// 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.

// Garbage collector: marking and scanning

package runtime

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

const (
	fixedRootFinalizers = iota
	fixedRootFreeGStacks
	fixedRootCleanups
	fixedRootCount

	// rootBlockBytes is the number of bytes to scan per data or
	// BSS root.
	rootBlockBytes = 256 << 10

	// maxObletBytes is the maximum bytes of an object to scan at
	// once. Larger objects will be split up into "oblets" of at
	// most this size. Since we can scan 1–2 MB/ms, 128 KB bounds
	// scan preemption at ~100 µs.
	//
	// This must be > _MaxSmallSize so that the object base is the
	// span base.
	maxObletBytes = 128 << 10

	// drainCheckThreshold specifies how many units of work to do
	// between self-preemption checks in gcDrain. Assuming a scan
	// rate of 1 MB/ms, this is ~100 µs. Lower values have higher
	// overhead in the scan loop (the scheduler check may perform
	// a syscall, so its overhead is nontrivial). Higher values
	// make the system less responsive to incoming work.
	drainCheckThreshold = 100000

	// pagesPerSpanRoot indicates how many pages to scan from a span root
	// at a time. Used by special root marking.
	//
	// Higher values improve throughput by increasing locality, but
	// increase the minimum latency of a marking operation.
	//
	// Must be a multiple of the pageInUse bitmap element size and
	// must also evenly divide pagesPerArena.
	pagesPerSpanRoot = min(512, pagesPerArena)
)

// internalBlocked returns true if the goroutine is blocked due to an
// internal (non-leaking) waitReason, e.g. waiting for the netpoller or garbage collector.
// Such goroutines are never leak detection candidates according to the GC.
//
//go:nosplit
func (gp *g) internalBlocked() bool {
	reason := gp.waitreason
	return reason < waitReasonChanReceiveNilChan || waitReasonSyncWaitGroupWait < reason
}

// allGsSnapshotSortedForGC takes a snapshot of allgs and returns a sorted
// array of Gs. The array is sorted by the G's status, with running Gs
// first, followed by blocked Gs. The returned index indicates the cutoff
// between runnable and blocked Gs.
//
// The world must be stopped or allglock must be held.
func allGsSnapshotSortedForGC() ([]*g, int) {
	assertWorldStoppedOrLockHeld(&allglock)

	// Reset the status of leaked goroutines in order to improve
	// the precision of goroutine leak detection.
	for _, gp := range allgs {
		gp.atomicstatus.CompareAndSwap(_Gleaked, _Gwaiting)
	}

	allgsSorted := make([]*g, len(allgs))

	// Indices cutting off runnable and blocked Gs.
	var currIndex, blockedIndex = 0, len(allgsSorted) - 1
	for _, gp := range allgs {
		// not sure if we need atomic load because we are stopping the world,
		// but do it just to be safe for now
		if status := readgstatus(gp); status != _Gwaiting || gp.internalBlocked() {
			allgsSorted[currIndex] = gp
			currIndex++
		} else {
			allgsSorted[blockedIndex] = gp
			blockedIndex--
		}
	}

	// Because the world is stopped or allglock is held, allgadd
	// cannot happen concurrently with this. allgs grows
	// monotonically and existing entries never change, so we can
	// simply return a copy of the slice header. For added safety,
	// we trim everything past len because that can still change.
	return allgsSorted, blockedIndex + 1
}

// gcPrepareMarkRoots queues root scanning jobs (stacks, globals, and
// some miscellany) and initializes scanning-related state.
//
// The world must be stopped.
func gcPrepareMarkRoots() {
	assertWorldStopped()

	// Compute how many data and BSS root blocks there are.
	nBlocks := func(bytes uintptr) int {
		return int(divRoundUp(bytes, rootBlockBytes))
	}

	work.nDataRoots = 0
	work.nBSSRoots = 0

	// Scan globals.
	for _, datap := range activeModules() {
		nDataRoots := nBlocks(datap.edata - datap.data)
		if nDataRoots > work.nDataRoots {
			work.nDataRoots = nDataRoots
		}

		nBSSRoots := nBlocks(datap.ebss - datap.bss)
		if nBSSRoots > work.nBSSRoots {
			work.nBSSRoots = nBSSRoots
		}
	}

	// Scan span roots for finalizer specials.
	//
	// We depend on addfinalizer to mark objects that get
	// finalizers after root marking.
	//
	// We're going to scan the whole heap (that was available at the time the
	// mark phase started, i.e. markArenas) for in-use spans which have specials.
	//
	// Break up the work into arenas, and further into chunks.
	//
	// Snapshot heapArenas as markArenas. This snapshot is safe because heapArenas
	// is append-only.
	mheap_.markArenas = mheap_.heapArenas[:len(mheap_.heapArenas):len(mheap_.heapArenas)]
	work.nSpanRoots = len(mheap_.markArenas) * (pagesPerArena / pagesPerSpanRoot)

	// Scan stacks.
	//
	// Gs may be created after this point, but it's okay that we
	// ignore them because they begin life without any roots, so
	// there's nothing to scan, and any roots they create during
	// the concurrent phase will be caught by the write barrier.
	if work.goroutineLeak.enabled {
		// goroutine leak finder GC --- only prepare runnable
		// goroutines for marking.
		work.stackRoots, work.nMaybeRunnableStackRoots = allGsSnapshotSortedForGC()
	} else {
		// regular GC --- scan every goroutine
		work.stackRoots = allGsSnapshot()
		work.nMaybeRunnableStackRoots = len(work.stackRoots)
	}

	work.nStackRoots = len(work.stackRoots)

	work.markrootNext.Store(0)
	work.markrootJobs.Store(uint32(fixedRootCount + work.nDataRoots + work.nBSSRoots + work.nSpanRoots + work.nMaybeRunnableStackRoots))

	// Calculate base indexes of each root type
	work.baseData = uint32(fixedRootCount)
	work.baseBSS = work.baseData + uint32(work.nDataRoots)
	work.baseSpans = work.baseBSS + uint32(work.nBSSRoots)
	work.baseStacks = work.baseSpans + uint32(work.nSpanRoots)
	work.baseEnd = work.baseStacks + uint32(work.nStackRoots)
}

// gcMarkRootCheck checks that all roots have been scanned. It is
// purely for debugging.
func gcMarkRootCheck() {
	if next, jobs := work.markrootNext.Load(), work.markrootJobs.Load(); next < jobs {
		print(next, " of ", jobs, " markroot jobs done\n")
		throw("left over markroot jobs")
	}

	// Check that stacks have been scanned.
	//
	// We only check the first nStackRoots Gs that we should have scanned.
	// Since we don't care about newer Gs (see comment in
	// gcPrepareMarkRoots), no locking is required.
	i := 0
	forEachGRace(func(gp *g) {
		if i >= work.nStackRoots {
			return
		}

		if !gp.gcscandone {
			println("gp", gp, "goid", gp.goid,
				"status", readgstatus(gp),
				"gcscandone", gp.gcscandone)
			throw("scan missed a g")
		}

		i++
	})
}

// oneptrmask for an allocation containing a single pointer.
var oneptrmask = [...]uint8{1}

