kernel/trace/ring_buffer.c

   1 /*
   2  * Generic ring buffer
   3  *
   4  * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
   5  */
   6 #include <linux/ring_buffer.h>
   7 #include <linux/trace_clock.h>
   8 #include <linux/spinlock.h>
   9 #include <linux/debugfs.h>
  10 #include <linux/uaccess.h>
  11 #include <linux/hardirq.h>
  12 #include <linux/kmemcheck.h>
  13 #include <linux/module.h>
  14 #include <linux/percpu.h>
  15 #include <linux/mutex.h>
  16 #include <linux/slab.h>
  17 #include <linux/init.h>
  18 #include <linux/hash.h>
  19 #include <linux/list.h>
  20 #include <linux/cpu.h>
  21 #include <linux/fs.h>
  22
  23 #include <asm/local.h>
  24 #include "trace.h"
  25
  26 /*
  27  * The ring buffer header is special. We must manually up keep it.
  28  */
  29 int ring_buffer_print_entry_header(struct trace_seq *s)
  30 {
  31         int ret;
  32
  33         ret = trace_seq_printf(s, "# compressed entry header\n");
  34         ret = trace_seq_printf(s, "\ttype_len    :    5 bits\n");
  35         ret = trace_seq_printf(s, "\ttime_delta  :   27 bits\n");
  36         ret = trace_seq_printf(s, "\tarray       :   32 bits\n");
  37         ret = trace_seq_printf(s, "\n");
  38         ret = trace_seq_printf(s, "\tpadding     : type == %d\n",
  39                                RINGBUF_TYPE_PADDING);
  40         ret = trace_seq_printf(s, "\ttime_extend : type == %d\n",
  41                                RINGBUF_TYPE_TIME_EXTEND);
  42         ret = trace_seq_printf(s, "\tdata max type_len  == %d\n",
  43                                RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
  44
  45         return ret;
  46 }
  47
  48 /*
  49  * The ring buffer is made up of a list of pages. A separate list of pages is
  50  * allocated for each CPU. A writer may only write to a buffer that is
  51  * associated with the CPU it is currently executing on.  A reader may read
  52  * from any per cpu buffer.
  53  *
  54  * The reader is special. For each per cpu buffer, the reader has its own
  55  * reader page. When a reader has read the entire reader page, this reader
  56  * page is swapped with another page in the ring buffer.
  57  *
  58  * Now, as long as the writer is off the reader page, the reader can do what
  59  * ever it wants with that page. The writer will never write to that page
  60  * again (as long as it is out of the ring buffer).
  61  *
  62  * Here's some silly ASCII art.
  63  *
  64  *   +------+
  65  *   |reader|          RING BUFFER
  66  *   |page  |
  67  *   +------+        +---+   +---+   +---+
  68  *                   |   |-->|   |-->|   |
  69  *                   +---+   +---+   +---+
  70  *                     ^               |
  71  *                     |               |
  72  *                     +---------------+
  73  *
  74  *
  75  *   +------+
  76  *   |reader|          RING BUFFER
  77  *   |page  |------------------v
  78  *   +------+        +---+   +---+   +---+
  79  *                   |   |-->|   |-->|   |
  80  *                   +---+   +---+   +---+
  81  *                     ^               |
  82  *                     |               |
  83  *                     +---------------+
  84  *
  85  *
  86  *   +------+
  87  *   |reader|          RING BUFFER
  88  *   |page  |------------------v
  89  *   +------+        +---+   +---+   +---+
  90  *      ^            |   |-->|   |-->|   |
  91  *      |            +---+   +---+   +---+
  92  *      |                              |
  93  *      |                              |
  94  *      +------------------------------+
  95  *
  96  *
  97  *   +------+
  98  *   |buffer|          RING BUFFER
  99  *   |page  |------------------v
 100  *   +------+        +---+   +---+   +---+
 101  *      ^            |   |   |   |-->|   |
 102  *      |   New      +---+   +---+   +---+
 103  *      |  Reader------^               |
 104  *      |   page                       |
 105  *      +------------------------------+
 106  *
 107  *
 108  * After we make this swap, the reader can hand this page off to the splice
 109  * code and be done with it. It can even allocate a new page if it needs to
 110  * and swap that into the ring buffer.
 111  *
 112  * We will be using cmpxchg soon to make all this lockless.
 113  *
 114  */
 115
 116 /*
 117  * A fast way to enable or disable all ring buffers is to
 118  * call tracing_on or tracing_off. Turning off the ring buffers
 119  * prevents all ring buffers from being recorded to.
 120  * Turning this switch on, makes it OK to write to the
 121  * ring buffer, if the ring buffer is enabled itself.
 122  *
 123  * There's three layers that must be on in order to write
 124  * to the ring buffer.
 125  *
 126  * 1) This global flag must be set.
 127  * 2) The ring buffer must be enabled for recording.
 128  * 3) The per cpu buffer must be enabled for recording.
 129  *
 130  * In case of an anomaly, this global flag has a bit set that
 131  * will permantly disable all ring buffers.
