blob: 3c7834c24e54e6630a286423cfd17ba3d6b61350 [file] [log] [blame]
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001/*
2 * Generic ring buffer
3 *
4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
5 */
6#include <linux/ring_buffer.h>
Ingo Molnar14131f22009-02-26 18:47:11 +01007#include <linux/trace_clock.h>
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04008#include <linux/spinlock.h>
9#include <linux/debugfs.h>
10#include <linux/uaccess.h>
Steven Rostedta81bd802009-02-06 01:45:16 -050011#include <linux/hardirq.h>
Vegard Nossum1744a212009-02-28 08:29:44 +010012#include <linux/kmemcheck.h>
Steven Rostedt7a8e76a2008-09-29 23:02:38 -040013#include <linux/module.h>
14#include <linux/percpu.h>
15#include <linux/mutex.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/slab.h>
Steven Rostedt7a8e76a2008-09-29 23:02:38 -040017#include <linux/init.h>
18#include <linux/hash.h>
19#include <linux/list.h>
Steven Rostedt554f7862009-03-11 22:00:13 -040020#include <linux/cpu.h>
Steven Rostedt7a8e76a2008-09-29 23:02:38 -040021#include <linux/fs.h>
22
Christoph Lameter79615762010-01-05 15:34:50 +090023#include <asm/local.h>
Steven Rostedt182e9f52008-11-03 23:15:56 -050024#include "trace.h"
25
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -070026static void update_pages_handler(struct work_struct *work);
27
Steven Rostedt033601a2008-11-21 12:41:55 -050028/*
Steven Rostedtd1b182a2009-04-15 16:53:47 -040029 * The ring buffer header is special. We must manually up keep it.
30 */
31int ring_buffer_print_entry_header(struct trace_seq *s)
32{
33 int ret;
34
Lai Jiangshan334d4162009-04-24 11:27:05 +080035 ret = trace_seq_printf(s, "# compressed entry header\n");
36 ret = trace_seq_printf(s, "\ttype_len : 5 bits\n");
Steven Rostedtd1b182a2009-04-15 16:53:47 -040037 ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n");
38 ret = trace_seq_printf(s, "\tarray : 32 bits\n");
39 ret = trace_seq_printf(s, "\n");
40 ret = trace_seq_printf(s, "\tpadding : type == %d\n",
41 RINGBUF_TYPE_PADDING);
42 ret = trace_seq_printf(s, "\ttime_extend : type == %d\n",
43 RINGBUF_TYPE_TIME_EXTEND);
Lai Jiangshan334d4162009-04-24 11:27:05 +080044 ret = trace_seq_printf(s, "\tdata max type_len == %d\n",
45 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
Steven Rostedtd1b182a2009-04-15 16:53:47 -040046
47 return ret;
48}
49
50/*
Steven Rostedt5cc98542009-03-12 22:24:17 -040051 * The ring buffer is made up of a list of pages. A separate list of pages is
52 * allocated for each CPU. A writer may only write to a buffer that is
53 * associated with the CPU it is currently executing on. A reader may read
54 * from any per cpu buffer.
55 *
56 * The reader is special. For each per cpu buffer, the reader has its own
57 * reader page. When a reader has read the entire reader page, this reader
58 * page is swapped with another page in the ring buffer.
59 *
60 * Now, as long as the writer is off the reader page, the reader can do what
61 * ever it wants with that page. The writer will never write to that page
62 * again (as long as it is out of the ring buffer).
63 *
64 * Here's some silly ASCII art.
65 *
66 * +------+
67 * |reader| RING BUFFER
68 * |page |
69 * +------+ +---+ +---+ +---+
70 * | |-->| |-->| |
71 * +---+ +---+ +---+
72 * ^ |
73 * | |
74 * +---------------+
75 *
76 *
77 * +------+
78 * |reader| RING BUFFER
79 * |page |------------------v
80 * +------+ +---+ +---+ +---+
81 * | |-->| |-->| |
82 * +---+ +---+ +---+
83 * ^ |
84 * | |
85 * +---------------+
86 *
87 *
88 * +------+
89 * |reader| RING BUFFER
90 * |page |------------------v
91 * +------+ +---+ +---+ +---+
92 * ^ | |-->| |-->| |
93 * | +---+ +---+ +---+
94 * | |
95 * | |
96 * +------------------------------+
97 *
98 *
99 * +------+
100 * |buffer| RING BUFFER
101 * |page |------------------v
102 * +------+ +---+ +---+ +---+
103 * ^ | | | |-->| |
104 * | New +---+ +---+ +---+
105 * | Reader------^ |
106 * | page |
107 * +------------------------------+
108 *
109 *
110 * After we make this swap, the reader can hand this page off to the splice
111 * code and be done with it. It can even allocate a new page if it needs to
112 * and swap that into the ring buffer.
113 *
114 * We will be using cmpxchg soon to make all this lockless.
115 *
116 */
117
118/*
Steven Rostedt033601a2008-11-21 12:41:55 -0500119 * A fast way to enable or disable all ring buffers is to
120 * call tracing_on or tracing_off. Turning off the ring buffers
121 * prevents all ring buffers from being recorded to.
122 * Turning this switch on, makes it OK to write to the
123 * ring buffer, if the ring buffer is enabled itself.
124 *
125 * There's three layers that must be on in order to write
126 * to the ring buffer.
127 *
128 * 1) This global flag must be set.
129 * 2) The ring buffer must be enabled for recording.
130 * 3) The per cpu buffer must be enabled for recording.
131 *
132 * In case of an anomaly, this global flag has a bit set that
133 * will permantly disable all ring buffers.
134 */
135
136/*
137 * Global flag to disable all recording to ring buffers
138 * This has two bits: ON, DISABLED
139 *
140 * ON DISABLED
141 * ---- ----------
142 * 0 0 : ring buffers are off
143 * 1 0 : ring buffers are on
144 * X 1 : ring buffers are permanently disabled
145 */
146
147enum {
148 RB_BUFFERS_ON_BIT = 0,
149 RB_BUFFERS_DISABLED_BIT = 1,
150};
151
152enum {
153 RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT,
154 RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT,
155};
156
Hannes Eder5e398412009-02-10 19:44:34 +0100157static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON;
Steven Rostedta3583242008-11-11 15:01:42 -0500158
Steven Rostedt499e5472012-02-22 15:50:28 -0500159/* Used for individual buffers (after the counter) */
160#define RB_BUFFER_OFF (1 << 20)
161
Steven Rostedt474d32b2009-03-03 19:51:40 -0500162#define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
163
Steven Rostedta3583242008-11-11 15:01:42 -0500164/**
Steven Rostedt033601a2008-11-21 12:41:55 -0500165 * tracing_off_permanent - permanently disable ring buffers
166 *
167 * This function, once called, will disable all ring buffers
Wenji Huangc3706f02009-02-10 01:03:18 -0500168 * permanently.
Steven Rostedt033601a2008-11-21 12:41:55 -0500169 */
170void tracing_off_permanent(void)
171{
172 set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags);
Steven Rostedta3583242008-11-11 15:01:42 -0500173}
174
Steven Rostedte3d6bf02009-03-03 13:53:07 -0500175#define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
Andrew Morton67d34722009-01-09 12:27:09 -0800176#define RB_ALIGNMENT 4U
Lai Jiangshan334d4162009-04-24 11:27:05 +0800177#define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
Steven Rostedtc7b09302009-06-11 11:12:00 -0400178#define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
Lai Jiangshan334d4162009-04-24 11:27:05 +0800179
Steven Rostedt22710482010-03-18 17:54:19 -0400180#if !defined(CONFIG_64BIT) || defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
181# define RB_FORCE_8BYTE_ALIGNMENT 0
182# define RB_ARCH_ALIGNMENT RB_ALIGNMENT
183#else
184# define RB_FORCE_8BYTE_ALIGNMENT 1
185# define RB_ARCH_ALIGNMENT 8U
186#endif
187
Lai Jiangshan334d4162009-04-24 11:27:05 +0800188/* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
189#define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400190
191enum {
192 RB_LEN_TIME_EXTEND = 8,
193 RB_LEN_TIME_STAMP = 16,
194};
195
Steven Rostedt69d1b832010-10-07 18:18:05 -0400196#define skip_time_extend(event) \
197 ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
198
Tom Zanussi2d622712009-03-22 03:30:49 -0500199static inline int rb_null_event(struct ring_buffer_event *event)
200{
Steven Rostedta1863c22009-09-03 10:23:58 -0400201 return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
Tom Zanussi2d622712009-03-22 03:30:49 -0500202}
203
204static void rb_event_set_padding(struct ring_buffer_event *event)
205{
Steven Rostedta1863c22009-09-03 10:23:58 -0400206 /* padding has a NULL time_delta */
Lai Jiangshan334d4162009-04-24 11:27:05 +0800207 event->type_len = RINGBUF_TYPE_PADDING;
Tom Zanussi2d622712009-03-22 03:30:49 -0500208 event->time_delta = 0;
209}
210
Tom Zanussi2d622712009-03-22 03:30:49 -0500211static unsigned
212rb_event_data_length(struct ring_buffer_event *event)
213{
214 unsigned length;
215
Lai Jiangshan334d4162009-04-24 11:27:05 +0800216 if (event->type_len)
217 length = event->type_len * RB_ALIGNMENT;
Tom Zanussi2d622712009-03-22 03:30:49 -0500218 else
219 length = event->array[0];
220 return length + RB_EVNT_HDR_SIZE;
221}
222
Steven Rostedt69d1b832010-10-07 18:18:05 -0400223/*
224 * Return the length of the given event. Will return
225 * the length of the time extend if the event is a
226 * time extend.
227 */
228static inline unsigned
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400229rb_event_length(struct ring_buffer_event *event)
230{
Lai Jiangshan334d4162009-04-24 11:27:05 +0800231 switch (event->type_len) {
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400232 case RINGBUF_TYPE_PADDING:
Tom Zanussi2d622712009-03-22 03:30:49 -0500233 if (rb_null_event(event))
234 /* undefined */
235 return -1;
Lai Jiangshan334d4162009-04-24 11:27:05 +0800236 return event->array[0] + RB_EVNT_HDR_SIZE;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400237
238 case RINGBUF_TYPE_TIME_EXTEND:
239 return RB_LEN_TIME_EXTEND;
240
241 case RINGBUF_TYPE_TIME_STAMP:
242 return RB_LEN_TIME_STAMP;
243
244 case RINGBUF_TYPE_DATA:
Tom Zanussi2d622712009-03-22 03:30:49 -0500245 return rb_event_data_length(event);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400246 default:
247 BUG();
248 }
249 /* not hit */
250 return 0;
251}
252
Steven Rostedt69d1b832010-10-07 18:18:05 -0400253/*
254 * Return total length of time extend and data,
255 * or just the event length for all other events.
256 */
257static inline unsigned
258rb_event_ts_length(struct ring_buffer_event *event)
259{
260 unsigned len = 0;
261
262 if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) {
263 /* time extends include the data event after it */
264 len = RB_LEN_TIME_EXTEND;
265 event = skip_time_extend(event);
266 }
267 return len + rb_event_length(event);
268}
269
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400270/**
271 * ring_buffer_event_length - return the length of the event
272 * @event: the event to get the length of
Steven Rostedt69d1b832010-10-07 18:18:05 -0400273 *
274 * Returns the size of the data load of a data event.
275 * If the event is something other than a data event, it
276 * returns the size of the event itself. With the exception
277 * of a TIME EXTEND, where it still returns the size of the
278 * data load of the data event after it.
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400279 */
280unsigned ring_buffer_event_length(struct ring_buffer_event *event)
281{
Steven Rostedt69d1b832010-10-07 18:18:05 -0400282 unsigned length;
283
284 if (event->type_len == RINGBUF_TYPE_TIME_EXTEND)
285 event = skip_time_extend(event);
286
287 length = rb_event_length(event);
Lai Jiangshan334d4162009-04-24 11:27:05 +0800288 if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
Robert Richter465634a2009-01-07 15:32:11 +0100289 return length;
290 length -= RB_EVNT_HDR_SIZE;
291 if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
292 length -= sizeof(event->array[0]);
293 return length;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400294}
Robert Richterc4f50182008-12-11 16:49:22 +0100295EXPORT_SYMBOL_GPL(ring_buffer_event_length);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400296
297/* inline for ring buffer fast paths */
Andrew Morton34a148b2009-01-09 12:27:09 -0800298static void *
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400299rb_event_data(struct ring_buffer_event *event)
300{
Steven Rostedt69d1b832010-10-07 18:18:05 -0400301 if (event->type_len == RINGBUF_TYPE_TIME_EXTEND)
302 event = skip_time_extend(event);
Lai Jiangshan334d4162009-04-24 11:27:05 +0800303 BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400304 /* If length is in len field, then array[0] has the data */
Lai Jiangshan334d4162009-04-24 11:27:05 +0800305 if (event->type_len)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400306 return (void *)&event->array[0];
307 /* Otherwise length is in array[0] and array[1] has the data */
308 return (void *)&event->array[1];
309}
310
311/**
312 * ring_buffer_event_data - return the data of the event
313 * @event: the event to get the data from
314 */
315void *ring_buffer_event_data(struct ring_buffer_event *event)
316{
317 return rb_event_data(event);
318}
Robert Richterc4f50182008-12-11 16:49:22 +0100319EXPORT_SYMBOL_GPL(ring_buffer_event_data);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400320
321#define for_each_buffer_cpu(buffer, cpu) \
Rusty Russell9e01c1b2009-01-01 10:12:22 +1030322 for_each_cpu(cpu, buffer->cpumask)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400323
324#define TS_SHIFT 27
325#define TS_MASK ((1ULL << TS_SHIFT) - 1)
326#define TS_DELTA_TEST (~TS_MASK)
327
Steven Rostedt66a8cb92010-03-31 13:21:56 -0400328/* Flag when events were overwritten */
329#define RB_MISSED_EVENTS (1 << 31)
Steven Rostedtff0ff842010-03-31 22:11:42 -0400330/* Missed count stored at end */
331#define RB_MISSED_STORED (1 << 30)
Steven Rostedt66a8cb92010-03-31 13:21:56 -0400332
Steven Rostedtabc9b562008-12-02 15:34:06 -0500333struct buffer_data_page {
Steven Rostedte4c2ce82008-10-01 11:14:54 -0400334 u64 time_stamp; /* page time stamp */
Wenji Huangc3706f02009-02-10 01:03:18 -0500335 local_t commit; /* write committed index */
Steven Rostedtabc9b562008-12-02 15:34:06 -0500336 unsigned char data[]; /* data of buffer page */
337};
338
Steven Rostedt77ae3652009-03-27 11:00:29 -0400339/*
340 * Note, the buffer_page list must be first. The buffer pages
341 * are allocated in cache lines, which means that each buffer
342 * page will be at the beginning of a cache line, and thus
343 * the least significant bits will be zero. We use this to
344 * add flags in the list struct pointers, to make the ring buffer
345 * lockless.
346 */
Steven Rostedtabc9b562008-12-02 15:34:06 -0500347struct buffer_page {
Steven Rostedt778c55d2009-05-01 18:44:45 -0400348 struct list_head list; /* list of buffer pages */
Steven Rostedtabc9b562008-12-02 15:34:06 -0500349 local_t write; /* index for next write */
Steven Rostedt6f807ac2008-10-04 02:00:58 -0400350 unsigned read; /* index for next read */
Steven Rostedt778c55d2009-05-01 18:44:45 -0400351 local_t entries; /* entries on this page */
Steven Rostedtff0ff842010-03-31 22:11:42 -0400352 unsigned long real_end; /* real end of data */
Steven Rostedtabc9b562008-12-02 15:34:06 -0500353 struct buffer_data_page *page; /* Actual data page */
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400354};
355
Steven Rostedt77ae3652009-03-27 11:00:29 -0400356/*
357 * The buffer page counters, write and entries, must be reset
358 * atomically when crossing page boundaries. To synchronize this
359 * update, two counters are inserted into the number. One is
360 * the actual counter for the write position or count on the page.
361 *
362 * The other is a counter of updaters. Before an update happens
363 * the update partition of the counter is incremented. This will
364 * allow the updater to update the counter atomically.
365 *
366 * The counter is 20 bits, and the state data is 12.
367 */
368#define RB_WRITE_MASK 0xfffff
369#define RB_WRITE_INTCNT (1 << 20)
370
Steven Rostedt044fa782008-12-02 23:50:03 -0500371static void rb_init_page(struct buffer_data_page *bpage)
Steven Rostedtabc9b562008-12-02 15:34:06 -0500372{
Steven Rostedt044fa782008-12-02 23:50:03 -0500373 local_set(&bpage->commit, 0);
Steven Rostedtabc9b562008-12-02 15:34:06 -0500374}
375
Steven Rostedt474d32b2009-03-03 19:51:40 -0500376/**
377 * ring_buffer_page_len - the size of data on the page.
378 * @page: The page to read
379 *
380 * Returns the amount of data on the page, including buffer page header.
381 */
Steven Rostedtef7a4a12009-03-03 00:27:49 -0500382size_t ring_buffer_page_len(void *page)
383{
Steven Rostedt474d32b2009-03-03 19:51:40 -0500384 return local_read(&((struct buffer_data_page *)page)->commit)
385 + BUF_PAGE_HDR_SIZE;
Steven Rostedtef7a4a12009-03-03 00:27:49 -0500386}
387
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400388/*
Steven Rostedted568292008-09-29 23:02:40 -0400389 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
390 * this issue out.
391 */
Andrew Morton34a148b2009-01-09 12:27:09 -0800392static void free_buffer_page(struct buffer_page *bpage)
Steven Rostedted568292008-09-29 23:02:40 -0400393{
Andrew Morton34a148b2009-01-09 12:27:09 -0800394 free_page((unsigned long)bpage->page);
Steven Rostedte4c2ce82008-10-01 11:14:54 -0400395 kfree(bpage);
Steven Rostedted568292008-09-29 23:02:40 -0400396}
397
398/*
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400399 * We need to fit the time_stamp delta into 27 bits.
400 */
401static inline int test_time_stamp(u64 delta)
402{
403 if (delta & TS_DELTA_TEST)
404 return 1;
405 return 0;
406}
407
Steven Rostedt474d32b2009-03-03 19:51:40 -0500408#define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400409
Steven Rostedtbe957c42009-05-11 14:42:53 -0400410/* Max payload is BUF_PAGE_SIZE - header (8bytes) */
411#define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
412
Steven Rostedtd1b182a2009-04-15 16:53:47 -0400413int ring_buffer_print_page_header(struct trace_seq *s)
414{
415 struct buffer_data_page field;
416 int ret;
417
418 ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t"
Tom Zanussi26a50742009-10-06 01:09:50 -0500419 "offset:0;\tsize:%u;\tsigned:%u;\n",
420 (unsigned int)sizeof(field.time_stamp),
421 (unsigned int)is_signed_type(u64));
Steven Rostedtd1b182a2009-04-15 16:53:47 -0400422
423 ret = trace_seq_printf(s, "\tfield: local_t commit;\t"
Tom Zanussi26a50742009-10-06 01:09:50 -0500424 "offset:%u;\tsize:%u;\tsigned:%u;\n",
Steven Rostedtd1b182a2009-04-15 16:53:47 -0400425 (unsigned int)offsetof(typeof(field), commit),
Tom Zanussi26a50742009-10-06 01:09:50 -0500426 (unsigned int)sizeof(field.commit),
427 (unsigned int)is_signed_type(long));
Steven Rostedtd1b182a2009-04-15 16:53:47 -0400428
Steven Rostedt66a8cb92010-03-31 13:21:56 -0400429 ret = trace_seq_printf(s, "\tfield: int overwrite;\t"
430 "offset:%u;\tsize:%u;\tsigned:%u;\n",
431 (unsigned int)offsetof(typeof(field), commit),
432 1,
433 (unsigned int)is_signed_type(long));
434
Steven Rostedtd1b182a2009-04-15 16:53:47 -0400435 ret = trace_seq_printf(s, "\tfield: char data;\t"
Tom Zanussi26a50742009-10-06 01:09:50 -0500436 "offset:%u;\tsize:%u;\tsigned:%u;\n",
Steven Rostedtd1b182a2009-04-15 16:53:47 -0400437 (unsigned int)offsetof(typeof(field), data),
Tom Zanussi26a50742009-10-06 01:09:50 -0500438 (unsigned int)BUF_PAGE_SIZE,
439 (unsigned int)is_signed_type(char));
Steven Rostedtd1b182a2009-04-15 16:53:47 -0400440
441 return ret;
442}
443
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400444/*
445 * head_page == tail_page && head == tail then buffer is empty.
446 */
447struct ring_buffer_per_cpu {
448 int cpu;
Richard Kennedy985023d2010-03-25 11:27:36 +0000449 atomic_t record_disabled;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400450 struct ring_buffer *buffer;
Thomas Gleixner5389f6f2009-07-25 17:13:33 +0200451 raw_spinlock_t reader_lock; /* serialize readers */
Thomas Gleixner445c8952009-12-02 19:49:50 +0100452 arch_spinlock_t lock;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400453 struct lock_class_key lock_key;
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -0800454 unsigned int nr_pages;
Steven Rostedt3adc54f2009-03-30 15:32:01 -0400455 struct list_head *pages;
Steven Rostedt6f807ac2008-10-04 02:00:58 -0400456 struct buffer_page *head_page; /* read from head */
457 struct buffer_page *tail_page; /* write to tail */
Wenji Huangc3706f02009-02-10 01:03:18 -0500458 struct buffer_page *commit_page; /* committed pages */
Steven Rostedtd7690412008-10-01 00:29:53 -0400459 struct buffer_page *reader_page;
Steven Rostedt66a8cb92010-03-31 13:21:56 -0400460 unsigned long lost_events;
461 unsigned long last_overrun;
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -0700462 local_t entries_bytes;
Steven Rostedte4906ef2009-04-30 20:49:44 -0400463 local_t entries;
Slava Pestov884bfe82011-07-15 14:23:58 -0700464 local_t overrun;
465 local_t commit_overrun;
466 local_t dropped_events;
Steven Rostedtfa743952009-06-16 12:37:57 -0400467 local_t committing;
468 local_t commits;
Steven Rostedt77ae3652009-03-27 11:00:29 -0400469 unsigned long read;
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -0700470 unsigned long read_bytes;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400471 u64 write_stamp;
472 u64 read_stamp;
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -0800473 /* ring buffer pages to update, > 0 to add, < 0 to remove */
474 int nr_pages_to_update;
475 struct list_head new_pages; /* new pages to add */
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -0700476 struct work_struct update_pages_work;
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -0700477 struct completion update_done;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400478};
479
480struct ring_buffer {
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400481 unsigned flags;
482 int cpus;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400483 atomic_t record_disabled;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -0700484 atomic_t resize_disabled;
Arnaldo Carvalho de Melo00f62f62009-02-09 17:04:06 -0200485 cpumask_var_t cpumask;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400486
Peter Zijlstra1f8a6a12009-06-08 18:18:39 +0200487 struct lock_class_key *reader_lock_key;
488
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400489 struct mutex mutex;
490
491 struct ring_buffer_per_cpu **buffers;
Steven Rostedt554f7862009-03-11 22:00:13 -0400492
Steven Rostedt59222ef2009-03-12 11:46:03 -0400493#ifdef CONFIG_HOTPLUG_CPU
Steven Rostedt554f7862009-03-11 22:00:13 -0400494 struct notifier_block cpu_notify;
495#endif
Steven Rostedt37886f62009-03-17 17:22:06 -0400496 u64 (*clock)(void);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400497};
498
499struct ring_buffer_iter {
500 struct ring_buffer_per_cpu *cpu_buffer;
501 unsigned long head;
502 struct buffer_page *head_page;
Steven Rostedt492a74f2010-01-25 15:17:47 -0500503 struct buffer_page *cache_reader_page;
504 unsigned long cache_read;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400505 u64 read_stamp;
506};
507
Steven Rostedtf536aaf2008-11-10 23:07:30 -0500508/* buffer may be either ring_buffer or ring_buffer_per_cpu */
Steven Rostedt077c5402009-09-03 19:53:46 -0400509#define RB_WARN_ON(b, cond) \
510 ({ \
511 int _____ret = unlikely(cond); \
512 if (_____ret) { \
513 if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
514 struct ring_buffer_per_cpu *__b = \
515 (void *)b; \
516 atomic_inc(&__b->buffer->record_disabled); \
517 } else \
518 atomic_inc(&b->record_disabled); \
519 WARN_ON(1); \
520 } \
521 _____ret; \
Steven Rostedt3e89c7b2008-11-11 15:28:41 -0500522 })
Steven Rostedtf536aaf2008-11-10 23:07:30 -0500523
Steven Rostedt37886f62009-03-17 17:22:06 -0400524/* Up this if you want to test the TIME_EXTENTS and normalization */
525#define DEBUG_SHIFT 0
526
Jiri Olsa6d3f1e12009-10-23 19:36:19 -0400527static inline u64 rb_time_stamp(struct ring_buffer *buffer)
Steven Rostedt88eb0122009-05-11 16:28:23 -0400528{
529 /* shift to debug/test normalization and TIME_EXTENTS */
530 return buffer->clock() << DEBUG_SHIFT;
531}
532
Steven Rostedt37886f62009-03-17 17:22:06 -0400533u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
534{
535 u64 time;
536
537 preempt_disable_notrace();
Jiri Olsa6d3f1e12009-10-23 19:36:19 -0400538 time = rb_time_stamp(buffer);
Steven Rostedt37886f62009-03-17 17:22:06 -0400539 preempt_enable_no_resched_notrace();
540
541 return time;
542}
543EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
544
545void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer,
546 int cpu, u64 *ts)
547{
548 /* Just stupid testing the normalize function and deltas */
549 *ts >>= DEBUG_SHIFT;
550}
551EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
552
Steven Rostedt77ae3652009-03-27 11:00:29 -0400553/*
554 * Making the ring buffer lockless makes things tricky.