// markroot scans the i'th root.
//
// Preemption must be disabled (because this uses a gcWork).
//
// Returns the amount of GC work credit produced by the operation.
// If flushBgCredit is true, then that credit is also flushed
// to the background credit pool.
//
// nowritebarrier is only advisory here.
//
//go:nowritebarrier
func markroot(gcw *gcWork, i uint32, flushBgCredit bool) int64 {
	// Note: if you add a case here, please also update heapdump.go:dumproots.
	var workDone int64
	var workCounter *atomic.Int64
	switch {
	case work.baseData <= i && i < work.baseBSS:
		workCounter = &gcController.globalsScanWork
		for _, datap := range activeModules() {
			workDone += markrootBlock(datap.data, datap.edata-datap.data, datap.gcdatamask.bytedata, gcw, int(i-work.baseData))
		}

	case work.baseBSS <= i && i < work.baseSpans:
		workCounter = &gcController.globalsScanWork
		for _, datap := range activeModules() {
			workDone += markrootBlock(datap.bss, datap.ebss-datap.bss, datap.gcbssmask.bytedata, gcw, int(i-work.baseBSS))
		}

	case i == fixedRootFinalizers:
		for fb := allfin; fb != nil; fb = fb.alllink {
			cnt := uintptr(atomic.Load(&fb.cnt))
			scanblock(uintptr(unsafe.Pointer(&fb.fin[0])), cnt*unsafe.Sizeof(fb.fin[0]), &finptrmask[0], gcw, nil)
		}

	case i == fixedRootFreeGStacks:
		// Switch to the system stack so we can call
		// stackfree.
		systemstack(markrootFreeGStacks)

	case i == fixedRootCleanups:
		for cb := (*cleanupBlock)(gcCleanups.all.Load()); cb != nil; cb = cb.alllink {
			// N.B. This only needs to synchronize with cleanup execution, which only resets these blocks.
			// All cleanup queueing happens during sweep.
			n := uintptr(atomic.Load(&cb.n))
			scanblock(uintptr(unsafe.Pointer(&cb.cleanups[0])), n*unsafe.Sizeof(cleanupFn{}), &cleanupBlockPtrMask[0], gcw, nil)
		}

	case work.baseSpans <= i && i < work.baseStacks:
		// mark mspan.specials
		markrootSpans(gcw, int(i-work.baseSpans))

	default:
		// the rest is scanning goroutine stacks
		workCounter = &gcController.stackScanWork
		if i < work.baseStacks || work.baseEnd <= i {
			printlock()
			print("runtime: markroot index ", i, " not in stack roots range [", work.baseStacks, ", ", work.baseEnd, ")\n")
			throw("markroot: bad index")
		}
		gp := work.stackRoots[i-work.baseStacks]

		// remember when we've first observed the G blocked
		// needed only to output in traceback
		status := readgstatus(gp) // We are not in a scan state
		if (status == _Gwaiting || status == _Gsyscall) && gp.waitsince == 0 {
			gp.waitsince = work.tstart
		}

		// scanstack must be done on the system stack in case
		// we're trying to scan our own stack.
		systemstack(func() {
			// If this is a self-scan, put the user G in
			// _Gwaiting to prevent self-deadlock. It may
			// already be in _Gwaiting if this is a mark
			// worker or we're in mark termination.
			userG := getg().m.curg
			selfScan := gp == userG && readgstatus(userG) == _Grunning
			if selfScan {
				casGToWaitingForSuspendG(userG, _Grunning, waitReasonGarbageCollectionScan)
			}

			// TODO: suspendG blocks (and spins) until gp
			// stops, which may take a while for
			// running goroutines. Consider doing this in
			// two phases where the first is non-blocking:
			// we scan the stacks we can and ask running
			// goroutines to scan themselves; and the
			// second blocks.
			stopped := suspendG(gp)
			if stopped.dead {
				gp.gcscandone = true
				return
			}
			if gp.gcscandone {
				throw("g already scanned")
			}
			workDone += scanstack(gp, gcw)
			gp.gcscandone = true
			resumeG(stopped)

			if selfScan {
				casgstatus(userG, _Gwaiting, _Grunning)
			}
		})
	}
	if workCounter != nil && workDone != 0 {
		workCounter.Add(workDone)
		if flushBgCredit {
			gcFlushBgCredit(workDone)
		}
	}
	return workDone
}

// markrootBlock scans the shard'th shard of the block of memory [b0,
// b0+n0), with the given pointer mask.
//
// Returns the amount of work done.
//
//go:nowritebarrier
func markrootBlock(b0, n0 uintptr, ptrmask0 *uint8, gcw *gcWork, shard int) int64 {
	if rootBlockBytes%(8*goarch.PtrSize) != 0 {
		// This is necessary to pick byte offsets in ptrmask0.
		throw("rootBlockBytes must be a multiple of 8*ptrSize")
	}

	// Note that if b0 is toward the end of the address space,
	// then b0 + rootBlockBytes might wrap around.
	// These tests are written to avoid any possible overflow.
	off := uintptr(shard) * rootBlockBytes
	if off >= n0 {
		return 0
	}
	b := b0 + off
	ptrmask := (*uint8)(add(unsafe.Pointer(ptrmask0), uintptr(shard)*(rootBlockBytes/(8*goarch.PtrSize))))
	n := uintptr(rootBlockBytes)
	if off+n > n0 {
		n = n0 - off
	}

	// Scan this shard.
	scanblock(b, n, ptrmask, gcw, nil)
	return int64(n)
}

// markrootFreeGStacks frees stacks of dead Gs.
//
// This does not free stacks of dead Gs cached on Ps, but having a few
// cached stacks around isn't a problem.
func markrootFreeGStacks() {
	// Take list of dead Gs with stacks.
	lock(&sched.gFree.lock)
	list := sched.gFree.stack
	sched.gFree.stack = gList{}
	unlock(&sched.gFree.lock)
	if list.empty() {
		return
	}

	// Free stacks.
	var tail *g
	for gp := list.head.ptr(); gp != nil; gp = gp.schedlink.ptr() {
		tail = gp
		stackfree(gp.stack)
		gp.stack.lo = 0
		gp.stack.hi = 0
		if valgrindenabled {
			valgrindDeregisterStack(gp.valgrindStackID)
			gp.valgrindStackID = 0
		}
	}

	q := gQueue{list.head, tail.guintptr(), list.size}

	// Put Gs back on the free list.
	lock(&sched.gFree.lock)
	sched.gFree.noStack.pushAll(q)
	unlock(&sched.gFree.lock)
}

// markrootSpans marks roots for one shard of markArenas.
//
//go:nowritebarrier
func markrootSpans(gcw *gcWork, shard int) {
	// Objects with finalizers have two GC-related invariants:
	//
	// 1) Everything reachable from the object must be marked.
	// This ensures that when we pass the object to its finalizer,
	// everything the finalizer can reach will be retained.
	//
	// 2) Finalizer specials (which are not in the garbage
	// collected heap) are roots. In practice, this means the fn
	// field must be scanned.
	//
	// Objects with weak handles have only one invariant related
	// to this function: weak handle specials (which are not in the
	// garbage collected heap) are roots. In practice, this means
	// the handle field must be scanned. Note that the value the
	// handle pointer referenced does *not* need to be scanned. See
	// the definition of specialWeakHandle for details.
	sg := mheap_.sweepgen

	// Find the arena and page index into that arena for this shard.
	ai := mheap_.markArenas[shard/(pagesPerArena/pagesPerSpanRoot)]
	ha := mheap_.arenas[ai.l1()][ai.l2()]
	arenaPage := uint(uintptr(shard) * pagesPerSpanRoot % pagesPerArena)

	// Construct slice of bitmap which we'll iterate over.
	specialsbits := ha.pageSpecials[arenaPage/8:]
	specialsbits = specialsbits[:pagesPerSpanRoot/8]
	for i := range specialsbits {
		// Find set bits, which correspond to spans with specials.
		specials := atomic.Load8(&specialsbits[i])
		if specials == 0 {
			continue
		}
		for j := uint(0); j < 8; j++ {
			if specials&(1<<j) == 0 {
				continue
			}
			// Find the span for this bit.
			//
			// This value is guaranteed to be non-nil because having
			// specials implies that the span is in-use, and since we're
			// currently marking we can be sure that we don't have to worry
			// about the span being freed and re-used.
			s := ha.spans[arenaPage+uint(i)*8+j]