 132  */
 133
 134 /*
 135  * Global flag to disable all recording to ring buffers
 136  *  This has two bits: ON, DISABLED
 137  *
 138  *  ON   DISABLED
 139  * ---- ----------
 140  *   0      0        : ring buffers are off
 141  *   1      0        : ring buffers are on
 142  *   X      1        : ring buffers are permanently disabled
 143  */
 144
 145 enum {
 146         RB_BUFFERS_ON_BIT       = 0,
 147         RB_BUFFERS_DISABLED_BIT = 1,
 148 };
 149
 150 enum {
 151         RB_BUFFERS_ON           = 1 << RB_BUFFERS_ON_BIT,
 152         RB_BUFFERS_DISABLED     = 1 << RB_BUFFERS_DISABLED_BIT,
 153 };
 154
 155 static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON;
 156
 157 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
 158
 159 /**
 160  * tracing_on - enable all tracing buffers
 161  *
 162  * This function enables all tracing buffers that may have been
 163  * disabled with tracing_off.
 164  */
 165 void tracing_on(void)
 166 {
 167         set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
 168 }
 169 EXPORT_SYMBOL_GPL(tracing_on);
 170
 171 /**
 172  * tracing_off - turn off all tracing buffers
 173  *
 174  * This function stops all tracing buffers from recording data.
 175  * It does not disable any overhead the tracers themselves may
 176  * be causing. This function simply causes all recording to
 177  * the ring buffers to fail.
 178  */
 179 void tracing_off(void)
 180 {
 181         clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
 182 }
 183 EXPORT_SYMBOL_GPL(tracing_off);
 184
 185 /**
 186  * tracing_off_permanent - permanently disable ring buffers
 187  *
 188  * This function, once called, will disable all ring buffers
 189  * permanently.
 190  */
 191 void tracing_off_permanent(void)
 192 {
 193         set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags);
 194 }
 195
 196 /**
 197  * tracing_is_on - show state of ring buffers enabled
 198  */
 199 int tracing_is_on(void)
 200 {
 201         return ring_buffer_flags == RB_BUFFERS_ON;
 202 }
 203 EXPORT_SYMBOL_GPL(tracing_is_on);
 204
 205 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
 206 #define RB_ALIGNMENT            4U
 207 #define RB_MAX_SMALL_DATA       (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
 208 #define RB_EVNT_MIN_SIZE        8U      /* two 32bit words */
 209
 210 #if !defined(CONFIG_64BIT) || defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
 211 # define RB_FORCE_8BYTE_ALIGNMENT       0
 212 # define RB_ARCH_ALIGNMENT              RB_ALIGNMENT
 213 #else
 214 # define RB_FORCE_8BYTE_ALIGNMENT       1
 215 # define RB_ARCH_ALIGNMENT              8U
 216 #endif
 217
 218 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
 219 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
 220
 221 enum {
 222         RB_LEN_TIME_EXTEND = 8,
 223         RB_LEN_TIME_STAMP = 16,
 224 };
 225
 226 #define skip_time_extend(event) \
 227         ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
 228
 229 static inline int rb_null_event(struct ring_buffer_event *event)
 230 {
 231         return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
 232 }
 233
 234 static void rb_event_set_padding(struct ring_buffer_event *event)
 235 {
 236         /* padding has a NULL time_delta */
 237         event->type_len = RINGBUF_TYPE_PADDING;
 238         event->time_delta = 0;
 239 }
 240
 241 static unsigned
 242 rb_event_data_length(struct ring_buffer_event *event)
 243 {
 244         unsigned length;
 245
 246         if (event->type_len)
 247                 length = event->type_len * RB_ALIGNMENT;
 248         else
 249                 length = event->array[0];
 250         return length + RB_EVNT_HDR_SIZE;
 251 }
 252
 253 /*
 254  * Return the length of the given event. Will return
 255  * the length of the time extend if the event is a
 256  * time extend.
 257  */
 258 static inline unsigned
 259 rb_event_length(struct ring_buffer_event *event)
 260 {
 261         switch (event->type_len) {
 262         case RINGBUF_TYPE_PADDING:
 263                 if (rb_null_event(event))
 264                         /* undefined */
 265                         return -1;
 266                 return  event->array[0] + RB_EVNT_HDR_SIZE;
 267
 268         case RINGBUF_TYPE_TIME_EXTEND:
 269                 return RB_LEN_TIME_EXTEND;
 270
 271         case RINGBUF_TYPE_TIME_STAMP:
 272                 return RB_LEN_TIME_STAMP;
 273
 274         case RINGBUF_TYPE_DATA:
 275                 return rb_event_data_length(event);
 276         default:
 277                 BUG();
 278         }
 279         /* not hit */
 280         return 0;
 281 }
 282
 283 /*
 284  * Return total length of time extend and data,
 285  *   or just the event length for all other events.
 286  */
 287 static inline unsigned
 288 rb_event_ts_length(struct ring_buffer_event *event)
 289 {
 290         unsigned len = 0;
 291
 292         if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) {
 293                 /* time extends include the data event after it */
 294                 len = RB_LEN_TIME_EXTEND;
 295                 event = skip_time_extend(event);
 296         }
 297         return len + rb_event_length(event);
 298 }
 299
 300 /**
 301  * ring_buffer_event_length - return the length of the event
 302  * @event: the event to get the length of
 303  *
 304  * Returns the size of the data load of a data event.
