aboutsummaryrefslogtreecommitdiffstats
path: root/tools/perf/builtin-kmem.c
diff options
context:
space:
mode:
authorLi Zefan <lizf@cn.fujitsu.com>2009-11-20 02:53:25 -0500
committerIngo Molnar <mingo@elte.hu>2009-11-20 03:51:41 -0500
commitba77c9e11111a172c9e8687fe16a6a173a61916f (patch)
tree54dae585a10100f977c525399b7598855b3ca7df /tools/perf/builtin-kmem.c
parentce64c62074d945fe5f8a7f01bdc30125f994ea67 (diff)
perf: Add 'perf kmem' tool
This tool is mostly a perf version of kmemtrace-user. The following information is provided by this tool: - the total amount of memory allocated and fragmentation per call-site - the total amount of memory allocated and fragmentation per allocation - total memory allocated and fragmentation in the collected dataset - ... Sample output: # ./perf kmem record ^C # ./perf kmem --stat caller --stat alloc -l 10 ------------------------------------------------------------------------------ Callsite | Total_alloc/Per | Total_req/Per | Hit | Fragmentation ------------------------------------------------------------------------------ 0xc052f37a | 790528/4096 | 790528/4096 | 193 | 0.000% 0xc0541d70 | 524288/4096 | 524288/4096 | 128 | 0.000% 0xc051cc68 | 481600/200 | 481600/200 | 2408 | 0.000% 0xc0572623 | 297444/676 | 297440/676 | 440 | 0.001% 0xc05399f1 | 73476/164 | 73472/164 | 448 | 0.005% 0xc05243bf | 51456/256 | 51456/256 | 201 | 0.000% 0xc0730d0e | 31844/497 | 31808/497 | 64 | 0.113% 0xc0734c4e | 17152/256 | 17152/256 | 67 | 0.000% 0xc0541a6d | 16384/128 | 16384/128 | 128 | 0.000% 0xc059c217 | 13120/40 | 13120/40 | 328 | 0.000% 0xc0501ee6 | 11264/88 | 11264/88 | 128 | 0.000% 0xc04daef0 | 7504/682 | 7128/648 | 11 | 5.011% 0xc04e14a3 | 4216/191 | 4216/191 | 22 | 0.000% 0xc05041ca | 3524/44 | 3520/44 | 80 | 0.114% 0xc0734fa3 | 2104/701 | 1620/540 | 3 | 23.004% 0xc05ec9f1 | 2024/289 | 2016/288 | 7 | 0.395% 0xc06a1999 | 1792/256 | 1792/256 | 7 | 0.000% 0xc0463b9a | 1584/144 | 1584/144 | 11 | 0.000% 0xc0541eb0 | 1024/16 | 1024/16 | 64 | 0.000% 0xc06a19ac | 896/128 | 896/128 | 7 | 0.000% 0xc05721c0 | 772/12 | 768/12 | 64 | 0.518% 0xc054d1e6 | 288/57 | 280/56 | 5 | 2.778% 0xc04b562e | 157/31 | 154/30 | 5 | 1.911% 0xc04b536f | 80/16 | 80/16 | 5 | 0.000% 0xc05855a0 | 64/64 | 36/36 | 1 | 43.750% ------------------------------------------------------------------------------ ------------------------------------------------------------------------------ Alloc Ptr | Total_alloc/Per | Total_req/Per | Hit | Fragmentation ------------------------------------------------------------------------------ 0xda884000 | 1052672/4096 | 1052672/4096 | 257 | 0.000% 0xda886000 | 262144/4096 | 262144/4096 | 64 | 0.000% 0xf60c7c00 | 16512/128 | 16512/128 | 129 | 0.000% 0xf59a4118 | 13120/40 | 13120/40 | 328 | 0.000% 0xdfd4b2c0 | 11264/88 | 11264/88 | 128 | 0.000% 0xf5274600 | 7680/256 | 7680/256 | 30 | 0.000% 0xe8395000 | 5948/594 | 5464/546 | 10 | 8.137% 0xe59c3c00 | 5748/479 | 5712/476 | 12 | 0.626% 0xf4cd1a80 | 3524/44 | 3520/44 | 80 | 0.114% 0xe5bd1600 | 2892/482 | 2856/476 | 6 | 1.245% ... | ... | ... | ... | ... ------------------------------------------------------------------------------ SUMMARY ======= Total bytes requested: 2333626 Total bytes allocated: 2353712 Total bytes wasted on internal fragmentation: 20086 Internal fragmentation: 0.853375% TODO: - show sym+offset in 'callsite' column - show cross node allocation stats - collect more useful stats? - ... Signed-off-by: Li Zefan <lizf@cn.fujitsu.com> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro> Cc: linux-mm@kvack.org <linux-mm@kvack.org> LKML-Reference: <4B064AF5.9060208@cn.fujitsu.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'tools/perf/builtin-kmem.c')
