[linux-block.git] / tools / perf / util / evlist.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 <poll.h>
#include "cpumap.h"
#include "thread_map.h"
#include "evlist.h"
#include "evsel.h"
#include "util.h"

#include <sys/mman.h>

#include <linux/bitops.h>
#include <linux/hash.h>

#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
#define SID(e, x, y) xyarray__entry(e->id, x, y)

void perf_evlist__init(struct perf_evlist *evlist)
{
	int i;

	for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i)
		INIT_HLIST_HEAD(&evlist->heads[i]);
	INIT_LIST_HEAD(&evlist->entries);
}

struct perf_evlist *perf_evlist__new(void)
{
	struct perf_evlist *evlist = zalloc(sizeof(*evlist));

	if (evlist != NULL)
		perf_evlist__init(evlist);

	return evlist;
}

static void perf_evlist__purge(struct perf_evlist *evlist)
{
	struct perf_evsel *pos, *n;

	list_for_each_entry_safe(pos, n, &evlist->entries, node) {
		list_del_init(&pos->node);
		perf_evsel__delete(pos);
	}

	evlist->nr_entries = 0;
}

void perf_evlist__exit(struct perf_evlist *evlist)
{
	free(evlist->mmap);
	free(evlist->pollfd);
	evlist->mmap = NULL;
	evlist->pollfd = NULL;
}

void perf_evlist__delete(struct perf_evlist *evlist)
{
	perf_evlist__purge(evlist);
	perf_evlist__exit(evlist);
	free(evlist);
}

void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry)
{
	list_add_tail(&entry->node, &evlist->entries);
	++evlist->nr_entries;
}

int perf_evlist__add_default(struct perf_evlist *evlist)
{
	struct perf_event_attr attr = {
		.type = PERF_TYPE_HARDWARE,
		.config = PERF_COUNT_HW_CPU_CYCLES,
	};
	struct perf_evsel *evsel = perf_evsel__new(&attr, 0);

	if (evsel == NULL)
		return -ENOMEM;

	perf_evlist__add(evlist, evsel);
	return 0;
}

int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads)
{
	int nfds = ncpus * nthreads * evlist->nr_entries;
	evlist->pollfd = malloc(sizeof(struct pollfd) * nfds);
	return evlist->pollfd != NULL ? 0 : -ENOMEM;
}

void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd)
{
	fcntl(fd, F_SETFL, O_NONBLOCK);
	evlist->pollfd[evlist->nr_fds].fd = fd;
	evlist->pollfd[evlist->nr_fds].events = POLLIN;
	evlist->nr_fds++;
}

static int perf_evlist__id_hash(struct perf_evlist *evlist, struct perf_evsel *evsel,
			       int cpu, int thread, int fd)
{
	struct perf_sample_id *sid;
	u64 read_data[4] = { 0, };
	int hash, id_idx = 1; /* The first entry is the counter value */

	if (!(evsel->attr.read_format & PERF_FORMAT_ID) ||
	    read(fd, &read_data, sizeof(read_data)) == -1)
		return -1;

	if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
		++id_idx;
	if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
		++id_idx;

	sid = SID(evsel, cpu, thread);
	sid->id = read_data[id_idx];
	sid->evsel = evsel;
	hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS);
	hlist_add_head(&sid->node, &evlist->heads[hash]);
	return 0;
}

struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id)
{
	struct hlist_head *head;
	struct hlist_node *pos;
	struct perf_sample_id *sid;
	int hash;

	if (evlist->nr_entries == 1)
		return list_entry(evlist->entries.next, struct perf_evsel, node);

	hash = hash_64(id, PERF_EVLIST__HLIST_BITS);
	head = &evlist->heads[hash];

	hlist_for_each_entry(sid, pos, head, node)
		if (sid->id == id)
			return sid->evsel;
	return NULL;
}

union perf_event *perf_evlist__read_on_cpu(struct perf_evlist *evlist, int cpu)
{
	/* XXX Move this to perf.c, making it generally available */
	unsigned int page_size = sysconf(_SC_PAGE_SIZE);
	struct perf_mmap *md = &evlist->mmap[cpu];
	unsigned int head = perf_mmap__read_head(md);
	unsigned int old = md->prev;
	unsigned char *data = md->base + page_size;
	union perf_event *event = NULL;

	if (evlist->overwrite) {
		/*
		 * If we're further behind than half the buffer, there's a chance
		 * the writer will bite our tail and mess up the samples under us.
		 *
		 * If we somehow ended up ahead of the head, we got messed up.
		 *
		 * In either case, truncate and restart at head.
		 */
		int diff = head - old;
		if (diff > md->mask / 2 || diff < 0) {
			fprintf(stderr, "WARNING: failed to keep up with mmap data.\n");