 305  * If the event is something other than a data event, it
 306  * returns the size of the event itself. With the exception
 307  * of a TIME EXTEND, where it still returns the size of the
 308  * data load of the data event after it.
 309  */
 310 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
 311 {
 312         unsigned length;
 313
 314         if (event->type_len == RINGBUF_TYPE_TIME_EXTEND)
 315                 event = skip_time_extend(event);
 316
 317         length = rb_event_length(event);
 318         if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
 319                 return length;
 320         length -= RB_EVNT_HDR_SIZE;
 321         if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
 322                 length -= sizeof(event->array[0]);
 323         return length;
 324 }
 325 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
 326
 327 /* inline for ring buffer fast paths */
 328 static void *
 329 rb_event_data(struct ring_buffer_event *event)
 330 {
 331         if (event->type_len == RINGBUF_TYPE_TIME_EXTEND)
 332                 event = skip_time_extend(event);
 333         BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
 334         /* If length is in len field, then array[0] has the data */
 335         if (event->type_len)
 336                 return (void *)&event->array[0];
 337         /* Otherwise length is in array[0] and array[1] has the data */
 338         return (void *)&event->array[1];
 339 }
 340
 341 /**
 342  * ring_buffer_event_data - return the data of the event
 343  * @event: the event to get the data from
 344  */
 345 void *ring_buffer_event_data(struct ring_buffer_event *event)
 346 {
 347         return rb_event_data(event);
 348 }
 349 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
 350
 351 #define for_each_buffer_cpu(buffer, cpu)                \
 352         for_each_cpu(cpu, buffer->cpumask)
 353
 354 #define TS_SHIFT        27
 355 #define TS_MASK         ((1ULL << TS_SHIFT) - 1)
 356 #define TS_DELTA_TEST   (~TS_MASK)
 357
 358 /* Flag when events were overwritten */
 359 #define RB_MISSED_EVENTS        (1 << 31)
 360 /* Missed count stored at end */
 361 #define RB_MISSED_STORED        (1 << 30)
 362
 363 struct buffer_data_page {
 364         u64              time_stamp;    /* page time stamp */
 365         local_t          commit;        /* write committed index */
 366         unsigned char    data[];        /* data of buffer page */
 367 };
 368
 369 /*
 370  * Note, the buffer_page list must be first. The buffer pages
 371  * are allocated in cache lines, which means that each buffer
 372  * page will be at the beginning of a cache line, and thus
 373  * the least significant bits will be zero. We use this to
 374  * add flags in the list struct pointers, to make the ring buffer
 375  * lockless.
 376  */
 377 struct buffer_page {
 378         struct list_head list;          /* list of buffer pages */
 379         local_t          write;         /* index for next write */
 380         unsigned         read;          /* index for next read */
 381         local_t          entries;       /* entries on this page */
 382         unsigned long    real_end;      /* real end of data */
 383         struct buffer_data_page *page;  /* Actual data page */
 384 };
 385
 386 /*
 387  * The buffer page counters, write and entries, must be reset
 388  * atomically when crossing page boundaries. To synchronize this
 389  * update, two counters are inserted into the number. One is
 390  * the actual counter for the write position or count on the page.
 391  *
 392  * The other is a counter of updaters. Before an update happens
 393  * the update partition of the counter is incremented. This will
 394  * allow the updater to update the counter atomically.
 395  *
 396  * The counter is 20 bits, and the state data is 12.
 397  */
 398 #define RB_WRITE_MASK           0xfffff
 399 #define RB_WRITE_INTCNT         (1 << 20)
 400
 401 static void rb_init_page(struct buffer_data_page *bpage)
 402 {
 403         local_set(&bpage->commit, 0);
 404 }
 405
 406 /**
 407  * ring_buffer_page_len - the size of data on the page.
 408  * @page: The page to read
 409  *
 410  * Returns the amount of data on the page, including buffer page header.
 411  */
 412 size_t ring_buffer_page_len(void *page)
 413 {
 414         return local_read(&((struct buffer_data_page *)page)->commit)
 415                 + BUF_PAGE_HDR_SIZE;
 416 }
 417
 418 /*
 419  * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
 420  * this issue out.
 421  */
 422 static void free_buffer_page(struct buffer_page *bpage)
 423 {
 424         free_page((unsigned long)bpage->page);
 425         kfree(bpage);
 426 }
 427
 428 /*
 429  * We need to fit the time_stamp delta into 27 bits.
