• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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->info_linear);
146 		free(node);
147 	}
148 
149 	env->bpf_progs.infos_cnt = 0;
150 
151 	root = &env->bpf_progs.btfs;
152 	next = rb_first(root);
153 
154 	while (next) {
155 		struct btf_node *node;
156 
157 		node = rb_entry(next, struct btf_node, rb_node);
158 		next = rb_next(&node->rb_node);
159 		rb_erase(&node->rb_node, root);
160 		free(node);
161 	}
162 
163 	env->bpf_progs.btfs_cnt = 0;
164 
165 	up_write(&env->bpf_progs.lock);
166 }
167 
perf_env__exit(struct perf_env * env)168 void perf_env__exit(struct perf_env *env)
169 {
170 	int i;
171 
172 	perf_env__purge_bpf(env);
173 	perf_env__purge_cgroups(env);
174 	zfree(&env->hostname);
175 	zfree(&env->os_release);
176 	zfree(&env->version);
177 	zfree(&env->arch);
178 	zfree(&env->cpu_desc);
179 	zfree(&env->cpuid);
180 	zfree(&env->cmdline);
181 	zfree(&env->cmdline_argv);
182 	zfree(&env->sibling_dies);
183 	zfree(&env->sibling_cores);
184 	zfree(&env->sibling_threads);
185 	zfree(&env->pmu_mappings);
186 	zfree(&env->cpu);
187 	zfree(&env->cpu_pmu_caps);
188 	zfree(&env->numa_map);
189 
190 	for (i = 0; i < env->nr_numa_nodes; i++)
191 		perf_cpu_map__put(env->numa_nodes[i].map);
192 	zfree(&env->numa_nodes);
193 
194 	for (i = 0; i < env->caches_cnt; i++)
195 		cpu_cache_level__free(&env->caches[i]);
196 	zfree(&env->caches);
197 
198 	for (i = 0; i < env->nr_memory_nodes; i++)
199 		zfree(&env->memory_nodes[i].set);
200 	zfree(&env->memory_nodes);
201 }
202 
perf_env__init(struct perf_env * env)203 void perf_env__init(struct perf_env *env)
204 {
205 	env->bpf_progs.infos = RB_ROOT;
206 	env->bpf_progs.btfs = RB_ROOT;
207 	init_rwsem(&env->bpf_progs.lock);
208 }
209 
perf_env__set_cmdline(struct perf_env * env,int argc,const char * argv[])210 int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
211 {
212 	int i;
213 
214 	/* do not include NULL termination */
215 	env->cmdline_argv = calloc(argc, sizeof(char *));
216 	if (env->cmdline_argv == NULL)
217 		goto out_enomem;
218 
219 	/*
220 	 * Must copy argv contents because it gets moved around during option
221 	 * parsing:
222 	 */
223 	for (i = 0; i < argc ; i++) {
224 		env->cmdline_argv[i] = argv[i];
225 		if (env->cmdline_argv[i] == NULL)
226 			goto out_free;
227 	}
228 
229 	env->nr_cmdline = argc;
230 
231 	return 0;
232 out_free:
233 	zfree(&env->cmdline_argv);
234 out_enomem:
235 	return -ENOMEM;
236 }
237 
perf_env__read_cpu_topology_map(struct perf_env * env)238 int perf_env__read_cpu_topology_map(struct perf_env *env)
239 {
240 	int cpu, nr_cpus;
241 
242 	if (env->cpu != NULL)
243 		return 0;
244 
245 	if (env->nr_cpus_avail == 0)
246 		env->nr_cpus_avail = cpu__max_present_cpu();
247 
248 	nr_cpus = env->nr_cpus_avail;
249 	if (nr_cpus == -1)
250 		return -EINVAL;
251 
252 	env->cpu = calloc(nr_cpus, sizeof(env->cpu[0]));
253 	if (env->cpu == NULL)
254 		return -ENOMEM;
255 
256 	for (cpu = 0; cpu < nr_cpus; ++cpu) {
257 		env->cpu[cpu].core_id	= cpu_map__get_core_id(cpu);
258 		env->cpu[cpu].socket_id	= cpu_map__get_socket_id(cpu);
259 		env->cpu[cpu].die_id	= cpu_map__get_die_id(cpu);
260 	}
261 
262 	env->nr_cpus_avail = nr_cpus;
263 	return 0;
264 }
265 
perf_env__read_cpuid(struct perf_env * env)266 int perf_env__read_cpuid(struct perf_env *env)
267 {
268 	char cpuid[128];
269 	int err = get_cpuid(cpuid, sizeof(cpuid));
270 
271 	if (err)
272 		return err;
273 
274 	free(env->cpuid);
275 	env->cpuid = strdup(cpuid);
276 	if (env->cpuid == NULL)
277 		return ENOMEM;
