[linux-2.6-block.git] / Documentation / perf_counter / builtin-top.c

/*
 * builtin-top.c
 *
 * Builtin top command: Display a continuously updated profile of
 * any workload, CPU or specific PID.
 *
 * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
 *
 * Improvements and fixes by:
 *
 *   Arjan van de Ven <arjan@linux.intel.com>
 *   Yanmin Zhang <yanmin.zhang@intel.com>
 *   Wu Fengguang <fengguang.wu@intel.com>
 *   Mike Galbraith <efault@gmx.de>
 *   Paul Mackerras <paulus@samba.org>
 *
 * Released under the GPL v2. (and only v2, not any later version)
 */
#include "builtin.h"

#include "perf.h"

#include "util/symbol.h"
#include "util/util.h"
#include "util/rbtree.h"
#include "util/parse-options.h"
#include "util/parse-events.h"

#include <assert.h>
#include <fcntl.h>

#include <stdio.h>

#include <errno.h>
#include <time.h>
#include <sched.h>
#include <pthread.h>

#include <sys/syscall.h>
#include <sys/ioctl.h>
#include <sys/poll.h>
#include <sys/prctl.h>
#include <sys/wait.h>
#include <sys/uio.h>
#include <sys/mman.h>

#include <linux/unistd.h>
#include <linux/types.h>

static int			system_wide			=  0;

static __u64			default_event_id[MAX_COUNTERS]		= {
	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_TASK_CLOCK),
	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_CONTEXT_SWITCHES),
	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_CPU_MIGRATIONS),
	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_PAGE_FAULTS),

	EID(PERF_TYPE_HARDWARE, PERF_COUNT_CPU_CYCLES),
	EID(PERF_TYPE_HARDWARE, PERF_COUNT_INSTRUCTIONS),
	EID(PERF_TYPE_HARDWARE, PERF_COUNT_CACHE_REFERENCES),
	EID(PERF_TYPE_HARDWARE, PERF_COUNT_CACHE_MISSES),
};
static int			default_interval = 100000;
static int			event_count[MAX_COUNTERS];
static int			fd[MAX_NR_CPUS][MAX_COUNTERS];

static __u64			count_filter		       = 100;

static int			target_pid				= -1;
static int			profile_cpu			= -1;
static int			nr_cpus				=  0;
static unsigned int		realtime_prio			=  0;
static int			group				=  0;
static unsigned int		page_size;
static unsigned int		mmap_pages			=  16;
static int			use_mmap			= 0;
static int			use_munmap			= 0;
static int			freq				= 0;

static char			*sym_filter;
static unsigned long		filter_start;
static unsigned long		filter_end;

static int			delay_secs			=  2;
static int			zero;
static int			dump_symtab;

static const unsigned int default_count[] = {
	1000000,
	1000000,
	  10000,
	  10000,
	1000000,
	  10000,
};

/*
 * Symbols
 */

static uint64_t			min_ip;
static uint64_t			max_ip = -1ll;

struct sym_entry {
	struct rb_node		rb_node;
	struct list_head	node;
	unsigned long		count[MAX_COUNTERS];
	unsigned long		snap_count;
	double			weight;
	int			skip;
};

struct sym_entry		*sym_filter_entry;

struct dso *kernel_dso;

/*
 * Symbols will be added here in record_ip and will get out
 * after decayed.
 */
static LIST_HEAD(active_symbols);
static pthread_mutex_t active_symbols_lock = PTHREAD_MUTEX_INITIALIZER;

/*
 * Ordering weight: count-1 * count-2 * ... / count-n
 */
static double sym_weight(const struct sym_entry *sym)
{
	double weight = sym->snap_count;
	int counter;

	for (counter = 1; counter < nr_counters-1; counter++)
		weight *= sym->count[counter];

	weight /= (sym->count[counter] + 1);

	return weight;
}

static long			events;
static long			userspace_events;
static const char		CONSOLE_CLEAR[] = "\e[H\e[2J";

static void __list_insert_active_sym(struct sym_entry *syme)
{
	list_add(&syme->node, &active_symbols);
}

static void list_remove_active_sym(struct sym_entry *syme)
{
	pthread_mutex_lock(&active_symbols_lock);
	list_del_init(&syme->node);
	pthread_mutex_unlock(&active_symbols_lock);
}

static void rb_insert_active_sym(struct rb_root *tree, struct sym_entry *se)
{
	struct rb_node **p = &tree->rb_node;
	struct rb_node *parent = NULL;
	struct sym_entry *iter;

	while (*p != NULL) {
		parent = *p;
		iter = rb_entry(parent, struct sym_entry, rb_node);

		if (se->weight > iter->weight)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}

	rb_link_node(&se->rb_node, parent, p);
	rb_insert_color(&se->rb_node, tree);
}

static void print_sym_table(void)
{
	int printed, j;
	int counter;
	float events_per_sec = events/delay_secs;
	float kevents_per_sec = (events-userspace_events)/delay_secs;
	float sum_kevents = 0.0;
	struct sym_entry *syme, *n;
	struct rb_root tmp = RB_ROOT;
	struct rb_node *nd;

	events = userspace_events = 0;

