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 #include "util/tool.h"
12 #include "util/parse-options.h"
13 #include "util/trace-event.h"
15 #include "util/debug.h"
17 #include <linux/rbtree.h>
20 typedef int (*sort_fn_t
)(struct alloc_stat
*, struct alloc_stat
*);
22 static const char *input_name
;
24 static int alloc_flag
;
25 static int caller_flag
;
27 static int alloc_lines
= -1;
28 static int caller_lines
= -1;
32 static char default_sort_order
[] = "frag,hit,bytes";
34 static int *cpunode_map
;
35 static int max_cpu_num
;
50 static struct rb_root root_alloc_stat
;
51 static struct rb_root root_alloc_sorted
;
52 static struct rb_root root_caller_stat
;
53 static struct rb_root root_caller_sorted
;
55 static unsigned long total_requested
, total_allocated
;
56 static unsigned long nr_allocs
, nr_cross_allocs
;
58 #define PATH_SYS_NODE "/sys/devices/system/node"
60 static void init_cpunode_map(void)
65 fp
= fopen("/sys/devices/system/cpu/kernel_max", "r");
71 if (fscanf(fp
, "%d", &max_cpu_num
) < 1)
72 die("Failed to read 'kernel_max' from sysfs");
75 cpunode_map
= calloc(max_cpu_num
, sizeof(int));
78 for (i
= 0; i
< max_cpu_num
; i
++)
83 static void setup_cpunode_map(void)
85 struct dirent
*dent1
, *dent2
;
87 unsigned int cpu
, mem
;
92 dir1
= opendir(PATH_SYS_NODE
);
96 while ((dent1
= readdir(dir1
)) != NULL
) {
97 if (dent1
->d_type
!= DT_DIR
||
98 sscanf(dent1
->d_name
, "node%u", &mem
) < 1)
101 snprintf(buf
, PATH_MAX
, "%s/%s", PATH_SYS_NODE
, dent1
->d_name
);
105 while ((dent2
= readdir(dir2
)) != NULL
) {
106 if (dent2
->d_type
!= DT_LNK
||
107 sscanf(dent2
->d_name
, "cpu%u", &cpu
) < 1)
109 cpunode_map
[cpu
] = mem
;
116 static void insert_alloc_stat(unsigned long call_site
, unsigned long ptr
,
117 int bytes_req
, int bytes_alloc
, int cpu
)
119 struct rb_node
**node
= &root_alloc_stat
.rb_node
;
120 struct rb_node
*parent
= NULL
;
121 struct alloc_stat
*data
= NULL
;
125 data
= rb_entry(*node
, struct alloc_stat
, node
);
128 node
= &(*node
)->rb_right
;
129 else if (ptr
< data
->ptr
)
130 node
= &(*node
)->rb_left
;
135 if (data
&& data
->ptr
== ptr
) {
137 data
->bytes_req
+= bytes_req
;
138 data
->bytes_alloc
+= bytes_alloc
;
140 data
= malloc(sizeof(*data
));
146 data
->bytes_req
= bytes_req
;
147 data
->bytes_alloc
= bytes_alloc
;
149 rb_link_node(&data
->node
, parent
, node
);
150 rb_insert_color(&data
->node
, &root_alloc_stat
);
152 data
->call_site
= call_site
;
153 data
->alloc_cpu
= cpu
;
156 static void insert_caller_stat(unsigned long call_site
,
157 int bytes_req
, int bytes_alloc
)
159 struct rb_node
**node
= &root_caller_stat
.rb_node
;
160 struct rb_node
*parent
= NULL
;
161 struct alloc_stat
*data
= NULL
;
165 data
= rb_entry(*node
, struct alloc_stat
, node
);
167 if (call_site
> data
->call_site
)
168 node
= &(*node
)->rb_right
;
169 else if (call_site
< data
->call_site
)
170 node
= &(*node
)->rb_left
;
175 if (data
&& data
->call_site
== call_site
) {
177 data
->bytes_req
+= bytes_req
;
178 data
->bytes_alloc
+= bytes_alloc
;
180 data
= malloc(sizeof(*data
));
183 data
->call_site
= call_site
;
186 data
->bytes_req
= bytes_req
;
187 data
->bytes_alloc
= bytes_alloc
;
189 rb_link_node(&data
->node
, parent
, node
);
190 rb_insert_color(&data
->node
, &root_caller_stat
);
194 static void process_alloc_event(void *data
,
197 u64 timestamp __used
,
198 struct thread
*thread __used
,
201 unsigned long call_site
;
207 ptr
= raw_field_value(event
, "ptr", data
);
208 call_site
= raw_field_value(event
, "call_site", data
);
209 bytes_req
= raw_field_value(event
, "bytes_req", data
);
210 bytes_alloc
= raw_field_value(event
, "bytes_alloc", data
);
212 insert_alloc_stat(call_site
, ptr
, bytes_req
, bytes_alloc