555 * Although writes only happen on the CPU that they are on,
556 * and they only need to worry about interrupts. Reads can
557 * happen on any CPU.
558 *
559 * The reader page is always off the ring buffer, but when the
560 * reader finishes with a page, it needs to swap its page with
561 * a new one from the buffer. The reader needs to take from
562 * the head (writes go to the tail). But if a writer is in overwrite
563 * mode and wraps, it must push the head page forward.
564 *
565 * Here lies the problem.
566 *
567 * The reader must be careful to replace only the head page, and
568 * not another one. As described at the top of the file in the
569 * ASCII art, the reader sets its old page to point to the next
570 * page after head. It then sets the page after head to point to
571 * the old reader page. But if the writer moves the head page
572 * during this operation, the reader could end up with the tail.
573 *
574 * We use cmpxchg to help prevent this race. We also do something
575 * special with the page before head. We set the LSB to 1.
576 *
577 * When the writer must push the page forward, it will clear the
578 * bit that points to the head page, move the head, and then set
579 * the bit that points to the new head page.
580 *
581 * We also don't want an interrupt coming in and moving the head
582 * page on another writer. Thus we use the second LSB to catch
583 * that too. Thus:
584 *
585 * head->list->prev->next bit 1 bit 0
586 * ------- -------
587 * Normal page 0 0
588 * Points to head page 0 1
589 * New head page 1 0
590 *
591 * Note we can not trust the prev pointer of the head page, because:
592 *
593 * +----+ +-----+ +-----+
594 * | |------>| T |---X--->| N |
595 * | |<------| | | |
596 * +----+ +-----+ +-----+
597 * ^ ^ |
598 * | +-----+ | |
599 * +----------| R |----------+ |
600 * | |<-----------+
601 * +-----+
602 *
603 * Key: ---X--> HEAD flag set in pointer
604 * T Tail page
605 * R Reader page
606 * N Next page
607 *
608 * (see __rb_reserve_next() to see where this happens)
609 *
610 * What the above shows is that the reader just swapped out
611 * the reader page with a page in the buffer, but before it
612 * could make the new header point back to the new page added
613 * it was preempted by a writer. The writer moved forward onto
614 * the new page added by the reader and is about to move forward
615 * again.
616 *
617 * You can see, it is legitimate for the previous pointer of
618 * the head (or any page) not to point back to itself. But only
619 * temporarially.
620 */
621
622#define RB_PAGE_NORMAL 0UL
623#define RB_PAGE_HEAD 1UL
624#define RB_PAGE_UPDATE 2UL
625
626
627#define RB_FLAG_MASK 3UL
628
629/* PAGE_MOVED is not part of the mask */
630#define RB_PAGE_MOVED 4UL
631
632/*
633 * rb_list_head - remove any bit
634 */
635static struct list_head *rb_list_head(struct list_head *list)
636{
637 unsigned long val = (unsigned long)list;
638
639 return (struct list_head *)(val & ~RB_FLAG_MASK);
640}
641
642/*
Jiri Olsa6d3f1e12009-10-23 19:36:19 -0400643 * rb_is_head_page - test if the given page is the head page
Steven Rostedt77ae3652009-03-27 11:00:29 -0400644 *
645 * Because the reader may move the head_page pointer, we can
646 * not trust what the head page is (it may be pointing to
647 * the reader page). But if the next page is a header page,
648 * its flags will be non zero.
649 */
Jesper Juhl42b16b32011-01-17 00:09:38 +0100650static inline int
Steven Rostedt77ae3652009-03-27 11:00:29 -0400651rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer,
652 struct buffer_page *page, struct list_head *list)
653{
654 unsigned long val;
655
656 val = (unsigned long)list->next;
657
658 if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
659 return RB_PAGE_MOVED;
660
661 return val & RB_FLAG_MASK;
662}
663
664/*
665 * rb_is_reader_page
666 *
667 * The unique thing about the reader page, is that, if the
668 * writer is ever on it, the previous pointer never points
669 * back to the reader page.
670 */
671static int rb_is_reader_page(struct buffer_page *page)
672{
673 struct list_head *list = page->list.prev;
674
675 return rb_list_head(list->next) != &page->list;
676}
677
678/*
679 * rb_set_list_to_head - set a list_head to be pointing to head.
680 */
681static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer,
682 struct list_head *list)
683{
684 unsigned long *ptr;
685
686 ptr = (unsigned long *)&list->next;
687 *ptr |= RB_PAGE_HEAD;
688 *ptr &= ~RB_PAGE_UPDATE;
689}
690
691/*
692 * rb_head_page_activate - sets up head page
693 */
694static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
695{
696 struct buffer_page *head;
697
698 head = cpu_buffer->head_page;
699 if (!head)
700 return;
701
702 /*
703 * Set the previous list pointer to have the HEAD flag.
704 */
705 rb_set_list_to_head(cpu_buffer, head->list.prev);
706}
707
708static void rb_list_head_clear(struct list_head *list)
709{
710 unsigned long *ptr = (unsigned long *)&list->next;
711
712 *ptr &= ~RB_FLAG_MASK;
713}
714
715/*
716 * rb_head_page_dactivate - clears head page ptr (for free list)
717 */
718static void
719rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
720{
721 struct list_head *hd;
722
723 /* Go through the whole list and clear any pointers found. */
724 rb_list_head_clear(cpu_buffer->pages);
725
726 list_for_each(hd, cpu_buffer->pages)
727 rb_list_head_clear(hd);
728}
729
730static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
731 struct buffer_page *head,
732 struct buffer_page *prev,
733 int old_flag, int new_flag)
734{
735 struct list_head *list;
736 unsigned long val = (unsigned long)&head->list;
737 unsigned long ret;
738
739 list = &prev->list;
740
741 val &= ~RB_FLAG_MASK;
742
Steven Rostedt08a40812009-09-14 09:31:35 -0400743 ret = cmpxchg((unsigned long *)&list->next,
744 val | old_flag, val | new_flag);
Steven Rostedt77ae3652009-03-27 11:00:29 -0400745
746 /* check if the reader took the page */
747 if ((ret & ~RB_FLAG_MASK) != val)
748 return RB_PAGE_MOVED;
749
750 return ret & RB_FLAG_MASK;
751}
752
753static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
754 struct buffer_page *head,
755 struct buffer_page *prev,
756 int old_flag)
757{
758 return rb_head_page_set(cpu_buffer, head, prev,
759 old_flag, RB_PAGE_UPDATE);
760}
761
762static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
763 struct buffer_page *head,
764 struct buffer_page *prev,
765 int old_flag)
766{
767 return rb_head_page_set(cpu_buffer, head, prev,
768 old_flag, RB_PAGE_HEAD);
769}
770
771static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
772 struct buffer_page *head,
773 struct buffer_page *prev,
774 int old_flag)
775{
776 return rb_head_page_set(cpu_buffer, head, prev,
777 old_flag, RB_PAGE_NORMAL);
778}
779
780static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
781 struct buffer_page **bpage)
782{
783 struct list_head *p = rb_list_head((*bpage)->list.next);
784
785 *bpage = list_entry(p, struct buffer_page, list);
786}
787
788static struct buffer_page *
789rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
790{
791 struct buffer_page *head;
792 struct buffer_page *page;
793 struct list_head *list;
794 int i;
795
796 if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
797 return NULL;
798
799 /* sanity check */
800 list = cpu_buffer->pages;
801 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
802 return NULL;
803
804 page = head = cpu_buffer->head_page;
805 /*
806 * It is possible that the writer moves the header behind
807 * where we started, and we miss in one loop.
808 * A second loop should grab the header, but we'll do
809 * three loops just because I'm paranoid.
810 */
811 for (i = 0; i < 3; i++) {
812 do {
813 if (rb_is_head_page(cpu_buffer, page, page->list.prev)) {
814 cpu_buffer->head_page = page;
815 return page;
816 }
817 rb_inc_page(cpu_buffer, &page);
818 } while (page != head);
819 }
820
821 RB_WARN_ON(cpu_buffer, 1);
822
823 return NULL;
824}
825
826static int rb_head_page_replace(struct buffer_page *old,
827 struct buffer_page *new)
828{
829 unsigned long *ptr = (unsigned long *)&old->list.prev->next;
830 unsigned long val;
831 unsigned long ret;
832
833 val = *ptr & ~RB_FLAG_MASK;
834 val |= RB_PAGE_HEAD;
835
Steven Rostedt08a40812009-09-14 09:31:35 -0400836 ret = cmpxchg(ptr, val, (unsigned long)&new->list);
Steven Rostedt77ae3652009-03-27 11:00:29 -0400837
838 return ret == val;
839}
840
841/*
842 * rb_tail_page_update - move the tail page forward
843 *
844 * Returns 1 if moved tail page, 0 if someone else did.
845 */
846static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
847 struct buffer_page *tail_page,
848 struct buffer_page *next_page)
849{
850 struct buffer_page *old_tail;
851 unsigned long old_entries;
852 unsigned long old_write;
853 int ret = 0;
854
855 /*
856 * The tail page now needs to be moved forward.
857 *
858 * We need to reset the tail page, but without messing
859 * with possible erasing of data brought in by interrupts
860 * that have moved the tail page and are currently on it.
861 *
862 * We add a counter to the write field to denote this.
863 */
864 old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write);
865 old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries);
866
867 /*
868 * Just make sure we have seen our old_write and synchronize
869 * with any interrupts that come in.
870 */
871 barrier();
872
873 /*
874 * If the tail page is still the same as what we think
875 * it is, then it is up to us to update the tail
876 * pointer.
877 */
878 if (tail_page == cpu_buffer->tail_page) {
879 /* Zero the write counter */
880 unsigned long val = old_write & ~RB_WRITE_MASK;
881 unsigned long eval = old_entries & ~RB_WRITE_MASK;
882
883 /*
884 * This will only succeed if an interrupt did
885 * not come in and change it. In which case, we
886 * do not want to modify it.
Lai Jiangshanda706d82009-07-15 16:27:30 +0800887 *
888 * We add (void) to let the compiler know that we do not care
889 * about the return value of these functions. We use the
890 * cmpxchg to only update if an interrupt did not already
891 * do it for us. If the cmpxchg fails, we don't care.
Steven Rostedt77ae3652009-03-27 11:00:29 -0400892 */
Lai Jiangshanda706d82009-07-15 16:27:30 +0800893 (void)local_cmpxchg(&next_page->write, old_write, val);
894 (void)local_cmpxchg(&next_page->entries, old_entries, eval);
Steven Rostedt77ae3652009-03-27 11:00:29 -0400895
896 /*
897 * No need to worry about races with clearing out the commit.
898 * it only can increment when a commit takes place. But that
899 * only happens in the outer most nested commit.
900 */
901 local_set(&next_page->page->commit, 0);
902
903 old_tail = cmpxchg(&cpu_buffer->tail_page,
904 tail_page, next_page);
905
906 if (old_tail == tail_page)
907 ret = 1;
908 }
909
910 return ret;
911}
912
913static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
914 struct buffer_page *bpage)
915{
916 unsigned long val = (unsigned long)bpage;
917
918 if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK))
919 return 1;
920
921 return 0;
922}
923
924/**
925 * rb_check_list - make sure a pointer to a list has the last bits zero
926 */
927static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer,
928 struct list_head *list)
929{
930 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev))
931 return 1;
932 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next))
933 return 1;
934 return 0;
935}
936
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400937/**
938 * check_pages - integrity check of buffer pages
939 * @cpu_buffer: CPU buffer with pages to test
940 *
Wenji Huangc3706f02009-02-10 01:03:18 -0500941 * As a safety measure we check to make sure the data pages have not
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400942 * been corrupted.
943 */
944static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
945{
Steven Rostedt3adc54f2009-03-30 15:32:01 -0400946 struct list_head *head = cpu_buffer->pages;
Steven Rostedt044fa782008-12-02 23:50:03 -0500947 struct buffer_page *bpage, *tmp;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400948
Steven Rostedt308f7ee2012-05-16 19:46:32 -0400949 /* Reset the head page if it exists */
950 if (cpu_buffer->head_page)
951 rb_set_head_page(cpu_buffer);
952
Steven Rostedt77ae3652009-03-27 11:00:29 -0400953 rb_head_page_deactivate(cpu_buffer);
954
Steven Rostedt3e89c7b2008-11-11 15:28:41 -0500955 if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
956 return -1;
957 if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
958 return -1;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400959
Steven Rostedt77ae3652009-03-27 11:00:29 -0400960 if (rb_check_list(cpu_buffer, head))
961 return -1;
962
Steven Rostedt044fa782008-12-02 23:50:03 -0500963 list_for_each_entry_safe(bpage, tmp, head, list) {
Steven Rostedt3e89c7b2008-11-11 15:28:41 -0500964 if (RB_WARN_ON(cpu_buffer,
Steven Rostedt044fa782008-12-02 23:50:03 -0500965 bpage->list.next->prev != &bpage->list))
Steven Rostedt3e89c7b2008-11-11 15:28:41 -0500966 return -1;
967 if (RB_WARN_ON(cpu_buffer,
Steven Rostedt044fa782008-12-02 23:50:03 -0500968 bpage->list.prev->next != &bpage->list))
Steven Rostedt3e89c7b2008-11-11 15:28:41 -0500969 return -1;
Steven Rostedt77ae3652009-03-27 11:00:29 -0400970 if (rb_check_list(cpu_buffer, &bpage->list))
971 return -1;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400972 }
973
Steven Rostedt77ae3652009-03-27 11:00:29 -0400974 rb_head_page_activate(cpu_buffer);
975
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400976 return 0;
977}
978
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -0800979static int __rb_allocate_pages(int nr_pages, struct list_head *pages, int cpu)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400980{
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -0800981 int i;
Steven Rostedt044fa782008-12-02 23:50:03 -0500982 struct buffer_page *bpage, *tmp;
Steven Rostedt3adc54f2009-03-30 15:32:01 -0400983
Steven Rostedt7a8e76a2008-09-29 23:02:38 -0400984 for (i = 0; i < nr_pages; i++) {
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -0700985 struct page *page;
Vaibhav Nagarnaikd7ec4bf2011-06-07 17:01:42 -0700986 /*
987 * __GFP_NORETRY flag makes sure that the allocation fails
988 * gracefully without invoking oom-killer and the system is
989 * not destabilized.
990 */
Steven Rostedt044fa782008-12-02 23:50:03 -0500991 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
Vaibhav Nagarnaikd7ec4bf2011-06-07 17:01:42 -0700992 GFP_KERNEL | __GFP_NORETRY,
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -0800993 cpu_to_node(cpu));
Steven Rostedt044fa782008-12-02 23:50:03 -0500994 if (!bpage)
Steven Rostedte4c2ce82008-10-01 11:14:54 -0400995 goto free_pages;
Steven Rostedt77ae3652009-03-27 11:00:29 -0400996
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -0800997 list_add(&bpage->list, pages);
Steven Rostedt77ae3652009-03-27 11:00:29 -0400998
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -0800999 page = alloc_pages_node(cpu_to_node(cpu),
Vaibhav Nagarnaikd7ec4bf2011-06-07 17:01:42 -07001000 GFP_KERNEL | __GFP_NORETRY, 0);
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07001001 if (!page)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001002 goto free_pages;
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07001003 bpage->page = page_address(page);
Steven Rostedt044fa782008-12-02 23:50:03 -05001004 rb_init_page(bpage->page);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001005 }
1006
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001007 return 0;
1008
1009free_pages:
1010 list_for_each_entry_safe(bpage, tmp, pages, list) {
1011 list_del_init(&bpage->list);
1012 free_buffer_page(bpage);
1013 }
1014
1015 return -ENOMEM;
1016}
1017
1018static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
1019 unsigned nr_pages)
1020{
1021 LIST_HEAD(pages);
1022
1023 WARN_ON(!nr_pages);
1024
1025 if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu))
1026 return -ENOMEM;
1027
Steven Rostedt3adc54f2009-03-30 15:32:01 -04001028 /*
1029 * The ring buffer page list is a circular list that does not
1030 * start and end with a list head. All page list items point to
1031 * other pages.
1032 */
1033 cpu_buffer->pages = pages.next;
1034 list_del(&pages);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001035
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001036 cpu_buffer->nr_pages = nr_pages;
1037
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001038 rb_check_pages(cpu_buffer);
1039
1040 return 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001041}
1042
1043static struct ring_buffer_per_cpu *
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001044rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001045{
1046 struct ring_buffer_per_cpu *cpu_buffer;
Steven Rostedt044fa782008-12-02 23:50:03 -05001047 struct buffer_page *bpage;
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07001048 struct page *page;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001049 int ret;
1050
1051 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
1052 GFP_KERNEL, cpu_to_node(cpu));
1053 if (!cpu_buffer)
1054 return NULL;
1055
1056 cpu_buffer->cpu = cpu;
1057 cpu_buffer->buffer = buffer;
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02001058 raw_spin_lock_init(&cpu_buffer->reader_lock);
Peter Zijlstra1f8a6a12009-06-08 18:18:39 +02001059 lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
Thomas Gleixneredc35bd2009-12-03 12:38:57 +01001060 cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001061 INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler);
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001062 init_completion(&cpu_buffer->update_done);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001063
Steven Rostedt044fa782008-12-02 23:50:03 -05001064 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
Steven Rostedte4c2ce82008-10-01 11:14:54 -04001065 GFP_KERNEL, cpu_to_node(cpu));
Steven Rostedt044fa782008-12-02 23:50:03 -05001066 if (!bpage)
Steven Rostedte4c2ce82008-10-01 11:14:54 -04001067 goto fail_free_buffer;
1068
Steven Rostedt77ae3652009-03-27 11:00:29 -04001069 rb_check_bpage(cpu_buffer, bpage);
1070
Steven Rostedt044fa782008-12-02 23:50:03 -05001071 cpu_buffer->reader_page = bpage;
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07001072 page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0);
1073 if (!page)
Steven Rostedte4c2ce82008-10-01 11:14:54 -04001074 goto fail_free_reader;
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07001075 bpage->page = page_address(page);
Steven Rostedt044fa782008-12-02 23:50:03 -05001076 rb_init_page(bpage->page);
Steven Rostedte4c2ce82008-10-01 11:14:54 -04001077
Steven Rostedtd7690412008-10-01 00:29:53 -04001078 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
Vaibhav Nagarnaik44b99462012-06-22 11:50:05 -07001079 INIT_LIST_HEAD(&cpu_buffer->new_pages);
Steven Rostedtd7690412008-10-01 00:29:53 -04001080
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001081 ret = rb_allocate_pages(cpu_buffer, nr_pages);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001082 if (ret < 0)
Steven Rostedtd7690412008-10-01 00:29:53 -04001083 goto fail_free_reader;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001084
1085 cpu_buffer->head_page
Steven Rostedt3adc54f2009-03-30 15:32:01 -04001086 = list_entry(cpu_buffer->pages, struct buffer_page, list);
Steven Rostedtbf41a152008-10-04 02:00:59 -04001087 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001088
Steven Rostedt77ae3652009-03-27 11:00:29 -04001089 rb_head_page_activate(cpu_buffer);
1090
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001091 return cpu_buffer;
1092
Steven Rostedtd7690412008-10-01 00:29:53 -04001093 fail_free_reader:
1094 free_buffer_page(cpu_buffer->reader_page);
1095
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001096 fail_free_buffer:
1097 kfree(cpu_buffer);
1098 return NULL;
1099}
1100
1101static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
1102{
Steven Rostedt3adc54f2009-03-30 15:32:01 -04001103 struct list_head *head = cpu_buffer->pages;
Steven Rostedt044fa782008-12-02 23:50:03 -05001104 struct buffer_page *bpage, *tmp;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001105
Steven Rostedtd7690412008-10-01 00:29:53 -04001106 free_buffer_page(cpu_buffer->reader_page);
1107
Steven Rostedt77ae3652009-03-27 11:00:29 -04001108 rb_head_page_deactivate(cpu_buffer);
1109
Steven Rostedt3adc54f2009-03-30 15:32:01 -04001110 if (head) {
1111 list_for_each_entry_safe(bpage, tmp, head, list) {
1112 list_del_init(&bpage->list);
1113 free_buffer_page(bpage);
1114 }
1115 bpage = list_entry(head, struct buffer_page, list);
Steven Rostedt044fa782008-12-02 23:50:03 -05001116 free_buffer_page(bpage);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001117 }
Steven Rostedt3adc54f2009-03-30 15:32:01 -04001118
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001119 kfree(cpu_buffer);
1120}
1121
Steven Rostedt59222ef2009-03-12 11:46:03 -04001122#ifdef CONFIG_HOTPLUG_CPU
Frederic Weisbecker09c9e842009-03-21 04:33:36 +01001123static int rb_cpu_notify(struct notifier_block *self,
1124 unsigned long action, void *hcpu);
Steven Rostedt554f7862009-03-11 22:00:13 -04001125#endif
1126
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001127/**
1128 * ring_buffer_alloc - allocate a new ring_buffer
Robert Richter68814b52008-11-24 12:24:12 +01001129 * @size: the size in bytes per cpu that is needed.
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001130 * @flags: attributes to set for the ring buffer.
1131 *
1132 * Currently the only flag that is available is the RB_FL_OVERWRITE
1133 * flag. This flag means that the buffer will overwrite old data
1134 * when the buffer wraps. If this flag is not set, the buffer will
1135 * drop data when the tail hits the head.
