// system. We never release spine memory because there could be
	// concurrent lock-free access and we're likely to reuse it
	// anyway. (In principle, we could do this during STW.)

	spineLock mutex
	spine     atomicSpanSetSpinePointer // *[N]atomic.Pointer[spanSetBlock]
	spineLen  atomic.Uintptr            // Spine array length

	}
	t.Errorf(`time.Sleep did not contribute enough to "idle" class: minimum idle time = %.5fs`, minIdleCPUSeconds)
}

// Call f() and verify that the correct STW metrics increment. If isGC is true,
// fn triggers a GC STW. Otherwise, fn triggers an other STW.
func testSchedPauseMetrics(t *testing.T, fn func(t *testing.T), isGC bool) {
	m := []metrics.Sample{
		{Name: "/sched/pauses/stopping/gc:seconds"},

	trace.seqGC++
}

// STWStart traces a STWBegin event.
func (tl traceLocker) STWStart(reason stwReason) {
	// Although the current P may be in _Pgcstop here, we model the P as running during the STW. This deviates from the
	// runtime's state tracking, but it's more accurate and doesn't result in any loss of information.

  the locked region, reads do not need to be atomic, but the write
  does. Outside the locked region, reads need to be atomic.

* Reads that only happen during STW, where no writes can happen during
  STW, do not need to be atomic.

That said, the advice from the Go memory model stands: "Don't be
[too] clever." The performance of the runtime matters, but its
robustness matters more.

		# MB globals scannable global size, or /gc/scan/globals:bytes
		# P          number of processors used, or /sched/gomaxprocs:threads
	The phases are stop-the-world (STW) sweep termination, concurrent
	mark and scan, and STW mark termination. The CPU times
	for mark/scan are broken down in to assist time (GC performed in
	line with allocation), background GC time, and idle GC time.

	case LWA, LWAX, LWAUX:
		return true
	case LD, LDU, LDX, LDUX:
		return true
	case LQ:
		return true
	case STB, STBU, STBX, STBUX:
		return true
	case STH, STHU, STHX, STHUX:
		return true
	case STW, STWU, STWX, STWUX:
		return true
	case STD, STDU, STDX, STDUX:
		return true
	case STQ:
		return true
	case LHBRX, LWBRX, STHBRX, STWBRX:
		return true
	case LBARX, LWARX, LHARX, LDARX:
		return true

	MOVWZ (R3+R5), R4		<=>	lwzx r4,r3,r5
	MOVHZ  (R3), R4		<=>	lhz r4,0(r3)
	MOVHU 2(R3), R4		<=>	lhau r4,2(r3)
	MOVBZ (R3), R4		<=>	lbz r4,0(r3)

	MOVD R4,(R3)		<=>	std r4,0(r3)
	MOVW R4,(R3)		<=>	stw r4,0(r3)
	MOVW R4,(R3+R5)		<=>	stwx r4,r3,r5
	MOVWU R4,4(R3)		<=>	stwu r4,4(r3)
	MOVH R4,2(R3)		<=>	sth r4,2(r3)
	MOVBU R4,(R3)(R5)		<=>	stbux r4,r3,r5

4. Compares

}

const (
	// Block reasons
	sForever = iota
	sPreempted
	sGosched
	sSleep
	sChanSend
	sChanRecv
	sNetwork
	sSync
	sSyncCond
	sSelect
	sEmpty
	sMarkAssistWait

	// STW kinds
	sSTWUnknown
	sSTWGCMarkTermination
	sSTWGCSweepTermination
	sSTWWriteHeapDump
	sSTWGoroutineProfile
	sSTWGoroutineProfileCleanup
	sSTWAllGoroutinesStackTrace
	sSTWReadMemStats
	sSTWAllThreadsSyscall

// compute populates the cpuStatsAggregate with values from the runtime.
func (a *cpuStatsAggregate) compute() {
	a.cpuStats = work.cpuStats
	// TODO(mknyszek): Update the CPU stats again so that we're not
	// just relying on the STW snapshot. The issue here is that currently
	// this will cause non-monotonicity in the "user" CPU time metric.
	//
	// a.cpuStats.accumulate(nanotime(), gcphase == _GCmark)
}

			}
			evGC = nil
		case EvSTWStart:
			evp := &evSTW
			if *evp != nil {
				return fmt.Errorf("previous STW is not ended before a new one (time %d)", ev.Ts)
			}
			*evp = ev
		case EvSTWDone:
			evp := &evSTW
			if *evp == nil {
				return fmt.Errorf("bogus STW end (time %d)", ev.Ts)
			}
			*evp = nil
		case EvGCSweepStart:
			p := ps[ev.P]
			if p.evSweep != nil {