	/* Sort the active symbols */
	pthread_mutex_lock(&active_symbols_lock);
	syme = list_entry(active_symbols.next, struct sym_entry, node);
	pthread_mutex_unlock(&active_symbols_lock);

	list_for_each_entry_safe_from(syme, n, &active_symbols, node) {
		syme->snap_count = syme->count[0];
		if (syme->snap_count != 0) {
			syme->weight = sym_weight(syme);
			rb_insert_active_sym(&tmp, syme);
			sum_kevents += syme->snap_count;

			for (j = 0; j < nr_counters; j++)
				syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8;
		} else
			list_remove_active_sym(syme);
	}

	write(1, CONSOLE_CLEAR, strlen(CONSOLE_CLEAR));

	printf(
"------------------------------------------------------------------------------\n");
	printf( " KernelTop:%8.0f irqs/sec  kernel:%4.1f%% [",
		events_per_sec,
		100.0 - (100.0*((events_per_sec-kevents_per_sec)/events_per_sec)));

	if (nr_counters == 1)
		printf("%d ", event_count[0]);

	for (counter = 0; counter < nr_counters; counter++) {
		if (counter)
			printf("/");

		printf("%s", event_name(counter));
	}

	printf( "], ");

	if (target_pid != -1)
		printf(" (target_pid: %d", target_pid);
	else
		printf(" (all");

	if (profile_cpu != -1)
		printf(", cpu: %d)\n", profile_cpu);
	else {
		if (target_pid != -1)
			printf(")\n");
		else
			printf(", %d CPUs)\n", nr_cpus);
	}

	printf("------------------------------------------------------------------------------\n\n");

	if (nr_counters == 1)
		printf("             events    pcnt");
	else
		printf("  weight     events    pcnt");

	printf("         RIP          kernel function\n"
	       	       "  ______     ______   _____   ________________   _______________\n\n"
	);

	for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
		struct sym_entry *syme = rb_entry(nd, struct sym_entry, rb_node);
		struct symbol *sym = (struct symbol *)(syme + 1);
		float pcnt;

		if (++printed > 18 || syme->snap_count < count_filter)
			continue;

		pcnt = 100.0 - (100.0 * ((sum_kevents - syme->snap_count) /
					 sum_kevents));

		if (nr_counters == 1)
			printf("%19.2f - %4.1f%% - %016llx : %s\n",
				syme->weight, pcnt, sym->start, sym->name);
		else
			printf("%8.1f %10ld - %4.1f%% - %016llx : %s\n",
				syme->weight, syme->snap_count,
				pcnt, sym->start, sym->name);
	}

	{
		struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };

		if (poll(&stdin_poll, 1, 0) == 1) {
			printf("key pressed - exiting.\n");
			exit(0);
		}
	}
}

static void *display_thread(void *arg)
{
	printf("KernelTop refresh period: %d seconds\n", delay_secs);

	while (!sleep(delay_secs))
		print_sym_table();

	return NULL;
}

static int symbol_filter(struct dso *self, struct symbol *sym)
{
	static int filter_match;
	struct sym_entry *syme;
	const char *name = sym->name;

	if (!strcmp(name, "_text") ||
	    !strcmp(name, "_etext") ||
	    !strcmp(name, "_sinittext") ||
	    !strncmp("init_module", name, 11) ||
	    !strncmp("cleanup_module", name, 14) ||
	    strstr(name, "_text_start") ||
	    strstr(name, "_text_end"))
		return 1;

	syme = dso__sym_priv(self, sym);
	/* Tag events to be skipped. */
	if (!strcmp("default_idle", name) ||
	    !strcmp("cpu_idle", name) ||
	    !strcmp("enter_idle", name) ||
	    !strcmp("exit_idle", name) ||
	    !strcmp("mwait_idle", name))
		syme->skip = 1;

	if (filter_match == 1) {
		filter_end = sym->start;
		filter_match = -1;
		if (filter_end - filter_start > 10000) {
			fprintf(stderr,
				"hm, too large filter symbol <%s> - skipping.\n",
				sym_filter);
			fprintf(stderr, "symbol filter start: %016lx\n",
				filter_start);
			fprintf(stderr, "                end: %016lx\n",
				filter_end);
			filter_end = filter_start = 0;
			sym_filter = NULL;
			sleep(1);
		}
	}

	if (filter_match == 0 && sym_filter && !strcmp(name, sym_filter)) {
		filter_match = 1;
		filter_start = sym->start;
	}


	return 0;
}

static int parse_symbols(void)
{
	struct rb_node *node;
	struct symbol  *sym;

	kernel_dso = dso__new("[kernel]", sizeof(struct sym_entry));
	if (kernel_dso == NULL)
		return -1;

	if (dso__load_kernel(kernel_dso, NULL, symbol_filter) != 0)
		goto out_delete_dso;

	node = rb_first(&kernel_dso->syms);
	sym = rb_entry(node, struct symbol, rb_node);
	min_ip = sym->start;

	node = rb_last(&kernel_dso->syms);
	sym = rb_entry(node, struct symbol, rb_node);
	max_ip = sym->end;

	if (dump_symtab)
		dso__fprintf(kernel_dso, stderr);

	return 0;

out_delete_dso:
	dso__delete(kernel_dso);
	kernel_dso = NULL;
	return -1;
}

#define TRACE_COUNT     3

/*
 * Binary search in the histogram table and record the hit:
 */
static void record_ip(uint64_t ip, int counter)
{
	struct symbol *sym = dso__find_symbol(kernel_dso, ip);

	if (sym != NULL) {
		struct sym_entry *syme = dso__sym_priv(kernel_dso, sym);

		if (!syme->skip) {
			syme->count[counter]++;
			pthread_mutex_lock(&active_symbols_lock);
			if (list_empty(&syme->node) || !syme->node.next)
				__list_insert_active_sym(syme);
			pthread_mutex_unlock(&active_symbols_lock);
			return;
		}
	}

	events--;
}

static void process_event(uint64_t ip, int counter)
{
	events++;

	if (ip < min_ip || ip > max_ip) {
		userspace_events++;
		return;
	}

	record_ip(ip, counter);
}

struct mmap_data {
	int counter;
	void *base;
	unsigned int mask;
	unsigned int prev;
};

static unsigned int mmap_read_head(struct mmap_data *md)
{
	struct perf_counter_mmap_page *pc = md->base;
	int head;

	head = pc->data_head;
	rmb();

	return head;
}

struct timeval last_read, this_read;

static void mmap_read(struct mmap_data *md)
{
	unsigned int head = mmap_read_head(md);
	unsigned int old = md->prev;
	unsigned char *data = md->base + page_size;
	int diff;

	gettimeofday(&this_read, NULL);