1136 */
Peter Zijlstra1f8a6a12009-06-08 18:18:39 +02001137struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
1138 struct lock_class_key *key)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001139{
1140 struct ring_buffer *buffer;
1141 int bsize;
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001142 int cpu, nr_pages;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001143
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001144 /* keep it in its own cache line */
1145 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
1146 GFP_KERNEL);
1147 if (!buffer)
1148 return NULL;
1149
Rusty Russell9e01c1b2009-01-01 10:12:22 +10301150 if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
1151 goto fail_free_buffer;
1152
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001153 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001154 buffer->flags = flags;
Steven Rostedt37886f62009-03-17 17:22:06 -04001155 buffer->clock = trace_clock_local;
Peter Zijlstra1f8a6a12009-06-08 18:18:39 +02001156 buffer->reader_lock_key = key;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001157
1158 /* need at least two pages */
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001159 if (nr_pages < 2)
1160 nr_pages = 2;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001161
Frederic Weisbecker3bf832c2009-03-19 14:47:33 +01001162 /*
1163 * In case of non-hotplug cpu, if the ring-buffer is allocated
1164 * in early initcall, it will not be notified of secondary cpus.
1165 * In that off case, we need to allocate for all possible cpus.
1166 */
1167#ifdef CONFIG_HOTPLUG_CPU
Steven Rostedt554f7862009-03-11 22:00:13 -04001168 get_online_cpus();
1169 cpumask_copy(buffer->cpumask, cpu_online_mask);
Frederic Weisbecker3bf832c2009-03-19 14:47:33 +01001170#else
1171 cpumask_copy(buffer->cpumask, cpu_possible_mask);
1172#endif
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001173 buffer->cpus = nr_cpu_ids;
1174
1175 bsize = sizeof(void *) * nr_cpu_ids;
1176 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
1177 GFP_KERNEL);
1178 if (!buffer->buffers)
Rusty Russell9e01c1b2009-01-01 10:12:22 +10301179 goto fail_free_cpumask;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001180
1181 for_each_buffer_cpu(buffer, cpu) {
1182 buffer->buffers[cpu] =
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001183 rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001184 if (!buffer->buffers[cpu])
1185 goto fail_free_buffers;
1186 }
1187
Steven Rostedt59222ef2009-03-12 11:46:03 -04001188#ifdef CONFIG_HOTPLUG_CPU
Steven Rostedt554f7862009-03-11 22:00:13 -04001189 buffer->cpu_notify.notifier_call = rb_cpu_notify;
1190 buffer->cpu_notify.priority = 0;
1191 register_cpu_notifier(&buffer->cpu_notify);
1192#endif
1193
1194 put_online_cpus();
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001195 mutex_init(&buffer->mutex);
1196
1197 return buffer;
1198
1199 fail_free_buffers:
1200 for_each_buffer_cpu(buffer, cpu) {
1201 if (buffer->buffers[cpu])
1202 rb_free_cpu_buffer(buffer->buffers[cpu]);
1203 }
1204 kfree(buffer->buffers);
1205
Rusty Russell9e01c1b2009-01-01 10:12:22 +10301206 fail_free_cpumask:
1207 free_cpumask_var(buffer->cpumask);
Steven Rostedt554f7862009-03-11 22:00:13 -04001208 put_online_cpus();
Rusty Russell9e01c1b2009-01-01 10:12:22 +10301209
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001210 fail_free_buffer:
1211 kfree(buffer);
1212 return NULL;
1213}
Peter Zijlstra1f8a6a12009-06-08 18:18:39 +02001214EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001215
1216/**
1217 * ring_buffer_free - free a ring buffer.
1218 * @buffer: the buffer to free.
1219 */
1220void
1221ring_buffer_free(struct ring_buffer *buffer)
1222{
1223 int cpu;
1224
Steven Rostedt554f7862009-03-11 22:00:13 -04001225 get_online_cpus();
1226
Steven Rostedt59222ef2009-03-12 11:46:03 -04001227#ifdef CONFIG_HOTPLUG_CPU
Steven Rostedt554f7862009-03-11 22:00:13 -04001228 unregister_cpu_notifier(&buffer->cpu_notify);
1229#endif
1230
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001231 for_each_buffer_cpu(buffer, cpu)
1232 rb_free_cpu_buffer(buffer->buffers[cpu]);
1233
Steven Rostedt554f7862009-03-11 22:00:13 -04001234 put_online_cpus();
1235
Eric Dumazetbd3f0222009-08-07 12:49:29 +02001236 kfree(buffer->buffers);
Rusty Russell9e01c1b2009-01-01 10:12:22 +10301237 free_cpumask_var(buffer->cpumask);
1238
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001239 kfree(buffer);
1240}
Robert Richterc4f50182008-12-11 16:49:22 +01001241EXPORT_SYMBOL_GPL(ring_buffer_free);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001242
Steven Rostedt37886f62009-03-17 17:22:06 -04001243void ring_buffer_set_clock(struct ring_buffer *buffer,
1244 u64 (*clock)(void))
1245{
1246 buffer->clock = clock;
1247}
1248
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001249static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
1250
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001251static inline unsigned long rb_page_entries(struct buffer_page *bpage)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001252{
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001253 return local_read(&bpage->entries) & RB_WRITE_MASK;
1254}
1255
1256static inline unsigned long rb_page_write(struct buffer_page *bpage)
1257{
1258 return local_read(&bpage->write) & RB_WRITE_MASK;
1259}
1260
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001261static int
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001262rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages)
1263{
1264 struct list_head *tail_page, *to_remove, *next_page;
1265 struct buffer_page *to_remove_page, *tmp_iter_page;
1266 struct buffer_page *last_page, *first_page;
1267 unsigned int nr_removed;
1268 unsigned long head_bit;
1269 int page_entries;
1270
1271 head_bit = 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001272
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02001273 raw_spin_lock_irq(&cpu_buffer->reader_lock);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001274 atomic_inc(&cpu_buffer->record_disabled);
1275 /*
1276 * We don't race with the readers since we have acquired the reader
1277 * lock. We also don't race with writers after disabling recording.
1278 * This makes it easy to figure out the first and the last page to be
1279 * removed from the list. We unlink all the pages in between including
1280 * the first and last pages. This is done in a busy loop so that we
1281 * lose the least number of traces.
1282 * The pages are freed after we restart recording and unlock readers.
1283 */
1284 tail_page = &cpu_buffer->tail_page->list;
Steven Rostedt77ae3652009-03-27 11:00:29 -04001285
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001286 /*
1287 * tail page might be on reader page, we remove the next page
1288 * from the ring buffer
1289 */
1290 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
1291 tail_page = rb_list_head(tail_page->next);
1292 to_remove = tail_page;
1293
1294 /* start of pages to remove */
1295 first_page = list_entry(rb_list_head(to_remove->next),
1296 struct buffer_page, list);
1297
1298 for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) {
1299 to_remove = rb_list_head(to_remove)->next;
1300 head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001301 }
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001302
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001303 next_page = rb_list_head(to_remove)->next;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001304
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001305 /*
1306 * Now we remove all pages between tail_page and next_page.
1307 * Make sure that we have head_bit value preserved for the
1308 * next page
1309 */
1310 tail_page->next = (struct list_head *)((unsigned long)next_page |
1311 head_bit);
1312 next_page = rb_list_head(next_page);
1313 next_page->prev = tail_page;
1314
1315 /* make sure pages points to a valid page in the ring buffer */
1316 cpu_buffer->pages = next_page;
1317
1318 /* update head page */
1319 if (head_bit)
1320 cpu_buffer->head_page = list_entry(next_page,
1321 struct buffer_page, list);
1322
1323 /*
1324 * change read pointer to make sure any read iterators reset
1325 * themselves
1326 */
1327 cpu_buffer->read = 0;
1328
1329 /* pages are removed, resume tracing and then free the pages */
1330 atomic_dec(&cpu_buffer->record_disabled);
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02001331 raw_spin_unlock_irq(&cpu_buffer->reader_lock);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001332
1333 RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages));
1334
1335 /* last buffer page to remove */
1336 last_page = list_entry(rb_list_head(to_remove), struct buffer_page,
1337 list);
1338 tmp_iter_page = first_page;
1339
1340 do {
1341 to_remove_page = tmp_iter_page;
1342 rb_inc_page(cpu_buffer, &tmp_iter_page);
1343
1344 /* update the counters */
1345 page_entries = rb_page_entries(to_remove_page);
1346 if (page_entries) {
1347 /*
1348 * If something was added to this page, it was full
1349 * since it is not the tail page. So we deduct the
1350 * bytes consumed in ring buffer from here.
Vaibhav Nagarnaik48fdc722012-06-29 12:31:41 -07001351 * Increment overrun to account for the lost events.
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001352 */
Vaibhav Nagarnaik48fdc722012-06-29 12:31:41 -07001353 local_add(page_entries, &cpu_buffer->overrun);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001354 local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
1355 }
1356
1357 /*
1358 * We have already removed references to this list item, just
1359 * free up the buffer_page and its page
1360 */
1361 free_buffer_page(to_remove_page);
1362 nr_removed--;
1363
1364 } while (to_remove_page != last_page);
1365
1366 RB_WARN_ON(cpu_buffer, nr_removed);
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001367
1368 return nr_removed == 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001369}
1370
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001371static int
1372rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001373{
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001374 struct list_head *pages = &cpu_buffer->new_pages;
1375 int retries, success;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001376
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02001377 raw_spin_lock_irq(&cpu_buffer->reader_lock);
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001378 /*
1379 * We are holding the reader lock, so the reader page won't be swapped
1380 * in the ring buffer. Now we are racing with the writer trying to
1381 * move head page and the tail page.
1382 * We are going to adapt the reader page update process where:
1383 * 1. We first splice the start and end of list of new pages between
1384 * the head page and its previous page.
1385 * 2. We cmpxchg the prev_page->next to point from head page to the
1386 * start of new pages list.
1387 * 3. Finally, we update the head->prev to the end of new list.
1388 *
1389 * We will try this process 10 times, to make sure that we don't keep
1390 * spinning.
1391 */
1392 retries = 10;
1393 success = 0;
1394 while (retries--) {
1395 struct list_head *head_page, *prev_page, *r;
1396 struct list_head *last_page, *first_page;
1397 struct list_head *head_page_with_bit;
Steven Rostedt77ae3652009-03-27 11:00:29 -04001398
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001399 head_page = &rb_set_head_page(cpu_buffer)->list;
1400 prev_page = head_page->prev;
1401
1402 first_page = pages->next;
1403 last_page = pages->prev;
1404
1405 head_page_with_bit = (struct list_head *)
1406 ((unsigned long)head_page | RB_PAGE_HEAD);
1407
1408 last_page->next = head_page_with_bit;
1409 first_page->prev = prev_page;
1410
1411 r = cmpxchg(&prev_page->next, head_page_with_bit, first_page);
1412
1413 if (r == head_page_with_bit) {
1414 /*
1415 * yay, we replaced the page pointer to our new list,
1416 * now, we just have to update to head page's prev
1417 * pointer to point to end of list
1418 */
1419 head_page->prev = last_page;
1420 success = 1;
1421 break;
1422 }
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001423 }
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001424
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001425 if (success)
1426 INIT_LIST_HEAD(pages);
1427 /*
1428 * If we weren't successful in adding in new pages, warn and stop
1429 * tracing
1430 */
1431 RB_WARN_ON(cpu_buffer, !success);
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02001432 raw_spin_unlock_irq(&cpu_buffer->reader_lock);
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001433
1434 /* free pages if they weren't inserted */
1435 if (!success) {
1436 struct buffer_page *bpage, *tmp;
1437 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
1438 list) {
1439 list_del_init(&bpage->list);
1440 free_buffer_page(bpage);
1441 }
1442 }
1443 return success;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001444}
1445
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001446static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer)
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001447{
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001448 int success;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001449
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07001450 if (cpu_buffer->nr_pages_to_update > 0)
1451 success = rb_insert_pages(cpu_buffer);
1452 else
1453 success = rb_remove_pages(cpu_buffer,
1454 -cpu_buffer->nr_pages_to_update);
1455
1456 if (success)
1457 cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001458}
1459
1460static void update_pages_handler(struct work_struct *work)
1461{
1462 struct ring_buffer_per_cpu *cpu_buffer = container_of(work,
1463 struct ring_buffer_per_cpu, update_pages_work);
1464 rb_update_pages(cpu_buffer);
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001465 complete(&cpu_buffer->update_done);
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001466}
1467
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001468/**
1469 * ring_buffer_resize - resize the ring buffer
1470 * @buffer: the buffer to resize.
1471 * @size: the new size.
1472 *
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001473 * Minimum size is 2 * BUF_PAGE_SIZE.
1474 *
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001475 * Returns 0 on success and < 0 on failure.
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001476 */
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001477int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size,
1478 int cpu_id)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001479{
1480 struct ring_buffer_per_cpu *cpu_buffer;
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001481 unsigned nr_pages;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001482 int cpu, err = 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001483
Ingo Molnaree51a1d2008-11-13 14:58:31 +01001484 /*
1485 * Always succeed at resizing a non-existent buffer:
1486 */
1487 if (!buffer)
1488 return size;
1489
Steven Rostedt6a31e1f2012-05-23 15:35:17 -04001490 /* Make sure the requested buffer exists */
1491 if (cpu_id != RING_BUFFER_ALL_CPUS &&
1492 !cpumask_test_cpu(cpu_id, buffer->cpumask))
1493 return size;
1494
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001495 size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1496 size *= BUF_PAGE_SIZE;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001497
1498 /* we need a minimum of two pages */
1499 if (size < BUF_PAGE_SIZE * 2)
1500 size = BUF_PAGE_SIZE * 2;
1501
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001502 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1503
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001504 /*
1505 * Don't succeed if resizing is disabled, as a reader might be
1506 * manipulating the ring buffer and is expecting a sane state while
1507 * this is true.
1508 */
1509 if (atomic_read(&buffer->resize_disabled))
1510 return -EBUSY;
1511
1512 /* prevent another thread from changing buffer sizes */
1513 mutex_lock(&buffer->mutex);
1514
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001515 if (cpu_id == RING_BUFFER_ALL_CPUS) {
1516 /* calculate the pages to update */
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001517 for_each_buffer_cpu(buffer, cpu) {
1518 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001519
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001520 cpu_buffer->nr_pages_to_update = nr_pages -
1521 cpu_buffer->nr_pages;
Vaibhav Nagarnaikd7ec4bf2011-06-07 17:01:42 -07001522 /*
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001523 * nothing more to do for removing pages or no update
Vaibhav Nagarnaikd7ec4bf2011-06-07 17:01:42 -07001524 */
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001525 if (cpu_buffer->nr_pages_to_update <= 0)
1526 continue;
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001527 /*
1528 * to add pages, make sure all new pages can be
1529 * allocated without receiving ENOMEM
1530 */
1531 INIT_LIST_HEAD(&cpu_buffer->new_pages);
1532 if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update,
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001533 &cpu_buffer->new_pages, cpu)) {
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001534 /* not enough memory for new pages */
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001535 err = -ENOMEM;
1536 goto out_err;
1537 }
1538 }
1539
1540 get_online_cpus();
1541 /*
1542 * Fire off all the required work handlers
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001543 * We can't schedule on offline CPUs, but it's not necessary
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001544 * since we can change their buffer sizes without any race.
1545 */
1546 for_each_buffer_cpu(buffer, cpu) {
1547 cpu_buffer = buffer->buffers[cpu];
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001548 if (!cpu_buffer->nr_pages_to_update)
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001549 continue;
1550
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001551 if (cpu_online(cpu))
1552 schedule_work_on(cpu,
1553 &cpu_buffer->update_pages_work);
1554 else
1555 rb_update_pages(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001556 }
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001557
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001558 /* wait for all the updates to complete */
1559 for_each_buffer_cpu(buffer, cpu) {
1560 cpu_buffer = buffer->buffers[cpu];
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001561 if (!cpu_buffer->nr_pages_to_update)
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001562 continue;
1563
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001564 if (cpu_online(cpu))
1565 wait_for_completion(&cpu_buffer->update_done);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001566 cpu_buffer->nr_pages_to_update = 0;
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001567 }
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001568
1569 put_online_cpus();
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001570 } else {
Vaibhav Nagarnaik8e49f412012-10-10 16:40:27 -07001571 /* Make sure this CPU has been intitialized */
1572 if (!cpumask_test_cpu(cpu_id, buffer->cpumask))
1573 goto out;
1574
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001575 cpu_buffer = buffer->buffers[cpu_id];
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001576
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001577 if (nr_pages == cpu_buffer->nr_pages)
1578 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001579
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001580 cpu_buffer->nr_pages_to_update = nr_pages -
1581 cpu_buffer->nr_pages;
1582
1583 INIT_LIST_HEAD(&cpu_buffer->new_pages);
1584 if (cpu_buffer->nr_pages_to_update > 0 &&
1585 __rb_allocate_pages(cpu_buffer->nr_pages_to_update,
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001586 &cpu_buffer->new_pages, cpu_id)) {
1587 err = -ENOMEM;
1588 goto out_err;
1589 }
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001590
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001591 get_online_cpus();
1592
1593 if (cpu_online(cpu_id)) {
1594 schedule_work_on(cpu_id,
1595 &cpu_buffer->update_pages_work);
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001596 wait_for_completion(&cpu_buffer->update_done);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001597 } else
1598 rb_update_pages(cpu_buffer);
1599
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001600 cpu_buffer->nr_pages_to_update = 0;
Vaibhav Nagarnaik05fdd702012-05-18 13:29:51 -07001601 put_online_cpus();
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001602 }
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001603
1604 out:
Steven Rostedt659f4512012-05-14 17:02:33 -04001605 /*
1606 * The ring buffer resize can happen with the ring buffer
1607 * enabled, so that the update disturbs the tracing as little
1608 * as possible. But if the buffer is disabled, we do not need
1609 * to worry about that, and we can take the time to verify
1610 * that the buffer is not corrupt.
1611 */
1612 if (atomic_read(&buffer->record_disabled)) {
1613 atomic_inc(&buffer->record_disabled);
1614 /*
1615 * Even though the buffer was disabled, we must make sure
1616 * that it is truly disabled before calling rb_check_pages.
1617 * There could have been a race between checking
1618 * record_disable and incrementing it.
1619 */
1620 synchronize_sched();
1621 for_each_buffer_cpu(buffer, cpu) {
1622 cpu_buffer = buffer->buffers[cpu];
1623 rb_check_pages(cpu_buffer);
1624 }
1625 atomic_dec(&buffer->record_disabled);
1626 }
1627
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001628 mutex_unlock(&buffer->mutex);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001629 return size;
1630
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001631 out_err:
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001632 for_each_buffer_cpu(buffer, cpu) {
1633 struct buffer_page *bpage, *tmp;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001634
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001635 cpu_buffer = buffer->buffers[cpu];
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001636 cpu_buffer->nr_pages_to_update = 0;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001637
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001638 if (list_empty(&cpu_buffer->new_pages))
1639 continue;
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001640
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001641 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
1642 list) {
1643 list_del_init(&bpage->list);
1644 free_buffer_page(bpage);
1645 }
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001646 }
Vegard Nossum641d2f62008-11-18 19:22:13 +01001647 mutex_unlock(&buffer->mutex);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07001648 return err;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001649}
Robert Richterc4f50182008-12-11 16:49:22 +01001650EXPORT_SYMBOL_GPL(ring_buffer_resize);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001651
David Sharp750912f2010-12-08 13:46:47 -08001652void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val)
1653{
1654 mutex_lock(&buffer->mutex);
1655 if (val)
1656 buffer->flags |= RB_FL_OVERWRITE;
1657 else
1658 buffer->flags &= ~RB_FL_OVERWRITE;
1659 mutex_unlock(&buffer->mutex);
1660}
1661EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
1662
Steven Rostedt8789a9e2008-12-02 15:34:07 -05001663static inline void *
Steven Rostedt044fa782008-12-02 23:50:03 -05001664__rb_data_page_index(struct buffer_data_page *bpage, unsigned index)
Steven Rostedt8789a9e2008-12-02 15:34:07 -05001665{
Steven Rostedt044fa782008-12-02 23:50:03 -05001666 return bpage->data + index;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05001667}
1668
Steven Rostedt044fa782008-12-02 23:50:03 -05001669static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001670{
Steven Rostedt044fa782008-12-02 23:50:03 -05001671 return bpage->page->data + index;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001672}
1673
1674static inline struct ring_buffer_event *
Steven Rostedtd7690412008-10-01 00:29:53 -04001675rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001676{
Steven Rostedt6f807ac2008-10-04 02:00:58 -04001677 return __rb_page_index(cpu_buffer->reader_page,
1678 cpu_buffer->reader_page->read);
1679}
1680
1681static inline struct ring_buffer_event *
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001682rb_iter_head_event(struct ring_buffer_iter *iter)
1683{
Steven Rostedt6f807ac2008-10-04 02:00:58 -04001684 return __rb_page_index(iter->head_page, iter->head);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001685}
1686
Steven Rostedtbf41a152008-10-04 02:00:59 -04001687static inline unsigned rb_page_commit(struct buffer_page *bpage)
1688{
Steven Rostedtabc9b562008-12-02 15:34:06 -05001689 return local_read(&bpage->page->commit);
Steven Rostedtbf41a152008-10-04 02:00:59 -04001690}
1691
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001692/* Size is determined by what has been committed */
Steven Rostedtbf41a152008-10-04 02:00:59 -04001693static inline unsigned rb_page_size(struct buffer_page *bpage)
1694{
1695 return rb_page_commit(bpage);
1696}
1697
1698static inline unsigned
1699rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
1700{
1701 return rb_page_commit(cpu_buffer->commit_page);
1702}
1703
Steven Rostedtbf41a152008-10-04 02:00:59 -04001704static inline unsigned
1705rb_event_index(struct ring_buffer_event *event)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001706{
Steven Rostedtbf41a152008-10-04 02:00:59 -04001707 unsigned long addr = (unsigned long)event;
1708
Steven Rostedt22f470f2009-06-11 09:29:58 -04001709 return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001710}
1711
Steven Rostedt0f0c85f2009-05-11 16:08:00 -04001712static inline int
Steven Rostedtfa743952009-06-16 12:37:57 -04001713rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
1714 struct ring_buffer_event *event)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001715{
Steven Rostedtbf41a152008-10-04 02:00:59 -04001716 unsigned long addr = (unsigned long)event;
1717 unsigned long index;
1718
1719 index = rb_event_index(event);
1720 addr &= PAGE_MASK;
1721
1722 return cpu_buffer->commit_page->page == (void *)addr &&
1723 rb_commit_index(cpu_buffer) == index;
1724}
1725
Andrew Morton34a148b2009-01-09 12:27:09 -08001726static void
Steven Rostedtbf41a152008-10-04 02:00:59 -04001727rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
1728{
Steven Rostedt77ae3652009-03-27 11:00:29 -04001729 unsigned long max_count;
1730
Steven Rostedtbf41a152008-10-04 02:00:59 -04001731 /*
1732 * We only race with interrupts and NMIs on this CPU.
1733 * If we own the commit event, then we can commit
1734 * all others that interrupted us, since the interruptions
1735 * are in stack format (they finish before they come
1736 * back to us). This allows us to do a simple loop to
1737 * assign the commit to the tail.
