1 // SPDX-License-Identifier: GPL-2.0 2 #include "cpumap.h" 3 #include "debug.h" 4 #include "env.h" 5 #include "util/header.h" 6 #include <linux/ctype.h> 7 #include <linux/zalloc.h> 8 #include "bpf-event.h" 9 #include <errno.h> 10 #include <sys/utsname.h> 11 #include <bpf/libbpf.h> 12 #include <stdlib.h> 13 #include <string.h> 14 15 struct perf_env perf_env; 16 17 void perf_env__insert_bpf_prog_info(struct perf_env *env, 18 struct bpf_prog_info_node *info_node) 19 { 20 __u32 prog_id = info_node->info_linear->info.id; 21 struct bpf_prog_info_node *node; 22 struct rb_node *parent = NULL; 23 struct rb_node **p; 24 25 down_write(&env->bpf_progs.lock); 26 p = &env->bpf_progs.infos.rb_node; 27 28 while (*p != NULL) { 29 parent = *p; 30 node = rb_entry(parent, struct bpf_prog_info_node, rb_node); 31 if (prog_id < node->info_linear->info.id) { 32 p = &(*p)->rb_left; 33 } else if (prog_id > node->info_linear->info.id) { 34 p = &(*p)->rb_right; 35 } else { 36 pr_debug("duplicated bpf prog info %u\n", prog_id); 37 goto out; 38 } 39 } 40 41 rb_link_node(&info_node->rb_node, parent, p); 42 rb_insert_color(&info_node->rb_node, &env->bpf_progs.infos); 43 env->bpf_progs.infos_cnt++; 44 out: 45 up_write(&env->bpf_progs.lock); 46 } 47 48 struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env, 49 __u32 prog_id) 50 { 51 struct bpf_prog_info_node *node = NULL; 52 struct rb_node *n; 53 54 down_read(&env->bpf_progs.lock); 55 n = env->bpf_progs.infos.rb_node; 56 57 while (n) { 58 node = rb_entry(n, struct bpf_prog_info_node, rb_node); 59 if (prog_id < node->info_linear->info.id) 60 n = n->rb_left; 61 else if (prog_id > node->info_linear->info.id) 62 n = n->rb_right; 63 else 64 goto out; 65 } 66 node = NULL; 67 68 out: 69 up_read(&env->bpf_progs.lock); 70 return node; 71 } 72 73 void perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node) 74 { 75 struct rb_node *parent = NULL; 76 __u32 btf_id = btf_node->id; 77 struct btf_node *node; 78 struct rb_node **p; 79 80 down_write(&env->bpf_progs.lock); 81 p = &env->bpf_progs.btfs.rb_node; 82 83 while (*p != NULL) { 84 parent = *p; 85 node = rb_entry(parent, struct btf_node, rb_node); 86 if (btf_id < node->id) { 87 p = &(*p)->rb_left; 88 } else if (btf_id > node->id) { 89 p = &(*p)->rb_right; 90 } else { 91 pr_debug("duplicated btf %u\n", btf_id); 92 goto out; 93 } 94 } 95 96 rb_link_node(&btf_node->rb_node, parent, p); 97 rb_insert_color(&btf_node->rb_node, &env->bpf_progs.btfs); 98 env->bpf_progs.btfs_cnt++; 99 out: 100 up_write(&env->bpf_progs.lock); 101 } 102 103 struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id) 104 { 105 struct btf_node *node = NULL; 106 struct rb_node *n; 107 108 down_read(&env->bpf_progs.lock); 109 n = env->bpf_progs.btfs.rb_node; 110 111 while (n) { 112 node = rb_entry(n, struct btf_node, rb_node); 113 if (btf_id < node->id) 114 n = n->rb_left; 115 else if (btf_id > node->id) 116 n = n->rb_right; 117 else 118 goto out; 119 } 120 node = NULL; 121 122 out: 123 up_read(&env->bpf_progs.lock); 124 return node; 125 } 126 127 /* purge data in bpf_progs.infos tree */ 128 static void perf_env__purge_bpf(struct perf_env *env) 129 { 130 struct rb_root *root; 131 struct