Source File
	arena.go

Belonging Package
	runtime

// Copyright 2022 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.

// Implementation of (safe) user arenas.
//
// This file contains the implementation of user arenas wherein Go values can
// be manually allocated and freed in bulk. The act of manually freeing memory,
// potentially before a GC cycle, means that a garbage collection cycle can be
// delayed, improving efficiency by reducing GC cycle frequency. There are other
// potential efficiency benefits, such as improved locality and access to a more
// efficient allocation strategy.
//
// What makes the arenas here safe is that once they are freed, accessing the
// arena's memory will cause an explicit program fault, and the arena's address
// space will not be reused until no more pointers into it are found. There's one
// exception to this: if an arena allocated memory that isn't exhausted, it's placed
// back into a pool for reuse. This means that a crash is not always guaranteed.
//
// While this may seem unsafe, it still prevents memory corruption, and is in fact
// necessary in order to make new(T) a valid implementation of arenas. Such a property
// is desirable to allow for a trivial implementation. (It also avoids complexities
// that arise from synchronization with the GC when trying to set the arena chunks to
// fault while the GC is active.)
//
// The implementation works in layers. At the bottom, arenas are managed in chunks.
// Each chunk must be a multiple of the heap arena size, or the heap arena size must
// be divisible by the arena chunks. The address space for each chunk, and each
// corresponding heapArena for that address space, are eternally reserved for use as
// arena chunks. That is, they can never be used for the general heap. Each chunk
// is also represented by a single mspan, and is modeled as a single large heap
// allocation. It must be, because each chunk contains ordinary Go values that may
// point into the heap, so it must be scanned just like any other object. Any
// pointer into a chunk will therefore always cause the whole chunk to be scanned
// while its corresponding arena is still live.
//
// Chunks may be allocated either from new memory mapped by the OS on our behalf,
// or by reusing old freed chunks. When chunks are freed, their underlying memory
// is returned to the OS, set to fault on access, and may not be reused until the
// program doesn't point into the chunk anymore (the code refers to this state as
// "quarantined"), a property checked by the GC.
//
// The sweeper handles moving chunks out of this quarantine state to be ready for
// reuse. When the chunk is placed into the quarantine state, its corresponding
// span is marked as noscan so that the GC doesn't try to scan memory that would
// cause a fault.
//
// At the next layer are the user arenas themselves. They consist of a single
// active chunk which new Go values are bump-allocated into and a list of chunks
// that were exhausted when allocating into the arena. Once the arena is freed,
// it frees all full chunks it references, and places the active one onto a reuse
// list for a future arena to use. Each arena keeps its list of referenced chunks
// explicitly live until it is freed. Each user arena also maps to an object which
// has a finalizer attached that ensures the arena's chunks are all freed even if
// the arena itself is never explicitly freed.
//
// Pointer-ful memory is bump-allocated from low addresses to high addresses in each
// chunk, while pointer-free memory is bump-allocated from high address to low
// addresses. The reason for this is to take advantage of a GC optimization wherein
// the GC will stop scanning an object when there are no more pointers in it, which
// also allows us to elide clearing the heap bitmap for pointer-free Go values
// allocated into arenas.
//
// Note that arenas are not safe to use concurrently.
//
// In summary, there are 2 resources: arenas, and arena chunks. They exist in the
// following lifecycle:
//
// (1) A new arena is created via newArena.
// (2) Chunks are allocated to hold memory allocated into the arena with new or slice.
//    (a) Chunks are first allocated from the reuse list of partially-used chunks.
//    (b) If there are no such chunks, then chunks on the ready list are taken.
//    (c) Failing all the above, memory for a new chunk is mapped.
// (3) The arena is freed, or all references to it are dropped, triggering its finalizer.
//    (a) If the GC is not active, exhausted chunks are set to fault and placed on a
//        quarantine list.
//    (b) If the GC is active, exhausted chunks are placed on a fault list and will
//        go through step (a) at a later point in time.
//    (c) Any remaining partially-used chunk is placed on a reuse list.
// (4) Once no more pointers are found into quarantined arena chunks, the sweeper
//     takes these chunks out of quarantine and places them on the ready list.

package runtime

import (
	"internal/abi"
	"internal/goarch"
	"internal/runtime/atomic"
	"runtime/internal/math"
	"runtime/internal/sys"
	"unsafe"
)

// Functions starting with arena_ are meant to be exported to downstream users
// of arenas. They should wrap these functions in a higher-lever API.
//
// The underlying arena and its resources are managed through an opaque unsafe.Pointer.

