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 "cgroup.h"
10 #include <errno.h>
11 #include <sys/utsname.h>
12 #include <bpf/libbpf.h>
13 #include <stdlib.h>
14 #include <string.h>
15
16 struct perf_env perf_env;
17
perf_env__insert_bpf_prog_info(struct perf_env * env,struct bpf_prog_info_node * info_node)18 void perf_env__insert_bpf_prog_info(struct perf_env *env,
19 struct bpf_prog_info_node *info_node)
20 {
21 __u32 prog_id = info_node->info_linear->info.id;
22 struct bpf_prog_info_node *node;
23 struct rb_node *parent = NULL;
24 struct rb_node **p;
25
26 down_write(&env->bpf_progs.lock);
27 p = &env->bpf_progs.infos.rb_node;
28
29 while (*p != NULL) {
30 parent = *p;
31 node = rb_entry(parent, struct bpf_prog_info_node, rb_node);
32 if (prog_id < node->info_linear->info.id) {
33 p = &(*p)->rb_left;
34 } else if (prog_id > node->info_linear->info.id) {
35 p = &(*p)->rb_right;
36 } else {
37 pr_debug("duplicated bpf prog info %u\n", prog_id);
38 goto out;
39 }
40 }
41
42 rb_link_node(&info_node->rb_node, parent, p);
43 rb_insert_color(&info_node->rb_node, &env->bpf_progs.infos);
44 env->bpf_progs.infos_cnt++;
45 out:
46 up_write(&env->bpf_progs.lock);
47 }
48
perf_env__find_bpf_prog_info(struct perf_env * env,__u32 prog_id)49 struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
50 __u32 prog_id)
51 {
52 struct bpf_prog_info_node *node = NULL;
53 struct rb_node *n;
54
55 down_read(&env->bpf_progs.lock);
56 n = env->bpf_progs.infos.rb_node;
57
58 while (n) {
59 node = rb_entry(n, struct bpf_prog_info_node, rb_node);
60 if (prog_id < node->info_linear->info.id)
61 n = n->rb_left;
62 else if (prog_id > node->info_linear->info.id)
63 n = n->rb_right;
64 else
65 goto out;
66 }
67 node = NULL;
68
69 out:
70 up_read(&env->bpf_progs.lock);
71 return node;
72 }
73
perf_env__insert_btf(struct perf_env * env,struct btf_node * btf_node)74 void perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
75 {
76 struct rb_node *parent = NULL;
77 __u32 btf_id = btf_node->id;
78 struct btf_node *node;
79 struct rb_node **p;
80
81 down_write(&env->bpf_progs.lock);
82 p = &env->bpf_progs.btfs.rb_node;
83
84 while (*p != NULL) {
85 parent = *p;
86 node = rb_entry(parent, struct btf_node, rb_node);
87 if (btf_id < node->id) {
88 p = &(*p)->rb_left;
89 } else if (btf_id > node->id) {
90 p = &(*p)->rb_right;
91 } else {
92 pr_debug("duplicated btf %u\n", btf_id);
93 goto out;
94 }
95 }
96
97 rb_link_node(&btf_node->rb_node, parent, p);
98 rb_insert_color(&btf_node->rb_node, &env->bpf_progs.btfs);
99 env->bpf_progs.btfs_cnt++;
100 out:
101 up_write(&env->bpf_progs.lock);
102 }
103
perf_env__find_btf(struct perf_env * env,__u32 btf_id)104 struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id)
105 {
106 struct btf_node *node = NULL;
107 struct rb_node *n;
108
109 down_read(&env->bpf_progs.lock);
110 n = env->bpf_progs.btfs.rb_node;
111
112 while (n) {
113 node = rb_entry(n, struct btf_node, rb_node);
114 if (btf_id < node->id)
115 n = n->rb_left;
116 else if (btf_id > node->id)
117 n = n->rb_right;
118 else
119 goto out;
120 }
121 node = NULL;
122
123 out:
124 up_read(&env->bpf_progs.lock);
125 return node;
126 }
127
128 /* purge data in bpf_progs.infos tree */
perf_env__purge_bpf(struct perf_env * env)129 static void perf_env__purge_bpf(struct perf_env *env)
130 {
131 struct rb_root *root;
132 struct rb_node *next;
133
134 down_write(&env->bpf_progs.lock);
135
136 root = &env->bpf_progs.infos;
