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 "cgroup.h"
9 #include <errno.h>
10 #include <sys/utsname.h>
11 #include <stdlib.h>
12 #include <string.h>
13 #include "strbuf.h"
14
15 struct perf_env perf_env;
16
17 #ifdef HAVE_LIBBPF_SUPPORT
18 #include "bpf-event.h"
19 #include <bpf/libbpf.h>
20
perf_env__insert_bpf_prog_info(struct perf_env * env,struct bpf_prog_info_node * info_node)21 void perf_env__insert_bpf_prog_info(struct perf_env *env,
22 struct bpf_prog_info_node *info_node)
23 {
24 down_write(&env->bpf_progs.lock);
25 __perf_env__insert_bpf_prog_info(env, info_node);
26 up_write(&env->bpf_progs.lock);
27 }
28
__perf_env__insert_bpf_prog_info(struct perf_env * env,struct bpf_prog_info_node * info_node)29 void __perf_env__insert_bpf_prog_info(struct perf_env *env, struct bpf_prog_info_node *info_node)
30 {
31 __u32 prog_id = info_node->info_linear->info.id;
32 struct bpf_prog_info_node *node;
33 struct rb_node *parent = NULL;
34 struct rb_node **p;
35
36 p = &env->bpf_progs.infos.rb_node;
37
38 while (*p != NULL) {
39 parent = *p;
40 node = rb_entry(parent, struct bpf_prog_info_node, rb_node);
41 if (prog_id < node->info_linear->info.id) {
42 p = &(*p)->rb_left;
43 } else if (prog_id > node->info_linear->info.id) {
44 p = &(*p)->rb_right;
45 } else {
46 pr_debug("duplicated bpf prog info %u\n", prog_id);
47 return;
48 }
49 }
50
51 rb_link_node(&info_node->rb_node, parent, p);
52 rb_insert_color(&info_node->rb_node, &env->bpf_progs.infos);
53 env->bpf_progs.infos_cnt++;
54 }
55
perf_env__find_bpf_prog_info(struct perf_env * env,__u32 prog_id)56 struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
57 __u32 prog_id)
58 {
59 struct bpf_prog_info_node *node = NULL;
60 struct rb_node *n;
61
62 down_read(&env->bpf_progs.lock);
63 n = env->bpf_progs.infos.rb_node;
64
65 while (n) {
66 node = rb_entry(n, struct bpf_prog_info_node, rb_node);
67 if (prog_id < node->info_linear->info.id)
68 n = n->rb_left;
69 else if (prog_id > node->info_linear->info.id)
70 n = n->rb_right;
71 else
72 goto out;
73 }
74 node = NULL;
75
76 out:
77 up_read(&env->bpf_progs.lock);
78 return node;
79 }
80
perf_env__insert_btf(struct perf_env * env,struct btf_node * btf_node)81 bool perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
82 {
83 bool ret;
84
85 down_write(&env->bpf_progs.lock);
86 ret = __perf_env__insert_btf(env, btf_node);
87 up_write(&env->bpf_progs.lock);
88 return ret;
89 }
90
__perf_env__insert_btf(struct perf_env * env,struct btf_node * btf_node)91 bool __perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
92 {
93 struct rb_node *parent = NULL;
94 __u32 btf_id = btf_node->id;
95 struct btf_node *node;
96 struct rb_node **p;
97
98 p = &env->bpf_progs.btfs.rb_node;
99
100 while (*p != NULL) {
101 parent = *p;
102 node = rb_entry(parent, struct btf_node, rb_node);
103 if (btf_id < node->id) {
104 p = &(*p)->rb_left;
105 } else if (btf_id > node->id) {
106 p = &(*p)->rb_right;
107 } else {
108 pr_debug("duplicated btf %u\n", btf_id);
109 return false;
110 }
111 }
112
113 rb_link_node(&btf_node->rb_node, parent, p);
114 rb_insert_color(&btf_node->rb_node, &env->bpf_progs.btfs);
115 env->bpf_progs.btfs_cnt++;
116 return true;
117 }
118
perf_env__find_btf(struct perf_env * env,__u32 btf_id)119 struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id)
120 {
121 struct btf_node *res;
122
123 down_read(&env->bpf_progs.lock);
124 res = __perf_env__find_btf(env, btf_id);
125 up_read(&env->bpf_progs.lock);
126 return res;
127 }
128
__perf_env__find_btf(struct perf_env * env,__u32 btf_id)129 struct btf_node *__perf_env__find_btf(struct perf_env *env, __u32 btf_id)
130 {
131 struct btf_node *node = NULL;
132 struct rb_node *n;
133
134 n = env->bpf_progs.btfs.rb_node;
135
136 while (n) {