// arena_newArena is a wrapper around newUserArena.
//
//go:linkname arena_newArena arena.runtime_arena_newArena
func arena_newArena() unsafe.Pointer {
	return unsafe.Pointer(newUserArena())
}

// arena_arena_New is a wrapper around (*userArena).new, except that typ
// is an any (must be a *_type, still) and typ must be a type descriptor
// for a pointer to the type to actually be allocated, i.e. pass a *T
// to allocate a T. This is necessary because this function returns a *T.
//
//go:linkname arena_arena_New arena.runtime_arena_arena_New
func arena_arena_New(arena unsafe.Pointer, typ any) any {
	t := (*_type)(efaceOf(&typ).data)
	if t.Kind_&abi.KindMask != abi.Pointer {
		throw("arena_New: non-pointer type")
	}
	te := (*ptrtype)(unsafe.Pointer(t)).Elem
	x := ((*userArena)(arena)).new(te)
	var result any
	e := efaceOf(&result)
	e._type = t
	e.data = x
	return result
}

// arena_arena_Slice is a wrapper around (*userArena).slice.
//
//go:linkname arena_arena_Slice arena.runtime_arena_arena_Slice
func arena_arena_Slice(arena unsafe.Pointer, slice any, cap int) {
	((*userArena)(arena)).slice(slice, cap)
}

// arena_arena_Free is a wrapper around (*userArena).free.
//
//go:linkname arena_arena_Free arena.runtime_arena_arena_Free
func arena_arena_Free(arena unsafe.Pointer) {
	((*userArena)(arena)).free()
}

// arena_heapify takes a value that lives in an arena and makes a copy
// of it on the heap. Values that don't live in an arena are returned unmodified.
//
//go:linkname arena_heapify arena.runtime_arena_heapify
func arena_heapify(s any) any {
	var v unsafe.Pointer
	e := efaceOf(&s)
	t := e._type
	switch t.Kind_ & abi.KindMask {
	case abi.String:
		v = stringStructOf((*string)(e.data)).str
	case abi.Slice:
		v = (*slice)(e.data).array
	case abi.Pointer:
		v = e.data
	default:
		panic("arena: Clone only supports pointers, slices, and strings")
	}
	span := spanOf(uintptr(v))
	if span == nil || !span.isUserArenaChunk {
		// Not stored in a user arena chunk.
		return s
	}
	// Heap-allocate storage for a copy.
	var x any
	switch t.Kind_ & abi.KindMask {
	case abi.String:
		s1 := s.(string)
		s2, b := rawstring(len(s1))
		copy(b, s1)
		x = s2
	case abi.Slice:
		len := (*slice)(e.data).len
		et := (*slicetype)(unsafe.Pointer(t)).Elem
		sl := new(slice)
		*sl = slice{makeslicecopy(et, len, len, (*slice)(e.data).array), len, len}
		xe := efaceOf(&x)
		xe._type = t
		xe.data = unsafe.Pointer(sl)
	case abi.Pointer:
		et := (*ptrtype)(unsafe.Pointer(t)).Elem
		e2 := newobject(et)
		typedmemmove(et, e2, e.data)
		xe := efaceOf(&x)
		xe._type = t
		xe.data = e2
	}
	return x
}

const (
	// userArenaChunkBytes is the size of a user arena chunk.
	userArenaChunkBytesMax = 8 << 20
	userArenaChunkBytes    = uintptr(int64(userArenaChunkBytesMax-heapArenaBytes)&(int64(userArenaChunkBytesMax-heapArenaBytes)>>63) + heapArenaBytes) // min(userArenaChunkBytesMax, heapArenaBytes)

	// userArenaChunkPages is the number of pages a user arena chunk uses.
	userArenaChunkPages = userArenaChunkBytes / pageSize

	// userArenaChunkMaxAllocBytes is the maximum size of an object that can
	// be allocated from an arena. This number is chosen to cap worst-case
	// fragmentation of user arenas to 25%. Larger allocations are redirected
	// to the heap.
	userArenaChunkMaxAllocBytes = userArenaChunkBytes / 4
)

func init() {
	if userArenaChunkPages*pageSize != userArenaChunkBytes {
		throw("user arena chunk size is not a multiple of the page size")
	}
	if userArenaChunkBytes%physPageSize != 0 {
		throw("user arena chunk size is not a multiple of the physical page size")
	}
	if userArenaChunkBytes < heapArenaBytes {
		if heapArenaBytes%userArenaChunkBytes != 0 {
			throw("user arena chunk size is smaller than a heap arena, but doesn't divide it")
		}
	} else {
		if userArenaChunkBytes%heapArenaBytes != 0 {
			throw("user arena chunks size is larger than a heap arena, but not a multiple")
		}
	}
	lockInit(&userArenaState.lock, lockRankUserArenaState)
}

// userArenaChunkReserveBytes returns the amount of additional bytes to reserve for
// heap metadata.
func userArenaChunkReserveBytes() uintptr {
	// In the allocation headers experiment, we reserve the end of the chunk for
	// a pointer/scalar bitmap. We also reserve space for a dummy _type that
	// refers to the bitmap. The PtrBytes field of the dummy _type indicates how
	// many of those bits are valid.
	return userArenaChunkBytes/goarch.PtrSize/8 + unsafe.Sizeof(_type{})
}

type userArena struct {
	// fullList is a list of full chunks that have not enough free memory left, and
	// that we'll free once this user arena is freed.
	//
	// Can't use mSpanList here because it's not-in-heap.
	fullList *mspan

	// active is the user arena chunk we're currently allocating into.
	active *mspan

	// refs is a set of references to the arena chunks so that they're kept alive.
	//
	// The last reference in the list always refers to active, while the rest of
	// them correspond to fullList. Specifically, the head of fullList is the
	// second-to-last one, fullList.next is the third-to-last, and so on.
	//
	// In other words, every time a new chunk becomes active, its appended to this
	// list.
	refs []unsafe.Pointer

