~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/tools/perf/util/mmap.c

Version: ~ [ linux-5.8 ] ~ [ linux-5.7.14 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.57 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.138 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.193 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.232 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.232 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2011-2017, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
  3  *
  4  * Parts came from evlist.c 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 
 10 #include <sys/mman.h>
 11 #include <inttypes.h>
 12 #include <asm/bug.h>
 13 #include "debug.h"
 14 #include "event.h"
 15 #include "mmap.h"
 16 #include "util.h" /* page_size */
 17 
 18 size_t perf_mmap__mmap_len(struct perf_mmap *map)
 19 {
 20         return map->mask + 1 + page_size;
 21 }
 22 
 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 *startp, u64 end)
 26 {
 27         unsigned char *data = map->base + page_size;
 28         union perf_event *event = NULL;
 29         int diff = end - *startp;
 30 
 31         if (diff >= (int)sizeof(event->header)) {
 32                 size_t size;
 33 
 34                 event = (union perf_event *)&data[*startp & map->mask];
 35                 size = event->header.size;
 36 
 37                 if (size < sizeof(event->header) || diff < (int)size)
 38                         return NULL;
 39 
 40                 /*
 41                  * Event straddles the mmap boundary -- header should always
 42                  * be inside due to u64 alignment of output.
 43                  */
 44                 if ((*startp & map->mask) + size != ((*startp + size) & map->mask)) {
 45                         unsigned int offset = *startp;
 46                         unsigned int len = min(sizeof(*event), size), cpy;
 47                         void *dst = map->event_copy;
 48 
 49                         do {
 50                                 cpy = min(map->mask + 1 - (offset & map->mask), len);
 51                                 memcpy(dst, &data[offset & map->mask], cpy);
 52                                 offset += cpy;
 53                                 dst += cpy;
 54                                 len -= cpy;
 55                         } while (len);
 56 
 57                         event = (union perf_event *)map->event_copy;
 58                 }
 59 
 60                 *startp += size;
 61         }
 62 
 63         return event;
 64 }
 65 
 66 /*
 67  * Read event from ring buffer one by one.
 68  * Return one event for each call.
 69  *
 70  * Usage:
 71  * perf_mmap__read_init()
 72  * while(event = perf_mmap__read_event()) {
 73  *      //process the event
 74  *      perf_mmap__consume()
 75  * }
 76  * perf_mmap__read_done()
 77  */
 78 union perf_event *perf_mmap__read_event(struct perf_mmap *map)
 79 {
 80         union perf_event *event;
 81 
 82         /*
 83          * Check if event was unmapped due to a POLLHUP/POLLERR.
 84          */
 85         if (!refcount_read(&map->refcnt))
 86                 return NULL;
 87 
 88         /* non-overwirte doesn't pause the ringbuffer */
 89         if (!map->overwrite)
 90                 map->end = perf_mmap__read_head(map);
 91 
 92         event = perf_mmap__read(map, &map->start, map->end);
 93 
 94         if (!map->overwrite)
 95                 map->prev = map->start;
 96 
 97         return event;
 98 }
 99 
100 static bool perf_mmap__empty(struct perf_mmap *map)
101 {
102         return perf_mmap__read_head(map) == map->prev && !map->auxtrace_mmap.base;
103 }
104 
105 void perf_mmap__get(struct perf_mmap *map)
106 {
107         refcount_inc(&map->refcnt);
108 }
109 
110 void perf_mmap__put(struct perf_mmap *map)
111 {
112         BUG_ON(map->base && refcount_read(&map->refcnt) == 0);
113 
114         if (refcount_dec_and_test(&map->refcnt))
115                 perf_mmap__munmap(map);
116 }
117 
118 void perf_mmap__consume(struct perf_mmap *map)
119 {
120         if (!map->overwrite) {
121                 u64 old = map->prev;
122 
123                 perf_mmap__write_tail(map, old);
124         }
125 
126         if (refcount_read(&map->refcnt) == 1 && perf_mmap__empty(map))
127                 perf_mmap__put(map);
128 }
129 
130 int __weak auxtrace_mmap__mmap(struct auxtrace_mmap *mm __maybe_unused,
131                                struct auxtrace_mmap_params *mp __maybe_unused,
132                                void *userpg __maybe_unused,
133                                int fd __maybe_unused)
