[deliverable/linux.git] / tools / perf / util / evsel.c

/*
 * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
 *
 * Parts came from builtin-{top,stat,record}.c, see those files for further
 * copyright notes.
 *
 * Released under the GPL v2. (and only v2, not any later version)
 */

#include <byteswap.h>
#include "asm/bug.h"
#include "evsel.h"
#include "evlist.h"
#include "util.h"
#include "cpumap.h"
#include "thread_map.h"

#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
#define GROUP_FD(group_fd, cpu) (*(int *)xyarray__entry(group_fd, cpu, 0))

int __perf_evsel__sample_size(u64 sample_type)
{
	u64 mask = sample_type & PERF_SAMPLE_MASK;
	int size = 0;
	int i;

	for (i = 0; i < 64; i++) {
		if (mask & (1ULL << i))
			size++;
	}

	size *= sizeof(u64);

	return size;
}

void hists__init(struct hists *hists)
{
	memset(hists, 0, sizeof(*hists));
	hists->entries_in_array[0] = hists->entries_in_array[1] = RB_ROOT;
	hists->entries_in = &hists->entries_in_array[0];
	hists->entries_collapsed = RB_ROOT;
	hists->entries = RB_ROOT;
	pthread_mutex_init(&hists->lock, NULL);
}

void perf_evsel__init(struct perf_evsel *evsel,
		      struct perf_event_attr *attr, int idx)
{
	evsel->idx	   = idx;
	evsel->attr	   = *attr;
	INIT_LIST_HEAD(&evsel->node);
	hists__init(&evsel->hists);
}

struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx)
{
	struct perf_evsel *evsel = zalloc(sizeof(*evsel));

	if (evsel != NULL)
		perf_evsel__init(evsel, attr, idx);

	return evsel;
}

void perf_evsel__config(struct perf_evsel *evsel, struct perf_record_opts *opts,
			struct perf_evsel *first)
{
	struct perf_event_attr *attr = &evsel->attr;
	int track = !evsel->idx; /* only the first counter needs these */

	attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1;
	attr->inherit	    = !opts->no_inherit;
	attr->read_format   = PERF_FORMAT_TOTAL_TIME_ENABLED |
			      PERF_FORMAT_TOTAL_TIME_RUNNING |
			      PERF_FORMAT_ID;

	attr->sample_type  |= PERF_SAMPLE_IP | PERF_SAMPLE_TID;

	/*
	 * We default some events to a 1 default interval. But keep
	 * it a weak assumption overridable by the user.
	 */
	if (!attr->sample_period || (opts->user_freq != UINT_MAX &&
				     opts->user_interval != ULLONG_MAX)) {
		if (opts->freq) {
			attr->sample_type	|= PERF_SAMPLE_PERIOD;
			attr->freq		= 1;
			attr->sample_freq	= opts->freq;
		} else {
			attr->sample_period = opts->default_interval;
		}
	}

	if (opts->no_samples)
		attr->sample_freq = 0;

	if (opts->inherit_stat)
		attr->inherit_stat = 1;

	if (opts->sample_address) {
		attr->sample_type	|= PERF_SAMPLE_ADDR;
		attr->mmap_data = track;
	}

	if (opts->call_graph)
		attr->sample_type	|= PERF_SAMPLE_CALLCHAIN;

	if (opts->target.system_wide)
		attr->sample_type	|= PERF_SAMPLE_CPU;

	if (opts->period)
		attr->sample_type	|= PERF_SAMPLE_PERIOD;

	if (!opts->sample_id_all_missing &&
	    (opts->sample_time || opts->target.system_wide ||
	     !opts->no_inherit || opts->target.cpu_list))
		attr->sample_type	|= PERF_SAMPLE_TIME;

	if (opts->raw_samples) {
		attr->sample_type	|= PERF_SAMPLE_TIME;
		attr->sample_type	|= PERF_SAMPLE_RAW;
		attr->sample_type	|= PERF_SAMPLE_CPU;
	}

	if (opts->no_delay) {
		attr->watermark = 0;
		attr->wakeup_events = 1;
	}
	if (opts->branch_stack) {
		attr->sample_type	|= PERF_SAMPLE_BRANCH_STACK;
		attr->branch_sample_type = opts->branch_stack;
	}

	attr->mmap = track;
	attr->comm = track;

	if (!opts->target.pid && !opts->target.tid &&
	    !opts->target.system_wide && (!opts->group || evsel == first)) {
		attr->disabled = 1;
		attr->enable_on_exec = 1;
	}
}

int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads)
{
	int cpu, thread;
	evsel->fd = xyarray__new(ncpus, nthreads, sizeof(int));

	if (evsel->fd) {
		for (cpu = 0; cpu < ncpus; cpu++) {
			for (thread = 0; thread < nthreads; thread++) {
				FD(evsel, cpu, thread) = -1;
			}
		}
	}

	return evsel->fd != NULL ? 0 : -ENOMEM;
}

int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads)
{
	evsel->sample_id = xyarray__new(ncpus, nthreads, sizeof(struct perf_sample_id));
	if (evsel->sample_id == NULL)
		return -ENOMEM;

	evsel->id = zalloc(ncpus * nthreads * sizeof(u64));
	if (evsel->id == NULL) {
		xyarray__delete(evsel->sample_id);
		evsel->sample_id = NULL;
		return -ENOMEM;
	}