	// defunct is true if free has been called on this arena.
	//
	// This is just a best-effort way to discover a concurrent allocation
	// and free. Also used to detect a double-free.
	defunct atomic.Bool
}

// newUserArena creates a new userArena ready to be used.
func newUserArena() *userArena {
	a := new(userArena)
	SetFinalizer(a, func(a *userArena) {
		// If arena handle is dropped without being freed, then call
		// free on the arena, so the arena chunks are never reclaimed
		// by the garbage collector.
		a.free()
	})
	a.refill()
	return a
}

// new allocates a new object of the provided type into the arena, and returns
// its pointer.
//
// This operation is not safe to call concurrently with other operations on the
// same arena.
func (a *userArena) new(typ *_type) unsafe.Pointer {
	return a.alloc(typ, -1)
}

// slice allocates a new slice backing store. slice must be a pointer to a slice
// (i.e. *[]T), because userArenaSlice will update the slice directly.
//
// cap determines the capacity of the slice backing store and must be non-negative.
//
// This operation is not safe to call concurrently with other operations on the
// same arena.
func (a *userArena) slice(sl any, cap int) {
	if cap < 0 {
		panic("userArena.slice: negative cap")
	}
	i := efaceOf(&sl)
	typ := i._type
	if typ.Kind_&abi.KindMask != abi.Pointer {
		panic("slice result of non-ptr type")
	}
	typ = (*ptrtype)(unsafe.Pointer(typ)).Elem
	if typ.Kind_&abi.KindMask != abi.Slice {
		panic("slice of non-ptr-to-slice type")
	}
	typ = (*slicetype)(unsafe.Pointer(typ)).Elem
	// t is now the element type of the slice we want to allocate.

	*((*slice)(i.data)) = slice{a.alloc(typ, cap), cap, cap}
}

// free returns the userArena's chunks back to mheap and marks it as defunct.
//
// Must be called at most once for any given arena.
//
// This operation is not safe to call concurrently with other operations on the
// same arena.
func (a *userArena) free() {
	// Check for a double-free.
	if a.defunct.Load() {
		panic("arena double free")
	}

	// Mark ourselves as defunct.
	a.defunct.Store(true)
	SetFinalizer(a, nil)

	// Free all the full arenas.
	//
	// The refs on this list are in reverse order from the second-to-last.
	s := a.fullList
	i := len(a.refs) - 2
	for s != nil {
		a.fullList = s.next
		s.next = nil
		freeUserArenaChunk(s, a.refs[i])
		s = a.fullList
		i--
	}
	if a.fullList != nil || i >= 0 {
		// There's still something left on the full list, or we
		// failed to actually iterate over the entire refs list.
		throw("full list doesn't match refs list in length")
	}

	// Put the active chunk onto the reuse list.
	//
	// Note that active's reference is always the last reference in refs.
	s = a.active
	if s != nil {
		if raceenabled || msanenabled || asanenabled {
			// Don't reuse arenas with sanitizers enabled. We want to catch
			// any use-after-free errors aggressively.
			freeUserArenaChunk(s, a.refs[len(a.refs)-1])
		} else {
			lock(&userArenaState.lock)
			userArenaState.reuse = append(userArenaState.reuse, liveUserArenaChunk{s, a.refs[len(a.refs)-1]})
			unlock(&userArenaState.lock)
		}
	}
	// nil out a.active so that a race with freeing will more likely cause a crash.
	a.active = nil
	a.refs = nil
}

// alloc reserves space in the current chunk or calls refill and reserves space
// in a new chunk. If cap is negative, the type will be taken literally, otherwise
// it will be considered as an element type for a slice backing store with capacity
// cap.
func (a *userArena) alloc(typ *_type, cap int) unsafe.Pointer {
	s := a.active
	var x unsafe.Pointer
	for {
		x = s.userArenaNextFree(typ, cap)
		if x != nil {
			break
		}
		s = a.refill()
	}
	return x
}

// refill inserts the current arena chunk onto the full list and obtains a new
// one, either from the partial list or allocating a new one, both from mheap.
func (a *userArena) refill() *mspan {
	// If there's an active chunk, assume it's full.
	s := a.active
	if s != nil {
		if s.userArenaChunkFree.size() > userArenaChunkMaxAllocBytes {
			// It's difficult to tell when we're actually out of memory
			// in a chunk because the allocation that failed may still leave
			// some free space available. However, that amount of free space
			// should never exceed the maximum allocation size.
			throw("wasted too much memory in an arena chunk")
		}
		s.next = a.fullList
		a.fullList = s
		a.active = nil
		s = nil
	}
	var x unsafe.Pointer

	// Check the partially-used list.
	lock(&userArenaState.lock)
	if len(userArenaState.reuse) > 0 {
		// Pick off the last arena chunk from the list.
		n := len(userArenaState.reuse) - 1
		x = userArenaState.reuse[n].x
		s = userArenaState.reuse[n].mspan
		userArenaState.reuse[n].x = nil
		userArenaState.reuse[n].mspan = nil
		userArenaState.reuse = userArenaState.reuse[:n]
	}
	unlock(&userArenaState.lock)
	if s == nil {
		// Allocate a new one.
		x, s = newUserArenaChunk()
		if s == nil {
			throw("out of memory")
		}
	}
	a.refs = append(a.refs, x)
	a.active = s
	return s
}

type liveUserArenaChunk struct {
	*mspan // Must represent a user arena chunk.