	/*
	 * If we're further behind than half the buffer, there's a chance
	 * the writer will bite our tail and screw up the events under us.
	 *
	 * If we somehow ended up ahead of the head, we got messed up.
	 *
	 * In either case, truncate and restart at head.
	 */
	diff = head - old;
	if (diff > md->mask / 2 || diff < 0) {
		struct timeval iv;
		unsigned long msecs;

		timersub(&this_read, &last_read, &iv);
		msecs = iv.tv_sec*1000 + iv.tv_usec/1000;

		fprintf(stderr, "WARNING: failed to keep up with mmap data."
				"  Last read %lu msecs ago.\n", msecs);

		/*
		 * head points to a known good entry, start there.
		 */
		old = head;
	}

	last_read = this_read;

	for (; old != head;) {
		struct ip_event {
			struct perf_event_header header;
			__u64 ip;
			__u32 pid, target_pid;
		};
		struct mmap_event {
			struct perf_event_header header;
			__u32 pid, target_pid;
			__u64 start;
			__u64 len;
			__u64 pgoff;
			char filename[PATH_MAX];
		};

		typedef union event_union {
			struct perf_event_header header;
			struct ip_event ip;
			struct mmap_event mmap;
		} event_t;

		event_t *event = (event_t *)&data[old & md->mask];

		event_t event_copy;

		size_t size = event->header.size;

		/*
		 * Event straddles the mmap boundary -- header should always
		 * be inside due to u64 alignment of output.
		 */
		if ((old & md->mask) + size != ((old + size) & md->mask)) {
			unsigned int offset = old;
			unsigned int len = min(sizeof(*event), size), cpy;
			void *dst = &event_copy;

			do {
				cpy = min(md->mask + 1 - (offset & md->mask), len);
				memcpy(dst, &data[offset & md->mask], cpy);
				offset += cpy;
				dst += cpy;
				len -= cpy;
			} while (len);

			event = &event_copy;
		}

		old += size;

		if (event->header.misc & PERF_EVENT_MISC_OVERFLOW) {
			if (event->header.type & PERF_SAMPLE_IP)
				process_event(event->ip.ip, md->counter);
		} else {
			switch (event->header.type) {
				case PERF_EVENT_MMAP:
				case PERF_EVENT_MUNMAP:
					printf("%s: %Lu %Lu %Lu %s\n",
							event->header.type == PERF_EVENT_MMAP
							? "mmap" : "munmap",
							event->mmap.start,
							event->mmap.len,
							event->mmap.pgoff,
							event->mmap.filename);
					break;
			}
		}
	}

	md->prev = old;
}

static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS];
static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS];

static int __cmd_top(void)
{
	struct perf_counter_attr attr;
	pthread_t thread;
	int i, counter, group_fd, nr_poll = 0;
	unsigned int cpu;
	int ret;

	for (i = 0; i < nr_cpus; i++) {
		group_fd = -1;
		for (counter = 0; counter < nr_counters; counter++) {

			cpu	= profile_cpu;
			if (target_pid == -1 && profile_cpu == -1)
				cpu = i;

			memset(&attr, 0, sizeof(attr));
			attr.config		= event_id[counter];
			attr.sample_period	= event_count[counter];
			attr.sample_type	= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
			attr.mmap		= use_mmap;
			attr.munmap		= use_munmap;
			attr.freq		= freq;

			fd[i][counter] = sys_perf_counter_open(&attr, target_pid, cpu, group_fd, 0);
			if (fd[i][counter] < 0) {
				int err = errno;
				printf("kerneltop error: syscall returned with %d (%s)\n",
					fd[i][counter], strerror(err));
				if (err == EPERM)
					printf("Are you root?\n");
				exit(-1);
			}
			assert(fd[i][counter] >= 0);
			fcntl(fd[i][counter], F_SETFL, O_NONBLOCK);

			/*
			 * First counter acts as the group leader:
			 */
			if (group && group_fd == -1)
				group_fd = fd[i][counter];

			event_array[nr_poll].fd = fd[i][counter];
			event_array[nr_poll].events = POLLIN;
			nr_poll++;

			mmap_array[i][counter].counter = counter;
			mmap_array[i][counter].prev = 0;
			mmap_array[i][counter].mask = mmap_pages*page_size - 1;
			mmap_array[i][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
					PROT_READ, MAP_SHARED, fd[i][counter], 0);
			if (mmap_array[i][counter].base == MAP_FAILED) {
				printf("kerneltop error: failed to mmap with %d (%s)\n",
						errno, strerror(errno));
				exit(-1);
			}
		}
	}

	if (pthread_create(&thread, NULL, display_thread, NULL)) {
		printf("Could not create display thread.\n");
		exit(-1);
	}

	if (realtime_prio) {
		struct sched_param param;

		param.sched_priority = realtime_prio;
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
			printf("Could not set realtime priority.\n");
			exit(-1);
		}
	}

	while (1) {
		int hits = events;

		for (i = 0; i < nr_cpus; i++) {
			for (counter = 0; counter < nr_counters; counter++)
				mmap_read(&mmap_array[i][counter]);
		}

		if (hits == events)
			ret = poll(event_array, nr_poll, 100);
	}

	return 0;
}

static const char * const top_usage[] = {
	"perf top [<options>]",
	NULL
};

static char events_help_msg[EVENTS_HELP_MAX];