rb_node *next; 132 133 down_write(&env->bpf_progs.lock); 134 135 root = &env->bpf_progs.infos; 136 next = rb_first(root); 137 138 while (next) { 139 struct bpf_prog_info_node *node; 140 141 node = rb_entry(next, struct bpf_prog_info_node, rb_node); 142 next = rb_next(&node->rb_node); 143 rb_erase(&node->rb_node, root); 144 free(node); 145 } 146 147 env->bpf_progs.infos_cnt = 0; 148 149 root = &env->bpf_progs.btfs; 150 next = rb_first(root); 151 152 while (next) { 153 struct btf_node *node; 154 155 node = rb_entry(next, struct btf_node, rb_node); 156 next = rb_next(&node->rb_node); 157 rb_erase(&node->rb_node, root); 158 free(node); 159 } 160 161 env->bpf_progs.btfs_cnt = 0; 162 163 up_write(&env->bpf_progs.lock); 164 } 165 166 void perf_env__exit(struct perf_env *env) 167 { 168 int i; 169 170 perf_env__purge_bpf(env); 171 zfree(&env->hostname); 172 zfree(&env->os_release); 173 zfree(&env->version); 174 zfree(&env->arch); 175 zfree(&env->cpu_desc); 176 zfree(&env->cpuid); 177 zfree(&env->cmdline); 178 zfree(&env->cmdline_argv); 179 zfree(&env->sibling_cores); 180 zfree(&env->sibling_threads); 181 zfree(&env->pmu_mappings); 182 zfree(&env->cpu); 183 zfree(&env->numa_map); 184 185 for (i = 0; i < env->nr_numa_nodes; i++) 186 perf_cpu_map__put(env->numa_nodes[i].map); 187 zfree(&env->numa_nodes); 188 189 for (i = 0; i < env->caches_cnt; i++) 190 cpu_cache_level__free(&env->caches[i]); 191 zfree(&env->caches); 192 193 for (i = 0; i < env->nr_memory_nodes; i++) 194 zfree(&env->memory_nodes[i].set); 195 zfree(&env->memory_nodes); 196 } 197 198 void perf_env__init(struct perf_env *env) 199 { 200 env->bpf_progs.infos = RB_ROOT; 201 env->bpf_progs.btfs = RB_ROOT; 202 init_rwsem(&env->bpf_progs.lock); 203 } 204 205 int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[]) 206 { 207 int i; 208 209 /* do not include NULL termination */ 210 env->cmdline_argv = calloc(argc, sizeof(char *)); 211 if (env->cmdline_argv == NULL) 212 goto out_enomem; 213 214 /* 215 * Must copy argv contents because it gets moved around during option 216 * parsing: 217 */ 218 for (i = 0; i < argc ; i++) { 219 env->cmdline_argv[i] = argv[i]; 220 if (env->cmdline_argv[i] == NULL) 221 goto out_free; 222 } 223 224 env->nr_cmdline = argc; 225 226 return 0; 227 out_free: 228 zfree(&env->cmdline_argv); 229 out_enomem: 230 return -ENOMEM; 231 } 232 233 int perf_env__read_cpu_topology_map(struct perf_env *env) 234 { 235 int cpu, nr_cpus; 236 237 if (env->cpu != NULL) 238 return 0; 239 240 if (env->nr_cpus_avail == 0) 241 env->nr_cpus_avail = cpu__max_present_cpu(); 242 243 nr_cpus = env->nr_cpus_avail; 244 if (nr_cpus == -1) 245 return -EINVAL; 246 247 env->cpu = calloc(nr_cpus, sizeof(env->cpu[0])); 248 if (env->cpu == NULL) 249 return -ENOMEM; 250 251 for (cpu = 0; cpu < nr_cpus; ++cpu) { 252 env->cpu[cpu].core_id = cpu_map__get_core_id(cpu); 253 env->cpu[cpu].socket_id = cpu_map__get_socket_id(cpu); 254 env->cpu[cpu].die_id = cpu_map__get_die_id(cpu); 255 } 256 257 env->nr_cpus_avail = nr_cpus; 258 return 0; 259 } 260 261 int perf_env__read_cpuid(struct perf_env *env) 262 { 263 char cpuid[128]; 264 int err = get_cpuid(cpuid, sizeof(cpuid)); 265 266 if (err) 267 return err; 268 269 free(env->cpuid); 270 env->cpuid = strdup(cpuid); 271 if (env->cpuid == NULL) 272 return ENOMEM; 273 return 0; 274 } 275 276 static int perf_env__read_arch(struct perf_env *env) 277 { 278 struct utsname uts; 279 280 if (env->arch) 281 return 0; 282 283 if (!uname(&uts)) 284 env->arch = strdup(uts.machine); 285 286 return env->arch ? 