept.Load()-sweptBasis) {
   935  		if sweepone() == ^uintptr(0) {
   936  			mheap_.sweepPagesPerByte = 0
   937  			break
   938  		}
   939  		if mheap_.pagesSweptBasis.Load() != sweptBasis {
   940  			// Sweep pacing changed. Recompute debt.
   941  			goto retry
   942  		}
   943  	}
   944  
   945  	trace = traceAcquire()
   946  	if trace.ok() {
   947  		trace.GCSweepDone()
   948  		traceRelease(trace)
   949  	}
   950  }
   951  
   952  // clobberfree sets the memory content at x to bad content, for debugging
   953  // purposes.
   954  func clobberfree(x unsafe.Pointer, size uintptr) {
   955  	// size (span.elemsize) is always a multiple of 4.
   956  	for i := uintptr(0); i < size; i += 4 {
   957  		*(*uint32)(add(x, i)) = 0xdeadbeef
   958  	}
   959  }
   960  
   961  // gcPaceSweeper updates the sweeper's pacing parameters.
   962  //
   963  // Must be called whenever the GC's pacing is updated.
   964  //
   965  // The world must be stopped, or mheap_.lock must be held.
   966  func gcPaceSweeper(trigger uint64) {
   967  	assertWorldStoppedOrLockHeld(&mheap_.lock)
   968  
   969  	// Update sweep pacing.
   970  	if isSweepDone() {
   971  		mheap_.sweepPagesPerByte = 0
   972  	} else {
   973  		// Concurrent sweep needs to sweep all of the in-use
   974  		// pages by the time the allocated heap reaches the GC
   975  		// trigger. Compute the ratio of in-use pages to sweep
   976  		// per byte allocated, accounting for the fact that
   977  		// some might already be swept.
   978  		heapLiveBasis := gcController.heapLive.Load()
   979  		heapDistance := int64(trigger) - int64(heapLiveBasis)
   980  		// Add a little margin so rounding errors and
   981  		// concurrent sweep are less likely to leave pages
   982  		// unswept when GC starts.
   983  		heapDistance -= 1024 * 1024
   984  		if heapDistance < _PageSize {
   985  			// Avoid setting the sweep ratio extremely high
   986  			heapDistance = _PageSize
   987  		}
   988  		pagesSwept := mheap_.pagesSwept.Load()
   989  		pagesInUse := mheap_.pagesInUse.Load()
   990  		sweepDistancePages := int64(pagesInUse) - int64(pagesSwept)
   991  		if sweepDistancePages <= 0 {
   992  			mheap_.sweepPagesPerByte = 0
   993  		} else {
   994  			mheap_.sweepPagesPerByte = float64(sweepDistancePages) / float64(heapDistance)
   995  			mheap_.sweepHeapLiveBasis = heapLiveBasis
   996  			// Write pagesSweptBasis last, since this
   997  			// signals concurrent sweeps to recompute
   998  			// their debt.
   999  			mheap_.pagesSweptBasis.Store(pagesSwept)
  1000  		}
  1001  	}
  1002  }
  1003  

View as plain text