static const struct option options[] = {
	OPT_CALLBACK('e', "event", NULL, "event",
		     events_help_msg, parse_events),
	OPT_INTEGER('c', "count", &default_interval,
		    "event period to sample"),
	OPT_INTEGER('p', "pid", &target_pid,
		    "profile events on existing pid"),
	OPT_BOOLEAN('a', "all-cpus", &system_wide,
			    "system-wide collection from all CPUs"),
	OPT_INTEGER('C', "CPU", &profile_cpu,
		    "CPU to profile on"),
	OPT_INTEGER('m', "mmap-pages", &mmap_pages,
		    "number of mmap data pages"),
	OPT_INTEGER('r', "realtime", &realtime_prio,
		    "collect data with this RT SCHED_FIFO priority"),
	OPT_INTEGER('d', "delay", &delay_secs,
		    "number of seconds to delay between refreshes"),
	OPT_BOOLEAN('D', "dump-symtab", &dump_symtab,
			    "dump the symbol table used for profiling"),
	OPT_INTEGER('f', "--count-filter", &count_filter,
		    "only display functions with more events than this"),
	OPT_BOOLEAN('g', "group", &group,
			    "put the counters into a counter group"),
	OPT_STRING('s', "sym-filter", &sym_filter, "pattern",
		    "only display symbols matchig this pattern"),
	OPT_BOOLEAN('z', "zero", &group,
		    "zero history across updates"),
	OPT_BOOLEAN('M', "use-mmap", &use_mmap,
		    "track mmap events"),
	OPT_BOOLEAN('U', "use-munmap", &use_munmap,
		    "track munmap events"),
	OPT_INTEGER('F', "--freq", &freq,
		    "profile at this frequency"),
	OPT_END()
};

int cmd_top(int argc, const char **argv, const char *prefix)
{
	int counter;

	page_size = sysconf(_SC_PAGE_SIZE);

	create_events_help(events_help_msg);
	memcpy(event_id, default_event_id, sizeof(default_event_id));

	argc = parse_options(argc, argv, options, top_usage, 0);
	if (argc)
		usage_with_options(top_usage, options);

	if (freq) {
		default_interval = freq;
		freq = 1;
	}

	/* CPU and PID are mutually exclusive */
	if (target_pid != -1 && profile_cpu != -1) {
		printf("WARNING: PID switch overriding CPU\n");
		sleep(1);
		profile_cpu = -1;
	}

	if (!nr_counters) {
		nr_counters = 1;
		event_id[0] = 0;
	}

	for (counter = 0; counter < nr_counters; counter++) {
		if (event_count[counter])
			continue;

		event_count[counter] = default_interval;
	}

	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	assert(nr_cpus <= MAX_NR_CPUS);
	assert(nr_cpus >= 0);

	if (target_pid != -1 || profile_cpu != -1)
		nr_cpus = 1;

	parse_symbols();

