xref: /openbmc/linux/tools/perf/builtin-kmem.c (revision 4800cd83)
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 #include "util/session.h"
10 
11 #include "util/parse-options.h"
12 #include "util/trace-event.h"
13 
14 #include "util/debug.h"
15 
16 #include <linux/rbtree.h>
17 
18 struct alloc_stat;
19 typedef int (*sort_fn_t)(struct alloc_stat *, struct alloc_stat *);
20 
21 static char const		*input_name = "perf.data";
22 
23 static int			alloc_flag;
24 static int			caller_flag;
25 
26 static int			alloc_lines = -1;
27 static int			caller_lines = -1;
28 
29 static bool			raw_ip;
30 
31 static char			default_sort_order[] = "frag,hit,bytes";
32 
33 static int			*cpunode_map;
34 static int			max_cpu_num;
35 
36 struct alloc_stat {
37 	u64	call_site;
38 	u64	ptr;
39 	u64	bytes_req;
40 	u64	bytes_alloc;
41 	u32	hit;
42 	u32	pingpong;
43 
44 	short	alloc_cpu;
45 
46 	struct rb_node node;
47 };
48 
49 static struct rb_root root_alloc_stat;
50 static struct rb_root root_alloc_sorted;
51 static struct rb_root root_caller_stat;
52 static struct rb_root root_caller_sorted;
53 
54 static unsigned long total_requested, total_allocated;
55 static unsigned long nr_allocs, nr_cross_allocs;
56 
57 #define PATH_SYS_NODE	"/sys/devices/system/node"
58 
59 static void init_cpunode_map(void)
60 {
61 	FILE *fp;
62 	int i;
63 
64 	fp = fopen("/sys/devices/system/cpu/kernel_max", "r");
65 	if (!fp) {
66 		max_cpu_num = 4096;
67 		return;
68 	}
69 
70 	if (fscanf(fp, "%d", &max_cpu_num) < 1)
71 		die("Failed to read 'kernel_max' from sysfs");
72 	max_cpu_num++;
73 
74 	cpunode_map = calloc(max_cpu_num, sizeof(int));
75 	if (!cpunode_map)
76 		die("calloc");
77 	for (i = 0; i < max_cpu_num; i++)
78 		cpunode_map[i] = -1;
79 	fclose(fp);
80 }
81 
82 static void setup_cpunode_map(void)
83 {
84 	struct dirent *dent1, *dent2;
85 	DIR *dir1, *dir2;
86 	unsigned int cpu, mem;
87 	char buf[PATH_MAX];
88 
89 	init_cpunode_map();
90 
91 	dir1 = opendir(PATH_SYS_NODE);
92 	if (!dir1)
93 		return;
94 
95 	while ((dent1 = readdir(dir1)) != NULL) {
96 		if (dent1->d_type != DT_DIR ||
97 		    sscanf(dent1->d_name, "node%u", &mem) < 1)
98 			continue;
99 
100 		snprintf(buf, PATH_MAX, "%s/%s", PATH_SYS_NODE, dent1->d_name);
101 		dir2 = opendir(buf);
102 		if (!dir2)
103 			continue;
104 		while ((dent2 = readdir(dir2)) != NULL) {
105 			if (dent2->d_type != DT_LNK ||
106 			    sscanf(dent2->d_name, "cpu%u", &cpu) < 1)
107 				continue;
108 			cpunode_map[cpu] = mem;
109 		}
110 	}
111 }
112 
113 static void insert_alloc_stat(unsigned long call_site, unsigned long ptr,
114 			      int bytes_req, int bytes_alloc, int cpu)
115 {
116 	struct rb_node **node = &root_alloc_stat.rb_node;
117 	struct rb_node *parent = NULL;
118 	struct alloc_stat *data = NULL;
119 
120 	while (*node) {
121 		parent = *node;
122 		data = rb_entry(*node, struct alloc_stat, node);
123 
124 		if (ptr > data->ptr)
125 			node = &(*node)->rb_right;
126 		else if (ptr < data->ptr)
127 			node = &(*node)->rb_left;
128 		else
129 			break;
130 	}
131 
132 	if (data && data->ptr == ptr) {
133 		data->hit++;
134 		data->bytes_req += bytes_req;
135 		data->bytes_alloc += bytes_alloc;
136 	} else {
137 		data = malloc(sizeof(*data));
138 		if (!data)
139 			die("malloc");
140 		data->ptr = ptr;
141 		data->pingpong = 0;
142 		data->hit = 1;
143 		data->bytes_req = bytes_req;
144 		data->bytes_alloc = bytes_alloc;