-rw-r--r--tools/perf/builtin-kmem.c578
1 files changed, 578 insertions, 0 deletions
diff --git a/tools/perf/builtin-kmem.c b/tools/perf/builtin-kmem.c
new file mode 100644
index 000000000000..f315b052f819
--- /dev/null
+++ b/tools/perf/builtin-kmem.c
@@ -0,0 +1,578 @@
1#include "builtin.h"
2#include "perf.h"
3
4#include "util/util.h"
5#include "util/cache.h"
6#include "util/symbol.h"
7#include "util/thread.h"
8#include "util/header.h"
9
10#include "util/parse-options.h"
11#include "util/trace-event.h"
12
13#include "util/debug.h"
14#include "util/data_map.h"
15
16#include <linux/rbtree.h>
17
18struct alloc_stat;
19typedef int (*sort_fn_t)(struct alloc_stat *, struct alloc_stat *);
20
21static char const *input_name = "perf.data";
22
23static struct perf_header *header;
24static u64 sample_type;
25
26static int alloc_flag;
27static int caller_flag;
28
29sort_fn_t alloc_sort_fn;
30sort_fn_t caller_sort_fn;
31
32static int alloc_lines = -1;
33static int caller_lines = -1;
34
35static char *cwd;
36static int cwdlen;
37
38struct alloc_stat {
39 union {
40 struct {
41 char *name;
42 u64 call_site;
43 };
44 u64 ptr;
45 };
46 u64 bytes_req;
47 u64 bytes_alloc;
48 u32 hit;
49
50 struct rb_node node;
51};
52
53static struct rb_root root_alloc_stat;
54static struct rb_root root_alloc_sorted;
55static struct rb_root root_caller_stat;
56static struct rb_root root_caller_sorted;
57
58static unsigned long total_requested, total_allocated;
59
60struct raw_event_sample {
61 u32 size;
62 char data[0];
63};
64
65static int
66process_comm_event(event_t *event, unsigned long offset, unsigned long head)
67{
68 struct thread *thread = threads__findnew(event->comm.pid);
69
70 dump_printf("%p [%p]: PERF_RECORD_COMM: %s:%d\n",
71 (void *)(offset + head),
72 (void *)(long)(event->header.size),
73 event->comm.comm, event->comm.pid);
74
75 if (thread == NULL ||
76 thread__set_comm(thread, event->comm.comm)) {
77 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
78 return -1;
79 }
80
81 return 0;
82}
83
84static void insert_alloc_stat(unsigned long ptr,
85 int bytes_req, int bytes_alloc)
86{
87 struct rb_node **node = &root_alloc_stat.rb_node;
88 struct rb_node *parent = NULL;
89 struct alloc_stat *data = NULL;
90
91 if (!alloc_flag)
92 return;
93
94 while (*node) {
95 parent = *node;
96 data = rb_entry(*node, struct alloc_stat, node);
97
98 if (ptr > data->ptr)
99 node = &(*node)->rb_right;
100 else if (ptr < data->ptr)
101 node = &(*node)->rb_left;
102 else
103 break;
104 }
105
106 if (data && data->ptr == ptr) {
107 data->hit++;
108 data->bytes_req += bytes_req;
109 data->bytes_alloc += bytes_req;
110 } else {
111 data = malloc(sizeof(*data));
112 data->ptr = ptr;
113 data->hit = 1;
114 data->bytes_req = bytes_req;
115 data->bytes_alloc = bytes_alloc;
116
117 rb_link_node(&data->node, parent, node);
118 rb_insert_color(&data->node, &root_alloc_stat);
119 }
120}
121
122static void insert_caller_stat(unsigned long call_site,
123 int bytes_req, int bytes_alloc)
124{
125 struct rb_node **node = &root_caller_stat.rb_node;
126 struct rb_node *parent = NULL;
127 struct alloc_stat *data = NULL;
128
129 if (!caller_flag)
130 return;
131
132 while (*node) {
133 parent = *node;
134 data = rb_entry(*node, struct alloc_stat, node);
135
136 if (call_site > data->call_site)
137 node = &(*node)->rb_right;
138 else if (call_site < data->call_site)
139 node = &(*node)->rb_left;
140 else
141 break;
142 }
143
144 if (data && data->call_site == call_site) {
145 data->hit++;
146 data->bytes_req += bytes_req;
147 data->bytes_alloc += bytes_req;
148 } else {