1738 */
Steven Rostedta8ccf1d2008-12-23 11:32:24 -05001739 again:
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08001740 max_count = cpu_buffer->nr_pages * 100;
Steven Rostedt77ae3652009-03-27 11:00:29 -04001741
Steven Rostedtbf41a152008-10-04 02:00:59 -04001742 while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
Steven Rostedt77ae3652009-03-27 11:00:29 -04001743 if (RB_WARN_ON(cpu_buffer, !(--max_count)))
1744 return;
1745 if (RB_WARN_ON(cpu_buffer,
1746 rb_is_reader_page(cpu_buffer->tail_page)))
1747 return;
1748 local_set(&cpu_buffer->commit_page->page->commit,
1749 rb_page_write(cpu_buffer->commit_page));
Steven Rostedtbf41a152008-10-04 02:00:59 -04001750 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
Steven Rostedtabc9b562008-12-02 15:34:06 -05001751 cpu_buffer->write_stamp =
1752 cpu_buffer->commit_page->page->time_stamp;
Steven Rostedtbf41a152008-10-04 02:00:59 -04001753 /* add barrier to keep gcc from optimizing too much */
1754 barrier();
1755 }
1756 while (rb_commit_index(cpu_buffer) !=
1757 rb_page_write(cpu_buffer->commit_page)) {
Steven Rostedt77ae3652009-03-27 11:00:29 -04001758
1759 local_set(&cpu_buffer->commit_page->page->commit,
1760 rb_page_write(cpu_buffer->commit_page));
1761 RB_WARN_ON(cpu_buffer,
1762 local_read(&cpu_buffer->commit_page->page->commit) &
1763 ~RB_WRITE_MASK);
Steven Rostedtbf41a152008-10-04 02:00:59 -04001764 barrier();
1765 }
Steven Rostedta8ccf1d2008-12-23 11:32:24 -05001766
1767 /* again, keep gcc from optimizing */
1768 barrier();
1769
1770 /*
1771 * If an interrupt came in just after the first while loop
1772 * and pushed the tail page forward, we will be left with
1773 * a dangling commit that will never go forward.
1774 */
1775 if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page))
1776 goto again;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001777}
1778
Steven Rostedtd7690412008-10-01 00:29:53 -04001779static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001780{
Steven Rostedtabc9b562008-12-02 15:34:06 -05001781 cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp;
Steven Rostedt6f807ac2008-10-04 02:00:58 -04001782 cpu_buffer->reader_page->read = 0;
Steven Rostedtd7690412008-10-01 00:29:53 -04001783}
1784
Andrew Morton34a148b2009-01-09 12:27:09 -08001785static void rb_inc_iter(struct ring_buffer_iter *iter)
Steven Rostedtd7690412008-10-01 00:29:53 -04001786{
1787 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1788
1789 /*
1790 * The iterator could be on the reader page (it starts there).
1791 * But the head could have moved, since the reader was
1792 * found. Check for this case and assign the iterator
1793 * to the head page instead of next.
1794 */
1795 if (iter->head_page == cpu_buffer->reader_page)
Steven Rostedt77ae3652009-03-27 11:00:29 -04001796 iter->head_page = rb_set_head_page(cpu_buffer);
Steven Rostedtd7690412008-10-01 00:29:53 -04001797 else
1798 rb_inc_page(cpu_buffer, &iter->head_page);
1799
Steven Rostedtabc9b562008-12-02 15:34:06 -05001800 iter->read_stamp = iter->head_page->page->time_stamp;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001801 iter->head = 0;
1802}
1803
Steven Rostedt69d1b832010-10-07 18:18:05 -04001804/* Slow path, do not inline */
1805static noinline struct ring_buffer_event *
1806rb_add_time_stamp(struct ring_buffer_event *event, u64 delta)
1807{
1808 event->type_len = RINGBUF_TYPE_TIME_EXTEND;
1809
1810 /* Not the first event on the page? */
1811 if (rb_event_index(event)) {
1812 event->time_delta = delta & TS_MASK;
1813 event->array[0] = delta >> TS_SHIFT;
1814 } else {
1815 /* nope, just zero it */
1816 event->time_delta = 0;
1817 event->array[0] = 0;
1818 }
1819
1820 return skip_time_extend(event);
1821}
1822
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001823/**
David Sharp01e3e712012-06-07 16:46:24 -07001824 * rb_update_event - update event type and data
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001825 * @event: the even to update
1826 * @type: the type of event
1827 * @length: the size of the event field in the ring buffer
1828 *
1829 * Update the type and data fields of the event. The length
1830 * is the actual size that is written to the ring buffer,
1831 * and with this, we can determine what to place into the
1832 * data field.
1833 */
Andrew Morton34a148b2009-01-09 12:27:09 -08001834static void
Steven Rostedt69d1b832010-10-07 18:18:05 -04001835rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
1836 struct ring_buffer_event *event, unsigned length,
1837 int add_timestamp, u64 delta)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001838{
Steven Rostedt69d1b832010-10-07 18:18:05 -04001839 /* Only a commit updates the timestamp */
1840 if (unlikely(!rb_event_is_commit(cpu_buffer, event)))
1841 delta = 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001842
Steven Rostedt69d1b832010-10-07 18:18:05 -04001843 /*
1844 * If we need to add a timestamp, then we
1845 * add it to the start of the resevered space.
1846 */
1847 if (unlikely(add_timestamp)) {
1848 event = rb_add_time_stamp(event, delta);
1849 length -= RB_LEN_TIME_EXTEND;
1850 delta = 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001851 }
Steven Rostedt69d1b832010-10-07 18:18:05 -04001852
1853 event->time_delta = delta;
1854 length -= RB_EVNT_HDR_SIZE;
1855 if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) {
1856 event->type_len = 0;
1857 event->array[0] = length;
1858 } else
1859 event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04001860}
1861
Steven Rostedt77ae3652009-03-27 11:00:29 -04001862/*
1863 * rb_handle_head_page - writer hit the head page
1864 *
1865 * Returns: +1 to retry page
1866 * 0 to continue
1867 * -1 on error
1868 */
1869static int
1870rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
1871 struct buffer_page *tail_page,
1872 struct buffer_page *next_page)
1873{
1874 struct buffer_page *new_head;
1875 int entries;
1876 int type;
1877 int ret;
1878
1879 entries = rb_page_entries(next_page);
1880
1881 /*
1882 * The hard part is here. We need to move the head
1883 * forward, and protect against both readers on
1884 * other CPUs and writers coming in via interrupts.
1885 */
1886 type = rb_head_page_set_update(cpu_buffer, next_page, tail_page,
1887 RB_PAGE_HEAD);
1888
1889 /*
1890 * type can be one of four:
1891 * NORMAL - an interrupt already moved it for us
1892 * HEAD - we are the first to get here.
1893 * UPDATE - we are the interrupt interrupting
1894 * a current move.
1895 * MOVED - a reader on another CPU moved the next
1896 * pointer to its reader page. Give up
1897 * and try again.
1898 */
1899
1900 switch (type) {
1901 case RB_PAGE_HEAD:
1902 /*
1903 * We changed the head to UPDATE, thus
1904 * it is our responsibility to update
1905 * the counters.
1906 */
1907 local_add(entries, &cpu_buffer->overrun);
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07001908 local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
Steven Rostedt77ae3652009-03-27 11:00:29 -04001909
1910 /*
1911 * The entries will be zeroed out when we move the
1912 * tail page.
1913 */
1914
1915 /* still more to do */
1916 break;
1917
1918 case RB_PAGE_UPDATE:
1919 /*
1920 * This is an interrupt that interrupt the
1921 * previous update. Still more to do.
1922 */
1923 break;
1924 case RB_PAGE_NORMAL:
1925 /*
1926 * An interrupt came in before the update
1927 * and processed this for us.
1928 * Nothing left to do.
1929 */
1930 return 1;
1931 case RB_PAGE_MOVED:
1932 /*
1933 * The reader is on another CPU and just did
1934 * a swap with our next_page.
1935 * Try again.
1936 */
1937 return 1;
1938 default:
1939 RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
1940 return -1;
1941 }
1942
1943 /*
1944 * Now that we are here, the old head pointer is
1945 * set to UPDATE. This will keep the reader from
1946 * swapping the head page with the reader page.
1947 * The reader (on another CPU) will spin till
1948 * we are finished.
1949 *
1950 * We just need to protect against interrupts
1951 * doing the job. We will set the next pointer
1952 * to HEAD. After that, we set the old pointer
1953 * to NORMAL, but only if it was HEAD before.
1954 * otherwise we are an interrupt, and only
1955 * want the outer most commit to reset it.
1956 */
1957 new_head = next_page;
1958 rb_inc_page(cpu_buffer, &new_head);
1959
1960 ret = rb_head_page_set_head(cpu_buffer, new_head, next_page,
1961 RB_PAGE_NORMAL);
1962
1963 /*
1964 * Valid returns are:
1965 * HEAD - an interrupt came in and already set it.
1966 * NORMAL - One of two things:
1967 * 1) We really set it.
1968 * 2) A bunch of interrupts came in and moved
1969 * the page forward again.
1970 */
1971 switch (ret) {
1972 case RB_PAGE_HEAD:
1973 case RB_PAGE_NORMAL:
1974 /* OK */
1975 break;
1976 default:
1977 RB_WARN_ON(cpu_buffer, 1);
1978 return -1;
1979 }
1980
1981 /*
1982 * It is possible that an interrupt came in,
1983 * set the head up, then more interrupts came in
1984 * and moved it again. When we get back here,
1985 * the page would have been set to NORMAL but we
1986 * just set it back to HEAD.
1987 *
1988 * How do you detect this? Well, if that happened
1989 * the tail page would have moved.
1990 */
1991 if (ret == RB_PAGE_NORMAL) {
1992 /*
1993 * If the tail had moved passed next, then we need
1994 * to reset the pointer.
1995 */
1996 if (cpu_buffer->tail_page != tail_page &&
1997 cpu_buffer->tail_page != next_page)
1998 rb_head_page_set_normal(cpu_buffer, new_head,
1999 next_page,
2000 RB_PAGE_HEAD);
2001 }
2002
2003 /*
2004 * If this was the outer most commit (the one that
2005 * changed the original pointer from HEAD to UPDATE),
2006 * then it is up to us to reset it to NORMAL.
2007 */
2008 if (type == RB_PAGE_HEAD) {
2009 ret = rb_head_page_set_normal(cpu_buffer, next_page,
2010 tail_page,
2011 RB_PAGE_UPDATE);
2012 if (RB_WARN_ON(cpu_buffer,
2013 ret != RB_PAGE_UPDATE))
2014 return -1;
2015 }
2016
2017 return 0;
2018}
2019
Andrew Morton34a148b2009-01-09 12:27:09 -08002020static unsigned rb_calculate_event_length(unsigned length)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002021{
2022 struct ring_buffer_event event; /* Used only for sizeof array */
2023
2024 /* zero length can cause confusions */
2025 if (!length)
2026 length = 1;
2027
Steven Rostedt22710482010-03-18 17:54:19 -04002028 if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002029 length += sizeof(event.array[0]);
2030
2031 length += RB_EVNT_HDR_SIZE;
Steven Rostedt22710482010-03-18 17:54:19 -04002032 length = ALIGN(length, RB_ARCH_ALIGNMENT);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002033
2034 return length;
2035}
2036
Steven Rostedtc7b09302009-06-11 11:12:00 -04002037static inline void
2038rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
2039 struct buffer_page *tail_page,
2040 unsigned long tail, unsigned long length)
2041{
2042 struct ring_buffer_event *event;
2043
2044 /*
2045 * Only the event that crossed the page boundary
2046 * must fill the old tail_page with padding.
2047 */
2048 if (tail >= BUF_PAGE_SIZE) {
Steven Rostedtb3230c82010-05-21 11:55:21 -04002049 /*
2050 * If the page was filled, then we still need
2051 * to update the real_end. Reset it to zero
2052 * and the reader will ignore it.
2053 */
2054 if (tail == BUF_PAGE_SIZE)
2055 tail_page->real_end = 0;
2056
Steven Rostedtc7b09302009-06-11 11:12:00 -04002057 local_sub(length, &tail_page->write);
2058 return;
2059 }
2060
2061 event = __rb_page_index(tail_page, tail);
Linus Torvaldsb0b70652009-06-20 10:56:46 -07002062 kmemcheck_annotate_bitfield(event, bitfield);
Steven Rostedtc7b09302009-06-11 11:12:00 -04002063
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002064 /* account for padding bytes */
2065 local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes);
2066
Steven Rostedtc7b09302009-06-11 11:12:00 -04002067 /*
Steven Rostedtff0ff842010-03-31 22:11:42 -04002068 * Save the original length to the meta data.
2069 * This will be used by the reader to add lost event
2070 * counter.
2071 */
2072 tail_page->real_end = tail;
2073
2074 /*
Steven Rostedtc7b09302009-06-11 11:12:00 -04002075 * If this event is bigger than the minimum size, then
2076 * we need to be careful that we don't subtract the
2077 * write counter enough to allow another writer to slip
2078 * in on this page.
2079 * We put in a discarded commit instead, to make sure
2080 * that this space is not used again.
2081 *
2082 * If we are less than the minimum size, we don't need to
2083 * worry about it.
2084 */
2085 if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
2086 /* No room for any events */
2087
2088 /* Mark the rest of the page with padding */
2089 rb_event_set_padding(event);
2090
2091 /* Set the write back to the previous setting */
2092 local_sub(length, &tail_page->write);
2093 return;
2094 }
2095
2096 /* Put in a discarded event */
2097 event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
2098 event->type_len = RINGBUF_TYPE_PADDING;
2099 /* time delta must be non zero */
2100 event->time_delta = 1;
Steven Rostedtc7b09302009-06-11 11:12:00 -04002101
2102 /* Set write to end of buffer */
2103 length = (tail + length) - BUF_PAGE_SIZE;
2104 local_sub(length, &tail_page->write);
2105}
Steven Rostedt6634ff22009-05-06 15:30:07 -04002106
Steven Rostedt747e94a2010-10-08 13:51:48 -04002107/*
2108 * This is the slow path, force gcc not to inline it.
2109 */
2110static noinline struct ring_buffer_event *
Steven Rostedt6634ff22009-05-06 15:30:07 -04002111rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
2112 unsigned long length, unsigned long tail,
Steven Rostedte8bc43e2010-10-20 10:58:02 -04002113 struct buffer_page *tail_page, u64 ts)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002114{
Steven Rostedt5a50e332009-11-17 08:43:01 -05002115 struct buffer_page *commit_page = cpu_buffer->commit_page;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002116 struct ring_buffer *buffer = cpu_buffer->buffer;
Steven Rostedt77ae3652009-03-27 11:00:29 -04002117 struct buffer_page *next_page;
2118 int ret;
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002119
2120 next_page = tail_page;
2121
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002122 rb_inc_page(cpu_buffer, &next_page);
2123
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002124 /*
2125 * If for some reason, we had an interrupt storm that made
2126 * it all the way around the buffer, bail, and warn
2127 * about it.
2128 */
2129 if (unlikely(next_page == commit_page)) {
Steven Rostedt77ae3652009-03-27 11:00:29 -04002130 local_inc(&cpu_buffer->commit_overrun);
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002131 goto out_reset;
2132 }
2133
Steven Rostedt77ae3652009-03-27 11:00:29 -04002134 /*
2135 * This is where the fun begins!
2136 *
2137 * We are fighting against races between a reader that
2138 * could be on another CPU trying to swap its reader
2139 * page with the buffer head.
2140 *
2141 * We are also fighting against interrupts coming in and
2142 * moving the head or tail on us as well.
2143 *
2144 * If the next page is the head page then we have filled
2145 * the buffer, unless the commit page is still on the
2146 * reader page.
2147 */
2148 if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) {
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002149
Steven Rostedt77ae3652009-03-27 11:00:29 -04002150 /*
2151 * If the commit is not on the reader page, then
2152 * move the header page.
2153 */
2154 if (!rb_is_reader_page(cpu_buffer->commit_page)) {
2155 /*
2156 * If we are not in overwrite mode,
2157 * this is easy, just stop here.
2158 */
Slava Pestov884bfe82011-07-15 14:23:58 -07002159 if (!(buffer->flags & RB_FL_OVERWRITE)) {
2160 local_inc(&cpu_buffer->dropped_events);
Steven Rostedt77ae3652009-03-27 11:00:29 -04002161 goto out_reset;
Slava Pestov884bfe82011-07-15 14:23:58 -07002162 }
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002163
Steven Rostedt77ae3652009-03-27 11:00:29 -04002164 ret = rb_handle_head_page(cpu_buffer,
2165 tail_page,
2166 next_page);
2167 if (ret < 0)
2168 goto out_reset;
2169 if (ret)
2170 goto out_again;
2171 } else {
2172 /*
2173 * We need to be careful here too. The
2174 * commit page could still be on the reader
2175 * page. We could have a small buffer, and
2176 * have filled up the buffer with events
2177 * from interrupts and such, and wrapped.
2178 *
2179 * Note, if the tail page is also the on the
2180 * reader_page, we let it move out.
2181 */
2182 if (unlikely((cpu_buffer->commit_page !=
2183 cpu_buffer->tail_page) &&
2184 (cpu_buffer->commit_page ==
2185 cpu_buffer->reader_page))) {
2186 local_inc(&cpu_buffer->commit_overrun);
2187 goto out_reset;
2188 }
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002189 }
2190 }
2191
Steven Rostedt77ae3652009-03-27 11:00:29 -04002192 ret = rb_tail_page_update(cpu_buffer, tail_page, next_page);
2193 if (ret) {
2194 /*
2195 * Nested commits always have zero deltas, so
2196 * just reread the time stamp
2197 */
Steven Rostedte8bc43e2010-10-20 10:58:02 -04002198 ts = rb_time_stamp(buffer);
2199 next_page->page->time_stamp = ts;
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002200 }
2201
Steven Rostedt77ae3652009-03-27 11:00:29 -04002202 out_again:
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002203
Steven Rostedt77ae3652009-03-27 11:00:29 -04002204 rb_reset_tail(cpu_buffer, tail_page, tail, length);
Steven Rostedtaa20ae82009-05-05 21:16:11 -04002205
2206 /* fail and let the caller try again */
2207 return ERR_PTR(-EAGAIN);
2208
Steven Rostedt45141d42009-02-12 13:19:48 -05002209 out_reset:
Lai Jiangshan6f3b3442009-01-12 11:06:18 +08002210 /* reset write */
Steven Rostedtc7b09302009-06-11 11:12:00 -04002211 rb_reset_tail(cpu_buffer, tail_page, tail, length);
Lai Jiangshan6f3b3442009-01-12 11:06:18 +08002212
Steven Rostedtbf41a152008-10-04 02:00:59 -04002213 return NULL;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002214}
2215
Steven Rostedt6634ff22009-05-06 15:30:07 -04002216static struct ring_buffer_event *
2217__rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
Steven Rostedt69d1b832010-10-07 18:18:05 -04002218 unsigned long length, u64 ts,
2219 u64 delta, int add_timestamp)
Steven Rostedt6634ff22009-05-06 15:30:07 -04002220{
Steven Rostedt5a50e332009-11-17 08:43:01 -05002221 struct buffer_page *tail_page;
Steven Rostedt6634ff22009-05-06 15:30:07 -04002222 struct ring_buffer_event *event;
2223 unsigned long tail, write;
2224
Steven Rostedt69d1b832010-10-07 18:18:05 -04002225 /*
2226 * If the time delta since the last event is too big to
2227 * hold in the time field of the event, then we append a
2228 * TIME EXTEND event ahead of the data event.
2229 */
2230 if (unlikely(add_timestamp))
2231 length += RB_LEN_TIME_EXTEND;
2232
Steven Rostedt6634ff22009-05-06 15:30:07 -04002233 tail_page = cpu_buffer->tail_page;
2234 write = local_add_return(length, &tail_page->write);
Steven Rostedt77ae3652009-03-27 11:00:29 -04002235
2236 /* set write to only the index of the write */
2237 write &= RB_WRITE_MASK;
Steven Rostedt6634ff22009-05-06 15:30:07 -04002238 tail = write - length;
2239
2240 /* See if we shot pass the end of this buffer page */
Steven Rostedt747e94a2010-10-08 13:51:48 -04002241 if (unlikely(write > BUF_PAGE_SIZE))
Steven Rostedt6634ff22009-05-06 15:30:07 -04002242 return rb_move_tail(cpu_buffer, length, tail,
Steven Rostedt5a50e332009-11-17 08:43:01 -05002243 tail_page, ts);
Steven Rostedt6634ff22009-05-06 15:30:07 -04002244
2245 /* We reserved something on the buffer */
2246
Steven Rostedt6634ff22009-05-06 15:30:07 -04002247 event = __rb_page_index(tail_page, tail);
Vegard Nossum1744a212009-02-28 08:29:44 +01002248 kmemcheck_annotate_bitfield(event, bitfield);
Steven Rostedt69d1b832010-10-07 18:18:05 -04002249 rb_update_event(cpu_buffer, event, length, add_timestamp, delta);
Steven Rostedt6634ff22009-05-06 15:30:07 -04002250
Steven Rostedt69d1b832010-10-07 18:18:05 -04002251 local_inc(&tail_page->entries);
Steven Rostedt6634ff22009-05-06 15:30:07 -04002252
2253 /*
Steven Rostedtfa743952009-06-16 12:37:57 -04002254 * If this is the first commit on the page, then update
2255 * its timestamp.
Steven Rostedt6634ff22009-05-06 15:30:07 -04002256 */
Steven Rostedtfa743952009-06-16 12:37:57 -04002257 if (!tail)
Steven Rostedte8bc43e2010-10-20 10:58:02 -04002258 tail_page->page->time_stamp = ts;
Steven Rostedt6634ff22009-05-06 15:30:07 -04002259
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002260 /* account for these added bytes */
2261 local_add(length, &cpu_buffer->entries_bytes);
2262
Steven Rostedt6634ff22009-05-06 15:30:07 -04002263 return event;
2264}
2265
Steven Rostedtedd813bf2009-06-02 23:00:53 -04002266static inline int
2267rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
2268 struct ring_buffer_event *event)
2269{
2270 unsigned long new_index, old_index;
2271 struct buffer_page *bpage;
2272 unsigned long index;
2273 unsigned long addr;
2274
2275 new_index = rb_event_index(event);
Steven Rostedt69d1b832010-10-07 18:18:05 -04002276 old_index = new_index + rb_event_ts_length(event);
Steven Rostedtedd813bf2009-06-02 23:00:53 -04002277 addr = (unsigned long)event;
2278 addr &= PAGE_MASK;
2279
2280 bpage = cpu_buffer->tail_page;
2281
2282 if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
Steven Rostedt77ae3652009-03-27 11:00:29 -04002283 unsigned long write_mask =
2284 local_read(&bpage->write) & ~RB_WRITE_MASK;
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002285 unsigned long event_length = rb_event_length(event);
Steven Rostedtedd813bf2009-06-02 23:00:53 -04002286 /*
2287 * This is on the tail page. It is possible that
2288 * a write could come in and move the tail page
2289 * and write to the next page. That is fine
2290 * because we just shorten what is on this page.
2291 */
Steven Rostedt77ae3652009-03-27 11:00:29 -04002292 old_index += write_mask;
2293 new_index += write_mask;
Steven Rostedtedd813bf2009-06-02 23:00:53 -04002294 index = local_cmpxchg(&bpage->write, old_index, new_index);
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002295 if (index == old_index) {
2296 /* update counters */
2297 local_sub(event_length, &cpu_buffer->entries_bytes);
Steven Rostedtedd813bf2009-06-02 23:00:53 -04002298 return 1;
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002299 }
Steven Rostedtedd813bf2009-06-02 23:00:53 -04002300 }
2301
2302 /* could not discard */
2303 return 0;
2304}
2305
Steven Rostedtfa743952009-06-16 12:37:57 -04002306static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
2307{
2308 local_inc(&cpu_buffer->committing);
2309 local_inc(&cpu_buffer->commits);
2310}
2311
Steven Rostedtd9abde22010-10-19 13:17:08 -04002312static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
Steven Rostedtfa743952009-06-16 12:37:57 -04002313{
2314 unsigned long commits;
2315
2316 if (RB_WARN_ON(cpu_buffer,
2317 !local_read(&cpu_buffer->committing)))
2318 return;
2319
2320 again:
2321 commits = local_read(&cpu_buffer->commits);
2322 /* synchronize with interrupts */
2323 barrier();
2324 if (local_read(&cpu_buffer->committing) == 1)
2325 rb_set_commit_to_write(cpu_buffer);
2326
2327 local_dec(&cpu_buffer->committing);
2328
2329 /* synchronize with interrupts */
2330 barrier();
2331
2332 /*
2333 * Need to account for interrupts coming in between the
2334 * updating of the commit page and the clearing of the
2335 * committing counter.