	// Reference to mspan.base() to keep the chunk alive.
	x unsafe.Pointer
}

var userArenaState struct {
	lock mutex

	// reuse contains a list of partially-used and already-live
	// user arena chunks that can be quickly reused for another
	// arena.
	//
	// Protected by lock.
	reuse []liveUserArenaChunk

	// fault contains full user arena chunks that need to be faulted.
	//
	// Protected by lock.
	fault []liveUserArenaChunk
}

// userArenaNextFree reserves space in the user arena for an item of the specified
// type. If cap is not -1, this is for an array of cap elements of type t.
func (s *mspan) userArenaNextFree(typ *_type, cap int) unsafe.Pointer {
	size := typ.Size_
	if cap > 0 {
		if size > ^uintptr(0)/uintptr(cap) {
			// Overflow.
			throw("out of memory")
		}
		size *= uintptr(cap)
	}
	if size == 0 || cap == 0 {
		return unsafe.Pointer(&zerobase)
	}
	if size > userArenaChunkMaxAllocBytes {
		// Redirect allocations that don't fit into a chunk well directly
		// from the heap.
		if cap >= 0 {
			return newarray(typ, cap)
		}
		return newobject(typ)
	}

	// Prevent preemption as we set up the space for a new object.
	//
	// Act like we're allocating.
	mp := acquirem()
	if mp.mallocing != 0 {
		throw("malloc deadlock")
	}
	if mp.gsignal == getg() {
		throw("malloc during signal")
	}
	mp.mallocing = 1

	var ptr unsafe.Pointer
	if !typ.Pointers() {
		// Allocate pointer-less objects from the tail end of the chunk.
		v, ok := s.userArenaChunkFree.takeFromBack(size, typ.Align_)
		if ok {
			ptr = unsafe.Pointer(v)
		}
	} else {
		v, ok := s.userArenaChunkFree.takeFromFront(size, typ.Align_)
		if ok {
			ptr = unsafe.Pointer(v)
		}
	}
	if ptr == nil {
		// Failed to allocate.
		mp.mallocing = 0
		releasem(mp)
		return nil
	}
	if s.needzero != 0 {
		throw("arena chunk needs zeroing, but should already be zeroed")
	}
	// Set up heap bitmap and do extra accounting.
	if typ.Pointers() {
		if cap >= 0 {
			userArenaHeapBitsSetSliceType(typ, cap, ptr, s)
		} else {
			userArenaHeapBitsSetType(typ, ptr, s)
		}
		c := getMCache(mp)
		if c == nil {
			throw("mallocgc called without a P or outside bootstrapping")
		}
		if cap > 0 {
			c.scanAlloc += size - (typ.Size_ - typ.PtrBytes)
		} else {
			c.scanAlloc += typ.PtrBytes
		}
	}

	// Ensure that the stores above that initialize x to
	// type-safe memory and set the heap bits occur before
	// the caller can make ptr observable to the garbage
	// collector. Otherwise, on weakly ordered machines,
	// the garbage collector could follow a pointer to x,
	// but see uninitialized memory or stale heap bits.
	publicationBarrier()

	mp.mallocing = 0
	releasem(mp)

	return ptr
}

// userArenaHeapBitsSetSliceType is the equivalent of heapBitsSetType but for
// Go slice backing store values allocated in a user arena chunk. It sets up the
// heap bitmap for n consecutive values with type typ allocated at address ptr.
func userArenaHeapBitsSetSliceType(typ *_type, n int, ptr unsafe.Pointer, s *mspan) {
	mem, overflow := math.MulUintptr(typ.Size_, uintptr(n))
	if overflow || n < 0 || mem > maxAlloc {
		panic(plainError("runtime: allocation size out of range"))
	}
	for i := 0; i < n; i++ {
		userArenaHeapBitsSetType(typ, add(ptr, uintptr(i)*typ.Size_), s)
	}
}

// userArenaHeapBitsSetType is the equivalent of heapSetType but for
// non-slice-backing-store Go values allocated in a user arena chunk. It
// sets up the type metadata for the value with type typ allocated at address ptr.
// base is the base address of the arena chunk.
func userArenaHeapBitsSetType(typ *_type, ptr unsafe.Pointer, s *mspan) {
	base := s.base()
	h := s.writeUserArenaHeapBits(uintptr(ptr))

	p := typ.GCData // start of 1-bit pointer mask (or GC program)
	var gcProgBits uintptr
	if typ.Kind_&abi.KindGCProg != 0 {
		// Expand gc program, using the object itself for storage.
		gcProgBits = runGCProg(addb(p, 4), (*byte)(ptr))
		p = (*byte)(ptr)
	}
	nb := typ.PtrBytes / goarch.PtrSize