278 	return 0;
279 }
280 
perf_env__read_arch(struct perf_env * env)281 static int perf_env__read_arch(struct perf_env *env)
282 {
283 	struct utsname uts;
284 
285 	if (env->arch)
286 		return 0;
287 
288 	if (!uname(&uts))
289 		env->arch = strdup(uts.machine);
290 
291 	return env->arch ? 0 : -ENOMEM;
292 }
293 
perf_env__read_nr_cpus_avail(struct perf_env * env)294 static int perf_env__read_nr_cpus_avail(struct perf_env *env)
295 {
296 	if (env->nr_cpus_avail == 0)
297 		env->nr_cpus_avail = cpu__max_present_cpu();
298 
299 	return env->nr_cpus_avail ? 0 : -ENOENT;
300 }
301 
perf_env__raw_arch(struct perf_env * env)302 const char *perf_env__raw_arch(struct perf_env *env)
303 {
304 	return env && !perf_env__read_arch(env) ? env->arch : "unknown";
305 }
306 
perf_env__nr_cpus_avail(struct perf_env * env)307 int perf_env__nr_cpus_avail(struct perf_env *env)
308 {
309 	return env && !perf_env__read_nr_cpus_avail(env) ? env->nr_cpus_avail : 0;
310 }
311 
cpu_cache_level__free(struct cpu_cache_level * cache)312 void cpu_cache_level__free(struct cpu_cache_level *cache)
313 {
314 	zfree(&cache->type);
315 	zfree(&cache->map);
316 	zfree(&cache->size);
317 }
318 
319 /*
320  * Return architecture name in a normalized form.
321  * The conversion logic comes from the Makefile.
322  */
normalize_arch(char * arch)323 static const char *normalize_arch(char *arch)
324 {
325 	if (!strcmp(arch, "x86_64"))
326 		return "x86";
327 	if (arch[0] == 'i' && arch[2] == '8' && arch[3] == '6')
328 		return "x86";
329 	if (!strcmp(arch, "sun4u") || !strncmp(arch, "sparc", 5))
330 		return "sparc";
331 	if (!strncmp(arch, "aarch64", 7) || !strncmp(arch, "arm64", 5))
332 		return "arm64";
333 	if (!strncmp(arch, "arm", 3) || !strcmp(arch, "sa110"))
334 		return "arm";
335 	if (!strncmp(arch, "s390", 4))
336 		return "s390";
337 	if (!strncmp(arch, "parisc", 6))
338 		return "parisc";
339 	if (!strncmp(arch, "powerpc", 7) || !strncmp(arch, "ppc", 3))
340 		return "powerpc";
341 	if (!strncmp(arch, "mips", 4))
342 		return "mips";
343 	if (!strncmp(arch, "sh", 2) && isdigit(arch[2]))
344 		return "sh";
345 
346 	return arch;
347 }
348 
perf_env__arch(struct perf_env * env)349 const char *perf_env__arch(struct perf_env *env)
350 {
351 	char *arch_name;
352 
353 	if (!env || !env->arch) { /* Assume local operation */
354 		static struct utsname uts = { .machine[0] = '\0', };
355 		if (uts.machine[0] == '\0' && uname(&uts) < 0)
356 			return NULL;
357 		arch_name = uts.machine;
358 	} else
359 		arch_name = env->arch;
360 
361 	return normalize_arch(arch_name);
362 }
363 
364 
perf_env__numa_node(struct perf_env * env,int cpu)365 int perf_env__numa_node(struct perf_env *env, int cpu)
366 {
367 	if (!env->nr_numa_map) {
368 		struct numa_node *nn;
369 		int i, nr = 0;
370 
371 		for (i = 0; i < env->nr_numa_nodes; i++) {
372 			nn = &env->numa_nodes[i];
373 			nr = max(nr, perf_cpu_map__max(nn->map));
374 		}
375 
376 		nr++;
377 
378 		/*
379 		 * We initialize the numa_map array to prepare
380 		 * it for missing cpus, which return node -1
381 		 */
382 		env->numa_map = malloc(nr * sizeof(int));
383 		if (!env->numa_map)
384 			return -1;
385 
386 		for (i = 0; i < nr; i++)
387 			env->numa_map[i] = -1;
388 
389 		env->nr_numa_map = nr;
390 
391 		for (i = 0; i < env->nr_numa_nodes; i++) {
392 			int tmp, j;
393 
394 			nn = &env->numa_nodes[i];
395 			perf_cpu_map__for_each_cpu(j, tmp, nn->map)
396 				env->numa_map[j] = i;
397 		}
398 	}
399 
400 	return cpu >= 0 && cpu < env->nr_numa_map ? env->numa_map[cpu] : -1;
401 }
402