134 {
135         return 0;
136 }
137 
138 void __weak auxtrace_mmap__munmap(struct auxtrace_mmap *mm __maybe_unused)
139 {
140 }
141 
142 void __weak auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp __maybe_unused,
143                                        off_t auxtrace_offset __maybe_unused,
144                                        unsigned int auxtrace_pages __maybe_unused,
145                                        bool auxtrace_overwrite __maybe_unused)
146 {
147 }
148 
149 void __weak auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp __maybe_unused,
150                                           struct perf_evlist *evlist __maybe_unused,
151                                           int idx __maybe_unused,
152                                           bool per_cpu __maybe_unused)
153 {
154 }
155 
156 #ifdef HAVE_AIO_SUPPORT
157 static int perf_mmap__aio_mmap(struct perf_mmap *map, struct mmap_params *mp)
158 {
159         int delta_max, i, prio;
160 
161         map->aio.nr_cblocks = mp->nr_cblocks;
162         if (map->aio.nr_cblocks) {
163                 map->aio.aiocb = calloc(map->aio.nr_cblocks, sizeof(struct aiocb *));
164                 if (!map->aio.aiocb) {
165                         pr_debug2("failed to allocate aiocb for data buffer, error %m\n");
166                         return -1;
167                 }
168                 map->aio.cblocks = calloc(map->aio.nr_cblocks, sizeof(struct aiocb));
169                 if (!map->aio.cblocks) {
170                         pr_debug2("failed to allocate cblocks for data buffer, error %m\n");
171                         return -1;
172                 }
173                 map->aio.data = calloc(map->aio.nr_cblocks, sizeof(void *));
174                 if (!map->aio.data) {
175                         pr_debug2("failed to allocate data buffer, error %m\n");
176                         return -1;
177                 }
178                 delta_max = sysconf(_SC_AIO_PRIO_DELTA_MAX);
179                 for (i = 0; i < map->aio.nr_cblocks; ++i) {
180                         map->aio.data[i] = malloc(perf_mmap__mmap_len(map));
181                         if (!map->aio.data[i]) {
182                                 pr_debug2("failed to allocate data buffer area, error %m");
183                                 return -1;
184                         }
185                         /*
186                          * Use cblock.aio_fildes value different from -1
187                          * to denote started aio write operation on the
188                          * cblock so it requires explicit record__aio_sync()
189                          * call prior the cblock may be reused again.
190                          */
191                         map->aio.cblocks[i].aio_fildes = -1;
192                         /*
193                          * Allocate cblocks with priority delta to have
194                          * faster aio write system calls because queued requests
195                          * are kept in separate per-prio queues and adding
196                          * a new request will iterate thru shorter per-prio
197                          * list. Blocks with numbers higher than
198                          *  _SC_AIO_PRIO_DELTA_MAX go with priority 0.
199                          */
200                         prio = delta_max - i;
201                         map->aio.cblocks[i].aio_reqprio = prio >= 0 ? prio : 0;
202                 }
203         }
204 
205         return 0;
206 }
207 
208 static void perf_mmap__aio_munmap(struct perf_mmap *map)
209 {
210         int i;
211 
212         for (i = 0; i < map->aio.nr_cblocks; ++i)
213                 zfree(&map->aio.data[i]);
214         if (map->aio.data)
215                 zfree(&map->aio.data);
216         zfree(&map->aio.cblocks);
217         zfree(&map->aio.aiocb);
218 }
219 
220 int perf_mmap__aio_push(struct perf_mmap *md, void *to, int idx,
221                         int push(void *to, struct aiocb *cblock, void *buf, size_t size, off_t off),
222                         off_t *off)
223 {
224         u64 head = perf_mmap__read_head(md);
225         unsigned char *data = md->base + page_size;
226         unsigned long size, size0 = 0;
227         void *buf;
228         int rc = 0;
229 
230         rc = perf_mmap__read_init(md);
231         if (rc < 0)
232                 return (rc == -EAGAIN) ? 0 : -1;
233 
234         /*
235          * md->base data is copied into md->data[idx] buffer to
236          * release space in the kernel buffer as fast as possible,
237          * thru perf_mmap__consume() below.