, cpu
);
213 insert_caller_stat(call_site
, bytes_req
, bytes_alloc
);
215 total_requested
+= bytes_req
;
216 total_allocated
+= bytes_alloc
;
219 node1
= cpunode_map
[cpu
];
220 node2
= raw_field_value(event
, "node", data
);
227 static int ptr_cmp(struct alloc_stat
*, struct alloc_stat
*);
228 static int callsite_cmp(struct alloc_stat
*, struct alloc_stat
*);
230 static struct alloc_stat
*search_alloc_stat(unsigned long ptr
,
231 unsigned long call_site
,
232 struct rb_root
*root
,
235 struct rb_node
*node
= root
->rb_node
;
236 struct alloc_stat key
= { .ptr
= ptr
, .call_site
= call_site
};
239 struct alloc_stat
*data
;
242 data
= rb_entry(node
, struct alloc_stat
, node
);
244 cmp
= sort_fn(&key
, data
);
246 node
= node
->rb_left
;
248 node
= node
->rb_right
;
255 static void process_free_event(void *data
,
258 u64 timestamp __used
,
259 struct thread
*thread __used
)
262 struct alloc_stat
*s_alloc
, *s_caller
;
264 ptr
= raw_field_value(event
, "ptr", data
);
266 s_alloc
= search_alloc_stat(ptr
, 0, &root_alloc_stat
, ptr_cmp
);
270 if (cpu
!= s_alloc
->alloc_cpu
) {
273 s_caller
= search_alloc_stat(0, s_alloc
->call_site
,
274 &root_caller_stat
, callsite_cmp
);
276 s_caller
->pingpong
++;
278 s_alloc
->alloc_cpu
= -1;
281 static void process_raw_event(union perf_event
*raw_event __used
, void *data
,
282 int cpu
, u64 timestamp
, struct thread
*thread
)
287 type
= trace_parse_common_type(data
);
288 event
= trace_find_event(type
);
290 if (!strcmp(event
->name
, "kmalloc") ||
291 !strcmp(event
->name
, "kmem_cache_alloc")) {
292 process_alloc_event(data
, event
, cpu
, timestamp
, thread
, 0);
296 if (!strcmp(event
->name
, "kmalloc_node") ||
297 !strcmp(event
->name
, "kmem_cache_alloc_node")) {
298 process_alloc_event(data
, event
, cpu
, timestamp
, thread
, 1);
302 if (!strcmp(event
->name
, "kfree") ||
303 !strcmp(event
->name
, "kmem_cache_free")) {
304 process_free_event(data
, event
, cpu
, timestamp
, thread
);
309 static int process_sample_event(struct perf_tool
*tool __used
,
310 union perf_event
*event
,
311 struct perf_sample
*sample
,
312 struct perf_evsel
*evsel __used
,
313 struct machine
*machine
)
315 struct thread
*thread
= machine__findnew_thread(machine
, event
->ip
.pid
);
317 if (thread
== NULL
) {
318 pr_debug("problem processing %d event, skipping it.\n",
323 dump_printf(" ... thread: %s:%d\n", thread
->comm
, thread
->pid
);
325 process_raw_event(event
, sample
->raw_data
, sample
->cpu
,
326 sample
->time
, thread
);
331 static struct perf_tool perf_kmem
= {
332 .sample
= process_sample_event
,
333 .comm
= perf_event__process_comm
,
334 .ordered_samples
= true,
337 static double fragmentation(unsigned long n_req
, unsigned long n_alloc
)
342 return 100.0 - (100.0 * n_req
/ n_alloc
);
345 static void __print_result(struct rb_root
*root
, struct perf_session
*session
,
346 int n_lines
, int is_caller
)
348 struct rb_node
*next
;
349 struct machine
*machine
;
351 printf("%.102s\n", graph_dotted_line
);
352 printf(" %-34s |", is_caller
? "Callsite": "Alloc Ptr");
353 printf(" Total_alloc/Per | Total_req/Per | Hit | Ping-pong | Frag\n");
354 printf("%.102s\n", graph_dotted_line
);
356 next
= rb_first(root
);
358 machine
= perf_session__find_host_machine(session
);
360 pr_err("__print_result: couldn't find kernel information\n");
363 while (next
&& n_lines
--) {
364 struct alloc_stat
*data
= rb_entry(next
, struct alloc_stat
,
366 struct symbol
*sym
= NULL
;
372 addr
= data
->call_site
;
374 sym
= machine__find_kernel_function(machine
, addr
, &map
, NULL
);
379 snprintf(buf
, sizeof(buf
), "%s+%" PRIx64
"", sym
->name
,
380 addr
- map
->unmap_ip(map
, sym
->start
));
382 snprintf(buf
, sizeof(buf
), "%#" PRIx64
