From 834cc80d160ab56359e2fd98c74d60fe4208d67d Mon Sep 17 00:00:00 2001
From: claudio <claudio@openbsd.org>
Date: Thu, 3 Aug 2023 16:12:08 +0000
Subject: [PATCH] Remove the per-cpu loadavg calculation. The current scheduler
 useage is highly questionable and probably not helpful. OK kettenis@ cheloha@
 deraadt@

---
 sys/kern/kern_sched.c | 13 ++-----------
 sys/sys/sched.h       |  3 +--
 sys/uvm/uvm_meter.c   | 23 ++++++-----------------
 3 files changed, 9 insertions(+), 30 deletions(-)

diff --git a/sys/kern/kern_sched.c b/sys/kern/kern_sched.c
index c94438ecf09..b6d1ea68431 100644
--- a/sys/kern/kern_sched.c
+++ b/sys/kern/kern_sched.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sched.c,v 1.81 2023/07/27 17:52:53 cheloha Exp $	*/
+/*	$OpenBSD: kern_sched.c,v 1.82 2023/08/03 16:12:08 claudio Exp $	*/
 /*
  * Copyright (c) 2007, 2008 Artur Grabowski <art@openbsd.org>
  *
@@ -373,7 +373,6 @@ sched_choosecpu_fork(struct proc *parent, int flags)
 {
 #ifdef MULTIPROCESSOR
 	struct cpu_info *choice = NULL;
-	fixpt_t load, best_load = ~0;
 	int run, best_run = INT_MAX;
 	struct cpu_info *ci;
 	struct cpuset set;
@@ -407,13 +406,10 @@ sched_choosecpu_fork(struct proc *parent, int flags)
 	while ((ci = cpuset_first(&set)) != NULL) {
 		cpuset_del(&set, ci);
 
-		load = ci->ci_schedstate.spc_ldavg;
 		run = ci->ci_schedstate.spc_nrun;
 
-		if (choice == NULL || run < best_run ||
-		    (run == best_run &&load < best_load)) {
+		if (choice == NULL || run < best_run) {
 			choice = ci;
-			best_load = load;
 			best_run = run;
 		}
 	}
@@ -606,11 +602,6 @@ sched_proc_to_cpu_cost(struct cpu_info *ci, struct proc *p)
 	if (CPU_IS_PRIMARY(ci))
 		cost += sched_cost_runnable;
 
-	/*
-	 * Higher load on the destination means we don't want to go there.
-	 */
-	cost += ((sched_cost_load * spc->spc_ldavg) >> FSHIFT);
-
 	/*
 	 * If the proc is on this cpu already, lower the cost by how much
 	 * it has been running and an estimate of its footprint.
diff --git a/sys/sys/sched.h b/sys/sys/sched.h
index 8673205dc7c..c53b0843120 100644
--- a/sys/sys/sched.h
+++ b/sys/sys/sched.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sched.h,v 1.58 2023/07/25 18:16:19 cheloha Exp $	*/
+/*	$OpenBSD: sched.h,v 1.59 2023/08/03 16:12:08 claudio Exp $	*/
 /* $NetBSD: sched.h,v 1.2 1999/02/28 18:14:58 ross Exp $ */
 
 /*-
@@ -110,7 +110,6 @@ struct schedstate_percpu {
 	struct clockintr *spc_profclock; /* [o] profclock handle */
 
 	u_int spc_nrun;			/* procs on the run queues */
-	fixpt_t spc_ldavg;		/* shortest load avg. for this cpu */
 
 	volatile uint32_t spc_whichqs;
 	volatile u_int spc_spinning;	/* this cpu is currently spinning */
diff --git a/sys/uvm/uvm_meter.c b/sys/uvm/uvm_meter.c
index da3908233b8..e0c5d41ae8d 100644
--- a/sys/uvm/uvm_meter.c
+++ b/sys/uvm/uvm_meter.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_meter.c,v 1.47 2023/08/03 16:08:12 claudio Exp $	*/
+/*	$OpenBSD: uvm_meter.c,v 1.48 2023/08/03 16:12:08 claudio Exp $	*/
 /*	$NetBSD: uvm_meter.c,v 1.21 2001/07/14 06:36:03 matt Exp $	*/
 
 /*
@@ -98,27 +98,16 @@ uvm_meter(void)
 static void
 uvm_loadav(struct loadavg *avg)
 {
+	extern struct cpuset sched_idle_cpus;
 	CPU_INFO_ITERATOR cii;
 	struct cpu_info *ci;
-	struct schedstate_percpu *spc;
-	u_int i, nrun = 0, nrun_cpu;
-	int s;
+	u_int i, nrun = 0;
 
-
-	SCHED_LOCK(s);
 	CPU_INFO_FOREACH(cii, ci) {
-		spc = &ci->ci_schedstate;
-		nrun_cpu = spc->spc_nrun;
-		if (ci->ci_curproc != spc->spc_idleproc)
-			nrun_cpu++;
-		if (nrun_cpu == 0)
-			continue;
-		spc->spc_ldavg = (cexp[0] * spc->spc_ldavg +
-		    nrun_cpu * FSCALE *
-		    (FSCALE - cexp[0])) >> FSHIFT;
-		nrun += nrun_cpu;
+		if (!cpuset_isset(&sched_idle_cpus, ci))
+			nrun++;
+		nrun += ci->ci_schedstate.spc_nrun;
 	}
-	SCHED_UNLOCK(s);
 
 	for (i = 0; i < 3; i++) {
 		avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
-- 
2.20.1