src/runtime/mheap.go

// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

// Page heap.
//
// See malloc.go for overview.

package runtime

import (
	"internal/cpu"
	"runtime/internal/atomic"
	"runtime/internal/sys"
	"unsafe"
)

// minPhysPageSize is a lower-bound on the physical page size. The
// true physical page size may be larger than this. In contrast,
// sys.PhysPageSize is an upper-bound on the physical page size.
const minPhysPageSize = 4096

// Main malloc heap.
// The heap itself is the "free[]" and "large" arrays,
// but all the other global data is here too.
//
// mheap must not be heap-allocated because it contains mSpanLists,
// which must not be heap-allocated.
//
//go:notinheap
type mheap struct {
	lock      mutex
	free      mTreap    // free treap of spans
	busy      mSpanList // busy list of spans
	sweepgen  uint32    // sweep generation, see comment in mspan
	sweepdone uint32    // all spans are swept
	sweepers  uint32    // number of active sweepone calls

	// allspans is a slice of all mspans ever created. Each mspan
	// appears exactly once.
	//
	// The memory for allspans is manually managed and can be
	// reallocated and move as the heap grows.
	//
	// In general, allspans is protected by mheap_.lock, which
	// prevents concurrent access as well as freeing the backing
	// store. Accesses during STW might not hold the lock, but
	// must ensure that allocation cannot happen around the
	// access (since that may free the backing store).
	allspans []*mspan // all spans out there

	// sweepSpans contains two mspan stacks: one of swept in-use
	// spans, and one of unswept in-use spans. These two trade
	// roles on each GC cycle. Since the sweepgen increases by 2
	// on each cycle, this means the swept spans are in
	// sweepSpans[sweepgen/2%2] and the unswept spans are in
	// sweepSpans[1-sweepgen/2%2]. Sweeping pops spans from the
	// unswept stack and pushes spans that are still in-use on the
	// swept stack. Likewise, allocating an in-use span pushes it
	// on the swept stack.
	sweepSpans [2]gcSweepBuf

	//_ uint32 // align uint64 fields on 32-bit for atomics

	// Proportional sweep
	//
	// These parameters represent a linear function from heap_live
	// to page sweep count. The proportional sweep system works to
	// stay in the black by keeping the current page sweep count
	// above this line at the current heap_live.
	//
	// The line has slope sweepPagesPerByte and passes through a
	// basis point at (sweepHeapLiveBasis, pagesSweptBasis). At
	// any given time, the system is at (memstats.heap_live,
	// pagesSwept) in this space.
	//
	// It's important that the line pass through a point we
	// control rather than simply starting at a (0,0) origin
	// because that lets us adjust sweep pacing at any time while
	// accounting for current progress. If we could only adjust
	// the slope, it would create a discontinuity in debt if any
	// progress has already been made.
	pagesInUse         uint64  // pages of spans in stats mSpanInUse; R/W with mheap.lock
	pagesSwept         uint64  // pages swept this cycle; updated atomically
	pagesSweptBasis    uint64  // pagesSwept to use as the origin of the sweep ratio; updated atomically
	sweepHeapLiveBasis uint64  // value of heap_live to use as the origin of sweep ratio; written with lock, read without
	sweepPagesPerByte  float64 // proportional sweep ratio; written with lock, read without
	// TODO(austin): pagesInUse should be a uintptr, but the 386
	// compiler can't 8-byte align fields.

	// Malloc stats.
	largealloc  uint64                  // bytes allocated for large objects
	nlargealloc uint64                  // number of large object allocations
	largefree   uint64                  // bytes freed for large objects (>maxsmallsize)
	nlargefree  uint64                  // number of frees for large objects (>maxsmallsize)
	nsmallfree  [_NumSizeClasses]uint64 // number of frees for small objects (<=maxsmallsize)

	// arenas is the heap arena map. It points to the metadata for
	// the heap for every arena frame of the entire usable virtual
	// address space.
	//
	// Use arenaIndex to compute indexes into this array.
	//
	// For regions of the address space that are not backed by the
	// Go heap, the arena map contains nil.
	//
	// Modifications are protected by mheap_.lock. Reads can be
	// performed without locking; however, a given entry can
	// transition from nil to non-nil at any time when the lock
	// isn't held. (Entries never transitions back to nil.)
	//
	// In general, this is a two-level mapping consisting of an L1
	// map and possibly many L2 maps. This saves space when there
	// are a huge number of arena frames. However, on many
	// platforms (even 64-bit), arenaL1Bits is 0, making this
	// effectively a single-level map. In this case, arenas[0]
	// will never be nil.
	arenas [1 << arenaL1Bits]*[1 << arenaL2Bits]*heapArena

	// heapArenaAlloc is pre-reserved space for allocating heapArena
	// objects. This is only used on 32-bit, where we pre-reserve
	// this space to avoid interleaving it with the heap itself.
	heapArenaAlloc linearAlloc

	// arenaHints is a list of addresses at which to attempt to
	// add more heap arenas. This is initially populated with a
	// set of general hint addresses, and grown with the bounds of
	// actual heap arena ranges.
	arenaHints *arenaHint

	// arena is a pre-reserved space for allocating heap arenas
	// (the actual arenas). This is only used on 32-bit.
	arena linearAlloc

	//_ uint32 // ensure 64-bit alignment of central

	// central free lists for small size classes.
	// the padding makes sure that the MCentrals are
	// spaced CacheLinePadSize bytes apart, so that each MCentral.lock
	// gets its own cache line.
	// central is indexed by spanClass.
	central [numSpanClasses]struct {
		mcentral mcentral
		pad      [cpu.CacheLinePadSize - unsafe.Sizeof(mcentral{})%cpu.CacheLinePadSize]byte
	}

	spanalloc             fixalloc // allocator for span*
	cachealloc            fixalloc // allocator for mcache*
	treapalloc            fixalloc // allocator for treapNodes* used by large objects
	specialfinalizeralloc fixalloc // allocator for specialfinalizer*
	specialprofilealloc   fixalloc // allocator for specialprofile*
	speciallock           mutex    // lock for special record allocators.
	arenaHintAlloc        fixalloc // allocator for arenaHints

	unused *specialfinalizer // never set, just here to force the specialfinalizer type into DWARF
}

var mheap_ mheap

// A heapArena stores metadata for a heap arena. heapArenas are stored
// outside of the Go heap and accessed via the mheap_.arenas index.
//
// This gets allocated directly from the OS, so ideally it should be a
// multiple of the system page size. For example, avoid adding small
// fields.
//
//go:notinheap
type heapArena struct {
	// bitmap stores the pointer/scalar bitmap for the words in
	// this arena. See mbitmap.go for a description. Use the
	// heapBits type to access this.
	bitmap [heapArenaBitmapBytes]byte