238          *
239          * That lets the kernel to proceed with storing more
240          * profiling data into the kernel buffer earlier than other
241          * per-cpu kernel buffers are handled.
242          *
243          * Coping can be done in two steps in case the chunk of
244          * profiling data crosses the upper bound of the kernel buffer.
245          * In this case we first move part of data from md->start
246          * till the upper bound and then the reminder from the
247          * beginning of the kernel buffer till the end of
248          * the data chunk.
249          */
250 
251         size = md->end - md->start;
252 
253         if ((md->start & md->mask) + size != (md->end & md->mask)) {
254                 buf = &data[md->start & md->mask];
255                 size = md->mask + 1 - (md->start & md->mask);
256                 md->start += size;
257                 memcpy(md->aio.data[idx], buf, size);
258                 size0 = size;
259         }
260 
261         buf = &data[md->start & md->mask];
262         size = md->end - md->start;
263         md->start += size;
264         memcpy(md->aio.data[idx] + size0, buf, size);
265 
266         /*
267          * Increment md->refcount to guard md->data[idx] buffer
268          * from premature deallocation because md object can be
269          * released earlier than aio write request started
270          * on mmap->data[idx] is complete.
271          *
272          * perf_mmap__put() is done at record__aio_complete()
273          * after started request completion.
274          */
275         perf_mmap__get(md);
276 
277         md->prev = head;
278         perf_mmap__consume(md);
279 
280         rc = push(to, &md->aio.cblocks[idx], md->aio.data[idx], size0 + size, *off);
281         if (!rc) {
282                 *off += size0 + size;
283         } else {
284                 /*
285                  * Decrement md->refcount back if aio write
286                  * operation failed to start.
287                  */
288                 perf_mmap__put(md);
289         }
290 
291         return rc;
292 }
293 #else
294 static int perf_mmap__aio_mmap(struct perf_mmap *map __maybe_unused,
295                                struct mmap_params *mp __maybe_unused)
296 {
297         return 0;
298 }
299 
300 static void perf_mmap__aio_munmap(struct perf_mmap *map __maybe_unused)
301 {
302 }
303 #endif
304 
305 void perf_mmap__munmap(struct perf_mmap *map)
306 {
307         perf_mmap__aio_munmap(map);
308         if (map->base != NULL) {
309                 munmap(map->base, perf_mmap__mmap_len(map));
310                 map->base = NULL;
311                 map->fd = -1;
312                 refcount_set(&map->refcnt, 0);
313         }
314         auxtrace_mmap__munmap(&map->auxtrace_mmap);
315 }
316 
317 int perf_mmap__mmap(struct perf_mmap *map, struct mmap_params *mp, int fd, int cpu)
318 {
319         /*
320          * The last one will be done at perf_mmap__consume(), so that we
321          * make sure we don't prevent tools from consuming every last event in
322          * the ring buffer.
323          *
324          * I.e. we can get the POLLHUP meaning that the fd doesn't exist
325          * anymore, but the last events for it are still in the ring buffer,
326          * waiting to be consumed.
327          *
328          * Tools can chose to ignore this at their own discretion, but the
329          * evlist layer can't just drop it when filtering events in
330          * perf_evlist__filter_pollfd().