			// The state must be mSpanInUse if the specials bit is set, so
			// sanity check that.
			if state := s.state.get(); state != mSpanInUse {
				print("s.state = ", state, "\n")
				throw("non in-use span found with specials bit set")
			}
			// Check that this span was swept (it may be cached or uncached).
			if !useCheckmark && !(s.sweepgen == sg || s.sweepgen == sg+3) {
				// sweepgen was updated (+2) during non-checkmark GC pass
				print("sweep ", s.sweepgen, " ", sg, "\n")
				throw("gc: unswept span")
			}

			// Lock the specials to prevent a special from being
			// removed from the list while we're traversing it.
			lock(&s.speciallock)
			for sp := s.specials; sp != nil; sp = sp.next {
				switch sp.kind {
				case _KindSpecialFinalizer:
					gcScanFinalizer((*specialfinalizer)(unsafe.Pointer(sp)), s, gcw)
				case _KindSpecialWeakHandle:
					// The special itself is a root.
					spw := (*specialWeakHandle)(unsafe.Pointer(sp))
					scanblock(uintptr(unsafe.Pointer(&spw.handle)), goarch.PtrSize, &oneptrmask[0], gcw, nil)
				case _KindSpecialCleanup:
					gcScanCleanup((*specialCleanup)(unsafe.Pointer(sp)), gcw)
				}
			}
			unlock(&s.speciallock)
		}
	}
}

// gcScanFinalizer scans the relevant parts of a finalizer special as a root.
func gcScanFinalizer(spf *specialfinalizer, s *mspan, gcw *gcWork) {
	// Don't mark finalized object, but scan it so we retain everything it points to.

	// A finalizer can be set for an inner byte of an object, find object beginning.
	p := s.base() + spf.special.offset/s.elemsize*s.elemsize

	// Mark everything that can be reached from
	// the object (but *not* the object itself or
	// we'll never collect it).
	if !s.spanclass.noscan() {
		scanObject(p, gcw)
	}

	// The special itself is also a root.
	scanblock(uintptr(unsafe.Pointer(&spf.fn)), goarch.PtrSize, &oneptrmask[0], gcw, nil)
}

// gcScanCleanup scans the relevant parts of a cleanup special as a root.
func gcScanCleanup(spc *specialCleanup, gcw *gcWork) {
	// The special itself is a root.
	scanblock(uintptr(unsafe.Pointer(&spc.cleanup)), unsafe.Sizeof(cleanupFn{}), &cleanupFnPtrMask[0], gcw, nil)
}

// gcAssistAlloc performs GC work to make gp's assist debt positive.
// gp must be the calling user goroutine.
//
// This must be called with preemption enabled.
func gcAssistAlloc(gp *g) {
	// Don't assist in non-preemptible contexts. These are
	// generally fragile and won't allow the assist to block.
	if getg() == gp.m.g0 {
		return
	}
	if mp := getg().m; mp.locks > 0 || mp.preemptoff != "" {
		return
	}

	if gp := getg(); gp.bubble != nil {
		// Disassociate the G from its synctest bubble while allocating.
		// This is less elegant than incrementing the group's active count,
		// but avoids any contamination between GC assist and synctest.
		bubble := gp.bubble
		gp.bubble = nil
		defer func() {
			gp.bubble = bubble
		}()
	}

	// This extremely verbose boolean indicates whether we've
	// entered mark assist from the perspective of the tracer.
	//
	// In the tracer, this is just before we call gcAssistAlloc1
	// *regardless* of whether tracing is enabled. This is because
	// the tracer allows for tracing to begin (and advance
	// generations) in the middle of a GC mark phase, so we need to
	// record some state so that the tracer can pick it up to ensure
	// a consistent trace result.
	//
	// TODO(mknyszek): Hide the details of inMarkAssist in tracer
	// functions and simplify all the state tracking. This is a lot.
	enteredMarkAssistForTracing := false
retry:
	if gcCPULimiter.limiting() {
		// If the CPU limiter is enabled, intentionally don't
		// assist to reduce the amount of CPU time spent in the GC.
		if enteredMarkAssistForTracing {
			trace := traceAcquire()
			if trace.ok() {
				trace.GCMarkAssistDone()
				// Set this *after* we trace the end to make sure
				// that we emit an in-progress event if this is
				// the first event for the goroutine in the trace
				// or trace generation. Also, do this between
				// acquire/release because this is part of the
				// goroutine's trace state, and it must be atomic
				// with respect to the tracer.
				gp.inMarkAssist = false
				traceRelease(trace)
			} else {
				// This state is tracked even if tracing isn't enabled.
				// It's only used by the new tracer.
				// See the comment on enteredMarkAssistForTracing.
				gp.inMarkAssist = false
			}
		}
		return
	}
	// Compute the amount of scan work we need to do to make the
	// balance positive. When the required amount of work is low,
	// we over-assist to build up credit for future allocations
	// and amortize the cost of assisting.
	assistWorkPerByte := gcController.assistWorkPerByte.Load()
	assistBytesPerWork := gcController.assistBytesPerWork.Load()
	debtBytes := -gp.gcAssistBytes
	scanWork := int64(assistWorkPerByte * float64(debtBytes))
	if scanWork < gcOverAssistWork {
		scanWork = gcOverAssistWork
		debtBytes = int64(assistBytesPerWork * float64(scanWork))
	}

	// Steal as much credit as we can from the background GC's
	// scan credit. This is racy and may drop the background
	// credit below 0 if two mutators steal at the same time. This
	// will just cause steals to fail until credit is accumulated
	// again, so in the long run it doesn't really matter, but we
	// do have to handle the negative credit case.
	bgScanCredit := gcController.bgScanCredit.Load()
	stolen := int64(0)
	if bgScanCredit > 0 {
		if bgScanCredit < scanWork {
			stolen = bgScanCredit
			gp.gcAssistBytes += 1 + int64(assistBytesPerWork*float64(stolen))
		} else {
			stolen = scanWork
			gp.gcAssistBytes += debtBytes
		}
		gcController.bgScanCredit.Add(-stolen)

		scanWork -= stolen

		if scanWork == 0 {
			// We were able to steal all of the credit we
			// needed.
			if enteredMarkAssistForTracing {
				trace := traceAcquire()
				if trace.ok() {
					trace.GCMarkAssistDone()
					// Set this *after* we trace the end to make sure
					// that we emit an in-progress event if this is
					// the first event for the goroutine in the trace
					// or trace generation. Also, do this between
					// acquire/release because this is part of the
					// goroutine's trace state, and it must be atomic
					// with respect to the tracer.
					gp.inMarkAssist = false
					traceRelease(trace)
				} else {
					// This state is tracked even if tracing isn't enabled.
					// It's only used by the new tracer.
					// See the comment on enteredMarkAssistForTracing.
					gp.inMarkAssist = false
				}
			}
			return
		}
	}
	if !enteredMarkAssistForTracing {
		trace := traceAcquire()
		if trace.ok() {
			trace.GCMarkAssistStart()
			// Set this *after* we trace the start, otherwise we may
			// emit an in-progress event for an assist we're about to start.
			gp.inMarkAssist = true
			traceRelease(trace)
		} else {
			gp.inMarkAssist = true
		}
		// In the new tracer, set enter mark assist tracing if we
		// ever pass this point, because we must manage inMarkAssist
		// correctly.
		//
		// See the comment on enteredMarkAssistForTracing.
		enteredMarkAssistForTracing = true
	}