2336 */
2337 if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
2338 !local_read(&cpu_buffer->committing)) {
2339 local_inc(&cpu_buffer->committing);
2340 goto again;
2341 }
2342}
2343
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002344static struct ring_buffer_event *
Steven Rostedt62f0b3e2009-09-04 14:11:34 -04002345rb_reserve_next_event(struct ring_buffer *buffer,
2346 struct ring_buffer_per_cpu *cpu_buffer,
Steven Rostedt1cd8d732009-05-11 14:08:09 -04002347 unsigned long length)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002348{
2349 struct ring_buffer_event *event;
Steven Rostedt69d1b832010-10-07 18:18:05 -04002350 u64 ts, delta;
Steven Rostedt818e3dd2008-10-31 09:58:35 -04002351 int nr_loops = 0;
Steven Rostedt69d1b832010-10-07 18:18:05 -04002352 int add_timestamp;
Steven Rostedt140ff892010-10-08 10:50:30 -04002353 u64 diff;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002354
Steven Rostedtfa743952009-06-16 12:37:57 -04002355 rb_start_commit(cpu_buffer);
2356
Steven Rostedt85bac322009-09-04 14:24:40 -04002357#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
Steven Rostedt62f0b3e2009-09-04 14:11:34 -04002358 /*
2359 * Due to the ability to swap a cpu buffer from a buffer
2360 * it is possible it was swapped before we committed.
2361 * (committing stops a swap). We check for it here and
2362 * if it happened, we have to fail the write.
2363 */
2364 barrier();
2365 if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) {
2366 local_dec(&cpu_buffer->committing);
2367 local_dec(&cpu_buffer->commits);
2368 return NULL;
2369 }
Steven Rostedt85bac322009-09-04 14:24:40 -04002370#endif
Steven Rostedt62f0b3e2009-09-04 14:11:34 -04002371
Steven Rostedtbe957c42009-05-11 14:42:53 -04002372 length = rb_calculate_event_length(length);
Steven Rostedtbf41a152008-10-04 02:00:59 -04002373 again:
Steven Rostedt69d1b832010-10-07 18:18:05 -04002374 add_timestamp = 0;
2375 delta = 0;
2376
Steven Rostedt818e3dd2008-10-31 09:58:35 -04002377 /*
2378 * We allow for interrupts to reenter here and do a trace.
2379 * If one does, it will cause this original code to loop
2380 * back here. Even with heavy interrupts happening, this
2381 * should only happen a few times in a row. If this happens
2382 * 1000 times in a row, there must be either an interrupt
2383 * storm or we have something buggy.
2384 * Bail!
2385 */
Steven Rostedt3e89c7b2008-11-11 15:28:41 -05002386 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
Steven Rostedtfa743952009-06-16 12:37:57 -04002387 goto out_fail;
Steven Rostedt818e3dd2008-10-31 09:58:35 -04002388
Jiri Olsa6d3f1e12009-10-23 19:36:19 -04002389 ts = rb_time_stamp(cpu_buffer->buffer);
Steven Rostedt140ff892010-10-08 10:50:30 -04002390 diff = ts - cpu_buffer->write_stamp;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002391
Steven Rostedt140ff892010-10-08 10:50:30 -04002392 /* make sure this diff is calculated here */
2393 barrier();
Steven Rostedtbf41a152008-10-04 02:00:59 -04002394
Steven Rostedt140ff892010-10-08 10:50:30 -04002395 /* Did the write stamp get updated already? */
2396 if (likely(ts >= cpu_buffer->write_stamp)) {
Steven Rostedt168b6b12009-05-11 22:11:05 -04002397 delta = diff;
2398 if (unlikely(test_time_stamp(delta))) {
Jiri Olsa31274d72011-02-18 15:52:19 +01002399 int local_clock_stable = 1;
2400#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
2401 local_clock_stable = sched_clock_stable;
2402#endif
Steven Rostedt69d1b832010-10-07 18:18:05 -04002403 WARN_ONCE(delta > (1ULL << 59),
Jiri Olsa31274d72011-02-18 15:52:19 +01002404 KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s",
Steven Rostedt69d1b832010-10-07 18:18:05 -04002405 (unsigned long long)delta,
2406 (unsigned long long)ts,
Jiri Olsa31274d72011-02-18 15:52:19 +01002407 (unsigned long long)cpu_buffer->write_stamp,
2408 local_clock_stable ? "" :
2409 "If you just came from a suspend/resume,\n"
2410 "please switch to the trace global clock:\n"
2411 " echo global > /sys/kernel/debug/tracing/trace_clock\n");
Steven Rostedt69d1b832010-10-07 18:18:05 -04002412 add_timestamp = 1;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002413 }
Steven Rostedt168b6b12009-05-11 22:11:05 -04002414 }
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002415
Steven Rostedt69d1b832010-10-07 18:18:05 -04002416 event = __rb_reserve_next(cpu_buffer, length, ts,
2417 delta, add_timestamp);
Steven Rostedt168b6b12009-05-11 22:11:05 -04002418 if (unlikely(PTR_ERR(event) == -EAGAIN))
Steven Rostedtbf41a152008-10-04 02:00:59 -04002419 goto again;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002420
Steven Rostedtfa743952009-06-16 12:37:57 -04002421 if (!event)
2422 goto out_fail;
Steven Rostedtbf41a152008-10-04 02:00:59 -04002423
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002424 return event;
Steven Rostedtfa743952009-06-16 12:37:57 -04002425
2426 out_fail:
2427 rb_end_commit(cpu_buffer);
2428 return NULL;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002429}
2430
Paul Mundt1155de42009-06-25 14:30:12 +09002431#ifdef CONFIG_TRACING
2432
Steven Rostedtaa18efb2009-04-20 16:16:11 -04002433#define TRACE_RECURSIVE_DEPTH 16
Steven Rostedt261842b2009-04-16 21:41:52 -04002434
Steven Rostedtd9abde22010-10-19 13:17:08 -04002435/* Keep this code out of the fast path cache */
2436static noinline void trace_recursive_fail(void)
Steven Rostedt261842b2009-04-16 21:41:52 -04002437{
Steven Rostedtaa18efb2009-04-20 16:16:11 -04002438 /* Disable all tracing before we do anything else */
2439 tracing_off_permanent();
Frederic Weisbeckere057a5e2009-04-19 23:38:12 +02002440
Steven Rostedt7d7d2b82009-04-27 12:37:49 -04002441 printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:"
Steven Rostedtaa18efb2009-04-20 16:16:11 -04002442 "HC[%lu]:SC[%lu]:NMI[%lu]\n",
Steven Rostedtb1cff0a2011-05-25 14:27:43 -04002443 trace_recursion_buffer(),
Steven Rostedtaa18efb2009-04-20 16:16:11 -04002444 hardirq_count() >> HARDIRQ_SHIFT,
2445 softirq_count() >> SOFTIRQ_SHIFT,
2446 in_nmi());
Frederic Weisbeckere057a5e2009-04-19 23:38:12 +02002447
Steven Rostedtaa18efb2009-04-20 16:16:11 -04002448 WARN_ON_ONCE(1);
Steven Rostedtd9abde22010-10-19 13:17:08 -04002449}
2450
2451static inline int trace_recursive_lock(void)
2452{
Steven Rostedtb1cff0a2011-05-25 14:27:43 -04002453 trace_recursion_inc();
Steven Rostedtd9abde22010-10-19 13:17:08 -04002454
Steven Rostedtb1cff0a2011-05-25 14:27:43 -04002455 if (likely(trace_recursion_buffer() < TRACE_RECURSIVE_DEPTH))
Steven Rostedtd9abde22010-10-19 13:17:08 -04002456 return 0;
2457
2458 trace_recursive_fail();
2459
Steven Rostedtaa18efb2009-04-20 16:16:11 -04002460 return -1;
Steven Rostedt261842b2009-04-16 21:41:52 -04002461}
2462
Steven Rostedtd9abde22010-10-19 13:17:08 -04002463static inline void trace_recursive_unlock(void)
Steven Rostedt261842b2009-04-16 21:41:52 -04002464{
Steven Rostedtb1cff0a2011-05-25 14:27:43 -04002465 WARN_ON_ONCE(!trace_recursion_buffer());
Steven Rostedt261842b2009-04-16 21:41:52 -04002466
Steven Rostedtb1cff0a2011-05-25 14:27:43 -04002467 trace_recursion_dec();
Steven Rostedt261842b2009-04-16 21:41:52 -04002468}
2469
Paul Mundt1155de42009-06-25 14:30:12 +09002470#else
2471
2472#define trace_recursive_lock() (0)
2473#define trace_recursive_unlock() do { } while (0)
2474
2475#endif
2476
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002477/**
2478 * ring_buffer_lock_reserve - reserve a part of the buffer
2479 * @buffer: the ring buffer to reserve from
2480 * @length: the length of the data to reserve (excluding event header)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002481 *
2482 * Returns a reseverd event on the ring buffer to copy directly to.
2483 * The user of this interface will need to get the body to write into
2484 * and can use the ring_buffer_event_data() interface.
2485 *
2486 * The length is the length of the data needed, not the event length
2487 * which also includes the event header.
2488 *
2489 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
2490 * If NULL is returned, then nothing has been allocated or locked.
2491 */
2492struct ring_buffer_event *
Arnaldo Carvalho de Melo0a987752009-02-05 16:12:56 -02002493ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002494{
2495 struct ring_buffer_per_cpu *cpu_buffer;
2496 struct ring_buffer_event *event;
Steven Rostedt5168ae52010-06-03 09:36:50 -04002497 int cpu;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002498
Steven Rostedt033601a2008-11-21 12:41:55 -05002499 if (ring_buffer_flags != RB_BUFFERS_ON)
Steven Rostedta3583242008-11-11 15:01:42 -05002500 return NULL;
2501
Steven Rostedtbf41a152008-10-04 02:00:59 -04002502 /* If we are tracing schedule, we don't want to recurse */
Steven Rostedt5168ae52010-06-03 09:36:50 -04002503 preempt_disable_notrace();
Steven Rostedtbf41a152008-10-04 02:00:59 -04002504
Lai Jiangshan52fbe9c2010-03-08 14:50:43 +08002505 if (atomic_read(&buffer->record_disabled))
2506 goto out_nocheck;
2507
Steven Rostedt261842b2009-04-16 21:41:52 -04002508 if (trace_recursive_lock())
2509 goto out_nocheck;
2510
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002511 cpu = raw_smp_processor_id();
2512
Rusty Russell9e01c1b2009-01-01 10:12:22 +10302513 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedtd7690412008-10-01 00:29:53 -04002514 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002515
2516 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002517
2518 if (atomic_read(&cpu_buffer->record_disabled))
Steven Rostedtd7690412008-10-01 00:29:53 -04002519 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002520
Steven Rostedtbe957c42009-05-11 14:42:53 -04002521 if (length > BUF_MAX_DATA_SIZE)
Steven Rostedtbf41a152008-10-04 02:00:59 -04002522 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002523
Steven Rostedt62f0b3e2009-09-04 14:11:34 -04002524 event = rb_reserve_next_event(buffer, cpu_buffer, length);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002525 if (!event)
Steven Rostedtd7690412008-10-01 00:29:53 -04002526 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002527
2528 return event;
2529
Steven Rostedtd7690412008-10-01 00:29:53 -04002530 out:
Steven Rostedt261842b2009-04-16 21:41:52 -04002531 trace_recursive_unlock();
2532
2533 out_nocheck:
Steven Rostedt5168ae52010-06-03 09:36:50 -04002534 preempt_enable_notrace();
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002535 return NULL;
2536}
Robert Richterc4f50182008-12-11 16:49:22 +01002537EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002538
Steven Rostedta1863c22009-09-03 10:23:58 -04002539static void
2540rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer,
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002541 struct ring_buffer_event *event)
2542{
Steven Rostedt69d1b832010-10-07 18:18:05 -04002543 u64 delta;
2544
Steven Rostedtfa743952009-06-16 12:37:57 -04002545 /*
2546 * The event first in the commit queue updates the
2547 * time stamp.
2548 */
Steven Rostedt69d1b832010-10-07 18:18:05 -04002549 if (rb_event_is_commit(cpu_buffer, event)) {
2550 /*
2551 * A commit event that is first on a page
2552 * updates the write timestamp with the page stamp
2553 */
2554 if (!rb_event_index(event))
2555 cpu_buffer->write_stamp =
2556 cpu_buffer->commit_page->page->time_stamp;
2557 else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) {
2558 delta = event->array[0];
2559 delta <<= TS_SHIFT;
2560 delta += event->time_delta;
2561 cpu_buffer->write_stamp += delta;
2562 } else
2563 cpu_buffer->write_stamp += event->time_delta;
2564 }
Steven Rostedta1863c22009-09-03 10:23:58 -04002565}
Steven Rostedtbf41a152008-10-04 02:00:59 -04002566
Steven Rostedta1863c22009-09-03 10:23:58 -04002567static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
2568 struct ring_buffer_event *event)
2569{
2570 local_inc(&cpu_buffer->entries);
2571 rb_update_write_stamp(cpu_buffer, event);
Steven Rostedtfa743952009-06-16 12:37:57 -04002572 rb_end_commit(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002573}
2574
2575/**
2576 * ring_buffer_unlock_commit - commit a reserved
2577 * @buffer: The buffer to commit to
2578 * @event: The event pointer to commit.
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002579 *
2580 * This commits the data to the ring buffer, and releases any locks held.
2581 *
2582 * Must be paired with ring_buffer_lock_reserve.
2583 */
2584int ring_buffer_unlock_commit(struct ring_buffer *buffer,
Arnaldo Carvalho de Melo0a987752009-02-05 16:12:56 -02002585 struct ring_buffer_event *event)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002586{
2587 struct ring_buffer_per_cpu *cpu_buffer;
2588 int cpu = raw_smp_processor_id();
2589
2590 cpu_buffer = buffer->buffers[cpu];
2591
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002592 rb_commit(cpu_buffer, event);
2593
Steven Rostedt261842b2009-04-16 21:41:52 -04002594 trace_recursive_unlock();
2595
Steven Rostedt5168ae52010-06-03 09:36:50 -04002596 preempt_enable_notrace();
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002597
2598 return 0;
2599}
Robert Richterc4f50182008-12-11 16:49:22 +01002600EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002601
Frederic Weisbeckerf3b9aae2009-04-19 23:39:33 +02002602static inline void rb_event_discard(struct ring_buffer_event *event)
2603{
Steven Rostedt69d1b832010-10-07 18:18:05 -04002604 if (event->type_len == RINGBUF_TYPE_TIME_EXTEND)
2605 event = skip_time_extend(event);
2606
Lai Jiangshan334d4162009-04-24 11:27:05 +08002607 /* array[0] holds the actual length for the discarded event */
2608 event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
2609 event->type_len = RINGBUF_TYPE_PADDING;
Frederic Weisbeckerf3b9aae2009-04-19 23:39:33 +02002610 /* time delta must be non zero */
2611 if (!event->time_delta)
2612 event->time_delta = 1;
2613}
2614
Steven Rostedta1863c22009-09-03 10:23:58 -04002615/*
2616 * Decrement the entries to the page that an event is on.
2617 * The event does not even need to exist, only the pointer
2618 * to the page it is on. This may only be called before the commit
2619 * takes place.
2620 */
2621static inline void
2622rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
2623 struct ring_buffer_event *event)
2624{
2625 unsigned long addr = (unsigned long)event;
2626 struct buffer_page *bpage = cpu_buffer->commit_page;
2627 struct buffer_page *start;
2628
2629 addr &= PAGE_MASK;
2630
2631 /* Do the likely case first */
2632 if (likely(bpage->page == (void *)addr)) {
2633 local_dec(&bpage->entries);
2634 return;
2635 }
2636
2637 /*
2638 * Because the commit page may be on the reader page we
2639 * start with the next page and check the end loop there.
2640 */
2641 rb_inc_page(cpu_buffer, &bpage);
2642 start = bpage;
2643 do {
2644 if (bpage->page == (void *)addr) {
2645 local_dec(&bpage->entries);
2646 return;
2647 }
2648 rb_inc_page(cpu_buffer, &bpage);
2649 } while (bpage != start);
2650
2651 /* commit not part of this buffer?? */
2652 RB_WARN_ON(cpu_buffer, 1);
2653}
2654
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002655/**
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002656 * ring_buffer_commit_discard - discard an event that has not been committed
2657 * @buffer: the ring buffer
2658 * @event: non committed event to discard
2659 *
Steven Rostedtdc892f72009-09-03 15:33:41 -04002660 * Sometimes an event that is in the ring buffer needs to be ignored.
2661 * This function lets the user discard an event in the ring buffer
2662 * and then that event will not be read later.
2663 *
2664 * This function only works if it is called before the the item has been
2665 * committed. It will try to free the event from the ring buffer
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002666 * if another event has not been added behind it.
2667 *
2668 * If another event has been added behind it, it will set the event
2669 * up as discarded, and perform the commit.
2670 *
2671 * If this function is called, do not call ring_buffer_unlock_commit on
2672 * the event.
2673 */
2674void ring_buffer_discard_commit(struct ring_buffer *buffer,
2675 struct ring_buffer_event *event)
2676{
2677 struct ring_buffer_per_cpu *cpu_buffer;
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002678 int cpu;
2679
2680 /* The event is discarded regardless */
Frederic Weisbeckerf3b9aae2009-04-19 23:39:33 +02002681 rb_event_discard(event);
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002682
Steven Rostedtfa743952009-06-16 12:37:57 -04002683 cpu = smp_processor_id();
2684 cpu_buffer = buffer->buffers[cpu];
2685
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002686 /*
2687 * This must only be called if the event has not been
2688 * committed yet. Thus we can assume that preemption
2689 * is still disabled.
2690 */
Steven Rostedtfa743952009-06-16 12:37:57 -04002691 RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002692
Steven Rostedta1863c22009-09-03 10:23:58 -04002693 rb_decrement_entry(cpu_buffer, event);
Steven Rostedt0f2541d2009-08-05 12:02:48 -04002694 if (rb_try_to_discard(cpu_buffer, event))
Steven Rostedtedd813bf2009-06-02 23:00:53 -04002695 goto out;
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002696
2697 /*
2698 * The commit is still visible by the reader, so we
Steven Rostedta1863c22009-09-03 10:23:58 -04002699 * must still update the timestamp.
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002700 */
Steven Rostedta1863c22009-09-03 10:23:58 -04002701 rb_update_write_stamp(cpu_buffer, event);
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002702 out:
Steven Rostedtfa743952009-06-16 12:37:57 -04002703 rb_end_commit(cpu_buffer);
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002704
Frederic Weisbeckerf3b9aae2009-04-19 23:39:33 +02002705 trace_recursive_unlock();
2706
Steven Rostedt5168ae52010-06-03 09:36:50 -04002707 preempt_enable_notrace();
Steven Rostedtfa1b47d2009-04-02 00:09:41 -04002708
2709}
2710EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
2711
2712/**
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002713 * ring_buffer_write - write data to the buffer without reserving
2714 * @buffer: The ring buffer to write to.
2715 * @length: The length of the data being written (excluding the event header)
2716 * @data: The data to write to the buffer.
2717 *
2718 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
2719 * one function. If you already have the data to write to the buffer, it
2720 * may be easier to simply call this function.
2721 *
2722 * Note, like ring_buffer_lock_reserve, the length is the length of the data
2723 * and not the length of the event which would hold the header.
2724 */
2725int ring_buffer_write(struct ring_buffer *buffer,
David Sharp01e3e712012-06-07 16:46:24 -07002726 unsigned long length,
2727 void *data)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002728{
2729 struct ring_buffer_per_cpu *cpu_buffer;
2730 struct ring_buffer_event *event;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002731 void *body;
2732 int ret = -EBUSY;
Steven Rostedt5168ae52010-06-03 09:36:50 -04002733 int cpu;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002734
Steven Rostedt033601a2008-11-21 12:41:55 -05002735 if (ring_buffer_flags != RB_BUFFERS_ON)
Steven Rostedta3583242008-11-11 15:01:42 -05002736 return -EBUSY;
2737
Steven Rostedt5168ae52010-06-03 09:36:50 -04002738 preempt_disable_notrace();
Steven Rostedtbf41a152008-10-04 02:00:59 -04002739
Lai Jiangshan52fbe9c2010-03-08 14:50:43 +08002740 if (atomic_read(&buffer->record_disabled))
2741 goto out;
2742
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002743 cpu = raw_smp_processor_id();
2744
Rusty Russell9e01c1b2009-01-01 10:12:22 +10302745 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedtd7690412008-10-01 00:29:53 -04002746 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002747
2748 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002749
2750 if (atomic_read(&cpu_buffer->record_disabled))
2751 goto out;
2752
Steven Rostedtbe957c42009-05-11 14:42:53 -04002753 if (length > BUF_MAX_DATA_SIZE)
2754 goto out;
2755
Steven Rostedt62f0b3e2009-09-04 14:11:34 -04002756 event = rb_reserve_next_event(buffer, cpu_buffer, length);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002757 if (!event)
2758 goto out;
2759
2760 body = rb_event_data(event);
2761
2762 memcpy(body, data, length);
2763
2764 rb_commit(cpu_buffer, event);
2765
2766 ret = 0;
2767 out:
Steven Rostedt5168ae52010-06-03 09:36:50 -04002768 preempt_enable_notrace();
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002769
2770 return ret;
2771}
Robert Richterc4f50182008-12-11 16:49:22 +01002772EXPORT_SYMBOL_GPL(ring_buffer_write);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002773
Andrew Morton34a148b2009-01-09 12:27:09 -08002774static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
Steven Rostedtbf41a152008-10-04 02:00:59 -04002775{
2776 struct buffer_page *reader = cpu_buffer->reader_page;
Steven Rostedt77ae3652009-03-27 11:00:29 -04002777 struct buffer_page *head = rb_set_head_page(cpu_buffer);
Steven Rostedtbf41a152008-10-04 02:00:59 -04002778 struct buffer_page *commit = cpu_buffer->commit_page;
2779
Steven Rostedt77ae3652009-03-27 11:00:29 -04002780 /* In case of error, head will be NULL */
2781 if (unlikely(!head))
2782 return 1;
2783
Steven Rostedtbf41a152008-10-04 02:00:59 -04002784 return reader->read == rb_page_commit(reader) &&
2785 (commit == reader ||
2786 (commit == head &&
2787 head->read == rb_page_commit(commit)));
2788}
2789
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002790/**
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002791 * ring_buffer_record_disable - stop all writes into the buffer
2792 * @buffer: The ring buffer to stop writes to.
2793 *
2794 * This prevents all writes to the buffer. Any attempt to write
2795 * to the buffer after this will fail and return NULL.
2796 *
2797 * The caller should call synchronize_sched() after this.