 430  */
 431 static inline int test_time_stamp(u64 delta)
 432 {
 433         if (delta & TS_DELTA_TEST)
 434                 return 1;
 435         return 0;
 436 }
 437
 438 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
 439
 440 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
 441 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
 442
 443 int ring_buffer_print_page_header(struct trace_seq *s)
 444 {
 445         struct buffer_data_page field;
 446         int ret;
 447
 448         ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t"
 449                                "offset:0;\tsize:%u;\tsigned:%u;\n",
 450                                (unsigned int)sizeof(field.time_stamp),
 451                                (unsigned int)is_signed_type(u64));
 452
 453         ret = trace_seq_printf(s, "\tfield: local_t commit;\t"
 454                                "offset:%u;\tsize:%u;\tsigned:%u;\n",
 455                                (unsigned int)offsetof(typeof(field), commit),
 456                                (unsigned int)sizeof(field.commit),
 457                                (unsigned int)is_signed_type(long));
 458
 459         ret = trace_seq_printf(s, "\tfield: int overwrite;\t"
 460                                "offset:%u;\tsize:%u;\tsigned:%u;\n",
 461                                (unsigned int)offsetof(typeof(field), commit),
 462                                1,
 463                                (unsigned int)is_signed_type(long));
 464
 465         ret = trace_seq_printf(s, "\tfield: char data;\t"
 466                                "offset:%u;\tsize:%u;\tsigned:%u;\n",
 467                                (unsigned int)offsetof(typeof(field), data),
 468                                (unsigned int)BUF_PAGE_SIZE,
 469                                (unsigned int)is_signed_type(char));
 470
 471         return ret;
 472 }
 473
 474 /*
 475  * head_page == tail_page && head == tail then buffer is empty.
 476  */
 477 struct ring_buffer_per_cpu {
 478         int                             cpu;
 479         atomic_t                        record_disabled;
 480         struct ring_buffer              *buffer;
 481         raw_spinlock_t                  reader_lock;    /* serialize readers */
 482         arch_spinlock_t                 lock;
 483         struct lock_class_key           lock_key;
 484         struct list_head                *pages;
 485         struct buffer_page              *head_page;     /* read from head */
 486         struct buffer_page              *tail_page;     /* write to tail */
 487         struct buffer_page              *commit_page;   /* committed pages */
 488         struct buffer_page              *reader_page;
 489         unsigned long                   lost_events;
 490         unsigned long                   last_overrun;
 491         local_t                         entries_bytes;
 492         local_t                         commit_overrun;
 493         local_t                         overrun;
 494         local_t                         entries;
 495         local_t                         committing;
 496         local_t                         commits;
 497         unsigned long                   read;
 498         unsigned long                   read_bytes;
 499         u64                             write_stamp;
 500         u64                             read_stamp;
 501 };
 502
 503 struct ring_buffer {
 504         unsigned                        pages;
 505         unsigned                        flags;
 506         int                             cpus;
 507         atomic_t                        record_disabled;
 508         cpumask_var_t                   cpumask;
 509
 510         struct lock_class_key           *reader_lock_key;
 511
 512         struct mutex                    mutex;
 513
 514         struct ring_buffer_per_cpu      **buffers;
 515
 516 #ifdef CONFIG_HOTPLUG_CPU
 517         struct notifier_block           cpu_notify;
 518 #endif
 519         u64                             (*clock)(void);
 520 };
 521
 522 struct ring_buffer_iter {
 523         struct ring_buffer_per_cpu      *cpu_buffer;
 524         unsigned long                   head;
 525         struct buffer_page              *head_page;
 526         struct buffer_page              *cache_reader_page;
 527         unsigned long                   cache_read;
 528         u64                             read_stamp;
 529 };
 530
 531 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
 532 #define RB_WARN_ON(b, cond)                                             \
 533         ({                                                              \
 534                 int _____ret = unlikely(cond);                          \
 535                 if (_____ret) {                                         \
 536                         if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
 537                                 struct ring_buffer_per_cpu *__b =       \
 538                                         (void *)b;                      \
 539                                 atomic_inc(&__b->buffer->record_disabled); \
 540                         } else                                          \
 541                                 atomic_inc(&b->record_disabled);        \
 542                         WARN_ON(1);                                     \
 543                 }                                                       \
 544                 _____ret;                                               \
 545         })
 546
 547 /* Up this if you want to test the TIME_EXTENTS and normalization */
 548 #define DEBUG_SHIFT 0
 549
 550 static inline u64 rb_time_stamp(struct ring_buffer *buffer)
 551 {
 552         /* shift to debug/test normalization and TIME_EXTENTS */
 553         return buffer->clock() << DEBUG_SHIFT;
 554 }
 555
 556 u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
 557 {
 558         u64 time;
 559
 560         preempt_disable_notrace();
 561         time = rb_time_stamp(buffer);
 562         preempt_enable_no_resched_notrace();
 563
 564         return time;
 565 }
 566 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
 567
 568 void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer,
 569                                       int cpu, u64 *ts)
 570 {
 571         /* Just stupid testing the normalize function and deltas */
 572         *ts >>= DEBUG_SHIFT;
 573 }
 574 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
 575
 576 /*
 577  * Making the ring buffer lockless makes things tricky.
 578  * Although writes only happen on the CPU that they are on,
 579  * and they only need to worry about interrupts. Reads can
 580  * happen on any CPU.
 581  *
 582  * The reader page is always off the ring buffer, but when the
 583  * reader finishes with a page, it needs to swap its page with
 584  * a new one from the buffer. The reader needs to take from
 585  * the head (writes go to the tail). But if a writer is in overwrite
 586  * mode and wraps, it must push the head page forward.
