ring_buffer.c 53 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242
  1. /*
  2. * Generic ring buffer
  3. *
  4. * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
  5. */
  6. #include <linux/ring_buffer.h>
  7. #include <linux/spinlock.h>
  8. #include <linux/debugfs.h>
  9. #include <linux/uaccess.h>
  10. #include <linux/module.h>
  11. #include <linux/percpu.h>
  12. #include <linux/mutex.h>
  13. #include <linux/sched.h> /* used for sched_clock() (for now) */
  14. #include <linux/init.h>
  15. #include <linux/hash.h>
  16. #include <linux/list.h>
  17. #include <linux/fs.h>
  18. #include "trace.h"
  19. /* Global flag to disable all recording to ring buffers */
  20. static int ring_buffers_off __read_mostly;
  21. /**
  22. * tracing_on - enable all tracing buffers
  23. *
  24. * This function enables all tracing buffers that may have been
  25. * disabled with tracing_off.
  26. */
  27. void tracing_on(void)
  28. {
  29. ring_buffers_off = 0;
  30. }
  31. /**
  32. * tracing_off - turn off all tracing buffers
  33. *
  34. * This function stops all tracing buffers from recording data.
  35. * It does not disable any overhead the tracers themselves may
  36. * be causing. This function simply causes all recording to
  37. * the ring buffers to fail.
  38. */
  39. void tracing_off(void)
  40. {
  41. ring_buffers_off = 1;
  42. }
  43. #include "trace.h"
  44. /* Up this if you want to test the TIME_EXTENTS and normalization */
  45. #define DEBUG_SHIFT 0
  46. /* FIXME!!! */
  47. u64 ring_buffer_time_stamp(int cpu)
  48. {
  49. u64 time;
  50. preempt_disable_notrace();
  51. /* shift to debug/test normalization and TIME_EXTENTS */
  52. time = sched_clock() << DEBUG_SHIFT;
  53. preempt_enable_notrace();
  54. return time;
  55. }
  56. void ring_buffer_normalize_time_stamp(int cpu, u64 *ts)
  57. {
  58. /* Just stupid testing the normalize function and deltas */
  59. *ts >>= DEBUG_SHIFT;
  60. }
  61. #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event))
  62. #define RB_ALIGNMENT_SHIFT 2
  63. #define RB_ALIGNMENT (1 << RB_ALIGNMENT_SHIFT)
  64. #define RB_MAX_SMALL_DATA 28
  65. enum {
  66. RB_LEN_TIME_EXTEND = 8,
  67. RB_LEN_TIME_STAMP = 16,
  68. };
  69. /* inline for ring buffer fast paths */
  70. static inline unsigned
  71. rb_event_length(struct ring_buffer_event *event)
  72. {
  73. unsigned length;
  74. switch (event->type) {
  75. case RINGBUF_TYPE_PADDING:
  76. /* undefined */
  77. return -1;
  78. case RINGBUF_TYPE_TIME_EXTEND:
  79. return RB_LEN_TIME_EXTEND;
  80. case RINGBUF_TYPE_TIME_STAMP:
  81. return RB_LEN_TIME_STAMP;
  82. case RINGBUF_TYPE_DATA:
  83. if (event->len)
  84. length = event->len << RB_ALIGNMENT_SHIFT;
  85. else
  86. length = event->array[0];
  87. return length + RB_EVNT_HDR_SIZE;
  88. default:
  89. BUG();
  90. }
  91. /* not hit */
  92. return 0;
  93. }
  94. /**
  95. * ring_buffer_event_length - return the length of the event
  96. * @event: the event to get the length of
  97. */
  98. unsigned ring_buffer_event_length(struct ring_buffer_event *event)
  99. {
  100. return rb_event_length(event);
  101. }
  102. /* inline for ring buffer fast paths */
  103. static inline void *
  104. rb_event_data(struct ring_buffer_event *event)
  105. {
  106. BUG_ON(event->type != RINGBUF_TYPE_DATA);
  107. /* If length is in len field, then array[0] has the data */
  108. if (event->len)
  109. return (void *)&event->array[0];
  110. /* Otherwise length is in array[0] and array[1] has the data */
  111. return (void *)&event->array[1];
  112. }
  113. /**
  114. * ring_buffer_event_data - return the data of the event
  115. * @event: the event to get the data from
  116. */
  117. void *ring_buffer_event_data(struct ring_buffer_event *event)
  118. {
  119. return rb_event_data(event);
  120. }
  121. #define for_each_buffer_cpu(buffer, cpu) \
  122. for_each_cpu_mask(cpu, buffer->cpumask)
  123. #define TS_SHIFT 27
  124. #define TS_MASK ((1ULL << TS_SHIFT) - 1)
  125. #define TS_DELTA_TEST (~TS_MASK)
  126. /*
  127. * This hack stolen from mm/slob.c.
  128. * We can store per page timing information in the page frame of the page.
  129. * Thanks to Peter Zijlstra for suggesting this idea.
  130. */
  131. struct buffer_page {
  132. u64 time_stamp; /* page time stamp */
  133. local_t write; /* index for next write */
  134. local_t commit; /* write commited index */
  135. unsigned read; /* index for next read */
  136. struct list_head list; /* list of free pages */
  137. void *page; /* Actual data page */
  138. };
  139. /*
  140. * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
  141. * this issue out.
  142. */
  143. static inline void free_buffer_page(struct buffer_page *bpage)
  144. {
  145. if (bpage->page)
  146. free_page((unsigned long)bpage->page);
  147. kfree(bpage);
  148. }
  149. /*
  150. * We need to fit the time_stamp delta into 27 bits.
  151. */
  152. static inline int test_time_stamp(u64 delta)
  153. {
  154. if (delta & TS_DELTA_TEST)
  155. return 1;
  156. return 0;
  157. }
  158. #define BUF_PAGE_SIZE PAGE_SIZE
  159. /*
  160. * head_page == tail_page && head == tail then buffer is empty.
  161. */
  162. struct ring_buffer_per_cpu {
  163. int cpu;
  164. struct ring_buffer *buffer;
  165. spinlock_t reader_lock; /* serialize readers */
  166. raw_spinlock_t lock;
  167. struct lock_class_key lock_key;
  168. struct list_head pages;
  169. struct buffer_page *head_page; /* read from head */
  170. struct buffer_page *tail_page; /* write to tail */
  171. struct buffer_page *commit_page; /* commited pages */
  172. struct buffer_page *reader_page;
  173. unsigned long overrun;
  174. unsigned long entries;
  175. u64 write_stamp;
  176. u64 read_stamp;
  177. atomic_t record_disabled;
  178. };
  179. struct ring_buffer {
  180. unsigned long size;
  181. unsigned pages;
  182. unsigned flags;
  183. int cpus;
  184. cpumask_t cpumask;
  185. atomic_t record_disabled;
  186. struct mutex mutex;
  187. struct ring_buffer_per_cpu **buffers;
  188. };
  189. struct ring_buffer_iter {
  190. struct ring_buffer_per_cpu *cpu_buffer;
  191. unsigned long head;
  192. struct buffer_page *head_page;
  193. u64 read_stamp;
  194. };
  195. /* buffer may be either ring_buffer or ring_buffer_per_cpu */
  196. #define RB_WARN_ON(buffer, cond) \
  197. ({ \
  198. int _____ret = unlikely(cond); \
  199. if (_____ret) { \
  200. atomic_inc(&buffer->record_disabled); \
  201. WARN_ON(1); \
  202. } \
  203. _____ret; \
  204. })
  205. /**
  206. * check_pages - integrity check of buffer pages
  207. * @cpu_buffer: CPU buffer with pages to test
  208. *
  209. * As a safty measure we check to make sure the data pages have not
  210. * been corrupted.