2798 */
2799void ring_buffer_record_disable(struct ring_buffer *buffer)
2800{
2801 atomic_inc(&buffer->record_disabled);
2802}
Robert Richterc4f50182008-12-11 16:49:22 +01002803EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002804
2805/**
2806 * ring_buffer_record_enable - enable writes to the buffer
2807 * @buffer: The ring buffer to enable writes
2808 *
2809 * Note, multiple disables will need the same number of enables
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002810 * to truly enable the writing (much like preempt_disable).
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002811 */
2812void ring_buffer_record_enable(struct ring_buffer *buffer)
2813{
2814 atomic_dec(&buffer->record_disabled);
2815}
Robert Richterc4f50182008-12-11 16:49:22 +01002816EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002817
2818/**
Steven Rostedt499e5472012-02-22 15:50:28 -05002819 * ring_buffer_record_off - stop all writes into the buffer
2820 * @buffer: The ring buffer to stop writes to.
2821 *
2822 * This prevents all writes to the buffer. Any attempt to write
2823 * to the buffer after this will fail and return NULL.
2824 *
2825 * This is different than ring_buffer_record_disable() as
Wang Tianhong87abb3b2012-08-02 14:02:00 +08002826 * it works like an on/off switch, where as the disable() version
Steven Rostedt499e5472012-02-22 15:50:28 -05002827 * must be paired with a enable().
2828 */
2829void ring_buffer_record_off(struct ring_buffer *buffer)
2830{
2831 unsigned int rd;
2832 unsigned int new_rd;
2833
2834 do {
2835 rd = atomic_read(&buffer->record_disabled);
2836 new_rd = rd | RB_BUFFER_OFF;
2837 } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
2838}
2839EXPORT_SYMBOL_GPL(ring_buffer_record_off);
2840
2841/**
2842 * ring_buffer_record_on - restart writes into the buffer
2843 * @buffer: The ring buffer to start writes to.
2844 *
2845 * This enables all writes to the buffer that was disabled by
2846 * ring_buffer_record_off().
2847 *
2848 * This is different than ring_buffer_record_enable() as
Wang Tianhong87abb3b2012-08-02 14:02:00 +08002849 * it works like an on/off switch, where as the enable() version
Steven Rostedt499e5472012-02-22 15:50:28 -05002850 * must be paired with a disable().
2851 */
2852void ring_buffer_record_on(struct ring_buffer *buffer)
2853{
2854 unsigned int rd;
2855 unsigned int new_rd;
2856
2857 do {
2858 rd = atomic_read(&buffer->record_disabled);
2859 new_rd = rd & ~RB_BUFFER_OFF;
2860 } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
2861}
2862EXPORT_SYMBOL_GPL(ring_buffer_record_on);
2863
2864/**
2865 * ring_buffer_record_is_on - return true if the ring buffer can write
2866 * @buffer: The ring buffer to see if write is enabled
2867 *
2868 * Returns true if the ring buffer is in a state that it accepts writes.
2869 */
2870int ring_buffer_record_is_on(struct ring_buffer *buffer)
2871{
2872 return !atomic_read(&buffer->record_disabled);
2873}
2874
2875/**
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002876 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
2877 * @buffer: The ring buffer to stop writes to.
2878 * @cpu: The CPU buffer to stop
2879 *
2880 * This prevents all writes to the buffer. Any attempt to write
2881 * to the buffer after this will fail and return NULL.
2882 *
2883 * The caller should call synchronize_sched() after this.
2884 */
2885void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
2886{
2887 struct ring_buffer_per_cpu *cpu_buffer;
2888
Rusty Russell9e01c1b2009-01-01 10:12:22 +10302889 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04002890 return;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002891
2892 cpu_buffer = buffer->buffers[cpu];
2893 atomic_inc(&cpu_buffer->record_disabled);
2894}
Robert Richterc4f50182008-12-11 16:49:22 +01002895EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002896
2897/**
2898 * ring_buffer_record_enable_cpu - enable writes to the buffer
2899 * @buffer: The ring buffer to enable writes
2900 * @cpu: The CPU to enable.
2901 *
2902 * Note, multiple disables will need the same number of enables
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05002903 * to truly enable the writing (much like preempt_disable).
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002904 */
2905void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
2906{
2907 struct ring_buffer_per_cpu *cpu_buffer;
2908
Rusty Russell9e01c1b2009-01-01 10:12:22 +10302909 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04002910 return;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002911
2912 cpu_buffer = buffer->buffers[cpu];
2913 atomic_dec(&cpu_buffer->record_disabled);
2914}
Robert Richterc4f50182008-12-11 16:49:22 +01002915EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002916
Steven Rostedtf6195aa2010-09-01 12:23:12 -04002917/*
2918 * The total entries in the ring buffer is the running counter
2919 * of entries entered into the ring buffer, minus the sum of
2920 * the entries read from the ring buffer and the number of
2921 * entries that were overwritten.
2922 */
2923static inline unsigned long
2924rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
2925{
2926 return local_read(&cpu_buffer->entries) -
2927 (local_read(&cpu_buffer->overrun) + cpu_buffer->read);
2928}
2929
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002930/**
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002931 * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
2932 * @buffer: The ring buffer
2933 * @cpu: The per CPU buffer to read from.
2934 */
Yoshihiro YUNOMAE50ecf2c2012-10-11 16:27:54 -07002935u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu)
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002936{
2937 unsigned long flags;
2938 struct ring_buffer_per_cpu *cpu_buffer;
2939 struct buffer_page *bpage;
Yoshihiro YUNOMAE50ecf2c2012-10-11 16:27:54 -07002940 u64 ret;
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002941
2942 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2943 return 0;
2944
2945 cpu_buffer = buffer->buffers[cpu];
Linus Torvalds7115e3f2011-10-26 17:03:38 +02002946 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002947 /*
2948 * if the tail is on reader_page, oldest time stamp is on the reader
2949 * page
2950 */
2951 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
2952 bpage = cpu_buffer->reader_page;
2953 else
2954 bpage = rb_set_head_page(cpu_buffer);
2955 ret = bpage->page->time_stamp;
Linus Torvalds7115e3f2011-10-26 17:03:38 +02002956 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07002957
2958 return ret;
2959}
2960EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
2961
2962/**
2963 * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
2964 * @buffer: The ring buffer
2965 * @cpu: The per CPU buffer to read from.
2966 */
2967unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu)
2968{
2969 struct ring_buffer_per_cpu *cpu_buffer;
2970 unsigned long ret;
2971
2972 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2973 return 0;
2974
2975 cpu_buffer = buffer->buffers[cpu];
2976 ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
2977
2978 return ret;
2979}
2980EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
2981
2982/**
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002983 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
2984 * @buffer: The ring buffer
2985 * @cpu: The per CPU buffer to get the entries from.
2986 */
2987unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
2988{
2989 struct ring_buffer_per_cpu *cpu_buffer;
2990
Rusty Russell9e01c1b2009-01-01 10:12:22 +10302991 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04002992 return 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002993
2994 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt554f7862009-03-11 22:00:13 -04002995
Steven Rostedtf6195aa2010-09-01 12:23:12 -04002996 return rb_num_of_entries(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002997}
Robert Richterc4f50182008-12-11 16:49:22 +01002998EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04002999
3000/**
Slava Pestov884bfe82011-07-15 14:23:58 -07003001 * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
3002 * buffer wrapping around (only if RB_FL_OVERWRITE is on).
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003003 * @buffer: The ring buffer
3004 * @cpu: The per CPU buffer to get the number of overruns from
3005 */
3006unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
3007{
3008 struct ring_buffer_per_cpu *cpu_buffer;
Steven Rostedt8aabee52009-03-12 13:13:49 -04003009 unsigned long ret;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003010
Rusty Russell9e01c1b2009-01-01 10:12:22 +10303011 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04003012 return 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003013
3014 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt77ae3652009-03-27 11:00:29 -04003015 ret = local_read(&cpu_buffer->overrun);
Steven Rostedt554f7862009-03-11 22:00:13 -04003016
3017 return ret;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003018}
Robert Richterc4f50182008-12-11 16:49:22 +01003019EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003020
3021/**
Slava Pestov884bfe82011-07-15 14:23:58 -07003022 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
3023 * commits failing due to the buffer wrapping around while there are uncommitted
3024 * events, such as during an interrupt storm.
Steven Rostedtf0d2c682009-04-29 13:43:37 -04003025 * @buffer: The ring buffer
3026 * @cpu: The per CPU buffer to get the number of overruns from
3027 */
3028unsigned long
3029ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu)
3030{
3031 struct ring_buffer_per_cpu *cpu_buffer;
3032 unsigned long ret;
3033
3034 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3035 return 0;
3036
3037 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt77ae3652009-03-27 11:00:29 -04003038 ret = local_read(&cpu_buffer->commit_overrun);
Steven Rostedtf0d2c682009-04-29 13:43:37 -04003039
3040 return ret;
3041}
3042EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
3043
3044/**
Slava Pestov884bfe82011-07-15 14:23:58 -07003045 * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
3046 * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
3047 * @buffer: The ring buffer
3048 * @cpu: The per CPU buffer to get the number of overruns from
3049 */
3050unsigned long
3051ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu)
3052{
3053 struct ring_buffer_per_cpu *cpu_buffer;
3054 unsigned long ret;
3055
3056 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3057 return 0;
3058
3059 cpu_buffer = buffer->buffers[cpu];
3060 ret = local_read(&cpu_buffer->dropped_events);
3061
3062 return ret;
3063}
3064EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu);
3065
3066/**
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003067 * ring_buffer_entries - get the number of entries in a buffer
3068 * @buffer: The ring buffer
3069 *
3070 * Returns the total number of entries in the ring buffer
3071 * (all CPU entries)
3072 */
3073unsigned long ring_buffer_entries(struct ring_buffer *buffer)
3074{
3075 struct ring_buffer_per_cpu *cpu_buffer;
3076 unsigned long entries = 0;
3077 int cpu;
3078
3079 /* if you care about this being correct, lock the buffer */
3080 for_each_buffer_cpu(buffer, cpu) {
3081 cpu_buffer = buffer->buffers[cpu];
Steven Rostedtf6195aa2010-09-01 12:23:12 -04003082 entries += rb_num_of_entries(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003083 }
3084
3085 return entries;
3086}
Robert Richterc4f50182008-12-11 16:49:22 +01003087EXPORT_SYMBOL_GPL(ring_buffer_entries);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003088
3089/**
Jiri Olsa67b394f2009-10-23 19:36:18 -04003090 * ring_buffer_overruns - get the number of overruns in buffer
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003091 * @buffer: The ring buffer
3092 *
3093 * Returns the total number of overruns in the ring buffer
3094 * (all CPU entries)
3095 */
3096unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
3097{
3098 struct ring_buffer_per_cpu *cpu_buffer;
3099 unsigned long overruns = 0;
3100 int cpu;
3101
3102 /* if you care about this being correct, lock the buffer */
3103 for_each_buffer_cpu(buffer, cpu) {
3104 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt77ae3652009-03-27 11:00:29 -04003105 overruns += local_read(&cpu_buffer->overrun);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003106 }
3107
3108 return overruns;
3109}
Robert Richterc4f50182008-12-11 16:49:22 +01003110EXPORT_SYMBOL_GPL(ring_buffer_overruns);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003111
Steven Rostedt642edba2008-11-12 00:01:26 -05003112static void rb_iter_reset(struct ring_buffer_iter *iter)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003113{
3114 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3115
Steven Rostedtd7690412008-10-01 00:29:53 -04003116 /* Iterator usage is expected to have record disabled */
3117 if (list_empty(&cpu_buffer->reader_page->list)) {
Steven Rostedt77ae3652009-03-27 11:00:29 -04003118 iter->head_page = rb_set_head_page(cpu_buffer);
3119 if (unlikely(!iter->head_page))
3120 return;
3121 iter->head = iter->head_page->read;
Steven Rostedtd7690412008-10-01 00:29:53 -04003122 } else {
3123 iter->head_page = cpu_buffer->reader_page;
Steven Rostedt6f807ac2008-10-04 02:00:58 -04003124 iter->head = cpu_buffer->reader_page->read;
Steven Rostedtd7690412008-10-01 00:29:53 -04003125 }
3126 if (iter->head)
3127 iter->read_stamp = cpu_buffer->read_stamp;
3128 else
Steven Rostedtabc9b562008-12-02 15:34:06 -05003129 iter->read_stamp = iter->head_page->page->time_stamp;
Steven Rostedt492a74f2010-01-25 15:17:47 -05003130 iter->cache_reader_page = cpu_buffer->reader_page;
3131 iter->cache_read = cpu_buffer->read;
Steven Rostedt642edba2008-11-12 00:01:26 -05003132}
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003133
Steven Rostedt642edba2008-11-12 00:01:26 -05003134/**
3135 * ring_buffer_iter_reset - reset an iterator
3136 * @iter: The iterator to reset
3137 *
3138 * Resets the iterator, so that it will start from the beginning
3139 * again.
3140 */
3141void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
3142{
Steven Rostedt554f7862009-03-11 22:00:13 -04003143 struct ring_buffer_per_cpu *cpu_buffer;
Steven Rostedt642edba2008-11-12 00:01:26 -05003144 unsigned long flags;
3145
Steven Rostedt554f7862009-03-11 22:00:13 -04003146 if (!iter)
3147 return;
3148
3149 cpu_buffer = iter->cpu_buffer;
3150
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003151 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
Steven Rostedt642edba2008-11-12 00:01:26 -05003152 rb_iter_reset(iter);
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003153 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003154}
Robert Richterc4f50182008-12-11 16:49:22 +01003155EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003156
3157/**
3158 * ring_buffer_iter_empty - check if an iterator has no more to read
3159 * @iter: The iterator to check
3160 */
3161int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
3162{
3163 struct ring_buffer_per_cpu *cpu_buffer;
3164
3165 cpu_buffer = iter->cpu_buffer;
3166
Steven Rostedtbf41a152008-10-04 02:00:59 -04003167 return iter->head_page == cpu_buffer->commit_page &&
3168 iter->head == rb_commit_index(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003169}
Robert Richterc4f50182008-12-11 16:49:22 +01003170EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003171
3172static void
3173rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
3174 struct ring_buffer_event *event)
3175{
3176 u64 delta;
3177
Lai Jiangshan334d4162009-04-24 11:27:05 +08003178 switch (event->type_len) {
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003179 case RINGBUF_TYPE_PADDING:
3180 return;
3181
3182 case RINGBUF_TYPE_TIME_EXTEND:
3183 delta = event->array[0];
3184 delta <<= TS_SHIFT;
3185 delta += event->time_delta;
3186 cpu_buffer->read_stamp += delta;
3187 return;
3188
3189 case RINGBUF_TYPE_TIME_STAMP:
3190 /* FIXME: not implemented */
3191 return;
3192
3193 case RINGBUF_TYPE_DATA:
3194 cpu_buffer->read_stamp += event->time_delta;
3195 return;
3196
3197 default:
3198 BUG();
3199 }
3200 return;
3201}
3202
3203static void
3204rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
3205 struct ring_buffer_event *event)
3206{
3207 u64 delta;
3208
Lai Jiangshan334d4162009-04-24 11:27:05 +08003209 switch (event->type_len) {
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003210 case RINGBUF_TYPE_PADDING:
3211 return;
3212
3213 case RINGBUF_TYPE_TIME_EXTEND:
3214 delta = event->array[0];
3215 delta <<= TS_SHIFT;
3216 delta += event->time_delta;
3217 iter->read_stamp += delta;
3218 return;
3219
3220 case RINGBUF_TYPE_TIME_STAMP:
3221 /* FIXME: not implemented */
3222 return;
3223
3224 case RINGBUF_TYPE_DATA:
3225 iter->read_stamp += event->time_delta;
3226 return;
3227
3228 default:
3229 BUG();
3230 }
3231 return;
3232}
3233
Steven Rostedtd7690412008-10-01 00:29:53 -04003234static struct buffer_page *
3235rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003236{
Steven Rostedtd7690412008-10-01 00:29:53 -04003237 struct buffer_page *reader = NULL;
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003238 unsigned long overwrite;
Steven Rostedtd7690412008-10-01 00:29:53 -04003239 unsigned long flags;
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003240 int nr_loops = 0;
Steven Rostedt77ae3652009-03-27 11:00:29 -04003241 int ret;
Steven Rostedtd7690412008-10-01 00:29:53 -04003242
Steven Rostedt3e03fb72008-11-06 00:09:43 -05003243 local_irq_save(flags);
Thomas Gleixner0199c4e2009-12-02 20:01:25 +01003244 arch_spin_lock(&cpu_buffer->lock);
Steven Rostedtd7690412008-10-01 00:29:53 -04003245
3246 again:
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003247 /*
3248 * This should normally only loop twice. But because the
3249 * start of the reader inserts an empty page, it causes
3250 * a case where we will loop three times. There should be no
3251 * reason to loop four times (that I know of).
3252 */
Steven Rostedt3e89c7b2008-11-11 15:28:41 -05003253 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003254 reader = NULL;
3255 goto out;
3256 }
3257
Steven Rostedtd7690412008-10-01 00:29:53 -04003258 reader = cpu_buffer->reader_page;
3259
3260 /* If there's more to read, return this page */
Steven Rostedtbf41a152008-10-04 02:00:59 -04003261 if (cpu_buffer->reader_page->read < rb_page_size(reader))
Steven Rostedtd7690412008-10-01 00:29:53 -04003262 goto out;
3263
3264 /* Never should we have an index greater than the size */
Steven Rostedt3e89c7b2008-11-11 15:28:41 -05003265 if (RB_WARN_ON(cpu_buffer,
3266 cpu_buffer->reader_page->read > rb_page_size(reader)))
3267 goto out;
Steven Rostedtd7690412008-10-01 00:29:53 -04003268
3269 /* check if we caught up to the tail */
3270 reader = NULL;
Steven Rostedtbf41a152008-10-04 02:00:59 -04003271 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
Steven Rostedtd7690412008-10-01 00:29:53 -04003272 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003273
Steven Rostedta5fb8332012-06-28 13:35:04 -04003274 /* Don't bother swapping if the ring buffer is empty */
3275 if (rb_num_of_entries(cpu_buffer) == 0)
3276 goto out;
3277
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003278 /*
Steven Rostedtd7690412008-10-01 00:29:53 -04003279 * Reset the reader page to size zero.
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003280 */
Steven Rostedt77ae3652009-03-27 11:00:29 -04003281 local_set(&cpu_buffer->reader_page->write, 0);
3282 local_set(&cpu_buffer->reader_page->entries, 0);
3283 local_set(&cpu_buffer->reader_page->page->commit, 0);
Steven Rostedtff0ff842010-03-31 22:11:42 -04003284 cpu_buffer->reader_page->real_end = 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003285
Steven Rostedt77ae3652009-03-27 11:00:29 -04003286 spin:
3287 /*
3288 * Splice the empty reader page into the list around the head.
3289 */
3290 reader = rb_set_head_page(cpu_buffer);
Steven Rostedt0e1ff5d2010-01-06 20:40:44 -05003291 cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
Steven Rostedtd7690412008-10-01 00:29:53 -04003292 cpu_buffer->reader_page->list.prev = reader->list.prev;
Steven Rostedtbf41a152008-10-04 02:00:59 -04003293
Steven Rostedt3adc54f2009-03-30 15:32:01 -04003294 /*
3295 * cpu_buffer->pages just needs to point to the buffer, it
3296 * has no specific buffer page to point to. Lets move it out
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003297 * of our way so we don't accidentally swap it.
Steven Rostedt3adc54f2009-03-30 15:32:01 -04003298 */
3299 cpu_buffer->pages = reader->list.prev;
3300
Steven Rostedt77ae3652009-03-27 11:00:29 -04003301 /* The reader page will be pointing to the new head */
3302 rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list);
Steven Rostedtd7690412008-10-01 00:29:53 -04003303
3304 /*
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003305 * We want to make sure we read the overruns after we set up our
3306 * pointers to the next object. The writer side does a
3307 * cmpxchg to cross pages which acts as the mb on the writer
3308 * side. Note, the reader will constantly fail the swap
3309 * while the writer is updating the pointers, so this
3310 * guarantees that the overwrite recorded here is the one we
3311 * want to compare with the last_overrun.
3312 */
3313 smp_mb();
3314 overwrite = local_read(&(cpu_buffer->overrun));
3315
3316 /*
Steven Rostedt77ae3652009-03-27 11:00:29 -04003317 * Here's the tricky part.
3318 *
3319 * We need to move the pointer past the header page.
3320 * But we can only do that if a writer is not currently
3321 * moving it. The page before the header page has the
3322 * flag bit '1' set if it is pointing to the page we want.
3323 * but if the writer is in the process of moving it
3324 * than it will be '2' or already moved '0'.
Steven Rostedtd7690412008-10-01 00:29:53 -04003325 */
Steven Rostedtd7690412008-10-01 00:29:53 -04003326
Steven Rostedt77ae3652009-03-27 11:00:29 -04003327 ret = rb_head_page_replace(reader, cpu_buffer->reader_page);
3328
3329 /*
3330 * If we did not convert it, then we must try again.
3331 */
3332 if (!ret)
3333 goto spin;
3334
3335 /*
3336 * Yeah! We succeeded in replacing the page.
3337 *
3338 * Now make the new head point back to the reader page.
3339 */
David Sharp5ded3dc62010-01-06 17:12:07 -08003340 rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
Steven Rostedt77ae3652009-03-27 11:00:29 -04003341 rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
Steven Rostedtd7690412008-10-01 00:29:53 -04003342
3343 /* Finally update the reader page to the new head */
3344 cpu_buffer->reader_page = reader;
3345 rb_reset_reader_page(cpu_buffer);
3346
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003347 if (overwrite != cpu_buffer->last_overrun) {
3348 cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
3349 cpu_buffer->last_overrun = overwrite;
3350 }
3351
Steven Rostedtd7690412008-10-01 00:29:53 -04003352 goto again;
3353
3354 out:
Thomas Gleixner0199c4e2009-12-02 20:01:25 +01003355 arch_spin_unlock(&cpu_buffer->lock);
Steven Rostedt3e03fb72008-11-06 00:09:43 -05003356 local_irq_restore(flags);
Steven Rostedtd7690412008-10-01 00:29:53 -04003357
3358 return reader;
3359}
3360
3361static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
3362{
3363 struct ring_buffer_event *event;
3364 struct buffer_page *reader;
3365 unsigned length;
3366
3367 reader = rb_get_reader_page(cpu_buffer);
3368
3369 /* This function should not be called when buffer is empty */
Steven Rostedt3e89c7b2008-11-11 15:28:41 -05003370 if (RB_WARN_ON(cpu_buffer, !reader))
3371 return;
Steven Rostedtd7690412008-10-01 00:29:53 -04003372
3373 event = rb_reader_event(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003374
Steven Rostedta1863c22009-09-03 10:23:58 -04003375 if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
Steven Rostedte4906ef2009-04-30 20:49:44 -04003376 cpu_buffer->read++;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003377
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003378 rb_update_read_stamp(cpu_buffer, event);
3379
Steven Rostedtd7690412008-10-01 00:29:53 -04003380 length = rb_event_length(event);
Steven Rostedt6f807ac2008-10-04 02:00:58 -04003381 cpu_buffer->reader_page->read += length;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003382}
3383
3384static void rb_advance_iter(struct ring_buffer_iter *iter)
3385{
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003386 struct ring_buffer_per_cpu *cpu_buffer;
3387 struct ring_buffer_event *event;
3388 unsigned length;
3389
3390 cpu_buffer = iter->cpu_buffer;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003391
3392 /*
3393 * Check if we are at the end of the buffer.