	// spans maps from virtual address page ID within this arena to *mspan.
	// For allocated spans, their pages map to the span itself.
	// For free spans, only the lowest and highest pages map to the span itself.
	// Internal pages map to an arbitrary span.
	// For pages that have never been allocated, spans entries are nil.
	//
	// Modifications are protected by mheap.lock. Reads can be
	// performed without locking, but ONLY from indexes that are
	// known to contain in-use or stack spans. This means there
	// must not be a safe-point between establishing that an
	// address is live and looking it up in the spans array.
	spans [pagesPerArena]*mspan
}

// arenaHint is a hint for where to grow the heap arenas. See
// mheap_.arenaHints.
//
//go:notinheap
type arenaHint struct {
	addr uintptr
	down bool
	next *arenaHint
}

// An MSpan is a run of pages.
//
// When a MSpan is in the heap free list, state == mSpanFree
// and heapmap(s->start) == span, heapmap(s->start+s->npages-1) == span.
//
// When a MSpan is allocated, state == mSpanInUse or mSpanManual
// and heapmap(i) == span for all s->start <= i < s->start+s->npages.

// Every MSpan is in one doubly-linked list,
// either one of the MHeap's free lists or one of the
// MCentral's span lists.

// An MSpan representing actual memory has state mSpanInUse,
// mSpanManual, or mSpanFree. Transitions between these states are
// constrained as follows:
//
// * A span may transition from free to in-use or manual during any GC
//   phase.
//
// * During sweeping (gcphase == _GCoff), a span may transition from
//   in-use to free (as a result of sweeping) or manual to free (as a
//   result of stacks being freed).
//
// * During GC (gcphase != _GCoff), a span *must not* transition from
//   manual or in-use to free. Because concurrent GC may read a pointer
//   and then look up its span, the span state must be monotonic.
type mSpanState uint8

const (
	mSpanDead   mSpanState = iota
	mSpanInUse             // allocated for garbage collected heap
	mSpanManual            // allocated for manual management (e.g., stack allocator)
	mSpanFree
)

// mSpanStateNames are the names of the span states, indexed by
// mSpanState.
var mSpanStateNames = []string{
	"mSpanDead",
	"mSpanInUse",
	"mSpanManual",
	"mSpanFree",
}

// mSpanList heads a linked list of spans.
//
//go:notinheap
type mSpanList struct {
	first *mspan // first span in list, or nil if none
	last  *mspan // last span in list, or nil if none
}

//go:notinheap
type mspan struct {
	next *mspan     // next span in list, or nil if none
	prev *mspan     // previous span in list, or nil if none
	list *mSpanList // For debugging. TODO: Remove.

	startAddr uintptr // address of first byte of span aka s.base()
	npages    uintptr // number of pages in span

	manualFreeList gclinkptr // list of free objects in mSpanManual spans

	// freeindex is the slot index between 0 and nelems at which to begin scanning
	// for the next free object in this span.
	// Each allocation scans allocBits starting at freeindex until it encounters a 0
	// indicating a free object. freeindex is then adjusted so that subsequent scans begin
	// just past the newly discovered free object.
	//
	// If freeindex == nelem, this span has no free objects.
	//
	// allocBits is a bitmap of objects in this span.
	// If n >= freeindex and allocBits[n/8] & (1<<(n%8)) is 0
	// then object n is free;
	// otherwise, object n is allocated. Bits starting at nelem are
	// undefined and should never be referenced.
	//
	// Object n starts at address n*elemsize + (start << pageShift).
	freeindex uintptr
	// TODO: Look up nelems from sizeclass and remove this field if it
	// helps performance.
	nelems uintptr // number of object in the span.

	// Cache of the allocBits at freeindex. allocCache is shifted
	// such that the lowest bit corresponds to the bit freeindex.
	// allocCache holds the complement of allocBits, thus allowing
	// ctz (count trailing zero) to use it directly.
	// allocCache may contain bits beyond s.nelems; the caller must ignore
	// these.
	allocCache uint64

	// allocBits and gcmarkBits hold pointers to a span's mark and
	// allocation bits. The pointers are 8 byte aligned.
	// There are three arenas where this data is held.
	// free: Dirty arenas that are no longer accessed
	//       and can be reused.
	// next: Holds information to be used in the next GC cycle.
	// current: Information being used during this GC cycle.
	// previous: Information being used during the last GC cycle.
	// A new GC cycle starts with the call to finishsweep_m.
	// finishsweep_m moves the previous arena to the free arena,
	// the current arena to the previous arena, and
	// the next arena to the current arena.
	// The next arena is populated as the spans request
	// memory to hold gcmarkBits for the next GC cycle as well
	// as allocBits for newly allocated spans.
	//
	// The pointer arithmetic is done "by hand" instead of using
	// arrays to avoid bounds checks along critical performance
	// paths.
	// The sweep will free the old allocBits and set allocBits to the
	// gcmarkBits. The gcmarkBits are replaced with a fresh zeroed
	// out memory.
	allocBits  *gcBits
	gcmarkBits *gcBits

	// sweep generation:
	// if sweepgen == h->sweepgen - 2, the span needs sweeping
	// if sweepgen == h->sweepgen - 1, the span is currently being swept
	// if sweepgen == h->sweepgen, the span is swept and ready to use
	// if sweepgen == h->sweepgen + 1, the span was cached before sweep began and is still cached, and needs sweeping
	// if sweepgen == h->sweepgen + 3, the span was swept and then cached and is still cached
	// h->sweepgen is incremented by 2 after every GC

	sweepgen    uint32
	divMul      uint16     // for divide by elemsize - divMagic.mul
	baseMask    uint16     // if non-0, elemsize is a power of 2, & this will get object allocation base
	allocCount  uint16     // number of allocated objects
	spanclass   spanClass  // size class and noscan (uint8)
	state       mSpanState // mspaninuse etc
	needzero    uint8      // needs to be zeroed before allocation
	divShift    uint8      // for divide by elemsize - divMagic.shift
	divShift2   uint8      // for divide by elemsize - divMagic.shift2
	elemsize    uintptr    // computed from sizeclass or from npages
	unusedsince int64      // first time spotted by gc in mspanfree state
	npreleased  uintptr    // number of pages released to the os
	limit       uintptr    // end of data in span
	speciallock mutex      // guards specials list
	specials    *special   // linked list of special records sorted by offset.
}

func (s *mspan) base() uintptr {
	return s.startAddr
}

func (s *mspan) layout() (size, n, total uintptr) {
	total = s.npages << _PageShift
	size = s.elemsize
	if size > 0 {
		n = total / size
	}
	return
}

func (s *mspan) scavenge() uintptr {
	start := s.base()
	end := start + s.npages<<_PageShift
	if physPageSize > _PageSize {
		// We can only release pages in
		// physPageSize blocks, so round start
		// and end in. (Otherwise, madvise
		// will round them *out* and release
		// more memory than we want.)
		start = (start + physPageSize - 1) &^ (physPageSize - 1)
		end &^= physPageSize - 1
		if end <= start {
			// start and end don't span a
			// whole physical page.
			return 0
		}
	}
	len := end - start
	released := len - (s.npreleased << _PageShift)
	if physPageSize > _PageSize && released == 0 {
		return 0
	}
	memstats.heap_released += uint64(released)
	s.npreleased = len >> _PageShift
	sysUnused(unsafe.Pointer(start), len)
	return released
}