145 
146 		rb_link_node(&data->node, parent, node);
147 		rb_insert_color(&data->node, &root_alloc_stat);
148 	}
149 	data->call_site = call_site;
150 	data->alloc_cpu = cpu;
151 }
152 
153 static void insert_caller_stat(unsigned long call_site,
154 			      int bytes_req, int bytes_alloc)
155 {
156 	struct rb_node **node = &root_caller_stat.rb_node;
157 	struct rb_node *parent = NULL;
158 	struct alloc_stat *data = NULL;
159 
160 	while (*node) {
161 		parent = *node;
162 		data = rb_entry(*node, struct alloc_stat, node);
163 
164 		if (call_site > data->call_site)
165 			node = &(*node)->rb_right;
166 		else if (call_site < data->call_site)
167 			node = &(*node)->rb_left;
168 		else
169 			break;
170 	}
171 
172 	if (data && data->call_site == call_site) {
173 		data->hit++;
174 		data->bytes_req += bytes_req;
175 		data->bytes_alloc += bytes_alloc;
176 	} else {
177 		data = malloc(sizeof(*data));
178 		if (!data)
179 			die("malloc");
180 		data->call_site = call_site;
181 		data->pingpong = 0;
182 		data->hit = 1;
183 		data->bytes_req = bytes_req;
184 		data->bytes_alloc = bytes_alloc;
185 
186 		rb_link_node(&data->node, parent, node);
187 		rb_insert_color(&data->node, &root_caller_stat);
188 	}
189 }
190 
191 static void process_alloc_event(void *data,
192 				struct event *event,
193 				int cpu,
194 				u64 timestamp __used,
195 				struct thread *thread __used,
196 				int node)
197 {
198 	unsigned long call_site;
199 	unsigned long ptr;
200 	int bytes_req;
201 	int bytes_alloc;
202 	int node1, node2;
203 
204 	ptr = raw_field_value(event, "ptr", data);
205 	call_site = raw_field_value(event, "call_site", data);
206 	bytes_req = raw_field_value(event, "bytes_req", data);
207 	bytes_alloc = raw_field_value(event, "bytes_alloc", data);
208 
209 	insert_alloc_stat(call_site, ptr, bytes_req, bytes_alloc, cpu);
210 	insert_caller_stat(call_site, bytes_req, bytes_alloc);
211 
212 	total_requested += bytes_req;
213 	total_allocated += bytes_alloc;
214 
215 	if (node) {
216 		node1 = cpunode_map[cpu];
217 		node2 = raw_field_value(event, "node", data);
218 		if (node1 != node2)
219 			nr_cross_allocs++;
220 	}
221 	nr_allocs++;
222 }
223 
224 static int ptr_cmp(struct alloc_stat *, struct alloc_stat *);
225 static int callsite_cmp(struct alloc_stat *, struct alloc_stat *);
226 
227 static struct alloc_stat *search_alloc_stat(unsigned long ptr,
228 					    unsigned long call_site,
229 					    struct rb_root *root,
230 					    sort_fn_t sort_fn)
231 {
232 	struct rb_node *node = root->rb_node;
233 	struct alloc_stat key = { .ptr = ptr, .call_site = call_site };
234 
235 	while (node) {
236 		struct alloc_stat *data;
237 		int cmp;
238 
239 		data = rb_entry(node, struct alloc_stat, node);
240 
241 		cmp = sort_fn(&key, data);
242 		if (cmp < 0)
243 			node = node->rb_left;
244 		else if (cmp > 0)
245 			node = node->rb_right;
246 		else
247 			return data;
248 	}
249 	return NULL;
250 }
251 
252 static void process_free_event(void *data,
253 			       struct event *event,
254 			       int cpu,
255 			       u64 timestamp __used,
256 			       struct thread *thread __used)
257 {
258 	unsigned long ptr;
259 	struct alloc_stat *s_alloc, *s_caller;
260 
261 	ptr = raw_field_value(event, "ptr", data);
262 
263 	s_alloc = search_alloc_stat(ptr, 0, &root_alloc_stat, ptr_cmp);
264 	if (!s_alloc)
265 		return;
266 
267 	if (cpu != s_alloc->alloc_cpu) {