	for i := uintptr(0); i < nb; i += ptrBits {
		k := nb - i
		if k > ptrBits {
			k = ptrBits
		}
		// N.B. On big endian platforms we byte swap the data that we
		// read from GCData, which is always stored in little-endian order
		// by the compiler. writeUserArenaHeapBits handles data in
		// a platform-ordered way for efficiency, but stores back the
		// data in little endian order, since we expose the bitmap through
		// a dummy type.
		h = h.write(s, readUintptr(addb(p, i/8)), k)
	}
	// Note: we call pad here to ensure we emit explicit 0 bits
	// for the pointerless tail of the object. This ensures that
	// there's only a single noMorePtrs mark for the next object
	// to clear. We don't need to do this to clear stale noMorePtrs
	// markers from previous uses because arena chunk pointer bitmaps
	// are always fully cleared when reused.
	h = h.pad(s, typ.Size_-typ.PtrBytes)
	h.flush(s, uintptr(ptr), typ.Size_)

	if typ.Kind_&abi.KindGCProg != 0 {
		// Zero out temporary ptrmask buffer inside object.
		memclrNoHeapPointers(ptr, (gcProgBits+7)/8)
	}

	// Update the PtrBytes value in the type information. After this
	// point, the GC will observe the new bitmap.
	s.largeType.PtrBytes = uintptr(ptr) - base + typ.PtrBytes

	// Double-check that the bitmap was written out correctly.
	const doubleCheck = false
	if doubleCheck {
		doubleCheckHeapPointersInterior(uintptr(ptr), uintptr(ptr), typ.Size_, typ.Size_, typ, &s.largeType, s)
	}
}

type writeUserArenaHeapBits struct {
	offset uintptr // offset in span that the low bit of mask represents the pointer state of.
	mask   uintptr // some pointer bits starting at the address addr.
	valid  uintptr // number of bits in buf that are valid (including low)
	low    uintptr // number of low-order bits to not overwrite
}

func (s *mspan) writeUserArenaHeapBits(addr uintptr) (h writeUserArenaHeapBits) {
	offset := addr - s.base()

	// We start writing bits maybe in the middle of a heap bitmap word.
	// Remember how many bits into the word we started, so we can be sure
	// not to overwrite the previous bits.
	h.low = offset / goarch.PtrSize % ptrBits

	// round down to heap word that starts the bitmap word.
	h.offset = offset - h.low*goarch.PtrSize

	// We don't have any bits yet.
	h.mask = 0
	h.valid = h.low

	return
}

// write appends the pointerness of the next valid pointer slots
// using the low valid bits of bits. 1=pointer, 0=scalar.
func (h writeUserArenaHeapBits) write(s *mspan, bits, valid uintptr) writeUserArenaHeapBits {
	if h.valid+valid <= ptrBits {
		// Fast path - just accumulate the bits.
		h.mask |= bits << h.valid
		h.valid += valid
		return h
	}
	// Too many bits to fit in this word. Write the current word
	// out and move on to the next word.

	data := h.mask | bits<<h.valid       // mask for this word
	h.mask = bits >> (ptrBits - h.valid) // leftover for next word
	h.valid += valid - ptrBits           // have h.valid+valid bits, writing ptrBits of them

	// Flush mask to the memory bitmap.
	idx := h.offset / (ptrBits * goarch.PtrSize)
	m := uintptr(1)<<h.low - 1
	bitmap := s.heapBits()
	bitmap[idx] = bswapIfBigEndian(bswapIfBigEndian(bitmap[idx])&m | data)
	// Note: no synchronization required for this write because
	// the allocator has exclusive access to the page, and the bitmap
	// entries are all for a single page. Also, visibility of these
	// writes is guaranteed by the publication barrier in mallocgc.

	// Move to next word of bitmap.
	h.offset += ptrBits * goarch.PtrSize
	h.low = 0
	return h
}

// Add padding of size bytes.
func (h writeUserArenaHeapBits) pad(s *mspan, size uintptr) writeUserArenaHeapBits {
	if size == 0 {
		return h
	}
	words := size / goarch.PtrSize
	for words > ptrBits {
		h = h.write(s, 0, ptrBits)
		words -= ptrBits
	}
	return h.write(s, 0, words)
}

// Flush the bits that have been written, and add zeros as needed
// to cover the full object [addr, addr+size).
func (h writeUserArenaHeapBits) flush(s *mspan, addr, size uintptr) {
	offset := addr - s.base()

	// zeros counts the number of bits needed to represent the object minus the
	// number of bits we've already written. This is the number of 0 bits
	// that need to be added.
	zeros := (offset+size-h.offset)/goarch.PtrSize - h.valid

	// Add zero bits up to the bitmap word boundary
	if zeros > 0 {
		z := ptrBits - h.valid
		if z > zeros {
			z = zeros
		}
		h.valid += z
		zeros -= z
	}

	// Find word in bitmap that we're going to write.
	bitmap := s.heapBits()
	idx := h.offset / (ptrBits * goarch.PtrSize)

	// Write remaining bits.
	if h.valid != h.low {
		m := uintptr(1)<<h.low - 1      // don't clear existing bits below "low"
		m |= ^(uintptr(1)<<h.valid - 1) // don't clear existing bits above "valid"
		bitmap[idx] = bswapIfBigEndian(bswapIfBigEndian(bitmap[idx])&m | h.mask)
	}
	if zeros == 0 {
		return
	}