	return 0;
}

int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus)
{
	evsel->counts = zalloc((sizeof(*evsel->counts) +
				(ncpus * sizeof(struct perf_counts_values))));
	return evsel->counts != NULL ? 0 : -ENOMEM;
}

void perf_evsel__free_fd(struct perf_evsel *evsel)
{
	xyarray__delete(evsel->fd);
	evsel->fd = NULL;
}

void perf_evsel__free_id(struct perf_evsel *evsel)
{
	xyarray__delete(evsel->sample_id);
	evsel->sample_id = NULL;
	free(evsel->id);
	evsel->id = NULL;
}

void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads)
{
	int cpu, thread;

	for (cpu = 0; cpu < ncpus; cpu++)
		for (thread = 0; thread < nthreads; ++thread) {
			close(FD(evsel, cpu, thread));
			FD(evsel, cpu, thread) = -1;
		}
}

void perf_evsel__exit(struct perf_evsel *evsel)
{
	assert(list_empty(&evsel->node));
	xyarray__delete(evsel->fd);
	xyarray__delete(evsel->sample_id);
	free(evsel->id);
}

void perf_evsel__delete(struct perf_evsel *evsel)
{
	perf_evsel__exit(evsel);
	close_cgroup(evsel->cgrp);
	free(evsel->name);
	free(evsel);
}

int __perf_evsel__read_on_cpu(struct perf_evsel *evsel,
			      int cpu, int thread, bool scale)
{
	struct perf_counts_values count;
	size_t nv = scale ? 3 : 1;

	if (FD(evsel, cpu, thread) < 0)
		return -EINVAL;

	if (evsel->counts == NULL && perf_evsel__alloc_counts(evsel, cpu + 1) < 0)
		return -ENOMEM;

	if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0)
		return -errno;

	if (scale) {
		if (count.run == 0)
			count.val = 0;
		else if (count.run < count.ena)
			count.val = (u64)((double)count.val * count.ena / count.run + 0.5);
	} else
		count.ena = count.run = 0;

	evsel->counts->cpu[cpu] = count;
	return 0;
}

int __perf_evsel__read(struct perf_evsel *evsel,
		       int ncpus, int nthreads, bool scale)
{
	size_t nv = scale ? 3 : 1;
	int cpu, thread;
	struct perf_counts_values *aggr = &evsel->counts->aggr, count;

	aggr->val = aggr->ena = aggr->run = 0;

	for (cpu = 0; cpu < ncpus; cpu++) {
		for (thread = 0; thread < nthreads; thread++) {
			if (FD(evsel, cpu, thread) < 0)
				continue;

			if (readn(FD(evsel, cpu, thread),
				  &count, nv * sizeof(u64)) < 0)
				return -errno;

			aggr->val += count.val;
			if (scale) {
				aggr->ena += count.ena;
				aggr->run += count.run;
			}
		}
	}

	evsel->counts->scaled = 0;
	if (scale) {
		if (aggr->run == 0) {
			evsel->counts->scaled = -1;
			aggr->val = 0;
			return 0;
		}

		if (aggr->run < aggr->ena) {
			evsel->counts->scaled = 1;
			aggr->val = (u64)((double)aggr->val * aggr->ena / aggr->run + 0.5);
		}
	} else
		aggr->ena = aggr->run = 0;

	return 0;
}

static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
			      struct thread_map *threads, bool group,
			      struct xyarray *group_fds)
{
	int cpu, thread;
	unsigned long flags = 0;
	int pid = -1, err;

	if (evsel->fd == NULL &&
	    perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0)
		return -ENOMEM;

	if (evsel->cgrp) {
		flags = PERF_FLAG_PID_CGROUP;
		pid = evsel->cgrp->fd;
	}

	for (cpu = 0; cpu < cpus->nr; cpu++) {
		int group_fd = group_fds ? GROUP_FD(group_fds, cpu) : -1;

		for (thread = 0; thread < threads->nr; thread++) {

			if (!evsel->cgrp)
				pid = threads->map[thread];

			FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
								     pid,
								     cpus->map[cpu],
								     group_fd, flags);
			if (FD(evsel, cpu, thread) < 0) {
				err = -errno;
				goto out_close;
			}

			if (group && group_fd == -1)
				group_fd = FD(evsel, cpu, thread);
		}
	}

	return 0;

out_close:
	do {
		while (--thread >= 0) {
			close(FD(evsel, cpu, thread));
			FD(evsel, cpu, thread) = -1;
		}
		thread = threads->nr;
	} while (--cpu >= 0);
	return err;
}

void perf_evsel__close(struct perf_evsel *evsel, int ncpus, int nthreads)
{
	if (evsel->fd == NULL)
		return;

	perf_evsel__close_fd(evsel, ncpus, nthreads);
	perf_evsel__free_fd(evsel);
	evsel->fd = NULL;
}

static struct {
	struct cpu_map map;
	int cpus[1];
} empty_cpu_map = {
	.map.nr	= 1,
	.cpus	= { -1, },
};

static struct {
	struct thread_map map;
	int threads[1];
} empty_thread_map = {
	.map.nr	 = 1,
	.threads = { -1, },
};

int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
		     struct thread_map *threads, bool group,
		     struct xyarray *group_fd)
{
	if (cpus == NULL) {
		/* Work around old compiler warnings about strict aliasing */
		cpus = &empty_cpu_map.map;
	}

	if (threads == NULL)
		threads = &empty_thread_map.map;