149 data = malloc(sizeof(*data));
150 data->call_site = call_site;
151 data->hit = 1;
152 data->bytes_req = bytes_req;
153 data->bytes_alloc = bytes_alloc;
154
155 rb_link_node(&data->node, parent, node);
156 rb_insert_color(&data->node, &root_caller_stat);
157 }
158}
159
160static void process_alloc_event(struct raw_event_sample *raw,
161 struct event *event,
162 int cpu __used,
163 u64 timestamp __used,
164 struct thread *thread __used,
165 int node __used)
166{
167 unsigned long call_site;
168 unsigned long ptr;
169 int bytes_req;
170 int bytes_alloc;
171
172 ptr = raw_field_value(event, "ptr", raw->data);
173 call_site = raw_field_value(event, "call_site", raw->data);
174 bytes_req = raw_field_value(event, "bytes_req", raw->data);
175 bytes_alloc = raw_field_value(event, "bytes_alloc", raw->data);
176
177 insert_alloc_stat(ptr, bytes_req, bytes_alloc);
178 insert_caller_stat(call_site, bytes_req, bytes_alloc);
179
180 total_requested += bytes_req;
181 total_allocated += bytes_alloc;
182}
183
184static void process_free_event(struct raw_event_sample *raw __used,
185 struct event *event __used,
186 int cpu __used,
187 u64 timestamp __used,
188 struct thread *thread __used)
189{
190}
191
192static void
193process_raw_event(event_t *raw_event __used, void *more_data,
194 int cpu, u64 timestamp, struct thread *thread)
195{
196 struct raw_event_sample *raw = more_data;
197 struct event *event;
198 int type;
199
200 type = trace_parse_common_type(raw->data);
201 event = trace_find_event(type);
202
203 if (!strcmp(event->name, "kmalloc") ||
204 !strcmp(event->name, "kmem_cache_alloc")) {
205 process_alloc_event(raw, event, cpu, timestamp, thread, 0);
206 return;
207 }
208
209 if (!strcmp(event->name, "kmalloc_node") ||
210 !strcmp(event->name, "kmem_cache_alloc_node")) {
211 process_alloc_event(raw, event, cpu, timestamp, thread, 1);
212 return;
213 }
214
215 if (!strcmp(event->name, "kfree") ||
216 !strcmp(event->name, "kmem_cache_free")) {
217 process_free_event(raw, event, cpu, timestamp, thread);
218 return;
219 }
220}
221
222static int
223process_sample_event(event_t *event, unsigned long offset, unsigned long head)
224{
225 u64 ip = event->ip.ip;
226 u64 timestamp = -1;
227 u32 cpu = -1;
228 u64 period = 1;
229 void *more_data = event->ip.__more_data;
230 struct thread *thread = threads__findnew(event->ip.pid);
231
232 if (sample_type & PERF_SAMPLE_TIME) {
233 timestamp = *(u64 *)more_data;
234 more_data += sizeof(u64);
235 }
236
237 if (sample_type & PERF_SAMPLE_CPU) {
238 cpu = *(u32 *)more_data;
239 more_data += sizeof(u32);
240 more_data += sizeof(u32); /* reserved */
241 }
242
243 if (sample_type & PERF_SAMPLE_PERIOD) {
244 period = *(u64 *)more_data;
245 more_data += sizeof(u64);
246 }
247
248 dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
249 (void *)(offset + head),
250 (void *)(long)(event->header.size),
251 event->header.misc,
252 event->ip.pid, event->ip.tid,
253 (void *)(long)ip,
254 (long long)period);
255
256 if (thread == NULL) {
257 pr_debug("problem processing %d event, skipping it.\n",
258 event->header.type);
259 return -1;
260 }
261
262 dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
263
264 process_raw_event(event, more_data, cpu, timestamp, thread);
265
266 return 0;
267}
268
269static int sample_type_check(u64 type)
270{
271 sample_type = type;
272
273 if (!(sample_type & PERF_SAMPLE_RAW)) {
274 fprintf(stderr,
275 "No trace sample to read. Did you call perf record "
276 "without -R?");
277 return -1;
278 }
279
280 return 0;
281}
282
283static struct perf_file_handler file_handler = {
284 .process_sample_event = process_sample_event,