	// Advance to next bitmap word.
	h.offset += ptrBits * goarch.PtrSize

	// Continue on writing zeros for the rest of the object.
	// For standard use of the ptr bits this is not required, as
	// the bits are read from the beginning of the object. Some uses,
	// like noscan spans, oblets, bulk write barriers, and cgocheck, might
	// start mid-object, so these writes are still required.
	for {
		// Write zero bits.
		idx := h.offset / (ptrBits * goarch.PtrSize)
		if zeros < ptrBits {
			bitmap[idx] = bswapIfBigEndian(bswapIfBigEndian(bitmap[idx]) &^ (uintptr(1)<<zeros - 1))
			break
		} else if zeros == ptrBits {
			bitmap[idx] = 0
			break
		} else {
			bitmap[idx] = 0
			zeros -= ptrBits
		}
		h.offset += ptrBits * goarch.PtrSize
	}
}

// bswapIfBigEndian swaps the byte order of the uintptr on goarch.BigEndian platforms,
// and leaves it alone elsewhere.
func bswapIfBigEndian(x uintptr) uintptr {
	if goarch.BigEndian {
		if goarch.PtrSize == 8 {
			return uintptr(sys.Bswap64(uint64(x)))
		}
		return uintptr(sys.Bswap32(uint32(x)))
	}
	return x
}

// newUserArenaChunk allocates a user arena chunk, which maps to a single
// heap arena and single span. Returns a pointer to the base of the chunk
// (this is really important: we need to keep the chunk alive) and the span.
func newUserArenaChunk() (unsafe.Pointer, *mspan) {
	if gcphase == _GCmarktermination {
		throw("newUserArenaChunk called with gcphase == _GCmarktermination")
	}

	// Deduct assist credit. Because user arena chunks are modeled as one
	// giant heap object which counts toward heapLive, we're obligated to
	// assist the GC proportionally (and it's worth noting that the arena
	// does represent additional work for the GC, but we also have no idea
	// what that looks like until we actually allocate things into the
	// arena).
	deductAssistCredit(userArenaChunkBytes)

	// Set mp.mallocing to keep from being preempted by GC.
	mp := acquirem()
	if mp.mallocing != 0 {
		throw("malloc deadlock")
	}
	if mp.gsignal == getg() {
		throw("malloc during signal")
	}
	mp.mallocing = 1

	// Allocate a new user arena.
	var span *mspan
	systemstack(func() {
		span = mheap_.allocUserArenaChunk()
	})
	if span == nil {
		throw("out of memory")
	}
	x := unsafe.Pointer(span.base())

	// Allocate black during GC.
	// All slots hold nil so no scanning is needed.
	// This may be racing with GC so do it atomically if there can be
	// a race marking the bit.
	if gcphase != _GCoff {
		gcmarknewobject(span, span.base())
	}

	if raceenabled {
		// TODO(mknyszek): Track individual objects.
		racemalloc(unsafe.Pointer(span.base()), span.elemsize)
	}

	if msanenabled {
		// TODO(mknyszek): Track individual objects.
		msanmalloc(unsafe.Pointer(span.base()), span.elemsize)
	}

	if asanenabled {
		// TODO(mknyszek): Track individual objects.
		rzSize := computeRZlog(span.elemsize)
		span.elemsize -= rzSize
		span.largeType.Size_ = span.elemsize
		rzStart := span.base() + span.elemsize
		span.userArenaChunkFree = makeAddrRange(span.base(), rzStart)
		asanpoison(unsafe.Pointer(rzStart), span.limit-rzStart)
		asanunpoison(unsafe.Pointer(span.base()), span.elemsize)
	}

	if rate := MemProfileRate; rate > 0 {
		c := getMCache(mp)
		if c == nil {
			throw("newUserArenaChunk called without a P or outside bootstrapping")
		}
		// Note cache c only valid while m acquired; see #47302
		if rate != 1 && userArenaChunkBytes < c.nextSample {
			c.nextSample -= userArenaChunkBytes
		} else {
			profilealloc(mp, unsafe.Pointer(span.base()), userArenaChunkBytes)
		}
	}
	mp.mallocing = 0
	releasem(mp)

	// Again, because this chunk counts toward heapLive, potentially trigger a GC.
	if t := (gcTrigger{kind: gcTriggerHeap}); t.test() {
		gcStart(t)
	}

	if debug.malloc {
		if inittrace.active && inittrace.id == getg().goid {
			// Init functions are executed sequentially in a single goroutine.
			inittrace.bytes += uint64(userArenaChunkBytes)
		}
	}

	// Double-check it's aligned to the physical page size. Based on the current
	// implementation this is trivially true, but it need not be in the future.
	// However, if it's not aligned to the physical page size then we can't properly
	// set it to fault later.
	if uintptr(x)%physPageSize != 0 {
		throw("user arena chunk is not aligned to the physical page size")
	}

	return x, span
}

// isUnusedUserArenaChunk indicates that the arena chunk has been set to fault
// and doesn't contain any scannable memory anymore. However, it might still be
// mSpanInUse as it sits on the quarantine list, since it needs to be swept.
//
// This is not safe to execute unless the caller has ownership of the mspan or
// the world is stopped (preemption is prevented while the relevant state changes).
//
// This is really only meant to be used by accounting tests in the runtime to
// distinguish when a span shouldn't be counted (since mSpanInUse might not be
// enough).
func (s *mspan) isUnusedUserArenaChunk() bool {
	return s.isUserArenaChunk && s.spanclass == makeSpanClass(0, true)
}