268 		s_alloc->pingpong++;
269 
270 		s_caller = search_alloc_stat(0, s_alloc->call_site,
271 					     &root_caller_stat, callsite_cmp);
272 		assert(s_caller);
273 		s_caller->pingpong++;
274 	}
275 	s_alloc->alloc_cpu = -1;
276 }
277 
278 static void
279 process_raw_event(event_t *raw_event __used, void *data,
280 		  int cpu, u64 timestamp, struct thread *thread)
281 {
282 	struct event *event;
283 	int type;
284 
285 	type = trace_parse_common_type(data);
286 	event = trace_find_event(type);
287 
288 	if (!strcmp(event->name, "kmalloc") ||
289 	    !strcmp(event->name, "kmem_cache_alloc")) {
290 		process_alloc_event(data, event, cpu, timestamp, thread, 0);
291 		return;
292 	}
293 
294 	if (!strcmp(event->name, "kmalloc_node") ||
295 	    !strcmp(event->name, "kmem_cache_alloc_node")) {
296 		process_alloc_event(data, event, cpu, timestamp, thread, 1);
297 		return;
298 	}
299 
300 	if (!strcmp(event->name, "kfree") ||
301 	    !strcmp(event->name, "kmem_cache_free")) {
302 		process_free_event(data, event, cpu, timestamp, thread);
303 		return;
304 	}
305 }
306 
307 static int process_sample_event(event_t *event, struct sample_data *sample,
308 				struct perf_session *session)
309 {
310 	struct thread *thread = perf_session__findnew(session, event->ip.pid);
311 
312 	if (thread == NULL) {
313 		pr_debug("problem processing %d event, skipping it.\n",
314 			 event->header.type);
315 		return -1;
316 	}
317 
318 	dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
319 
320 	process_raw_event(event, sample->raw_data, sample->cpu,
321 			  sample->time, thread);
322 
323 	return 0;
324 }
325 
326 static struct perf_event_ops event_ops = {
327 	.sample			= process_sample_event,
328 	.comm			= event__process_comm,
329 	.ordered_samples	= true,
330 };
331 
332 static double fragmentation(unsigned long n_req, unsigned long n_alloc)
333 {
334 	if (n_alloc == 0)
335 		return 0.0;
336 	else
337 		return 100.0 - (100.0 * n_req / n_alloc);
338 }
339 
340 static void __print_result(struct rb_root *root, struct perf_session *session,
341 			   int n_lines, int is_caller)
342 {
343 	struct rb_node *next;
344 	struct machine *machine;
345 
346 	printf("%.102s\n", graph_dotted_line);
347 	printf(" %-34s |",  is_caller ? "Callsite": "Alloc Ptr");
348 	printf(" Total_alloc/Per | Total_req/Per   | Hit      | Ping-pong | Frag\n");
349 	printf("%.102s\n", graph_dotted_line);
350 
351 	next = rb_first(root);
352 
353 	machine = perf_session__find_host_machine(session);
354 	if (!machine) {
355 		pr_err("__print_result: couldn't find kernel information\n");
356 		return;
357 	}
358 	while (next && n_lines--) {
359 		struct alloc_stat *data = rb_entry(next, struct alloc_stat,
360 						   node);
361 		struct symbol *sym = NULL;
362 		struct map *map;
363 		char buf[BUFSIZ];
364 		u64 addr;
365 
366 		if (is_caller) {
367 			addr = data->call_site;
368 			if (!raw_ip)
369 				sym = machine__find_kernel_function(machine, addr, &map, NULL);
370 		} else
371 			addr = data->ptr;
372 
373 		if (sym != NULL)
374 			snprintf(buf, sizeof(buf), "%s+%" PRIx64 "", sym->name,
375 				 addr - map->unmap_ip(map, sym->start));
376 		else
377 			snprintf(buf, sizeof(buf), "%#" PRIx64 "", addr);
378 		printf(" %-34s |", buf);
379 
380 		printf(" %9llu/%-5lu | %9llu/%-5lu | %8lu | %8lu | %6.3f%%\n",
381 		       (unsigned long long)data->bytes_alloc,