137 node = rb_entry(n, struct btf_node, rb_node);
138 if (btf_id < node->id)
139 n = n->rb_left;
140 else if (btf_id > node->id)
141 n = n->rb_right;
142 else
143 return node;
144 }
145 return NULL;
146 }
147
148 /* purge data in bpf_progs.infos tree */
perf_env__purge_bpf(struct perf_env * env)149 static void perf_env__purge_bpf(struct perf_env *env)
150 {
151 struct rb_root *root;
152 struct rb_node *next;
153
154 down_write(&env->bpf_progs.lock);
155
156 root = &env->bpf_progs.infos;
157 next = rb_first(root);
158
159 while (next) {
160 struct bpf_prog_info_node *node;
161
162 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
163 next = rb_next(&node->rb_node);
164 rb_erase(&node->rb_node, root);
165 free(node->info_linear);
166 free(node);
167 }
168
169 env->bpf_progs.infos_cnt = 0;
170
171 root = &env->bpf_progs.btfs;
172 next = rb_first(root);
173
174 while (next) {
175 struct btf_node *node;
176
177 node = rb_entry(next, struct btf_node, rb_node);
178 next = rb_next(&node->rb_node);
179 rb_erase(&node->rb_node, root);
180 free(node);
181 }
182
183 env->bpf_progs.btfs_cnt = 0;
184
185 up_write(&env->bpf_progs.lock);
186 }
187 #else // HAVE_LIBBPF_SUPPORT
perf_env__purge_bpf(struct perf_env * env __maybe_unused)188 static void perf_env__purge_bpf(struct perf_env *env __maybe_unused)
189 {
190 }
191 #endif // HAVE_LIBBPF_SUPPORT
192
perf_env__exit(struct perf_env * env)193 void perf_env__exit(struct perf_env *env)
194 {
195 int i;
196
197 perf_env__purge_bpf(env);
198 perf_env__purge_cgroups(env);
199 zfree(&env->hostname);
200 zfree(&env->os_release);
201 zfree(&env->version);
202 zfree(&env->arch);
203 zfree(&env->cpu_desc);
204 zfree(&env->cpuid);
205 zfree(&env->cmdline);
206 zfree(&env->cmdline_argv);
207 zfree(&env->sibling_dies);
208 zfree(&env->sibling_cores);
209 zfree(&env->sibling_threads);
210 zfree(&env->pmu_mappings);
211 zfree(&env->cpu);
212 zfree(&env->cpu_pmu_caps);
213 zfree(&env->numa_map);
214
215 for (i = 0; i < env->nr_numa_nodes; i++)
216 perf_cpu_map__put(env->numa_nodes[i].map);
217 zfree(&env->numa_nodes);
218
219 for (i = 0; i < env->caches_cnt; i++)
220 cpu_cache_level__free(&env->caches[i]);
221 zfree(&env->caches);
222
223 for (i = 0; i < env->nr_memory_nodes; i++)
224 zfree(&env->memory_nodes[i].set);
225 zfree(&env->memory_nodes);
226
227 for (i = 0; i < env->nr_hybrid_nodes; i++) {
228 zfree(&env->hybrid_nodes[i].pmu_name);
229 zfree(&env->hybrid_nodes[i].cpus);
230 }
231 zfree(&env->hybrid_nodes);
232
233 for (i = 0; i < env->nr_hybrid_cpc_nodes; i++) {
234 zfree(&env->hybrid_cpc_nodes[i].cpu_pmu_caps);
235 zfree(&env->hybrid_cpc_nodes[i].pmu_name);
236 }
237 zfree(&env->hybrid_cpc_nodes);
238 }
239
perf_env__init(struct perf_env * env)240 void perf_env__init(struct perf_env *env)
241 {
242 #ifdef HAVE_LIBBPF_SUPPORT
243 env->bpf_progs.infos = RB_ROOT;
244 env->bpf_progs.btfs = RB_ROOT;
245 init_rwsem(&env->bpf_progs.lock);
246 #endif
247 env->kernel_is_64_bit = -1;
248 }
249
perf_env__init_kernel_mode(struct perf_env * env)250 static void perf_env__init_kernel_mode(struct perf_env *env)
251 {
252 const char *arch = perf_env__raw_arch(env);
253
254 if (!strncmp(arch, "x86_64", 6) || !strncmp(arch, "aarch64", 7) ||
255 !strncmp(arch, "arm64", 5) || !strncmp(arch, "mips64", 6) ||
256 !strncmp(arch, "parisc64", 8) || !strncmp(arch, "riscv64", 7) ||
257 !strncmp(arch, "s390x", 5) || !strncmp(arch, "sparc64", 7))
258 env->kernel_is_64_bit = 1;
259 else
260 env->kernel_is_64_bit = 0;
261 }
262
perf_env__kernel_is_64_bit(struct perf_env * env)263 int perf_env__kernel_is_64_bit(struct perf_env *env)
264 {
265 if (env->kernel_is_64_bit == -1)
266 perf_env__init_kernel_mode(env);
267
268 return env->kernel_is_64_bit;
269 }
270