// setUserArenaChunkToFault sets the address space for the user arena chunk to fault
// and releases any underlying memory resources.
//
// Must be in a non-preemptible state to ensure the consistency of statistics
// exported to MemStats.
func (s *mspan) setUserArenaChunkToFault() {
	if !s.isUserArenaChunk {
		throw("invalid span in heapArena for user arena")
	}
	if s.npages*pageSize != userArenaChunkBytes {
		throw("span on userArena.faultList has invalid size")
	}

	// Update the span class to be noscan. What we want to happen is that
	// any pointer into the span keeps it from getting recycled, so we want
	// the mark bit to get set, but we're about to set the address space to fault,
	// so we have to prevent the GC from scanning this memory.
	//
	// It's OK to set it here because (1) a GC isn't in progress, so the scanning code
	// won't make a bad decision, (2) we're currently non-preemptible and in the runtime,
	// so a GC is blocked from starting. We might race with sweeping, which could
	// put it on the "wrong" sweep list, but really don't care because the chunk is
	// treated as a large object span and there's no meaningful difference between scan
	// and noscan large objects in the sweeper. The STW at the start of the GC acts as a
	// barrier for this update.
	s.spanclass = makeSpanClass(0, true)

	// Actually set the arena chunk to fault, so we'll get dangling pointer errors.
	// sysFault currently uses a method on each OS that forces it to evacuate all
	// memory backing the chunk.
	sysFault(unsafe.Pointer(s.base()), s.npages*pageSize)

	// Everything on the list is counted as in-use, however sysFault transitions to
	// Reserved, not Prepared, so we skip updating heapFree or heapReleased and just
	// remove the memory from the total altogether; it's just address space now.
	gcController.heapInUse.add(-int64(s.npages * pageSize))

	// Count this as a free of an object right now as opposed to when
	// the span gets off the quarantine list. The main reason is so that the
	// amount of bytes allocated doesn't exceed how much is counted as
	// "mapped ready," which could cause a deadlock in the pacer.
	gcController.totalFree.Add(int64(s.elemsize))

	// Update consistent stats to match.
	//
	// We're non-preemptible, so it's safe to update consistent stats (our P
	// won't change out from under us).
	stats := memstats.heapStats.acquire()
	atomic.Xaddint64(&stats.committed, -int64(s.npages*pageSize))
	atomic.Xaddint64(&stats.inHeap, -int64(s.npages*pageSize))
	atomic.Xadd64(&stats.largeFreeCount, 1)
	atomic.Xadd64(&stats.largeFree, int64(s.elemsize))
	memstats.heapStats.release()

	// This counts as a free, so update heapLive.
	gcController.update(-int64(s.elemsize), 0)

	// Mark it as free for the race detector.
	if raceenabled {
		racefree(unsafe.Pointer(s.base()), s.elemsize)
	}

	systemstack(func() {
		// Add the user arena to the quarantine list.
		lock(&mheap_.lock)
		mheap_.userArena.quarantineList.insert(s)
		unlock(&mheap_.lock)
	})
}

// inUserArenaChunk returns true if p points to a user arena chunk.
func inUserArenaChunk(p uintptr) bool {
	s := spanOf(p)
	if s == nil {
		return false
	}
	return s.isUserArenaChunk
}

// freeUserArenaChunk releases the user arena represented by s back to the runtime.
//
// x must be a live pointer within s.
//
// The runtime will set the user arena to fault once it's safe (the GC is no longer running)
// and then once the user arena is no longer referenced by the application, will allow it to
// be reused.
func freeUserArenaChunk(s *mspan, x unsafe.Pointer) {
	if !s.isUserArenaChunk {
		throw("span is not for a user arena")
	}
	if s.npages*pageSize != userArenaChunkBytes {
		throw("invalid user arena span size")
	}

	// Mark the region as free to various sanitizers immediately instead
	// of handling them at sweep time.
	if raceenabled {
		racefree(unsafe.Pointer(s.base()), s.elemsize)
	}
	if msanenabled {
		msanfree(unsafe.Pointer(s.base()), s.elemsize)
	}
	if asanenabled {
		asanpoison(unsafe.Pointer(s.base()), s.elemsize)
	}

	// Make ourselves non-preemptible as we manipulate state and statistics.
	//
	// Also required by setUserArenaChunksToFault.
	mp := acquirem()

	// We can only set user arenas to fault if we're in the _GCoff phase.
	if gcphase == _GCoff {
		lock(&userArenaState.lock)
		faultList := userArenaState.fault
		userArenaState.fault = nil
		unlock(&userArenaState.lock)

		s.setUserArenaChunkToFault()
		for _, lc := range faultList {
			lc.mspan.setUserArenaChunkToFault()
		}

		// Until the chunks are set to fault, keep them alive via the fault list.
		KeepAlive(x)
		KeepAlive(faultList)
	} else {
		// Put the user arena on the fault list.
		lock(&userArenaState.lock)
		userArenaState.fault = append(userArenaState.fault, liveUserArenaChunk{s, x})
		unlock(&userArenaState.lock)
	}
	releasem(mp)
}