	return __perf_evsel__open(evsel, cpus, threads, group, group_fd);
}

int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
			     struct cpu_map *cpus, bool group,
			     struct xyarray *group_fd)
{
	return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group,
				  group_fd);
}

int perf_evsel__open_per_thread(struct perf_evsel *evsel,
				struct thread_map *threads, bool group,
				struct xyarray *group_fd)
{
	return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group,
				  group_fd);
}

static int perf_event__parse_id_sample(const union perf_event *event, u64 type,
				       struct perf_sample *sample)
{
	const u64 *array = event->sample.array;

	array += ((event->header.size -
		   sizeof(event->header)) / sizeof(u64)) - 1;

	if (type & PERF_SAMPLE_CPU) {
		u32 *p = (u32 *)array;
		sample->cpu = *p;
		array--;
	}

	if (type & PERF_SAMPLE_STREAM_ID) {
		sample->stream_id = *array;
		array--;
	}

	if (type & PERF_SAMPLE_ID) {
		sample->id = *array;
		array--;
	}

	if (type & PERF_SAMPLE_TIME) {
		sample->time = *array;
		array--;
	}

	if (type & PERF_SAMPLE_TID) {
		u32 *p = (u32 *)array;
		sample->pid = p[0];
		sample->tid = p[1];
	}

	return 0;
}

static bool sample_overlap(const union perf_event *event,
			   const void *offset, u64 size)
{
	const void *base = event;

	if (offset + size > base + event->header.size)
		return true;

	return false;
}

int perf_event__parse_sample(const union perf_event *event, u64 type,
			     int sample_size, bool sample_id_all,
			     struct perf_sample *data, bool swapped)
{
	const u64 *array;

	/*
	 * used for cross-endian analysis. See git commit 65014ab3
	 * for why this goofiness is needed.
	 */
	union {
		u64 val64;
		u32 val32[2];
	} u;

	memset(data, 0, sizeof(*data));
	data->cpu = data->pid = data->tid = -1;
	data->stream_id = data->id = data->time = -1ULL;
	data->period = 1;

	if (event->header.type != PERF_RECORD_SAMPLE) {
		if (!sample_id_all)
			return 0;
		return perf_event__parse_id_sample(event, type, data);
	}

	array = event->sample.array;

	if (sample_size + sizeof(event->header) > event->header.size)
		return -EFAULT;

	if (type & PERF_SAMPLE_IP) {
		data->ip = event->ip.ip;
		array++;
	}

	if (type & PERF_SAMPLE_TID) {
		u.val64 = *array;
		if (swapped) {
			/* undo swap of u64, then swap on individual u32s */
			u.val64 = bswap_64(u.val64);
			u.val32[0] = bswap_32(u.val32[0]);
			u.val32[1] = bswap_32(u.val32[1]);
		}

		data->pid = u.val32[0];
		data->tid = u.val32[1];
		array++;
	}

	if (type & PERF_SAMPLE_TIME) {
		data->time = *array;
		array++;
	}

	data->addr = 0;
	if (type & PERF_SAMPLE_ADDR) {
		data->addr = *array;
		array++;
	}

	data->id = -1ULL;
	if (type & PERF_SAMPLE_ID) {
		data->id = *array;
		array++;
	}

	if (type & PERF_SAMPLE_STREAM_ID) {
		data->stream_id = *array;
		array++;
	}

	if (type & PERF_SAMPLE_CPU) {

		u.val64 = *array;
		if (swapped) {
			/* undo swap of u64, then swap on individual u32s */
			u.val64 = bswap_64(u.val64);
			u.val32[0] = bswap_32(u.val32[0]);
		}

		data->cpu = u.val32[0];
		array++;
	}

	if (type & PERF_SAMPLE_PERIOD) {
		data->period = *array;
		array++;
	}

	if (type & PERF_SAMPLE_READ) {
		fprintf(stderr, "PERF_SAMPLE_READ is unsupported for now\n");
		return -1;
	}

	if (type & PERF_SAMPLE_CALLCHAIN) {
		if (sample_overlap(event, array, sizeof(data->callchain->nr)))
			return -EFAULT;

		data->callchain = (struct ip_callchain *)array;

		if (sample_overlap(event, array, data->callchain->nr))
			return -EFAULT;

		array += 1 + data->callchain->nr;
	}

	if (type & PERF_SAMPLE_RAW) {
		const u64 *pdata;

		u.val64 = *array;
		if (WARN_ONCE(swapped,
			      "Endianness of raw data not corrected!\n")) {
			/* undo swap of u64, then swap on individual u32s */
			u.val64 = bswap_64(u.val64);
			u.val32[0] = bswap_32(u.val32[0]);
			u.val32[1] = bswap_32(u.val32[1]);
		}

		if (sample_overlap(event, array, sizeof(u32)))
			return -EFAULT;

		data->raw_size = u.val32[0];
		pdata = (void *) array + sizeof(u32);

		if (sample_overlap(event, pdata, data->raw_size))
			return -EFAULT;

		data->raw_data = (void *) pdata;

		array = (void *)array + data->raw_size + sizeof(u32);
	}

	if (type & PERF_SAMPLE_BRANCH_STACK) {
		u64 sz;

		data->branch_stack = (struct branch_stack *)array;
		array++; /* nr */

		sz = data->branch_stack->nr * sizeof(struct branch_entry);
		sz /= sizeof(u64);
		array += sz;
	}
	return 0;
}

int perf_event__synthesize_sample(union perf_event *event, u64 type,
				  const struct perf_sample *sample,
				  bool swapped)
{
	u64 *array;