  211. */
  212. static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
  213. {
  214. struct list_head *head = &cpu_buffer->pages;
  215. struct buffer_page *page, *tmp;
  216. if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
  217. return -1;
  218. if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
  219. return -1;
  220. list_for_each_entry_safe(page, tmp, head, list) {
  221. if (RB_WARN_ON(cpu_buffer,
  222. page->list.next->prev != &page->list))
  223. return -1;
  224. if (RB_WARN_ON(cpu_buffer,
  225. page->list.prev->next != &page->list))
  226. return -1;
  227. }
  228. return 0;
  229. }
  230. static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
  231. unsigned nr_pages)
  232. {
  233. struct list_head *head = &cpu_buffer->pages;
  234. struct buffer_page *page, *tmp;
  235. unsigned long addr;
  236. LIST_HEAD(pages);
  237. unsigned i;
  238. for (i = 0; i < nr_pages; i++) {
  239. page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
  240. GFP_KERNEL, cpu_to_node(cpu_buffer->cpu));
  241. if (!page)
  242. goto free_pages;
  243. list_add(&page->list, &pages);
  244. addr = __get_free_page(GFP_KERNEL);
  245. if (!addr)
  246. goto free_pages;
  247. page->page = (void *)addr;
  248. }
  249. list_splice(&pages, head);
  250. rb_check_pages(cpu_buffer);
  251. return 0;
  252. free_pages:
  253. list_for_each_entry_safe(page, tmp, &pages, list) {
  254. list_del_init(&page->list);
  255. free_buffer_page(page);
  256. }
  257. return -ENOMEM;
  258. }
  259. static struct ring_buffer_per_cpu *
  260. rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
  261. {
  262. struct ring_buffer_per_cpu *cpu_buffer;
  263. struct buffer_page *page;
  264. unsigned long addr;
  265. int ret;
  266. cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
  267. GFP_KERNEL, cpu_to_node(cpu));
  268. if (!cpu_buffer)
  269. return NULL;
  270. cpu_buffer->cpu = cpu;
  271. cpu_buffer->buffer = buffer;
  272. spin_lock_init(&cpu_buffer->reader_lock);
  273. cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
  274. INIT_LIST_HEAD(&cpu_buffer->pages);
  275. page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
  276. GFP_KERNEL, cpu_to_node(cpu));
  277. if (!page)
  278. goto fail_free_buffer;
  279. cpu_buffer->reader_page = page;
  280. addr = __get_free_page(GFP_KERNEL);
  281. if (!addr)
  282. goto fail_free_reader;
  283. page->page = (void *)addr;
  284. INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
  285. ret = rb_allocate_pages(cpu_buffer, buffer->pages);
  286. if (ret < 0)
  287. goto fail_free_reader;
  288. cpu_buffer->head_page
  289. = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
  290. cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
  291. return cpu_buffer;
  292. fail_free_reader:
  293. free_buffer_page(cpu_buffer->reader_page);
  294. fail_free_buffer:
  295. kfree(cpu_buffer);
  296. return NULL;
  297. }
  298. static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
  299. {
  300. struct list_head *head = &cpu_buffer->pages;
  301. struct buffer_page *page, *tmp;
  302. list_del_init(&cpu_buffer->reader_page->list);
  303. free_buffer_page(cpu_buffer->reader_page);
  304. list_for_each_entry_safe(page, tmp, head, list) {
  305. list_del_init(&page->list);
  306. free_buffer_page(page);
  307. }
  308. kfree(cpu_buffer);
  309. }
  310. /*
  311. * Causes compile errors if the struct buffer_page gets bigger
  312. * than the struct page.
  313. */
  314. extern int ring_buffer_page_too_big(void);
  315. /**
  316. * ring_buffer_alloc - allocate a new ring_buffer
  317. * @size: the size in bytes that is needed.
  318. * @flags: attributes to set for the ring buffer.
  319. *
  320. * Currently the only flag that is available is the RB_FL_OVERWRITE
  321. * flag. This flag means that the buffer will overwrite old data
  322. * when the buffer wraps. If this flag is not set, the buffer will
  323. * drop data when the tail hits the head.
  324. */
  325. struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
  326. {
  327. struct ring_buffer *buffer;
  328. int bsize;
  329. int cpu;
  330. /* Paranoid! Optimizes out when all is well */
  331. if (sizeof(struct buffer_page) > sizeof(struct page))
  332. ring_buffer_page_too_big();
  333. /* keep it in its own cache line */
  334. buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
  335. GFP_KERNEL);
  336. if (!buffer)
  337. return NULL;
  338. buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
  339. buffer->flags = flags;
  340. /* need at least two pages */
  341. if (buffer->pages == 1)
  342. buffer->pages++;
  343. buffer->cpumask = cpu_possible_map;
  344. buffer->cpus = nr_cpu_ids;
  345. bsize = sizeof(void *) * nr_cpu_ids;
  346. buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
  347. GFP_KERNEL);
  348. if (!buffer->buffers)
  349. goto fail_free_buffer;
  350. for_each_buffer_cpu(buffer, cpu) {
  351. buffer->buffers[cpu] =
  352. rb_allocate_cpu_buffer(buffer, cpu);
  353. if (!buffer->buffers[cpu])
  354. goto fail_free_buffers;
  355. }
  356. mutex_init(&buffer->mutex);
  357. return buffer;
  358. fail_free_buffers:
  359. for_each_buffer_cpu(buffer, cpu) {
  360. if (buffer->buffers[cpu])
  361. rb_free_cpu_buffer(buffer->buffers[cpu]);
  362. }
  363. kfree(buffer->buffers);
  364. fail_free_buffer:
  365. kfree(buffer);
  366. return NULL;
  367. }
  368. /**
  369. * ring_buffer_free - free a ring buffer.
  370. * @buffer: the buffer to free.
  371. */
  372. void
  373. ring_buffer_free(struct ring_buffer *buffer)
  374. {
  375. int cpu;
  376. for_each_buffer_cpu(buffer, cpu)
  377. rb_free_cpu_buffer(buffer->buffers[cpu]);
  378. kfree(buffer);
  379. }
  380. static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
  381. static void
  382. rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
  383. {
  384. struct buffer_page *page;
  385. struct list_head *p;
  386. unsigned i;
  387. atomic_inc(&cpu_buffer->record_disabled);
  388. synchronize_sched();
  389. for (i = 0; i < nr_pages; i++) {
  390. if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
  391. return;
  392. p = cpu_buffer->pages.next;
  393. page = list_entry(p, struct buffer_page, list);
  394. list_del_init(&page->list);
  395. free_buffer_page(page);
  396. }
  397. if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
  398. return;
  399. rb_reset_cpu(cpu_buffer);
  400. rb_check_pages(cpu_buffer);
  401. atomic_dec(&cpu_buffer->record_disabled);
  402. }
  403. static void
  404. rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer,
  405. struct list_head *pages, unsigned nr_pages)
  406. {
  407. struct buffer_page *page;
  408. struct list_head *p;
  409. unsigned i;
  410. atomic_inc(&cpu_buffer->record_disabled);
  411. synchronize_sched();
  412. for (i = 0; i < nr_pages; i++) {
  413. if (RB_WARN_ON(cpu_buffer, list_empty(pages)))
  414. return;
  415. p = pages->next;
  416. page = list_entry(p, struct buffer_page, list);
  417. list_del_init(&page->list);
  418. list_add_tail(&page->list, &cpu_buffer->pages);
  419. }
  420. rb_reset_cpu(cpu_buffer);
  421. rb_check_pages(cpu_buffer);
  422. atomic_dec(&cpu_buffer->record_disabled);
  423. }
  424. /**
  425. * ring_buffer_resize - resize the ring buffer
  426. * @buffer: the buffer to resize.
  427. * @size: the new size.
  428. *
  429. * The tracer is responsible for making sure that the buffer is
  430. * not being used while changing the size.
  431. * Note: We may be able to change the above requirement by using
  432. * RCU synchronizations.
  433. *
  434. * Minimum size is 2 * BUF_PAGE_SIZE.
  435. *
  436. * Returns -1 on failure.