// recordspan adds a newly allocated span to h.allspans.
//
// This only happens the first time a span is allocated from
// mheap.spanalloc (it is not called when a span is reused).
//
// Write barriers are disallowed here because it can be called from
// gcWork when allocating new workbufs. However, because it's an
// indirect call from the fixalloc initializer, the compiler can't see
// this.
//
//go:nowritebarrierrec
func recordspan(vh unsafe.Pointer, p unsafe.Pointer) {
	h := (*mheap)(vh)
	s := (*mspan)(p)
	if len(h.allspans) >= cap(h.allspans) {
		n := 64 * 1024 / sys.PtrSize
		if n < cap(h.allspans)*3/2 {
			n = cap(h.allspans) * 3 / 2
		}
		var new []*mspan
		sp := (*slice)(unsafe.Pointer(&new))
		sp.array = sysAlloc(uintptr(n)*sys.PtrSize, &memstats.other_sys)
		if sp.array == nil {
			throw("runtime: cannot allocate memory")
		}
		sp.len = len(h.allspans)
		sp.cap = n
		if len(h.allspans) > 0 {
			copy(new, h.allspans)
		}
		oldAllspans := h.allspans
		*(*notInHeapSlice)(unsafe.Pointer(&h.allspans)) = *(*notInHeapSlice)(unsafe.Pointer(&new))
		if len(oldAllspans) != 0 {
			sysFree(unsafe.Pointer(&oldAllspans[0]), uintptr(cap(oldAllspans))*unsafe.Sizeof(oldAllspans[0]), &memstats.other_sys)
		}
	}
	h.allspans = h.allspans[:len(h.allspans)+1]
	h.allspans[len(h.allspans)-1] = s
}

// A spanClass represents the size class and noscan-ness of a span.
//
// Each size class has a noscan spanClass and a scan spanClass. The
// noscan spanClass contains only noscan objects, which do not contain
// pointers and thus do not need to be scanned by the garbage
// collector.
type spanClass uint8

const (
	numSpanClasses = _NumSizeClasses << 1
	tinySpanClass  = spanClass(tinySizeClass<<1 | 1)
)

func makeSpanClass(sizeclass uint8, noscan bool) spanClass {
	return spanClass(sizeclass<<1) | spanClass(bool2int(noscan))
}

func (sc spanClass) sizeclass() int8 {
	return int8(sc >> 1)
}

func (sc spanClass) noscan() bool {
	return sc&1 != 0
}

// arenaIndex returns the index into mheap_.arenas of the arena
// containing metadata for p. This index combines of an index into the
// L1 map and an index into the L2 map and should be used as
// mheap_.arenas[ai.l1()][ai.l2()].
//
// If p is outside the range of valid heap addresses, either l1() or
// l2() will be out of bounds.
//
// It is nosplit because it's called by spanOf and several other
// nosplit functions.
//
//go:nosplit
func arenaIndex(p uintptr) arenaIdx {
	return arenaIdx((p + arenaBaseOffset) / heapArenaBytes)
}

// arenaBase returns the low address of the region covered by heap
// arena i.
func arenaBase(i arenaIdx) uintptr {
	return uintptr(i)*heapArenaBytes - arenaBaseOffset
}

type arenaIdx uint

func (i arenaIdx) l1() uint {
	if arenaL1Bits == 0 {
		// Let the compiler optimize this away if there's no
		// L1 map.
		return 0
	} else {
		return uint(i) >> arenaL1Shift
	}
}

func (i arenaIdx) l2() uint {
	if arenaL1Bits == 0 {
		return uint(i)
	} else {
		return uint(i) & (1<<arenaL2Bits - 1)
	}
}

// inheap reports whether b is a pointer into a (potentially dead) heap object.
// It returns false for pointers into mSpanManual spans.
// Non-preemptible because it is used by write barriers.
//go:nowritebarrier
//go:nosplit
func inheap(b uintptr) bool {
	return spanOfHeap(b) != nil
}

// inHeapOrStack is a variant of inheap that returns true for pointers
// into any allocated heap span.
//
//go:nowritebarrier
//go:nosplit
func inHeapOrStack(b uintptr) bool {
	s := spanOf(b)
	if s == nil || b < s.base() {
		return false
	}
	switch s.state {
	case mSpanInUse, mSpanManual:
		return b < s.limit
	default:
		return false
	}
}

// spanOf returns the span of p. If p does not point into the heap
// arena or no span has ever contained p, spanOf returns nil.
//
// If p does not point to allocated memory, this may return a non-nil
// span that does *not* contain p. If this is a possibility, the
// caller should either call spanOfHeap or check the span bounds
// explicitly.
//
// Must be nosplit because it has callers that are nosplit.
//
//go:nosplit
func spanOf(p uintptr) *mspan {
	// This function looks big, but we use a lot of constant
	// folding around arenaL1Bits to get it under the inlining
	// budget. Also, many of the checks here are safety checks
	// that Go needs to do anyway, so the generated code is quite
	// short.
	ri := arenaIndex(p)
	if arenaL1Bits == 0 {
		// If there's no L1, then ri.l1() can't be out of bounds but ri.l2() can.
		if ri.l2() >= uint(len(mheap_.arenas[0])) {
			return nil
		}
	} else {
		// If there's an L1, then ri.l1() can be out of bounds but ri.l2() can't.
		if ri.l1() >= uint(len(mheap_.arenas)) {
			return nil
		}
	}
	l2 := mheap_.arenas[ri.l1()]
	if arenaL1Bits != 0 && l2 == nil { // Should never happen if there's no L1.
		return nil
	}
	ha := l2[ri.l2()]
	if ha == nil {
		return nil
	}
	return ha.spans[(p/pageSize)%pagesPerArena]
}

// spanOfUnchecked is equivalent to spanOf, but the caller must ensure
// that p points into an allocated heap arena.
//
// Must be nosplit because it has callers that are nosplit.
//
//go:nosplit
func spanOfUnchecked(p uintptr) *mspan {
	ai := arenaIndex(p)
	return mheap_.arenas[ai.l1()][ai.l2()].spans[(p/pageSize)%pagesPerArena]
}

