clockintr_cpu_init: stagger clock interrupts by MAXCPUS

During clockintr_cpu_init(), we can't stagger by ncpus because not
every platform has fully incremented it yet.  Instead, stagger by
MAXCPUS.  The resulting intervals are smaller, but are probably still
sufficiently large to avoid aggravating lock contention, even on
platforms where MAXCPUS is large.

While here, don't bother staggering the statclock if it is randomized.

With input from claudio@.

1 files changed, 15 insertions, 8 deletions

diff --git a/sys/kern/kern_clockintr.c b/sys/kern/kern_clockintr.c
index 21fad46338b..187898caf07 100644
--- a/sys/kern/kern_clockintr.c
+++ b/sys/kern/kern_clockintr.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: kern_clockintr.c,v 1.25 2023/06/22 16:23:50 cheloha Exp $ */
+/* $OpenBSD: kern_clockintr.c,v 1.26 2023/07/02 00:55:18 cheloha Exp $ */
 /*
  * Copyright (c) 2003 Dale Rahn <drahn@openbsd.org>
  * Copyright (c) 2020 Mark Kettenis <kettenis@openbsd.org>
@@ -107,7 +107,7 @@ clockintr_init(u_int flags)
 void
 clockintr_cpu_init(const struct intrclock *ic)
 {
-	uint64_t multiplier = 0, offset;
+	uint64_t multiplier = 0;
 	struct cpu_info *ci = curcpu();
 	struct clockintr_queue *cq = &ci->ci_queue;
 	int reset_cq_intrclock = 0;
@@ -170,21 +170,28 @@ clockintr_cpu_init(const struct intrclock *ic)
 			clockintr_advance(cq->cq_hardclock, hardclock_period);
 	} else {
 		if (cq->cq_hardclock->cl_expiration == 0) {
-			offset = hardclock_period / ncpus * multiplier;
-			cq->cq_hardclock->cl_expiration =  offset;
+			clockintr_stagger(cq->cq_hardclock, hardclock_period,
+			     multiplier, MAXCPUS);
 		}
 		clockintr_advance(cq->cq_hardclock, hardclock_period);
 	}
 
 	/*
 	 * We can always advance the statclock and schedclock.
+	 * There is no reason to stagger a randomized statclock.
 	 */
-	offset = statclock_avg / ncpus * multiplier;
-	clockintr_schedule(cq->cq_statclock, offset);
+	if (!ISSET(clockintr_flags, CL_RNDSTAT)) {
+		if (cq->cq_statclock->cl_expiration == 0) {
+			clockintr_stagger(cq->cq_statclock, statclock_avg,
+			    multiplier, MAXCPUS);
+		}
+	}
 	clockintr_advance(cq->cq_statclock, statclock_avg);
 	if (schedhz != 0) {
-		offset = schedclock_period / ncpus * multiplier;
-		clockintr_schedule(cq->cq_schedclock, offset);
+		if (cq->cq_schedclock->cl_expiration == 0) {
+			clockintr_stagger(cq->cq_schedclock, schedclock_period,
+			    multiplier, MAXCPUS);
+		}
 		clockintr_advance(cq->cq_schedclock, schedclock_period);
 	}