  437. */
  438. int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
  439. {
  440. struct ring_buffer_per_cpu *cpu_buffer;
  441. unsigned nr_pages, rm_pages, new_pages;
  442. struct buffer_page *page, *tmp;
  443. unsigned long buffer_size;
  444. unsigned long addr;
  445. LIST_HEAD(pages);
  446. int i, cpu;
  447. /*
  448. * Always succeed at resizing a non-existent buffer:
  449. */
  450. if (!buffer)
  451. return size;
  452. size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
  453. size *= BUF_PAGE_SIZE;
  454. buffer_size = buffer->pages * BUF_PAGE_SIZE;
  455. /* we need a minimum of two pages */
  456. if (size < BUF_PAGE_SIZE * 2)
  457. size = BUF_PAGE_SIZE * 2;
  458. if (size == buffer_size)
  459. return size;
  460. mutex_lock(&buffer->mutex);
  461. nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
  462. if (size < buffer_size) {
  463. /* easy case, just free pages */
  464. if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) {
  465. mutex_unlock(&buffer->mutex);
  466. return -1;
  467. }
  468. rm_pages = buffer->pages - nr_pages;
  469. for_each_buffer_cpu(buffer, cpu) {
  470. cpu_buffer = buffer->buffers[cpu];
  471. rb_remove_pages(cpu_buffer, rm_pages);
  472. }
  473. goto out;
  474. }
  475. /*
  476. * This is a bit more difficult. We only want to add pages
  477. * when we can allocate enough for all CPUs. We do this
  478. * by allocating all the pages and storing them on a local
  479. * link list. If we succeed in our allocation, then we
  480. * add these pages to the cpu_buffers. Otherwise we just free
  481. * them all and return -ENOMEM;
  482. */
  483. if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) {
  484. mutex_unlock(&buffer->mutex);
  485. return -1;
  486. }
  487. new_pages = nr_pages - buffer->pages;
  488. for_each_buffer_cpu(buffer, cpu) {
  489. for (i = 0; i < new_pages; i++) {
  490. page = kzalloc_node(ALIGN(sizeof(*page),
  491. cache_line_size()),
  492. GFP_KERNEL, cpu_to_node(cpu));
  493. if (!page)
  494. goto free_pages;
  495. list_add(&page->list, &pages);
  496. addr = __get_free_page(GFP_KERNEL);
  497. if (!addr)
  498. goto free_pages;
  499. page->page = (void *)addr;
  500. }
  501. }
  502. for_each_buffer_cpu(buffer, cpu) {
  503. cpu_buffer = buffer->buffers[cpu];
  504. rb_insert_pages(cpu_buffer, &pages, new_pages);
  505. }
  506. if (RB_WARN_ON(buffer, !list_empty(&pages))) {
  507. mutex_unlock(&buffer->mutex);
  508. return -1;
  509. }
  510. out:
  511. buffer->pages = nr_pages;
  512. mutex_unlock(&buffer->mutex);
  513. return size;
  514. free_pages:
  515. list_for_each_entry_safe(page, tmp, &pages, list) {
  516. list_del_init(&page->list);
  517. free_buffer_page(page);
  518. }
  519. mutex_unlock(&buffer->mutex);
  520. return -ENOMEM;
  521. }
  522. static inline int rb_null_event(struct ring_buffer_event *event)
  523. {
  524. return event->type == RINGBUF_TYPE_PADDING;
  525. }
  526. static inline void *__rb_page_index(struct buffer_page *page, unsigned index)
  527. {
  528. return page->page + index;
  529. }
  530. static inline struct ring_buffer_event *
  531. rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
  532. {
  533. return __rb_page_index(cpu_buffer->reader_page,
  534. cpu_buffer->reader_page->read);
  535. }
  536. static inline struct ring_buffer_event *
  537. rb_head_event(struct ring_buffer_per_cpu *cpu_buffer)
  538. {
  539. return __rb_page_index(cpu_buffer->head_page,
  540. cpu_buffer->head_page->read);
  541. }
  542. static inline struct ring_buffer_event *
  543. rb_iter_head_event(struct ring_buffer_iter *iter)
  544. {
  545. return __rb_page_index(iter->head_page, iter->head);
  546. }
  547. static inline unsigned rb_page_write(struct buffer_page *bpage)
  548. {
  549. return local_read(&bpage->write);
  550. }
  551. static inline unsigned rb_page_commit(struct buffer_page *bpage)
  552. {
  553. return local_read(&bpage->commit);
  554. }
  555. /* Size is determined by what has been commited */
  556. static inline unsigned rb_page_size(struct buffer_page *bpage)
  557. {
  558. return rb_page_commit(bpage);
  559. }
  560. static inline unsigned
  561. rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
  562. {
  563. return rb_page_commit(cpu_buffer->commit_page);
  564. }
  565. static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
  566. {
  567. return rb_page_commit(cpu_buffer->head_page);
  568. }
  569. /*
  570. * When the tail hits the head and the buffer is in overwrite mode,
  571. * the head jumps to the next page and all content on the previous
  572. * page is discarded. But before doing so, we update the overrun
  573. * variable of the buffer.
  574. */
  575. static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer)
  576. {
  577. struct ring_buffer_event *event;
  578. unsigned long head;
  579. for (head = 0; head < rb_head_size(cpu_buffer);
  580. head += rb_event_length(event)) {
  581. event = __rb_page_index(cpu_buffer->head_page, head);
  582. if (RB_WARN_ON(cpu_buffer, rb_null_event(event)))
  583. return;
  584. /* Only count data entries */
  585. if (event->type != RINGBUF_TYPE_DATA)
  586. continue;
  587. cpu_buffer->overrun++;
  588. cpu_buffer->entries--;
  589. }
  590. }
  591. static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
  592. struct buffer_page **page)
  593. {
  594. struct list_head *p = (*page)->list.next;
  595. if (p == &cpu_buffer->pages)
  596. p = p->next;
  597. *page = list_entry(p, struct buffer_page, list);
  598. }
  599. static inline unsigned
  600. rb_event_index(struct ring_buffer_event *event)
  601. {
  602. unsigned long addr = (unsigned long)event;
  603. return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE);
  604. }
  605. static inline int
  606. rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
  607. struct ring_buffer_event *event)
  608. {
  609. unsigned long addr = (unsigned long)event;
  610. unsigned long index;
  611. index = rb_event_index(event);
  612. addr &= PAGE_MASK;
  613. return cpu_buffer->commit_page->page == (void *)addr &&
  614. rb_commit_index(cpu_buffer) == index;
  615. }
  616. static inline void
  617. rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer,
  618. struct ring_buffer_event *event)
  619. {
  620. unsigned long addr = (unsigned long)event;
  621. unsigned long index;
  622. index = rb_event_index(event);
  623. addr &= PAGE_MASK;
  624. while (cpu_buffer->commit_page->page != (void *)addr) {
  625. if (RB_WARN_ON(cpu_buffer,
  626. cpu_buffer->commit_page == cpu_buffer->tail_page))
  627. return;
  628. cpu_buffer->commit_page->commit =
  629. cpu_buffer->commit_page->write;
  630. rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
  631. cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp;
  632. }
  633. /* Now set the commit to the event's index */
  634. local_set(&cpu_buffer->commit_page->commit, index);
  635. }
  636. static inline void
  637. rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
  638. {
  639. /*
  640. * We only race with interrupts and NMIs on this CPU.
  641. * If we own the commit event, then we can commit
  642. * all others that interrupted us, since the interruptions
  643. * are in stack format (they finish before they come
  644. * back to us). This allows us to do a simple loop to
  645. * assign the commit to the tail.
  646. */
  647. while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
  648. cpu_buffer->commit_page->commit =
  649. cpu_buffer->commit_page->write;
  650. rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
  651. cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp;
  652. /* add barrier to keep gcc from optimizing too much */
  653. barrier();
  654. }
  655. while (rb_commit_index(cpu_buffer) !=
  656. rb_page_write(cpu_buffer->commit_page)) {
  657. cpu_buffer->commit_page->commit =
  658. cpu_buffer->commit_page->write;
  659. barrier();
  660. }
  661. }
  662. static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
  663. {
  664. cpu_buffer->read_stamp = cpu_buffer->reader_page->time_stamp;
  665. cpu_buffer->reader_page->read = 0;
  666. }
  667. static inline void rb_inc_iter(struct ring_buffer_iter *iter)
  668. {
  669. struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
  670. /*
  671. * The iterator could be on the reader page (it starts there).
  672. * But the head could have moved, since the reader was
  673. * found. Check for this case and assign the iterator
  674. * to the head page instead of next.
  675. */
  676. if (iter->head_page == cpu_buffer->reader_page)
  677. iter->head_page = cpu_buffer->head_page;
  678. else
  679. rb_inc_page(cpu_buffer, &iter->head_page);
  680. iter->read_stamp = iter->head_page->time_stamp;
  681. iter->head = 0;
  682. }
  683. /**
  684. * ring_buffer_update_event - update event type and data
  685. * @event: the even to update
  686. * @type: the type of event
  687. * @length: the size of the event field in the ring buffer
  688. *
  689. * Update the type and data fields of the event. The length
  690. * is the actual size that is written to the ring buffer,
  691. * and with this, we can determine what to place into the
  692. * data field.