 587  *
 588  * Here lies the problem.
 589  *
 590  * The reader must be careful to replace only the head page, and
 591  * not another one. As described at the top of the file in the
 592  * ASCII art, the reader sets its old page to point to the next
 593  * page after head. It then sets the page after head to point to
 594  * the old reader page. But if the writer moves the head page
 595  * during this operation, the reader could end up with the tail.
 596  *
 597  * We use cmpxchg to help prevent this race. We also do something
 598  * special with the page before head. We set the LSB to 1.
 599  *
 600  * When the writer must push the page forward, it will clear the
 601  * bit that points to the head page, move the head, and then set
 602  * the bit that points to the new head page.
 603  *
 604  * We also don't want an interrupt coming in and moving the head
 605  * page on another writer. Thus we use the second LSB to catch
 606  * that too. Thus:
 607  *
 608  * head->list->prev->next        bit 1          bit 0
 609  *                              -------        -------
 610  * Normal page                     0              0
 611  * Points to head page             0              1
 612  * New head page                   1              0
 613  *
 614  * Note we can not trust the prev pointer of the head page, because:
 615  *
 616  * +----+       +-----+        +-----+
 617  * |    |------>|  T  |---X--->|  N  |
 618  * |    |<------|     |        |     |
 619  * +----+       +-----+        +-----+
 620  *   ^                           ^ |
 621  *   |          +-----+          | |
 622  *   +----------|  R  |----------+ |
 623  *              |     |<-----------+
 624  *              +-----+
 625  *
 626  * Key:  ---X-->  HEAD flag set in pointer
 627  *         T      Tail page
 628  *         R      Reader page
 629  *         N      Next page
 630  *
 631  * (see __rb_reserve_next() to see where this happens)
 632  *
 633  *  What the above shows is that the reader just swapped out
 634  *  the reader page with a page in the buffer, but before it
 635  *  could make the new header point back to the new page added
 636  *  it was preempted by a writer. The writer moved forward onto
 637  *  the new page added by the reader and is about to move forward
 638  *  again.
 639  *
 640  *  You can see, it is legitimate for the previous pointer of
 641  *  the head (or any page) not to point back to itself. But only
 642  *  temporarially.
 643  */
 644
 645 #define RB_PAGE_NORMAL          0UL
 646 #define RB_PAGE_HEAD            1UL
 647 #define RB_PAGE_UPDATE          2UL
 648
 649
 650 #define RB_FLAG_MASK            3UL
 651
 652 /* PAGE_MOVED is not part of the mask */
 653 #define RB_PAGE_MOVED           4UL
 654
 655 /*
 656  * rb_list_head - remove any bit
 657  */
 658 static struct list_head *rb_list_head(struct list_head *list)
 659 {
 660         unsigned long val = (unsigned long)list;
 661
 662         return (struct list_head *)(val & ~RB_FLAG_MASK);
 663 }
 664
 665 /*
 666  * rb_is_head_page - test if the given page is the head page
 667  *
 668  * Because the reader may move the head_page pointer, we can
 669  * not trust what the head page is (it may be pointing to
 670  * the reader page). But if the next page is a header page,
 671  * its flags will be non zero.
 672  */
 673 static inline int
 674 rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer,
 675                 struct buffer_page *page, struct list_head *list)
 676 {
 677         unsigned long val;
 678
 679         val = (unsigned long)list->next;
 680
 681         if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
 682                 return RB_PAGE_MOVED;
 683
 684         return val & RB_FLAG_MASK;
 685 }
 686
 687 /*
 688  * rb_is_reader_page
 689  *
 690  * The unique thing about the reader page, is that, if the
 691  * writer is ever on it, the previous pointer never points
 692  * back to the reader page.
 693  */
 694 static int rb_is_reader_page(struct buffer_page *page)
 695 {
 696         struct list_head *list = page->list.prev;
 697
 698         return rb_list_head(list->next) != &page->list;
 699 }
 700
 701 /*
 702  * rb_set_list_to_head - set a list_head to be pointing to head.
 703  */
 704 static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer,
 705                                 struct list_head *list)
 706 {
 707         unsigned long *ptr;
 708
 709         ptr = (unsigned long *)&list->next;
 710         *ptr |= RB_PAGE_HEAD;
 711         *ptr &= ~RB_PAGE_UPDATE;
 712 }
 713
 714 /*
 715  * rb_head_page_activate - sets up head page
 716  */
 717 static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
 718 {
 719         struct buffer_page *head;
 720
 721         head = cpu_buffer->head_page;
 722         if (!head)
 723                 return;
 724
 725         /*
 726          * Set the previous list pointer to have the HEAD flag.