	return __cmd_top();
}
Commit	Line	Data
07800601	1	/*
bf9e1876 IM	2	* builtin-top.c
	3	*
	4	* Builtin top command: Display a continuously updated profile of
	5	* any workload, CPU or specific PID.
	6	*
	7	* Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
	8	*
	9	* Improvements and fixes by:
	10	*
	11	* Arjan van de Ven <arjan@linux.intel.com>
	12	* Yanmin Zhang <yanmin.zhang@intel.com>
	13	* Wu Fengguang <fengguang.wu@intel.com>
	14	* Mike Galbraith <efault@gmx.de>
	15	* Paul Mackerras <paulus@samba.org>
	16	*
	17	* Released under the GPL v2. (and only v2, not any later version)
07800601	18	*/
bf9e1876	19	#include "builtin.h"
07800601	20
1a482f38	21	#include "perf.h"
bf9e1876	22
de04687f	23	#include "util/symbol.h"
148be2c1	24	#include "util/util.h"
de04687f	25	#include "util/rbtree.h"
b456bae0 IM	26	#include "util/parse-options.h"
b456bae0 IM	27	#include "util/parse-events.h"
07800601	28
07800601 IM	29	#include <assert.h>
07800601 IM	30	#include <fcntl.h>
0e9b20b8	31
07800601	32	#include <stdio.h>
0e9b20b8	33
07800601	34	#include <errno.h>
07800601 IM	35	#include <time.h>
	36	#include <sched.h>
	37	#include <pthread.h>
	38
	39	#include <sys/syscall.h>
	40	#include <sys/ioctl.h>
	41	#include <sys/poll.h>
	42	#include <sys/prctl.h>
	43	#include <sys/wait.h>
	44	#include <sys/uio.h>
	45	#include <sys/mman.h>
	46
	47	#include <linux/unistd.h>
	48	#include <linux/types.h>
	49
07800601 IM	50	static int system_wide = 0;
07800601 IM	51
b456bae0	52	static __u64 default_event_id[MAX_COUNTERS] = {
07800601 IM	53	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_TASK_CLOCK),
	54	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_CONTEXT_SWITCHES),
	55	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_CPU_MIGRATIONS),
	56	EID(PERF_TYPE_SOFTWARE, PERF_COUNT_PAGE_FAULTS),
	57
	58	EID(PERF_TYPE_HARDWARE, PERF_COUNT_CPU_CYCLES),
	59	EID(PERF_TYPE_HARDWARE, PERF_COUNT_INSTRUCTIONS),
	60	EID(PERF_TYPE_HARDWARE, PERF_COUNT_CACHE_REFERENCES),
	61	EID(PERF_TYPE_HARDWARE, PERF_COUNT_CACHE_MISSES),
	62	};
	63	static int default_interval = 100000;
	64	static int event_count[MAX_COUNTERS];
	65	static int fd[MAX_NR_CPUS][MAX_COUNTERS];
	66
	67	static __u64 count_filter = 100;
	68
b456bae0	69	static int target_pid = -1;
07800601 IM	70	static int profile_cpu = -1;
07800601 IM	71	static int nr_cpus = 0;
07800601 IM	72	static unsigned int realtime_prio = 0;
	73	static int group = 0;
	74	static unsigned int page_size;
	75	static unsigned int mmap_pages = 16;
	76	static int use_mmap = 0;
	77	static int use_munmap = 0;
f5456a6b	78	static int freq = 0;
07800601	79
07800601 IM	80	static char *sym_filter;
	81	static unsigned long filter_start;
	82	static unsigned long filter_end;
	83
	84	static int delay_secs = 2;
	85	static int zero;
	86	static int dump_symtab;
	87
ddcacfa0	88	static const unsigned int default_count[] = {
07800601 IM	89	1000000,
	90	1000000,
	91	10000,
	92	10000,
	93	1000000,
	94	10000,
	95	};
	96
07800601 IM	97	/*
	98	* Symbols
	99	*/
	100
	101	static uint64_t min_ip;
	102	static uint64_t max_ip = -1ll;
	103
	104	struct sym_entry {
de04687f ACM	105	struct rb_node rb_node;
de04687f ACM	106	struct list_head node;
07800601	107	unsigned long count[MAX_COUNTERS];
c44613a4 ACM	108	unsigned long snap_count;
c44613a4 ACM	109	double weight;
07800601	110	int skip;
07800601 IM	111	};
07800601 IM	112
07800601 IM	113	struct sym_entry *sym_filter_entry;
07800601 IM	114
de04687f ACM	115	struct dso *kernel_dso;
	116
	117	/*
	118	* Symbols will be added here in record_ip and will get out
	119	* after decayed.
	120	*/
	121	static LIST_HEAD(active_symbols);
c44613a4	122	static pthread_mutex_t active_symbols_lock = PTHREAD_MUTEX_INITIALIZER;
07800601	123
07800601 IM	124	/*
	125	* Ordering weight: count-1 * count-2 * ... / count-n
	126	*/
	127	static double sym_weight(const struct sym_entry *sym)
	128	{
c44613a4	129	double weight = sym->snap_count;
07800601 IM	130	int counter;
07800601 IM	131
07800601 IM	132	for (counter = 1; counter < nr_counters-1; counter++)
	133	weight *= sym->count[counter];
	134
	135	weight /= (sym->count[counter] + 1);
	136
	137	return weight;
	138	}
	139
07800601 IM	140	static long events;
	141	static long userspace_events;
	142	static const char CONSOLE_CLEAR[] = "\e[H\e[2J";
	143
c44613a4	144	static void __list_insert_active_sym(struct sym_entry *syme)
de04687f ACM	145	{
	146	list_add(&syme->node, &active_symbols);
	147	}
	148
c44613a4 ACM	149	static void list_remove_active_sym(struct sym_entry *syme)
	150	{
	151	pthread_mutex_lock(&active_symbols_lock);
	152	list_del_init(&syme->node);
	153	pthread_mutex_unlock(&active_symbols_lock);
	154	}
	155
de04687f ACM	156	static void rb_insert_active_sym(struct rb_root tree, struct sym_entry se)
	157	{
	158	struct rb_node **p = &tree->rb_node;
	159	struct rb_node *parent = NULL;
	160	struct sym_entry *iter;
	161
	162	while (*p != NULL) {
	163	parent = *p;
	164	iter = rb_entry(parent, struct sym_entry, rb_node);
	165
c44613a4	166	if (se->weight > iter->weight)
de04687f ACM	167	p = &(*p)->rb_left;
	168	else
	169	p = &(*p)->rb_right;
	170	}
	171
	172	rb_link_node(&se->rb_node, parent, p);
	173	rb_insert_color(&se->rb_node, tree);
	174	}
07800601 IM	175
	176	static void print_sym_table(void)
	177	{
de04687f	178	int printed, j;
07800601 IM	179	int counter;