	/*
	 * used for cross-endian analysis. See git commit 65014ab3
	 * for why this goofiness is needed.
	 */
	union {
		u64 val64;
		u32 val32[2];
	} u;

	array = event->sample.array;

	if (type & PERF_SAMPLE_IP) {
		event->ip.ip = sample->ip;
		array++;
	}

	if (type & PERF_SAMPLE_TID) {
		u.val32[0] = sample->pid;
		u.val32[1] = sample->tid;
		if (swapped) {
			/*
			 * Inverse of what is done in perf_event__parse_sample
			 */
			u.val32[0] = bswap_32(u.val32[0]);
			u.val32[1] = bswap_32(u.val32[1]);
			u.val64 = bswap_64(u.val64);
		}

		*array = u.val64;
		array++;
	}

	if (type & PERF_SAMPLE_TIME) {
		*array = sample->time;
		array++;
	}

	if (type & PERF_SAMPLE_ADDR) {
		*array = sample->addr;
		array++;
	}

	if (type & PERF_SAMPLE_ID) {
		*array = sample->id;
		array++;
	}

	if (type & PERF_SAMPLE_STREAM_ID) {
		*array = sample->stream_id;
		array++;
	}

	if (type & PERF_SAMPLE_CPU) {
		u.val32[0] = sample->cpu;
		if (swapped) {
			/*
			 * Inverse of what is done in perf_event__parse_sample
			 */
			u.val32[0] = bswap_32(u.val32[0]);
			u.val64 = bswap_64(u.val64);
		}
		*array = u.val64;
		array++;
	}