285 .process_comm_event = process_comm_event,
286 .sample_type_check = sample_type_check,
287};
288
289static int read_events(void)
290{
291 register_idle_thread();
292 register_perf_file_handler(&file_handler);
293
294 return mmap_dispatch_perf_file(&header, input_name, 0, 0,
295 &cwdlen, &cwd);
296}
297
298static double fragmentation(unsigned long n_req, unsigned long n_alloc)
299{
300 if (n_alloc == 0)
301 return 0.0;
302 else
303 return 100.0 - (100.0 * n_req / n_alloc);
304}
305
306static void __print_result(struct rb_root *root, int n_lines, int is_caller)
307{
308 struct rb_node *next;
309
310 printf("\n ------------------------------------------------------------------------------\n");
311 if (is_caller)
312 printf(" Callsite |");
313 else
314 printf(" Alloc Ptr |");
315 printf(" Total_alloc/Per | Total_req/Per | Hit | Fragmentation\n");
316 printf(" ------------------------------------------------------------------------------\n");
317
318 next = rb_first(root);
319
320 while (next && n_lines--) {
321 struct alloc_stat *data;
322
323 data = rb_entry(next, struct alloc_stat, node);
324
325 printf(" %-16p | %8llu/%-6lu | %8llu/%-6lu | %6lu | %8.3f%%\n",
326 is_caller ? (void *)(unsigned long)data->call_site :
327 (void *)(unsigned long)data->ptr,
328 (unsigned long long)data->bytes_alloc,
329 (unsigned long)data->bytes_alloc / data->hit,
330 (unsigned long long)data->bytes_req,
331 (unsigned long)data->bytes_req / data->hit,
332 (unsigned long)data->hit,
333 fragmentation(data->bytes_req, data->bytes_alloc));
334
335 next = rb_next(next);
336 }
337
338 if (n_lines == -1)
339 printf(" ... | ... | ... | ... | ... \n");
340
341 printf(" ------------------------------------------------------------------------------\n");
342}
343
344static void print_summary(void)
345{
346 printf("\nSUMMARY\n=======\n");
347 printf("Total bytes requested: %lu\n", total_requested);
348 printf("Total bytes allocated: %lu\n", total_allocated);
349 printf("Total bytes wasted on internal fragmentation: %lu\n",
350 total_allocated - total_requested);
351 printf("Internal fragmentation: %f%%\n",
352 fragmentation(total_requested, total_allocated));
353}
354
355static void print_result(void)
356{
357 if (caller_flag)
358 __print_result(&root_caller_sorted, caller_lines, 1);
359 if (alloc_flag)
360 __print_result(&root_alloc_sorted, alloc_lines, 0);
361 print_summary();
362}
363
364static void sort_insert(struct rb_root *root, struct alloc_stat *data,
365 sort_fn_t sort_fn)
366{
367 struct rb_node **new = &(root->rb_node);
368 struct rb_node *parent = NULL;
369
370 while (*new) {
371 struct alloc_stat *this;
372 int cmp;
373
374 this = rb_entry(*new, struct alloc_stat, node);
375 parent = *new;
376
377 cmp = sort_fn(data, this);
378
379 if (cmp > 0)
380 new = &((*new)->rb_left);
381 else
382 new = &((*new)->rb_right);
383 }
384
385 rb_link_node(&data->node, parent, new);
386 rb_insert_color(&data->node, root);
387}
388
389static void __sort_result(struct rb_root *root, struct rb_root *root_sorted,
390 sort_fn_t sort_fn)
391{
392 struct rb_node *node;
393 struct alloc_stat *data;
394
395 for (;;) {
396 node = rb_first(root);
397 if (!node)
398 break;
399
400 rb_erase(node, root);
401 data = rb_entry(node, struct alloc_stat, node);
402 sort_insert(root_sorted, data, sort_fn);
403 }
404}
405
406static void sort_result(void)
407{
408 __sort_result(&root_alloc_stat, &root_alloc_sorted, alloc_sort_fn);
409 __sort_result(&root_caller_stat, &root_caller_sorted, caller_sort_fn);
410}
411
412static int __cmd_kmem(void)
413{
414 setup_pager();
415 read_events();
416 sort_result();
417 print_result();
418
419 return 0;
420}
421
