1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Copyright (c) 2022 Meta Platforms, Inc. and affiliates.
4 * Copyright (c) 2022 Tejun Heo <tj@kernel.org>
5 * Copyright (c) 2022 David Vernet <dvernet@meta.com>
6 */
7 #include <stdio.h>
8 #include <unistd.h>
9 #include <inttypes.h>
10 #include <signal.h>
11 #include <assert.h>
12 #include <libgen.h>
13 #include <bpf/bpf.h>
14 #include <scx/common.h>
15 #include "scx_pair.h"
16 #include "scx_pair.bpf.skel.h"
17
18 const char help_fmt[] =
19 "A demo sched_ext core-scheduler which always makes every sibling CPU pair\n"
20 "execute from the same CPU cgroup.\n"
21 "\n"
22 "See the top-level comment in .bpf.c for more details.\n"
23 "\n"
24 "Usage: %s [-S STRIDE]\n"
25 "\n"
26 " -S STRIDE Override CPU pair stride (default: nr_cpus_ids / 2)\n"
27 " -v Print libbpf debug messages\n"
28 " -h Display this help and exit\n";
29
30 static bool verbose;
31 static volatile int exit_req;
32
libbpf_print_fn(enum libbpf_print_level level,const char * format,va_list args)33 static int libbpf_print_fn(enum libbpf_print_level level, const char *format, va_list args)
34 {
35 if (level == LIBBPF_DEBUG && !verbose)
36 return 0;
37 return vfprintf(stderr, format, args);
38 }
39
sigint_handler(int dummy)40 static void sigint_handler(int dummy)
41 {
42 exit_req = 1;
43 }
44
main(int argc,char ** argv)45 int main(int argc, char **argv)
46 {
47 struct scx_pair *skel;
48 struct bpf_link *link;
49 __u64 seq = 0, ecode;
50 __s32 stride, i, opt, outer_fd;
51
52 libbpf_set_print(libbpf_print_fn);
53 signal(SIGINT, sigint_handler);
54 signal(SIGTERM, sigint_handler);
55 restart:
56 optind = 1;
57 skel = SCX_OPS_OPEN(pair_ops, scx_pair);
58
59 skel->rodata->nr_cpu_ids = libbpf_num_possible_cpus();
60 skel->rodata->pair_batch_dur_ns = __COMPAT_ENUM_OR_ZERO("scx_public_consts", "SCX_SLICE_DFL");
61
62 /* pair up the earlier half to the latter by default, override with -s */
63 stride = skel->rodata->nr_cpu_ids / 2;
64
65 while ((opt = getopt(argc, argv, "S:vh")) != -1) {
66 switch (opt) {
67 case 'S':
68 stride = strtoul(optarg, NULL, 0);
69 break;
70 case 'v':
71 verbose = true;
72 break;
73 default:
74 fprintf(stderr, help_fmt, basename(argv[0]));
75 return opt != 'h';
76 }
77 }
78
79 /* Stride must be positive to pair distinct CPUs. */
80 if (stride <= 0) {
81 fprintf(stderr, "Invalid stride %d, must be positive\n", stride);
82 scx_pair__destroy(skel);
83 return -1;
84 }
85 bpf_map__set_max_entries(skel->maps.pair_ctx, skel->rodata->nr_cpu_ids / 2);
86
87 /* Resize arrays so their element count is equal to cpu count. */
88 RESIZE_ARRAY(skel, rodata, pair_cpu, skel->rodata->nr_cpu_ids);
89 RESIZE_ARRAY(skel, rodata, pair_id, skel->rodata->nr_cpu_ids);
90 RESIZE_ARRAY(skel, rodata, in_pair_idx, skel->rodata->nr_cpu_ids);
91
92 for (i = 0; i < skel->rodata->nr_cpu_ids; i++)