	if (type & PERF_SAMPLE_PERIOD) {
		*array = sample->period;
		array++;
	}

	return 0;
}
Commit	Line	Data
f8a95309 ACM	1	/*
	2	* Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
	3	*
	4	* Parts came from builtin-{top,stat,record}.c, see those files for further
	5	* copyright notes.
	6	*
	7	* Released under the GPL v2. (and only v2, not any later version)
	8	*/
	9
936be503 DA	10	#include <byteswap.h>
936be503 DA	11	#include "asm/bug.h"
69aad6f1	12	#include "evsel.h"
70082dd9	13	#include "evlist.h"
69aad6f1	14	#include "util.h"
86bd5e86	15	#include "cpumap.h"
fd78260b	16	#include "thread_map.h"
69aad6f1	17
c52b12ed	18	#define FD(e, x, y) ((int )xyarray__entry(e->fd, x, y))
727ab04e	19	#define GROUP_FD(group_fd, cpu) ((int )xyarray__entry(group_fd, cpu, 0))
c52b12ed	20
c2a70653 ACM	21	int __perf_evsel__sample_size(u64 sample_type)
	22	{
	23	u64 mask = sample_type & PERF_SAMPLE_MASK;
	24	int size = 0;
	25	int i;
	26
	27	for (i = 0; i < 64; i++) {
	28	if (mask & (1ULL << i))
	29	size++;
	30	}
	31
	32	size *= sizeof(u64);
	33
	34	return size;
	35	}
	36
4bf9ce1b	37	void hists__init(struct hists *hists)
0e2a5f10 ACM	38	{
	39	memset(hists, 0, sizeof(*hists));
	40	hists->entries_in_array[0] = hists->entries_in_array[1] = RB_ROOT;
	41	hists->entries_in = &hists->entries_in_array[0];
	42	hists->entries_collapsed = RB_ROOT;
	43	hists->entries = RB_ROOT;
	44	pthread_mutex_init(&hists->lock, NULL);
	45	}
	46
ef1d1af2 ACM	47	void perf_evsel__init(struct perf_evsel *evsel,
	48	struct perf_event_attr *attr, int idx)
	49	{
	50	evsel->idx = idx;
	51	evsel->attr = *attr;
	52	INIT_LIST_HEAD(&evsel->node);
1980c2eb	53	hists__init(&evsel->hists);
ef1d1af2 ACM	54	}
ef1d1af2 ACM	55
23a2f3ab	56	struct perf_evsel perf_evsel__new(struct perf_event_attr attr, int idx)
69aad6f1 ACM	57	{
	58	struct perf_evsel evsel = zalloc(sizeof(evsel));
	59
ef1d1af2 ACM	60	if (evsel != NULL)
ef1d1af2 ACM	61	perf_evsel__init(evsel, attr, idx);
69aad6f1 ACM	62
	63	return evsel;
	64	}
	65
5090c6ae NK	66	void perf_evsel__config(struct perf_evsel evsel, struct perf_record_opts opts,
5090c6ae NK	67	struct perf_evsel *first)
0f82ebc4 ACM	68	{
	69	struct perf_event_attr *attr = &evsel->attr;
	70	int track = !evsel->idx; /* only the first counter needs these */
	71
808e1226	72	attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1;
0f82ebc4 ACM	73	attr->inherit = !opts->no_inherit;
	74	attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED \|
	75	PERF_FORMAT_TOTAL_TIME_RUNNING \|
	76	PERF_FORMAT_ID;
	77
	78	attr->sample_type \|= PERF_SAMPLE_IP \| PERF_SAMPLE_TID;
	79
	80	/*
	81	* We default some events to a 1 default interval. But keep
	82	* it a weak assumption overridable by the user.
	83	*/
	84	if (!attr->sample_period \|\| (opts->user_freq != UINT_MAX &&
	85	opts->user_interval != ULLONG_MAX)) {
	86	if (opts->freq) {
	87	attr->sample_type \|= PERF_SAMPLE_PERIOD;
	88	attr->freq = 1;
	89	attr->sample_freq = opts->freq;
	90	} else {
	91	attr->sample_period = opts->default_interval;
	92	}
	93	}
	94
	95	if (opts->no_samples)
	96	attr->sample_freq = 0;
	97
	98	if (opts->inherit_stat)
	99	attr->inherit_stat = 1;
	100
	101	if (opts->sample_address) {
	102	attr->sample_type \|= PERF_SAMPLE_ADDR;
	103	attr->mmap_data = track;
	104	}
	105
	106	if (opts->call_graph)
	107	attr->sample_type \|= PERF_SAMPLE_CALLCHAIN;
	108
bea03405	109	if (opts->target.system_wide)
0f82ebc4 ACM	110	attr->sample_type \|= PERF_SAMPLE_CPU;
0f82ebc4 ACM	111
3e76ac78 AV	112	if (opts->period)
	113	attr->sample_type \|= PERF_SAMPLE_PERIOD;
	114
808e1226	115	if (!opts->sample_id_all_missing &&
bea03405 NK	116	(opts->sample_time \|\| opts->target.system_wide \|\|
bea03405 NK	117	!opts->no_inherit \|\| opts->target.cpu_list))
0f82ebc4 ACM	118	attr->sample_type \|= PERF_SAMPLE_TIME;
	119
	120	if (opts->raw_samples) {
	121	attr->sample_type \|= PERF_SAMPLE_TIME;
	122	attr->sample_type \|= PERF_SAMPLE_RAW;
	123	attr->sample_type \|= PERF_SAMPLE_CPU;
	124	}
	125
	126	if (opts->no_delay) {
	127	attr->watermark = 0;
	128	attr->wakeup_events = 1;
	129	}
bdfebd84 RAV	130	if (opts->branch_stack) {
	131	attr->sample_type \|= PERF_SAMPLE_BRANCH_STACK;
	132	attr->branch_sample_type = opts->branch_stack;
	133	}
0f82ebc4 ACM	134
	135	attr->mmap = track;
	136	attr->comm = track;
	137
bea03405 NK	138	if (!opts->target.pid && !opts->target.tid &&
bea03405 NK	139	!opts->target.system_wide && (!opts->group \|\| evsel == first)) {
0f82ebc4 ACM	140	attr->disabled = 1;
	141	attr->enable_on_exec = 1;
	142	}
	143	}
	144
69aad6f1 ACM	145	int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads)
69aad6f1 ACM	146	{
4af4c955	147	int cpu, thread;
69aad6f1	148	evsel->fd = xyarray__new(ncpus, nthreads, sizeof(int));