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

	if (old != head) {
		size_t size;

		event = (union perf_event *)&data[old & md->mask];
		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 = &evlist->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 = &evlist->event_copy;
		}

		old += size;
	}

	md->prev = old;

	if (!evlist->overwrite)
		perf_mmap__write_tail(md, old);

	return event;
}

void perf_evlist__munmap(struct perf_evlist *evlist, int ncpus)
{
	int cpu;

	for (cpu = 0; cpu < ncpus; cpu++) {
		if (evlist->mmap[cpu].base != NULL) {
			munmap(evlist->mmap[cpu].base, evlist->mmap_len);
			evlist->mmap[cpu].base = NULL;
		}
	}
}

int perf_evlist__alloc_mmap(struct perf_evlist *evlist, int ncpus)
{
	evlist->mmap = zalloc(ncpus * sizeof(struct perf_mmap));
	return evlist->mmap != NULL ? 0 : -ENOMEM;
}

static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot,
			       int mask, int fd)
{
	evlist->mmap[cpu].prev = 0;
	evlist->mmap[cpu].mask = mask;
	evlist->mmap[cpu].base = mmap(NULL, evlist->mmap_len, prot,
				      MAP_SHARED, fd, 0);
	if (evlist->mmap[cpu].base == MAP_FAILED)
		return -1;

	perf_evlist__add_pollfd(evlist, fd);
	return 0;
}

/** perf_evlist__mmap - Create per cpu maps to receive events
 *
 * @evlist - list of events
 * @cpus - cpu map being monitored
 * @threads - threads map being monitored
 * @pages - map length in pages
 * @overwrite - overwrite older events?
 *
 * If overwrite is false the user needs to signal event consuption using:
 *
 *	struct perf_mmap *m = &evlist->mmap[cpu];
 *	unsigned int head = perf_mmap__read_head(m);
 *
 *	perf_mmap__write_tail(m, head)
 */
int perf_evlist__mmap(struct perf_evlist *evlist, struct cpu_map *cpus,
		      struct thread_map *threads, int pages, bool overwrite)
{
	unsigned int page_size = sysconf(_SC_PAGE_SIZE);
	int mask = pages * page_size - 1, cpu;
	struct perf_evsel *first_evsel, *evsel;
	int thread, prot = PROT_READ | (overwrite ? 0 : PROT_WRITE);

	if (evlist->mmap == NULL &&
	    perf_evlist__alloc_mmap(evlist, cpus->nr) < 0)
		return -ENOMEM;

	if (evlist->pollfd == NULL &&
	    perf_evlist__alloc_pollfd(evlist, cpus->nr, threads->nr) < 0)
		return -ENOMEM;

	evlist->overwrite = overwrite;
	evlist->mmap_len = (pages + 1) * page_size;
	first_evsel = list_entry(evlist->entries.next, struct perf_evsel, node);

	list_for_each_entry(evsel, &evlist->entries, node) {
		if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
		    evsel->id == NULL &&
		    perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0)
			return -ENOMEM;

		for (cpu = 0; cpu < cpus->nr; cpu++) {
			for (thread = 0; thread < threads->nr; thread++) {
				int fd = FD(evsel, cpu, thread);

				if (evsel->idx || thread) {
					if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT,
						  FD(first_evsel, cpu, 0)) != 0)
						goto out_unmap;
				} else if (__perf_evlist__mmap(evlist, cpu, prot, mask, fd) < 0)
					goto out_unmap;

				if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
				    perf_evlist__id_hash(evlist, evsel, cpu, thread, fd) < 0)
					goto out_unmap;
			}
		}
	}