137 next = rb_first(root);
138
139 while (next) {
140 struct bpf_prog_info_node *node;
141
142 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
143 next = rb_next(&node->rb_node);
144 rb_erase(&node->rb_node, root);
145 free(node);
146 }
147
148 env->bpf_progs.infos_cnt = 0;
149
150 root = &env->bpf_progs.btfs;
151 next = rb_first(root);
152
153 while (next) {
154 struct btf_node *node;
155
156 node = rb_entry(next, struct btf_node, rb_node);
157 next = rb_next(&node->rb_node);
158 rb_erase(&node->rb_node, root);
159 free(node);
160 }
161
162 env->bpf_progs.btfs_cnt = 0;
163
164 up_write(&env->bpf_progs.lock);
165 }
166
perf_env__exit(struct perf_env * env)167 void perf_env__exit(struct perf_env *env)
168 {
169 int i;
170
171 perf_env__purge_bpf(env);
172 perf_env__purge_cgroups(env);
173 zfree(&env->hostname);
174 zfree(&env->os_release);
175 zfree(&env->version);
176 zfree(&env->arch);
177 zfree(&env->cpu_desc);
178 zfree(&env->cpuid);
179 zfree(&env->cmdline);
180 zfree(&env->cmdline_argv);
181 zfree(&env->sibling_cores);
182 zfree(&env->sibling_threads);
183 zfree(&env->pmu_mappings);
184 zfree(&env->cpu);
185 zfree(&env->numa_map);
186
187 for (i = 0; i < env->nr_numa_nodes; i++)
188 perf_cpu_map__put(env->numa_nodes[i].map);
189 zfree(&env->numa_nodes);
190
191 for (i = 0; i < env->caches_cnt; i++)
192 cpu_cache_level__free(&env->caches[i]);
193 zfree(&env->caches);
194
195 for (i = 0; i < env->nr_memory_nodes; i++)
196 zfree(&env->memory_nodes[i].set);
197 zfree(&env->memory_nodes);
198 }
199
perf_env__init(struct perf_env * env)200 void perf_env__init(struct perf_env *env)
201 {
202 env->bpf_progs.infos = RB_ROOT;
203 env->bpf_progs.btfs = RB_ROOT;
204 init_rwsem(&env->bpf_progs.lock);
205 }
206
perf_env__set_cmdline(struct perf_env * env,int argc,const char * argv[])207 int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
208 {
209 int i;
210
211 /* do not include NULL termination */
212 env->cmdline_argv = calloc(argc, sizeof(char *));
213 if (env->cmdline_argv == NULL)
214 goto out_enomem;
215
216 /*
217 * Must copy argv contents because it gets moved around during option
218 * parsing:
219 */
220 for (i = 0; i < argc ; i++) {
221 env->cmdline_argv[i] = argv[i];
222 if (env->cmdline_argv[i] == NULL)
223 goto out_free;
224 }
225
226 env->nr_cmdline = argc;
227
228 return 0;
229 out_free:
230 zfree(&env->cmdline_argv);
231 out_enomem:
232 return -ENOMEM;
233 }
234
perf_env__read_cpu_topology_map(struct perf_env * env)235 int perf_env__read_cpu_topology_map(struct perf_env *env)
236 {
237 int cpu, nr_cpus;
238
239 if (env->cpu != NULL)
240 return 0;
241
242 if (env->nr_cpus_avail == 0)
243 env->nr_cpus_avail = cpu__max_present_cpu();
244
245 nr_cpus = env->nr_cpus_avail;
246 if (nr_cpus == -1)
247 return -EINVAL;
248
249 env->cpu = calloc(nr_cpus, sizeof(env->cpu[0]));
250 if (env->cpu == NULL)
251 return -ENOMEM;
252
253 for (cpu = 0; cpu < nr_cpus; ++cpu) {
254 env->cpu[cpu].core_id = cpu_map__get_core_id(cpu);
255 env->cpu[cpu].socket_id = cpu_map__get_socket_id(cpu);
256 env->cpu[cpu].die_id = cpu_map__get_die_id(cpu);
257 }
258
259 env->nr_cpus_avail = nr_cpus;
260 return 0;
261 }
262
perf_env__read_cpuid(struct perf_env * env)263 int perf_env__read_cpuid(struct perf_env *env)
264 {
265 char cpuid[128];
266 int err = get_cpuid(cpuid, sizeof(cpuid));
267
268 if (err)
269 return err;
270
271 free(env->cpuid);
272 env->cpuid = strdup(cpuid);