  693. */
  694. static inline void
  695. rb_update_event(struct ring_buffer_event *event,
  696. unsigned type, unsigned length)
  697. {
  698. event->type = type;
  699. switch (type) {
  700. case RINGBUF_TYPE_PADDING:
  701. break;
  702. case RINGBUF_TYPE_TIME_EXTEND:
  703. event->len =
  704. (RB_LEN_TIME_EXTEND + (RB_ALIGNMENT-1))
  705. >> RB_ALIGNMENT_SHIFT;
  706. break;
  707. case RINGBUF_TYPE_TIME_STAMP:
  708. event->len =
  709. (RB_LEN_TIME_STAMP + (RB_ALIGNMENT-1))
  710. >> RB_ALIGNMENT_SHIFT;
  711. break;
  712. case RINGBUF_TYPE_DATA:
  713. length -= RB_EVNT_HDR_SIZE;
  714. if (length > RB_MAX_SMALL_DATA) {
  715. event->len = 0;
  716. event->array[0] = length;
  717. } else
  718. event->len =
  719. (length + (RB_ALIGNMENT-1))
  720. >> RB_ALIGNMENT_SHIFT;
  721. break;
  722. default:
  723. BUG();
  724. }
  725. }
  726. static inline unsigned rb_calculate_event_length(unsigned length)
  727. {
  728. struct ring_buffer_event event; /* Used only for sizeof array */
  729. /* zero length can cause confusions */
  730. if (!length)
  731. length = 1;
  732. if (length > RB_MAX_SMALL_DATA)
  733. length += sizeof(event.array[0]);
  734. length += RB_EVNT_HDR_SIZE;
  735. length = ALIGN(length, RB_ALIGNMENT);
  736. return length;
  737. }
  738. static struct ring_buffer_event *
  739. __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
  740. unsigned type, unsigned long length, u64 *ts)
  741. {
  742. struct buffer_page *tail_page, *head_page, *reader_page;
  743. unsigned long tail, write;
  744. struct ring_buffer *buffer = cpu_buffer->buffer;
  745. struct ring_buffer_event *event;
  746. unsigned long flags;
  747. tail_page = cpu_buffer->tail_page;
  748. write = local_add_return(length, &tail_page->write);
  749. tail = write - length;
  750. /* See if we shot pass the end of this buffer page */
  751. if (write > BUF_PAGE_SIZE) {
  752. struct buffer_page *next_page = tail_page;
  753. local_irq_save(flags);
  754. __raw_spin_lock(&cpu_buffer->lock);
  755. rb_inc_page(cpu_buffer, &next_page);
  756. head_page = cpu_buffer->head_page;
  757. reader_page = cpu_buffer->reader_page;
  758. /* we grabbed the lock before incrementing */
  759. if (RB_WARN_ON(cpu_buffer, next_page == reader_page))
  760. goto out_unlock;
  761. /*
  762. * If for some reason, we had an interrupt storm that made
  763. * it all the way around the buffer, bail, and warn
  764. * about it.
  765. */
  766. if (unlikely(next_page == cpu_buffer->commit_page)) {
  767. WARN_ON_ONCE(1);
  768. goto out_unlock;
  769. }
  770. if (next_page == head_page) {
  771. if (!(buffer->flags & RB_FL_OVERWRITE)) {
  772. /* reset write */
  773. if (tail <= BUF_PAGE_SIZE)
  774. local_set(&tail_page->write, tail);
  775. goto out_unlock;
  776. }
  777. /* tail_page has not moved yet? */
  778. if (tail_page == cpu_buffer->tail_page) {
  779. /* count overflows */
  780. rb_update_overflow(cpu_buffer);
  781. rb_inc_page(cpu_buffer, &head_page);
  782. cpu_buffer->head_page = head_page;
  783. cpu_buffer->head_page->read = 0;
  784. }
  785. }
  786. /*
  787. * If the tail page is still the same as what we think
  788. * it is, then it is up to us to update the tail
  789. * pointer.
  790. */
  791. if (tail_page == cpu_buffer->tail_page) {
  792. local_set(&next_page->write, 0);
  793. local_set(&next_page->commit, 0);
  794. cpu_buffer->tail_page = next_page;
  795. /* reread the time stamp */
  796. *ts = ring_buffer_time_stamp(cpu_buffer->cpu);
  797. cpu_buffer->tail_page->time_stamp = *ts;
  798. }
  799. /*
  800. * The actual tail page has moved forward.
  801. */
  802. if (tail < BUF_PAGE_SIZE) {
  803. /* Mark the rest of the page with padding */
  804. event = __rb_page_index(tail_page, tail);
  805. event->type = RINGBUF_TYPE_PADDING;
  806. }
  807. if (tail <= BUF_PAGE_SIZE)
  808. /* Set the write back to the previous setting */
  809. local_set(&tail_page->write, tail);
  810. /*
  811. * If this was a commit entry that failed,
  812. * increment that too
  813. */
  814. if (tail_page == cpu_buffer->commit_page &&
  815. tail == rb_commit_index(cpu_buffer)) {
  816. rb_set_commit_to_write(cpu_buffer);
  817. }
  818. __raw_spin_unlock(&cpu_buffer->lock);
  819. local_irq_restore(flags);
  820. /* fail and let the caller try again */
  821. return ERR_PTR(-EAGAIN);
  822. }
  823. /* We reserved something on the buffer */
  824. if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE))
  825. return NULL;
  826. event = __rb_page_index(tail_page, tail);
  827. rb_update_event(event, type, length);
  828. /*
  829. * If this is a commit and the tail is zero, then update
  830. * this page's time stamp.
  831. */
  832. if (!tail && rb_is_commit(cpu_buffer, event))
  833. cpu_buffer->commit_page->time_stamp = *ts;
  834. return event;
  835. out_unlock:
  836. __raw_spin_unlock(&cpu_buffer->lock);
  837. local_irq_restore(flags);
  838. return NULL;
  839. }
  840. static int
  841. rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer,
  842. u64 *ts, u64 *delta)
  843. {
  844. struct ring_buffer_event *event;
  845. static int once;
  846. int ret;
  847. if (unlikely(*delta > (1ULL << 59) && !once++)) {
  848. printk(KERN_WARNING "Delta way too big! %llu"
  849. " ts=%llu write stamp = %llu\n",
  850. (unsigned long long)*delta,
  851. (unsigned long long)*ts,
  852. (unsigned long long)cpu_buffer->write_stamp);
  853. WARN_ON(1);
  854. }
  855. /*
  856. * The delta is too big, we to add a
  857. * new timestamp.
  858. */
  859. event = __rb_reserve_next(cpu_buffer,
  860. RINGBUF_TYPE_TIME_EXTEND,
  861. RB_LEN_TIME_EXTEND,
  862. ts);
  863. if (!event)
  864. return -EBUSY;
  865. if (PTR_ERR(event) == -EAGAIN)
  866. return -EAGAIN;
  867. /* Only a commited time event can update the write stamp */
  868. if (rb_is_commit(cpu_buffer, event)) {
  869. /*
  870. * If this is the first on the page, then we need to
  871. * update the page itself, and just put in a zero.
  872. */
  873. if (rb_event_index(event)) {
  874. event->time_delta = *delta & TS_MASK;
  875. event->array[0] = *delta >> TS_SHIFT;
  876. } else {
  877. cpu_buffer->commit_page->time_stamp = *ts;
  878. event->time_delta = 0;
  879. event->array[0] = 0;
  880. }
  881. cpu_buffer->write_stamp = *ts;
  882. /* let the caller know this was the commit */
  883. ret = 1;
  884. } else {
  885. /* Darn, this is just wasted space */
  886. event->time_delta = 0;
  887. event->array[0] = 0;
  888. ret = 0;
  889. }
  890. *delta = 0;
  891. return ret;
  892. }
  893. static struct ring_buffer_event *
  894. rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer,
  895. unsigned type, unsigned long length)
  896. {
  897. struct ring_buffer_event *event;
  898. u64 ts, delta;
  899. int commit = 0;
  900. int nr_loops = 0;
  901. again:
  902. /*
  903. * We allow for interrupts to reenter here and do a trace.
  904. * If one does, it will cause this original code to loop
  905. * back here. Even with heavy interrupts happening, this
  906. * should only happen a few times in a row. If this happens
  907. * 1000 times in a row, there must be either an interrupt
  908. * storm or we have something buggy.
  909. * Bail!
  910. */
  911. if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
  912. return NULL;
  913. ts = ring_buffer_time_stamp(cpu_buffer->cpu);
  914. /*
  915. * Only the first commit can update the timestamp.
  916. * Yes there is a race here. If an interrupt comes in
  917. * just after the conditional and it traces too, then it
  918. * will also check the deltas. More than one timestamp may
  919. * also be made. But only the entry that did the actual
  920. * commit will be something other than zero.
  921. */
  922. if (cpu_buffer->tail_page == cpu_buffer->commit_page &&
  923. rb_page_write(cpu_buffer->tail_page) ==
  924. rb_commit_index(cpu_buffer)) {
  925. delta = ts - cpu_buffer->write_stamp;
  926. /* make sure this delta is calculated here */
  927. barrier();
  928. /* Did the write stamp get updated already? */
  929. if (unlikely(ts < cpu_buffer->write_stamp))
  930. delta = 0;
  931. if (test_time_stamp(delta)) {
  932. commit = rb_add_time_stamp(cpu_buffer, &ts, &delta);
  933. if (commit == -EBUSY)
  934. return NULL;
  935. if (commit == -EAGAIN)
  936. goto again;
  937. RB_WARN_ON(cpu_buffer, commit < 0);
  938. }
  939. } else
  940. /* Non commits have zero deltas */
  941. delta = 0;
  942. event = __rb_reserve_next(cpu_buffer, type, length, &ts);
  943. if (PTR_ERR(event) == -EAGAIN)
  944. goto again;
  945. if (!event) {
  946. if (unlikely(commit))
  947. /*
  948. * Ouch! We needed a timestamp and it was commited. But
  949. * we didn't get our event reserved.