 727          */
 728         rb_set_list_to_head(cpu_buffer, head->list.prev);
 729 }
 730
 731 static void rb_list_head_clear(struct list_head *list)
 732 {
 733         unsigned long *ptr = (unsigned long *)&list->next;
 734
 735         *ptr &= ~RB_FLAG_MASK;
 736 }
 737
 738 /*
 739  * rb_head_page_dactivate - clears head page ptr (for free list)
 740  */
 741 static void
 742 rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
 743 {
 744         struct list_head *hd;
 745
 746         /* Go through the whole list and clear any pointers found. */
 747         rb_list_head_clear(cpu_buffer->pages);
 748
 749         list_for_each(hd, cpu_buffer->pages)
 750                 rb_list_head_clear(hd);
 751 }
 752
 753 static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
 754                             struct buffer_page *head,
 755                             struct buffer_page *prev,
 756                             int old_flag, int new_flag)
 757 {
 758         struct list_head *list;
 759         unsigned long val = (unsigned long)&head->list;
 760         unsigned long ret;
 761
 762         list = &prev->list;
 763
 764         val &= ~RB_FLAG_MASK;
 765
 766         ret = cmpxchg((unsigned long *)&list->next,
 767                       val | old_flag, val | new_flag);
 768
 769         /* check if the reader took the page */
 770         if ((ret & ~RB_FLAG_MASK) != val)
 771                 return RB_PAGE_MOVED;
 772
 773         return ret & RB_FLAG_MASK;
 774 }
 775
 776 static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
 777                                    struct buffer_page *head,
 778                                    struct buffer_page *prev,
 779                                    int old_flag)
 780 {
 781         return rb_head_page_set(cpu_buffer, head, prev,
 782                                 old_flag, RB_PAGE_UPDATE);
 783 }
 784
 785 static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
 786                                  struct buffer_page *head,
 787                                  struct buffer_page *prev,
 788                                  int old_flag)
 789 {
 790         return rb_head_page_set(cpu_buffer, head, prev,
 791                                 old_flag, RB_PAGE_HEAD);
 792 }
 793
 794 static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
 795                                    struct buffer_page *head,
 796                                    struct buffer_page *prev,
 797                                    int old_flag)
 798 {
 799         return rb_head_page_set(cpu_buffer, head, prev,
 800                                 old_flag, RB_PAGE_NORMAL);
 801 }
 802
 803 static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
 804                                struct buffer_page **bpage)
 805 {
 806         struct list_head *p = rb_list_head((*bpage)->list.next);
 807
 808         *bpage = list_entry(p, struct buffer_page, list);
 809 }
 810
 811 static struct buffer_page *
 812 rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
 813 {
 814         struct buffer_page *head;
 815         struct buffer_page *page;
 816         struct list_head *list;
 817         int i;
 818
 819         if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
 820                 return NULL;
 821
 822         /* sanity check */
 823         list = cpu_buffer->pages;
 824         if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
 825                 return NULL;
 826
 827         page = head = cpu_buffer->head_page;
 828         /*
 829          * It is possible that the writer moves the header behind
 830          * where we started, and we miss in one loop.
 831          * A second loop should grab the header, but we'll do
 832          * three loops just because I'm paranoid.
 833          */
 834         for (i = 0; i < 3; i++) {
 835                 do {
 836                         if (rb_is_head_page(cpu_buffer, page, page->list.prev)) {
 837                                 cpu_buffer->head_page = page;
 838                                 return page;
 839                         }
 840                         rb_inc_page(cpu_buffer, &page);
 841                 } while (page != head);
 842         }
 843
 844         RB_WARN_ON(cpu_buffer, 1);
 845
 846         return NULL;
 847 }
 848
 849 static int rb_head_page_replace(struct buffer_page *old,
 850                                 struct buffer_page *new)
 851 {
 852         unsigned long *ptr = (unsigned long *)&old->list.prev->next;
 853         unsigned long val;
 854         unsigned long ret;
 855
 856         val = *ptr & ~RB_FLAG_MASK;
 857         val |= RB_PAGE_HEAD;
 858
 859         ret = cmpxchg(ptr, val, (unsigned long)&new->list);
 860
 861         return ret == val;
 862 }
 863
 864 /*
 865  * rb_tail_page_update - move the tail page forward
 866  *
 867  * Returns 1 if moved tail page, 0 if someone else did.
 868  */
 869 static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
 870                                struct buffer_page *tail_page,
 871                                struct buffer_page *next_page)
 872 {
 873         struct buffer_page *old_tail;
 874         unsigned long old_entries;
 875         unsigned long old_write;
 876         int ret = 0;
 877
 878         /*
 879          * The tail page now needs to be moved forward.
 880          *
 881          * We need to reset the tail page, but without messing
 882          * with possible erasing of data brought in by interrupts
 883          * that have moved the tail page and are currently on it.
 884          *
 885          * We add a counter to the write field to denote this.
 886          */
 887         old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write);
 888         old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries);
 889
 890         /*
 891          * Just make sure we have seen our old_write and synchronize
 892          * with any interrupts that come in.
 893          */
 894         barrier();
 895
 896         /*
 897          * If the tail page is still the same as what we think
 898          * it is, then it is up to us to update the tail
 899          * pointer.
 900          */
 901         if (tail_page == cpu_buffer->tail_page) {
 902                 /* Zero the write counter */
 903                 unsigned long val = old_write & ~RB_WRITE_MASK;
 904                 unsigned long eval = old_entries & ~RB_WRITE_MASK;
 905
 906                 /*
 907                  * This will only succeed if an interrupt did
 908                  * not come in and change it. In which case, we
 909                  * do not want to modify it.