	// Perform assist work
	systemstack(func() {
		gcAssistAlloc1(gp, scanWork)
		// The user stack may have moved, so this can't touch
		// anything on it until it returns from systemstack.
	})

	completed := gp.param != nil
	gp.param = nil
	if completed {
		gcMarkDone()
	}

	if gp.gcAssistBytes < 0 {
		// We were unable steal enough credit or perform
		// enough work to pay off the assist debt. We need to
		// do one of these before letting the mutator allocate
		// more to prevent over-allocation.
		//
		// If this is because we were preempted, reschedule
		// and try some more.
		if gp.preempt {
			Gosched()
			goto retry
		}

		// Add this G to an assist queue and park. When the GC
		// has more background credit, it will satisfy queued
		// assists before flushing to the global credit pool.
		//
		// Note that this does *not* get woken up when more
		// work is added to the work list. The theory is that
		// there wasn't enough work to do anyway, so we might
		// as well let background marking take care of the
		// work that is available.
		if !gcParkAssist() {
			goto retry
		}

		// At this point either background GC has satisfied
		// this G's assist debt, or the GC cycle is over.
	}
	if enteredMarkAssistForTracing {
		trace := traceAcquire()
		if trace.ok() {
			trace.GCMarkAssistDone()
			// Set this *after* we trace the end to make sure
			// that we emit an in-progress event if this is
			// the first event for the goroutine in the trace
			// or trace generation. Also, do this between
			// acquire/release because this is part of the
			// goroutine's trace state, and it must be atomic
			// with respect to the tracer.
			gp.inMarkAssist = false
			traceRelease(trace)
		} else {
			// This state is tracked even if tracing isn't enabled.
			// It's only used by the new tracer.
			// See the comment on enteredMarkAssistForTracing.
			gp.inMarkAssist = false
		}
	}
}

// gcAssistAlloc1 is the part of gcAssistAlloc that runs on the system
// stack. This is a separate function to make it easier to see that
// we're not capturing anything from the user stack, since the user
// stack may move while we're in this function.
//
// gcAssistAlloc1 indicates whether this assist completed the mark
// phase by setting gp.param to non-nil. This can't be communicated on
// the stack since it may move.
//
//go:systemstack
func gcAssistAlloc1(gp *g, scanWork int64) {
	// Clear the flag indicating that this assist completed the
	// mark phase.
	gp.param = nil

	if atomic.Load(&gcBlackenEnabled) == 0 {
		// The gcBlackenEnabled check in malloc races with the
		// store that clears it but an atomic check in every malloc
		// would be a performance hit.
		// Instead we recheck it here on the non-preemptible system
		// stack to determine if we should perform an assist.

		// GC is done, so ignore any remaining debt.
		gp.gcAssistBytes = 0
		return
	}

	// Track time spent in this assist. Since we're on the
	// system stack, this is non-preemptible, so we can
	// just measure start and end time.
	//
	// Limiter event tracking might be disabled if we end up here
	// while on a mark worker.
	startTime := nanotime()
	trackLimiterEvent := gp.m.p.ptr().limiterEvent.start(limiterEventMarkAssist, startTime)

	gcBeginWork()

	// gcDrainN requires the caller to be preemptible.
	casGToWaitingForSuspendG(gp, _Grunning, waitReasonGCAssistMarking)

	// drain own cached work first in the hopes that it
	// will be more cache friendly.
	gcw := &getg().m.p.ptr().gcw
	workDone := gcDrainN(gcw, scanWork)

	casgstatus(gp, _Gwaiting, _Grunning)

	// Record that we did this much scan work.
	//
	// Back out the number of bytes of assist credit that
	// this scan work counts for. The "1+" is a poor man's
	// round-up, to ensure this adds credit even if
	// assistBytesPerWork is very low.
	assistBytesPerWork := gcController.assistBytesPerWork.Load()
	gp.gcAssistBytes += 1 + int64(assistBytesPerWork*float64(workDone))

	// If this is the last worker and we ran out of work,
	// signal a completion point.
	if gcEndWork() {
		// This has reached a background completion point. Set
		// gp.param to a non-nil value to indicate this. It
		// doesn't matter what we set it to (it just has to be
		// a valid pointer).
		gp.param = unsafe.Pointer(gp)
	}
	now := nanotime()
	duration := now - startTime
	pp := gp.m.p.ptr()
	pp.gcAssistTime += duration
	if trackLimiterEvent {
		pp.limiterEvent.stop(limiterEventMarkAssist, now)
	}
	if pp.gcAssistTime > gcAssistTimeSlack {
		gcController.assistTime.Add(pp.gcAssistTime)
		gcCPULimiter.update(now)
		pp.gcAssistTime = 0
	}
}

// gcWakeAllAssists wakes all currently blocked assists. This is used
// at the end of a GC cycle. gcBlackenEnabled must be false to prevent
// new assists from going to sleep after this point.
func gcWakeAllAssists() {
	lock(&work.assistQueue.lock)
	list := work.assistQueue.q.popList()
	injectglist(&list)
	unlock(&work.assistQueue.lock)
}

// gcParkAssist puts the current goroutine on the assist queue and parks.
//
// gcParkAssist reports whether the assist is now satisfied. If it
// returns false, the caller must retry the assist.
func gcParkAssist() bool {
	lock(&work.assistQueue.lock)
	// If the GC cycle finished while we were getting the lock,
	// exit the assist. The cycle can't finish while we hold the
	// lock.
	if atomic.Load(&gcBlackenEnabled) == 0 {
		unlock(&work.assistQueue.lock)
		return true
	}

	gp := getg()
	oldList := work.assistQueue.q
	work.assistQueue.q.pushBack(gp)

	// Recheck for background credit now that this G is in
	// the queue, but can still back out. This avoids a
	// race in case background marking has flushed more
	// credit since we checked above.
	if gcController.bgScanCredit.Load() > 0 {
		work.assistQueue.q = oldList
		if oldList.tail != 0 {
			oldList.tail.ptr().schedlink.set(nil)
		}
		unlock(&work.assistQueue.lock)
		return false
	}
	// Park.
	goparkunlock(&work.assistQueue.lock, waitReasonGCAssistWait, traceBlockGCMarkAssist, 2)
	return true
}

// gcFlushBgCredit flushes scanWork units of background scan work
// credit. This first satisfies blocked assists on the
// work.assistQueue and then flushes any remaining credit to
// gcController.bgScanCredit.
//
// Write barriers are disallowed because this is used by gcDrain after
// it has ensured that all work is drained and this must preserve that
// condition.
//
//go:nowritebarrierrec
func gcFlushBgCredit(scanWork int64) {
	if work.assistQueue.q.empty() {
		// Fast path; there are no blocked assists. There's a
		// small window here where an assist may add itself to
		// the blocked queue and park. If that happens, we'll
		// just get it on the next flush.
		gcController.bgScanCredit.Add(scanWork)
		return
	}

	assistBytesPerWork := gcController.assistBytesPerWork.Load()
	scanBytes := int64(float64(scanWork) * assistBytesPerWork)

	lock(&work.assistQueue.lock)
	for !work.assistQueue.q.empty() && scanBytes > 0 {
		gp := work.assistQueue.q.pop()
		// Note that gp.gcAssistBytes is negative because gp
		// is in debt. Think carefully about the signs below.
		if scanBytes+gp.gcAssistBytes >= 0 {
			// Satisfy this entire assist debt.
			scanBytes += gp.gcAssistBytes
			gp.gcAssistBytes = 0
			// It's important that we *not* put gp in
			// runnext. Otherwise, it's possible for user
			// code to exploit the GC worker's high
			// scheduler priority to get itself always run
			// before other goroutines and always in the
			// fresh quantum started by GC.
			ready(gp, 0, false)
		} else {
			// Partially satisfy this assist.
			gp.gcAssistBytes += scanBytes
			scanBytes = 0
			// As a heuristic, we move this assist to the
			// back of the queue so that large assists
			// can't clog up the assist queue and
			// substantially delay small assists.
			work.assistQueue.q.pushBack(gp)
			break
		}
	}

	if scanBytes > 0 {
		// Convert from scan bytes back to work.
		assistWorkPerByte := gcController.assistWorkPerByte.Load()
		scanWork = int64(float64(scanBytes) * assistWorkPerByte)
		gcController.bgScanCredit.Add(scanWork)
	}
	unlock(&work.assistQueue.lock)
}