// allocUserArenaChunk attempts to reuse a free user arena chunk represented
// as a span.
//
// Must be in a non-preemptible state to ensure the consistency of statistics
// exported to MemStats.
//
// Acquires the heap lock. Must run on the system stack for that reason.
//
//go:systemstack
func (h *mheap) allocUserArenaChunk() *mspan {
	var s *mspan
	var base uintptr

	// First check the free list.
	lock(&h.lock)
	if !h.userArena.readyList.isEmpty() {
		s = h.userArena.readyList.first
		h.userArena.readyList.remove(s)
		base = s.base()
	} else {
		// Free list was empty, so allocate a new arena.
		hintList := &h.userArena.arenaHints
		if raceenabled {
			// In race mode just use the regular heap hints. We might fragment
			// the address space, but the race detector requires that the heap
			// is mapped contiguously.
			hintList = &h.arenaHints
		}
		v, size := h.sysAlloc(userArenaChunkBytes, hintList, false)
		if size%userArenaChunkBytes != 0 {
			throw("sysAlloc size is not divisible by userArenaChunkBytes")
		}
		if size > userArenaChunkBytes {
			// We got more than we asked for. This can happen if
			// heapArenaSize > userArenaChunkSize, or if sysAlloc just returns
			// some extra as a result of trying to find an aligned region.
			//
			// Divide it up and put it on the ready list.
			for i := userArenaChunkBytes; i < size; i += userArenaChunkBytes {
				s := h.allocMSpanLocked()
				s.init(uintptr(v)+i, userArenaChunkPages)
				h.userArena.readyList.insertBack(s)
			}
			size = userArenaChunkBytes
		}
		base = uintptr(v)
		if base == 0 {
			// Out of memory.
			unlock(&h.lock)
			return nil
		}
		s = h.allocMSpanLocked()
	}
	unlock(&h.lock)

	// sysAlloc returns Reserved address space, and any span we're
	// reusing is set to fault (so, also Reserved), so transition
	// it to Prepared and then Ready.
	//
	// Unlike (*mheap).grow, just map in everything that we
	// asked for. We're likely going to use it all.
	sysMap(unsafe.Pointer(base), userArenaChunkBytes, &gcController.heapReleased)
	sysUsed(unsafe.Pointer(base), userArenaChunkBytes, userArenaChunkBytes)

	// Model the user arena as a heap span for a large object.
	spc := makeSpanClass(0, false)
	h.initSpan(s, spanAllocHeap, spc, base, userArenaChunkPages)
	s.isUserArenaChunk = true
	s.elemsize -= userArenaChunkReserveBytes()
	s.limit = s.base() + s.elemsize
	s.freeindex = 1
	s.allocCount = 1

	// Account for this new arena chunk memory.
	gcController.heapInUse.add(int64(userArenaChunkBytes))
	gcController.heapReleased.add(-int64(userArenaChunkBytes))

	stats := memstats.heapStats.acquire()
	atomic.Xaddint64(&stats.inHeap, int64(userArenaChunkBytes))
	atomic.Xaddint64(&stats.committed, int64(userArenaChunkBytes))

	// Model the arena as a single large malloc.
	atomic.Xadd64(&stats.largeAlloc, int64(s.elemsize))
	atomic.Xadd64(&stats.largeAllocCount, 1)
	memstats.heapStats.release()

	// Count the alloc in inconsistent, internal stats.
	gcController.totalAlloc.Add(int64(s.elemsize))

	// Update heapLive.
	gcController.update(int64(s.elemsize), 0)

	// This must clear the entire heap bitmap so that it's safe
	// to allocate noscan data without writing anything out.
	s.initHeapBits(true)

	// Clear the span preemptively. It's an arena chunk, so let's assume
	// everything is going to be used.
	//
	// This also seems to make a massive difference as to whether or
	// not Linux decides to back this memory with transparent huge
	// pages. There's latency involved in this zeroing, but the hugepage
	// gains are almost always worth it. Note: it's important that we
	// clear even if it's freshly mapped and we know there's no point
	// to zeroing as *that* is the critical signal to use huge pages.
	memclrNoHeapPointers(unsafe.Pointer(s.base()), s.elemsize)
	s.needzero = 0

	s.freeIndexForScan = 1

	// Set up the range for allocation.
	s.userArenaChunkFree = makeAddrRange(base, base+s.elemsize)

	// Put the large span in the mcentral swept list so that it's
	// visible to the background sweeper.
	h.central[spc].mcentral.fullSwept(h.sweepgen).push(s)

	// Set up an allocation header. Avoid write barriers here because this type
	// is not a real type, and it exists in an invalid location.
	*(*uintptr)(unsafe.Pointer(&s.largeType)) = uintptr(unsafe.Pointer(s.limit))
	*(*uintptr)(unsafe.Pointer(&s.largeType.GCData)) = s.limit + unsafe.Sizeof(_type{})
	s.largeType.PtrBytes = 0
	s.largeType.Size_ = s.elemsize

	return s
}

The pages are generated with Golds v0.7.0-preview. (GOOS=linux GOARCH=amd64)
Golds is a Go 101 project developed by Tapir Liu.
PR and bug reports are welcome and can be submitted to the issue list.
Please follow @zigo_101 (reachable from the left QR code) to get the latest news of Golds.