]>
Commit | Line | Data |
---|---|---|
76369139 FW |
1 | /* |
2 | * Performance events ring-buffer code: | |
3 | * | |
4 | * Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de> | |
5 | * Copyright (C) 2008-2011 Red Hat, Inc., Ingo Molnar | |
6 | * Copyright (C) 2008-2011 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> | |
d36b6910 | 7 | * Copyright © 2009 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> |
76369139 FW |
8 | * |
9 | * For licensing details see kernel-base/COPYING | |
10 | */ | |
11 | ||
12 | #include <linux/perf_event.h> | |
13 | #include <linux/vmalloc.h> | |
14 | #include <linux/slab.h> | |
26c86da8 | 15 | #include <linux/circ_buf.h> |
7c60fc0e | 16 | #include <linux/poll.h> |
76369139 FW |
17 | |
18 | #include "internal.h" | |
19 | ||
76369139 FW |
20 | static void perf_output_wakeup(struct perf_output_handle *handle) |
21 | { | |
7c60fc0e | 22 | atomic_set(&handle->rb->poll, POLLIN); |
76369139 | 23 | |
a8b0ca17 PZ |
24 | handle->event->pending_wakeup = 1; |
25 | irq_work_queue(&handle->event->pending); | |
76369139 FW |
26 | } |
27 | ||
28 | /* | |
29 | * We need to ensure a later event_id doesn't publish a head when a former | |
30 | * event isn't done writing. However since we need to deal with NMIs we | |
31 | * cannot fully serialize things. | |
32 | * | |
33 | * We only publish the head (and generate a wakeup) when the outer-most | |
34 | * event completes. | |
35 | */ | |
36 | static void perf_output_get_handle(struct perf_output_handle *handle) | |
37 | { | |
38 | struct ring_buffer *rb = handle->rb; | |
39 | ||
40 | preempt_disable(); | |
41 | local_inc(&rb->nest); | |
42 | handle->wakeup = local_read(&rb->wakeup); | |
43 | } | |
44 | ||
45 | static void perf_output_put_handle(struct perf_output_handle *handle) | |
46 | { | |
47 | struct ring_buffer *rb = handle->rb; | |
48 | unsigned long head; | |
49 | ||
50 | again: | |
51 | head = local_read(&rb->head); | |
52 | ||
53 | /* | |
54 | * IRQ/NMI can happen here, which means we can miss a head update. | |
55 | */ | |
56 | ||
57 | if (!local_dec_and_test(&rb->nest)) | |
58 | goto out; | |
59 | ||
60 | /* | |
bf378d34 PZ |
61 | * Since the mmap() consumer (userspace) can run on a different CPU: |
62 | * | |
63 | * kernel user | |
64 | * | |
c7f2e3cd PZ |
65 | * if (LOAD ->data_tail) { LOAD ->data_head |
66 | * (A) smp_rmb() (C) | |
67 | * STORE $data LOAD $data | |
68 | * smp_wmb() (B) smp_mb() (D) | |
69 | * STORE ->data_head STORE ->data_tail | |
70 | * } | |
bf378d34 PZ |
71 | * |
72 | * Where A pairs with D, and B pairs with C. | |
73 | * | |
c7f2e3cd PZ |
74 | * In our case (A) is a control dependency that separates the load of |
75 | * the ->data_tail and the stores of $data. In case ->data_tail | |
76 | * indicates there is no room in the buffer to store $data we do not. | |
bf378d34 | 77 | * |
c7f2e3cd | 78 | * D needs to be a full barrier since it separates the data READ |
bf378d34 PZ |
79 | * from the tail WRITE. |
80 | * | |
81 | * For B a WMB is sufficient since it separates two WRITEs, and for C | |
82 | * an RMB is sufficient since it separates two READs. | |
83 | * | |
84 | * See perf_output_begin(). | |
76369139 | 85 | */ |
c7f2e3cd | 86 | smp_wmb(); /* B, matches C */ |
76369139 FW |
87 | rb->user_page->data_head = head; |
88 | ||
89 | /* | |
394570b7 PZ |
90 | * Now check if we missed an update -- rely on previous implied |
91 | * compiler barriers to force a re-read. | |
76369139 FW |
92 | */ |
93 | if (unlikely(head != local_read(&rb->head))) { | |
94 | local_inc(&rb->nest); | |
95 | goto again; | |
96 | } | |
97 | ||
98 | if (handle->wakeup != local_read(&rb->wakeup)) | |
99 | perf_output_wakeup(handle); | |
100 | ||
101 | out: | |
102 | preempt_enable(); | |
103 | } | |
104 | ||
105 | int perf_output_begin(struct perf_output_handle *handle, | |
a7ac67ea | 106 | struct perf_event *event, unsigned int size) |
76369139 FW |
107 | { |
108 | struct ring_buffer *rb; | |
109 | unsigned long tail, offset, head; | |
524feca5 | 110 | int have_lost, page_shift; |
76369139 FW |
111 | struct { |
112 | struct perf_event_header header; | |
113 | u64 id; | |
114 | u64 lost; | |
115 | } lost_event; | |
116 | ||
117 | rcu_read_lock(); | |
118 | /* | |
119 | * For inherited events we send all the output towards the parent. | |
120 | */ | |
121 | if (event->parent) | |
122 | event = event->parent; | |
123 | ||
124 | rb = rcu_dereference(event->rb); | |
c72b42a3 | 125 | if (unlikely(!rb)) |
76369139 FW |
126 | goto out; |
127 | ||
c72b42a3 | 128 | if (unlikely(!rb->nr_pages)) |
76369139 FW |
129 | goto out; |
130 | ||
c72b42a3 PZ |
131 | handle->rb = rb; |
132 | handle->event = event; | |
133 | ||
76369139 | 134 | have_lost = local_read(&rb->lost); |
c72b42a3 | 135 | if (unlikely(have_lost)) { |
d20a973f PZ |
136 | size += sizeof(lost_event); |
137 | if (event->attr.sample_id_all) | |
138 | size += event->id_header_size; | |
76369139 FW |
139 | } |
140 | ||
141 | perf_output_get_handle(handle); | |
142 | ||
143 | do { | |
76369139 | 144 | tail = ACCESS_ONCE(rb->user_page->data_tail); |
76369139 | 145 | offset = head = local_read(&rb->head); |
26c86da8 PZ |
146 | if (!rb->overwrite && |
147 | unlikely(CIRC_SPACE(head, tail, perf_data_size(rb)) < size)) | |
76369139 | 148 | goto fail; |
c7f2e3cd PZ |
149 | |
150 | /* | |
151 | * The above forms a control dependency barrier separating the | |
152 | * @tail load above from the data stores below. Since the @tail | |
153 | * load is required to compute the branch to fail below. | |
154 | * | |
155 | * A, matches D; the full memory barrier userspace SHOULD issue | |
156 | * after reading the data and before storing the new tail | |
157 | * position. | |
158 | * | |
159 | * See perf_output_put_handle(). | |
160 | */ | |
161 | ||
26c86da8 | 162 | head += size; |
76369139 FW |
163 | } while (local_cmpxchg(&rb->head, offset, head) != offset); |
164 | ||
85f59edf | 165 | /* |
c7f2e3cd PZ |
166 | * We rely on the implied barrier() by local_cmpxchg() to ensure |
167 | * none of the data stores below can be lifted up by the compiler. | |
85f59edf | 168 | */ |
85f59edf | 169 | |
c72b42a3 | 170 | if (unlikely(head - local_read(&rb->wakeup) > rb->watermark)) |
76369139 FW |
171 | local_add(rb->watermark, &rb->wakeup); |
172 | ||
524feca5 PZ |
173 | page_shift = PAGE_SHIFT + page_order(rb); |
174 | ||
175 | handle->page = (offset >> page_shift) & (rb->nr_pages - 1); | |
176 | offset &= (1UL << page_shift) - 1; | |
177 | handle->addr = rb->data_pages[handle->page] + offset; | |
178 | handle->size = (1UL << page_shift) - offset; | |
76369139 | 179 | |
c72b42a3 | 180 | if (unlikely(have_lost)) { |
d20a973f PZ |
181 | struct perf_sample_data sample_data; |
182 | ||
183 | lost_event.header.size = sizeof(lost_event); | |
76369139 FW |
184 | lost_event.header.type = PERF_RECORD_LOST; |
185 | lost_event.header.misc = 0; | |
186 | lost_event.id = event->id; | |
187 | lost_event.lost = local_xchg(&rb->lost, 0); | |
188 | ||
d20a973f PZ |
189 | perf_event_header__init_id(&lost_event.header, |
190 | &sample_data, event); | |
76369139 FW |
191 | perf_output_put(handle, lost_event); |
192 | perf_event__output_id_sample(event, handle, &sample_data); | |
193 | } | |
194 | ||
195 | return 0; | |
196 | ||
197 | fail: | |
198 | local_inc(&rb->lost); | |
199 | perf_output_put_handle(handle); | |
200 | out: | |
201 | rcu_read_unlock(); | |
202 | ||
203 | return -ENOSPC; | |
204 | } | |
205 | ||
91d7753a | 206 | unsigned int perf_output_copy(struct perf_output_handle *handle, |
76369139 FW |
207 | const void *buf, unsigned int len) |
208 | { | |
91d7753a | 209 | return __output_copy(handle, buf, len); |
76369139 FW |
210 | } |
211 | ||
5685e0ff JO |
212 | unsigned int perf_output_skip(struct perf_output_handle *handle, |
213 | unsigned int len) | |
214 | { | |
215 | return __output_skip(handle, NULL, len); | |
216 | } | |
217 | ||
76369139 FW |
218 | void perf_output_end(struct perf_output_handle *handle) |
219 | { | |
76369139 FW |
220 | perf_output_put_handle(handle); |
221 | rcu_read_unlock(); | |
222 | } | |
223 | ||
224 | static void | |
225 | ring_buffer_init(struct ring_buffer *rb, long watermark, int flags) | |
226 | { | |
227 | long max_size = perf_data_size(rb); | |
228 | ||
229 | if (watermark) | |
230 | rb->watermark = min(max_size, watermark); | |
231 | ||
232 | if (!rb->watermark) | |
233 | rb->watermark = max_size / 2; | |
234 | ||
235 | if (flags & RING_BUFFER_WRITABLE) | |
dd9c086d SE |
236 | rb->overwrite = 0; |
237 | else | |
238 | rb->overwrite = 1; | |
76369139 FW |
239 | |
240 | atomic_set(&rb->refcount, 1); | |
10c6db11 PZ |
241 | |
242 | INIT_LIST_HEAD(&rb->event_list); | |
243 | spin_lock_init(&rb->event_lock); | |
76369139 FW |
244 | } |
245 | ||
fdc26706 AS |
246 | /* |
247 | * This is called before hardware starts writing to the AUX area to | |
248 | * obtain an output handle and make sure there's room in the buffer. | |
249 | * When the capture completes, call perf_aux_output_end() to commit | |
250 | * the recorded data to the buffer. | |
251 | * | |
252 | * The ordering is similar to that of perf_output_{begin,end}, with | |
253 | * the exception of (B), which should be taken care of by the pmu | |
254 | * driver, since ordering rules will differ depending on hardware. | |
255 | */ | |
256 | void *perf_aux_output_begin(struct perf_output_handle *handle, | |
257 | struct perf_event *event) | |
258 | { | |
259 | struct perf_event *output_event = event; | |
260 | unsigned long aux_head, aux_tail; | |
261 | struct ring_buffer *rb; | |
262 | ||
263 | if (output_event->parent) | |
264 | output_event = output_event->parent; | |
265 | ||
266 | /* | |
267 | * Since this will typically be open across pmu::add/pmu::del, we | |
268 | * grab ring_buffer's refcount instead of holding rcu read lock | |
269 | * to make sure it doesn't disappear under us. | |
270 | */ | |
271 | rb = ring_buffer_get(output_event); | |
272 | if (!rb) | |
273 | return NULL; | |
274 | ||
275 | if (!rb_has_aux(rb) || !atomic_inc_not_zero(&rb->aux_refcount)) | |
276 | goto err; | |
277 | ||
278 | /* | |
279 | * Nesting is not supported for AUX area, make sure nested | |
280 | * writers are caught early | |
281 | */ | |
282 | if (WARN_ON_ONCE(local_xchg(&rb->aux_nest, 1))) | |
283 | goto err_put; | |
284 | ||
285 | aux_head = local_read(&rb->aux_head); | |
286 | aux_tail = ACCESS_ONCE(rb->user_page->aux_tail); | |
287 | ||
288 | handle->rb = rb; | |
289 | handle->event = event; | |
290 | handle->head = aux_head; | |
291 | if (aux_head - aux_tail < perf_aux_size(rb)) | |
292 | handle->size = CIRC_SPACE(aux_head, aux_tail, perf_aux_size(rb)); | |
293 | else | |
294 | handle->size = 0; | |
295 | ||
296 | /* | |
297 | * handle->size computation depends on aux_tail load; this forms a | |
298 | * control dependency barrier separating aux_tail load from aux data | |
299 | * store that will be enabled on successful return | |
300 | */ | |
301 | if (!handle->size) { /* A, matches D */ | |
302 | event->pending_disable = 1; | |
303 | perf_output_wakeup(handle); | |
304 | local_set(&rb->aux_nest, 0); | |
305 | goto err_put; | |
306 | } | |
307 | ||
308 | return handle->rb->aux_priv; | |
309 | ||
310 | err_put: | |
311 | rb_free_aux(rb); | |
312 | ||
313 | err: | |
314 | ring_buffer_put(rb); | |
315 | handle->event = NULL; | |
316 | ||
317 | return NULL; | |
318 | } | |
319 | ||
320 | /* | |
321 | * Commit the data written by hardware into the ring buffer by adjusting | |
322 | * aux_head and posting a PERF_RECORD_AUX into the perf buffer. It is the | |
323 | * pmu driver's responsibility to observe ordering rules of the hardware, | |
324 | * so that all the data is externally visible before this is called. | |
325 | */ | |
326 | void perf_aux_output_end(struct perf_output_handle *handle, unsigned long size, | |
327 | bool truncated) | |
328 | { | |
329 | struct ring_buffer *rb = handle->rb; | |
330 | unsigned long aux_head = local_read(&rb->aux_head); | |
331 | u64 flags = 0; | |
332 | ||
333 | if (truncated) | |
334 | flags |= PERF_AUX_FLAG_TRUNCATED; | |
335 | ||
336 | local_add(size, &rb->aux_head); | |
337 | ||
338 | if (size || flags) { | |
339 | /* | |
340 | * Only send RECORD_AUX if we have something useful to communicate | |
341 | */ | |
342 | ||
343 | perf_event_aux_event(handle->event, aux_head, size, flags); | |
344 | } | |
345 | ||
346 | rb->user_page->aux_head = local_read(&rb->aux_head); | |
347 | ||
348 | perf_output_wakeup(handle); | |
349 | handle->event = NULL; | |
350 | ||
351 | local_set(&rb->aux_nest, 0); | |
352 | rb_free_aux(rb); | |
353 | ring_buffer_put(rb); | |
354 | } | |
355 | ||
356 | /* | |
357 | * Skip over a given number of bytes in the AUX buffer, due to, for example, | |
358 | * hardware's alignment constraints. | |
359 | */ | |
360 | int perf_aux_output_skip(struct perf_output_handle *handle, unsigned long size) | |
361 | { | |
362 | struct ring_buffer *rb = handle->rb; | |
363 | unsigned long aux_head; | |
364 | ||
365 | if (size > handle->size) | |
366 | return -ENOSPC; | |
367 | ||
368 | local_add(size, &rb->aux_head); | |
369 | ||
370 | handle->head = aux_head; | |
371 | handle->size -= size; | |
372 | ||
373 | return 0; | |
374 | } | |
375 | ||
376 | void *perf_get_aux(struct perf_output_handle *handle) | |
377 | { | |
378 | /* this is only valid between perf_aux_output_begin and *_end */ | |
379 | if (!handle->event) | |
380 | return NULL; | |
381 | ||
382 | return handle->rb->aux_priv; | |
383 | } | |
384 | ||
0a4e38e6 AS |
385 | #define PERF_AUX_GFP (GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY) |
386 | ||
387 | static struct page *rb_alloc_aux_page(int node, int order) | |
388 | { | |
389 | struct page *page; | |
390 | ||
391 | if (order > MAX_ORDER) | |
392 | order = MAX_ORDER; | |
393 | ||
394 | do { | |
395 | page = alloc_pages_node(node, PERF_AUX_GFP, order); | |
396 | } while (!page && order--); | |
397 | ||
398 | if (page && order) { | |
399 | /* | |
400 | * Communicate the allocation size to the driver | |
401 | */ | |
402 | split_page(page, order); | |
403 | SetPagePrivate(page); | |
404 | set_page_private(page, order); | |
405 | } | |
406 | ||
407 | return page; | |
408 | } | |
409 | ||
410 | static void rb_free_aux_page(struct ring_buffer *rb, int idx) | |
411 | { | |
412 | struct page *page = virt_to_page(rb->aux_pages[idx]); | |
413 | ||
414 | ClearPagePrivate(page); | |
415 | page->mapping = NULL; | |
416 | __free_page(page); | |
417 | } | |
418 | ||
45bfb2e5 PZ |
419 | int rb_alloc_aux(struct ring_buffer *rb, struct perf_event *event, |
420 | pgoff_t pgoff, int nr_pages, int flags) | |
421 | { | |
422 | bool overwrite = !(flags & RING_BUFFER_WRITABLE); | |
423 | int node = (event->cpu == -1) ? -1 : cpu_to_node(event->cpu); | |
0a4e38e6 | 424 | int ret = -ENOMEM, max_order = 0; |
45bfb2e5 PZ |
425 | |
426 | if (!has_aux(event)) | |
427 | return -ENOTSUPP; | |
428 | ||
6a279230 | 429 | if (event->pmu->capabilities & PERF_PMU_CAP_AUX_NO_SG) { |
0a4e38e6 AS |
430 | /* |
431 | * We need to start with the max_order that fits in nr_pages, | |
432 | * not the other way around, hence ilog2() and not get_order. | |
433 | */ | |
434 | max_order = ilog2(nr_pages); | |
435 | ||
6a279230 AS |
436 | /* |
437 | * PMU requests more than one contiguous chunks of memory | |
438 | * for SW double buffering | |
439 | */ | |
440 | if ((event->pmu->capabilities & PERF_PMU_CAP_AUX_SW_DOUBLEBUF) && | |
441 | !overwrite) { | |
442 | if (!max_order) | |
443 | return -EINVAL; | |
444 | ||
445 | max_order--; | |
446 | } | |
447 | } | |
448 | ||
45bfb2e5 PZ |
449 | rb->aux_pages = kzalloc_node(nr_pages * sizeof(void *), GFP_KERNEL, node); |
450 | if (!rb->aux_pages) | |
451 | return -ENOMEM; | |
452 | ||
453 | rb->free_aux = event->pmu->free_aux; | |
0a4e38e6 | 454 | for (rb->aux_nr_pages = 0; rb->aux_nr_pages < nr_pages;) { |
45bfb2e5 | 455 | struct page *page; |
0a4e38e6 | 456 | int last, order; |
45bfb2e5 | 457 | |
0a4e38e6 AS |
458 | order = min(max_order, ilog2(nr_pages - rb->aux_nr_pages)); |
459 | page = rb_alloc_aux_page(node, order); | |
45bfb2e5 PZ |
460 | if (!page) |
461 | goto out; | |
462 | ||
0a4e38e6 AS |
463 | for (last = rb->aux_nr_pages + (1 << page_private(page)); |
464 | last > rb->aux_nr_pages; rb->aux_nr_pages++) | |
465 | rb->aux_pages[rb->aux_nr_pages] = page_address(page++); | |
45bfb2e5 PZ |
466 | } |
467 | ||
468 | rb->aux_priv = event->pmu->setup_aux(event->cpu, rb->aux_pages, nr_pages, | |
469 | overwrite); | |
470 | if (!rb->aux_priv) | |
471 | goto out; | |
472 | ||
473 | ret = 0; | |
474 | ||
475 | /* | |
476 | * aux_pages (and pmu driver's private data, aux_priv) will be | |
477 | * referenced in both producer's and consumer's contexts, thus | |
478 | * we keep a refcount here to make sure either of the two can | |
479 | * reference them safely. | |
480 | */ | |
481 | atomic_set(&rb->aux_refcount, 1); | |
482 | ||
483 | out: | |
484 | if (!ret) | |
485 | rb->aux_pgoff = pgoff; | |
486 | else | |
487 | rb_free_aux(rb); | |
488 | ||
489 | return ret; | |
490 | } | |
491 | ||
492 | static void __rb_free_aux(struct ring_buffer *rb) | |
493 | { | |
494 | int pg; | |
495 | ||
496 | if (rb->aux_priv) { | |
497 | rb->free_aux(rb->aux_priv); | |
498 | rb->free_aux = NULL; | |
499 | rb->aux_priv = NULL; | |
500 | } | |
501 | ||
502 | for (pg = 0; pg < rb->aux_nr_pages; pg++) | |
0a4e38e6 | 503 | rb_free_aux_page(rb, pg); |
45bfb2e5 PZ |
504 | |
505 | kfree(rb->aux_pages); | |
506 | rb->aux_nr_pages = 0; | |
507 | } | |
508 | ||
509 | void rb_free_aux(struct ring_buffer *rb) | |
510 | { | |
511 | if (atomic_dec_and_test(&rb->aux_refcount)) | |
512 | __rb_free_aux(rb); | |
513 | } | |
514 | ||
76369139 FW |
515 | #ifndef CONFIG_PERF_USE_VMALLOC |
516 | ||
517 | /* | |
518 | * Back perf_mmap() with regular GFP_KERNEL-0 pages. | |
519 | */ | |
520 | ||
45bfb2e5 PZ |
521 | static struct page * |
522 | __perf_mmap_to_page(struct ring_buffer *rb, unsigned long pgoff) | |
76369139 FW |
523 | { |
524 | if (pgoff > rb->nr_pages) | |
525 | return NULL; | |
526 | ||
527 | if (pgoff == 0) | |
528 | return virt_to_page(rb->user_page); | |
529 | ||
530 | return virt_to_page(rb->data_pages[pgoff - 1]); | |
531 | } | |
532 | ||
533 | static void *perf_mmap_alloc_page(int cpu) | |
534 | { | |
535 | struct page *page; | |
536 | int node; | |
537 | ||
538 | node = (cpu == -1) ? cpu : cpu_to_node(cpu); | |
539 | page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0); | |
540 | if (!page) | |
541 | return NULL; | |
542 | ||
543 | return page_address(page); | |
544 | } | |
545 | ||
546 | struct ring_buffer *rb_alloc(int nr_pages, long watermark, int cpu, int flags) | |
547 | { | |
548 | struct ring_buffer *rb; | |
549 | unsigned long size; | |
550 | int i; | |
551 | ||
552 | size = sizeof(struct ring_buffer); | |
553 | size += nr_pages * sizeof(void *); | |
554 | ||
555 | rb = kzalloc(size, GFP_KERNEL); | |
556 | if (!rb) | |
557 | goto fail; | |
558 | ||
559 | rb->user_page = perf_mmap_alloc_page(cpu); | |
560 | if (!rb->user_page) | |
561 | goto fail_user_page; | |
562 | ||
563 | for (i = 0; i < nr_pages; i++) { | |
564 | rb->data_pages[i] = perf_mmap_alloc_page(cpu); | |
565 | if (!rb->data_pages[i]) | |
566 | goto fail_data_pages; | |
567 | } | |
568 | ||
569 | rb->nr_pages = nr_pages; | |
570 | ||
571 | ring_buffer_init(rb, watermark, flags); | |
572 | ||
573 | return rb; | |
574 | ||
575 | fail_data_pages: | |
576 | for (i--; i >= 0; i--) | |
577 | free_page((unsigned long)rb->data_pages[i]); | |
578 | ||
579 | free_page((unsigned long)rb->user_page); | |
580 | ||
581 | fail_user_page: | |
582 | kfree(rb); | |
583 | ||
584 | fail: | |
585 | return NULL; | |
586 | } | |
587 | ||
588 | static void perf_mmap_free_page(unsigned long addr) | |
589 | { | |
590 | struct page *page = virt_to_page((void *)addr); | |
591 | ||
592 | page->mapping = NULL; | |
593 | __free_page(page); | |
594 | } | |
595 | ||
596 | void rb_free(struct ring_buffer *rb) | |
597 | { | |
598 | int i; | |
599 | ||
600 | perf_mmap_free_page((unsigned long)rb->user_page); | |
601 | for (i = 0; i < rb->nr_pages; i++) | |
602 | perf_mmap_free_page((unsigned long)rb->data_pages[i]); | |
603 | kfree(rb); | |
604 | } | |
605 | ||
606 | #else | |
5919b309 JO |
607 | static int data_page_nr(struct ring_buffer *rb) |
608 | { | |
609 | return rb->nr_pages << page_order(rb); | |
610 | } | |
76369139 | 611 | |
45bfb2e5 PZ |
612 | static struct page * |
613 | __perf_mmap_to_page(struct ring_buffer *rb, unsigned long pgoff) | |
76369139 | 614 | { |
5919b309 JO |
615 | /* The '>' counts in the user page. */ |
616 | if (pgoff > data_page_nr(rb)) | |
76369139 FW |
617 | return NULL; |
618 | ||
619 | return vmalloc_to_page((void *)rb->user_page + pgoff * PAGE_SIZE); | |
620 | } | |
621 | ||
622 | static void perf_mmap_unmark_page(void *addr) | |
623 | { | |
624 | struct page *page = vmalloc_to_page(addr); | |
625 | ||
626 | page->mapping = NULL; | |
627 | } | |
628 | ||
629 | static void rb_free_work(struct work_struct *work) | |
630 | { | |
631 | struct ring_buffer *rb; | |
632 | void *base; | |
633 | int i, nr; | |
634 | ||
635 | rb = container_of(work, struct ring_buffer, work); | |
5919b309 | 636 | nr = data_page_nr(rb); |
76369139 FW |
637 | |
638 | base = rb->user_page; | |
5919b309 JO |
639 | /* The '<=' counts in the user page. */ |
640 | for (i = 0; i <= nr; i++) | |
76369139 FW |
641 | perf_mmap_unmark_page(base + (i * PAGE_SIZE)); |
642 | ||
643 | vfree(base); | |
644 | kfree(rb); | |
645 | } | |
646 | ||
647 | void rb_free(struct ring_buffer *rb) | |
648 | { | |
649 | schedule_work(&rb->work); | |
650 | } | |
651 | ||
652 | struct ring_buffer *rb_alloc(int nr_pages, long watermark, int cpu, int flags) | |
653 | { | |
654 | struct ring_buffer *rb; | |
655 | unsigned long size; | |
656 | void *all_buf; | |
657 | ||
658 | size = sizeof(struct ring_buffer); | |
659 | size += sizeof(void *); | |
660 | ||
661 | rb = kzalloc(size, GFP_KERNEL); | |
662 | if (!rb) | |
663 | goto fail; | |
664 | ||
665 | INIT_WORK(&rb->work, rb_free_work); | |
666 | ||
667 | all_buf = vmalloc_user((nr_pages + 1) * PAGE_SIZE); | |
668 | if (!all_buf) | |
669 | goto fail_all_buf; | |
670 | ||
671 | rb->user_page = all_buf; | |
672 | rb->data_pages[0] = all_buf + PAGE_SIZE; | |
673 | rb->page_order = ilog2(nr_pages); | |
5919b309 | 674 | rb->nr_pages = !!nr_pages; |
76369139 FW |
675 | |
676 | ring_buffer_init(rb, watermark, flags); | |
677 | ||
678 | return rb; | |
679 | ||
680 | fail_all_buf: | |
681 | kfree(rb); | |
682 | ||
683 | fail: | |
684 | return NULL; | |
685 | } | |
686 | ||
687 | #endif | |
45bfb2e5 PZ |
688 | |
689 | struct page * | |
690 | perf_mmap_to_page(struct ring_buffer *rb, unsigned long pgoff) | |
691 | { | |
692 | if (rb->aux_nr_pages) { | |
693 | /* above AUX space */ | |
694 | if (pgoff > rb->aux_pgoff + rb->aux_nr_pages) | |
695 | return NULL; | |
696 | ||
697 | /* AUX space */ | |
698 | if (pgoff >= rb->aux_pgoff) | |
699 | return virt_to_page(rb->aux_pages[pgoff - rb->aux_pgoff]); | |
700 | } | |
701 | ||
702 | return __perf_mmap_to_page(rb, pgoff); | |
703 | } |