273 if (env->cpuid == NULL)
274 return ENOMEM;
275 return 0;
276 }
277
perf_env__read_arch(struct perf_env * env)278 static int perf_env__read_arch(struct perf_env *env)
279 {
280 struct utsname uts;
281
282 if (env->arch)
283 return 0;
284
285 if (!uname(&uts))
286 env->arch = strdup(uts.machine);
287
288 return env->arch ? 0 : -ENOMEM;
289 }
290
perf_env__read_nr_cpus_avail(struct perf_env * env)291 static int perf_env__read_nr_cpus_avail(struct perf_env *env)
292 {
293 if (env->nr_cpus_avail == 0)
294 env->nr_cpus_avail = cpu__max_present_cpu();
295
296 return env->nr_cpus_avail ? 0 : -ENOENT;
297 }
298
perf_env__raw_arch(struct perf_env * env)299 const char *perf_env__raw_arch(struct perf_env *env)
300 {
301 return env && !perf_env__read_arch(env) ? env->arch : "unknown";
302 }
303
perf_env__nr_cpus_avail(struct perf_env * env)304 int perf_env__nr_cpus_avail(struct perf_env *env)
305 {
306 return env && !perf_env__read_nr_cpus_avail(env) ? env->nr_cpus_avail : 0;
307 }
308
cpu_cache_level__free(struct cpu_cache_level * cache)309 void cpu_cache_level__free(struct cpu_cache_level *cache)
310 {
311 zfree(&cache->type);
312 zfree(&cache->map);
313 zfree(&cache->size);
314 }
315
316 /*
317 * Return architecture name in a normalized form.
318 * The conversion logic comes from the Makefile.
319 */
normalize_arch(char * arch)320 static const char *normalize_arch(char *arch)
321 {
322 if (!strcmp(arch, "x86_64"))
323 return "x86";
324 if (arch[0] == 'i' && arch[2] == '8' && arch[3] == '6')
325 return "x86";
326 if (!strcmp(arch, "sun4u") || !strncmp(arch, "sparc", 5))
327 return "sparc";
328 if (!strcmp(arch, "aarch64") || !strcmp(arch, "arm64"))
329 return "arm64";
330 if (!strncmp(arch, "arm", 3) || !strcmp(arch, "sa110"))
331 return "arm";
332 if (!strncmp(arch, "s390", 4))
333 return "s390";
334 if (!strncmp(arch, "parisc", 6))
335 return "parisc";
336 if (!strncmp(arch, "powerpc", 7) || !strncmp(arch, "ppc", 3))
337 return "powerpc";
338 if (!strncmp(arch, "mips", 4))
339 return "mips";
340 if (!strncmp(arch, "sh", 2) && isdigit(arch[2]))
341 return "sh";
342
343 return arch;
344 }
345
perf_env__arch(struct perf_env * env)346 const char *perf_env__arch(struct perf_env *env)
347 {
348 char *arch_name;
349
350 if (!env || !env->arch) { /* Assume local operation */
351 static struct utsname uts = { .machine[0] = '\0', };
352 if (uts.machine[0] == '\0' && uname(&uts) < 0)
353 return NULL;
354 arch_name = uts.machine;
355 } else
356 arch_name = env->arch;
357
358 return normalize_arch(arch_name);
359 }
360
361
perf_env__numa_node(struct perf_env * env,int cpu)362 int perf_env__numa_node(struct perf_env *env, int cpu)
363 {
364 if (!env->nr_numa_map) {
365 struct numa_node *nn;
366 int i, nr = 0;
367
368 for (i = 0; i < env->nr_numa_nodes; i++) {
369 nn = &env->numa_nodes[i];
370 nr = max(nr, perf_cpu_map__max(nn->map));
371 }
372
373 nr++;
374
375 /*
376 * We initialize the numa_map array to prepare
377 * it for missing cpus, which return node -1
378 */
379 env->numa_map = malloc(nr * sizeof(int));
380 if (!env->numa_map)
381 return -1;
382
383 for (i = 0; i < nr; i++)
384 env->numa_map[i] = -1;
385
386 env->nr_numa_map = nr;
387
388 for (i = 0; i < env->nr_numa_nodes; i++) {
389 int tmp, j;
390
391 nn = &env->numa_nodes[i];
392 perf_cpu_map__for_each_cpu(j, tmp, nn->map)
393 env->numa_map[j] = i;
394 }
395 }
396
397 return cpu >= 0 && cpu < env->nr_numa_map ? env->numa_map[cpu] : -1;
398 }
399