	180	float events_per_sec = events/delay_secs;
	181	float kevents_per_sec = (events-userspace_events)/delay_secs;
	182	float sum_kevents = 0.0;
de04687f ACM	183	struct sym_entry syme, n;
	184	struct rb_root tmp = RB_ROOT;
	185	struct rb_node *nd;
07800601 IM	186
07800601 IM	187	events = userspace_events = 0;
07800601	188
de04687f	189	/* Sort the active symbols */
c44613a4 ACM	190	pthread_mutex_lock(&active_symbols_lock);
	191	syme = list_entry(active_symbols.next, struct sym_entry, node);
	192	pthread_mutex_unlock(&active_symbols_lock);
	193
	194	list_for_each_entry_safe_from(syme, n, &active_symbols, node) {
	195	syme->snap_count = syme->count[0];
	196	if (syme->snap_count != 0) {
	197	syme->weight = sym_weight(syme);
de04687f	198	rb_insert_active_sym(&tmp, syme);
c44613a4	199	sum_kevents += syme->snap_count;
d94b9430 MG	200
d94b9430 MG	201	for (j = 0; j < nr_counters; j++)
de04687f ACM	202	syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8;
de04687f ACM	203	} else
c44613a4	204	list_remove_active_sym(syme);
d94b9430 MG	205	}
d94b9430 MG	206
07800601 IM	207	write(1, CONSOLE_CLEAR, strlen(CONSOLE_CLEAR));
	208
	209	printf(
	210	"------------------------------------------------------------------------------\n");
f91183fe	211	printf( " KernelTop:%8.0f irqs/sec kernel:%4.1f%% [",
07800601	212	events_per_sec,
f91183fe	213	100.0 - (100.0*((events_per_sec-kevents_per_sec)/events_per_sec)));
07800601 IM	214
	215	if (nr_counters == 1)
	216	printf("%d ", event_count[0]);
	217
	218	for (counter = 0; counter < nr_counters; counter++) {
	219	if (counter)
	220	printf("/");
	221
	222	printf("%s", event_name(counter));
	223	}
	224
	225	printf( "], ");
	226
b456bae0 IM	227	if (target_pid != -1)
b456bae0 IM	228	printf(" (target_pid: %d", target_pid);
07800601 IM	229	else
	230	printf(" (all");
	231
	232	if (profile_cpu != -1)
	233	printf(", cpu: %d)\n", profile_cpu);
	234	else {
b456bae0	235	if (target_pid != -1)
07800601 IM	236	printf(")\n");
	237	else
	238	printf(", %d CPUs)\n", nr_cpus);
	239	}
	240
	241	printf("------------------------------------------------------------------------------\n\n");
	242
	243	if (nr_counters == 1)
	244	printf(" events pcnt");
	245	else
	246	printf(" weight events pcnt");
	247
	248	printf(" RIP kernel function\n"
	249	" ______ ______ _____ ________________ _______________\n\n"
	250	);
	251
de04687f ACM	252	for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
	253	struct sym_entry *syme = rb_entry(nd, struct sym_entry, rb_node);
	254	struct symbol sym = (struct symbol )(syme + 1);
07800601	255	float pcnt;
d94b9430	256
c44613a4 ACM	257	if (++printed > 18 \|\| syme->snap_count < count_filter)
c44613a4 ACM	258	continue;
d94b9430	259
c44613a4	260	pcnt = 100.0 - (100.0 * ((sum_kevents - syme->snap_count) /
de04687f	261	sum_kevents));
d94b9430 MG	262
	263	if (nr_counters == 1)
	264	printf("%19.2f - %4.1f%% - %016llx : %s\n",
c44613a4	265	syme->weight, pcnt, sym->start, sym->name);
d94b9430 MG	266	else
d94b9430 MG	267	printf("%8.1f %10ld - %4.1f%% - %016llx : %s\n",
c44613a4	268	syme->weight, syme->snap_count,
de04687f	269	pcnt, sym->start, sym->name);
07800601 IM	270	}
07800601 IM	271
07800601 IM	272	{
	273	struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
	274
	275	if (poll(&stdin_poll, 1, 0) == 1) {
	276	printf("key pressed - exiting.\n");
	277	exit(0);
	278	}
	279	}
	280	}
	281
	282	static void display_thread(void arg)
	283	{
	284	printf("KernelTop refresh period: %d seconds\n", delay_secs);
	285
	286	while (!sleep(delay_secs))
	287	print_sym_table();
	288
	289	return NULL;
	290	}
	291
de04687f	292	static int symbol_filter(struct dso self, struct symbol sym)
07800601	293	{
de04687f ACM	294	static int filter_match;
	295	struct sym_entry *syme;
	296	const char *name = sym->name;
	297
	298	if (!strcmp(name, "_text") \|\|
	299	!strcmp(name, "_etext") \|\|
	300	!strcmp(name, "_sinittext") \|\|
	301	!strncmp("init_module", name, 11) \|\|
	302	!strncmp("cleanup_module", name, 14) \|\|
	303	strstr(name, "_text_start") \|\|
	304	strstr(name, "_text_end"))
07800601	305	return 1;
07800601	306
de04687f	307	syme = dso__sym_priv(self, sym);
07800601	308	/* Tag events to be skipped. */
de04687f ACM	309	if (!strcmp("default_idle", name) \|\|
	310	!strcmp("cpu_idle", name) \|\|
	311	!strcmp("enter_idle", name) \|\|
	312	!strcmp("exit_idle", name) \|\|
	313	!strcmp("mwait_idle", name))
	314	syme->skip = 1;
07800601 IM	315
07800601 IM	316	if (filter_match == 1) {
de04687f	317	filter_end = sym->start;
07800601 IM	318	filter_match = -1;
07800601 IM	319	if (filter_end - filter_start > 10000) {
de04687f ACM	320	fprintf(stderr,
de04687f ACM	321	"hm, too large filter symbol <%s> - skipping.\n",
07800601	322	sym_filter);
de04687f ACM	323	fprintf(stderr, "symbol filter start: %016lx\n",
	324	filter_start);
	325	fprintf(stderr, " end: %016lx\n",
	326	filter_end);
07800601 IM	327	filter_end = filter_start = 0;
	328	sym_filter = NULL;
	329	sleep(1);
	330	}
	331	}
de04687f ACM	332
de04687f ACM	333	if (filter_match == 0 && sym_filter && !strcmp(name, sym_filter)) {
07800601	334	filter_match = 1;
de04687f	335	filter_start = sym->start;
07800601 IM	336	}
07800601 IM	337
de04687f	338
07800601 IM	339	return 0;
	340	}
	341
de04687f	342	static int parse_symbols(void)
07800601	343	{
de04687f ACM	344	struct rb_node *node;
de04687f ACM	345	struct symbol *sym;