// scanstack scans gp's stack, greying all pointers found on the stack.
//
// Returns the amount of scan work performed, but doesn't update
// gcController.stackScanWork or flush any credit. Any background credit produced
// by this function should be flushed by its caller. scanstack itself can't
// safely flush because it may result in trying to wake up a goroutine that
// was just scanned, resulting in a self-deadlock.
//
// scanstack will also shrink the stack if it is safe to do so. If it
// is not, it schedules a stack shrink for the next synchronous safe
// point.
//
// scanstack is marked go:systemstack because it must not be preempted
// while using a workbuf.
//
//go:nowritebarrier
//go:systemstack
func scanstack(gp *g, gcw *gcWork) int64 {
	if readgstatus(gp)&_Gscan == 0 {
		print("runtime:scanstack: gp=", gp, ", goid=", gp.goid, ", gp->atomicstatus=", hex(readgstatus(gp)), "\n")
		throw("scanstack - bad status")
	}

	switch readgstatus(gp) &^ _Gscan {
	default:
		print("runtime: gp=", gp, ", goid=", gp.goid, ", gp->atomicstatus=", readgstatus(gp), "\n")
		throw("mark - bad status")
	case _Gdead, _Gdeadextra:
		return 0
	case _Grunning:
		print("runtime: gp=", gp, ", goid=", gp.goid, ", gp->atomicstatus=", readgstatus(gp), "\n")
		throw("scanstack: goroutine not stopped")
	case _Grunnable, _Gsyscall, _Gwaiting, _Gleaked:
		// ok
	}

	if gp == getg() {
		throw("can't scan our own stack")
	}

	// scannedSize is the amount of work we'll be reporting.
	//
	// It is less than the allocated size (which is hi-lo).
	var sp uintptr
	if gp.syscallsp != 0 {
		sp = gp.syscallsp // If in a system call this is the stack pointer (gp.sched.sp can be 0 in this case on Windows).
	} else {
		sp = gp.sched.sp
	}
	scannedSize := gp.stack.hi - sp

	// Keep statistics for initial stack size calculation.
	// Note that this accumulates the scanned size, not the allocated size.
	p := getg().m.p.ptr()
	p.scannedStackSize += uint64(scannedSize)
	p.scannedStacks++

	if isShrinkStackSafe(gp) {
		// Shrink the stack if not much of it is being used.
		shrinkstack(gp)
	} else {
		// Otherwise, shrink the stack at the next sync safe point.
		gp.preemptShrink = true
	}

	var state stackScanState
	state.stack = gp.stack

	if stackTraceDebug {
		println("stack trace goroutine", gp.goid)
	}

	if debugScanConservative && gp.asyncSafePoint {
		print("scanning async preempted goroutine ", gp.goid, " stack [", hex(gp.stack.lo), ",", hex(gp.stack.hi), ")\n")
	}

	// Scan the saved context register. This is effectively a live
	// register that gets moved back and forth between the
	// register and sched.ctxt without a write barrier.
	if gp.sched.ctxt != nil {
		scanblock(uintptr(unsafe.Pointer(&gp.sched.ctxt)), goarch.PtrSize, &oneptrmask[0], gcw, &state)
	}

	// Scan conservatively the extended register state.
	if gp.asyncSafePoint {
		xRegScan(gp, gcw, &state)
	}

	// Scan the stack. Accumulate a list of stack objects.
	var u unwinder
	for u.init(gp, 0); u.valid(); u.next() {
		scanframeworker(&u.frame, &state, gcw)
	}

	// Find additional pointers that point into the stack from the heap.
	// Currently this includes defers and panics. See also function copystack.

	// Find and trace other pointers in defer records.
	for d := gp._defer; d != nil; d = d.link {
		if d.fn != nil {
			// Scan the func value, which could be a stack allocated closure.
			// See issue 30453.
			scanblock(uintptr(unsafe.Pointer(&d.fn)), goarch.PtrSize, &oneptrmask[0], gcw, &state)
		}
		if d.link != nil {
			// The link field of a stack-allocated defer record might point
			// to a heap-allocated defer record. Keep that heap record live.
			scanblock(uintptr(unsafe.Pointer(&d.link)), goarch.PtrSize, &oneptrmask[0], gcw, &state)
		}
		// Retain defers records themselves.
		// Defer records might not be reachable from the G through regular heap
		// tracing because the defer linked list might weave between the stack and the heap.
		if d.heap {
			scanblock(uintptr(unsafe.Pointer(&d)), goarch.PtrSize, &oneptrmask[0], gcw, &state)
		}
	}
	if gp._panic != nil {
		// Panics are always stack allocated.
		state.putPtr(uintptr(unsafe.Pointer(gp._panic)), false)
	}

	// Find and scan all reachable stack objects.
	//
	// The state's pointer queue prioritizes precise pointers over
	// conservative pointers so that we'll prefer scanning stack
	// objects precisely.
	state.buildIndex()
	for {
		p, conservative := state.getPtr()
		if p == 0 {
			break
		}
		obj := state.findObject(p)
		if obj == nil {
			continue
		}
		r := obj.r
		if r == nil {
			// We've already scanned this object.
			continue
		}
		obj.setRecord(nil) // Don't scan it again.
		if stackTraceDebug {
			printlock()
			print("  live stkobj at", hex(state.stack.lo+uintptr(obj.off)), "of size", obj.size)
			if conservative {
				print(" (conservative)")
			}
			println()
			printunlock()
		}
		ptrBytes, gcData := r.gcdata()
		b := state.stack.lo + uintptr(obj.off)
		if conservative {
			scanConservative(b, ptrBytes, gcData, gcw, &state)
		} else {
			scanblock(b, ptrBytes, gcData, gcw, &state)
		}
	}

	// Deallocate object buffers.
	// (Pointer buffers were all deallocated in the loop above.)
	for state.head != nil {
		x := state.head
		state.head = x.next
		if stackTraceDebug {
			for i := 0; i < x.nobj; i++ {
				obj := &x.obj[i]
				if obj.r == nil { // reachable
					continue
				}
				println("  dead stkobj at", hex(gp.stack.lo+uintptr(obj.off)), "of size", obj.r.size)
				// Note: not necessarily really dead - only reachable-from-ptr dead.
			}
		}
		x.nobj = 0
		putempty((*workbuf)(unsafe.Pointer(x)))
	}
	if state.buf != nil || state.cbuf != nil || state.freeBuf != nil {
		throw("remaining pointer buffers")
	}
	return int64(scannedSize)
}

// Scan a stack frame: local variables and function arguments/results.
//
//go:nowritebarrier
func scanframeworker(frame *stkframe, state *stackScanState, gcw *gcWork) {
	if _DebugGC > 1 && frame.continpc != 0 {
		print("scanframe ", funcname(frame.fn), "\n")
	}

	isAsyncPreempt := frame.fn.valid() && frame.fn.funcID == abi.FuncID_asyncPreempt
	isDebugCall := frame.fn.valid() && frame.fn.funcID == abi.FuncID_debugCallV2
	if state.conservative || isAsyncPreempt || isDebugCall {
		if debugScanConservative {
			println("conservatively scanning function", funcname(frame.fn), "at PC", hex(frame.continpc))
		}

