`krt` provides a framework for building _declarative_ controllers.
See the [design doc](https://docs.google.com/document/d/1-ywpCnOfubqg7WAXSPf4YgbaFDBEU9HIqMWcxZLhzwE/edit#heading=h.ffjmk8byb9gt) and [KubeCon talk](https://sched.co/1R2oY) for more background.

The framework aims to solve a few problems with writing controllers:

				{"sync.(*Cond).Signal", 0},
				{"main.main", 0},
			}},
			{trace.EventStateTransition, "Goroutine Running->Waiting", []frame{
				{"time.Sleep", 0},
				{"main.main", 0},
			}},
			{trace.EventMetric, "/sched/gomaxprocs:threads", []frame{
				{"runtime.startTheWorld", 0}, // this is when the current gomaxprocs is logged.
				{"runtime.startTheWorldGC", 0},
				{"runtime.GOMAXPROCS", 0},
				{"main.main", 0},
			}},
		}

//go:build linux

package syscall

import (
	"internal/itoa"
	"runtime"
	"unsafe"
)

// Linux unshare/clone/clone2/clone3 flags, architecture-independent,
// copied from linux/sched.h.
const (
	CLONE_VM             = 0x00000100 // set if VM shared between processes
	CLONE_FS             = 0x00000200 // set if fs info shared between processes

func getg() *g

// mcall switches from the g to the g0 stack and invokes fn(g),
// where g is the goroutine that made the call.
// mcall saves g's current PC/SP in g->sched so that it can be restored later.
// It is up to fn to arrange for that later execution, typically by recording
// g in a data structure, causing something to call ready(g) later.

		# MB stacks  estimated scannable stack size, or /gc/scan/stack:bytes
		# 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

				// saveblockevent)
				mp.profStack[0] -= 1
			}
			nstk += fpTracebackPCs(unsafe.Pointer(getfp()), mp.profStack[1:])
		} else {
			mp.profStack[1] = gp.m.curg.sched.pc
			nstk += 1 + fpTracebackPCs(unsafe.Pointer(gp.m.curg.sched.bp), mp.profStack[2:])
		}
	}

	saveBlockEventStack(cycles, rate, mp.profStack[:nstk], which)
	releasem(mp)
}

	for i := range events {
		ev := &events[i]
		lastEv = ev

		// Process the event.
		switch ev.Kind() {
		case EventSync:
			seenSync = true
		case EventMetric:
			m := ev.Metric()
			if m.Name != "/sched/gomaxprocs:threads" {
				break
			}
			gomaxprocs := int(m.Value.Uint64())
			if len(ps) > gomaxprocs {
				if flags&UtilPerProc != 0 {
					// End each P's series.
					for _, p := range ps[gomaxprocs:] {

//go:nowritebarrier
func (c *gcControllerState) enlistWorker() {
	// If there are idle Ps, wake one so it will run an idle worker.
	// NOTE: This is suspected of causing deadlocks. See golang.org/issue/19112.
	//
	//	if sched.npidle.Load() != 0 && sched.nmspinning.Load() == 0 {
	//		wakep()
	//		return
	//	}

	// There are no idle Ps. If we need more dedicated workers,
	// try to preempt a running P so it will switch to a worker.

func (e Event) Metric() Metric {
	if e.Kind() != EventMetric {
		panic("Metric called on non-Metric event")
	}
	var m Metric
	switch e.base.typ {
	case go122.EvProcsChange:
		m.Name = "/sched/gomaxprocs:threads"
		m.Value = Value{kind: ValueUint64, scalar: e.base.args[0]}
	case go122.EvHeapAlloc:
		m.Name = "/memory/classes/heap/objects:bytes"
		m.Value = Value{kind: ValueUint64, scalar: e.base.args[0]}

		KeepAlive(ep)
		return
	}

	// stack
	if gp := getg(); gp.m.curg.stack.lo <= uintptr(p) && uintptr(p) < gp.m.curg.stack.hi {
		found := false
		var u unwinder
		for u.initAt(gp.m.curg.sched.pc, gp.m.curg.sched.sp, 0, gp.m.curg, 0); u.valid(); u.next() {
			if u.frame.sp <= uintptr(p) && uintptr(p) < u.frame.varp {
				found = true
				break
			}
		}
		if found {