3394 */
Steven Rostedtbf41a152008-10-04 02:00:59 -04003395 if (iter->head >= rb_page_size(iter->head_page)) {
Steven Rostedtea05b572009-06-03 09:30:10 -04003396 /* discarded commits can make the page empty */
3397 if (iter->head_page == cpu_buffer->commit_page)
Steven Rostedt3e89c7b2008-11-11 15:28:41 -05003398 return;
Steven Rostedtd7690412008-10-01 00:29:53 -04003399 rb_inc_iter(iter);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003400 return;
3401 }
3402
3403 event = rb_iter_head_event(iter);
3404
3405 length = rb_event_length(event);
3406
3407 /*
3408 * This should not be called to advance the header if we are
3409 * at the tail of the buffer.
3410 */
Steven Rostedt3e89c7b2008-11-11 15:28:41 -05003411 if (RB_WARN_ON(cpu_buffer,
Steven Rostedtf536aaf2008-11-10 23:07:30 -05003412 (iter->head_page == cpu_buffer->commit_page) &&
Steven Rostedt3e89c7b2008-11-11 15:28:41 -05003413 (iter->head + length > rb_commit_index(cpu_buffer))))
3414 return;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003415
3416 rb_update_iter_read_stamp(iter, event);
3417
3418 iter->head += length;
3419
3420 /* check for end of page padding */
Steven Rostedtbf41a152008-10-04 02:00:59 -04003421 if ((iter->head >= rb_page_size(iter->head_page)) &&
3422 (iter->head_page != cpu_buffer->commit_page))
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003423 rb_advance_iter(iter);
3424}
3425
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003426static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
3427{
3428 return cpu_buffer->lost_events;
3429}
3430
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003431static struct ring_buffer_event *
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003432rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
3433 unsigned long *lost_events)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003434{
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003435 struct ring_buffer_event *event;
Steven Rostedtd7690412008-10-01 00:29:53 -04003436 struct buffer_page *reader;
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003437 int nr_loops = 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003438
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003439 again:
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003440 /*
Steven Rostedt69d1b832010-10-07 18:18:05 -04003441 * We repeat when a time extend is encountered.
3442 * Since the time extend is always attached to a data event,
3443 * we should never loop more than once.
3444 * (We never hit the following condition more than twice).
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003445 */
Steven Rostedt69d1b832010-10-07 18:18:05 -04003446 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003447 return NULL;
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003448
Steven Rostedtd7690412008-10-01 00:29:53 -04003449 reader = rb_get_reader_page(cpu_buffer);
3450 if (!reader)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003451 return NULL;
3452
Steven Rostedtd7690412008-10-01 00:29:53 -04003453 event = rb_reader_event(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003454
Lai Jiangshan334d4162009-04-24 11:27:05 +08003455 switch (event->type_len) {
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003456 case RINGBUF_TYPE_PADDING:
Tom Zanussi2d622712009-03-22 03:30:49 -05003457 if (rb_null_event(event))
3458 RB_WARN_ON(cpu_buffer, 1);
3459 /*
3460 * Because the writer could be discarding every
3461 * event it creates (which would probably be bad)
3462 * if we were to go back to "again" then we may never
3463 * catch up, and will trigger the warn on, or lock
3464 * the box. Return the padding, and we will release
3465 * the current locks, and try again.
3466 */
Tom Zanussi2d622712009-03-22 03:30:49 -05003467 return event;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003468
3469 case RINGBUF_TYPE_TIME_EXTEND:
3470 /* Internal data, OK to advance */
Steven Rostedtd7690412008-10-01 00:29:53 -04003471 rb_advance_reader(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003472 goto again;
3473
3474 case RINGBUF_TYPE_TIME_STAMP:
3475 /* FIXME: not implemented */
Steven Rostedtd7690412008-10-01 00:29:53 -04003476 rb_advance_reader(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003477 goto again;
3478
3479 case RINGBUF_TYPE_DATA:
3480 if (ts) {
3481 *ts = cpu_buffer->read_stamp + event->time_delta;
Robert Richterd8eeb2d2009-07-31 14:58:04 +02003482 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
Steven Rostedt37886f62009-03-17 17:22:06 -04003483 cpu_buffer->cpu, ts);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003484 }
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003485 if (lost_events)
3486 *lost_events = rb_lost_events(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003487 return event;
3488
3489 default:
3490 BUG();
3491 }
3492
3493 return NULL;
3494}
Robert Richterc4f50182008-12-11 16:49:22 +01003495EXPORT_SYMBOL_GPL(ring_buffer_peek);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003496
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003497static struct ring_buffer_event *
3498rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003499{
3500 struct ring_buffer *buffer;
3501 struct ring_buffer_per_cpu *cpu_buffer;
3502 struct ring_buffer_event *event;
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003503 int nr_loops = 0;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003504
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003505 cpu_buffer = iter->cpu_buffer;
3506 buffer = cpu_buffer->buffer;
3507
Steven Rostedt492a74f2010-01-25 15:17:47 -05003508 /*
3509 * Check if someone performed a consuming read to
3510 * the buffer. A consuming read invalidates the iterator
3511 * and we need to reset the iterator in this case.
3512 */
3513 if (unlikely(iter->cache_read != cpu_buffer->read ||
3514 iter->cache_reader_page != cpu_buffer->reader_page))
3515 rb_iter_reset(iter);
3516
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003517 again:
Steven Rostedt3c05d742010-01-26 16:14:08 -05003518 if (ring_buffer_iter_empty(iter))
3519 return NULL;
3520
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003521 /*
Steven Rostedt69d1b832010-10-07 18:18:05 -04003522 * We repeat when a time extend is encountered.
3523 * Since the time extend is always attached to a data event,
3524 * we should never loop more than once.
3525 * (We never hit the following condition more than twice).
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003526 */
Steven Rostedt69d1b832010-10-07 18:18:05 -04003527 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003528 return NULL;
Steven Rostedt818e3dd2008-10-31 09:58:35 -04003529
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003530 if (rb_per_cpu_empty(cpu_buffer))
3531 return NULL;
3532
Steven Rostedt3c05d742010-01-26 16:14:08 -05003533 if (iter->head >= local_read(&iter->head_page->page->commit)) {
3534 rb_inc_iter(iter);
3535 goto again;
3536 }
3537
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003538 event = rb_iter_head_event(iter);
3539
Lai Jiangshan334d4162009-04-24 11:27:05 +08003540 switch (event->type_len) {
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003541 case RINGBUF_TYPE_PADDING:
Tom Zanussi2d622712009-03-22 03:30:49 -05003542 if (rb_null_event(event)) {
3543 rb_inc_iter(iter);
3544 goto again;
3545 }
3546 rb_advance_iter(iter);
3547 return event;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003548
3549 case RINGBUF_TYPE_TIME_EXTEND:
3550 /* Internal data, OK to advance */
3551 rb_advance_iter(iter);
3552 goto again;
3553
3554 case RINGBUF_TYPE_TIME_STAMP:
3555 /* FIXME: not implemented */
3556 rb_advance_iter(iter);
3557 goto again;
3558
3559 case RINGBUF_TYPE_DATA:
3560 if (ts) {
3561 *ts = iter->read_stamp + event->time_delta;
Steven Rostedt37886f62009-03-17 17:22:06 -04003562 ring_buffer_normalize_time_stamp(buffer,
3563 cpu_buffer->cpu, ts);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003564 }
3565 return event;
3566
3567 default:
3568 BUG();
3569 }
3570
3571 return NULL;
3572}
Robert Richterc4f50182008-12-11 16:49:22 +01003573EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003574
Steven Rostedt8d707e82009-06-16 21:22:48 -04003575static inline int rb_ok_to_lock(void)
3576{
3577 /*
3578 * If an NMI die dumps out the content of the ring buffer
3579 * do not grab locks. We also permanently disable the ring
3580 * buffer too. A one time deal is all you get from reading
3581 * the ring buffer from an NMI.
3582 */
Steven Rostedt464e85e2009-08-05 15:26:37 -04003583 if (likely(!in_nmi()))
Steven Rostedt8d707e82009-06-16 21:22:48 -04003584 return 1;
3585
3586 tracing_off_permanent();
3587 return 0;
3588}
3589
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003590/**
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003591 * ring_buffer_peek - peek at the next event to be read
3592 * @buffer: The ring buffer to read
3593 * @cpu: The cpu to peak at
3594 * @ts: The timestamp counter of this event.
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003595 * @lost_events: a variable to store if events were lost (may be NULL)
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003596 *
3597 * This will return the event that will be read next, but does
3598 * not consume the data.
3599 */
3600struct ring_buffer_event *
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003601ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts,
3602 unsigned long *lost_events)
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003603{
3604 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
Steven Rostedt8aabee52009-03-12 13:13:49 -04003605 struct ring_buffer_event *event;
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003606 unsigned long flags;
Steven Rostedt8d707e82009-06-16 21:22:48 -04003607 int dolock;
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003608
Steven Rostedt554f7862009-03-11 22:00:13 -04003609 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04003610 return NULL;
Steven Rostedt554f7862009-03-11 22:00:13 -04003611
Steven Rostedt8d707e82009-06-16 21:22:48 -04003612 dolock = rb_ok_to_lock();
Tom Zanussi2d622712009-03-22 03:30:49 -05003613 again:
Steven Rostedt8d707e82009-06-16 21:22:48 -04003614 local_irq_save(flags);
3615 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003616 raw_spin_lock(&cpu_buffer->reader_lock);
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003617 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
Robert Richter469535a2009-07-30 19:19:18 +02003618 if (event && event->type_len == RINGBUF_TYPE_PADDING)
3619 rb_advance_reader(cpu_buffer);
Steven Rostedt8d707e82009-06-16 21:22:48 -04003620 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003621 raw_spin_unlock(&cpu_buffer->reader_lock);
Steven Rostedt8d707e82009-06-16 21:22:48 -04003622 local_irq_restore(flags);
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003623
Steven Rostedt1b959e12009-09-03 10:12:13 -04003624 if (event && event->type_len == RINGBUF_TYPE_PADDING)
Tom Zanussi2d622712009-03-22 03:30:49 -05003625 goto again;
Tom Zanussi2d622712009-03-22 03:30:49 -05003626
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003627 return event;
3628}
3629
3630/**
3631 * ring_buffer_iter_peek - peek at the next event to be read
3632 * @iter: The ring buffer iterator
3633 * @ts: The timestamp counter of this event.
3634 *
3635 * This will return the event that will be read next, but does
3636 * not increment the iterator.
3637 */
3638struct ring_buffer_event *
3639ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
3640{
3641 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3642 struct ring_buffer_event *event;
3643 unsigned long flags;
3644
Tom Zanussi2d622712009-03-22 03:30:49 -05003645 again:
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003646 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003647 event = rb_iter_peek(iter, ts);
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003648 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003649
Steven Rostedt1b959e12009-09-03 10:12:13 -04003650 if (event && event->type_len == RINGBUF_TYPE_PADDING)
Tom Zanussi2d622712009-03-22 03:30:49 -05003651 goto again;
Tom Zanussi2d622712009-03-22 03:30:49 -05003652
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003653 return event;
3654}
3655
3656/**
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003657 * ring_buffer_consume - return an event and consume it
3658 * @buffer: The ring buffer to get the next event from
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003659 * @cpu: the cpu to read the buffer from
3660 * @ts: a variable to store the timestamp (may be NULL)
3661 * @lost_events: a variable to store if events were lost (may be NULL)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003662 *
3663 * Returns the next event in the ring buffer, and that event is consumed.
3664 * Meaning, that sequential reads will keep returning a different event,
3665 * and eventually empty the ring buffer if the producer is slower.
3666 */
3667struct ring_buffer_event *
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003668ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts,
3669 unsigned long *lost_events)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003670{
Steven Rostedt554f7862009-03-11 22:00:13 -04003671 struct ring_buffer_per_cpu *cpu_buffer;
3672 struct ring_buffer_event *event = NULL;
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003673 unsigned long flags;
Steven Rostedt8d707e82009-06-16 21:22:48 -04003674 int dolock;
3675
3676 dolock = rb_ok_to_lock();
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003677
Tom Zanussi2d622712009-03-22 03:30:49 -05003678 again:
Steven Rostedt554f7862009-03-11 22:00:13 -04003679 /* might be called in atomic */
3680 preempt_disable();
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003681
Steven Rostedt554f7862009-03-11 22:00:13 -04003682 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3683 goto out;
3684
3685 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt8d707e82009-06-16 21:22:48 -04003686 local_irq_save(flags);
3687 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003688 raw_spin_lock(&cpu_buffer->reader_lock);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003689
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003690 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
3691 if (event) {
3692 cpu_buffer->lost_events = 0;
Robert Richter469535a2009-07-30 19:19:18 +02003693 rb_advance_reader(cpu_buffer);
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003694 }
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003695
Steven Rostedt8d707e82009-06-16 21:22:48 -04003696 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003697 raw_spin_unlock(&cpu_buffer->reader_lock);
Steven Rostedt8d707e82009-06-16 21:22:48 -04003698 local_irq_restore(flags);
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003699
Steven Rostedt554f7862009-03-11 22:00:13 -04003700 out:
3701 preempt_enable();
3702
Steven Rostedt1b959e12009-09-03 10:12:13 -04003703 if (event && event->type_len == RINGBUF_TYPE_PADDING)
Tom Zanussi2d622712009-03-22 03:30:49 -05003704 goto again;
Tom Zanussi2d622712009-03-22 03:30:49 -05003705
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003706 return event;
3707}
Robert Richterc4f50182008-12-11 16:49:22 +01003708EXPORT_SYMBOL_GPL(ring_buffer_consume);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003709
3710/**
David Miller72c9ddf2010-04-20 15:47:11 -07003711 * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003712 * @buffer: The ring buffer to read from
3713 * @cpu: The cpu buffer to iterate over
3714 *
David Miller72c9ddf2010-04-20 15:47:11 -07003715 * This performs the initial preparations necessary to iterate
3716 * through the buffer. Memory is allocated, buffer recording
3717 * is disabled, and the iterator pointer is returned to the caller.
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003718 *
David Miller72c9ddf2010-04-20 15:47:11 -07003719 * Disabling buffer recordng prevents the reading from being
3720 * corrupted. This is not a consuming read, so a producer is not
3721 * expected.
3722 *
3723 * After a sequence of ring_buffer_read_prepare calls, the user is
3724 * expected to make at least one call to ring_buffer_prepare_sync.
3725 * Afterwards, ring_buffer_read_start is invoked to get things going
3726 * for real.
3727 *
3728 * This overall must be paired with ring_buffer_finish.
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003729 */
3730struct ring_buffer_iter *
David Miller72c9ddf2010-04-20 15:47:11 -07003731ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003732{
3733 struct ring_buffer_per_cpu *cpu_buffer;
Steven Rostedt8aabee52009-03-12 13:13:49 -04003734 struct ring_buffer_iter *iter;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003735
Rusty Russell9e01c1b2009-01-01 10:12:22 +10303736 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04003737 return NULL;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003738
3739 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
3740 if (!iter)
Steven Rostedt8aabee52009-03-12 13:13:49 -04003741 return NULL;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003742
3743 cpu_buffer = buffer->buffers[cpu];
3744
3745 iter->cpu_buffer = cpu_buffer;
3746
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07003747 atomic_inc(&buffer->resize_disabled);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003748 atomic_inc(&cpu_buffer->record_disabled);
David Miller72c9ddf2010-04-20 15:47:11 -07003749
3750 return iter;
3751}
3752EXPORT_SYMBOL_GPL(ring_buffer_read_prepare);
3753
3754/**
3755 * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
3756 *
3757 * All previously invoked ring_buffer_read_prepare calls to prepare
3758 * iterators will be synchronized. Afterwards, read_buffer_read_start
3759 * calls on those iterators are allowed.
3760 */
3761void
3762ring_buffer_read_prepare_sync(void)
3763{
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003764 synchronize_sched();
David Miller72c9ddf2010-04-20 15:47:11 -07003765}
3766EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync);
3767
3768/**
3769 * ring_buffer_read_start - start a non consuming read of the buffer
3770 * @iter: The iterator returned by ring_buffer_read_prepare
3771 *
3772 * This finalizes the startup of an iteration through the buffer.
3773 * The iterator comes from a call to ring_buffer_read_prepare and
3774 * an intervening ring_buffer_read_prepare_sync must have been
3775 * performed.
3776 *
3777 * Must be paired with ring_buffer_finish.
3778 */
3779void
3780ring_buffer_read_start(struct ring_buffer_iter *iter)
3781{
3782 struct ring_buffer_per_cpu *cpu_buffer;
3783 unsigned long flags;
3784
3785 if (!iter)
3786 return;
3787
3788 cpu_buffer = iter->cpu_buffer;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003789
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003790 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
Thomas Gleixner0199c4e2009-12-02 20:01:25 +01003791 arch_spin_lock(&cpu_buffer->lock);
Steven Rostedt642edba2008-11-12 00:01:26 -05003792 rb_iter_reset(iter);
Thomas Gleixner0199c4e2009-12-02 20:01:25 +01003793 arch_spin_unlock(&cpu_buffer->lock);
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003794 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003795}
Robert Richterc4f50182008-12-11 16:49:22 +01003796EXPORT_SYMBOL_GPL(ring_buffer_read_start);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003797
3798/**
3799 * ring_buffer_finish - finish reading the iterator of the buffer
3800 * @iter: The iterator retrieved by ring_buffer_start
3801 *
3802 * This re-enables the recording to the buffer, and frees the
3803 * iterator.
3804 */
3805void
3806ring_buffer_read_finish(struct ring_buffer_iter *iter)
3807{
3808 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3809
Steven Rostedt659f4512012-05-14 17:02:33 -04003810 /*
3811 * Ring buffer is disabled from recording, here's a good place
3812 * to check the integrity of the ring buffer.
3813 */
3814 rb_check_pages(cpu_buffer);
3815
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003816 atomic_dec(&cpu_buffer->record_disabled);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07003817 atomic_dec(&cpu_buffer->buffer->resize_disabled);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003818 kfree(iter);
3819}
Robert Richterc4f50182008-12-11 16:49:22 +01003820EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003821
3822/**
3823 * ring_buffer_read - read the next item in the ring buffer by the iterator
3824 * @iter: The ring buffer iterator
3825 * @ts: The time stamp of the event read.
3826 *
3827 * This reads the next event in the ring buffer and increments the iterator.
3828 */
3829struct ring_buffer_event *
3830ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
3831{
3832 struct ring_buffer_event *event;
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003833 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3834 unsigned long flags;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003835
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003836 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
Steven Rostedt7e9391c2009-09-03 10:02:09 -04003837 again:
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003838 event = rb_iter_peek(iter, ts);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003839 if (!event)
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003840 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003841
Steven Rostedt7e9391c2009-09-03 10:02:09 -04003842 if (event->type_len == RINGBUF_TYPE_PADDING)
3843 goto again;
3844
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003845 rb_advance_iter(iter);
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003846 out:
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003847 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003848
3849 return event;
3850}
Robert Richterc4f50182008-12-11 16:49:22 +01003851EXPORT_SYMBOL_GPL(ring_buffer_read);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003852
3853/**
3854 * ring_buffer_size - return the size of the ring buffer (in bytes)
3855 * @buffer: The ring buffer.
3856 */
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08003857unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003858{
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08003859 /*
3860 * Earlier, this method returned
3861 * BUF_PAGE_SIZE * buffer->nr_pages
3862 * Since the nr_pages field is now removed, we have converted this to
3863 * return the per cpu buffer value.
3864 */
3865 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3866 return 0;
3867
3868 return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003869}
Robert Richterc4f50182008-12-11 16:49:22 +01003870EXPORT_SYMBOL_GPL(ring_buffer_size);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003871
3872static void
3873rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
3874{
Steven Rostedt77ae3652009-03-27 11:00:29 -04003875 rb_head_page_deactivate(cpu_buffer);
3876
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003877 cpu_buffer->head_page
Steven Rostedt3adc54f2009-03-30 15:32:01 -04003878 = list_entry(cpu_buffer->pages, struct buffer_page, list);
Steven Rostedtbf41a152008-10-04 02:00:59 -04003879 local_set(&cpu_buffer->head_page->write, 0);
Steven Rostedt778c55d2009-05-01 18:44:45 -04003880 local_set(&cpu_buffer->head_page->entries, 0);
Steven Rostedtabc9b562008-12-02 15:34:06 -05003881 local_set(&cpu_buffer->head_page->page->commit, 0);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003882
Steven Rostedt6f807ac2008-10-04 02:00:58 -04003883 cpu_buffer->head_page->read = 0;
Steven Rostedtbf41a152008-10-04 02:00:59 -04003884
3885 cpu_buffer->tail_page = cpu_buffer->head_page;
3886 cpu_buffer->commit_page = cpu_buffer->head_page;
3887
3888 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
Vaibhav Nagarnaik5040b4b2012-05-03 18:59:51 -07003889 INIT_LIST_HEAD(&cpu_buffer->new_pages);
Steven Rostedtbf41a152008-10-04 02:00:59 -04003890 local_set(&cpu_buffer->reader_page->write, 0);
Steven Rostedt778c55d2009-05-01 18:44:45 -04003891 local_set(&cpu_buffer->reader_page->entries, 0);
Steven Rostedtabc9b562008-12-02 15:34:06 -05003892 local_set(&cpu_buffer->reader_page->page->commit, 0);
Steven Rostedt6f807ac2008-10-04 02:00:58 -04003893 cpu_buffer->reader_page->read = 0;
Steven Rostedtd7690412008-10-01 00:29:53 -04003894
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07003895 local_set(&cpu_buffer->entries_bytes, 0);
Steven Rostedt77ae3652009-03-27 11:00:29 -04003896 local_set(&cpu_buffer->overrun, 0);
Slava Pestov884bfe82011-07-15 14:23:58 -07003897 local_set(&cpu_buffer->commit_overrun, 0);
3898 local_set(&cpu_buffer->dropped_events, 0);
Steven Rostedte4906ef2009-04-30 20:49:44 -04003899 local_set(&cpu_buffer->entries, 0);
Steven Rostedtfa743952009-06-16 12:37:57 -04003900 local_set(&cpu_buffer->committing, 0);
3901 local_set(&cpu_buffer->commits, 0);
Steven Rostedt77ae3652009-03-27 11:00:29 -04003902 cpu_buffer->read = 0;
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07003903 cpu_buffer->read_bytes = 0;
Steven Rostedt69507c02009-01-21 18:45:57 -05003904
3905 cpu_buffer->write_stamp = 0;
3906 cpu_buffer->read_stamp = 0;
Steven Rostedt77ae3652009-03-27 11:00:29 -04003907
Steven Rostedt66a8cb92010-03-31 13:21:56 -04003908 cpu_buffer->lost_events = 0;
3909 cpu_buffer->last_overrun = 0;
3910
Steven Rostedt77ae3652009-03-27 11:00:29 -04003911 rb_head_page_activate(cpu_buffer);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003912}
3913
3914/**
3915 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
3916 * @buffer: The ring buffer to reset a per cpu buffer of
3917 * @cpu: The CPU buffer to be reset
3918 */
3919void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
3920{
3921 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
3922 unsigned long flags;
3923
Rusty Russell9e01c1b2009-01-01 10:12:22 +10303924 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04003925 return;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003926
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07003927 atomic_inc(&buffer->resize_disabled);
Steven Rostedt41ede232009-05-01 20:26:54 -04003928 atomic_inc(&cpu_buffer->record_disabled);
3929
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07003930 /* Make sure all commits have finished */
3931 synchronize_sched();
3932
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003933 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003934
Steven Rostedt41b6a952009-09-02 09:59:48 -04003935 if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
3936 goto out;
3937
Thomas Gleixner0199c4e2009-12-02 20:01:25 +01003938 arch_spin_lock(&cpu_buffer->lock);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003939
3940 rb_reset_cpu(cpu_buffer);
3941
Thomas Gleixner0199c4e2009-12-02 20:01:25 +01003942 arch_spin_unlock(&cpu_buffer->lock);
Steven Rostedtf83c9d02008-11-11 18:47:44 +01003943
Steven Rostedt41b6a952009-09-02 09:59:48 -04003944 out:
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003945 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
Steven Rostedt41ede232009-05-01 20:26:54 -04003946
3947 atomic_dec(&cpu_buffer->record_disabled);
Vaibhav Nagarnaik83f40312012-05-03 18:59:50 -07003948 atomic_dec(&buffer->resize_disabled);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003949}
Robert Richterc4f50182008-12-11 16:49:22 +01003950EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003951
3952/**
3953 * ring_buffer_reset - reset a ring buffer
3954 * @buffer: The ring buffer to reset all cpu buffers
3955 */
3956void ring_buffer_reset(struct ring_buffer *buffer)
3957{
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003958 int cpu;
3959
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003960 for_each_buffer_cpu(buffer, cpu)
Steven Rostedtd7690412008-10-01 00:29:53 -04003961 ring_buffer_reset_cpu(buffer, cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003962}
Robert Richterc4f50182008-12-11 16:49:22 +01003963EXPORT_SYMBOL_GPL(ring_buffer_reset);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003964
3965/**
3966 * rind_buffer_empty - is the ring buffer empty?