 910                  *
 911                  * We add (void) to let the compiler know that we do not care
 912                  * about the return value of these functions. We use the
 913                  * cmpxchg to only update if an interrupt did not already
 914                  * do it for us. If the cmpxchg fails, we don't care.
 915                  */
 916                 (void)local_cmpxchg(&next_page->write, old_write, val);
 917                 (void)local_cmpxchg(&next_page->entries, old_entries, eval);
 918
 919                 /*
 920                  * No need to worry about races with clearing out the commit.
 921                  * it only can increment when a commit takes place. But that
 922                  * only happens in the outer most nested commit.
 923                  */
 924                 local_set(&next_page->page->commit, 0);
 925
 926                 old_tail = cmpxchg(&cpu_buffer->tail_page,
 927                                    tail_page, next_page);
 928
 929                 if (old_tail == tail_page)
 930                         ret = 1;
 931         }
 932
 933         return ret;
 934 }
 935
 936 static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
 937                           struct buffer_page *bpage)
 938 {
 939         unsigned long val = (unsigned long)bpage;
 940
 941         if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK))
 942                 return 1;
 943
 944         return 0;
 945 }
 946
 947 /**
 948  * rb_check_list - make sure a pointer to a list has the last bits zero
 949  */
 950 static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer,
 951                          struct list_head *list)
 952 {
 953         if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev))
 954                 return 1;
 955         if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next))
 956                 return 1;
 957         return 0;
 958 }
 959
 960 /**
 961  * check_pages - integrity check of buffer pages
 962  * @cpu_buffer: CPU buffer with pages to test
 963  *
 964  * As a safety measure we check to make sure the data pages have not
 965  * been corrupted.
 966  */
 967 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
 968 {
 969         struct list_head *head = cpu_buffer->pages;
 970         struct buffer_page *bpage, *tmp;
 971
 972         rb_head_page_deactivate(cpu_buffer);
 973
 974         if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
 975                 return -1;
 976         if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
 977                 return -1;
 978
 979         if (rb_check_list(cpu_buffer, head))
 980                 return -1;
 981
 982         list_for_each_entry_safe(bpage, tmp, head, list) {
 983                 if (RB_WARN_ON(cpu_buffer,
 984                                bpage->list.next->prev != &bpage->list))
 985                         return -1;
 986                 if (RB_WARN_ON(cpu_buffer,
 987                                bpage->list.prev->next != &bpage->list))
 988                         return -1;
 989                 if (rb_check_list(cpu_buffer, &bpage->list))
 990                         return -1;
 991         }
 992
 993         rb_head_page_activate(cpu_buffer);
 994
 995         return 0;
 996 }
 997
 998 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
 999                              unsigned nr_pages)
1000 {
1001         struct buffer_page *bpage, *tmp;
1002         LIST_HEAD(pages);
1003         unsigned i;
1004
1005         WARN_ON(!nr_pages);
1006
1007         for (i = 0; i < nr_pages; i++) {
1008                 struct page *page;
1009                 /*
1010                  * __GFP_NORETRY flag makes sure that the allocation fails
1011                  * gracefully without invoking oom-killer and the system is
1012                  * not destabilized.
1013                  */
1014                 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1015                                     GFP_KERNEL | __GFP_NORETRY,
1016                                     cpu_to_node(cpu_buffer->cpu));
1017                 if (!bpage)
1018                         goto free_pages;
1019
1020                 rb_check_bpage(cpu_buffer, bpage);
1021
1022                 list_add(&bpage->list, &pages);
1023
1024                 page = alloc_pages_node(cpu_to_node(cpu_buffer->cpu),
1025                                         GFP_KERNEL | __GFP_NORETRY, 0);
1026                 if (!page)
1027                         goto free_pages;
1028                 bpage->page = page_address(page);
1029                 rb_init_page(bpage->page);
1030         }
1031
1032         /*
1033          * The ring buffer page list is a circular list that does not
1034          * start and end with a list head. All page list items point to
1035          * other pages.