07800601	346
de04687f ACM	347	kernel_dso = dso__new("[kernel]", sizeof(struct sym_entry));
	348	if (kernel_dso == NULL)
	349	return -1;
07800601	350
de04687f ACM	351	if (dso__load_kernel(kernel_dso, NULL, symbol_filter) != 0)
de04687f ACM	352	goto out_delete_dso;
07800601	353
de04687f ACM	354	node = rb_first(&kernel_dso->syms);
	355	sym = rb_entry(node, struct symbol, rb_node);
	356	min_ip = sym->start;
07800601	357
de04687f ACM	358	node = rb_last(&kernel_dso->syms);
de04687f ACM	359	sym = rb_entry(node, struct symbol, rb_node);
da417a75	360	max_ip = sym->end;
07800601	361
de04687f	362	if (dump_symtab)
a3ec8d70	363	dso__fprintf(kernel_dso, stderr);
07800601	364
de04687f	365	return 0;
07800601	366
de04687f ACM	367	out_delete_dso:
	368	dso__delete(kernel_dso);
	369	kernel_dso = NULL;
	370	return -1;
07800601 IM	371	}
07800601 IM	372
07800601 IM	373	#define TRACE_COUNT 3
07800601 IM	374
07800601 IM	375	/*
	376	* Binary search in the histogram table and record the hit:
	377	*/
	378	static void record_ip(uint64_t ip, int counter)
	379	{
de04687f	380	struct symbol *sym = dso__find_symbol(kernel_dso, ip);
07800601	381
de04687f ACM	382	if (sym != NULL) {
de04687f ACM	383	struct sym_entry *syme = dso__sym_priv(kernel_dso, sym);
07800601	384
de04687f ACM	385	if (!syme->skip) {
de04687f ACM	386	syme->count[counter]++;
c44613a4	387	pthread_mutex_lock(&active_symbols_lock);
de04687f	388	if (list_empty(&syme->node) \|\| !syme->node.next)
c44613a4 ACM	389	__list_insert_active_sym(syme);
c44613a4 ACM	390	pthread_mutex_unlock(&active_symbols_lock);
de04687f	391	return;
07800601	392	}
07800601 IM	393	}
07800601 IM	394
de04687f	395	events--;
07800601 IM	396	}
	397
	398	static void process_event(uint64_t ip, int counter)
	399	{
	400	events++;
	401
	402	if (ip < min_ip \|\| ip > max_ip) {
	403	userspace_events++;
	404	return;
	405	}
	406
	407	record_ip(ip, counter);
	408	}
	409
07800601 IM	410	struct mmap_data {
	411	int counter;
	412	void *base;
	413	unsigned int mask;
	414	unsigned int prev;
	415	};
	416
	417	static unsigned int mmap_read_head(struct mmap_data *md)
	418	{
	419	struct perf_counter_mmap_page *pc = md->base;
	420	int head;
	421
	422	head = pc->data_head;
	423	rmb();
	424
	425	return head;
	426	}
	427
	428	struct timeval last_read, this_read;
	429
	430	static void mmap_read(struct mmap_data *md)
	431	{
	432	unsigned int head = mmap_read_head(md);
	433	unsigned int old = md->prev;
	434	unsigned char *data = md->base + page_size;
	435	int diff;
	436
	437	gettimeofday(&this_read, NULL);
	438
	439	/*
	440	* If we're further behind than half the buffer, there's a chance
	441	* the writer will bite our tail and screw up the events under us.
	442	*
	443	* If we somehow ended up ahead of the head, we got messed up.
	444	*
	445	* In either case, truncate and restart at head.
	446	*/
	447	diff = head - old;
	448	if (diff > md->mask / 2 \|\| diff < 0) {
	449	struct timeval iv;
	450	unsigned long msecs;
	451
	452	timersub(&this_read, &last_read, &iv);
	453	msecs = iv.tv_sec*1000 + iv.tv_usec/1000;
	454
	455	fprintf(stderr, "WARNING: failed to keep up with mmap data."
	456	" Last read %lu msecs ago.\n", msecs);
	457
	458	/*
	459	* head points to a known good entry, start there.
	460	*/
	461	old = head;
	462	}
	463
	464	last_read = this_read;
	465
	466	for (; old != head;) {
	467	struct ip_event {
	468	struct perf_event_header header;
	469	__u64 ip;
b456bae0	470	__u32 pid, target_pid;
07800601 IM	471	};
	472	struct mmap_event {
	473	struct perf_event_header header;
b456bae0	474	__u32 pid, target_pid;
07800601 IM	475	__u64 start;
	476	__u64 len;
	477	__u64 pgoff;
	478	char filename[PATH_MAX];
	479	};
	480
	481	typedef union event_union {
	482	struct perf_event_header header;
	483	struct ip_event ip;
	484	struct mmap_event mmap;
	485	} event_t;
	486
	487	event_t event = (event_t )&data[old & md->mask];
	488
	489	event_t event_copy;
	490
6f06ccbc	491	size_t size = event->header.size;
07800601 IM	492
	493	/*
	494	* Event straddles the mmap boundary -- header should always
	495	* be inside due to u64 alignment of output.
	496	*/
	497	if ((old & md->mask) + size != ((old + size) & md->mask)) {
	498	unsigned int offset = old;
	499	unsigned int len = min(sizeof(*event), size), cpy;
	500	void *dst = &event_copy;
	501
	502	do {
	503	cpy = min(md->mask + 1 - (offset & md->mask), len);
	504	memcpy(dst, &data[offset & md->mask], cpy);
	505	offset += cpy;
	506	dst += cpy;
	507	len -= cpy;
	508	} while (len);
	509
	510	event = &event_copy;
	511	}
	512
	513	old += size;
	514
	515	if (event->header.misc & PERF_EVENT_MISC_OVERFLOW) {
c70975bc	516	if (event->header.type & PERF_SAMPLE_IP)
07800601 IM	517	process_event(event->ip.ip, md->counter);
	518	} else {
	519	switch (event->header.type) {
	520	case PERF_EVENT_MMAP:
	521	case PERF_EVENT_MUNMAP:
	522	printf("%s: %Lu %Lu %Lu %s\n",
	523	event->header.type == PERF_EVENT_MMAP
	524	? "mmap" : "munmap",
	525	event->mmap.start,
	526	event->mmap.len,
	527	event->mmap.pgoff,
	528	event->mmap.filename);
	529	break;
	530	}
	531	}
	532	}
	533
	534	md->prev = old;
	535	}
	536
c2990a2a MG	537	static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS];
	538	static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS];
	539
b456bae0	540	static int __cmd_top(void)
07800601	541	{
