runtime: write much more direct test for semaphore waiter scalability

This test originally existed as two tests in test/locklinear.go, but
this checked against actual locks and was flaky. The test was checking
a property of a deep part of the runtime but from a much higher level,
and it's easy for nondeterminism due to scheduling to completely mess
that up, especially on an oversubscribed system.

That test was then moved to the sync package with a more rigorous
testing methodology, but it could still flake pretty easily.

Finally, this CL makes semtable more testable, exports it in
export_test.go, then writes a very direct scalability test for exactly
the situation the original test described. As far as I can tell, this is
much, much more stable, because it's single-threaded and is just
checking exactly the algorithm we need to check.

Don't bother trying to bring in a test that checks for O(log n) behavior
on the other kind of iteration. It'll be perpetually flaky because the
underlying data structure is a treap, so it's only _expected_ to be
O(log n), but it's very easy for it to get unlucky without a large
number of iterations that's too much for a simple test.

Fixes #53381.

Change-Id: Ia1cd2d2b0e36d552d5a8ae137077260a16016602
Reviewed-on: https://go-review.googlesource.com/c/go/+/412875
Reviewed-by: Michael Pratt <mpratt@google.com>

src/runtime/export_test.go

@@ -1163,10 +1163,38 @@ var Semacquire = semacquire
 var Semrelease1 = semrelease1
 
 func SemNwait(addr *uint32) uint32 {
-	root := semroot(addr)
+	root := semtable.rootFor(addr)
 	return atomic.Load(&root.nwait)
 }
 
+const SemTableSize = semTabSize
+
+// SemTable is a wrapper around semTable exported for testing.
+type SemTable struct {
+	semTable
+}
+
+// Enqueue simulates enqueuing a waiter for a semaphore (or lock) at addr.
+func (t *SemTable) Enqueue(addr *uint32) {
+	s := acquireSudog()
+	s.releasetime = 0
+	s.acquiretime = 0
+	s.ticket = 0
+	t.semTable.rootFor(addr).queue(addr, s, false)
+}
+
+// Dequeue simulates dequeuing a waiter for a semaphore (or lock) at addr.
+//
+// Returns true if there actually was a waiter to be dequeued.
+func (t *SemTable) Dequeue(addr *uint32) bool {
+	s, _ := t.semTable.rootFor(addr).dequeue(addr)
+	if s != nil {
+		releaseSudog(s)
+		return true
+	}
+	return false
+}
+
 // mspan wrapper for testing.
 //
 //go:notinheap

src/runtime/sema.go

@@ -35,22 +35,28 @@ import (
 // where n is the number of distinct addresses with goroutines blocked
 // on them that hash to the given semaRoot.
 // See golang.org/issue/17953 for a program that worked badly
-// before we introduced the second level of list, and test/locklinear.go
+// before we introduced the second level of list, and TestSemTableOneAddrCollisionLinear
 // for a test that exercises this.
 type semaRoot struct {
 	lock  mutex
 	treap *sudog // root of balanced tree of unique waiters.
 	nwait uint32 // Number of waiters. Read w/o the lock.
 }
 
+var semtable semTable
+
 // Prime to not correlate with any user patterns.
 const semTabSize = 251
 
-var semtable [semTabSize]struct {
+type semTable [semTabSize]struct {
 	root semaRoot
 	pad  [cpu.CacheLinePadSize - unsafe.Sizeof(semaRoot{})]byte
 }
 
+func (t *semTable) rootFor(addr *uint32) *semaRoot {
+	return &t[(uintptr(unsafe.Pointer(addr))>>3)%semTabSize].root
+}
+
 //go:linkname sync_runtime_Semacquire sync.runtime_Semacquire
 func sync_runtime_Semacquire(addr *uint32) {
 	semacquire1(addr, false, semaBlockProfile, 0)
@@ -113,7 +119,7 @@ func semacquire1(addr *uint32, lifo bool, profile semaProfileFlags, skipframes i
 	//	sleep
 	//	(waiter descriptor is dequeued by signaler)
 	s := acquireSudog()
-	root := semroot(addr)
+	root := semtable.rootFor(addr)
 	t0 := int64(0)
 	s.releasetime = 0
 	s.acquiretime = 0
@@ -157,7 +163,7 @@ func semrelease(addr *uint32) {
 }
 
 func semrelease1(addr *uint32, handoff bool, skipframes int) {
-	root := semroot(addr)
+	root := semtable.rootFor(addr)
 	atomic.Xadd(addr, 1)
 
 	// Easy case: no waiters?
@@ -214,10 +220,6 @@ func semrelease1(addr *uint32, handoff bool, skipframes int) {
 	}
 }
 