// spanOfHeap is like spanOf, but returns nil if p does not point to a
// heap object.
//
// Must be nosplit because it has callers that are nosplit.
//
//go:nosplit
func spanOfHeap(p uintptr) *mspan {
	s := spanOf(p)
	// If p is not allocated, it may point to a stale span, so we
	// have to check the span's bounds and state.
	if s == nil || p < s.base() || p >= s.limit || s.state != mSpanInUse {
		return nil
	}
	return s
}

// Initialize the heap.
func (h *mheap) init() {
	h.treapalloc.init(unsafe.Sizeof(treapNode{}), nil, nil, &memstats.other_sys)
	h.spanalloc.init(unsafe.Sizeof(mspan{}), recordspan, unsafe.Pointer(h), &memstats.mspan_sys)
	h.cachealloc.init(unsafe.Sizeof(mcache{}), nil, nil, &memstats.mcache_sys)
	h.specialfinalizeralloc.init(unsafe.Sizeof(specialfinalizer{}), nil, nil, &memstats.other_sys)
	h.specialprofilealloc.init(unsafe.Sizeof(specialprofile{}), nil, nil, &memstats.other_sys)
	h.arenaHintAlloc.init(unsafe.Sizeof(arenaHint{}), nil, nil, &memstats.other_sys)

	// Don't zero mspan allocations. Background sweeping can
	// inspect a span concurrently with allocating it, so it's
	// important that the span's sweepgen survive across freeing
	// and re-allocating a span to prevent background sweeping
	// from improperly cas'ing it from 0.
	//
	// This is safe because mspan contains no heap pointers.
	h.spanalloc.zero = false

	// h->mapcache needs no init
	h.busy.init()
	for i := range h.central {
		h.central[i].mcentral.init(spanClass(i))
	}
}

// Sweeps spans in list until reclaims at least npages into heap.
// Returns the actual number of pages reclaimed.
func (h *mheap) reclaimList(list *mSpanList, npages uintptr) uintptr {
	n := uintptr(0)
	sg := mheap_.sweepgen
retry:
	for s := list.first; s != nil; s = s.next {
		if s.sweepgen == sg-2 && atomic.Cas(&s.sweepgen, sg-2, sg-1) {
			list.remove(s)
			// swept spans are at the end of the list
			list.insertBack(s) // Puts it back on a busy list. s is not in the treap at this point.
			unlock(&h.lock)
			snpages := s.npages
			if s.sweep(false) {
				n += snpages
			}
			lock(&h.lock)
			if n >= npages {
				return n
			}
			// the span could have been moved elsewhere
			goto retry
		}
		if s.sweepgen == sg-1 {
			// the span is being swept by background sweeper, skip
			continue
		}
		// already swept empty span,
		// all subsequent ones must also be either swept or in process of sweeping
		break
	}
	return n
}

// Sweeps and reclaims at least npage pages into heap.
// Called before allocating npage pages.
func (h *mheap) reclaim(npage uintptr) {
	if h.reclaimList(&h.busy, npage) != 0 {
		return // Bingo!
	}

	// Now sweep everything that is not yet swept.
	var reclaimed uintptr
	unlock(&h.lock)
	for {
		n := sweepone()
		if n == ^uintptr(0) { // all spans are swept
			break
		}
		reclaimed += n
		if reclaimed >= npage {
			break
		}
	}
	lock(&h.lock)
}

// alloc_m is the internal implementation of mheap.alloc.
//
// alloc_m must run on the system stack because it locks the heap, so
// any stack growth during alloc_m would self-deadlock.
//
//go:systemstack
func (h *mheap) alloc_m(npage uintptr, spanclass spanClass, large bool) *mspan {
	_g_ := getg()
	lock(&h.lock)

	// To prevent excessive heap growth, before allocating n pages
	// we need to sweep and reclaim at least n pages.
	if h.sweepdone == 0 {
		// TODO(austin): This tends to sweep a large number of
		// spans in order to find a few completely free spans
		// (for example, in the garbage benchmark, this sweeps
		// ~30x the number of pages it's trying to allocate).
		// If GC kept a bit for whether there were any marks
		// in a span, we could release these free spans
		// at the end of GC and eliminate this entirely.
		if trace.enabled {
			traceGCSweepStart()
		}
		h.reclaim(npage)
		if trace.enabled {
			traceGCSweepDone()
		}
	}

	// transfer stats from cache to global
	memstats.heap_scan += uint64(_g_.m.mcache.local_scan)
	_g_.m.mcache.local_scan = 0
	memstats.tinyallocs += uint64(_g_.m.mcache.local_tinyallocs)
	_g_.m.mcache.local_tinyallocs = 0

	s := h.allocSpanLocked(npage, &memstats.heap_inuse)
	if s != nil {
		// Record span info, because gc needs to be
		// able to map interior pointer to containing span.
		atomic.Store(&s.sweepgen, h.sweepgen)
		h.sweepSpans[h.sweepgen/2%2].push(s) // Add to swept in-use list.
		s.state = mSpanInUse
		s.allocCount = 0
		s.spanclass = spanclass
		if sizeclass := spanclass.sizeclass(); sizeclass == 0 {
			s.elemsize = s.npages << _PageShift
			s.divShift = 0
			s.divMul = 0
			s.divShift2 = 0
			s.baseMask = 0
		} else {
			s.elemsize = uintptr(class_to_size[sizeclass])
			m := &class_to_divmagic[sizeclass]
			s.divShift = m.shift
			s.divMul = m.mul
			s.divShift2 = m.shift2
			s.baseMask = m.baseMask
		}

		// update stats, sweep lists
		h.pagesInUse += uint64(npage)
		if large {
			memstats.heap_objects++
			mheap_.largealloc += uint64(s.elemsize)
			mheap_.nlargealloc++
			atomic.Xadd64(&memstats.heap_live, int64(npage<<_PageShift))
			// Swept spans are at the end of lists.
			h.busy.insertBack(s)
		}
	}
	// heap_scan and heap_live were updated.
	if gcBlackenEnabled != 0 {
		gcController.revise()
	}

	if trace.enabled {
		traceHeapAlloc()
	}

	// h.spans is accessed concurrently without synchronization
	// from other threads. Hence, there must be a store/store
	// barrier here to ensure the writes to h.spans above happen
	// before the caller can publish a pointer p to an object
	// allocated from s. As soon as this happens, the garbage
	// collector running on another processor could read p and
	// look up s in h.spans. The unlock acts as the barrier to
	// order these writes. On the read side, the data dependency
	// between p and the index in h.spans orders the reads.
	unlock(&h.lock)
	return s
}