c70975bc	542	struct perf_counter_attr attr;
07800601 IM	543	pthread_t thread;
	544	int i, counter, group_fd, nr_poll = 0;
	545	unsigned int cpu;
	546	int ret;
	547
07800601 IM	548	for (i = 0; i < nr_cpus; i++) {
	549	group_fd = -1;
	550	for (counter = 0; counter < nr_counters; counter++) {
	551
	552	cpu = profile_cpu;
b456bae0	553	if (target_pid == -1 && profile_cpu == -1)
07800601 IM	554	cpu = i;
07800601 IM	555
c70975bc PZ	556	memset(&attr, 0, sizeof(attr));
	557	attr.config = event_id[counter];
	558	attr.sample_period = event_count[counter];
	559	attr.sample_type = PERF_SAMPLE_IP \| PERF_SAMPLE_TID;
	560	attr.mmap = use_mmap;
	561	attr.munmap = use_munmap;
	562	attr.freq = freq;
07800601	563
c70975bc	564	fd[i][counter] = sys_perf_counter_open(&attr, target_pid, cpu, group_fd, 0);
07800601 IM	565	if (fd[i][counter] < 0) {
	566	int err = errno;
	567	printf("kerneltop error: syscall returned with %d (%s)\n",
	568	fd[i][counter], strerror(err));
	569	if (err == EPERM)
	570	printf("Are you root?\n");
	571	exit(-1);
	572	}
	573	assert(fd[i][counter] >= 0);
	574	fcntl(fd[i][counter], F_SETFL, O_NONBLOCK);
	575
	576	/*
	577	* First counter acts as the group leader:
	578	*/
	579	if (group && group_fd == -1)
	580	group_fd = fd[i][counter];
	581
	582	event_array[nr_poll].fd = fd[i][counter];
	583	event_array[nr_poll].events = POLLIN;
	584	nr_poll++;
	585
	586	mmap_array[i][counter].counter = counter;
	587	mmap_array[i][counter].prev = 0;
	588	mmap_array[i][counter].mask = mmap_pages*page_size - 1;
	589	mmap_array[i][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
	590	PROT_READ, MAP_SHARED, fd[i][counter], 0);
	591	if (mmap_array[i][counter].base == MAP_FAILED) {
	592	printf("kerneltop error: failed to mmap with %d (%s)\n",
	593	errno, strerror(errno));
	594	exit(-1);
	595	}
	596	}
	597	}
	598
	599	if (pthread_create(&thread, NULL, display_thread, NULL)) {
	600	printf("Could not create display thread.\n");
	601	exit(-1);
	602	}
	603
	604	if (realtime_prio) {
	605	struct sched_param param;
	606
	607	param.sched_priority = realtime_prio;
	608	if (sched_setscheduler(0, SCHED_FIFO, &param)) {
	609	printf("Could not set realtime priority.\n");
	610	exit(-1);
	611	}
	612	}
	613
	614	while (1) {
	615	int hits = events;
	616
	617	for (i = 0; i < nr_cpus; i++) {
	618	for (counter = 0; counter < nr_counters; counter++)
	619	mmap_read(&mmap_array[i][counter]);
	620	}
	621
	622	if (hits == events)
	623	ret = poll(event_array, nr_poll, 100);
	624	}
	625
	626	return 0;
	627	}
b456bae0 IM	628
	629	static const char * const top_usage[] = {
	630	"perf top [<options>]",
	631	NULL
	632	};
	633
	634	static char events_help_msg[EVENTS_HELP_MAX];
	635
	636	static const struct option options[] = {
	637	OPT_CALLBACK('e', "event", NULL, "event",
	638	events_help_msg, parse_events),
	639	OPT_INTEGER('c', "count", &default_interval,
	640	"event period to sample"),
	641	OPT_INTEGER('p', "pid", &target_pid,
	642	"profile events on existing pid"),
	643	OPT_BOOLEAN('a', "all-cpus", &system_wide,
	644	"system-wide collection from all CPUs"),
	645	OPT_INTEGER('C', "CPU", &profile_cpu,
	646	"CPU to profile on"),
	647	OPT_INTEGER('m', "mmap-pages", &mmap_pages,
	648	"number of mmap data pages"),
	649	OPT_INTEGER('r', "realtime", &realtime_prio,
	650	"collect data with this RT SCHED_FIFO priority"),
db20c003	651	OPT_INTEGER('d', "delay", &delay_secs,
b456bae0 IM	652	"number of seconds to delay between refreshes"),
	653	OPT_BOOLEAN('D', "dump-symtab", &dump_symtab,
	654	"dump the symbol table used for profiling"),
	655	OPT_INTEGER('f', "--count-filter", &count_filter,
	656	"only display functions with more events than this"),
	657	OPT_BOOLEAN('g', "group", &group,
	658	"put the counters into a counter group"),
	659	OPT_STRING('s', "sym-filter", &sym_filter, "pattern",
	660	"only display symbols matchig this pattern"),
	661	OPT_BOOLEAN('z', "zero", &group,
	662	"zero history across updates"),
	663	OPT_BOOLEAN('M', "use-mmap", &use_mmap,
	664	"track mmap events"),
	665	OPT_BOOLEAN('U', "use-munmap", &use_munmap,
	666	"track munmap events"),
	667	OPT_INTEGER('F', "--freq", &freq,
	668	"profile at this frequency"),
	669	OPT_END()
	670	};
	671
	672	int cmd_top(int argc, const char *argv, const char prefix)
	673	{
	674	int counter;
	675
	676	page_size = sysconf(_SC_PAGE_SIZE);
	677
	678	create_events_help(events_help_msg);
	679	memcpy(event_id, default_event_id, sizeof(default_event_id));
	680
	681	argc = parse_options(argc, argv, options, top_usage, 0);
	682	if (argc)
	683	usage_with_options(top_usage, options);
	684
	685	if (freq) {
	686	default_interval = freq;
	687	freq = 1;
	688	}
	689
	690	/* CPU and PID are mutually exclusive */
	691	if (target_pid != -1 && profile_cpu != -1) {
	692	printf("WARNING: PID switch overriding CPU\n");
	693	sleep(1);
	694	profile_cpu = -1;
	695	}
	696
	697	if (!nr_counters) {
	698	nr_counters = 1;
	699	event_id[0] = 0;
	700	}
	701
	702	for (counter = 0; counter < nr_counters; counter++) {
	703	if (event_count[counter])
	704	continue;
	705
	706	event_count[counter] = default_interval;
	707	}
	708
	709	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	710	assert(nr_cpus <= MAX_NR_CPUS);
	711	assert(nr_cpus >= 0);
	712
	713	if (target_pid != -1 \|\| profile_cpu != -1)
	714	nr_cpus = 1;
	715
716	parse_symbols();
717
718	return __cmd_top();
719	}