		// Conservatively scan the frame. Unlike the precise
		// case, this includes the outgoing argument space
		// since we may have stopped while this function was
		// setting up a call.
		//
		// TODO: We could narrow this down if the compiler
		// produced a single map per function of stack slots
		// and registers that ever contain a pointer.
		if frame.varp != 0 {
			size := frame.varp - frame.sp
			if size > 0 {
				scanConservative(frame.sp, size, nil, gcw, state)
			}
		}

		// Scan arguments to this frame.
		if n := frame.argBytes(); n != 0 {
			// TODO: We could pass the entry argument map
			// to narrow this down further.
			scanConservative(frame.argp, n, nil, gcw, state)
		}

		if isAsyncPreempt || isDebugCall {
			// This function's frame contained the
			// registers for the asynchronously stopped
			// parent frame. Scan the parent
			// conservatively.
			state.conservative = true
		} else {
			// We only wanted to scan those two frames
			// conservatively. Clear the flag for future
			// frames.
			state.conservative = false
		}
		return
	}

	locals, args, objs := frame.getStackMap(false)

	// Scan local variables if stack frame has been allocated.
	if locals.n > 0 {
		size := uintptr(locals.n) * goarch.PtrSize
		scanblock(frame.varp-size, size, locals.bytedata, gcw, state)
	}

	// Scan arguments.
	if args.n > 0 {
		scanblock(frame.argp, uintptr(args.n)*goarch.PtrSize, args.bytedata, gcw, state)
	}

	// Add all stack objects to the stack object list.
	if frame.varp != 0 {
		// varp is 0 for defers, where there are no locals.
		// In that case, there can't be a pointer to its args, either.
		// (And all args would be scanned above anyway.)
		for i := range objs {
			obj := &objs[i]
			off := obj.off
			base := frame.varp // locals base pointer
			if off >= 0 {
				base = frame.argp // arguments and return values base pointer
			}
			ptr := base + uintptr(off)
			if ptr < frame.sp {
				// object hasn't been allocated in the frame yet.
				continue
			}
			if stackTraceDebug {
				println("stkobj at", hex(ptr), "of size", obj.size)
			}
			state.addObject(ptr, obj)
		}
	}
}

type gcDrainFlags int

const (
	gcDrainUntilPreempt gcDrainFlags = 1 << iota
	gcDrainFlushBgCredit
	gcDrainIdle
	gcDrainFractional
)

// gcDrainMarkWorkerIdle is a wrapper for gcDrain that exists to better account
// mark time in profiles.
func gcDrainMarkWorkerIdle(gcw *gcWork) {
	gcDrain(gcw, gcDrainIdle|gcDrainUntilPreempt|gcDrainFlushBgCredit)
}

// gcDrainMarkWorkerDedicated is a wrapper for gcDrain that exists to better account
// mark time in profiles.
func gcDrainMarkWorkerDedicated(gcw *gcWork, untilPreempt bool) {
	flags := gcDrainFlushBgCredit
	if untilPreempt {
		flags |= gcDrainUntilPreempt
	}
	gcDrain(gcw, flags)
}

// gcDrainMarkWorkerFractional is a wrapper for gcDrain that exists to better account
// mark time in profiles.
func gcDrainMarkWorkerFractional(gcw *gcWork) {
	gcDrain(gcw, gcDrainFractional|gcDrainUntilPreempt|gcDrainFlushBgCredit)
}

// gcNextMarkRoot safely increments work.markrootNext and returns the
// index of the next root job. The returned boolean is true if the root job
// is valid, and false if there are no more root jobs to be claimed,
// i.e. work.markrootNext >= work.markrootJobs.
func gcNextMarkRoot() (uint32, bool) {
	if !work.goroutineLeak.enabled {
		// If not running goroutine leak detection, assume regular GC behavior.
		job := work.markrootNext.Add(1) - 1
		return job, job < work.markrootJobs.Load()
	}

	// Otherwise, use a CAS loop to increment markrootNext.
	for next, jobs := work.markrootNext.Load(), work.markrootJobs.Load(); next < jobs; next = work.markrootNext.Load() {
		// There is still work available at the moment.
		if work.markrootNext.CompareAndSwap(next, next+1) {
			// We manage to snatch a root job. Return the root index.
			return next, true
		}
	}
	return 0, false
}

// gcDrain scans roots and objects in work buffers, blackening grey
// objects until it is unable to get more work. It may return before
// GC is done; it's the caller's responsibility to balance work from
// other Ps.
//
// If flags&gcDrainUntilPreempt != 0, gcDrain returns when g.preempt
// is set.
//
// If flags&gcDrainIdle != 0, gcDrain returns when there is other work
// to do.
//
// If flags&gcDrainFractional != 0, gcDrain self-preempts when
// pollFractionalWorkerExit() returns true. This implies
// gcDrainNoBlock.
//
// If flags&gcDrainFlushBgCredit != 0, gcDrain flushes scan work
// credit to gcController.bgScanCredit every gcCreditSlack units of
// scan work.
//
// gcDrain will always return if there is a pending STW or forEachP.
//
// Disabling write barriers is necessary to ensure that after we've
// confirmed that we've drained gcw, that we don't accidentally end
// up flipping that condition by immediately adding work in the form
// of a write barrier buffer flush.
//
// Don't set nowritebarrierrec because it's safe for some callees to
// have write barriers enabled.
//
//go:nowritebarrier
func gcDrain(gcw *gcWork, flags gcDrainFlags) {
	if !writeBarrier.enabled {
		throw("gcDrain phase incorrect")
	}

	// N.B. We must be running in a non-preemptible context, so it's
	// safe to hold a reference to our P here.
	gp := getg().m.curg
	pp := gp.m.p.ptr()
	preemptible := flags&gcDrainUntilPreempt != 0
	flushBgCredit := flags&gcDrainFlushBgCredit != 0
	idle := flags&gcDrainIdle != 0

	initScanWork := gcw.heapScanWork

	// checkWork is the scan work before performing the next
	// self-preempt check.
	checkWork := int64(1<<63 - 1)
	var check func() bool
	if flags&(gcDrainIdle|gcDrainFractional) != 0 {
		checkWork = initScanWork + drainCheckThreshold
		if idle {
			check = pollWork
		} else if flags&gcDrainFractional != 0 {
			check = pollFractionalWorkerExit
		}
	}

	if work.markrootNext.Load() < work.markrootJobs.Load() {
		// Stop if we're preemptible, if someone wants to STW, or if
		// someone is calling forEachP.
		for !(gp.preempt && (preemptible || sched.gcwaiting.Load() || pp.runSafePointFn != 0)) {
			job, ok := gcNextMarkRoot()
			if !ok {
				break
			}
			markroot(gcw, job, flushBgCredit)
			if check != nil && check() {
				goto done
			}

			// Spin up a new worker if requested.
			if goexperiment.GreenTeaGC && gcw.mayNeedWorker {
				gcw.mayNeedWorker = false
				if gcphase == _GCmark {
					gcController.enlistWorker()
				}
			}
		}
	}

	// Drain heap marking jobs.
	//
	// Stop if we're preemptible, if someone wants to STW, or if
	// someone is calling forEachP.
	//
	// TODO(mknyszek): Consider always checking gp.preempt instead
	// of having the preempt flag, and making an exception for certain
	// mark workers in retake. That might be simpler than trying to
	// enumerate all the reasons why we might want to preempt, even
	// if we're supposed to be mostly non-preemptible.
	for !(gp.preempt && (preemptible || sched.gcwaiting.Load() || pp.runSafePointFn != 0)) {
		// Try to keep work available on the global queue. We used to
		// check if there were waiting workers, but it's better to
		// just keep work available than to make workers wait. In the
		// worst case, we'll do O(log(_WorkbufSize)) unnecessary
		// balances.
		if work.full == 0 {
			gcw.balance()
		}