// alloc allocates a new span of npage pages from the GC'd heap.
//
// Either large must be true or spanclass must indicates the span's
// size class and scannability.
//
// If needzero is true, the memory for the returned span will be zeroed.
func (h *mheap) alloc(npage uintptr, spanclass spanClass, large bool, needzero bool) *mspan {
	// Don't do any operations that lock the heap on the G stack.
	// It might trigger stack growth, and the stack growth code needs
	// to be able to allocate heap.
	var s *mspan
	systemstack(func() {
		s = h.alloc_m(npage, spanclass, large)
	})

	if s != nil {
		if needzero && s.needzero != 0 {
			memclrNoHeapPointers(unsafe.Pointer(s.base()), s.npages<<_PageShift)
		}
		s.needzero = 0
	}
	return s
}

// allocManual allocates a manually-managed span of npage pages.
// allocManual returns nil if allocation fails.
//
// allocManual adds the bytes used to *stat, which should be a
// memstats in-use field. Unlike allocations in the GC'd heap, the
// allocation does *not* count toward heap_inuse or heap_sys.
//
// The memory backing the returned span may not be zeroed if
// span.needzero is set.
//
// allocManual must be called on the system stack to prevent stack
// growth. Since this is used by the stack allocator, stack growth
// during allocManual would self-deadlock.
//
//go:systemstack
func (h *mheap) allocManual(npage uintptr, stat *uint64) *mspan {
	lock(&h.lock)
	s := h.allocSpanLocked(npage, stat)
	if s != nil {
		s.state = mSpanManual
		s.manualFreeList = 0
		s.allocCount = 0
		s.spanclass = 0
		s.nelems = 0
		s.elemsize = 0
		s.limit = s.base() + s.npages<<_PageShift
		// Manually managed memory doesn't count toward heap_sys.
		memstats.heap_sys -= uint64(s.npages << _PageShift)
	}

	// This unlock acts as a release barrier. See mheap.alloc_m.
	unlock(&h.lock)

	return s
}

// setSpan modifies the span map so spanOf(base) is s.
func (h *mheap) setSpan(base uintptr, s *mspan) {
	ai := arenaIndex(base)
	h.arenas[ai.l1()][ai.l2()].spans[(base/pageSize)%pagesPerArena] = s
}

// setSpans modifies the span map so [spanOf(base), spanOf(base+npage*pageSize))
// is s.
func (h *mheap) setSpans(base, npage uintptr, s *mspan) {
	p := base / pageSize
	ai := arenaIndex(base)
	ha := h.arenas[ai.l1()][ai.l2()]
	for n := uintptr(0); n < npage; n++ {
		i := (p + n) % pagesPerArena
		if i == 0 {
			ai = arenaIndex(base + n*pageSize)
			ha = h.arenas[ai.l1()][ai.l2()]
		}
		ha.spans[i] = s
	}
}

// Allocates a span of the given size.  h must be locked.
// The returned span has been removed from the
// free list, but its state is still mSpanFree.
func (h *mheap) allocSpanLocked(npage uintptr, stat *uint64) *mspan {
	var s *mspan

	// Best fit in the treap of spans.
	s = h.free.remove(npage)
	if s == nil {
		if !h.grow(npage) {
			return nil
		}
		s = h.free.remove(npage)
		if s == nil {
			return nil
		}
	}

	// Mark span in use.
	if s.state != mSpanFree {
		throw("MHeap_AllocLocked - MSpan not free")
	}
	if s.npages < npage {
		throw("MHeap_AllocLocked - bad npages")
	}
	if s.npreleased > 0 {
		sysUsed(unsafe.Pointer(s.base()), s.npages<<_PageShift)
		memstats.heap_released -= uint64(s.npreleased << _PageShift)
		s.npreleased = 0
	}

	if s.npages > npage {
		// Trim extra and put it back in the heap.
		t := (*mspan)(h.spanalloc.alloc())
		t.init(s.base()+npage<<_PageShift, s.npages-npage)
		s.npages = npage
		h.setSpan(t.base()-1, s)
		h.setSpan(t.base(), t)
		h.setSpan(t.base()+t.npages*pageSize-1, t)
		t.needzero = s.needzero
		s.state = mSpanManual // prevent coalescing with s
		t.state = mSpanManual
		h.freeSpanLocked(t, false, false, s.unusedsince)
		s.state = mSpanFree
	}
	s.unusedsince = 0

	h.setSpans(s.base(), npage, s)

	*stat += uint64(npage << _PageShift)
	memstats.heap_idle -= uint64(npage << _PageShift)

	//println("spanalloc", hex(s.start<<_PageShift))
	if s.inList() {
		throw("still in list")
	}
	return s
}

// Try to add at least npage pages of memory to the heap,
// returning whether it worked.
//
// h must be locked.
func (h *mheap) grow(npage uintptr) bool {
	ask := npage << _PageShift
	v, size := h.sysAlloc(ask)
	if v == nil {
		print("runtime: out of memory: cannot allocate ", ask, "-byte block (", memstats.heap_sys, " in use)\n")
		return false
	}

	// Create a fake "in use" span and free it, so that the
	// right coalescing happens.
	s := (*mspan)(h.spanalloc.alloc())
	s.init(uintptr(v), size/pageSize)
	h.setSpans(s.base(), s.npages, s)
	atomic.Store(&s.sweepgen, h.sweepgen)
	s.state = mSpanInUse
	h.pagesInUse += uint64(s.npages)
	h.freeSpanLocked(s, false, true, 0)
	return true
}

// Free the span back into the heap.
//
// large must match the value of large passed to mheap.alloc. This is
// used for accounting.
func (h *mheap) freeSpan(s *mspan, large bool) {
	systemstack(func() {
		mp := getg().m
		lock(&h.lock)
		memstats.heap_scan += uint64(mp.mcache.local_scan)
		mp.mcache.local_scan = 0
		memstats.tinyallocs += uint64(mp.mcache.local_tinyallocs)
		mp.mcache.local_tinyallocs = 0
		if msanenabled {
			// Tell msan that this entire span is no longer in use.
			base := unsafe.Pointer(s.base())
			bytes := s.npages << _PageShift
			msanfree(base, bytes)
		}
		if large {
			// Match accounting done in mheap.alloc.
			memstats.heap_objects--
		}
		if gcBlackenEnabled != 0 {
			// heap_scan changed.
			gcController.revise()
		}
		h.freeSpanLocked(s, true, true, 0)
		unlock(&h.lock)
	})
}

// freeManual frees a manually-managed span returned by allocManual.
// stat must be the same as the stat passed to the allocManual that
// allocated s.
//
// This must only be called when gcphase == _GCoff. See mSpanState for
// an explanation.
//
// freeManual must be called on the system stack to prevent stack
// growth, just like allocManual.
//
//go:systemstack
func (h *mheap) freeManual(s *mspan, stat *uint64) {
	s.needzero = 1
	lock(&h.lock)
	*stat -= uint64(s.npages << _PageShift)
	memstats.heap_sys += uint64(s.npages << _PageShift)
	h.freeSpanLocked(s, false, true, 0)
	unlock(&h.lock)
}