0 : -ENOMEM; 287 } 288 289 static int perf_env__read_nr_cpus_avail(struct perf_env *env) 290 { 291 if (env->nr_cpus_avail == 0) 292 env->nr_cpus_avail = cpu__max_present_cpu(); 293 294 return env->nr_cpus_avail ? 0 : -ENOENT; 295 } 296 297 const char *perf_env__raw_arch(struct perf_env *env) 298 { 299 return env && !perf_env__read_arch(env) ? env->arch : "unknown"; 300 } 301 302 int perf_env__nr_cpus_avail(struct perf_env *env) 303 { 304 return env && !perf_env__read_nr_cpus_avail(env) ? env->nr_cpus_avail : 0; 305 } 306 307 void cpu_cache_level__free(struct cpu_cache_level *cache) 308 { 309 zfree(&cache->type); 310 zfree(&cache->map); 311 zfree(&cache->size); 312 } 313 314 /* 315 * Return architecture name in a normalized form. 316 * The conversion logic comes from the Makefile. 317 */ 318 static const char *normalize_arch(char *arch) 319 { 320 if (!strcmp(arch, "x86_64")) 321 return "x86"; 322 if (arch[0] == 'i' && arch[2] == '8' && arch[3] == '6') 323 return "x86"; 324 if (!strcmp(arch, "sun4u") || !strncmp(arch, "sparc", 5)) 325 return "sparc"; 326 if (!strcmp(arch, "aarch64") || !strcmp(arch, "arm64")) 327 return "arm64"; 328 if (!strncmp(arch, "arm", 3) || !strcmp(arch, "sa110")) 329 return "arm"; 330 if (!strncmp(arch, "s390", 4)) 331 return "s390"; 332 if (!strncmp(arch, "parisc", 6)) 333 return "parisc"; 334 if (!strncmp(arch, "powerpc", 7) || !strncmp(arch, "ppc", 3)) 335 return "powerpc"; 336 if (!strncmp(arch, "mips", 4)) 337 return "mips"; 338 if (!strncmp(arch, "sh", 2) && isdigit(arch[2])) 339 return "sh"; 340 341 return arch; 342 } 343 344 const char *perf_env__arch(struct perf_env *env) 345 { 346 char *arch_name; 347 348 if (!env || !env->arch) { /* Assume local operation */ 349 static struct utsname uts = { .machine[0] = '\0', }; 350 if (uts.machine[0] == '\0' && uname(&uts) < 0) 351 return NULL; 352 arch_name = uts.machine; 353 } else 354 arch_name = env->arch; 355 356 return normalize_arch(arch_name); 357 } 358 359 360 int perf_env__numa_node(struct perf_env *env, int cpu) 361 { 362 if (!env->nr_numa_map) { 363 struct numa_node *nn; 364 int i, nr = 0; 365 366 for (i = 0; i < env->nr_numa_nodes; i++) { 367 nn = &env->numa_nodes[i]; 368 nr = max(nr, perf_cpu_map__max(nn->map)); 369 } 370 371 nr++; 372 373 /* 374 * We initialize the numa_map array to prepare 375 * it for missing cpus, which return node -1 376 */ 377 env->numa_map = malloc(nr * sizeof(int)); 378 if (!env->numa_map) 379 return -1; 380 381 for (i = 0; i < nr; i++) 382 env->numa_map[i] = -1; 383 384 env->nr_numa_map = nr; 385 386 for (i = 0; i < env->nr_numa_nodes; i++) { 387 int tmp, j; 388 389 nn = &env->numa_nodes[i]; 390 perf_cpu_map__for_each_cpu(j, tmp, nn->map) 391 env->numa_map[j] = i; 392 } 393 } 394 395 return cpu >= 0 && cpu < env->nr_numa_map ? env->numa_map[cpu] : -1; 396 } 397