"", addr
);
383 printf(" %-34s |", buf
);
385 printf(" %9llu/%-5lu | %9llu/%-5lu | %8lu | %8lu | %6.3f%%\n",
386 (unsigned long long)data
->bytes_alloc
,
387 (unsigned long)data
->bytes_alloc
/ data
->hit
,
388 (unsigned long long)data
->bytes_req
,
389 (unsigned long)data
->bytes_req
/ data
->hit
,
390 (unsigned long)data
->hit
,
391 (unsigned long)data
->pingpong
,
392 fragmentation(data
->bytes_req
, data
->bytes_alloc
));
394 next
= rb_next(next
);
398 printf(" ... | ... | ... | ... | ... | ... \n");
400 printf("%.102s\n", graph_dotted_line
);
403 static void print_summary(void)
405 printf("\nSUMMARY\n=======\n");
406 printf("Total bytes requested: %lu\n", total_requested
);
407 printf("Total bytes allocated: %lu\n", total_allocated
);
408 printf("Total bytes wasted on internal fragmentation: %lu\n",
409 total_allocated
- total_requested
);
410 printf("Internal fragmentation: %f%%\n",
411 fragmentation(total_requested
, total_allocated
));
412 printf("Cross CPU allocations: %lu/%lu\n", nr_cross_allocs
, nr_allocs
);
415 static void print_result(struct perf_session
*session
)
418 __print_result(&root_caller_sorted
, session
, caller_lines
, 1);
420 __print_result(&root_alloc_sorted
, session
, alloc_lines
, 0);
424 struct sort_dimension
{
427 struct list_head list
;
430 static LIST_HEAD(caller_sort
);
431 static LIST_HEAD(alloc_sort
);
433 static void sort_insert(struct rb_root
*root
, struct alloc_stat
*data
,
434 struct list_head
*sort_list
)
436 struct rb_node
**new = &(root
->rb_node
);
437 struct rb_node
*parent
= NULL
;
438 struct sort_dimension
*sort
;
441 struct alloc_stat
*this;
444 this = rb_entry(*new, struct alloc_stat
, node
);
447 list_for_each_entry(sort
, sort_list
, list
) {
448 cmp
= sort
->cmp(data
, this);
454 new = &((*new)->rb_left
);
456 new = &((*new)->rb_right
);
459 rb_link_node(&data
->node
, parent
, new);
460 rb_insert_color(&data
->node
, root
);
463 static void __sort_result(struct rb_root
*root
, struct rb_root
*root_sorted
,
464 struct list_head
*sort_list
)
466 struct rb_node
*node
;
467 struct alloc_stat
*data
;
470 node
= rb_first(root
);
474 rb_erase(node
, root
);
475 data
= rb_entry(node
, struct alloc_stat
, node
);
476 sort_insert(root_sorted
, data
, sort_list
);
480 static void sort_result(void)
482 __sort_result(&root_alloc_stat
, &root_alloc_sorted
, &alloc_sort
);
483 __sort_result(&root_caller_stat
, &root_caller_sorted
, &caller_sort
);
486 static int __cmd_kmem(void)
489 struct perf_session
*session
= perf_session__new(input_name
, O_RDONLY
,
490 0, false, &perf_kmem
);
494 if (perf_session__create_kernel_maps(session
) < 0)
497 if (!perf_session__has_traces(session
, "kmem record"))
501 err
= perf_session__process_events(session
, &perf_kmem
);
505 print_result(session
);
507 perf_session__delete(session
);
511 static const char * const kmem_usage
[] = {
512 "perf kmem [<options>] {record|stat}",
516 static int ptr_cmp(struct alloc_stat
*l
, struct alloc_stat
*r
)
520 else if (l
->ptr
> r
->ptr
)
525 static struct sort_dimension ptr_sort_dimension
= {
530 static int callsite_cmp(struct alloc_stat
*l
, struct alloc_stat
*r
)
532 if (l
->call_site
< r
->call_site
)
534 else if (l
->call_site
> r
->call_site
)
539 static struct sort_dimension callsite_sort_dimension
= {
544 static int hit_cmp(struct alloc_stat
*l
, struct alloc_stat
*r
)
548 else if (l
->hit
> r
->hit
)
553 static struct sort_dimension hit_sort_dimension
= {
558 static int bytes_cmp(struct alloc_stat
*l
, struct alloc_stat
*r
)
560 if (l
->bytes_alloc
< r
->bytes_alloc
)
562 else if (l
->bytes_alloc
> r
->bytes_alloc
)
567 static struct sort_dimension bytes_sort_dimension
= {
572 static int frag_cmp(struct alloc_stat
*l
, struct alloc_stat
*r
)
576 x
= fragmentation(l
->bytes_req
, l
->bytes_alloc
);
577 y
= fragmentation(r
->bytes_req
, r
->bytes_alloc
);
586 static struct sort_dimension frag_sort_dimension
= {
591 static int pingpong_cmp(struct alloc_stat
*l
, struct alloc_stat
*r
)
593 if (l
->pingpong
< r
->pingpong
)
595 else if (l
->pingpong
> r
->pingpong
)
600 static struct sort_dimension pingpong_sort_dimension
= {
605 static struct sort_dimension
*avail_sorts
[] = {
607 &callsite_sort_dimension
,
609 &bytes_sort_dimension
,
610 &frag_sort_dimension
,
611 &pingpong_sort_dimension
,
614 #define NUM_AVAIL_SORTS \
615 (int)(sizeof(avail_sorts) / sizeof(struct sort_dimension *))
617 static int sort_dimension__add(const char *tok
, struct list_head
*list
)
619 struct sort_dimension
*sort
;
622 for (i
= 0; i
< NUM_AVAIL_SORTS
; i
++) {
623 if (!strcmp(avail_sorts
[i
]->name
, tok
)) {
624 sort
= malloc(sizeof(*sort
));
627 memcpy(sort
, avail_sorts
[i
], sizeof(*sort
));
628 list_add_tail(&sort
->list
, list
);
636 static int setup_sorting(struct list_head
*sort_list
, const char *arg
)
639 char *str
= strdup(arg
);
645 tok
= strsep(&str
, ",");
648 if (sort_dimension__add(tok
, sort_list
) < 0) {
649 error("Unknown --sort key: '%s'", tok
);
659 static int parse_sort_opt(const struct option
*opt __used
,
660 const char *arg
, int unset __used
)
665 if (caller_flag
> alloc_flag
)
666 return setup_sorting(&caller_sort
, arg
);
668 return setup_sorting(&alloc_sort
, arg
);
673 static int parse_caller_opt(const struct option
*opt __used
,
674 const char *arg __used
, int unset __used
)
676 caller_flag
= (alloc_flag
+ 1);
680 static int parse_alloc_opt(const struct option
*opt __used
,
681 const char *arg __used
, int unset __used
)
683 alloc_flag
= (caller_flag
+ 1);
687 static int parse_line_opt(const struct option
*opt __used
,
688 const char *arg
, int unset __used
)
695 lines
= strtoul(arg
, NULL
, 10);
697 if (caller_flag
> alloc_flag
)
698 caller_lines
= lines
;
705 static const struct option kmem_options
[] = {
706 OPT_STRING('i', "input", &input_name
, "file",
708 OPT_CALLBACK_NOOPT(0, "caller", NULL
, NULL
,
709 "show per-callsite statistics",
711 OPT_CALLBACK_NOOPT(0, "alloc", NULL
, NULL
,
712 "show per-allocation statistics",
714 OPT_CALLBACK('s', "sort", NULL
, "key[,key2...]",
715 "sort by keys: ptr, call_site, bytes, hit, pingpong, frag",
717 OPT_CALLBACK('l', "line", NULL
, "num",
720 OPT_BOOLEAN(0, "raw-ip", &raw_ip
, "show raw ip instead of symbol"),
724 static const char *record_args
[] = {
730 "-e", "kmem:kmalloc",
731 "-e", "kmem:kmalloc_node",
733 "-e", "kmem:kmem_cache_alloc",
734 "-e", "kmem:kmem_cache_alloc_node",
735 "-e", "kmem:kmem_cache_free",
738 static int __cmd_record(int argc
, const char **argv
)
740 unsigned int rec_argc
, i
, j
;
741 const char **rec_argv
;
743 rec_argc
= ARRAY_SIZE(record_args
) + argc
- 1;
744 rec_argv
= calloc(rec_argc
+ 1, sizeof(char *));
746 if (rec_argv
== NULL
)
749 for (i
= 0; i
< ARRAY_SIZE(record_args
); i
++)
750 rec_argv
[i
] = strdup(record_args
[i
]);
752 for (j
= 1; j
< (unsigned int)argc
; j
++, i
++)
753 rec_argv
[i
] = argv
[j
];
755 return cmd_record(i
, rec_argv
, NULL
);
758 int cmd_kmem(int argc
, const char **argv
, const char *prefix __used
)
760 argc
= parse_options(argc
, argv
, kmem_options
, kmem_usage
, 0);
763 usage_with_options(kmem_usage
, kmem_options
);
767 if (!strncmp(argv
[0], "rec", 3)) {
768 return __cmd_record(argc
, argv
);
769 } else if (!strcmp(argv
[0], "stat")) {
772 if (list_empty(&caller_sort
))
773 setup_sorting(&caller_sort
, default_sort_order
);
774 if (list_empty(&alloc_sort
))
775 setup_sorting(&alloc_sort
, default_sort_order
);
779 usage_with_options(kmem_usage
, kmem_options
);