// s must be on the busy list or unlinked.
func (h *mheap) freeSpanLocked(s *mspan, acctinuse, acctidle bool, unusedsince int64) {
	switch s.state {
	case mSpanManual:
		if s.allocCount != 0 {
			throw("MHeap_FreeSpanLocked - invalid stack free")
		}
	case mSpanInUse:
		if s.allocCount != 0 || s.sweepgen != h.sweepgen {
			print("MHeap_FreeSpanLocked - span ", s, " ptr ", hex(s.base()), " allocCount ", s.allocCount, " sweepgen ", s.sweepgen, "/", h.sweepgen, "\n")
			throw("MHeap_FreeSpanLocked - invalid free")
		}
		h.pagesInUse -= uint64(s.npages)
	default:
		throw("MHeap_FreeSpanLocked - invalid span state")
	}

	if acctinuse {
		memstats.heap_inuse -= uint64(s.npages << _PageShift)
	}
	if acctidle {
		memstats.heap_idle += uint64(s.npages << _PageShift)
	}
	s.state = mSpanFree
	if s.inList() {
		h.busy.remove(s)
	}

	// Stamp newly unused spans. The scavenger will use that
	// info to potentially give back some pages to the OS.
	s.unusedsince = unusedsince
	if unusedsince == 0 {
		s.unusedsince = nanotime()
	}
	s.npreleased = 0

	// Coalesce with earlier, later spans.
	if before := spanOf(s.base() - 1); before != nil && before.state == mSpanFree {
		// Now adjust s.
		s.startAddr = before.startAddr
		s.npages += before.npages
		s.npreleased = before.npreleased // absorb released pages
		s.needzero |= before.needzero
		h.setSpan(before.base(), s)
		// The size is potentially changing so the treap needs to delete adjacent nodes and
		// insert back as a combined node.
		h.free.removeSpan(before)
		before.state = mSpanDead
		h.spanalloc.free(unsafe.Pointer(before))
	}

	// Now check to see if next (greater addresses) span is free and can be coalesced.
	if after := spanOf(s.base() + s.npages*pageSize); after != nil && after.state == mSpanFree {
		s.npages += after.npages
		s.npreleased += after.npreleased
		s.needzero |= after.needzero
		h.setSpan(s.base()+s.npages*pageSize-1, s)
		h.free.removeSpan(after)
		after.state = mSpanDead
		h.spanalloc.free(unsafe.Pointer(after))
	}

	// Insert s into the free treap.
	h.free.insert(s)
}

func scavengeTreapNode(t *treapNode, now, limit uint64) uintptr {
	s := t.spanKey
	if (now-uint64(s.unusedsince)) > limit && s.npreleased != s.npages {
		if released := s.scavenge(); released != 0 {
			return released
		}
	}
	return 0
}

func (h *mheap) scavenge(k int32, now, limit uint64) {
	// Disallow malloc or panic while holding the heap lock. We do
	// this here because this is an non-mallocgc entry-point to
	// the mheap API.
	gp := getg()
	gp.m.mallocing++
	lock(&h.lock)
	sumreleased := scavengetreap(h.free.treap, now, limit)
	unlock(&h.lock)
	gp.m.mallocing--

	if debug.gctrace > 0 {
		if sumreleased > 0 {
			print("scvg", k, ": ", sumreleased>>20, " MB released\n")
		}
		print("scvg", k, ": inuse: ", memstats.heap_inuse>>20, ", idle: ", memstats.heap_idle>>20, ", sys: ", memstats.heap_sys>>20, ", released: ", memstats.heap_released>>20, ", consumed: ", (memstats.heap_sys-memstats.heap_released)>>20, " (MB)\n")
	}
}

//go:linkname runtime_debug_freeOSMemory runtime/debug.freeOSMemory
func runtime_debug_freeOSMemory() {
	GC()
	systemstack(func() { mheap_.scavenge(-1, ^uint64(0), 0) })
}

// Initialize a new span with the given start and npages.
func (span *mspan) init(base uintptr, npages uintptr) {
	// span is *not* zeroed.
	span.next = nil
	span.prev = nil
	span.list = nil
	span.startAddr = base
	span.npages = npages
	span.allocCount = 0
	span.spanclass = 0
	span.elemsize = 0
	span.state = mSpanDead
	span.unusedsince = 0
	span.npreleased = 0
	span.speciallock.key = 0
	span.specials = nil
	span.needzero = 0
	span.freeindex = 0
	span.allocBits = nil
	span.gcmarkBits = nil
}

func (span *mspan) inList() bool {
	return span.list != nil
}

// Initialize an empty doubly-linked list.
func (list *mSpanList) init() {
	list.first = nil
	list.last = nil
}

func (list *mSpanList) remove(span *mspan) {
	if span.list != list {
		print("runtime: failed MSpanList_Remove span.npages=", span.npages,
			" span=", span, " prev=", span.prev, " span.list=", span.list, " list=", list, "\n")
		throw("MSpanList_Remove")
	}
	if list.first == span {
		list.first = span.next
	} else {
		span.prev.next = span.next
	}
	if list.last == span {
		list.last = span.prev
	} else {
		span.next.prev = span.prev
	}
	span.next = nil
	span.prev = nil
	span.list = nil
}

func (list *mSpanList) isEmpty() bool {
	return list.first == nil
}

func (list *mSpanList) insert(span *mspan) {
	if span.next != nil || span.prev != nil || span.list != nil {
		println("runtime: failed MSpanList_Insert", span, span.next, span.prev, span.list)
		throw("MSpanList_Insert")
	}
	span.next = list.first
	if list.first != nil {
		// The list contains at least one span; link it in.
		// The last span in the list doesn't change.
		list.first.prev = span
	} else {
		// The list contains no spans, so this is also the last span.
		list.last = span
	}
	list.first = span
	span.list = list
}

func (list *mSpanList) insertBack(span *mspan) {
	if span.next != nil || span.prev != nil || span.list != nil {
		println("runtime: failed MSpanList_InsertBack", span, span.next, span.prev, span.list)
		throw("MSpanList_InsertBack")
	}
	span.prev = list.last
	if list.last != nil {
		// The list contains at least one span.
		list.last.next = span
	} else {
		// The list contains no spans, so this is also the first span.
		list.first = span
	}
	list.last = span
	span.list = list
}

// takeAll removes all spans from other and inserts them at the front
// of list.
func (list *mSpanList) takeAll(other *mSpanList) {
	if other.isEmpty() {
		return
	}

	// Reparent everything in other to list.
	for s := other.first; s != nil; s = s.next {
		s.list = list
	}

