2 * Copyright (C) 2011-2017, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
4 * Parts came from evlist.c builtin-{top,stat,record}.c, see those files for further
7 * Released under the GPL v2. (and only v2, not any later version)
16 #include "util.h" /* page_size */
18 size_t perf_mmap__mmap_len(struct perf_mmap *map)
20 return map->mask + 1 + page_size;
23 /* When check_messup is true, 'end' must points to a good entry */
24 static union perf_event *perf_mmap__read(struct perf_mmap *map,
25 u64 start, u64 end, u64 *prev)
27 unsigned char *data = map->base + page_size;
28 union perf_event *event = NULL;
29 int diff = end - start;
31 if (diff >= (int)sizeof(event->header)) {
34 event = (union perf_event *)&data[start & map->mask];
35 size = event->header.size;
37 if (size < sizeof(event->header) || diff < (int)size) {
43 * Event straddles the mmap boundary -- header should always
44 * be inside due to u64 alignment of output.
46 if ((start & map->mask) + size != ((start + size) & map->mask)) {
47 unsigned int offset = start;
48 unsigned int len = min(sizeof(*event), size), cpy;
49 void *dst = map->event_copy;
52 cpy = min(map->mask + 1 - (offset & map->mask), len);
53 memcpy(dst, &data[offset & map->mask], cpy);
59 event = (union perf_event *)map->event_copy;
72 union perf_event *perf_mmap__read_forward(struct perf_mmap *map)
78 * Check if event was unmapped due to a POLLHUP/POLLERR.
80 if (!refcount_read(&map->refcnt))
83 head = perf_mmap__read_head(map);
85 return perf_mmap__read(map, old, head, &map->prev);
88 union perf_event *perf_mmap__read_backward(struct perf_mmap *map)
91 u64 start = map->prev;
94 * Check if event was unmapped due to a POLLHUP/POLLERR.
96 if (!refcount_read(&map->refcnt))
99 head = perf_mmap__read_head(map);
104 * 'head' pointer starts from 0. Kernel minus sizeof(record) form
105 * it each time when kernel writes to it, so in fact 'head' is
106 * negative. 'end' pointer is made manually by adding the size of
107 * the ring buffer to 'head' pointer, means the validate data can
108 * read is the whole ring buffer. If 'end' is positive, the ring
109 * buffer has not fully filled, so we must adjust 'end' to 0.
111 * However, since both 'head' and 'end' is unsigned, we can't
112 * simply compare 'end' against 0. Here we compare '-head' and
113 * the size of the ring buffer, where -head is the number of bytes
114 * kernel write to the ring buffer.
116 if (-head < (u64)(map->mask + 1))
119 end = head + map->mask + 1;
121 return perf_mmap__read(map, start, end, &map->prev);
124 void perf_mmap__read_catchup(struct perf_mmap *map)
128 if (!refcount_read(&map->refcnt))
131 head = perf_mmap__read_head(map);
135 static bool perf_mmap__empty(struct perf_mmap *map)
137 return perf_mmap__read_head(map) == map->prev && !map->auxtrace_mmap.base;
140 void perf_mmap__get(struct perf_mmap *map)
142 refcount_inc(&map->refcnt);
145 void perf_mmap__put(struct perf_mmap *map)
147 BUG_ON(map->base && refcount_read(&map->refcnt) == 0);
149 if (refcount_dec_and_test(&map->refcnt))
150 perf_mmap__munmap(map);
153 void perf_mmap__consume(struct perf_mmap *map, bool overwrite)
158 perf_mmap__write_tail(map, old);
161 if (refcount_read(&map->refcnt) == 1 && perf_mmap__empty(map))
165 int __weak auxtrace_mmap__mmap(struct auxtrace_mmap *mm __maybe_unused,
166 struct auxtrace_mmap_params *mp __maybe_unused,
167 void *userpg __maybe_unused,
168 int fd __maybe_unused)
173 void __weak auxtrace_mmap__munmap(struct auxtrace_mmap *mm __maybe_unused)
177 void __weak auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp __maybe_unused,
178 off_t auxtrace_offset __maybe_unused,
179 unsigned int auxtrace_pages __maybe_unused,
180 bool auxtrace_overwrite __maybe_unused)
184 void __weak auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp __maybe_unused,
185 struct perf_evlist *evlist __maybe_unused,
186 int idx __maybe_unused,
187 bool per_cpu __maybe_unused)
191 void perf_mmap__munmap(struct perf_mmap *map)
193 if (map->base != NULL) {
194 munmap(map->base, perf_mmap__mmap_len(map));
197 refcount_set(&map->refcnt, 0);
199 auxtrace_mmap__munmap(&map->auxtrace_mmap);
202 int perf_mmap__mmap(struct perf_mmap *map, struct mmap_params *mp, int fd)
205 * The last one will be done at perf_evlist__mmap_consume(), so that we
206 * make sure we don't prevent tools from consuming every last event in
209 * I.e. we can get the POLLHUP meaning that the fd doesn't exist
210 * anymore, but the last events for it are still in the ring buffer,
211 * waiting to be consumed.