382 		       (unsigned long)data->bytes_alloc / data->hit,
383 		       (unsigned long long)data->bytes_req,
384 		       (unsigned long)data->bytes_req / data->hit,
385 		       (unsigned long)data->hit,
386 		       (unsigned long)data->pingpong,
387 		       fragmentation(data->bytes_req, data->bytes_alloc));
388 
389 		next = rb_next(next);
390 	}
391 
392 	if (n_lines == -1)
393 		printf(" ...                                | ...             | ...             | ...    | ...      | ...   \n");
394 
395 	printf("%.102s\n", graph_dotted_line);
396 }
397 
398 static void print_summary(void)
399 {
400 	printf("\nSUMMARY\n=======\n");
401 	printf("Total bytes requested: %lu\n", total_requested);
402 	printf("Total bytes allocated: %lu\n", total_allocated);
403 	printf("Total bytes wasted on internal fragmentation: %lu\n",
404 	       total_allocated - total_requested);
405 	printf("Internal fragmentation: %f%%\n",
406 	       fragmentation(total_requested, total_allocated));
407 	printf("Cross CPU allocations: %lu/%lu\n", nr_cross_allocs, nr_allocs);
408 }
409 
410 static void print_result(struct perf_session *session)
411 {
412 	if (caller_flag)
413 		__print_result(&root_caller_sorted, session, caller_lines, 1);
414 	if (alloc_flag)
415 		__print_result(&root_alloc_sorted, session, alloc_lines, 0);
416 	print_summary();
417 }
418 
419 struct sort_dimension {
420 	const char		name[20];
421 	sort_fn_t		cmp;
422 	struct list_head	list;
423 };
424 
425 static LIST_HEAD(caller_sort);
426 static LIST_HEAD(alloc_sort);
427 
428 static void sort_insert(struct rb_root *root, struct alloc_stat *data,
429 			struct list_head *sort_list)
430 {
431 	struct rb_node **new = &(root->rb_node);
432 	struct rb_node *parent = NULL;
433 	struct sort_dimension *sort;
434 
435 	while (*new) {
436 		struct alloc_stat *this;
437 		int cmp = 0;
438 
439 		this = rb_entry(*new, struct alloc_stat, node);
440 		parent = *new;
441 
442 		list_for_each_entry(sort, sort_list, list) {
443 			cmp = sort->cmp(data, this);
444 			if (cmp)
445 				break;
446 		}
447 
448 		if (cmp > 0)
449 			new = &((*new)->rb_left);
450 		else
451 			new = &((*new)->rb_right);
452 	}
453 
454 	rb_link_node(&data->node, parent, new);
455 	rb_insert_color(&data->node, root);
456 }
457 
458 static void __sort_result(struct rb_root *root, struct rb_root *root_sorted,
459 			  struct list_head *sort_list)
460 {
461 	struct rb_node *node;
462 	struct alloc_stat *data;
463 
464 	for (;;) {
465 		node = rb_first(root);
466 		if (!node)
467 			break;
468 
469 		rb_erase(node, root);
470 		data = rb_entry(node, struct alloc_stat, node);
471 		sort_insert(root_sorted, data, sort_list);
472 	}
473 }
474 
475 static void sort_result(void)
476 {
477 	__sort_result(&root_alloc_stat, &root_alloc_sorted, &alloc_sort);
478 	__sort_result(&root_caller_stat, &root_caller_sorted, &caller_sort);
479 }
480 
481 static int __cmd_kmem(void)
482 {
483 	int err = -EINVAL;
484 	struct perf_session *session = perf_session__new(input_name, O_RDONLY,
485 							 0, false, &event_ops);
486 	if (session == NULL)
487 		return -ENOMEM;
488 
489 	if (perf_session__create_kernel_maps(session) < 0)
490 		goto out_delete;
491 
492 	if (!perf_session__has_traces(session, "kmem record"))
493 		goto out_delete;
494 
495 	setup_pager();
496 	err = perf_session__process_events(session, &event_ops);
497 	if (err != 0)
498 		goto out_delete;
499 	sort_result();