	// Concatenate the lists.
	if list.isEmpty() {
		*list = *other
	} else {
		// Neither list is empty. Put other before list.
		other.last.next = list.first
		list.first.prev = other.last
		list.first = other.first
	}

	other.first, other.last = nil, nil
}

const (
	_KindSpecialFinalizer = 1
	_KindSpecialProfile   = 2
	// Note: The finalizer special must be first because if we're freeing
	// an object, a finalizer special will cause the freeing operation
	// to abort, and we want to keep the other special records around
	// if that happens.
)

//go:notinheap
type special struct {
	next   *special // linked list in span
	offset uint16   // span offset of object
	kind   byte     // kind of special
}

// Adds the special record s to the list of special records for
// the object p. All fields of s should be filled in except for
// offset & next, which this routine will fill in.
// Returns true if the special was successfully added, false otherwise.
// (The add will fail only if a record with the same p and s->kind
//  already exists.)
func addspecial(p unsafe.Pointer, s *special) bool {
	span := spanOfHeap(uintptr(p))
	if span == nil {
		throw("addspecial on invalid pointer")
	}

	// Ensure that the span is swept.
	// Sweeping accesses the specials list w/o locks, so we have
	// to synchronize with it. And it's just much safer.
	mp := acquirem()
	span.ensureSwept()

	offset := uintptr(p) - span.base()
	kind := s.kind

	lock(&span.speciallock)

	// Find splice point, check for existing record.
	t := &span.specials
	for {
		x := *t
		if x == nil {
			break
		}
		if offset == uintptr(x.offset) && kind == x.kind {
			unlock(&span.speciallock)
			releasem(mp)
			return false // already exists
		}
		if offset < uintptr(x.offset) || (offset == uintptr(x.offset) && kind < x.kind) {
			break
		}
		t = &x.next
	}

	// Splice in record, fill in offset.
	s.offset = uint16(offset)
	s.next = *t
	*t = s
	unlock(&span.speciallock)
	releasem(mp)

	return true
}

// Removes the Special record of the given kind for the object p.
// Returns the record if the record existed, nil otherwise.
// The caller must FixAlloc_Free the result.
func removespecial(p unsafe.Pointer, kind uint8) *special {
	span := spanOfHeap(uintptr(p))
	if span == nil {
		throw("removespecial on invalid pointer")
	}

	// Ensure that the span is swept.
	// Sweeping accesses the specials list w/o locks, so we have
	// to synchronize with it. And it's just much safer.
	mp := acquirem()
	span.ensureSwept()

	offset := uintptr(p) - span.base()

	lock(&span.speciallock)
	t := &span.specials
	for {
		s := *t
		if s == nil {
			break
		}
		// This function is used for finalizers only, so we don't check for
		// "interior" specials (p must be exactly equal to s->offset).
		if offset == uintptr(s.offset) && kind == s.kind {
			*t = s.next
			unlock(&span.speciallock)
			releasem(mp)
			return s
		}
		t = &s.next
	}
	unlock(&span.speciallock)
	releasem(mp)
	return nil
}

// The described object has a finalizer set for it.
//
// specialfinalizer is allocated from non-GC'd memory, so any heap
// pointers must be specially handled.
//
//go:notinheap
type specialfinalizer struct {
	special special
	fn      *funcval // May be a heap pointer.
	nret    uintptr
	fint    *_type   // May be a heap pointer, but always live.
	ot      *ptrtype // May be a heap pointer, but always live.
}

// Adds a finalizer to the object p. Returns true if it succeeded.
func addfinalizer(p unsafe.Pointer, f *funcval, nret uintptr, fint *_type, ot *ptrtype) bool {
	lock(&mheap_.speciallock)
	s := (*specialfinalizer)(mheap_.specialfinalizeralloc.alloc())
	unlock(&mheap_.speciallock)
	s.special.kind = _KindSpecialFinalizer
	s.fn = f
	s.nret = nret
	s.fint = fint
	s.ot = ot
	if addspecial(p, &s.special) {
		// This is responsible for maintaining the same
		// GC-related invariants as markrootSpans in any
		// situation where it's possible that markrootSpans
		// has already run but mark termination hasn't yet.
		if gcphase != _GCoff {
			base, _, _ := findObject(uintptr(p), 0, 0)
			mp := acquirem()
			gcw := &mp.p.ptr().gcw
			// Mark everything reachable from the object
			// so it's retained for the finalizer.
			scanobject(base, gcw)
			// Mark the finalizer itself, since the
			// special isn't part of the GC'd heap.
			scanblock(uintptr(unsafe.Pointer(&s.fn)), sys.PtrSize, &oneptrmask[0], gcw, nil)
			releasem(mp)
		}
		return true
	}

	// There was an old finalizer
	lock(&mheap_.speciallock)
	mheap_.specialfinalizeralloc.free(unsafe.Pointer(s))
	unlock(&mheap_.speciallock)
	return false
}

// Removes the finalizer (if any) from the object p.
func removefinalizer(p unsafe.Pointer) {
	s := (*specialfinalizer)(unsafe.Pointer(removespecial(p, _KindSpecialFinalizer)))
	if s == nil {
		return // there wasn't a finalizer to remove
	}
	lock(&mheap_.speciallock)
	mheap_.specialfinalizeralloc.free(unsafe.Pointer(s))
	unlock(&mheap_.speciallock)
}

// The described object is being heap profiled.
//
//go:notinheap
type specialprofile struct {
	special special
	b       *bucket
}

// Set the heap profile bucket associated with addr to b.
func setprofilebucket(p unsafe.Pointer, b *bucket) {
	lock(&mheap_.speciallock)
	s := (*specialprofile)(mheap_.specialprofilealloc.alloc())
	unlock(&mheap_.speciallock)
	s.special.kind = _KindSpecialProfile
	s.b = b
	if !addspecial(p, &s.special) {
		throw("setprofilebucket: profile already set")
	}
}

// Do whatever cleanup needs to be done to deallocate s. It has
// already been unlinked from the MSpan specials list.
func freespecial(s *special, p unsafe.Pointer, size uintptr) {
	switch s.kind {
	case _KindSpecialFinalizer:
		sf := (*specialfinalizer)(unsafe.Pointer(s))
		queuefinalizer(p, sf.fn, sf.nret, sf.fint, sf.ot)
		lock(&mheap_.speciallock)
		mheap_.specialfinalizeralloc.free(unsafe.Pointer(sf))
		unlock(&mheap_.speciallock)
	case _KindSpecialProfile:
		sp := (*specialprofile)(unsafe.Pointer(s))
		mProf_Free(sp.b, size)
		lock(&mheap_.speciallock)
		mheap_.specialprofilealloc.free(unsafe.Pointer(sp))
		unlock(&mheap_.speciallock)
	default:
		throw("bad special kind")
		panic("not reached")
	}
}

// gcBits is an alloc/mark bitmap. This is always used as *gcBits.
//
//go:notinheap
type gcBits uint8

// bytep returns a pointer to the n'th byte of b.
func (b *gcBits) bytep(n uintptr) *uint8 {
	return addb((*uint8)(b), n)
}