  950. */
  951. rb_set_commit_to_write(cpu_buffer);
  952. return NULL;
  953. }
  954. /*
  955. * If the timestamp was commited, make the commit our entry
  956. * now so that we will update it when needed.
  957. */
  958. if (commit)
  959. rb_set_commit_event(cpu_buffer, event);
  960. else if (!rb_is_commit(cpu_buffer, event))
  961. delta = 0;
  962. event->time_delta = delta;
  963. return event;
  964. }
  965. static DEFINE_PER_CPU(int, rb_need_resched);
  966. /**
  967. * ring_buffer_lock_reserve - reserve a part of the buffer
  968. * @buffer: the ring buffer to reserve from
  969. * @length: the length of the data to reserve (excluding event header)
  970. * @flags: a pointer to save the interrupt flags
  971. *
  972. * Returns a reseverd event on the ring buffer to copy directly to.
  973. * The user of this interface will need to get the body to write into
  974. * and can use the ring_buffer_event_data() interface.
  975. *
  976. * The length is the length of the data needed, not the event length
  977. * which also includes the event header.
  978. *
  979. * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
  980. * If NULL is returned, then nothing has been allocated or locked.
  981. */
  982. struct ring_buffer_event *
  983. ring_buffer_lock_reserve(struct ring_buffer *buffer,
  984. unsigned long length,
  985. unsigned long *flags)
  986. {
  987. struct ring_buffer_per_cpu *cpu_buffer;
  988. struct ring_buffer_event *event;
  989. int cpu, resched;
  990. if (ring_buffers_off)
  991. return NULL;
  992. if (atomic_read(&buffer->record_disabled))
  993. return NULL;
  994. /* If we are tracing schedule, we don't want to recurse */
  995. resched = ftrace_preempt_disable();
  996. cpu = raw_smp_processor_id();
  997. if (!cpu_isset(cpu, buffer->cpumask))
  998. goto out;
  999. cpu_buffer = buffer->buffers[cpu];
  1000. if (atomic_read(&cpu_buffer->record_disabled))
  1001. goto out;
  1002. length = rb_calculate_event_length(length);
  1003. if (length > BUF_PAGE_SIZE)
  1004. goto out;
  1005. event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length);
  1006. if (!event)
  1007. goto out;
  1008. /*
  1009. * Need to store resched state on this cpu.
  1010. * Only the first needs to.
  1011. */
  1012. if (preempt_count() == 1)
  1013. per_cpu(rb_need_resched, cpu) = resched;
  1014. return event;
  1015. out:
  1016. ftrace_preempt_enable(resched);
  1017. return NULL;
  1018. }
  1019. static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
  1020. struct ring_buffer_event *event)
  1021. {
  1022. cpu_buffer->entries++;
  1023. /* Only process further if we own the commit */
  1024. if (!rb_is_commit(cpu_buffer, event))
  1025. return;
  1026. cpu_buffer->write_stamp += event->time_delta;
  1027. rb_set_commit_to_write(cpu_buffer);
  1028. }
  1029. /**
  1030. * ring_buffer_unlock_commit - commit a reserved
  1031. * @buffer: The buffer to commit to
  1032. * @event: The event pointer to commit.
  1033. * @flags: the interrupt flags received from ring_buffer_lock_reserve.
  1034. *
  1035. * This commits the data to the ring buffer, and releases any locks held.
  1036. *
  1037. * Must be paired with ring_buffer_lock_reserve.
  1038. */
  1039. int ring_buffer_unlock_commit(struct ring_buffer *buffer,
  1040. struct ring_buffer_event *event,
  1041. unsigned long flags)
  1042. {
  1043. struct ring_buffer_per_cpu *cpu_buffer;
  1044. int cpu = raw_smp_processor_id();
  1045. cpu_buffer = buffer->buffers[cpu];
  1046. rb_commit(cpu_buffer, event);
  1047. /*
  1048. * Only the last preempt count needs to restore preemption.
  1049. */
  1050. if (preempt_count() == 1)
  1051. ftrace_preempt_enable(per_cpu(rb_need_resched, cpu));
  1052. else
  1053. preempt_enable_no_resched_notrace();
  1054. return 0;
  1055. }
  1056. /**
  1057. * ring_buffer_write - write data to the buffer without reserving
  1058. * @buffer: The ring buffer to write to.
  1059. * @length: The length of the data being written (excluding the event header)
  1060. * @data: The data to write to the buffer.
  1061. *
  1062. * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
  1063. * one function. If you already have the data to write to the buffer, it
  1064. * may be easier to simply call this function.
  1065. *
  1066. * Note, like ring_buffer_lock_reserve, the length is the length of the data
  1067. * and not the length of the event which would hold the header.
  1068. */
  1069. int ring_buffer_write(struct ring_buffer *buffer,
  1070. unsigned long length,
  1071. void *data)
  1072. {
  1073. struct ring_buffer_per_cpu *cpu_buffer;
  1074. struct ring_buffer_event *event;
  1075. unsigned long event_length;
  1076. void *body;
  1077. int ret = -EBUSY;
  1078. int cpu, resched;
  1079. if (ring_buffers_off)
  1080. return -EBUSY;
  1081. if (atomic_read(&buffer->record_disabled))
  1082. return -EBUSY;
  1083. resched = ftrace_preempt_disable();
  1084. cpu = raw_smp_processor_id();
  1085. if (!cpu_isset(cpu, buffer->cpumask))
  1086. goto out;
  1087. cpu_buffer = buffer->buffers[cpu];
  1088. if (atomic_read(&cpu_buffer->record_disabled))
  1089. goto out;
  1090. event_length = rb_calculate_event_length(length);
  1091. event = rb_reserve_next_event(cpu_buffer,
  1092. RINGBUF_TYPE_DATA, event_length);
  1093. if (!event)
  1094. goto out;
  1095. body = rb_event_data(event);
  1096. memcpy(body, data, length);
  1097. rb_commit(cpu_buffer, event);
  1098. ret = 0;
  1099. out:
  1100. ftrace_preempt_enable(resched);
  1101. return ret;
  1102. }
  1103. static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
  1104. {
  1105. struct buffer_page *reader = cpu_buffer->reader_page;
  1106. struct buffer_page *head = cpu_buffer->head_page;
  1107. struct buffer_page *commit = cpu_buffer->commit_page;
  1108. return reader->read == rb_page_commit(reader) &&
  1109. (commit == reader ||
  1110. (commit == head &&
  1111. head->read == rb_page_commit(commit)));
  1112. }
  1113. /**
  1114. * ring_buffer_record_disable - stop all writes into the buffer
  1115. * @buffer: The ring buffer to stop writes to.
  1116. *
  1117. * This prevents all writes to the buffer. Any attempt to write
  1118. * to the buffer after this will fail and return NULL.
  1119. *
  1120. * The caller should call synchronize_sched() after this.
  1121. */
  1122. void ring_buffer_record_disable(struct ring_buffer *buffer)
  1123. {
  1124. atomic_inc(&buffer->record_disabled);
  1125. }
  1126. /**
  1127. * ring_buffer_record_enable - enable writes to the buffer
  1128. * @buffer: The ring buffer to enable writes
  1129. *
  1130. * Note, multiple disables will need the same number of enables
  1131. * to truely enable the writing (much like preempt_disable).
  1132. */
  1133. void ring_buffer_record_enable(struct ring_buffer *buffer)
  1134. {
  1135. atomic_dec(&buffer->record_disabled);
  1136. }
  1137. /**
  1138. * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
  1139. * @buffer: The ring buffer to stop writes to.
  1140. * @cpu: The CPU buffer to stop
  1141. *
  1142. * This prevents all writes to the buffer. Any attempt to write
  1143. * to the buffer after this will fail and return NULL.
  1144. *
  1145. * The caller should call synchronize_sched() after this.
  1146. */
  1147. void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
  1148. {
  1149. struct ring_buffer_per_cpu *cpu_buffer;
  1150. if (!cpu_isset(cpu, buffer->cpumask))
  1151. return;
  1152. cpu_buffer = buffer->buffers[cpu];
  1153. atomic_inc(&cpu_buffer->record_disabled);
  1154. }
  1155. /**
  1156. * ring_buffer_record_enable_cpu - enable writes to the buffer
  1157. * @buffer: The ring buffer to enable writes
  1158. * @cpu: The CPU to enable.
  1159. *
  1160. * Note, multiple disables will need the same number of enables
  1161. * to truely enable the writing (much like preempt_disable).