4af4c955 DA	149
	150	if (evsel->fd) {
	151	for (cpu = 0; cpu < ncpus; cpu++) {
	152	for (thread = 0; thread < nthreads; thread++) {
	153	FD(evsel, cpu, thread) = -1;
	154	}
	155	}
	156	}
	157
69aad6f1 ACM	158	return evsel->fd != NULL ? 0 : -ENOMEM;
	159	}
	160
70db7533 ACM	161	int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads)
70db7533 ACM	162	{
a91e5431 ACM	163	evsel->sample_id = xyarray__new(ncpus, nthreads, sizeof(struct perf_sample_id));
	164	if (evsel->sample_id == NULL)
	165	return -ENOMEM;
	166
	167	evsel->id = zalloc(ncpus * nthreads * sizeof(u64));
	168	if (evsel->id == NULL) {
	169	xyarray__delete(evsel->sample_id);
	170	evsel->sample_id = NULL;
	171	return -ENOMEM;
	172	}
	173
	174	return 0;
70db7533 ACM	175	}
70db7533 ACM	176
c52b12ed ACM	177	int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus)
	178	{
	179	evsel->counts = zalloc((sizeof(*evsel->counts) +
	180	(ncpus * sizeof(struct perf_counts_values))));
	181	return evsel->counts != NULL ? 0 : -ENOMEM;
	182	}
	183
69aad6f1 ACM	184	void perf_evsel__free_fd(struct perf_evsel *evsel)
	185	{
	186	xyarray__delete(evsel->fd);
	187	evsel->fd = NULL;
	188	}
	189
70db7533 ACM	190	void perf_evsel__free_id(struct perf_evsel *evsel)
70db7533 ACM	191	{
a91e5431 ACM	192	xyarray__delete(evsel->sample_id);
	193	evsel->sample_id = NULL;
	194	free(evsel->id);
70db7533 ACM	195	evsel->id = NULL;
	196	}
	197
c52b12ed ACM	198	void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads)
	199	{
	200	int cpu, thread;
	201
	202	for (cpu = 0; cpu < ncpus; cpu++)
	203	for (thread = 0; thread < nthreads; ++thread) {
	204	close(FD(evsel, cpu, thread));
	205	FD(evsel, cpu, thread) = -1;
	206	}
	207	}
	208
ef1d1af2	209	void perf_evsel__exit(struct perf_evsel *evsel)
69aad6f1 ACM	210	{
	211	assert(list_empty(&evsel->node));
	212	xyarray__delete(evsel->fd);
a91e5431 ACM	213	xyarray__delete(evsel->sample_id);
a91e5431 ACM	214	free(evsel->id);
ef1d1af2 ACM	215	}
	216
	217	void perf_evsel__delete(struct perf_evsel *evsel)
	218	{
	219	perf_evsel__exit(evsel);
023695d9	220	close_cgroup(evsel->cgrp);
f0c55bcf	221	free(evsel->name);
69aad6f1 ACM	222	free(evsel);
69aad6f1 ACM	223	}
c52b12ed ACM	224
	225	int __perf_evsel__read_on_cpu(struct perf_evsel *evsel,
	226	int cpu, int thread, bool scale)
	227	{
	228	struct perf_counts_values count;
	229	size_t nv = scale ? 3 : 1;
	230
	231	if (FD(evsel, cpu, thread) < 0)
	232	return -EINVAL;
	233
4eed11d5 ACM	234	if (evsel->counts == NULL && perf_evsel__alloc_counts(evsel, cpu + 1) < 0)
	235	return -ENOMEM;
	236
c52b12ed ACM	237	if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0)
	238	return -errno;
	239
	240	if (scale) {
	241	if (count.run == 0)
	242	count.val = 0;
	243	else if (count.run < count.ena)
	244	count.val = (u64)((double)count.val * count.ena / count.run + 0.5);
	245	} else
	246	count.ena = count.run = 0;
	247
	248	evsel->counts->cpu[cpu] = count;
	249	return 0;
	250	}
	251
	252	int __perf_evsel__read(struct perf_evsel *evsel,
	253	int ncpus, int nthreads, bool scale)
	254	{
	255	size_t nv = scale ? 3 : 1;
	256	int cpu, thread;
	257	struct perf_counts_values *aggr = &evsel->counts->aggr, count;
	258
52bcd994	259	aggr->val = aggr->ena = aggr->run = 0;
c52b12ed ACM	260
	261	for (cpu = 0; cpu < ncpus; cpu++) {
	262	for (thread = 0; thread < nthreads; thread++) {
	263	if (FD(evsel, cpu, thread) < 0)
	264	continue;
	265
	266	if (readn(FD(evsel, cpu, thread),
	267	&count, nv * sizeof(u64)) < 0)
	268	return -errno;
	269
	270	aggr->val += count.val;
	271	if (scale) {
	272	aggr->ena += count.ena;
	273	aggr->run += count.run;
	274	}
	275	}
	276	}
	277
	278	evsel->counts->scaled = 0;
	279	if (scale) {
	280	if (aggr->run == 0) {
	281	evsel->counts->scaled = -1;
	282	aggr->val = 0;
	283	return 0;
	284	}
	285
	286	if (aggr->run < aggr->ena) {
	287	evsel->counts->scaled = 1;
	288	aggr->val = (u64)((double)aggr->val * aggr->ena / aggr->run + 0.5);
	289	}
	290	} else
	291	aggr->ena = aggr->run = 0;
	292
	293	return 0;
	294	}
48290609	295
0252208e	296	static int __perf_evsel__open(struct perf_evsel evsel, struct cpu_map cpus,
727ab04e ACM	297	struct thread_map *threads, bool group,
727ab04e ACM	298	struct xyarray *group_fds)
48290609	299	{
0252208e	300	int cpu, thread;
023695d9	301	unsigned long flags = 0;
727ab04e	302	int pid = -1, err;
48290609	303
0252208e ACM	304	if (evsel->fd == NULL &&
0252208e ACM	305	perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0)
727ab04e	306	return -ENOMEM;
4eed11d5	307
023695d9 SE	308	if (evsel->cgrp) {
	309	flags = PERF_FLAG_PID_CGROUP;
	310	pid = evsel->cgrp->fd;
	311	}
	312
86bd5e86	313	for (cpu = 0; cpu < cpus->nr; cpu++) {