perf_env__set_cmdline(struct perf_env * env,int argc,const char * argv[])271 int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
272 {
273 int i;
274
275 /* do not include NULL termination */
276 env->cmdline_argv = calloc(argc, sizeof(char *));
277 if (env->cmdline_argv == NULL)
278 goto out_enomem;
279
280 /*
281 * Must copy argv contents because it gets moved around during option
282 * parsing:
283 */
284 for (i = 0; i < argc ; i++) {
285 env->cmdline_argv[i] = argv[i];
286 if (env->cmdline_argv[i] == NULL)
287 goto out_free;
288 }
289
290 env->nr_cmdline = argc;
291
292 return 0;
293 out_free:
294 zfree(&env->cmdline_argv);
295 out_enomem:
296 return -ENOMEM;
297 }
298
perf_env__read_cpu_topology_map(struct perf_env * env)299 int perf_env__read_cpu_topology_map(struct perf_env *env)
300 {
301 int cpu, nr_cpus;
302
303 if (env->cpu != NULL)
304 return 0;
305
306 if (env->nr_cpus_avail == 0)
307 env->nr_cpus_avail = cpu__max_present_cpu();
308
309 nr_cpus = env->nr_cpus_avail;
310 if (nr_cpus == -1)
311 return -EINVAL;
312
313 env->cpu = calloc(nr_cpus, sizeof(env->cpu[0]));
314 if (env->cpu == NULL)
315 return -ENOMEM;
316
317 for (cpu = 0; cpu < nr_cpus; ++cpu) {
318 env->cpu[cpu].core_id = cpu_map__get_core_id(cpu);
319 env->cpu[cpu].socket_id = cpu_map__get_socket_id(cpu);
320 env->cpu[cpu].die_id = cpu_map__get_die_id(cpu);
321 }
322
323 env->nr_cpus_avail = nr_cpus;
324 return 0;
325 }
326
perf_env__read_pmu_mappings(struct perf_env * env)327 int perf_env__read_pmu_mappings(struct perf_env *env)
328 {
329 struct perf_pmu *pmu = NULL;
330 u32 pmu_num = 0;
331 struct strbuf sb;
332
333 while ((pmu = perf_pmu__scan(pmu))) {
334 if (!pmu->name)
335 continue;
336 pmu_num++;
337 }
338 if (!pmu_num) {
339 pr_debug("pmu mappings not available\n");
340 return -ENOENT;
341 }
342 env->nr_pmu_mappings = pmu_num;
343
344 if (strbuf_init(&sb, 128 * pmu_num) < 0)
345 return -ENOMEM;
346
347 while ((pmu = perf_pmu__scan(pmu))) {
348 if (!pmu->name)
349 continue;
350 if (strbuf_addf(&sb, "%u:%s", pmu->type, pmu->name) < 0)
351 goto error;
352 /* include a NULL character at the end */
353 if (strbuf_add(&sb, "", 1) < 0)
354 goto error;
355 }
356
357 env->pmu_mappings = strbuf_detach(&sb, NULL);
358
359 return 0;
360
361 error:
362 strbuf_release(&sb);
363 return -1;
364 }
365
perf_env__read_cpuid(struct perf_env * env)366 int perf_env__read_cpuid(struct perf_env *env)
367 {
368 char cpuid[128];
369 int err = get_cpuid(cpuid, sizeof(cpuid));
370
371 if (err)
372 return err;
373
374 free(env->cpuid);
375 env->cpuid = strdup(cpuid);
376 if (env->cpuid == NULL)
377 return ENOMEM;
378 return 0;
379 }
380
perf_env__read_arch(struct perf_env * env)381 static int perf_env__read_arch(struct perf_env *env)
382 {
383 struct utsname uts;
384
385 if (env->arch)
386 return 0;
387
388 if (!uname(&uts))
389 env->arch = strdup(uts.machine);
390
391 return env->arch ? 0 : -ENOMEM;
392 }
393
perf_env__read_nr_cpus_avail(struct perf_env * env)394 static int perf_env__read_nr_cpus_avail(struct perf_env *env)
395 {
396 if (env->nr_cpus_avail == 0)
397 env->nr_cpus_avail = cpu__max_present_cpu();
398
399 return env->nr_cpus_avail ? 0 : -ENOENT;
400 }
401
perf_env__raw_arch(struct perf_env * env)402 const char *perf_env__raw_arch(struct perf_env *env)
403 {
404 return env && !perf_env__read_arch(env) ? env->arch : "unknown";
405 }
406
perf_env__nr_cpus_avail(struct perf_env * env)407 int perf_env__nr_cpus_avail(struct perf_env *env)
408 {
409 return env && !perf_env__read_nr_cpus_avail(env) ? env->nr_cpus_avail : 0;
410 }
411
cpu_cache_level__free(struct cpu_cache_level * cache)412 void cpu_cache_level__free(struct cpu_cache_level *cache)
413 {
414 zfree(&cache->type);
415 zfree(&cache->map);
416 zfree(&cache->size);
417 }
418
419 /*
420 * Return architecture name in a normalized form.