	return 0;

out_unmap:
	for (cpu = 0; cpu < cpus->nr; cpu++) {
		if (evlist->mmap[cpu].base != NULL) {
			munmap(evlist->mmap[cpu].base, evlist->mmap_len);
			evlist->mmap[cpu].base = NULL;
		}
	}
	return -1;
}
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	*/
5c581041	9	#include <poll.h>
f8a95309 ACM	10	#include "cpumap.h"
f8a95309 ACM	11	#include "thread_map.h"
361c99a6 ACM	12	#include "evlist.h"
	13	#include "evsel.h"
	14	#include "util.h"
	15
f8a95309 ACM	16	#include <sys/mman.h>
f8a95309 ACM	17
70db7533 ACM	18	#include <linux/bitops.h>
	19	#include <linux/hash.h>
	20
f8a95309 ACM	21	#define FD(e, x, y) ((int )xyarray__entry(e->fd, x, y))
	22	#define SID(e, x, y) xyarray__entry(e->id, x, y)
	23
ef1d1af2 ACM	24	void perf_evlist__init(struct perf_evlist *evlist)
	25	{
	26	int i;
	27
	28	for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i)
	29	INIT_HLIST_HEAD(&evlist->heads[i]);
	30	INIT_LIST_HEAD(&evlist->entries);
	31	}
	32
361c99a6 ACM	33	struct perf_evlist *perf_evlist__new(void)
	34	{
	35	struct perf_evlist evlist = zalloc(sizeof(evlist));
	36
ef1d1af2 ACM	37	if (evlist != NULL)
ef1d1af2 ACM	38	perf_evlist__init(evlist);
361c99a6 ACM	39
	40	return evlist;
	41	}
	42
	43	static void perf_evlist__purge(struct perf_evlist *evlist)
	44	{
	45	struct perf_evsel pos, n;
	46
	47	list_for_each_entry_safe(pos, n, &evlist->entries, node) {
	48	list_del_init(&pos->node);
	49	perf_evsel__delete(pos);
	50	}
	51
	52	evlist->nr_entries = 0;
	53	}
	54
ef1d1af2	55	void perf_evlist__exit(struct perf_evlist *evlist)
361c99a6	56	{
70db7533	57	free(evlist->mmap);
5c581041	58	free(evlist->pollfd);
ef1d1af2 ACM	59	evlist->mmap = NULL;
	60	evlist->pollfd = NULL;
	61	}
	62
	63	void perf_evlist__delete(struct perf_evlist *evlist)
	64	{
	65	perf_evlist__purge(evlist);
	66	perf_evlist__exit(evlist);
361c99a6 ACM	67	free(evlist);
	68	}
	69
	70	void perf_evlist__add(struct perf_evlist evlist, struct perf_evsel entry)
	71	{
	72	list_add_tail(&entry->node, &evlist->entries);
	73	++evlist->nr_entries;
	74	}
	75
	76	int perf_evlist__add_default(struct perf_evlist *evlist)
	77	{
	78	struct perf_event_attr attr = {
	79	.type = PERF_TYPE_HARDWARE,
	80	.config = PERF_COUNT_HW_CPU_CYCLES,
	81	};
	82	struct perf_evsel *evsel = perf_evsel__new(&attr, 0);
	83
	84	if (evsel == NULL)
	85	return -ENOMEM;
	86
	87	perf_evlist__add(evlist, evsel);
	88	return 0;
	89	}
5c581041 ACM	90
	91	int perf_evlist__alloc_pollfd(struct perf_evlist *evlist, int ncpus, int nthreads)
	92	{
	93	int nfds = ncpus * nthreads * evlist->nr_entries;
	94	evlist->pollfd = malloc(sizeof(struct pollfd) * nfds);
	95	return evlist->pollfd != NULL ? 0 : -ENOMEM;
	96	}
70082dd9 ACM	97
	98	void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd)
	99	{
	100	fcntl(fd, F_SETFL, O_NONBLOCK);
	101	evlist->pollfd[evlist->nr_fds].fd = fd;
	102	evlist->pollfd[evlist->nr_fds].events = POLLIN;
	103	evlist->nr_fds++;
	104	}
70db7533	105
f8a95309 ACM	106	static int perf_evlist__id_hash(struct perf_evlist evlist, struct perf_evsel evsel,
	107	int cpu, int thread, int fd)
	108	{
	109	struct perf_sample_id *sid;
	110	u64 read_data[4] = { 0, };
	111	int hash, id_idx = 1; /* The first entry is the counter value */
	112
	113	if (!(evsel->attr.read_format & PERF_FORMAT_ID) \|\|
	114	read(fd, &read_data, sizeof(read_data)) == -1)
	115	return -1;
	116
	117	if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
	118	++id_idx;
	119	if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
	120	++id_idx;
	121
	122	sid = SID(evsel, cpu, thread);
	123	sid->id = read_data[id_idx];
	124	sid->evsel = evsel;
	125	hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS);
	126	hlist_add_head(&sid->node, &evlist->heads[hash]);
	127	return 0;
	128	}
	129
70db7533 ACM	130	struct perf_evsel perf_evlist__id2evsel(struct perf_evlist evlist, u64 id)
	131	{
	132	struct hlist_head *head;
	133	struct hlist_node *pos;
	134	struct perf_sample_id *sid;
	135	int hash;
	136
	137	if (evlist->nr_entries == 1)
	138	return list_entry(evlist->entries.next, struct perf_evsel, node);
	139
	140	hash = hash_64(id, PERF_EVLIST__HLIST_BITS);
	141	head = &evlist->heads[hash];
	142
	143	hlist_for_each_entry(sid, pos, head, node)
	144	if (sid->id == id)
	145	return sid->evsel;
	146	return NULL;
	147	}
04391deb	148
8115d60c	149	union perf_event perf_evlist__read_on_cpu(struct perf_evlist evlist, int cpu)
04391deb ACM	150	{
	151	/* XXX Move this to perf.c, making it generally available */
	152	unsigned int page_size = sysconf(_SC_PAGE_SIZE);
	153	struct perf_mmap *md = &evlist->mmap[cpu];
	154	unsigned int head = perf_mmap__read_head(md);
	155	unsigned int old = md->prev;
	156	unsigned char *data = md->base + page_size;
8115d60c	157	union perf_event *event = NULL;
04391deb	158
7bb41152	159	if (evlist->overwrite) {
04391deb	160	/*
7bb41152 ACM	161	* If we're further behind than half the buffer, there's a chance
	162	* the writer will bite our tail and mess up the samples under us.
	163	*
	164	* If we somehow ended up ahead of the head, we got messed up.
	165	*
	166	* In either case, truncate and restart at head.
04391deb	167	*/
7bb41152 ACM	168	int diff = head - old;
	169	if (diff > md->mask / 2 \|\| diff < 0) {
	170	fprintf(stderr, "WARNING: failed to keep up with mmap data.\n");
	171
	172	/*
	173	* head points to a known good entry, start there.
	174	*/
	175	old = head;
	176	}
04391deb ACM	177	}
	178
	179	if (old != head) {
	180	size_t size;
	181
8115d60c	182	event = (union perf_event *)&data[old & md->mask];
04391deb ACM	183	size = event->header.size;
	184
	185	/*
	186	* Event straddles the mmap boundary -- header should always
	187	* be inside due to u64 alignment of output.
	188	*/
	189	if ((old & md->mask) + size != ((old + size) & md->mask)) {
	190	unsigned int offset = old;
	191	unsigned int len = min(sizeof(*event), size), cpy;
	192	void *dst = &evlist->event_copy;
	193
	194	do {
	195	cpy = min(md->mask + 1 - (offset & md->mask), len);
	196	memcpy(dst, &data[offset & md->mask], cpy);
	197	offset += cpy;
	198	dst += cpy;
	199	len -= cpy;
	200	} while (len);
	201
	202	event = &evlist->event_copy;
	203	}
	204
	205	old += size;
	206	}
	207
	208	md->prev = old;
7bb41152 ACM	209
	210	if (!evlist->overwrite)
	211	perf_mmap__write_tail(md, old);
	212
04391deb ACM	213	return event;
04391deb ACM	214	}
f8a95309 ACM	215