331          */
332         refcount_set(&map->refcnt, 2);
333         map->prev = 0;
334         map->mask = mp->mask;
335         map->base = mmap(NULL, perf_mmap__mmap_len(map), mp->prot,
336                          MAP_SHARED, fd, 0);
337         if (map->base == MAP_FAILED) {
338                 pr_debug2("failed to mmap perf event ring buffer, error %d\n",
339                           errno);
340                 map->base = NULL;
341                 return -1;
342         }
343         map->fd = fd;
344         map->cpu = cpu;
345 
346         if (auxtrace_mmap__mmap(&map->auxtrace_mmap,
347                                 &mp->auxtrace_mp, map->base, fd))
348                 return -1;
349 
350         return perf_mmap__aio_mmap(map, mp);
351 }
352 
353 static int overwrite_rb_find_range(void *buf, int mask, u64 *start, u64 *end)
354 {
355         struct perf_event_header *pheader;
356         u64 evt_head = *start;
357         int size = mask + 1;
358 
359         pr_debug2("%s: buf=%p, start=%"PRIx64"\n", __func__, buf, *start);
360         pheader = (struct perf_event_header *)(buf + (*start & mask));
361         while (true) {
362                 if (evt_head - *start >= (unsigned int)size) {
363                         pr_debug("Finished reading overwrite ring buffer: rewind\n");
364                         if (evt_head - *start > (unsigned int)size)
365                                 evt_head -= pheader->size;
366                         *end = evt_head;
367                         return 0;
368                 }
369 
370                 pheader = (struct perf_event_header *)(buf + (evt_head & mask));
371 
372                 if (pheader->size == 0) {
373                         pr_debug("Finished reading overwrite ring buffer: get start\n");
374                         *end = evt_head;
375                         return 0;
376                 }
377 
378                 evt_head += pheader->size;
379                 pr_debug3("move evt_head: %"PRIx64"\n", evt_head);
380         }
381         WARN_ONCE(1, "Shouldn't get here\n");
382         return -1;
383 }
384 
385 /*
386  * Report the start and end of the available data in ringbuffer
387  */
388 static int __perf_mmap__read_init(struct perf_mmap *md)
389 {
390         u64 head = perf_mmap__read_head(md);
391         u64 old = md->prev;
392         unsigned char *data = md->base + page_size;
393         unsigned long size;
394 
395         md->start = md->overwrite ? head : old;
396         md->end = md->overwrite ? old : head;
397 
398         if (md->start == md->end)
399                 return -EAGAIN;
400 
401         size = md->end - md->start;
402         if (size > (unsigned long)(md->mask) + 1) {
403                 if (!md->overwrite) {
404                         WARN_ONCE(1, "failed to keep up with mmap data. (warn only once)\n");
405 
406                         md->prev = head;
407                         perf_mmap__consume(md);
408                         return -EAGAIN;
409                 }
410 
411                 /*
412                  * Backward ring buffer is full. We still have a chance to read
413                  * most of data from it.
414                  */
415                 if (overwrite_rb_find_range(data, md->mask, &md->start, &md->end))
416                         return -EINVAL;
417         }
418 
419         return 0;
420 }
421 
422 int perf_mmap__read_init(struct perf_mmap *map)
423 {
424         /*
425          * Check if event was unmapped due to a POLLHUP/POLLERR.
426          */
427         if (!refcount_read(&map->refcnt))
428                 return -ENOENT;
429 
430         return __perf_mmap__read_init(map);
431 }
432 
433 int perf_mmap__push(struct perf_mmap *md, void *to,
434                     int push(struct perf_mmap *map, void *to, void *buf, size_t size))
435 {
436         u64 head = perf_mmap__read_head(md);
437         unsigned char *data = md->base + page_size;
438         unsigned long size;
439         void *buf;
440         int rc = 0;
441 
442         rc = perf_mmap__read_init(md);
443         if (rc < 0)
444                 return (rc == -EAGAIN) ? 0 : -1;
445 
446         size = md->end - md->start;
447 
448         if ((md->start & md->mask) + size != (md->end & md->mask)) {
449                 buf = &data[md->start & md->mask];
450                 size = md->mask + 1 - (md->start & md->mask);
451                 md->start += size;
452 
453                 if (push(md, to, buf, size) < 0) {
454                         rc = -1;
455                         goto out;
456                 }
457         }
458 
459         buf = &data[md->start & md->mask];
460         size = md->end - md->start;
461         md->start += size;
462 
463         if (push(md, to, buf, size) < 0) {
464                 rc = -1;
465                 goto out;
466         }
467 
468         md->prev = head;
469         perf_mmap__consume(md);
470 out:
471         return rc;
472 }
473 
474 /*
475  * Mandatory for overwrite mode
476  * The direction of overwrite mode is backward.
477  * The last perf_mmap__read() will set tail to map->prev.
478  * Need to correct the map->prev to head which is the end of next read.
479  */
480 void perf_mmap__read_done(struct perf_mmap *map)
481 {
482         /*
483          * Check if event was unmapped due to a POLLHUP/POLLERR.
484          */
485         if (!refcount_read(&map->refcnt))
486                 return;
487 
488         map->prev = perf_mmap__read_head(map);
489 }
490 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp