Add wakeup on subbuffer delivery
[lttng-ust.git] / libringbuffer / ring_buffer_backend.c
1 /*
2 * ring_buffer_backend.c
3 *
4 * Copyright (C) 2005-2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * Dual LGPL v2.1/GPL v2 license.
7 */
8
9 #include <urcu/arch.h>
10
11 #include "ust/core.h"
12
13 #include <ust/ringbuffer-config.h>
14 #include "backend.h"
15 #include "frontend.h"
16 #include "smp.h"
17 #include "shm.h"
18
19 /**
20 * lib_ring_buffer_backend_allocate - allocate a channel buffer
21 * @config: ring buffer instance configuration
22 * @buf: the buffer struct
23 * @size: total size of the buffer
24 * @num_subbuf: number of subbuffers
25 * @extra_reader_sb: need extra subbuffer for reader
26 */
27 static
28 int lib_ring_buffer_backend_allocate(const struct lib_ring_buffer_config *config,
29 struct lib_ring_buffer_backend *bufb,
30 size_t size, size_t num_subbuf,
31 int extra_reader_sb,
32 struct shm_handle *handle,
33 struct shm_object *shmobj)
34 {
35 struct channel_backend *chanb = &shmp(handle, bufb->chan)->backend;
36 unsigned long subbuf_size, mmap_offset = 0;
37 unsigned long num_subbuf_alloc;
38 unsigned long i;
39
40 subbuf_size = chanb->subbuf_size;
41 num_subbuf_alloc = num_subbuf;
42
43 if (extra_reader_sb)
44 num_subbuf_alloc++;
45
46 align_shm(shmobj, __alignof__(struct lib_ring_buffer_backend_pages_shmp));
47 set_shmp(bufb->array, zalloc_shm(shmobj,
48 sizeof(struct lib_ring_buffer_backend_pages_shmp) * num_subbuf_alloc));
49 if (unlikely(!shmp(handle, bufb->array)))
50 goto array_error;
51
52 /*
53 * This is the largest element (the buffer pages) which needs to
54 * be aligned on PAGE_SIZE.
55 */
56 align_shm(shmobj, PAGE_SIZE);
57 set_shmp(bufb->memory_map, zalloc_shm(shmobj,
58 subbuf_size * num_subbuf_alloc));
59 if (unlikely(!shmp(handle, bufb->memory_map)))
60 goto memory_map_error;
61
62 /* Allocate backend pages array elements */
63 for (i = 0; i < num_subbuf_alloc; i++) {
64 align_shm(shmobj, __alignof__(struct lib_ring_buffer_backend_pages));
65 set_shmp(shmp(handle, bufb->array)[i].shmp,
66 zalloc_shm(shmobj,
67 sizeof(struct lib_ring_buffer_backend_pages)));
68 if (!shmp(handle, shmp(handle, bufb->array)[i].shmp))
69 goto free_array;
70 }
71
72 /* Allocate write-side subbuffer table */
73 align_shm(shmobj, __alignof__(struct lib_ring_buffer_backend_subbuffer));
74 set_shmp(bufb->buf_wsb, zalloc_shm(shmobj,
75 sizeof(struct lib_ring_buffer_backend_subbuffer)
76 * num_subbuf));
77 if (unlikely(!shmp(handle, bufb->buf_wsb)))
78 goto free_array;
79
80 for (i = 0; i < num_subbuf; i++)
81 shmp(handle, bufb->buf_wsb)[i].id = subbuffer_id(config, 0, 1, i);
82
83 /* Assign read-side subbuffer table */
84 if (extra_reader_sb)
85 bufb->buf_rsb.id = subbuffer_id(config, 0, 1,
86 num_subbuf_alloc - 1);
87 else
88 bufb->buf_rsb.id = subbuffer_id(config, 0, 1, 0);
89
90 /* Assign pages to page index */
91 for (i = 0; i < num_subbuf_alloc; i++) {
92 struct shm_ref ref;
93
94 ref.index = bufb->memory_map._ref.index;
95 ref.offset = bufb->memory_map._ref.offset;
96 ref.offset += i * subbuf_size;
97
98 set_shmp(shmp(handle, shmp(handle, bufb->array)[i].shmp)->p,
99 ref);
100 if (config->output == RING_BUFFER_MMAP) {
101 shmp(handle, shmp(handle, bufb->array)[i].shmp)->mmap_offset = mmap_offset;
102 mmap_offset += subbuf_size;
103 }
104 }
105
106 return 0;
107
108 free_array:
109 /* bufb->array[i] will be freed by shm teardown */
110 memory_map_error:
111 /* bufb->array will be freed by shm teardown */
112 array_error:
113 return -ENOMEM;
114 }
115
116 int lib_ring_buffer_backend_create(struct lib_ring_buffer_backend *bufb,
117 struct channel_backend *chanb, int cpu,
118 struct shm_handle *handle,
119 struct shm_object *shmobj)
120 {
121 const struct lib_ring_buffer_config *config = chanb->config;
122
123 set_shmp(bufb->chan, handle->chan._ref);
124 bufb->cpu = cpu;
125
126 return lib_ring_buffer_backend_allocate(config, bufb, chanb->buf_size,
127 chanb->num_subbuf,
128 chanb->extra_reader_sb,
129 handle, shmobj);
130 }
131
132 void lib_ring_buffer_backend_free(struct lib_ring_buffer_backend *bufb)
133 {
134 /* bufb->buf_wsb will be freed by shm teardown */
135 /* bufb->array[i] will be freed by shm teardown */
136 /* bufb->array will be freed by shm teardown */
137 bufb->allocated = 0;
138 }
139
140 void lib_ring_buffer_backend_reset(struct lib_ring_buffer_backend *bufb,
141 struct shm_handle *handle)
142 {
143 struct channel_backend *chanb = &shmp(handle, bufb->chan)->backend;
144 const struct lib_ring_buffer_config *config = chanb->config;
145 unsigned long num_subbuf_alloc;
146 unsigned int i;
147
148 num_subbuf_alloc = chanb->num_subbuf;
149 if (chanb->extra_reader_sb)
150 num_subbuf_alloc++;
151
152 for (i = 0; i < chanb->num_subbuf; i++)
153 shmp(handle, bufb->buf_wsb)[i].id = subbuffer_id(config, 0, 1, i);
154 if (chanb->extra_reader_sb)
155 bufb->buf_rsb.id = subbuffer_id(config, 0, 1,
156 num_subbuf_alloc - 1);
157 else
158 bufb->buf_rsb.id = subbuffer_id(config, 0, 1, 0);
159
160 for (i = 0; i < num_subbuf_alloc; i++) {
161 /* Don't reset mmap_offset */
162 v_set(config, &shmp(handle, shmp(handle, bufb->array)[i].shmp)->records_commit, 0);
163 v_set(config, &shmp(handle, shmp(handle, bufb->array)[i].shmp)->records_unread, 0);
164 shmp(handle, shmp(handle, bufb->array)[i].shmp)->data_size = 0;
165 /* Don't reset backend page and virt addresses */
166 }
167 /* Don't reset num_pages_per_subbuf, cpu, allocated */
168 v_set(config, &bufb->records_read, 0);
169 }
170
171 /*
172 * The frontend is responsible for also calling ring_buffer_backend_reset for
173 * each buffer when calling channel_backend_reset.
174 */
175 void channel_backend_reset(struct channel_backend *chanb)
176 {
177 struct channel *chan = caa_container_of(chanb, struct channel, backend);
178 const struct lib_ring_buffer_config *config = chanb->config;
179
180 /*
181 * Don't reset buf_size, subbuf_size, subbuf_size_order,
182 * num_subbuf_order, buf_size_order, extra_reader_sb, num_subbuf,
183 * priv, notifiers, config, cpumask and name.
184 */
185 chanb->start_tsc = config->cb.ring_buffer_clock_read(chan);
186 }
187
188 /**
189 * channel_backend_init - initialize a channel backend
190 * @chanb: channel backend
191 * @name: channel name
192 * @config: client ring buffer configuration
193 * @priv: client private data
194 * @parent: dentry of parent directory, %NULL for root directory
195 * @subbuf_size: size of sub-buffers (> PAGE_SIZE, power of 2)
196 * @num_subbuf: number of sub-buffers (power of 2)
197 * @shm_handle: shared memory handle
198 *
199 * Returns channel pointer if successful, %NULL otherwise.
200 *
201 * Creates per-cpu channel buffers using the sizes and attributes
202 * specified. The created channel buffer files will be named
203 * name_0...name_N-1. File permissions will be %S_IRUSR.
204 *
205 * Called with CPU hotplug disabled.
206 */
207 int channel_backend_init(struct channel_backend *chanb,
208 const char *name,
209 const struct lib_ring_buffer_config *config,
210 void *priv, size_t subbuf_size, size_t num_subbuf,
211 struct shm_handle *handle)
212 {
213 struct channel *chan = caa_container_of(chanb, struct channel, backend);
214 unsigned int i;
215 int ret;
216 size_t shmsize = 0, bufshmsize = 0, num_subbuf_alloc;
217
218 if (!name)
219 return -EPERM;
220
221 if (!(subbuf_size && num_subbuf))
222 return -EPERM;
223
224 /* Check that the subbuffer size is larger than a page. */
225 if (subbuf_size < PAGE_SIZE)
226 return -EINVAL;
227
228 /*
229 * Make sure the number of subbuffers and subbuffer size are power of 2.
230 */
231 CHAN_WARN_ON(chanb, hweight32(subbuf_size) != 1);
232 CHAN_WARN_ON(chanb, hweight32(num_subbuf) != 1);
233
234 ret = subbuffer_id_check_index(config, num_subbuf);
235 if (ret)
236 return ret;
237
238 chanb->priv = priv;
239 chanb->buf_size = num_subbuf * subbuf_size;
240 chanb->subbuf_size = subbuf_size;
241 chanb->buf_size_order = get_count_order(chanb->buf_size);
242 chanb->subbuf_size_order = get_count_order(subbuf_size);
243 chanb->num_subbuf_order = get_count_order(num_subbuf);
244 chanb->extra_reader_sb =
245 (config->mode == RING_BUFFER_OVERWRITE) ? 1 : 0;
246 chanb->num_subbuf = num_subbuf;
247 strncpy(chanb->name, name, NAME_MAX);
248 chanb->name[NAME_MAX - 1] = '\0';
249 chanb->config = config;
250
251 /* Per-cpu buffer size: control (prior to backend) */
252 shmsize = offset_align(shmsize, __alignof__(struct lib_ring_buffer));
253 shmsize += sizeof(struct lib_ring_buffer);
254
255 /* Per-cpu buffer size: backend */
256 /* num_subbuf + 1 is the worse case */
257 num_subbuf_alloc = num_subbuf + 1;
258 shmsize += offset_align(shmsize, __alignof__(struct lib_ring_buffer_backend_pages_shmp));
259 shmsize += sizeof(struct lib_ring_buffer_backend_pages_shmp) * num_subbuf_alloc;
260 shmsize += offset_align(bufshmsize, PAGE_SIZE);
261 shmsize += subbuf_size * num_subbuf_alloc;
262 shmsize += offset_align(bufshmsize, __alignof__(struct lib_ring_buffer_backend_pages));
263 shmsize += sizeof(struct lib_ring_buffer_backend_pages) * num_subbuf_alloc;
264 shmsize += offset_align(bufshmsize, __alignof__(struct lib_ring_buffer_backend_subbuffer));
265 shmsize += sizeof(struct lib_ring_buffer_backend_subbuffer) * num_subbuf;
266 /* Per-cpu buffer size: control (after backend) */
267 shmsize += offset_align(shmsize, __alignof__(struct commit_counters_hot));
268 shmsize += sizeof(struct commit_counters_hot) * num_subbuf;
269 shmsize += offset_align(shmsize, __alignof__(struct commit_counters_cold));
270 shmsize += sizeof(struct commit_counters_cold) * num_subbuf;
271
272 if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
273 struct lib_ring_buffer *buf;
274 /*
275 * We need to allocate for all possible cpus.
276 */
277 for_each_possible_cpu(i) {
278 struct shm_object *shmobj;
279
280 shmobj = shm_object_table_append(handle->table, shmsize);
281 align_shm(shmobj, __alignof__(struct lib_ring_buffer));
282 set_shmp(chanb->buf[i].shmp, zalloc_shm(shmobj, sizeof(struct lib_ring_buffer)));
283 buf = shmp(handle, chanb->buf[i].shmp);
284 if (!buf)
285 goto end;
286 set_shmp(buf->self, chanb->buf[i].shmp._ref);
287 ret = lib_ring_buffer_create(buf, chanb, i,
288 handle, shmobj);
289 if (ret)
290 goto free_bufs; /* cpu hotplug locked */
291 }
292 } else {
293 struct shm_object *shmobj;
294 struct lib_ring_buffer *buf;
295
296 shmobj = shm_object_table_append(handle->table, shmsize);
297 align_shm(shmobj, __alignof__(struct lib_ring_buffer));
298 set_shmp(chanb->buf[0].shmp, zalloc_shm(shmobj, sizeof(struct lib_ring_buffer)));
299 buf = shmp(handle, chanb->buf[0].shmp);
300 if (!buf)
301 goto end;
302 ret = lib_ring_buffer_create(buf, chanb, -1,
303 handle, shmobj);
304 if (ret)
305 goto free_bufs;
306 }
307 chanb->start_tsc = config->cb.ring_buffer_clock_read(chan);
308
309 return 0;
310
311 free_bufs:
312 if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
313 for_each_possible_cpu(i) {
314 struct lib_ring_buffer *buf = shmp(handle, chanb->buf[i].shmp);
315
316 if (!buf->backend.allocated)
317 continue;
318 lib_ring_buffer_free(buf, handle);
319 }
320 }
321 /* We only free the buffer data upon shm teardown */
322 end:
323 return -ENOMEM;
324 }
325
326 /**
327 * channel_backend_free - destroy the channel
328 * @chan: the channel
329 *
330 * Destroy all channel buffers and frees the channel.
331 */
332 void channel_backend_free(struct channel_backend *chanb,
333 struct shm_handle *handle)
334 {
335 const struct lib_ring_buffer_config *config = chanb->config;
336 unsigned int i;
337
338 if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
339 for_each_possible_cpu(i) {
340 struct lib_ring_buffer *buf = shmp(handle, chanb->buf[i].shmp);
341
342 if (!buf->backend.allocated)
343 continue;
344 lib_ring_buffer_free(buf, handle);
345 }
346 } else {
347 struct lib_ring_buffer *buf = shmp(handle, chanb->buf[0].shmp);
348
349 CHAN_WARN_ON(chanb, !buf->backend.allocated);
350 lib_ring_buffer_free(buf, handle);
351 }
352 /* We only free the buffer data upon shm teardown */
353 }
354
355 /**
356 * lib_ring_buffer_read - read data from ring_buffer_buffer.
357 * @bufb : buffer backend
358 * @offset : offset within the buffer
359 * @dest : destination address
360 * @len : length to copy to destination
361 *
362 * Should be protected by get_subbuf/put_subbuf.
363 * Returns the length copied.
364 */
365 size_t lib_ring_buffer_read(struct lib_ring_buffer_backend *bufb, size_t offset,
366 void *dest, size_t len, struct shm_handle *handle)
367 {
368 struct channel_backend *chanb = &shmp(handle, bufb->chan)->backend;
369 const struct lib_ring_buffer_config *config = chanb->config;
370 ssize_t orig_len;
371 struct lib_ring_buffer_backend_pages_shmp *rpages;
372 unsigned long sb_bindex, id;
373
374 orig_len = len;
375 offset &= chanb->buf_size - 1;
376
377 if (unlikely(!len))
378 return 0;
379 id = bufb->buf_rsb.id;
380 sb_bindex = subbuffer_id_get_index(config, id);
381 rpages = &shmp(handle, bufb->array)[sb_bindex];
382 /*
383 * Underlying layer should never ask for reads across
384 * subbuffers.
385 */
386 CHAN_WARN_ON(chanb, offset >= chanb->buf_size);
387 CHAN_WARN_ON(chanb, config->mode == RING_BUFFER_OVERWRITE
388 && subbuffer_id_is_noref(config, id));
389 memcpy(dest, shmp(handle, shmp(handle, rpages->shmp)->p) + (offset & ~(chanb->subbuf_size - 1)), len);
390 return orig_len;
391 }
392
393 /**
394 * lib_ring_buffer_read_cstr - read a C-style string from ring_buffer.
395 * @bufb : buffer backend
396 * @offset : offset within the buffer
397 * @dest : destination address
398 * @len : destination's length
399 *
400 * return string's length
401 * Should be protected by get_subbuf/put_subbuf.
402 */
403 int lib_ring_buffer_read_cstr(struct lib_ring_buffer_backend *bufb, size_t offset,
404 void *dest, size_t len, struct shm_handle *handle)
405 {
406 struct channel_backend *chanb = &shmp(handle, bufb->chan)->backend;
407 const struct lib_ring_buffer_config *config = chanb->config;
408 ssize_t string_len, orig_offset;
409 char *str;
410 struct lib_ring_buffer_backend_pages_shmp *rpages;
411 unsigned long sb_bindex, id;
412
413 offset &= chanb->buf_size - 1;
414 orig_offset = offset;
415 id = bufb->buf_rsb.id;
416 sb_bindex = subbuffer_id_get_index(config, id);
417 rpages = &shmp(handle, bufb->array)[sb_bindex];
418 /*
419 * Underlying layer should never ask for reads across
420 * subbuffers.
421 */
422 CHAN_WARN_ON(chanb, offset >= chanb->buf_size);
423 CHAN_WARN_ON(chanb, config->mode == RING_BUFFER_OVERWRITE
424 && subbuffer_id_is_noref(config, id));
425 str = (char *)shmp(handle, shmp(handle, rpages->shmp)->p) + (offset & ~(chanb->subbuf_size - 1));
426 string_len = strnlen(str, len);
427 if (dest && len) {
428 memcpy(dest, str, string_len);
429 ((char *)dest)[0] = 0;
430 }
431 return offset - orig_offset;
432 }
433
434 /**
435 * lib_ring_buffer_read_offset_address - get address of a buffer location
436 * @bufb : buffer backend
437 * @offset : offset within the buffer.
438 *
439 * Return the address where a given offset is located (for read).
440 * Should be used to get the current subbuffer header pointer. Given we know
441 * it's never on a page boundary, it's safe to write directly to this address,
442 * as long as the write is never bigger than a page size.
443 */
444 void *lib_ring_buffer_read_offset_address(struct lib_ring_buffer_backend *bufb,
445 size_t offset,
446 struct shm_handle *handle)
447 {
448 struct lib_ring_buffer_backend_pages_shmp *rpages;
449 struct channel_backend *chanb = &shmp(handle, bufb->chan)->backend;
450 const struct lib_ring_buffer_config *config = chanb->config;
451 unsigned long sb_bindex, id;
452
453 offset &= chanb->buf_size - 1;
454 id = bufb->buf_rsb.id;
455 sb_bindex = subbuffer_id_get_index(config, id);
456 rpages = &shmp(handle, bufb->array)[sb_bindex];
457 CHAN_WARN_ON(chanb, config->mode == RING_BUFFER_OVERWRITE
458 && subbuffer_id_is_noref(config, id));
459 return shmp(handle, shmp(handle, rpages->shmp)->p) + (offset & ~(chanb->subbuf_size - 1));
460 }
461
462 /**
463 * lib_ring_buffer_offset_address - get address of a location within the buffer
464 * @bufb : buffer backend
465 * @offset : offset within the buffer.
466 *
467 * Return the address where a given offset is located.
468 * Should be used to get the current subbuffer header pointer. Given we know
469 * it's always at the beginning of a page, it's safe to write directly to this
470 * address, as long as the write is never bigger than a page size.
471 */
472 void *lib_ring_buffer_offset_address(struct lib_ring_buffer_backend *bufb,
473 size_t offset,
474 struct shm_handle *handle)
475 {
476 size_t sbidx;
477 struct lib_ring_buffer_backend_pages_shmp *rpages;
478 struct channel_backend *chanb = &shmp(handle, bufb->chan)->backend;
479 const struct lib_ring_buffer_config *config = chanb->config;
480 unsigned long sb_bindex, id;
481
482 offset &= chanb->buf_size - 1;
483 sbidx = offset >> chanb->subbuf_size_order;
484 id = shmp(handle, bufb->buf_wsb)[sbidx].id;
485 sb_bindex = subbuffer_id_get_index(config, id);
486 rpages = &shmp(handle, bufb->array)[sb_bindex];
487 CHAN_WARN_ON(chanb, config->mode == RING_BUFFER_OVERWRITE
488 && subbuffer_id_is_noref(config, id));
489 return shmp(handle, shmp(handle, rpages->shmp)->p) + (offset & ~(chanb->subbuf_size - 1));
490 }
This page took 0.041601 seconds and 5 git commands to generate.