	216	void perf_evlist__munmap(struct perf_evlist *evlist, int ncpus)
	217	{
	218	int cpu;
	219
	220	for (cpu = 0; cpu < ncpus; cpu++) {
	221	if (evlist->mmap[cpu].base != NULL) {
	222	munmap(evlist->mmap[cpu].base, evlist->mmap_len);
	223	evlist->mmap[cpu].base = NULL;
	224	}
	225	}
	226	}
	227
	228	int perf_evlist__alloc_mmap(struct perf_evlist *evlist, int ncpus)
	229	{
	230	evlist->mmap = zalloc(ncpus * sizeof(struct perf_mmap));
	231	return evlist->mmap != NULL ? 0 : -ENOMEM;
	232	}
	233
	234	static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot,
	235	int mask, int fd)
	236	{
	237	evlist->mmap[cpu].prev = 0;
	238	evlist->mmap[cpu].mask = mask;
	239	evlist->mmap[cpu].base = mmap(NULL, evlist->mmap_len, prot,
	240	MAP_SHARED, fd, 0);
	241	if (evlist->mmap[cpu].base == MAP_FAILED)
	242	return -1;
	243
	244	perf_evlist__add_pollfd(evlist, fd);
	245	return 0;
	246	}
	247
	248	/** perf_evlist__mmap - Create per cpu maps to receive events
	249	*
	250	* @evlist - list of events
	251	* @cpus - cpu map being monitored
	252	* @threads - threads map being monitored
	253	* @pages - map length in pages
	254	* @overwrite - overwrite older events?
	255	*
	256	* If overwrite is false the user needs to signal event consuption using:
	257	*
	258	* struct perf_mmap *m = &evlist->mmap[cpu];
	259	* unsigned int head = perf_mmap__read_head(m);
	260	*
	261	* perf_mmap__write_tail(m, head)
	262	*/
	263	int perf_evlist__mmap(struct perf_evlist evlist, struct cpu_map cpus,
	264	struct thread_map *threads, int pages, bool overwrite)
	265	{
	266	unsigned int page_size = sysconf(_SC_PAGE_SIZE);
	267	int mask = pages * page_size - 1, cpu;
	268	struct perf_evsel first_evsel, evsel;
	269	int thread, prot = PROT_READ \| (overwrite ? 0 : PROT_WRITE);
	270
	271	if (evlist->mmap == NULL &&
	272	perf_evlist__alloc_mmap(evlist, cpus->nr) < 0)
	273	return -ENOMEM;
	274
	275	if (evlist->pollfd == NULL &&
	276	perf_evlist__alloc_pollfd(evlist, cpus->nr, threads->nr) < 0)
	277	return -ENOMEM;
	278
279	evlist->overwrite = overwrite;
280	evlist->mmap_len = (pages + 1) * page_size;
281	first_evsel = list_entry(evlist->entries.next, struct perf_evsel, node);
282
283	list_for_each_entry(evsel, &evlist->entries, node) {
284	if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
285	evsel->id == NULL &&
286	perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0)
287	return -ENOMEM;
288
289	for (cpu = 0; cpu < cpus->nr; cpu++) {
290	for (thread = 0; thread < threads->nr; thread++) {
291	int fd = FD(evsel, cpu, thread);
292
293	if (evsel->idx \|\| thread) {
294	if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT,
295	FD(first_evsel, cpu, 0)) != 0)
296	goto out_unmap;
297	} else if (__perf_evlist__mmap(evlist, cpu, prot, mask, fd) < 0)
298	goto out_unmap;
299
300	if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
301	perf_evlist__id_hash(evlist, evsel, cpu, thread, fd) < 0)
302	goto out_unmap;
303	}
304	}
305	}
306
307	return 0;
308
309	out_unmap:
310	for (cpu = 0; cpu < cpus->nr; cpu++) {
311	if (evlist->mmap[cpu].base != NULL) {
312	munmap(evlist->mmap[cpu].base, evlist->mmap_len);
313	evlist->mmap[cpu].base = NULL;
314	}
315	}
316	return -1;
317	}