500 	print_result(session);
501 out_delete:
502 	perf_session__delete(session);
503 	return err;
504 }
505 
506 static const char * const kmem_usage[] = {
507 	"perf kmem [<options>] {record|stat}",
508 	NULL
509 };
510 
511 static int ptr_cmp(struct alloc_stat *l, struct alloc_stat *r)
512 {
513 	if (l->ptr < r->ptr)
514 		return -1;
515 	else if (l->ptr > r->ptr)
516 		return 1;
517 	return 0;
518 }
519 
520 static struct sort_dimension ptr_sort_dimension = {
521 	.name	= "ptr",
522 	.cmp	= ptr_cmp,
523 };
524 
525 static int callsite_cmp(struct alloc_stat *l, struct alloc_stat *r)
526 {
527 	if (l->call_site < r->call_site)
528 		return -1;
529 	else if (l->call_site > r->call_site)
530 		return 1;
531 	return 0;
532 }
533 
534 static struct sort_dimension callsite_sort_dimension = {
535 	.name	= "callsite",
536 	.cmp	= callsite_cmp,
537 };
538 
539 static int hit_cmp(struct alloc_stat *l, struct alloc_stat *r)
540 {
541 	if (l->hit < r->hit)
542 		return -1;
543 	else if (l->hit > r->hit)
544 		return 1;
545 	return 0;
546 }
547 
548 static struct sort_dimension hit_sort_dimension = {
549 	.name	= "hit",
550 	.cmp	= hit_cmp,
551 };
552 
553 static int bytes_cmp(struct alloc_stat *l, struct alloc_stat *r)
554 {
555 	if (l->bytes_alloc < r->bytes_alloc)
556 		return -1;
557 	else if (l->bytes_alloc > r->bytes_alloc)
558 		return 1;
559 	return 0;
560 }
561 
562 static struct sort_dimension bytes_sort_dimension = {
563 	.name	= "bytes",
564 	.cmp	= bytes_cmp,
565 };
566 
567 static int frag_cmp(struct alloc_stat *l, struct alloc_stat *r)
568 {
569 	double x, y;
570 
571 	x = fragmentation(l->bytes_req, l->bytes_alloc);
572 	y = fragmentation(r->bytes_req, r->bytes_alloc);
573 
574 	if (x < y)
575 		return -1;
576 	else if (x > y)
577 		return 1;
578 	return 0;
579 }
580 
581 static struct sort_dimension frag_sort_dimension = {
582 	.name	= "frag",
583 	.cmp	= frag_cmp,
584 };
585 
586 static int pingpong_cmp(struct alloc_stat *l, struct alloc_stat *r)
587 {
588 	if (l->pingpong < r->pingpong)
589 		return -1;
590 	else if (l->pingpong > r->pingpong)
591 		return 1;
592 	return 0;
593 }
594 
595 static struct sort_dimension pingpong_sort_dimension = {
596 	.name	= "pingpong",
597 	.cmp	= pingpong_cmp,
598 };
599 
600 static struct sort_dimension *avail_sorts[] = {
601 	&ptr_sort_dimension,
602 	&callsite_sort_dimension,
603 	&hit_sort_dimension,
604 	&bytes_sort_dimension,
605 	&frag_sort_dimension,
606 	&pingpong_sort_dimension,
607 };
608 
609 #define NUM_AVAIL_SORTS	\
610 	(int)(sizeof(avail_sorts) / sizeof(struct sort_dimension *))
611 
612 static int sort_dimension__add(const char *tok, struct list_head *list)
613 {
614 	struct sort_dimension *sort;
615 	int i;
616 
617 	for (i = 0; i < NUM_AVAIL_SORTS; i++) {
618 		if (!strcmp(avail_sorts[i]->name, tok)) {
619 			sort = malloc(sizeof(*sort));
620 			if (!sort)
621 				die("malloc");
622 			memcpy(sort, avail_sorts[i], sizeof(*sort));
623 			list_add_tail(&sort->list, list);
624 			return 0;
625 		}
626 	}
627 
628 	return -1;
629 }
630 
631 static int setup_sorting(struct list_head *sort_list, const char *arg)
632 {
633 	char *tok;
634 	char *str = strdup(arg);
635 
636 	if (!str)
637 		die("strdup");
638 
639 	while (true) {
640 		tok = strsep(&str, ",");
641 		if (!tok)
642 			break;
643 		if (sort_dimension__add(tok, sort_list) < 0) {
644 			error("Unknown --sort key: '%s'", tok);
645 			return -1;
646 		}
647 	}
648 
649 	free(str);
650 	return 0;
651 }
652 
653 static int parse_sort_opt(const struct option *opt __used,
654 			  const char *arg, int unset __used)
655 {
656 	if (!arg)
657 		return -1;
658 
659 	if (caller_flag > alloc_flag)
660 		return setup_sorting(&caller_sort, arg);
661 	else
662 		return setup_sorting(&alloc_sort, arg);
663 
664 	return 0;
665 }
666 
667 static int parse_caller_opt(const struct option *opt __used,
668 			  const char *arg __used, int unset __used)
669 {
670 	caller_flag = (alloc_flag + 1);
671 	return 0;
672 }
673 
674 static int parse_alloc_opt(const struct option *opt __used,
675 			  const char *arg __used, int unset __used)
676 {
677 	alloc_flag = (caller_flag + 1);
678 	return 0;
679 }
680 
681 static int parse_line_opt(const struct option *opt __used,
682 			  const char *arg, int unset __used)
683 {
684 	int lines;
685 
686 	if (!arg)
687 		return -1;
688 
689 	lines = strtoul(arg, NULL, 10);
690 
691 	if (caller_flag > alloc_flag)
692 		caller_lines = lines;
693 	else
694 		alloc_lines = lines;
695 
696 	return 0;
697 }
698 
699 static const struct option kmem_options[] = {
700 	OPT_STRING('i', "input", &input_name, "file",
701 		   "input file name"),
702 	OPT_CALLBACK_NOOPT(0, "caller", NULL, NULL,
703 			   "show per-callsite statistics",
704 			   parse_caller_opt),
705 	OPT_CALLBACK_NOOPT(0, "alloc", NULL, NULL,
706 			   "show per-allocation statistics",
707 			   parse_alloc_opt),
708 	OPT_CALLBACK('s', "sort", NULL, "key[,key2...]",
709 		     "sort by keys: ptr, call_site, bytes, hit, pingpong, frag",
710 		     parse_sort_opt),
711 	OPT_CALLBACK('l', "line", NULL, "num",
712 		     "show n lines",
713 		     parse_line_opt),
714 	OPT_BOOLEAN(0, "raw-ip", &raw_ip, "show raw ip instead of symbol"),
715 	OPT_END()
716 };
717 
718 static const char *record_args[] = {
719 	"record",
720 	"-a",
721 	"-R",
722 	"-f",
723 	"-c", "1",
724 	"-e", "kmem:kmalloc",
725 	"-e", "kmem:kmalloc_node",
726 	"-e", "kmem:kfree",
727 	"-e", "kmem:kmem_cache_alloc",
728 	"-e", "kmem:kmem_cache_alloc_node",
729 	"-e", "kmem:kmem_cache_free",
730 };
731 
732 static int __cmd_record(int argc, const char **argv)
733 {
734 	unsigned int rec_argc, i, j;
735 	const char **rec_argv;
736 
737 	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
738 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
739 
740 	if (rec_argv == NULL)
741 		return -ENOMEM;
742 
743 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
744 		rec_argv[i] = strdup(record_args[i]);
745 
746 	for (j = 1; j < (unsigned int)argc; j++, i++)
747 		rec_argv[i] = argv[j];
748 
749 	return cmd_record(i, rec_argv, NULL);
750 }
751 
752 int cmd_kmem(int argc, const char **argv, const char *prefix __used)
753 {
754 	argc = parse_options(argc, argv, kmem_options, kmem_usage, 0);
755 
756 	if (!argc)
757 		usage_with_options(kmem_usage, kmem_options);
758 
759 	symbol__init();
760 
761 	if (!strncmp(argv[0], "rec", 3)) {
762 		return __cmd_record(argc, argv);
763 	} else if (!strcmp(argv[0], "stat")) {
764 		setup_cpunode_map();
765 
766 		if (list_empty(&caller_sort))
767 			setup_sorting(&caller_sort, default_sort_order);
768 		if (list_empty(&alloc_sort))
769 			setup_sorting(&alloc_sort, default_sort_order);
770 
771 		return __cmd_kmem();
772 	} else
773 		usage_with_options(kmem_usage, kmem_options);
774 
775 	return 0;
776 }
777 
778