93 skel->rodata_pair_cpu->pair_cpu[i] = -1;
94
95 printf("Pairs: ");
96 for (i = 0; i < skel->rodata->nr_cpu_ids; i++) {
97 int j = (i + stride) % skel->rodata->nr_cpu_ids;
98
99 if (skel->rodata_pair_cpu->pair_cpu[i] >= 0)
100 continue;
101
102 SCX_BUG_ON(i == j,
103 "Invalid stride %d - CPU%d wants to be its own pair",
104 stride, i);
105
106 SCX_BUG_ON(skel->rodata_pair_cpu->pair_cpu[j] >= 0,
107 "Invalid stride %d - three CPUs (%d, %d, %d) want to be a pair",
108 stride, i, j, skel->rodata_pair_cpu->pair_cpu[j]);
109
110 skel->rodata_pair_cpu->pair_cpu[i] = j;
111 skel->rodata_pair_cpu->pair_cpu[j] = i;
112 skel->rodata_pair_id->pair_id[i] = i;
113 skel->rodata_pair_id->pair_id[j] = i;
114 skel->rodata_in_pair_idx->in_pair_idx[i] = 0;
115 skel->rodata_in_pair_idx->in_pair_idx[j] = 1;
116
117 printf("[%d, %d] ", i, j);
118 }
119 printf("\n");
120
121 SCX_OPS_LOAD(skel, pair_ops, scx_pair, uei);
122
123 /*
124 * Populate the cgrp_q_arr map which is an array containing per-cgroup
125 * queues. It'd probably be better to do this from BPF but there are too
126 * many to initialize statically and there's no way to dynamically
127 * populate from BPF.
128 */
129 outer_fd = bpf_map__fd(skel->maps.cgrp_q_arr);
130 SCX_BUG_ON(outer_fd < 0, "Failed to get outer_fd: %d", outer_fd);
131
132 printf("Initializing");
133 for (i = 0; i < MAX_CGRPS; i++) {
134 __s32 inner_fd;
135
136 if (exit_req)
137 break;
138
139 inner_fd = bpf_map_create(BPF_MAP_TYPE_QUEUE, NULL, 0,
140 sizeof(__u32), MAX_QUEUED, NULL);
141 SCX_BUG_ON(inner_fd < 0, "Failed to get inner_fd: %d",
142 inner_fd);
143 SCX_BUG_ON(bpf_map_update_elem(outer_fd, &i, &inner_fd, BPF_ANY),
144 "Failed to set inner map");
145 close(inner_fd);
146
147 if (!(i % 10))
148 printf(".");
149 fflush(stdout);
150 }
151 printf("\n");
152
153 /*
154 * Fully initialized, attach and run.
155 */
156 link = SCX_OPS_ATTACH(skel, pair_ops, scx_pair);
157
158 while (!exit_req && !UEI_EXITED(skel, uei)) {
159 printf("[SEQ %llu]\n", seq++);
160 printf(" total:%10" PRIu64 " dispatch:%10" PRIu64 " missing:%10" PRIu64 "\n",
161 skel->bss->nr_total,
162 skel->bss->nr_dispatched,
163 skel->bss->nr_missing);
164 printf(" kicks:%10" PRIu64 " preemptions:%7" PRIu64 "\n",
165 skel->bss->nr_kicks,
166 skel->bss->nr_preemptions);
167 printf(" exp:%10" PRIu64 " exp_wait:%10" PRIu64 " exp_empty:%10" PRIu64 "\n",
168 skel->bss->nr_exps,
169 skel->bss->nr_exp_waits,
170 skel->bss->nr_exp_empty);
171 printf("cgnext:%10" PRIu64 " cgcoll:%10" PRIu64 " cgempty:%10" PRIu64 "\n",
172 skel->bss->nr_cgrp_next,
173 skel->bss->nr_cgrp_coll,
174 skel->bss->nr_cgrp_empty);
175 fflush(stdout);
176 sleep(1);
177 }
178
179 bpf_link__destroy(link);
180 ecode = UEI_REPORT(skel, uei);
181 scx_pair__destroy(skel);
182
183 if (UEI_ECODE_RESTART(ecode))
184 goto restart;
185 return 0;
186 }
187