727ab04e	314	int group_fd = group_fds ? GROUP_FD(group_fds, cpu) : -1;
9d04f178	315
0252208e	316	for (thread = 0; thread < threads->nr; thread++) {
023695d9 SE	317
	318	if (!evsel->cgrp)
	319	pid = threads->map[thread];
	320
0252208e	321	FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
023695d9	322	pid,
f08199d3	323	cpus->map[cpu],
023695d9	324	group_fd, flags);
727ab04e ACM	325	if (FD(evsel, cpu, thread) < 0) {
727ab04e ACM	326	err = -errno;
0252208e	327	goto out_close;
727ab04e	328	}
f08199d3 ACM	329
	330	if (group && group_fd == -1)
	331	group_fd = FD(evsel, cpu, thread);
0252208e	332	}
48290609 ACM	333	}
	334
	335	return 0;
	336
	337	out_close:
0252208e ACM	338	do {
	339	while (--thread >= 0) {
	340	close(FD(evsel, cpu, thread));
	341	FD(evsel, cpu, thread) = -1;
	342	}
	343	thread = threads->nr;
	344	} while (--cpu >= 0);
727ab04e ACM	345	return err;
	346	}
	347
	348	void perf_evsel__close(struct perf_evsel *evsel, int ncpus, int nthreads)
	349	{
	350	if (evsel->fd == NULL)
	351	return;
	352
	353	perf_evsel__close_fd(evsel, ncpus, nthreads);
	354	perf_evsel__free_fd(evsel);
	355	evsel->fd = NULL;
48290609 ACM	356	}
48290609 ACM	357
0252208e ACM	358	static struct {
	359	struct cpu_map map;
	360	int cpus[1];
	361	} empty_cpu_map = {
	362	.map.nr = 1,
	363	.cpus = { -1, },
	364	};
	365
	366	static struct {
	367	struct thread_map map;
	368	int threads[1];
	369	} empty_thread_map = {
	370	.map.nr = 1,
	371	.threads = { -1, },
	372	};
	373
f08199d3	374	int perf_evsel__open(struct perf_evsel evsel, struct cpu_map cpus,
727ab04e ACM	375	struct thread_map *threads, bool group,
727ab04e ACM	376	struct xyarray *group_fd)
48290609	377	{
0252208e ACM	378	if (cpus == NULL) {
	379	/* Work around old compiler warnings about strict aliasing */
	380	cpus = &empty_cpu_map.map;
48290609 ACM	381	}
48290609 ACM	382
0252208e ACM	383	if (threads == NULL)
0252208e ACM	384	threads = &empty_thread_map.map;
48290609	385
727ab04e	386	return __perf_evsel__open(evsel, cpus, threads, group, group_fd);
48290609 ACM	387	}
48290609 ACM	388
f08199d3	389	int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
727ab04e ACM	390	struct cpu_map *cpus, bool group,
727ab04e ACM	391	struct xyarray *group_fd)
48290609	392	{
727ab04e ACM	393	return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group,
727ab04e ACM	394	group_fd);
0252208e	395	}
48290609	396
f08199d3	397	int perf_evsel__open_per_thread(struct perf_evsel *evsel,
727ab04e ACM	398	struct thread_map *threads, bool group,
727ab04e ACM	399	struct xyarray *group_fd)
0252208e	400	{
727ab04e ACM	401	return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group,
727ab04e ACM	402	group_fd);
48290609	403	}
70082dd9	404
8115d60c ACM	405	static int perf_event__parse_id_sample(const union perf_event *event, u64 type,
8115d60c ACM	406	struct perf_sample *sample)
d0dd74e8 ACM	407	{
	408	const u64 *array = event->sample.array;
	409
	410	array += ((event->header.size -
	411	sizeof(event->header)) / sizeof(u64)) - 1;
	412
	413	if (type & PERF_SAMPLE_CPU) {
	414	u32 p = (u32 )array;
	415	sample->cpu = *p;
	416	array--;
	417	}
	418
	419	if (type & PERF_SAMPLE_STREAM_ID) {
	420	sample->stream_id = *array;
	421	array--;
	422	}
	423
	424	if (type & PERF_SAMPLE_ID) {
	425	sample->id = *array;
	426	array--;
	427	}
	428
	429	if (type & PERF_SAMPLE_TIME) {
	430	sample->time = *array;
	431	array--;
	432	}
	433
	434	if (type & PERF_SAMPLE_TID) {
	435	u32 p = (u32 )array;
	436	sample->pid = p[0];
	437	sample->tid = p[1];
	438	}
	439
	440	return 0;
	441	}
	442
98e1da90 FW	443	static bool sample_overlap(const union perf_event *event,
	444	const void *offset, u64 size)
	445	{
	446	const void *base = event;
	447
	448	if (offset + size > base + event->header.size)
	449	return true;
	450
	451	return false;
	452	}
	453
8115d60c	454	int perf_event__parse_sample(const union perf_event *event, u64 type,
a2854124	455	int sample_size, bool sample_id_all,
936be503	456	struct perf_sample *data, bool swapped)
d0dd74e8 ACM	457	{
	458	const u64 *array;
	459
936be503 DA	460	/*
	461	* used for cross-endian analysis. See git commit 65014ab3
	462	* for why this goofiness is needed.
	463	*/
	464	union {
	465	u64 val64;
	466	u32 val32[2];
	467	} u;
	468
f3bda2c9	469	memset(data, 0, sizeof(*data));
d0dd74e8 ACM	470	data->cpu = data->pid = data->tid = -1;
d0dd74e8 ACM	471	data->stream_id = data->id = data->time = -1ULL;
a4a03fc7	472	data->period = 1;
d0dd74e8 ACM	473
	474	if (event->header.type != PERF_RECORD_SAMPLE) {
	475	if (!sample_id_all)
	476	return 0;
8115d60c	477	return perf_event__parse_id_sample(event, type, data);
d0dd74e8 ACM	478	}
	479
	480	array = event->sample.array;
	481