  1162. */
  1163. void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
  1164. {
  1165. struct ring_buffer_per_cpu *cpu_buffer;
  1166. if (!cpu_isset(cpu, buffer->cpumask))
  1167. return;
  1168. cpu_buffer = buffer->buffers[cpu];
  1169. atomic_dec(&cpu_buffer->record_disabled);
  1170. }
  1171. /**
  1172. * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
  1173. * @buffer: The ring buffer
  1174. * @cpu: The per CPU buffer to get the entries from.
  1175. */
  1176. unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
  1177. {
  1178. struct ring_buffer_per_cpu *cpu_buffer;
  1179. if (!cpu_isset(cpu, buffer->cpumask))
  1180. return 0;
  1181. cpu_buffer = buffer->buffers[cpu];
  1182. return cpu_buffer->entries;
  1183. }
  1184. /**
  1185. * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
  1186. * @buffer: The ring buffer
  1187. * @cpu: The per CPU buffer to get the number of overruns from
  1188. */
  1189. unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
  1190. {
  1191. struct ring_buffer_per_cpu *cpu_buffer;
  1192. if (!cpu_isset(cpu, buffer->cpumask))
  1193. return 0;
  1194. cpu_buffer = buffer->buffers[cpu];
  1195. return cpu_buffer->overrun;
  1196. }
  1197. /**
  1198. * ring_buffer_entries - get the number of entries in a buffer
  1199. * @buffer: The ring buffer
  1200. *
  1201. * Returns the total number of entries in the ring buffer
  1202. * (all CPU entries)
  1203. */
  1204. unsigned long ring_buffer_entries(struct ring_buffer *buffer)
  1205. {
  1206. struct ring_buffer_per_cpu *cpu_buffer;
  1207. unsigned long entries = 0;
  1208. int cpu;
  1209. /* if you care about this being correct, lock the buffer */
  1210. for_each_buffer_cpu(buffer, cpu) {
  1211. cpu_buffer = buffer->buffers[cpu];
  1212. entries += cpu_buffer->entries;
  1213. }
  1214. return entries;
  1215. }
  1216. /**
  1217. * ring_buffer_overrun_cpu - get the number of overruns in buffer
  1218. * @buffer: The ring buffer
  1219. *
  1220. * Returns the total number of overruns in the ring buffer
  1221. * (all CPU entries)
  1222. */
  1223. unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
  1224. {
  1225. struct ring_buffer_per_cpu *cpu_buffer;
  1226. unsigned long overruns = 0;
  1227. int cpu;
  1228. /* if you care about this being correct, lock the buffer */
  1229. for_each_buffer_cpu(buffer, cpu) {
  1230. cpu_buffer = buffer->buffers[cpu];
  1231. overruns += cpu_buffer->overrun;
  1232. }
  1233. return overruns;
  1234. }
  1235. static void rb_iter_reset(struct ring_buffer_iter *iter)
  1236. {
  1237. struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
  1238. /* Iterator usage is expected to have record disabled */
  1239. if (list_empty(&cpu_buffer->reader_page->list)) {
  1240. iter->head_page = cpu_buffer->head_page;
  1241. iter->head = cpu_buffer->head_page->read;
  1242. } else {
  1243. iter->head_page = cpu_buffer->reader_page;
  1244. iter->head = cpu_buffer->reader_page->read;
  1245. }
  1246. if (iter->head)
  1247. iter->read_stamp = cpu_buffer->read_stamp;
  1248. else
  1249. iter->read_stamp = iter->head_page->time_stamp;
  1250. }
  1251. /**
  1252. * ring_buffer_iter_reset - reset an iterator
  1253. * @iter: The iterator to reset
  1254. *
  1255. * Resets the iterator, so that it will start from the beginning
  1256. * again.
  1257. */
  1258. void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
  1259. {
  1260. struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
  1261. unsigned long flags;
  1262. spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
  1263. rb_iter_reset(iter);
  1264. spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
  1265. }
  1266. /**
  1267. * ring_buffer_iter_empty - check if an iterator has no more to read
  1268. * @iter: The iterator to check
  1269. */
  1270. int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
  1271. {
  1272. struct ring_buffer_per_cpu *cpu_buffer;
  1273. cpu_buffer = iter->cpu_buffer;
  1274. return iter->head_page == cpu_buffer->commit_page &&
  1275. iter->head == rb_commit_index(cpu_buffer);
  1276. }
  1277. static void
  1278. rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
  1279. struct ring_buffer_event *event)
  1280. {
  1281. u64 delta;
  1282. switch (event->type) {
  1283. case RINGBUF_TYPE_PADDING:
  1284. return;
  1285. case RINGBUF_TYPE_TIME_EXTEND:
  1286. delta = event->array[0];
  1287. delta <<= TS_SHIFT;
  1288. delta += event->time_delta;
  1289. cpu_buffer->read_stamp += delta;
  1290. return;
  1291. case RINGBUF_TYPE_TIME_STAMP:
  1292. /* FIXME: not implemented */
  1293. return;
  1294. case RINGBUF_TYPE_DATA:
  1295. cpu_buffer->read_stamp += event->time_delta;
  1296. return;
  1297. default:
  1298. BUG();
  1299. }
  1300. return;
  1301. }
  1302. static void
  1303. rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
  1304. struct ring_buffer_event *event)
  1305. {
  1306. u64 delta;
  1307. switch (event->type) {
  1308. case RINGBUF_TYPE_PADDING:
  1309. return;
  1310. case RINGBUF_TYPE_TIME_EXTEND:
  1311. delta = event->array[0];
  1312. delta <<= TS_SHIFT;
  1313. delta += event->time_delta;
  1314. iter->read_stamp += delta;
  1315. return;
  1316. case RINGBUF_TYPE_TIME_STAMP:
  1317. /* FIXME: not implemented */
  1318. return;
  1319. case RINGBUF_TYPE_DATA:
  1320. iter->read_stamp += event->time_delta;
  1321. return;
  1322. default:
  1323. BUG();
  1324. }
  1325. return;
  1326. }
  1327. static struct buffer_page *
  1328. rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
  1329. {
  1330. struct buffer_page *reader = NULL;
  1331. unsigned long flags;
  1332. int nr_loops = 0;
  1333. local_irq_save(flags);
  1334. __raw_spin_lock(&cpu_buffer->lock);
  1335. again:
  1336. /*
  1337. * This should normally only loop twice. But because the
  1338. * start of the reader inserts an empty page, it causes
  1339. * a case where we will loop three times. There should be no
  1340. * reason to loop four times (that I know of).
  1341. */
  1342. if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
  1343. reader = NULL;
  1344. goto out;
  1345. }
  1346. reader = cpu_buffer->reader_page;
  1347. /* If there's more to read, return this page */
  1348. if (cpu_buffer->reader_page->read < rb_page_size(reader))
  1349. goto out;
  1350. /* Never should we have an index greater than the size */
  1351. if (RB_WARN_ON(cpu_buffer,
  1352. cpu_buffer->reader_page->read > rb_page_size(reader)))
  1353. goto out;
  1354. /* check if we caught up to the tail */
  1355. reader = NULL;
  1356. if (cpu_buffer->commit_page == cpu_buffer->reader_page)
  1357. goto out;
  1358. /*
  1359. * Splice the empty reader page into the list around the head.
  1360. * Reset the reader page to size zero.
  1361. */
  1362. reader = cpu_buffer->head_page;
  1363. cpu_buffer->reader_page->list.next = reader->list.next;
  1364. cpu_buffer->reader_page->list.prev = reader->list.prev;
  1365. local_set(&cpu_buffer->reader_page->write, 0);
  1366. local_set(&cpu_buffer->reader_page->commit, 0);
  1367. /* Make the reader page now replace the head */
  1368. reader->list.prev->next = &cpu_buffer->reader_page->list;
  1369. reader->list.next->prev = &cpu_buffer->reader_page->list;
  1370. /*
  1371. * If the tail is on the reader, then we must set the head
  1372. * to the inserted page, otherwise we set it one before.
  1373. */
  1374. cpu_buffer->head_page = cpu_buffer->reader_page;
  1375. if (cpu_buffer->commit_page != reader)
  1376. rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
  1377. /* Finally update the reader page to the new head */
  1378. cpu_buffer->reader_page = reader;
  1379. rb_reset_reader_page(cpu_buffer);
  1380. goto again;
  1381. out:
  1382. __raw_spin_unlock(&cpu_buffer->lock);
  1383. local_irq_restore(flags);
  1384. return reader;
  1385. }
  1386. static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
  1387. {
  1388. struct ring_buffer_event *event;
  1389. struct buffer_page *reader;
  1390. unsigned length;
  1391. reader = rb_get_reader_page(cpu_buffer);
  1392. /* This function should not be called when buffer is empty */
  1393. if (RB_WARN_ON(cpu_buffer, !reader))
  1394. return;
  1395. event = rb_reader_event(cpu_buffer);
  1396. if (event->type == RINGBUF_TYPE_DATA)
  1397. cpu_buffer->entries--;
  1398. rb_update_read_stamp(cpu_buffer, event);
  1399. length = rb_event_length(event);
  1400. cpu_buffer->reader_page->read += length;
  1401. }
  1402. static void rb_advance_iter(struct ring_buffer_iter *iter)
  1403. {
  1404. struct ring_buffer *buffer;
  1405. struct ring_buffer_per_cpu *cpu_buffer;
  1406. struct ring_buffer_event *event;
  1407. unsigned length;
  1408. cpu_buffer = iter->cpu_buffer;
  1409. buffer = cpu_buffer->buffer;
  1410. /*
  1411. * Check if we are at the end of the buffer.
