fix(drm): separate the frame wait reason from the memory-refusal run

commit: 7703bb9e384ac91419bb87626e931a3226b6e64c
author: brice · 11h ago · 2026-10-22 12:04 UTC
parent: d707400
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

2 files · +56 -3

M internal/adapters/drm/frame_lifecycle.go +19 -3


@@ -13,6 +13,7 @@ import (
13	13		// deadline ticks. Only Run uses the timer; the transitions take plain values.
14	14		type frameLifecycle struct {
15	15		pending bool
	16	+	waitKind frameWaitKind
16	17		pendingFrame pendingFrame
17	18		serials *atomic.Uint64 // shared across outputs on the card
18	19		pendingSerial, nextSerial uint64

@@ -27,6 +28,16 @@ type frameLifecycle struct {
27	28		stuckTimer *time.Timer
28	29		}
29	30
	31	+	// frameWaitKind is the current wait, independent of refusal-run summaries.
	32	+	type frameWaitKind uint8
	33	+
	34	+	const (
	35	+	waitNone frameWaitKind = iota
	36	+	waitFlip
	37	+	waitBusy
	38	+	waitNoMem
	39	+	)
	40	+
30	41		// pendingFrame is what the pending frame commit shows.
31	42		type pendingFrame struct {
32	43		security ports.SecurityState // admission epoch checked again at KMS boundary

@@ -94,6 +105,7 @@ func (l *frameLifecycle) userData(kind uint64) uint64 {
94	105		func (l *frameLifecycle) endPending() {
95	106		l.pendingFrame.closeFences()
96	107		l.pending, l.pendingFrame = false, pendingFrame{}
	108	+	l.waitKind = waitNone
97	109		}
98	110
99	111		// begin records a successful nonblocking commit awaiting its event.

@@ -101,6 +113,7 @@ func (l *frameLifecycle) begin(f pendingFrame, now time.Time) {
101	113		l.pendingSerial = l.nextSerial
102	114		l.pendingFrame.closeFences()
103	115		l.pending, l.pendingFrame, l.flipStart = true, f, now
	116	+	l.waitKind = waitFlip
104	117		l.stuckAt, l.fenceWarned = now.Add(l.stuckLimit()), false
105	118		l.busySince = time.Time{}
106	119		}

@@ -126,6 +139,7 @@ func (l *frameLifecycle) nomem(now time.Time) (attempt int, newRun bool) {
126	139		l.lastNomem = now
127	140		l.nomemAttempts++
128	141		l.pending, l.pendingSerial, l.flipStart = true, 0, now
	142	+	l.waitKind = waitNoMem
129	143		l.stuckAt = now.Add(l.nomemBackoff)
130	144		return l.nomemAttempts, newRun
131	145		}

@@ -158,6 +172,7 @@ func (l *frameLifecycle) busy(now time.Time) bool {
158	172		l.endPending()
159	173		l.pending, l.flipStart, l.stuckAt = true, now, now.Add(busyRetry)
160	174		l.pendingSerial = 0
	175	+	l.waitKind = waitBusy
161	176		newRun := l.busySince.IsZero() \|\| now.Sub(l.lastBusy) > 4*busyRetry
162	177		if newRun {
163	178		l.busySince = now

@@ -176,7 +191,7 @@ func (l *frameLifecycle) ours(user uint64) bool {
176	191		// EBUSY there is no known serial, so any CRTC event ends that wait. The
177	192		// returned start lets Output measure flip age after pending fences close.
178	193		func (l *frameLifecycle) flip(user uint64) (pendingFrame, time.Time, bool) {
179		-	if !l.pending \|\| l.pendingSerial == 0 && !l.nomemSince.IsZero() \|\| l.pendingSerial != 0 && user>>userKindBits != l.pendingSerial {
	194	+	if !l.pending \|\| l.waitKind == waitNoMem \|\| l.pendingSerial != 0 && user>>userKindBits != l.pendingSerial {
180	195		return pendingFrame{}, time.Time{}, false
181	196		}
182	197		f, start := l.pendingFrame, l.flipStart

@@ -211,11 +226,12 @@ func (l *frameLifecycle) timeout(now time.Time, fencesReady bool) timeoutResult
211	226		}
212	227		return timeoutEarly
213	228		}
	229	+	kind := l.waitKind
214	230		l.endPending()
215		-	if l.pendingSerial != 0 {
	231	+	if kind == waitFlip {
216	232		return timeoutMissing
217	233		}
218		-	if !l.nomemSince.IsZero() {
	234	+	if kind == waitNoMem {
219	235		return timeoutRetry
220	236		}
221	237		if l.busySince.IsZero() \|\| now.Sub(l.busySince) < l.stuckLimit() {

 // deadline ticks. Only Run uses the timer; the transitions take plain values.
 type frameLifecycle struct {
 	pending                   bool
+	waitKind                  frameWaitKind
 	pendingFrame              pendingFrame
 	serials                   *atomic.Uint64 // shared across outputs on the card
 	pendingSerial, nextSerial uint64
 	stuckTimer                *time.Timer
 }
+// frameWaitKind is the current wait, independent of refusal-run summaries.
+type frameWaitKind uint8
+const (
+	waitNone frameWaitKind = iota
+	waitFlip
+	waitBusy
+	waitNoMem
+)
 // pendingFrame is what the pending frame commit shows.
 type pendingFrame struct {
 	security ports.SecurityState // admission epoch checked again at KMS boundary
 func (l *frameLifecycle) endPending() {
 	l.pendingFrame.closeFences()
 	l.pending, l.pendingFrame = false, pendingFrame{}
+	l.waitKind = waitNone
 }
 // begin records a successful nonblocking commit awaiting its event.
 	l.pendingSerial = l.nextSerial
 	l.pendingFrame.closeFences()
 	l.pending, l.pendingFrame, l.flipStart = true, f, now
+	l.waitKind = waitFlip
 	l.stuckAt, l.fenceWarned = now.Add(l.stuckLimit()), false
 	l.busySince = time.Time{}
 }
 	l.lastNomem = now
 	l.nomemAttempts++
 	l.pending, l.pendingSerial, l.flipStart = true, 0, now
+	l.waitKind = waitNoMem
 	l.stuckAt = now.Add(l.nomemBackoff)
 	return l.nomemAttempts, newRun
 }
 	l.endPending()
 	l.pending, l.flipStart, l.stuckAt = true, now, now.Add(busyRetry)
 	l.pendingSerial = 0
+	l.waitKind = waitBusy
 	newRun := l.busySince.IsZero() || now.Sub(l.lastBusy) > 4*busyRetry
 	if newRun {
 		l.busySince = now
 // EBUSY there is no known serial, so any CRTC event ends that wait. The
 // returned start lets Output measure flip age after pending fences close.
 func (l *frameLifecycle) flip(user uint64) (pendingFrame, time.Time, bool) {
-	if !l.pending || l.pendingSerial == 0 && !l.nomemSince.IsZero() || l.pendingSerial != 0 && user>>userKindBits != l.pendingSerial {
+	if !l.pending || l.waitKind == waitNoMem || l.pendingSerial != 0 && user>>userKindBits != l.pendingSerial {
 		return pendingFrame{}, time.Time{}, false
 	}
 	f, start := l.pendingFrame, l.flipStart
 		}
 		return timeoutEarly
 	}
+	kind := l.waitKind
 	l.endPending()
-	if l.pendingSerial != 0 {
+	if kind == waitFlip {
 		return timeoutMissing
 	}
-	if !l.nomemSince.IsZero() {
+	if kind == waitNoMem {
 		return timeoutRetry
 	}
 	if l.busySince.IsZero() || now.Sub(l.busySince) < l.stuckLimit() {

M internal/adapters/drm/frame_lifecycle_test.go +37 -0


@@ -124,3 +124,40 @@ func TestFrameLifecycleNoMemBackoffAndRecovery(t *testing.T) {
124	124		t.Fatal("modeset did not clear run")
125	125		}
126	126		}
	127	+
	128	+	func TestFrameLifecycleNoMemThenBusyEvent(t *testing.T) {
	129	+	l := frameLifecycle{}
	130	+	start := time.Unix(100, 0)
	131	+	l.nomem(start)
	132	+	if got := l.timeout(l.stuckAt, true); got != timeoutRetry {
	133	+	t.Fatalf("memory wait: %v", got)
	134	+	}
	135	+	l.busy(start.Add(nomemRetryMin))
	136	+	if _, _, ok := l.flip(123 << userKindBits); !ok \|\| l.pending \|\| l.waitKind != waitNone {
	137	+	t.Fatal("memory history prevented a CRTC event from ending EBUSY")
	138	+	}
	139	+	if l.nomemAttempts != 1 \|\| l.nomemSince != start {
	140	+	t.Fatal("busy event lost memory recovery summary")
	141	+	}
	142	+	}
	143	+
	144	+	func TestFrameLifecycleNoMemThenBusyLimit(t *testing.T) {
	145	+	l := frameLifecycle{stuckAfter: 200 * time.Millisecond}
	146	+	start := time.Unix(100, 0)
	147	+	l.nomem(start)
	148	+	l.timeout(l.stuckAt, true)
	149	+	busyStart := start.Add(nomemRetryMin)
	150	+	for i := range 4 {
	151	+	l.busy(busyStart.Add(time.Duration(i) * busyRetry))
	152	+	want := timeoutRetry
	153	+	if i == 3 {
	154	+	want = timeoutBusy
	155	+	}
	156	+	if got := l.timeout(l.stuckAt, true); got != want {
	157	+	t.Fatalf("busy attempt %d: got %v want %v", i+1, got, want)
	158	+	}
	159	+	}
	160	+	if l.nomemAttempts != 1 \|\| l.nomemSince != start {
	161	+	t.Fatal("busy timeout lost memory recovery summary")
	162	+	}
	163	+	}


@@ -124,3 +124,40 @@ func TestFrameLifecycleNoMemBackoffAndRecovery(t *testing.T) {
124	t.Fatal("modeset did not clear run")	124		t.Fatal("modeset did not clear run")
125	}	125		}
126	}	126		}
		127	+
		128	+	func TestFrameLifecycleNoMemThenBusyEvent(t *testing.T) {
		129	+	l := frameLifecycle{}
		130	+	start := time.Unix(100, 0)
		131	+	l.nomem(start)
		132	+	if got := l.timeout(l.stuckAt, true); got != timeoutRetry {
		133	+	t.Fatalf("memory wait: %v", got)
		134	+	}
		135	+	l.busy(start.Add(nomemRetryMin))
		136	+	if _, _, ok := l.flip(123 << userKindBits); !ok \|\| l.pending \|\| l.waitKind != waitNone {
		137	+	t.Fatal("memory history prevented a CRTC event from ending EBUSY")
		138	+	}
		139	+	if l.nomemAttempts != 1 \|\| l.nomemSince != start {
		140	+	t.Fatal("busy event lost memory recovery summary")
		141	+	}
		142	+	}
		143	+
		144	+	func TestFrameLifecycleNoMemThenBusyLimit(t *testing.T) {
		145	+	l := frameLifecycle{stuckAfter: 200 * time.Millisecond}
		146	+	start := time.Unix(100, 0)
		147	+	l.nomem(start)
		148	+	l.timeout(l.stuckAt, true)
		149	+	busyStart := start.Add(nomemRetryMin)
		150	+	for i := range 4 {
		151	+	l.busy(busyStart.Add(time.Duration(i) * busyRetry))
		152	+	want := timeoutRetry
		153	+	if i == 3 {
		154	+	want = timeoutBusy
		155	+	}
		156	+	if got := l.timeout(l.stuckAt, true); got != want {
		157	+	t.Fatalf("busy attempt %d: got %v want %v", i+1, got, want)
		158	+	}
		159	+	}
		160	+	if l.nomemAttempts != 1 \|\| l.nomemSince != start {
		161	+	t.Fatal("busy timeout lost memory recovery summary")
		162	+	}
		163	+	}