// bitp returns a pointer to the byte containing bit n and a mask for
// selecting that bit from *bytep.
func (b *gcBits) bitp(n uintptr) (bytep *uint8, mask uint8) {
	return b.bytep(n / 8), 1 << (n % 8)
}

const gcBitsChunkBytes = uintptr(64 << 10)
const gcBitsHeaderBytes = unsafe.Sizeof(gcBitsHeader{})

type gcBitsHeader struct {
	free uintptr // free is the index into bits of the next free byte.
	next uintptr // *gcBits triggers recursive type bug. (issue 14620)
}

//go:notinheap
type gcBitsArena struct {
	// gcBitsHeader // side step recursive type bug (issue 14620) by including fields by hand.
	free uintptr // free is the index into bits of the next free byte; read/write atomically
	next *gcBitsArena
	bits [gcBitsChunkBytes - gcBitsHeaderBytes]gcBits
}

var gcBitsArenas struct {
	lock     mutex
	free     *gcBitsArena
	next     *gcBitsArena // Read atomically. Write atomically under lock.
	current  *gcBitsArena
	previous *gcBitsArena
}

// tryAlloc allocates from b or returns nil if b does not have enough room.
// This is safe to call concurrently.
func (b *gcBitsArena) tryAlloc(bytes uintptr) *gcBits {
	if b == nil || atomic.Loaduintptr(&b.free)+bytes > uintptr(len(b.bits)) {
		return nil
	}
	// Try to allocate from this block.
	end := atomic.Xadduintptr(&b.free, bytes)
	if end > uintptr(len(b.bits)) {
		return nil
	}
	// There was enough room.
	start := end - bytes
	return &b.bits[start]
}

// newMarkBits returns a pointer to 8 byte aligned bytes
// to be used for a span's mark bits.
func newMarkBits(nelems uintptr) *gcBits {
	blocksNeeded := uintptr((nelems + 63) / 64)
	bytesNeeded := blocksNeeded * 8

	// Try directly allocating from the current head arena.
	head := (*gcBitsArena)(atomic.Loadp(unsafe.Pointer(&gcBitsArenas.next)))
	if p := head.tryAlloc(bytesNeeded); p != nil {
		return p
	}

	// There's not enough room in the head arena. We may need to
	// allocate a new arena.
	lock(&gcBitsArenas.lock)
	// Try the head arena again, since it may have changed. Now
	// that we hold the lock, the list head can't change, but its
	// free position still can.
	if p := gcBitsArenas.next.tryAlloc(bytesNeeded); p != nil {
		unlock(&gcBitsArenas.lock)
		return p
	}

	// Allocate a new arena. This may temporarily drop the lock.
	fresh := newArenaMayUnlock()
	// If newArenaMayUnlock dropped the lock, another thread may
	// have put a fresh arena on the "next" list. Try allocating
	// from next again.
	if p := gcBitsArenas.next.tryAlloc(bytesNeeded); p != nil {
		// Put fresh back on the free list.
		// TODO: Mark it "already zeroed"
		fresh.next = gcBitsArenas.free
		gcBitsArenas.free = fresh
		unlock(&gcBitsArenas.lock)
		return p
	}

	// Allocate from the fresh arena. We haven't linked it in yet, so
	// this cannot race and is guaranteed to succeed.
	p := fresh.tryAlloc(bytesNeeded)
	if p == nil {
		throw("markBits overflow")
	}

	// Add the fresh arena to the "next" list.
	fresh.next = gcBitsArenas.next
	atomic.StorepNoWB(unsafe.Pointer(&gcBitsArenas.next), unsafe.Pointer(fresh))

	unlock(&gcBitsArenas.lock)
	return p
}

// newAllocBits returns a pointer to 8 byte aligned bytes
// to be used for this span's alloc bits.
// newAllocBits is used to provide newly initialized spans
// allocation bits. For spans not being initialized the
// mark bits are repurposed as allocation bits when
// the span is swept.
func newAllocBits(nelems uintptr) *gcBits {
	return newMarkBits(nelems)
}

// nextMarkBitArenaEpoch establishes a new epoch for the arenas
// holding the mark bits. The arenas are named relative to the
// current GC cycle which is demarcated by the call to finishweep_m.
//
// All current spans have been swept.
// During that sweep each span allocated room for its gcmarkBits in
// gcBitsArenas.next block. gcBitsArenas.next becomes the gcBitsArenas.current
// where the GC will mark objects and after each span is swept these bits
// will be used to allocate objects.
// gcBitsArenas.current becomes gcBitsArenas.previous where the span's
// gcAllocBits live until all the spans have been swept during this GC cycle.
// The span's sweep extinguishes all the references to gcBitsArenas.previous
// by pointing gcAllocBits into the gcBitsArenas.current.
// The gcBitsArenas.previous is released to the gcBitsArenas.free list.
func nextMarkBitArenaEpoch() {
	lock(&gcBitsArenas.lock)
	if gcBitsArenas.previous != nil {
		if gcBitsArenas.free == nil {
			gcBitsArenas.free = gcBitsArenas.previous
		} else {
			// Find end of previous arenas.
			last := gcBitsArenas.previous
			for last = gcBitsArenas.previous; last.next != nil; last = last.next {
			}
			last.next = gcBitsArenas.free
			gcBitsArenas.free = gcBitsArenas.previous
		}
	}
	gcBitsArenas.previous = gcBitsArenas.current
	gcBitsArenas.current = gcBitsArenas.next
	atomic.StorepNoWB(unsafe.Pointer(&gcBitsArenas.next), nil) // newMarkBits calls newArena when needed
	unlock(&gcBitsArenas.lock)
}

// newArenaMayUnlock allocates and zeroes a gcBits arena.
// The caller must hold gcBitsArena.lock. This may temporarily release it.
func newArenaMayUnlock() *gcBitsArena {
	var result *gcBitsArena
	if gcBitsArenas.free == nil {
		unlock(&gcBitsArenas.lock)
		result = (*gcBitsArena)(sysAlloc(gcBitsChunkBytes, &memstats.gc_sys))
		if result == nil {
			throw("runtime: cannot allocate memory")
		}
		lock(&gcBitsArenas.lock)
	} else {
		result = gcBitsArenas.free
		gcBitsArenas.free = gcBitsArenas.free.next
		memclrNoHeapPointers(unsafe.Pointer(result), gcBitsChunkBytes)
	}
	result.next = nil
	// If result.bits is not 8 byte aligned adjust index so
	// that &result.bits[result.free] is 8 byte aligned.
	if uintptr(unsafe.Offsetof(gcBitsArena{}.bits))&7 == 0 {
		result.free = 0
	} else {
		result.free = 8 - (uintptr(unsafe.Pointer(&result.bits[0])) & 7)
	}
	return result
}