422static const char * const kmem_usage[] = {
423 "perf kmem [<options>] {record}",
424 NULL
425};
426
427
428static int ptr_cmp(struct alloc_stat *l, struct alloc_stat *r)
429{
430 if (l->ptr < r->ptr)
431 return -1;
432 else if (l->ptr > r->ptr)
433 return 1;
434 return 0;
435}
436
437static int callsite_cmp(struct alloc_stat *l, struct alloc_stat *r)
438{
439 if (l->call_site < r->call_site)
440 return -1;
441 else if (l->call_site > r->call_site)
442 return 1;
443 return 0;
444}
445
446static int bytes_cmp(struct alloc_stat *l, struct alloc_stat *r)
447{
448 if (l->bytes_alloc < r->bytes_alloc)
449 return -1;
450 else if (l->bytes_alloc > r->bytes_alloc)
451 return 1;
452 return 0;
453}
454
455static int parse_sort_opt(const struct option *opt __used,
456 const char *arg, int unset __used)
457{
458 sort_fn_t sort_fn;
459
460 if (!arg)
461 return -1;
462
463 if (strcmp(arg, "ptr") == 0)
464 sort_fn = ptr_cmp;
465 else if (strcmp(arg, "call_site") == 0)
466 sort_fn = callsite_cmp;
467 else if (strcmp(arg, "bytes") == 0)
468 sort_fn = bytes_cmp;
469 else
470 return -1;
471
472 if (caller_flag > alloc_flag)
473 caller_sort_fn = sort_fn;
474 else
475 alloc_sort_fn = sort_fn;
476
477 return 0;
478}
479
480static int parse_stat_opt(const struct option *opt __used,
481 const char *arg, int unset __used)
482{
483 if (!arg)
484 return -1;
485
486 if (strcmp(arg, "alloc") == 0)
487 alloc_flag = (caller_flag + 1);
488 else if (strcmp(arg, "caller") == 0)
489 caller_flag = (alloc_flag + 1);
490 else
491 return -1;
492 return 0;
493}
494
495static int parse_line_opt(const struct option *opt __used,
496 const char *arg, int unset __used)
497{
498 int lines;
499
500 if (!arg)
501 return -1;
502
503 lines = strtoul(arg, NULL, 10);
504
505 if (caller_flag > alloc_flag)
506 caller_lines = lines;
507 else
508 alloc_lines = lines;
509
510 return 0;
511}
512
513static const struct option kmem_options[] = {
514 OPT_STRING('i', "input", &input_name, "file",
515 "input file name"),
516 OPT_CALLBACK(0, "stat", NULL, "<alloc>|<caller>",
517 "stat selector, Pass 'alloc' or 'caller'.",
518 parse_stat_opt),
519 OPT_CALLBACK('s', "sort", NULL, "key",
520 "sort by key: ptr, call_site, hit, bytes",
521 parse_sort_opt),
522 OPT_CALLBACK('l', "line", NULL, "num",
523 "show n lins",
524 parse_line_opt),
525 OPT_END()
526};
527
528static const char *record_args[] = {
529 "record",
530 "-a",
531 "-R",
532 "-M",
533 "-f",
534 "-c", "1",
535 "-e", "kmem:kmalloc",
536 "-e", "kmem:kmalloc_node",
537 "-e", "kmem:kfree",
538 "-e", "kmem:kmem_cache_alloc",
539 "-e", "kmem:kmem_cache_alloc_node",
540 "-e", "kmem:kmem_cache_free",
541};
542
543static int __cmd_record(int argc, const char **argv)
544{
545 unsigned int rec_argc, i, j;
546 const char **rec_argv;
547
548 rec_argc = ARRAY_SIZE(record_args) + argc - 1;
549 rec_argv = calloc(rec_argc + 1, sizeof(char *));
550
551 for (i = 0; i < ARRAY_SIZE(record_args); i++)
552 rec_argv[i] = strdup(record_args[i]);
553
554 for (j = 1; j < (unsigned int)argc; j++, i++)
555 rec_argv[i] = argv[j];
556
557 return cmd_record(i, rec_argv, NULL);
558}
559
560int cmd_kmem(int argc, const char **argv, const char *prefix __used)
561{
562 symbol__init(0);
563
564 argc = parse_options(argc, argv, kmem_options, kmem_usage, 0);
565
566 if (argc && !strncmp(argv[0], "rec", 3))
567 return __cmd_record(argc, argv);
568 else if (argc)
569 usage_with_options(kmem_usage, kmem_options);
570
571 if (!alloc_sort_fn)
572 alloc_sort_fn = bytes_cmp;
573 if (!caller_sort_fn)
574 caller_sort_fn = bytes_cmp;
575
576 return __cmd_kmem();
577}
578
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-10 07:20:32 -0500