a2854124 FW	482	if (sample_size + sizeof(event->header) > event->header.size)
	483	return -EFAULT;
	484
d0dd74e8 ACM	485	if (type & PERF_SAMPLE_IP) {
	486	data->ip = event->ip.ip;
	487	array++;
	488	}
	489
	490	if (type & PERF_SAMPLE_TID) {
936be503 DA	491	u.val64 = *array;
	492	if (swapped) {
	493	/* undo swap of u64, then swap on individual u32s */
	494	u.val64 = bswap_64(u.val64);
	495	u.val32[0] = bswap_32(u.val32[0]);
	496	u.val32[1] = bswap_32(u.val32[1]);
	497	}
	498
	499	data->pid = u.val32[0];
	500	data->tid = u.val32[1];
d0dd74e8 ACM	501	array++;
	502	}
	503
	504	if (type & PERF_SAMPLE_TIME) {
	505	data->time = *array;
	506	array++;
	507	}
	508
7cec0922	509	data->addr = 0;
d0dd74e8 ACM	510	if (type & PERF_SAMPLE_ADDR) {
	511	data->addr = *array;
	512	array++;
	513	}
	514
	515	data->id = -1ULL;
	516	if (type & PERF_SAMPLE_ID) {
	517	data->id = *array;
	518	array++;
	519	}
	520
	521	if (type & PERF_SAMPLE_STREAM_ID) {
	522	data->stream_id = *array;
	523	array++;
	524	}
	525
	526	if (type & PERF_SAMPLE_CPU) {
936be503 DA	527
	528	u.val64 = *array;
	529	if (swapped) {
	530	/* undo swap of u64, then swap on individual u32s */
	531	u.val64 = bswap_64(u.val64);
	532	u.val32[0] = bswap_32(u.val32[0]);
	533	}
	534
	535	data->cpu = u.val32[0];
d0dd74e8 ACM	536	array++;
	537	}
	538
	539	if (type & PERF_SAMPLE_PERIOD) {
	540	data->period = *array;
	541	array++;
	542	}
	543
	544	if (type & PERF_SAMPLE_READ) {
f9d36996	545	fprintf(stderr, "PERF_SAMPLE_READ is unsupported for now\n");
d0dd74e8 ACM	546	return -1;
	547	}
	548
	549	if (type & PERF_SAMPLE_CALLCHAIN) {
98e1da90 FW	550	if (sample_overlap(event, array, sizeof(data->callchain->nr)))
	551	return -EFAULT;
	552
d0dd74e8	553	data->callchain = (struct ip_callchain *)array;
98e1da90 FW	554
	555	if (sample_overlap(event, array, data->callchain->nr))
	556	return -EFAULT;
	557
d0dd74e8 ACM	558	array += 1 + data->callchain->nr;
	559	}
	560
	561	if (type & PERF_SAMPLE_RAW) {
8e303f20 JO	562	const u64 *pdata;
8e303f20 JO	563
936be503 DA	564	u.val64 = *array;
	565	if (WARN_ONCE(swapped,
	566	"Endianness of raw data not corrected!\n")) {
	567	/* undo swap of u64, then swap on individual u32s */
	568	u.val64 = bswap_64(u.val64);
	569	u.val32[0] = bswap_32(u.val32[0]);
	570	u.val32[1] = bswap_32(u.val32[1]);
	571	}
98e1da90 FW	572
	573	if (sample_overlap(event, array, sizeof(u32)))
	574	return -EFAULT;
	575
936be503	576	data->raw_size = u.val32[0];
8e303f20	577	pdata = (void *) array + sizeof(u32);
98e1da90	578
8e303f20	579	if (sample_overlap(event, pdata, data->raw_size))
98e1da90 FW	580	return -EFAULT;
98e1da90 FW	581
8e303f20	582	data->raw_data = (void *) pdata;
fa30c964 SE	583
fa30c964 SE	584	array = (void *)array + data->raw_size + sizeof(u32);
d0dd74e8 ACM	585	}
d0dd74e8 ACM	586
b5387528 RAV	587	if (type & PERF_SAMPLE_BRANCH_STACK) {
	588	u64 sz;
	589
	590	data->branch_stack = (struct branch_stack *)array;
	591	array++; /* nr */
	592
	593	sz = data->branch_stack->nr * sizeof(struct branch_entry);
	594	sz /= sizeof(u64);
	595	array += sz;
	596	}
d0dd74e8 ACM	597	return 0;
d0dd74e8 ACM	598	}
74eec26f AV	599
	600	int perf_event__synthesize_sample(union perf_event *event, u64 type,
	601	const struct perf_sample *sample,
	602	bool swapped)
	603	{
	604	u64 *array;
	605
	606	/*
	607	* used for cross-endian analysis. See git commit 65014ab3
	608	* for why this goofiness is needed.
	609	*/
	610	union {
	611	u64 val64;
	612	u32 val32[2];
	613	} u;
	614
	615	array = event->sample.array;
	616
	617	if (type & PERF_SAMPLE_IP) {
	618	event->ip.ip = sample->ip;
	619	array++;
	620	}
	621
	622	if (type & PERF_SAMPLE_TID) {
	623	u.val32[0] = sample->pid;
	624	u.val32[1] = sample->tid;
	625	if (swapped) {
	626	/*
	627	* Inverse of what is done in perf_event__parse_sample
	628	*/
	629	u.val32[0] = bswap_32(u.val32[0]);
	630	u.val32[1] = bswap_32(u.val32[1]);
	631	u.val64 = bswap_64(u.val64);
	632	}
	633
	634	*array = u.val64;
	635	array++;
	636	}
	637
	638	if (type & PERF_SAMPLE_TIME) {
	639	*array = sample->time;
	640	array++;
	641	}
	642
	643	if (type & PERF_SAMPLE_ADDR) {
	644	*array = sample->addr;
	645	array++;
	646	}
	647
	648	if (type & PERF_SAMPLE_ID) {
	649	*array = sample->id;
	650	array++;
	651	}
	652
	653	if (type & PERF_SAMPLE_STREAM_ID) {
	654	*array = sample->stream_id;
	655	array++;
	656	}
	657
	658	if (type & PERF_SAMPLE_CPU) {
	659	u.val32[0] = sample->cpu;
	660	if (swapped) {
	661	/*
	662	* Inverse of what is done in perf_event__parse_sample
663	*/
664	u.val32[0] = bswap_32(u.val32[0]);
665	u.val64 = bswap_64(u.val64);
666	}
667	*array = u.val64;
668	array++;
669	}
670
671	if (type & PERF_SAMPLE_PERIOD) {
672	*array = sample->period;
673	array++;
674	}
675
676	return 0;
677	}