		// See mgcwork.go for the rationale behind the order in which we check these queues.
		var b uintptr
		var s objptr
		if b = gcw.tryGetObjFast(); b == 0 {
			if s = gcw.tryGetSpanFast(); s == 0 {
				if b = gcw.tryGetObj(); b == 0 {
					if s = gcw.tryGetSpan(); s == 0 {
						// Flush the write barrier
						// buffer; this may create
						// more work.
						wbBufFlush()
						if b = gcw.tryGetObj(); b == 0 {
							if s = gcw.tryGetSpan(); s == 0 {
								s = gcw.tryStealSpan()
							}
						}
					}
				}
			}
		}
		if b != 0 {
			scanObject(b, gcw)
		} else if s != 0 {
			scanSpan(s, gcw)
		} else {
			// Unable to get work.
			break
		}

		// Spin up a new worker if requested.
		if goexperiment.GreenTeaGC && gcw.mayNeedWorker {
			gcw.mayNeedWorker = false
			if gcphase == _GCmark {
				gcController.enlistWorker()
			}
		}

		// Flush background scan work credit to the global
		// account if we've accumulated enough locally so
		// mutator assists can draw on it.
		if gcw.heapScanWork >= gcCreditSlack {
			gcController.heapScanWork.Add(gcw.heapScanWork)
			if flushBgCredit {
				gcFlushBgCredit(gcw.heapScanWork - initScanWork)
				initScanWork = 0
			}
			checkWork -= gcw.heapScanWork
			gcw.heapScanWork = 0

			if checkWork <= 0 {
				checkWork += drainCheckThreshold
				if check != nil && check() {
					break
				}
			}
		}
	}

done:
	// Flush remaining scan work credit.
	if gcw.heapScanWork > 0 {
		gcController.heapScanWork.Add(gcw.heapScanWork)
		if flushBgCredit {
			gcFlushBgCredit(gcw.heapScanWork - initScanWork)
		}
		gcw.heapScanWork = 0
	}
}

// gcDrainN blackens grey objects until it has performed roughly
// scanWork units of scan work or the G is preempted. This is
// best-effort, so it may perform less work if it fails to get a work
// buffer. Otherwise, it will perform at least n units of work, but
// may perform more because scanning is always done in whole object
// increments. It returns the amount of scan work performed.
//
// The caller goroutine must be in a preemptible state (e.g.,
// _Gwaiting) to prevent deadlocks during stack scanning. As a
// consequence, this must be called on the system stack.
//
//go:nowritebarrier
//go:systemstack
func gcDrainN(gcw *gcWork, scanWork int64) int64 {
	if !writeBarrier.enabled {
		throw("gcDrainN phase incorrect")
	}

	// There may already be scan work on the gcw, which we don't
	// want to claim was done by this call.
	workFlushed := -gcw.heapScanWork

	// In addition to backing out because of a preemption, back out
	// if the GC CPU limiter is enabled.
	gp := getg().m.curg
	for !gp.preempt && !gcCPULimiter.limiting() && workFlushed+gcw.heapScanWork < scanWork {
		// See gcDrain comment.
		if work.full == 0 {
			gcw.balance()
		}

		// See mgcwork.go for the rationale behind the order in which we check these queues.
		var b uintptr
		var s objptr
		if b = gcw.tryGetObjFast(); b == 0 {
			if s = gcw.tryGetSpanFast(); s == 0 {
				if b = gcw.tryGetObj(); b == 0 {
					if s = gcw.tryGetSpan(); s == 0 {
						// Flush the write barrier
						// buffer; this may create
						// more work.
						wbBufFlush()
						if b = gcw.tryGetObj(); b == 0 {
							if s = gcw.tryGetSpan(); s == 0 {
								// Try to do a root job.
								if work.markrootNext.Load() < work.markrootJobs.Load() {
									job, ok := gcNextMarkRoot()
									if ok {
										workFlushed += markroot(gcw, job, false)
										continue
									}
								}
								s = gcw.tryStealSpan()
							}
						}
					}
				}
			}
		}
		if b != 0 {
			scanObject(b, gcw)
		} else if s != 0 {
			scanSpan(s, gcw)
		} else {
			// Unable to get work.
			break
		}

		// Flush background scan work credit.
		if gcw.heapScanWork >= gcCreditSlack {
			gcController.heapScanWork.Add(gcw.heapScanWork)
			workFlushed += gcw.heapScanWork
			gcw.heapScanWork = 0
		}

		// Spin up a new worker if requested.
		if goexperiment.GreenTeaGC && gcw.mayNeedWorker {
			gcw.mayNeedWorker = false
			if gcphase == _GCmark {
				gcController.enlistWorker()
			}
		}
	}

	// Unlike gcDrain, there's no need to flush remaining work
	// here because this never flushes to bgScanCredit and
	// gcw.dispose will flush any remaining work to scanWork.

	return workFlushed + gcw.heapScanWork
}

// scanblock scans b as scanObject would, but using an explicit
// pointer bitmap instead of the heap bitmap.
//
// This is used to scan non-heap roots, so it does not update
// gcw.bytesMarked or gcw.heapScanWork.
//
// If stk != nil, possible stack pointers are also reported to stk.putPtr.
//
//go:nowritebarrier
func scanblock(b0, n0 uintptr, ptrmask *uint8, gcw *gcWork, stk *stackScanState) {
	// Use local copies of original parameters, so that a stack trace
	// due to one of the throws below shows the original block
	// base and extent.
	b := b0
	n := n0

	for i := uintptr(0); i < n; {
		// Find bits for the next word.
		bits := uint32(*addb(ptrmask, i/(goarch.PtrSize*8)))
		if bits == 0 {
			i += goarch.PtrSize * 8
			continue
		}
		for j := 0; j < 8 && i < n; j++ {
			if bits&1 != 0 {
				// Same work as in scanObject; see comments there.
				p := *(*uintptr)(unsafe.Pointer(b + i))
				if p != 0 {
					if stk != nil && p >= stk.stack.lo && p < stk.stack.hi {
						stk.putPtr(p, false)
					} else {
						if !tryDeferToSpanScan(p, gcw) {
							if obj, span, objIndex := findObject(p, b, i); obj != 0 {
								greyobject(obj, b, i, span, gcw, objIndex)
							}
						}
					}
				}
			}
			bits >>= 1
			i += goarch.PtrSize
		}
	}
}

// scanConservative scans block [b, b+n) conservatively, treating any
// pointer-like value in the block as a pointer.
//
// If ptrmask != nil, only words that are marked in ptrmask are
// considered as potential pointers.
//
// If state != nil, it's assumed that [b, b+n) is a block in the stack
// and may contain pointers to stack objects.
func scanConservative(b, n uintptr, ptrmask *uint8, gcw *gcWork, state *stackScanState) {
	if debugScanConservative {
		printlock()
		print("conservatively scanning [", hex(b), ",", hex(b+n), ")\n")
		hexdumpWords(b, n, func(p uintptr, m hexdumpMarker) {
			if ptrmask != nil {
				word := (p - b) / goarch.PtrSize
				bits := *addb(ptrmask, word/8)
				if (bits>>(word%8))&1 == 0 {
					return
				}
			}

			val := *(*uintptr)(unsafe.Pointer(p))
			if state != nil && state.stack.lo <= val && val < state.stack.hi {
				m.start()
				println("ptr to stack")
				return
			}

			span := spanOfHeap(val)
			if span == nil {
				return
			}
			idx := span.objIndex(val)
			if span.isFreeOrNewlyAllocated(idx) {
				return
			}
			m.start()
			println("ptr to heap")
		})
		printunlock()
	}

	for i := uintptr(0); i < n; i += goarch.PtrSize {
		if ptrmask != nil {
			word := i / goarch.PtrSize
			bits := *addb(ptrmask, word/8)
			if bits == 0 {
				// Skip 8 words (the loop increment will do the 8th)
				//
				// This must be the first time we've
				// seen this word of ptrmask, so i
				// must be 8-word-aligned, but check
				// our reasoning just in case.
				if i%(goarch.PtrSize*8) != 0 {
					throw("misaligned mask")
				}
				i += goarch.PtrSize*8 - goarch.PtrSize
				continue
			}
			if (bits>>(word%8))&1 == 0 {
				continue
			}
		}

		val := *(*uintptr)(unsafe.Pointer(b + i))

