scx_prev: a simple scheduler tested on OLTP workloads

scheds/c/README.md

@@ -164,6 +164,27 @@ simple scheduling policy.
 
 --------------------------------------------------------------------------------
 
+## scx_prev
+
+### Overview
+
+A variation on scx_simple with CPU selection that prioritizes an idle previous
+CPU over finding a fully idle core (as is done in scx_simple and scx_rusty).
+
+### Typical Use Case
+
+This scheduler outperforms the in-kernel fair class, scx_simple, and scx_rusty
+on OLTP workloads run on systems with simple topology (i.e. non-NUMA, single
+LLC).
+
+### Production Ready?
+
+scx_prev has not been tested in a production environment, but given its
+similarity to scx_simple, it might be production ready for specific workloads
+on hardware with simple topology.
+
+--------------------------------------------------------------------------------
+
 ## scx_userland
 
 ### Overview

scheds/c/meson.build

@@ -1,5 +1,5 @@
 c_scheds = ['scx_simple', 'scx_qmap', 'scx_central', 'scx_userland', 'scx_nest',
-            'scx_flatcg', 'scx_pair']
+            'scx_flatcg', 'scx_pair', 'scx_prev']
 
 c_scheds_lib = ['scx_sdt']
 

scheds/c/scx_prev.bpf.c

@@ -0,0 +1,79 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * A variation on scx_simple with CPU selection that prioritizes an idle
+ * previous CPU over finding a fully idle core (as is done in scx_simple and
+ * scx_rusty).
+ *
+ * Outperforms the in-kernel fair class (v6.12), scx_simple, and scx_rusty on
+ * OLTP workloads run on systems with simple topology (i.e. non-NUMA, single
+ * LLC).
+ *
+ * Copyright (c) 2025, Oracle and/or its affiliates.
+ * Copyright (c) 2025, Daniel Jordan <[email protected]>
+ */
+#include <scx/common.bpf.h>
+
+char _license[] SEC("license") = "GPL";
+
+UEI_DEFINE(uei);
+
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
+	__uint(key_size, sizeof(u32));
+	__uint(value_size, sizeof(u64));
+	__uint(max_entries, 4);			/* [local, select_fail, prev_cpu, idle_cpu] */
+} stats SEC(".maps");
+
+static void stat_inc(u32 idx)
+{
+	u64 *cnt_p = bpf_map_lookup_elem(&stats, &idx);
+	if (cnt_p)
+		(*cnt_p)++;
+}
+
+s32 BPF_STRUCT_OPS(prev_select_cpu, struct task_struct *p, s32 prev_cpu, u64 wake_flags)
+{
+	s32 cpu;
+
+	if (p->nr_cpus_allowed == 1) {
+		if (scx_bpf_test_and_clear_cpu_idle(prev_cpu)) {
+			cpu = prev_cpu;
+			goto insert;
+		} else {
+			return prev_cpu;
+		}
+	}
+
+	if (scx_bpf_test_and_clear_cpu_idle(prev_cpu)) {
+		stat_inc(2);	/* prev_cpu */
+		cpu = prev_cpu;
+		goto insert;
+	}
+
+	cpu = scx_bpf_pick_idle_cpu(p->cpus_ptr, 0);
+	if (cpu >= 0) {
+		stat_inc(3);	/* idle_cpu */
+		goto insert;
+	}
+
+	stat_inc(1);		/* select_fail */
+
+	return prev_cpu;
+
+insert:
+	stat_inc(0);		/* local */
+	scx_bpf_dsq_insert(p, SCX_DSQ_LOCAL, SCX_SLICE_DFL, 0);
+
+	return cpu;
+}
+
+void BPF_STRUCT_OPS(prev_exit, struct scx_exit_info *ei)
+{
+	UEI_RECORD(uei, ei);
+}
+
+SCX_OPS_DEFINE(prev_ops,
+	.select_cpu		= (void *)prev_select_cpu,
+	.exit			= (void *)prev_exit,
+	.name			= "prev"
+);

scheds/c/scx_prev.c

@@ -0,0 +1,112 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (c) 2025, Oracle and/or its affiliates.
+ * Copyright (c) 2025, Daniel Jordan <[email protected]>
+ */
+#include <stdio.h>
+#include <unistd.h>
+#include <signal.h>
+#include <libgen.h>
+#include <bpf/bpf.h>
+#include <scx/common.h>
+
+#include "scx_prev.bpf.skel.h"
+
+const char help_fmt[] =
+"A variation on scx_simple with CPU selection that prioritizes an idle\n"
+"previous CPU over finding a fully idle core.\n"
+"\n"
+"See the top-level comment in .bpf.c for more details.\n"
+"\n"
+"Usage: %s [-i sec] [-v]\n"
+"\n"
+"  -h            Display this help and exit\n"
+"  -i            Sampling interval for statistics in seconds\n"
+"  -v            Print libbpf debug messages\n";
+
+static bool verbose;
+static unsigned stat_interval = 1;
+static volatile int exit_req;
+
+static int libbpf_print_fn(enum libbpf_print_level level, const char *format, va_list args)
+{
+	if (level == LIBBPF_DEBUG && !verbose)
+		return 0;
+	return vfprintf(stderr, format, args);
+}
+
+static void sigint_handler(int unused)
+{
+	exit_req = 1;
+}
+
+static void read_stats(struct scx_prev *skel, __u64 *stats)
+{
+	int nr_cpus = libbpf_num_possible_cpus();
+	__u64 cnts[4][nr_cpus];
+	__u32 idx;
+
+	memset(stats, 0, sizeof(stats[0]) * 4);
+
+	for (idx = 0; idx < 4; idx++) {
+		int ret, cpu;
+
+		ret = bpf_map_lookup_elem(bpf_map__fd(skel->maps.stats),
+					  &idx, cnts[idx]);
+		if (ret < 0)
+			continue;
+		for (cpu = 0; cpu < nr_cpus; cpu++)
+			stats[idx] += cnts[idx][cpu];
+	}
+}
+
+int main(int argc, char **argv)
+{
+	struct scx_prev *skel;
+	struct bpf_link *link;
+	__u32 opt;
+	__u64 ecode;
+
+	libbpf_set_print(libbpf_print_fn);
+	signal(SIGINT, sigint_handler);
+	signal(SIGTERM, sigint_handler);
+restart:
+	skel = SCX_OPS_OPEN(prev_ops, scx_prev);
+
+	while ((opt = getopt(argc, argv, "hi:v")) != -1) {
+		switch (opt) {
+		case 'i':
+			stat_interval = strtoull(optarg, NULL, 0);
+			if (!stat_interval)
+				stat_interval = 1;
+			break;
+		case 'v':
+			verbose = true;
+			break;
+		default:
+			fprintf(stderr, help_fmt, basename(argv[0]));
+			return opt != 'h';
+		}
+	}
+
+	SCX_OPS_LOAD(skel, prev_ops, scx_prev, uei);
+	link = SCX_OPS_ATTACH(skel, prev_ops, scx_prev);
+
+	while (!exit_req && !UEI_EXITED(skel, uei)) {
+		__u64 stats[4];
+
+		read_stats(skel, stats);
+		printf("local=%llu select_fail=%llu prev_cpu=%llu idle_cpu=%llu\n",
+		       stats[0], stats[1], stats[2], stats[3]);
+		fflush(stdout);
+		sleep(stat_interval);
+	}
+
+	bpf_link__destroy(link);
+	ecode = UEI_REPORT(skel, uei);
+	scx_prev__destroy(skel);
+
+	if (UEI_ECODE_RESTART(ecode))
+		goto restart;
+	return 0;
+}