421 * The conversion logic comes from the Makefile.
422 */
normalize_arch(char * arch)423 static const char *normalize_arch(char *arch)
424 {
425 if (!strcmp(arch, "x86_64"))
426 return "x86";
427 if (arch[0] == 'i' && arch[2] == '8' && arch[3] == '6')
428 return "x86";
429 if (!strcmp(arch, "sun4u") || !strncmp(arch, "sparc", 5))
430 return "sparc";
431 if (!strncmp(arch, "aarch64", 7) || !strncmp(arch, "arm64", 5))
432 return "arm64";
433 if (!strncmp(arch, "arm", 3) || !strcmp(arch, "sa110"))
434 return "arm";
435 if (!strncmp(arch, "s390", 4))
436 return "s390";
437 if (!strncmp(arch, "parisc", 6))
438 return "parisc";
439 if (!strncmp(arch, "powerpc", 7) || !strncmp(arch, "ppc", 3))
440 return "powerpc";
441 if (!strncmp(arch, "mips", 4))
442 return "mips";
443 if (!strncmp(arch, "sh", 2) && isdigit(arch[2]))
444 return "sh";
445
446 return arch;
447 }
448
perf_env__arch(struct perf_env * env)449 const char *perf_env__arch(struct perf_env *env)
450 {
451 char *arch_name;
452
453 if (!env || !env->arch) { /* Assume local operation */
454 static struct utsname uts = { .machine[0] = '\0', };
455 if (uts.machine[0] == '\0' && uname(&uts) < 0)
456 return NULL;
457 arch_name = uts.machine;
458 } else
459 arch_name = env->arch;
460
461 return normalize_arch(arch_name);
462 }
463
perf_env__cpuid(struct perf_env * env)464 const char *perf_env__cpuid(struct perf_env *env)
465 {
466 int status;
467
468 if (!env || !env->cpuid) { /* Assume local operation */
469 status = perf_env__read_cpuid(env);
470 if (status)
471 return NULL;
472 }
473
474 return env->cpuid;
475 }
476
perf_env__nr_pmu_mappings(struct perf_env * env)477 int perf_env__nr_pmu_mappings(struct perf_env *env)
478 {
479 int status;
480
481 if (!env || !env->nr_pmu_mappings) { /* Assume local operation */
482 status = perf_env__read_pmu_mappings(env);
483 if (status)
484 return 0;
485 }
486
487 return env->nr_pmu_mappings;
488 }
489
perf_env__pmu_mappings(struct perf_env * env)490 const char *perf_env__pmu_mappings(struct perf_env *env)
491 {
492 int status;
493
494 if (!env || !env->pmu_mappings) { /* Assume local operation */
495 status = perf_env__read_pmu_mappings(env);
496 if (status)
497 return NULL;
498 }
499
500 return env->pmu_mappings;
501 }
502
perf_env__numa_node(struct perf_env * env,int cpu)503 int perf_env__numa_node(struct perf_env *env, int cpu)
504 {
505 if (!env->nr_numa_map) {
506 struct numa_node *nn;
507 int i, nr = 0;
508
509 for (i = 0; i < env->nr_numa_nodes; i++) {
510 nn = &env->numa_nodes[i];
511 nr = max(nr, perf_cpu_map__max(nn->map));
512 }
513
514 nr++;
515
516 /*
517 * We initialize the numa_map array to prepare
518 * it for missing cpus, which return node -1
519 */
520 env->numa_map = malloc(nr * sizeof(int));
521 if (!env->numa_map)
522 return -1;
523
524 for (i = 0; i < nr; i++)
525 env->numa_map[i] = -1;
526
527 env->nr_numa_map = nr;
528
529 for (i = 0; i < env->nr_numa_nodes; i++) {
530 int tmp, j;
531
532 nn = &env->numa_nodes[i];
533 perf_cpu_map__for_each_cpu(j, tmp, nn->map)
534 env->numa_map[j] = i;
535 }
536 }
537
538 return cpu >= 0 && cpu < env->nr_numa_map ? env->numa_map[cpu] : -1;
539 }
540