3967 * @buffer: The ring buffer to test
3968 */
3969int ring_buffer_empty(struct ring_buffer *buffer)
3970{
3971 struct ring_buffer_per_cpu *cpu_buffer;
Steven Rostedtd4788202009-06-17 00:39:43 -04003972 unsigned long flags;
Steven Rostedt8d707e82009-06-16 21:22:48 -04003973 int dolock;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003974 int cpu;
Steven Rostedtd4788202009-06-17 00:39:43 -04003975 int ret;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003976
Steven Rostedt8d707e82009-06-16 21:22:48 -04003977 dolock = rb_ok_to_lock();
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003978
3979 /* yes this is racy, but if you don't like the race, lock the buffer */
3980 for_each_buffer_cpu(buffer, cpu) {
3981 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt8d707e82009-06-16 21:22:48 -04003982 local_irq_save(flags);
3983 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003984 raw_spin_lock(&cpu_buffer->reader_lock);
Steven Rostedtd4788202009-06-17 00:39:43 -04003985 ret = rb_per_cpu_empty(cpu_buffer);
Steven Rostedt8d707e82009-06-16 21:22:48 -04003986 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02003987 raw_spin_unlock(&cpu_buffer->reader_lock);
Steven Rostedt8d707e82009-06-16 21:22:48 -04003988 local_irq_restore(flags);
3989
Steven Rostedtd4788202009-06-17 00:39:43 -04003990 if (!ret)
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003991 return 0;
3992 }
Steven Rostedt554f7862009-03-11 22:00:13 -04003993
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003994 return 1;
3995}
Robert Richterc4f50182008-12-11 16:49:22 +01003996EXPORT_SYMBOL_GPL(ring_buffer_empty);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04003997
3998/**
3999 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
4000 * @buffer: The ring buffer
4001 * @cpu: The CPU buffer to test
4002 */
4003int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
4004{
4005 struct ring_buffer_per_cpu *cpu_buffer;
Steven Rostedtd4788202009-06-17 00:39:43 -04004006 unsigned long flags;
Steven Rostedt8d707e82009-06-16 21:22:48 -04004007 int dolock;
Steven Rostedt8aabee52009-03-12 13:13:49 -04004008 int ret;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004009
Rusty Russell9e01c1b2009-01-01 10:12:22 +10304010 if (!cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt8aabee52009-03-12 13:13:49 -04004011 return 1;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004012
Steven Rostedt8d707e82009-06-16 21:22:48 -04004013 dolock = rb_ok_to_lock();
Steven Rostedt554f7862009-03-11 22:00:13 -04004014
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004015 cpu_buffer = buffer->buffers[cpu];
Steven Rostedt8d707e82009-06-16 21:22:48 -04004016 local_irq_save(flags);
4017 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02004018 raw_spin_lock(&cpu_buffer->reader_lock);
Steven Rostedt554f7862009-03-11 22:00:13 -04004019 ret = rb_per_cpu_empty(cpu_buffer);
Steven Rostedt8d707e82009-06-16 21:22:48 -04004020 if (dolock)
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02004021 raw_spin_unlock(&cpu_buffer->reader_lock);
Steven Rostedt8d707e82009-06-16 21:22:48 -04004022 local_irq_restore(flags);
Steven Rostedt554f7862009-03-11 22:00:13 -04004023
4024 return ret;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004025}
Robert Richterc4f50182008-12-11 16:49:22 +01004026EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004027
Steven Rostedt85bac322009-09-04 14:24:40 -04004028#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004029/**
4030 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
4031 * @buffer_a: One buffer to swap with
4032 * @buffer_b: The other buffer to swap with
4033 *
4034 * This function is useful for tracers that want to take a "snapshot"
4035 * of a CPU buffer and has another back up buffer lying around.
4036 * it is expected that the tracer handles the cpu buffer not being
4037 * used at the moment.
4038 */
4039int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
4040 struct ring_buffer *buffer_b, int cpu)
4041{
4042 struct ring_buffer_per_cpu *cpu_buffer_a;
4043 struct ring_buffer_per_cpu *cpu_buffer_b;
Steven Rostedt554f7862009-03-11 22:00:13 -04004044 int ret = -EINVAL;
4045
Rusty Russell9e01c1b2009-01-01 10:12:22 +10304046 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
4047 !cpumask_test_cpu(cpu, buffer_b->cpumask))
Steven Rostedt554f7862009-03-11 22:00:13 -04004048 goto out;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004049
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08004050 cpu_buffer_a = buffer_a->buffers[cpu];
4051 cpu_buffer_b = buffer_b->buffers[cpu];
4052
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004053 /* At least make sure the two buffers are somewhat the same */
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08004054 if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
Steven Rostedt554f7862009-03-11 22:00:13 -04004055 goto out;
4056
4057 ret = -EAGAIN;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004058
Steven Rostedt97b17ef2009-01-21 15:24:56 -05004059 if (ring_buffer_flags != RB_BUFFERS_ON)
Steven Rostedt554f7862009-03-11 22:00:13 -04004060 goto out;
Steven Rostedt97b17ef2009-01-21 15:24:56 -05004061
4062 if (atomic_read(&buffer_a->record_disabled))
Steven Rostedt554f7862009-03-11 22:00:13 -04004063 goto out;
Steven Rostedt97b17ef2009-01-21 15:24:56 -05004064
4065 if (atomic_read(&buffer_b->record_disabled))
Steven Rostedt554f7862009-03-11 22:00:13 -04004066 goto out;
Steven Rostedt97b17ef2009-01-21 15:24:56 -05004067
Steven Rostedt97b17ef2009-01-21 15:24:56 -05004068 if (atomic_read(&cpu_buffer_a->record_disabled))
Steven Rostedt554f7862009-03-11 22:00:13 -04004069 goto out;
Steven Rostedt97b17ef2009-01-21 15:24:56 -05004070
4071 if (atomic_read(&cpu_buffer_b->record_disabled))
Steven Rostedt554f7862009-03-11 22:00:13 -04004072 goto out;
Steven Rostedt97b17ef2009-01-21 15:24:56 -05004073
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004074 /*
4075 * We can't do a synchronize_sched here because this
4076 * function can be called in atomic context.
4077 * Normally this will be called from the same CPU as cpu.
4078 * If not it's up to the caller to protect this.
4079 */
4080 atomic_inc(&cpu_buffer_a->record_disabled);
4081 atomic_inc(&cpu_buffer_b->record_disabled);
4082
Steven Rostedt98277992009-09-02 10:56:15 -04004083 ret = -EBUSY;
4084 if (local_read(&cpu_buffer_a->committing))
4085 goto out_dec;
4086 if (local_read(&cpu_buffer_b->committing))
4087 goto out_dec;
4088
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004089 buffer_a->buffers[cpu] = cpu_buffer_b;
4090 buffer_b->buffers[cpu] = cpu_buffer_a;
4091
4092 cpu_buffer_b->buffer = buffer_a;
4093 cpu_buffer_a->buffer = buffer_b;
4094
Steven Rostedt98277992009-09-02 10:56:15 -04004095 ret = 0;
4096
4097out_dec:
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004098 atomic_dec(&cpu_buffer_a->record_disabled);
4099 atomic_dec(&cpu_buffer_b->record_disabled);
Steven Rostedt554f7862009-03-11 22:00:13 -04004100out:
Steven Rostedt554f7862009-03-11 22:00:13 -04004101 return ret;
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004102}
Robert Richterc4f50182008-12-11 16:49:22 +01004103EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
Steven Rostedt85bac322009-09-04 14:24:40 -04004104#endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
Steven Rostedt7a8e76a2008-09-29 23:02:38 -04004105
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004106/**
4107 * ring_buffer_alloc_read_page - allocate a page to read from buffer
4108 * @buffer: the buffer to allocate for.
4109 *
4110 * This function is used in conjunction with ring_buffer_read_page.
4111 * When reading a full page from the ring buffer, these functions
4112 * can be used to speed up the process. The calling function should
4113 * allocate a few pages first with this function. Then when it
4114 * needs to get pages from the ring buffer, it passes the result
4115 * of this function into ring_buffer_read_page, which will swap
4116 * the page that was allocated, with the read page of the buffer.
4117 *
4118 * Returns:
4119 * The page allocated, or NULL on error.
4120 */
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07004121void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu)
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004122{
Steven Rostedt044fa782008-12-02 23:50:03 -05004123 struct buffer_data_page *bpage;
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07004124 struct page *page;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004125
Vaibhav Nagarnaikd7ec4bf2011-06-07 17:01:42 -07004126 page = alloc_pages_node(cpu_to_node(cpu),
4127 GFP_KERNEL | __GFP_NORETRY, 0);
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07004128 if (!page)
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004129 return NULL;
4130
Vaibhav Nagarnaik7ea59062011-05-03 17:56:42 -07004131 bpage = page_address(page);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004132
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004133 rb_init_page(bpage);
4134
Steven Rostedt044fa782008-12-02 23:50:03 -05004135 return bpage;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004136}
Steven Rostedtd6ce96d2009-05-05 01:15:24 -04004137EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004138
4139/**
4140 * ring_buffer_free_read_page - free an allocated read page
4141 * @buffer: the buffer the page was allocate for
4142 * @data: the page to free
4143 *
4144 * Free a page allocated from ring_buffer_alloc_read_page.
4145 */
4146void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
4147{
4148 free_page((unsigned long)data);
4149}
Steven Rostedtd6ce96d2009-05-05 01:15:24 -04004150EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004151
4152/**
4153 * ring_buffer_read_page - extract a page from the ring buffer
4154 * @buffer: buffer to extract from
4155 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004156 * @len: amount to extract
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004157 * @cpu: the cpu of the buffer to extract
4158 * @full: should the extraction only happen when the page is full.
4159 *
4160 * This function will pull out a page from the ring buffer and consume it.
4161 * @data_page must be the address of the variable that was returned
4162 * from ring_buffer_alloc_read_page. This is because the page might be used
4163 * to swap with a page in the ring buffer.
4164 *
4165 * for example:
Lai Jiangshanb85fa012009-02-09 14:21:14 +08004166 * rpage = ring_buffer_alloc_read_page(buffer);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004167 * if (!rpage)
4168 * return error;
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004169 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
Lai Jiangshan667d2412009-02-09 14:21:17 +08004170 * if (ret >= 0)
4171 * process_page(rpage, ret);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004172 *
4173 * When @full is set, the function will not return true unless
4174 * the writer is off the reader page.
4175 *
4176 * Note: it is up to the calling functions to handle sleeps and wakeups.
4177 * The ring buffer can be used anywhere in the kernel and can not
4178 * blindly call wake_up. The layer that uses the ring buffer must be
4179 * responsible for that.
4180 *
4181 * Returns:
Lai Jiangshan667d2412009-02-09 14:21:17 +08004182 * >=0 if data has been transferred, returns the offset of consumed data.
4183 * <0 if no data has been transferred.
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004184 */
4185int ring_buffer_read_page(struct ring_buffer *buffer,
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004186 void **data_page, size_t len, int cpu, int full)
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004187{
4188 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4189 struct ring_buffer_event *event;
Steven Rostedt044fa782008-12-02 23:50:03 -05004190 struct buffer_data_page *bpage;
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004191 struct buffer_page *reader;
Steven Rostedtff0ff842010-03-31 22:11:42 -04004192 unsigned long missed_events;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004193 unsigned long flags;
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004194 unsigned int commit;
Lai Jiangshan667d2412009-02-09 14:21:17 +08004195 unsigned int read;
Steven Rostedt4f3640f2009-03-03 23:52:42 -05004196 u64 save_timestamp;
Lai Jiangshan667d2412009-02-09 14:21:17 +08004197 int ret = -1;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004198
Steven Rostedt554f7862009-03-11 22:00:13 -04004199 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4200 goto out;
4201
Steven Rostedt474d32b2009-03-03 19:51:40 -05004202 /*
4203 * If len is not big enough to hold the page header, then
4204 * we can not copy anything.
4205 */
4206 if (len <= BUF_PAGE_HDR_SIZE)
Steven Rostedt554f7862009-03-11 22:00:13 -04004207 goto out;
Steven Rostedt474d32b2009-03-03 19:51:40 -05004208
4209 len -= BUF_PAGE_HDR_SIZE;
4210
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004211 if (!data_page)
Steven Rostedt554f7862009-03-11 22:00:13 -04004212 goto out;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004213
Steven Rostedt044fa782008-12-02 23:50:03 -05004214 bpage = *data_page;
4215 if (!bpage)
Steven Rostedt554f7862009-03-11 22:00:13 -04004216 goto out;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004217
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02004218 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004219
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004220 reader = rb_get_reader_page(cpu_buffer);
4221 if (!reader)
Steven Rostedt554f7862009-03-11 22:00:13 -04004222 goto out_unlock;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004223
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004224 event = rb_reader_event(cpu_buffer);
Lai Jiangshan667d2412009-02-09 14:21:17 +08004225
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004226 read = reader->read;
4227 commit = rb_page_commit(reader);
4228
Steven Rostedt66a8cb92010-03-31 13:21:56 -04004229 /* Check if any events were dropped */
Steven Rostedtff0ff842010-03-31 22:11:42 -04004230 missed_events = cpu_buffer->lost_events;
Steven Rostedt66a8cb92010-03-31 13:21:56 -04004231
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004232 /*
Steven Rostedt474d32b2009-03-03 19:51:40 -05004233 * If this page has been partially read or
4234 * if len is not big enough to read the rest of the page or
4235 * a writer is still on the page, then
4236 * we must copy the data from the page to the buffer.
4237 * Otherwise, we can simply swap the page with the one passed in.
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004238 */
Steven Rostedt474d32b2009-03-03 19:51:40 -05004239 if (read || (len < (commit - read)) ||
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004240 cpu_buffer->reader_page == cpu_buffer->commit_page) {
Lai Jiangshan667d2412009-02-09 14:21:17 +08004241 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
Steven Rostedt474d32b2009-03-03 19:51:40 -05004242 unsigned int rpos = read;
4243 unsigned int pos = 0;
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004244 unsigned int size;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004245
4246 if (full)
Steven Rostedt554f7862009-03-11 22:00:13 -04004247 goto out_unlock;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004248
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004249 if (len > (commit - read))
4250 len = (commit - read);
4251
Steven Rostedt69d1b832010-10-07 18:18:05 -04004252 /* Always keep the time extend and data together */
4253 size = rb_event_ts_length(event);
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004254
4255 if (len < size)
Steven Rostedt554f7862009-03-11 22:00:13 -04004256 goto out_unlock;
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004257
Steven Rostedt4f3640f2009-03-03 23:52:42 -05004258 /* save the current timestamp, since the user will need it */
4259 save_timestamp = cpu_buffer->read_stamp;
4260
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004261 /* Need to copy one event at a time */
4262 do {
David Sharpe1e35922010-12-22 16:38:24 -08004263 /* We need the size of one event, because
4264 * rb_advance_reader only advances by one event,
4265 * whereas rb_event_ts_length may include the size of
4266 * one or two events.
4267 * We have already ensured there's enough space if this
4268 * is a time extend. */
4269 size = rb_event_length(event);
Steven Rostedt474d32b2009-03-03 19:51:40 -05004270 memcpy(bpage->data + pos, rpage->data + rpos, size);
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004271
4272 len -= size;
4273
4274 rb_advance_reader(cpu_buffer);
Steven Rostedt474d32b2009-03-03 19:51:40 -05004275 rpos = reader->read;
4276 pos += size;
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004277
Huang Ying18fab912010-07-28 14:14:01 +08004278 if (rpos >= commit)
4279 break;
4280
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004281 event = rb_reader_event(cpu_buffer);
Steven Rostedt69d1b832010-10-07 18:18:05 -04004282 /* Always keep the time extend and data together */
4283 size = rb_event_ts_length(event);
David Sharpe1e35922010-12-22 16:38:24 -08004284 } while (len >= size);
Lai Jiangshan667d2412009-02-09 14:21:17 +08004285
4286 /* update bpage */
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004287 local_set(&bpage->commit, pos);
Steven Rostedt4f3640f2009-03-03 23:52:42 -05004288 bpage->time_stamp = save_timestamp;
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004289
Steven Rostedt474d32b2009-03-03 19:51:40 -05004290 /* we copied everything to the beginning */
4291 read = 0;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004292 } else {
Steven Rostedtafbab762009-05-01 19:40:05 -04004293 /* update the entry counter */
Steven Rostedt77ae3652009-03-27 11:00:29 -04004294 cpu_buffer->read += rb_page_entries(reader);
Vaibhav Nagarnaikc64e1482011-08-16 14:46:16 -07004295 cpu_buffer->read_bytes += BUF_PAGE_SIZE;
Steven Rostedtafbab762009-05-01 19:40:05 -04004296
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004297 /* swap the pages */
Steven Rostedt044fa782008-12-02 23:50:03 -05004298 rb_init_page(bpage);
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004299 bpage = reader->page;
4300 reader->page = *data_page;
4301 local_set(&reader->write, 0);
Steven Rostedt778c55d2009-05-01 18:44:45 -04004302 local_set(&reader->entries, 0);
Steven Rostedtef7a4a12009-03-03 00:27:49 -05004303 reader->read = 0;
Steven Rostedt044fa782008-12-02 23:50:03 -05004304 *data_page = bpage;
Steven Rostedtff0ff842010-03-31 22:11:42 -04004305
4306 /*
4307 * Use the real_end for the data size,
4308 * This gives us a chance to store the lost events
4309 * on the page.
4310 */
4311 if (reader->real_end)
4312 local_set(&bpage->commit, reader->real_end);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004313 }
Lai Jiangshan667d2412009-02-09 14:21:17 +08004314 ret = read;
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004315
Steven Rostedt66a8cb92010-03-31 13:21:56 -04004316 cpu_buffer->lost_events = 0;
Steven Rostedt2711ca22010-05-21 13:32:26 -04004317
4318 commit = local_read(&bpage->commit);
Steven Rostedt66a8cb92010-03-31 13:21:56 -04004319 /*
4320 * Set a flag in the commit field if we lost events
4321 */
Steven Rostedtff0ff842010-03-31 22:11:42 -04004322 if (missed_events) {
Steven Rostedtff0ff842010-03-31 22:11:42 -04004323 /* If there is room at the end of the page to save the
4324 * missed events, then record it there.
4325 */
4326 if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) {
4327 memcpy(&bpage->data[commit], &missed_events,
4328 sizeof(missed_events));
4329 local_add(RB_MISSED_STORED, &bpage->commit);
Steven Rostedt2711ca22010-05-21 13:32:26 -04004330 commit += sizeof(missed_events);
Steven Rostedtff0ff842010-03-31 22:11:42 -04004331 }
Steven Rostedt66a8cb92010-03-31 13:21:56 -04004332 local_add(RB_MISSED_EVENTS, &bpage->commit);
Steven Rostedtff0ff842010-03-31 22:11:42 -04004333 }
Steven Rostedt66a8cb92010-03-31 13:21:56 -04004334
Steven Rostedt2711ca22010-05-21 13:32:26 -04004335 /*
4336 * This page may be off to user land. Zero it out here.
4337 */
4338 if (commit < BUF_PAGE_SIZE)
4339 memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit);
4340
Steven Rostedt554f7862009-03-11 22:00:13 -04004341 out_unlock:
Thomas Gleixner5389f6f2009-07-25 17:13:33 +02004342 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004343
Steven Rostedt554f7862009-03-11 22:00:13 -04004344 out:
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004345 return ret;
4346}
Steven Rostedtd6ce96d2009-05-05 01:15:24 -04004347EXPORT_SYMBOL_GPL(ring_buffer_read_page);
Steven Rostedt8789a9e2008-12-02 15:34:07 -05004348
Steven Rostedt59222ef2009-03-12 11:46:03 -04004349#ifdef CONFIG_HOTPLUG_CPU
Frederic Weisbecker09c9e842009-03-21 04:33:36 +01004350static int rb_cpu_notify(struct notifier_block *self,
4351 unsigned long action, void *hcpu)
Steven Rostedt554f7862009-03-11 22:00:13 -04004352{
4353 struct ring_buffer *buffer =
4354 container_of(self, struct ring_buffer, cpu_notify);
4355 long cpu = (long)hcpu;
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08004356 int cpu_i, nr_pages_same;
4357 unsigned int nr_pages;
Steven Rostedt554f7862009-03-11 22:00:13 -04004358
4359 switch (action) {
4360 case CPU_UP_PREPARE:
4361 case CPU_UP_PREPARE_FROZEN:
Rusty Russell3f237a72009-06-12 21:15:30 +09304362 if (cpumask_test_cpu(cpu, buffer->cpumask))
Steven Rostedt554f7862009-03-11 22:00:13 -04004363 return NOTIFY_OK;
4364
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08004365 nr_pages = 0;
4366 nr_pages_same = 1;
4367 /* check if all cpu sizes are same */
4368 for_each_buffer_cpu(buffer, cpu_i) {
4369 /* fill in the size from first enabled cpu */
4370 if (nr_pages == 0)
4371 nr_pages = buffer->buffers[cpu_i]->nr_pages;
4372 if (nr_pages != buffer->buffers[cpu_i]->nr_pages) {
4373 nr_pages_same = 0;
4374 break;
4375 }
4376 }
4377 /* allocate minimum pages, user can later expand it */
4378 if (!nr_pages_same)
4379 nr_pages = 2;
Steven Rostedt554f7862009-03-11 22:00:13 -04004380 buffer->buffers[cpu] =
Vaibhav Nagarnaik438ced12012-02-02 12:00:41 -08004381 rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
Steven Rostedt554f7862009-03-11 22:00:13 -04004382 if (!buffer->buffers[cpu]) {
4383 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
4384 cpu);
4385 return NOTIFY_OK;
4386 }
4387 smp_wmb();
Rusty Russell3f237a72009-06-12 21:15:30 +09304388 cpumask_set_cpu(cpu, buffer->cpumask);
Steven Rostedt554f7862009-03-11 22:00:13 -04004389 break;
4390 case CPU_DOWN_PREPARE:
4391 case CPU_DOWN_PREPARE_FROZEN:
4392 /*
4393 * Do nothing.
4394 * If we were to free the buffer, then the user would
4395 * lose any trace that was in the buffer.
4396 */
4397 break;
4398 default:
4399 break;
4400 }
4401 return NOTIFY_OK;
4402}
4403#endif