		// Check if val points into the stack.
		if state != nil && state.stack.lo <= val && val < state.stack.hi {
			// val may point to a stack object. This
			// object may be dead from last cycle and
			// hence may contain pointers to unallocated
			// objects, but unlike heap objects we can't
			// tell if it's already dead. Hence, if all
			// pointers to this object are from
			// conservative scanning, we have to scan it
			// defensively, too.
			state.putPtr(val, true)
			continue
		}

		// Check if val points to a heap span.
		span := spanOfHeap(val)
		if span == nil {
			continue
		}

		// Check if val points to an allocated object.
		//
		// Ignore objects allocated during the mark phase, they've
		// been allocated black.
		idx := span.objIndex(val)
		if span.isFreeOrNewlyAllocated(idx) {
			continue
		}

		// val points to an allocated object. Mark it.
		obj := span.base() + idx*span.elemsize
		if !tryDeferToSpanScan(obj, gcw) {
			greyobject(obj, b, i, span, gcw, idx)
		}
	}
}

// Shade the object if it isn't already.
// The object is not nil and known to be in the heap.
// Preemption must be disabled.
//
//go:nowritebarrier
func shade(b uintptr) {
	gcw := &getg().m.p.ptr().gcw
	if !tryDeferToSpanScan(b, gcw) {
		if obj, span, objIndex := findObject(b, 0, 0); obj != 0 {
			greyobject(obj, 0, 0, span, gcw, objIndex)
		}
	}
}

// obj is the start of an object with mark mbits.
// If it isn't already marked, mark it and enqueue into gcw.
// base and off are for debugging only and could be removed.
//
// See also wbBufFlush1, which partially duplicates this logic.
//
//go:nowritebarrierrec
func greyobject(obj, base, off uintptr, span *mspan, gcw *gcWork, objIndex uintptr) {
	// obj should be start of allocation, and so must be at least pointer-aligned.
	if obj&(goarch.PtrSize-1) != 0 {
		throw("greyobject: obj not pointer-aligned")
	}
	mbits := span.markBitsForIndex(objIndex)

	if useCheckmark {
		if setCheckmark(obj, base, off, mbits) {
			// Already marked.
			return
		}
		if debug.checkfinalizers > 1 {
			print("  mark ", hex(obj), " found at *(", hex(base), "+", hex(off), ")\n")
		}
	} else {
		if debug.gccheckmark > 0 && span.isFree(objIndex) {
			print("runtime: marking free object ", hex(obj), " found at *(", hex(base), "+", hex(off), ")\n")
			gcDumpObject("base", base, off)
			gcDumpObject("obj", obj, ^uintptr(0))
			getg().m.traceback = 2
			throw("marking free object")
		}

		// If marked we have nothing to do.
		if mbits.isMarked() {
			return
		}
		mbits.setMarked()

		// Mark span.
		arena, pageIdx, pageMask := pageIndexOf(span.base())
		if arena.pageMarks[pageIdx]&pageMask == 0 {
			atomic.Or8(&arena.pageMarks[pageIdx], pageMask)
		}
	}

	// If this is a noscan object, fast-track it to black
	// instead of greying it.
	if span.spanclass.noscan() {
		gcw.bytesMarked += uint64(span.elemsize)
		return
	}

	// We're adding obj to P's local workbuf, so it's likely
	// this object will be processed soon by the same P.
	// Even if the workbuf gets flushed, there will likely still be
	// some benefit on platforms with inclusive shared caches.
	sys.Prefetch(obj)
	// Queue the obj for scanning.
	if !gcw.putObjFast(obj) {
		gcw.putObj(obj)
	}
}

// gcDumpObject dumps the contents of obj for debugging and marks the
// field at byte offset off in obj.
func gcDumpObject(label string, obj, off uintptr) {
	s := spanOf(obj)
	print(label, "=", hex(obj))
	if s == nil {
		print(" s=nil\n")
		return
	}
	print(" s.base()=", hex(s.base()), " s.limit=", hex(s.limit), " s.spanclass=", s.spanclass, " s.elemsize=", s.elemsize, " s.state=")
	if state := s.state.get(); 0 <= state && int(state) < len(mSpanStateNames) {
		print(mSpanStateNames[state], "\n")
	} else {
		print("unknown(", state, ")\n")
	}

	skipped := false
	size := s.elemsize
	if s.state.get() == mSpanManual && size == 0 {
		// We're printing something from a stack frame. We
		// don't know how big it is, so just show up to an
		// including off.
		size = off + goarch.PtrSize
	}
	for i := uintptr(0); i < size; i += goarch.PtrSize {
		// For big objects, just print the beginning (because
		// that usually hints at the object's type) and the
		// fields around off.
		if !(i < 128*goarch.PtrSize || off-16*goarch.PtrSize < i && i < off+16*goarch.PtrSize) {
			skipped = true
			continue
		}
		if skipped {
			print(" ...\n")
			skipped = false
		}
		print(" *(", label, "+", i, ") = ", hex(*(*uintptr)(unsafe.Pointer(obj + i))))
		if i == off {
			print(" <==")
		}
		print("\n")
	}
	if skipped {
		print(" ...\n")
	}
}

// gcmarknewobject marks a newly allocated object black. obj must
// not contain any non-nil pointers.
//
// This is nosplit so it can manipulate a gcWork without preemption.
//
//go:nowritebarrier
//go:nosplit
func gcmarknewobject(span *mspan, obj uintptr) {
	if useCheckmark { // The world should be stopped so this should not happen.
		throw("gcmarknewobject called while doing checkmark")
	}
	if gcphase == _GCmarktermination {
		// Check this here instead of on the hot path.
		throw("mallocgc called with gcphase == _GCmarktermination")
	}

	// Mark object.
	objIndex := span.objIndex(obj)
	span.markBitsForIndex(objIndex).setMarked()
	if goexperiment.GreenTeaGC && gcUsesSpanInlineMarkBits(span.elemsize) {
		// No need to scan the new object.
		span.scannedBitsForIndex(objIndex).setMarked()
	}

	// Mark span.
	arena, pageIdx, pageMask := pageIndexOf(span.base())
	if arena.pageMarks[pageIdx]&pageMask == 0 {
		atomic.Or8(&arena.pageMarks[pageIdx], pageMask)
	}

	gcw := &getg().m.p.ptr().gcw
	gcw.bytesMarked += uint64(span.elemsize)
}

// gcMarkTinyAllocs greys all active tiny alloc blocks.
//
// The world must be stopped.
func gcMarkTinyAllocs() {
	assertWorldStopped()

	for _, p := range allp {
		c := p.mcache
		if c == nil || c.tiny == 0 {
			continue
		}
		gcw := &p.gcw
		if !tryDeferToSpanScan(c.tiny, gcw) {
			_, span, objIndex := findObject(c.tiny, 0, 0)
			greyobject(c.tiny, 0, 0, span, gcw, objIndex)
		}
	}
}