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 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 40 static void sigint_handler(int dummy) 41 { 42 exit_req = 1; 43 } 44 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 skel = SCX_OPS_OPEN(pair_ops, scx_pair); 57 58 skel->rodata->nr_cpu_ids = libbpf_num_possible_cpus(); 59 skel->rodata->pair_batch_dur_ns = __COMPAT_ENUM_OR_ZERO("scx_public_consts", "SCX_SLICE_DFL"); 60 61 /* pair up the earlier half to the latter by default, override with -s */ 62 stride = skel->rodata->nr_cpu_ids / 2; 63 64 while ((opt = getopt(argc, argv, "S:vh")) != -1) { 65 switch (opt) { 66 case 'S': 67 stride = strtoul(optarg, NULL, 0); 68 break; 69 case 'v': 70 verbose = true; 71 break; 72 default: 73 fprintf(stderr, help_fmt, basename(argv[0])); 74 return opt != 'h'; 75 } 76 } 77 78 /* Stride must be positive to pair distinct CPUs. */ 79 if (stride <= 0) { 80 fprintf(stderr, "Invalid stride %d, must be positive\n", stride); 81 scx_pair__destroy(skel); 82 return -1; 83 } 84 bpf_map__set_max_entries(skel->maps.pair_ctx, skel->rodata->nr_cpu_ids / 2); 85 86 /* Resize arrays so their element count is equal to cpu count. */ 87 RESIZE_ARRAY(skel, rodata, pair_cpu, skel->rodata->nr_cpu_ids); 88 RESIZE_ARRAY(skel, rodata, pair_id, skel->rodata->nr_cpu_ids); 89 RESIZE_ARRAY(skel, rodata, in_pair_idx, skel->rodata->nr_cpu_ids); 90 91 for (i = 0; i < skel->rodata->nr_cpu_ids; i++) 92 skel->rodata_pair_cpu->pair_cpu[i] = -1; 93 94 printf("Pairs: "); 95 for (i = 0; i < skel->rodata->nr_cpu_ids; i++) { 96 int j = (i + stride) % skel->rodata->nr_cpu_ids; 97 98 if (skel->rodata_pair_cpu->pair_cpu[i] >= 0) 99 continue; 100 101 SCX_BUG_ON(i == j, 102 "Invalid stride %d - CPU%d wants to be its own pair", 103 stride, i); 104 105 SCX_BUG_ON(skel->rodata_pair_cpu->pair_cpu[j] >= 0, 106 "Invalid stride %d - three CPUs (%d, %d, %d) want to be a pair", 107 stride, i, j, skel->rodata_pair_cpu->pair_cpu[j]); 108 109 skel->rodata_pair_cpu->pair_cpu[i] = j; 110 skel->rodata_pair_cpu->pair_cpu[j] = i; 111 skel->rodata_pair_id->pair_id[i] = i; 112 skel->rodata_pair_id->pair_id[j] = i; 113 skel->rodata_in_pair_idx->in_pair_idx[i] = 0; 114 skel->rodata_in_pair_idx->in_pair_idx[j] = 1; 115 116 printf("[%d, %d] ", i, j); 117 } 118 printf("\n"); 119 120 SCX_OPS_LOAD(skel, pair_ops, scx_pair, uei); 121 122 /* 123 * Populate the cgrp_q_arr map which is an array containing per-cgroup 124 * queues. It'd probably be better to do this from BPF but there are too 125 * many to initialize statically and there's no way to dynamically 126 * populate from BPF. 127 */ 128 outer_fd = bpf_map__fd(skel->maps.cgrp_q_arr); 129 SCX_BUG_ON(outer_fd < 0, "Failed to get outer_fd: %d", outer_fd); 130 131 printf("Initializing"); 132 for (i = 0; i < MAX_CGRPS; i++) { 133 __s32 inner_fd; 134 135 if (exit_req) 136 break; 137 138 inner_fd = bpf_map_create(BPF_MAP_TYPE_QUEUE, NULL, 0, 139 sizeof(__u32), MAX_QUEUED, NULL); 140 SCX_BUG_ON(inner_fd < 0, "Failed to get inner_fd: %d", 141 inner_fd); 142 SCX_BUG_ON(bpf_map_update_elem(outer_fd, &i, &inner_fd, BPF_ANY), 143 "Failed to set inner map"); 144 close(inner_fd); 145 146 if (!(i % 10)) 147 printf("."); 148 fflush(stdout); 149 } 150 printf("\n"); 151 152 /* 153 * Fully initialized, attach and run. 154 */ 155 link = SCX_OPS_ATTACH(skel, pair_ops, scx_pair); 156 157 while (!exit_req && !UEI_EXITED(skel, uei)) { 158 printf("[SEQ %llu]\n", seq++); 159 printf(" total:%10" PRIu64 " dispatch:%10" PRIu64 " missing:%10" PRIu64 "\n", 160 skel->bss->nr_total, 161 skel->bss->nr_dispatched, 162 skel->bss->nr_missing); 163 printf(" kicks:%10" PRIu64 " preemptions:%7" PRIu64 "\n", 164 skel->bss->nr_kicks, 165 skel->bss->nr_preemptions); 166 printf(" exp:%10" PRIu64 " exp_wait:%10" PRIu64 " exp_empty:%10" PRIu64 "\n", 167 skel->bss->nr_exps, 168 skel->bss->nr_exp_waits, 169 skel->bss->nr_exp_empty); 170 printf("cgnext:%10" PRIu64 " cgcoll:%10" PRIu64 " cgempty:%10" PRIu64 "\n", 171 skel->bss->nr_cgrp_next, 172 skel->bss->nr_cgrp_coll, 173 skel->bss->nr_cgrp_empty); 174 fflush(stdout); 175 sleep(1); 176 } 177 178 bpf_link__destroy(link); 179 ecode = UEI_REPORT(skel, uei); 180 scx_pair__destroy(skel); 181 182 if (UEI_ECODE_RESTART(ecode)) 183 goto restart; 184 return 0; 185 } 186