-func semroot(addr *uint32) *semaRoot {
-	return &semtable[(uintptr(unsafe.Pointer(addr))>>3)%semTabSize].root
-}
-
 func cansemacquire(addr *uint32) bool {
 	for {
 		v := atomic.Load(addr)

src/runtime/sema_test.go

@@ -5,6 +5,7 @@
 package runtime_test
 
 import (
+	"internal/testenv"
 	. "runtime"
 	"sync"
 	"sync/atomic"
@@ -101,3 +102,46 @@ func testSemaHandoff() bool {
 
 	return res == 1 // did the waiter run first?
 }
+
+func TestSemTableOneAddrCollisionLinear(t *testing.T) {
+	testenv.CheckLinear(t, func(scale float64) func(*testing.B) {
+		n := int(1000 * scale)
+		return func(b *testing.B) {
+			tab := Escape(new(SemTable))
+			u := make([]uint32, SemTableSize+1)
+
+			b.ResetTimer()
+
+			// Simulate two locks colliding on the same semaRoot.
+			//
+			// Specifically enqueue all the waiters for the first lock,
+			// then all the waiters for the second lock.
+			//
+			// Then, dequeue all the waiters from the first lock, then
+			// the second.
+			//
+			// Each enqueue/dequeue operation should be O(1), because
+			// there are exactly 2 locks. This could be O(n) if all
+			// the waiters for both locks are on the same list, as it
+			// once was.
+			for i := 0; i < n; i++ {
+				if i < n/2 {
+					tab.Enqueue(&u[0])
+				} else {
+					tab.Enqueue(&u[SemTableSize])
+				}
+			}
+			for i := 0; i < n; i++ {
+				var ok bool
+				if i < n/2 {
+					ok = tab.Dequeue(&u[0])
+				} else {
+					ok = tab.Dequeue(&u[SemTableSize])
+				}
+				if !ok {
+					b.Fatal("failed to dequeue")
+				}
+			}
+		}
+	})
+}

src/sync/mutex_test.go

@@ -333,115 +333,3 @@ func BenchmarkMutexSpin(b *testing.B) {
 		}
 	})
 }
-
-const runtimeSemaHashTableSize = 251 // known size of runtime hash table
-
-func TestMutexLinearOne(t *testing.T) {
-	testenv.CheckLinear(t, func(scale float64) func(*testing.B) {
-		n := int(1000 * scale)
-		return func(b *testing.B) {
-			ch := make(chan struct{})
-			locks := make([]RWMutex, runtimeSemaHashTableSize+1)
-
-			b.ResetTimer()
-
-			var wgStart, wgFinish WaitGroup
-			for i := 0; i < n; i++ {
-				wgStart.Add(1)
-				wgFinish.Add(1)
-				go func() {
-					wgStart.Done()
-					locks[0].Lock()
-					ch <- struct{}{}
-					wgFinish.Done()
-				}()
-			}
-			wgStart.Wait()
-
-			wgFinish.Add(1)
-			go func() {
-				for j := 0; j < n; j++ {
-					locks[1].Lock()
-					locks[runtimeSemaHashTableSize].Lock()
-					locks[1].Unlock()
-					runtime.Gosched()
-					locks[runtimeSemaHashTableSize].Unlock()
-				}
-				wgFinish.Done()
-			}()
-
-			for j := 0; j < n; j++ {
-				locks[1].Lock()
-				locks[runtimeSemaHashTableSize].Lock()
-				locks[1].Unlock()
-				runtime.Gosched()
-				locks[runtimeSemaHashTableSize].Unlock()
-			}
-
-			b.StopTimer()
-
-			for i := 0; i < n; i++ {
-				<-ch
-				locks[0].Unlock()
-			}
-
-			wgFinish.Wait()
-		}
-	})
-}
-
-func TestMutexLinearMany(t *testing.T) {
-	if runtime.GOARCH == "arm" && os.Getenv("GOARM") == "5" {
-		// stressLockMany reliably fails on the linux-arm-arm5spacemonkey
-		// builder. See https://golang.org/issue/24221.
-		return
-	}
-	testenv.CheckLinear(t, func(scale float64) func(*testing.B) {
-		n := int(1000 * scale)
-		return func(b *testing.B) {
-			locks := make([]RWMutex, n*runtimeSemaHashTableSize+1)
-
-			b.ResetTimer()
-
-			var wgStart, wgFinish WaitGroup
-			for i := 0; i < n; i++ {
-				wgStart.Add(1)
-				wgFinish.Add(1)
-				go func(i int) {
-					locks[(i+1)*runtimeSemaHashTableSize].Lock()
-					wgStart.Done()
-					locks[(i+1)*runtimeSemaHashTableSize].Lock()
-					locks[(i+1)*runtimeSemaHashTableSize].Unlock()
-					wgFinish.Done()
-				}(i)
-			}
-			wgStart.Wait()
-
-			go func() {
-				for j := 0; j < n; j++ {
-					locks[1].Lock()
-					locks[0].Lock()
-					locks[1].Unlock()
-					runtime.Gosched()
-					locks[0].Unlock()
-				}
-			}()
-
-			for j := 0; j < n; j++ {
-				locks[1].Lock()
-				locks[0].Lock()
-				locks[1].Unlock()
-				runtime.Gosched()
-				locks[0].Unlock()
-			}
-
-			b.StopTimer()
-
-			for i := 0; i < n; i++ {
-				locks[(i+1)*runtimeSemaHashTableSize].Unlock()
-			}
-
-			wgFinish.Wait()
-		}
-	})
-}