  1412. */
  1413. if (iter->head >= rb_page_size(iter->head_page)) {
  1414. if (RB_WARN_ON(buffer,
  1415. iter->head_page == cpu_buffer->commit_page))
  1416. return;
  1417. rb_inc_iter(iter);
  1418. return;
  1419. }
  1420. event = rb_iter_head_event(iter);
  1421. length = rb_event_length(event);
  1422. /*
  1423. * This should not be called to advance the header if we are
  1424. * at the tail of the buffer.
  1425. */
  1426. if (RB_WARN_ON(cpu_buffer,
  1427. (iter->head_page == cpu_buffer->commit_page) &&
  1428. (iter->head + length > rb_commit_index(cpu_buffer))))
  1429. return;
  1430. rb_update_iter_read_stamp(iter, event);
  1431. iter->head += length;
  1432. /* check for end of page padding */
  1433. if ((iter->head >= rb_page_size(iter->head_page)) &&
  1434. (iter->head_page != cpu_buffer->commit_page))
  1435. rb_advance_iter(iter);
  1436. }
  1437. static struct ring_buffer_event *
  1438. rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
  1439. {
  1440. struct ring_buffer_per_cpu *cpu_buffer;
  1441. struct ring_buffer_event *event;
  1442. struct buffer_page *reader;
  1443. int nr_loops = 0;
  1444. if (!cpu_isset(cpu, buffer->cpumask))
  1445. return NULL;
  1446. cpu_buffer = buffer->buffers[cpu];
  1447. again:
  1448. /*
  1449. * We repeat when a timestamp is encountered. It is possible
  1450. * to get multiple timestamps from an interrupt entering just
  1451. * as one timestamp is about to be written. The max times
  1452. * that this can happen is the number of nested interrupts we
  1453. * can have. Nesting 10 deep of interrupts is clearly
  1454. * an anomaly.
  1455. */
  1456. if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10))
  1457. return NULL;
  1458. reader = rb_get_reader_page(cpu_buffer);
  1459. if (!reader)
  1460. return NULL;
  1461. event = rb_reader_event(cpu_buffer);
  1462. switch (event->type) {
  1463. case RINGBUF_TYPE_PADDING:
  1464. RB_WARN_ON(cpu_buffer, 1);
  1465. rb_advance_reader(cpu_buffer);
  1466. return NULL;
  1467. case RINGBUF_TYPE_TIME_EXTEND:
  1468. /* Internal data, OK to advance */
  1469. rb_advance_reader(cpu_buffer);
  1470. goto again;
  1471. case RINGBUF_TYPE_TIME_STAMP:
  1472. /* FIXME: not implemented */
  1473. rb_advance_reader(cpu_buffer);
  1474. goto again;
  1475. case RINGBUF_TYPE_DATA:
  1476. if (ts) {
  1477. *ts = cpu_buffer->read_stamp + event->time_delta;
  1478. ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
  1479. }
  1480. return event;
  1481. default:
  1482. BUG();
  1483. }
  1484. return NULL;
  1485. }
  1486. static struct ring_buffer_event *
  1487. rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
  1488. {
  1489. struct ring_buffer *buffer;
  1490. struct ring_buffer_per_cpu *cpu_buffer;
  1491. struct ring_buffer_event *event;
  1492. int nr_loops = 0;
  1493. if (ring_buffer_iter_empty(iter))
  1494. return NULL;
  1495. cpu_buffer = iter->cpu_buffer;
  1496. buffer = cpu_buffer->buffer;
  1497. again:
  1498. /*
  1499. * We repeat when a timestamp is encountered. It is possible
  1500. * to get multiple timestamps from an interrupt entering just
  1501. * as one timestamp is about to be written. The max times
  1502. * that this can happen is the number of nested interrupts we
  1503. * can have. Nesting 10 deep of interrupts is clearly
  1504. * an anomaly.
  1505. */
  1506. if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10))
  1507. return NULL;
  1508. if (rb_per_cpu_empty(cpu_buffer))
  1509. return NULL;
  1510. event = rb_iter_head_event(iter);
  1511. switch (event->type) {
  1512. case RINGBUF_TYPE_PADDING:
  1513. rb_inc_iter(iter);
  1514. goto again;
  1515. case RINGBUF_TYPE_TIME_EXTEND:
  1516. /* Internal data, OK to advance */
  1517. rb_advance_iter(iter);
  1518. goto again;
  1519. case RINGBUF_TYPE_TIME_STAMP:
  1520. /* FIXME: not implemented */
  1521. rb_advance_iter(iter);
  1522. goto again;
  1523. case RINGBUF_TYPE_DATA:
  1524. if (ts) {
  1525. *ts = iter->read_stamp + event->time_delta;
  1526. ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
  1527. }
  1528. return event;
  1529. default:
  1530. BUG();
  1531. }
  1532. return NULL;
  1533. }
  1534. /**
  1535. * ring_buffer_peek - peek at the next event to be read
  1536. * @buffer: The ring buffer to read
  1537. * @cpu: The cpu to peak at
  1538. * @ts: The timestamp counter of this event.
  1539. *
  1540. * This will return the event that will be read next, but does
  1541. * not consume the data.
  1542. */
  1543. struct ring_buffer_event *
  1544. ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
  1545. {
  1546. struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
  1547. struct ring_buffer_event *event;
  1548. unsigned long flags;
  1549. spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
  1550. event = rb_buffer_peek(buffer, cpu, ts);
  1551. spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
  1552. return event;
  1553. }
  1554. /**
  1555. * ring_buffer_iter_peek - peek at the next event to be read
  1556. * @iter: The ring buffer iterator
  1557. * @ts: The timestamp counter of this event.
  1558. *
  1559. * This will return the event that will be read next, but does
  1560. * not increment the iterator.
  1561. */
  1562. struct ring_buffer_event *
  1563. ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
  1564. {
  1565. struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
  1566. struct ring_buffer_event *event;
  1567. unsigned long flags;
  1568. spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
  1569. event = rb_iter_peek(iter, ts);
  1570. spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
  1571. return event;
  1572. }
  1573. /**
  1574. * ring_buffer_consume - return an event and consume it
  1575. * @buffer: The ring buffer to get the next event from
  1576. *
  1577. * Returns the next event in the ring buffer, and that event is consumed.
  1578. * Meaning, that sequential reads will keep returning a different event,
  1579. * and eventually empty the ring buffer if the producer is slower.
  1580. */
  1581. struct ring_buffer_event *
  1582. ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
  1583. {
  1584. struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
  1585. struct ring_buffer_event *event;
  1586. unsigned long flags;
  1587. if (!cpu_isset(cpu, buffer->cpumask))
  1588. return NULL;
  1589. spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
  1590. event = rb_buffer_peek(buffer, cpu, ts);
  1591. if (!event)
  1592. goto out;
  1593. rb_advance_reader(cpu_buffer);
  1594. out:
  1595. spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
  1596. return event;
  1597. }
  1598. /**
  1599. * ring_buffer_read_start - start a non consuming read of the buffer
  1600. * @buffer: The ring buffer to read from
  1601. * @cpu: The cpu buffer to iterate over
  1602. *
  1603. * This starts up an iteration through the buffer. It also disables
  1604. * the recording to the buffer until the reading is finished.
  1605. * This prevents the reading from being corrupted. This is not
  1606. * a consuming read, so a producer is not expected.
  1607. *
  1608. * Must be paired with ring_buffer_finish.