213 * Tools can chose to ignore this at their own discretion, but the
214 * evlist layer can't just drop it when filtering events in
215 * perf_evlist__filter_pollfd().
217 refcount_set(&map->refcnt, 2);
219 map->mask = mp->mask;
220 map->base = mmap(NULL, perf_mmap__mmap_len(map), mp->prot,
222 if (map->base == MAP_FAILED) {
223 pr_debug2("failed to mmap perf event ring buffer, error %d\n",
230 if (auxtrace_mmap__mmap(&map->auxtrace_mmap,
231 &mp->auxtrace_mp, map->base, fd))
237 static int overwrite_rb_find_range(void *buf, int mask, u64 head, u64 *start, u64 *end)
239 struct perf_event_header *pheader;
243 pr_debug2("overwrite_rb_find_range: buf=%p, head=%"PRIx64"\n", buf, head);
244 pheader = (struct perf_event_header *)(buf + (head & mask));
247 if (evt_head - head >= (unsigned int)size) {
248 pr_debug("Finished reading overwrite ring buffer: rewind\n");
249 if (evt_head - head > (unsigned int)size)
250 evt_head -= pheader->size;
255 pheader = (struct perf_event_header *)(buf + (evt_head & mask));
257 if (pheader->size == 0) {
258 pr_debug("Finished reading overwrite ring buffer: get start\n");
263 evt_head += pheader->size;
264 pr_debug3("move evt_head: %"PRIx64"\n", evt_head);
266 WARN_ONCE(1, "Shouldn't get here\n");
271 * Report the start and end of the available data in ringbuffer
273 int perf_mmap__read_init(struct perf_mmap *md, bool overwrite,
274 u64 *startp, u64 *endp)
276 u64 head = perf_mmap__read_head(md);
278 unsigned char *data = md->base + page_size;
281 *startp = overwrite ? head : old;
282 *endp = overwrite ? old : head;
284 if (*startp == *endp)
287 size = *endp - *startp;
288 if (size > (unsigned long)(md->mask) + 1) {
290 WARN_ONCE(1, "failed to keep up with mmap data. (warn only once)\n");
293 perf_mmap__consume(md, overwrite);
298 * Backward ring buffer is full. We still have a chance to read
299 * most of data from it.
301 if (overwrite_rb_find_range(data, md->mask, head, startp, endp))
308 int perf_mmap__push(struct perf_mmap *md, bool overwrite,
309 void *to, int push(void *to, void *buf, size_t size))
311 u64 head = perf_mmap__read_head(md);
313 unsigned char *data = md->base + page_size;
318 rc = perf_mmap__read_init(md, overwrite, &start, &end);
320 return (rc == -EAGAIN) ? 0 : -1;
324 if ((start & md->mask) + size != (end & md->mask)) {
325 buf = &data[start & md->mask];
326 size = md->mask + 1 - (start & md->mask);
329 if (push(to, buf, size) < 0) {
335 buf = &data[start & md->mask];
339 if (push(to, buf, size) < 0) {
345 perf_mmap__consume(md, overwrite);