  1609. */
  1610. struct ring_buffer_iter *
  1611. ring_buffer_read_start(struct ring_buffer *buffer, int cpu)
  1612. {
  1613. struct ring_buffer_per_cpu *cpu_buffer;
  1614. struct ring_buffer_iter *iter;
  1615. unsigned long flags;
  1616. if (!cpu_isset(cpu, buffer->cpumask))
  1617. return NULL;
  1618. iter = kmalloc(sizeof(*iter), GFP_KERNEL);
  1619. if (!iter)
  1620. return NULL;
  1621. cpu_buffer = buffer->buffers[cpu];
  1622. iter->cpu_buffer = cpu_buffer;
  1623. atomic_inc(&cpu_buffer->record_disabled);
  1624. synchronize_sched();
  1625. spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
  1626. __raw_spin_lock(&cpu_buffer->lock);
  1627. rb_iter_reset(iter);
  1628. __raw_spin_unlock(&cpu_buffer->lock);
  1629. spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
  1630. return iter;
  1631. }
  1632. /**
  1633. * ring_buffer_finish - finish reading the iterator of the buffer
  1634. * @iter: The iterator retrieved by ring_buffer_start
  1635. *
  1636. * This re-enables the recording to the buffer, and frees the
  1637. * iterator.
  1638. */
  1639. void
  1640. ring_buffer_read_finish(struct ring_buffer_iter *iter)
  1641. {
  1642. struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
  1643. atomic_dec(&cpu_buffer->record_disabled);
  1644. kfree(iter);
  1645. }
  1646. /**
  1647. * ring_buffer_read - read the next item in the ring buffer by the iterator
  1648. * @iter: The ring buffer iterator
  1649. * @ts: The time stamp of the event read.
  1650. *
  1651. * This reads the next event in the ring buffer and increments the iterator.
  1652. */
  1653. struct ring_buffer_event *
  1654. ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
  1655. {
  1656. struct ring_buffer_event *event;
  1657. struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
  1658. unsigned long flags;
  1659. spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
  1660. event = rb_iter_peek(iter, ts);
  1661. if (!event)
  1662. goto out;
  1663. rb_advance_iter(iter);
  1664. out:
  1665. spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
  1666. return event;
  1667. }
  1668. /**
  1669. * ring_buffer_size - return the size of the ring buffer (in bytes)
  1670. * @buffer: The ring buffer.
  1671. */
  1672. unsigned long ring_buffer_size(struct ring_buffer *buffer)
  1673. {
  1674. return BUF_PAGE_SIZE * buffer->pages;
  1675. }
  1676. static void
  1677. rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
  1678. {
  1679. cpu_buffer->head_page
  1680. = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
  1681. local_set(&cpu_buffer->head_page->write, 0);
  1682. local_set(&cpu_buffer->head_page->commit, 0);
  1683. cpu_buffer->head_page->read = 0;
  1684. cpu_buffer->tail_page = cpu_buffer->head_page;
  1685. cpu_buffer->commit_page = cpu_buffer->head_page;
  1686. INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
  1687. local_set(&cpu_buffer->reader_page->write, 0);
  1688. local_set(&cpu_buffer->reader_page->commit, 0);
  1689. cpu_buffer->reader_page->read = 0;
  1690. cpu_buffer->overrun = 0;
  1691. cpu_buffer->entries = 0;
  1692. }
  1693. /**
  1694. * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
  1695. * @buffer: The ring buffer to reset a per cpu buffer of
  1696. * @cpu: The CPU buffer to be reset
  1697. */
  1698. void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
  1699. {
  1700. struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
  1701. unsigned long flags;
  1702. if (!cpu_isset(cpu, buffer->cpumask))
  1703. return;
  1704. spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
  1705. __raw_spin_lock(&cpu_buffer->lock);
  1706. rb_reset_cpu(cpu_buffer);
  1707. __raw_spin_unlock(&cpu_buffer->lock);
  1708. spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
  1709. }
  1710. /**
  1711. * ring_buffer_reset - reset a ring buffer
  1712. * @buffer: The ring buffer to reset all cpu buffers
  1713. */
  1714. void ring_buffer_reset(struct ring_buffer *buffer)
  1715. {
  1716. int cpu;
  1717. for_each_buffer_cpu(buffer, cpu)
  1718. ring_buffer_reset_cpu(buffer, cpu);
  1719. }
  1720. /**
  1721. * rind_buffer_empty - is the ring buffer empty?
  1722. * @buffer: The ring buffer to test
  1723. */
  1724. int ring_buffer_empty(struct ring_buffer *buffer)
  1725. {
  1726. struct ring_buffer_per_cpu *cpu_buffer;
  1727. int cpu;
  1728. /* yes this is racy, but if you don't like the race, lock the buffer */
  1729. for_each_buffer_cpu(buffer, cpu) {
  1730. cpu_buffer = buffer->buffers[cpu];
  1731. if (!rb_per_cpu_empty(cpu_buffer))
  1732. return 0;
  1733. }
  1734. return 1;
  1735. }
  1736. /**
  1737. * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
  1738. * @buffer: The ring buffer
  1739. * @cpu: The CPU buffer to test
  1740. */
  1741. int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
  1742. {
  1743. struct ring_buffer_per_cpu *cpu_buffer;
  1744. if (!cpu_isset(cpu, buffer->cpumask))
  1745. return 1;
  1746. cpu_buffer = buffer->buffers[cpu];
  1747. return rb_per_cpu_empty(cpu_buffer);
  1748. }
  1749. /**
  1750. * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
  1751. * @buffer_a: One buffer to swap with
  1752. * @buffer_b: The other buffer to swap with
  1753. *
  1754. * This function is useful for tracers that want to take a "snapshot"
  1755. * of a CPU buffer and has another back up buffer lying around.
  1756. * it is expected that the tracer handles the cpu buffer not being
  1757. * used at the moment.
  1758. */
  1759. int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
  1760. struct ring_buffer *buffer_b, int cpu)
  1761. {
  1762. struct ring_buffer_per_cpu *cpu_buffer_a;
  1763. struct ring_buffer_per_cpu *cpu_buffer_b;
  1764. if (!cpu_isset(cpu, buffer_a->cpumask) ||
  1765. !cpu_isset(cpu, buffer_b->cpumask))
  1766. return -EINVAL;
  1767. /* At least make sure the two buffers are somewhat the same */
  1768. if (buffer_a->size != buffer_b->size ||
  1769. buffer_a->pages != buffer_b->pages)
  1770. return -EINVAL;
  1771. cpu_buffer_a = buffer_a->buffers[cpu];
  1772. cpu_buffer_b = buffer_b->buffers[cpu];
  1773. /*
  1774. * We can't do a synchronize_sched here because this
  1775. * function can be called in atomic context.
  1776. * Normally this will be called from the same CPU as cpu.
  1777. * If not it's up to the caller to protect this.
  1778. */
  1779. atomic_inc(&cpu_buffer_a->record_disabled);
  1780. atomic_inc(&cpu_buffer_b->record_disabled);
  1781. buffer_a->buffers[cpu] = cpu_buffer_b;
  1782. buffer_b->buffers[cpu] = cpu_buffer_a;
  1783. cpu_buffer_b->buffer = buffer_a;
  1784. cpu_buffer_a->buffer = buffer_b;
  1785. atomic_dec(&cpu_buffer_a->record_disabled);
  1786. atomic_dec(&cpu_buffer_b->record_disabled);
  1787. return 0;
  1788. }
  1789. static ssize_t
  1790. rb_simple_read(struct file *filp, char __user *ubuf,
  1791. size_t cnt, loff_t *ppos)
  1792. {
  1793. int *p = filp->private_data;
  1794. char buf[64];
  1795. int r;
  1796. /* !ring_buffers_off == tracing_on */
  1797. r = sprintf(buf, "%d\n", !*p);
  1798. return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
  1799. }
  1800. static ssize_t
  1801. rb_simple_write(struct file *filp, const char __user *ubuf,
  1802. size_t cnt, loff_t *ppos)
  1803. {
  1804. int *p = filp->private_data;
  1805. char buf[64];
  1806. long val;
  1807. int ret;
  1808. if (cnt >= sizeof(buf))
  1809. return -EINVAL;
  1810. if (copy_from_user(&buf, ubuf, cnt))
  1811. return -EFAULT;
  1812. buf[cnt] = 0;
  1813. ret = strict_strtoul(buf, 10, &val);
  1814. if (ret < 0)
  1815. return ret;
  1816. /* !ring_buffers_off == tracing_on */
  1817. *p = !val;
  1818. (*ppos)++;
  1819. return cnt;
  1820. }
  1821. static struct file_operations rb_simple_fops = {
  1822. .open = tracing_open_generic,
  1823. .read = rb_simple_read,
  1824. .write = rb_simple_write,
  1825. };
  1826. static __init int rb_init_debugfs(void)
  1827. {
  1828. struct dentry *d_tracer;
  1829. struct dentry *entry;
  1830. d_tracer = tracing_init_dentry();
  1831. entry = debugfs_create_file("tracing_on", 0644, d_tracer,
  1832. &ring_buffers_off, &rb_simple_fops);
  1833. if (!entry)
  1834. pr_warning("Could not create debugfs 'tracing_on' entry\n");
  1835. return 0;
  1836. }
  1837. fs_initcall(rb_init_debugfs);