1036          */
1037         cpu_buffer->pages = pages.next;
1038         list_del(&pages);
1039
1040         rb_check_pages(cpu_buffer);
1041
1042         return 0;
1043
1044  free_pages:
1045         list_for_each_entry_safe(bpage, tmp, &pages, list) {
1046                 list_del_init(&bpage->list);
1047                 free_buffer_page(bpage);
1048         }
1049         return -ENOMEM;
1050 }
1051
1052 static struct ring_buffer_per_cpu *
1053 rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
1054 {
1055         struct ring_buffer_per_cpu *cpu_buffer;
1056         struct buffer_page *bpage;
1057         struct page *page;
1058         int ret;
1059
1060         cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
1061                                   GFP_KERNEL, cpu_to_node(cpu));
1062         if (!cpu_buffer)
1063                 return NULL;
1064
1065         cpu_buffer->cpu = cpu;
1066         cpu_buffer->buffer = buffer;
1067         raw_spin_lock_init(&cpu_buffer->reader_lock);
1068         lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
1069         cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1070
1071         bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1072                             GFP_KERNEL, cpu_to_node(cpu));
1073         if (!bpage)
1074                 goto fail_free_buffer;
1075
1076         rb_check_bpage(cpu_buffer, bpage);
1077
1078         cpu_buffer->reader_page = bpage;
1079         page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0);
1080         if (!page)
1081                 goto fail_free_reader;
1082         bpage->page = page_address(page);
1083         rb_init_page(bpage->page);
1084
1085         INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
1086
1087         ret = rb_allocate_pages(cpu_buffer, buffer->pages);
1088         if (ret < 0)
1089                 goto fail_free_reader;
1090
1091         cpu_buffer->head_page
1092                 = list_entry(cpu_buffer->pages, struct buffer_page, list);
1093         cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
1094
1095         rb_head_page_activate(cpu_buffer);
1096
1097         return cpu_buffer;
1098
1099  fail_free_reader:
1100         free_buffer_page(cpu_buffer->reader_page);
1101
1102  fail_free_buffer:
1103         kfree(cpu_buffer);
1104         return NULL;
1105 }
1106
1107 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
1108 {
1109         struct list_head *head = cpu_buffer->pages;
1110         struct buffer_page *bpage, *tmp;
1111
1112         free_buffer_page(cpu_buffer->reader_page);
1113
1114         rb_head_page_deactivate(cpu_buffer);
1115
1116         if (head) {
1117                 list_for_each_entry_safe(bpage, tmp, head, list) {
1118                         list_del_init(&bpage->list);
1119                         free_buffer_page(bpage);
1120                 }
1121                 bpage = list_entry(head, struct buffer_page, list);
1122                 free_buffer_page(bpage);
1123         }
1124
1125         kfree(cpu_buffer);
1126 }
1127
1128 #ifdef CONFIG_HOTPLUG_CPU
1129 static int rb_cpu_notify(struct notifier_block *self,
1130                          unsigned long action, void *hcpu);
1131 #endif
1132
1133 /**
1134  * ring_buffer_alloc - allocate a new ring_buffer
1135  * @size: the size in bytes per cpu that is needed.
1136  * @flags: attributes to set for the ring buffer.
1137  *
1138  * Currently the only flag that is available is the RB_FL_OVERWRITE
1139  * flag. This flag means that the buffer will overwrite old data
1140  * when the buffer wraps. If this flag is not set, the buffer will
1141  * drop data when the tail hits the head.
1142  */
1143 struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
1144                                         struct lock_class_key *key)
1145 {
1146         struct ring_buffer *buffer;
1147         int bsize;
1148         int cpu;
1149
1150         /* keep it in its own cache line */
1151         buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
1152                          GFP_KERNEL);
1153         if (!buffer)
1154                 return NULL;
1155
1156         if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
1157                 goto fail_free_buffer;
1158
1159         buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1160         buffer->flags = flags;
1161         buffer->clock = trace_clock_local;
1162         buffer->reader_lock_key = key;
1163
1164         /* need at least two pages */
1165         if (buffer->pages < 2)
1166                 buffer->pages = 2;
1167
1168         /*
1169          * In case of non-hotplug cpu, if the ring-buffer is allocated
1170          * in early initcall, it will not be notified of secondary cpus.
1171          * In that off case, we need to allocate for all possible cpus.
1172          */
1173 #ifdef CONFIG_HOTPLUG_CPU
1174         get_online_cpus();
1175         cpumask_copy(buffer->cpumask, cpu_online_mask);
1176 #else
1177         cpumask_copy(buffer->cpumask, cpu_possible_mask);
1178 #endif
1179         buffer->cpus = nr_cpu_ids;
1180
1181         bsize = sizeof(void *) * nr_cpu_ids;
1182         buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
1183                                   GFP_KERNEL);
1184         if (!buffer->buffers)
1185                 goto fail_free_cpumask;
1186
1187         for_each_buffer_cpu(buffer, cpu) {
1188                 buffer->buffers[cpu] =
1189                         rb_allocate_cpu_buffer(buffer, cpu);
1190                 if (!buffer->buffers[cpu])
1191                         goto fail_free_buffers;
1192         }
1193
1194 #ifdef CONFIG_HOTPLUG_CPU
1195         buffer->cpu_notify.notifier_call = rb_cpu_notify;
1196         buffer->cpu_notify.priority = 0;
1197         register_cpu_notifier(&buffer->cpu_notify);
1198 #endif
1199
1200         put_online_cpus();
1201         mutex_init(&buffer->mutex);
1202
1203         return buffer;
1204
1205  fail_free_buffers:
1206         for_each_buffer_cpu(buffer, cpu) {
1207                 if (buffer->buffers[cpu])
1208                         rb_free_cpu_buffer(buffer->buffers[cpu]);
1209         }
1210         kfree(buffer->buffers);
1211
1212  fail_free_cpumask:
1213         free_cpumask_var(buffer->cpumask);
1214         put_online_cpus();
1215
1216  fail_free_buffer:
1217         kfree(buffer);
1218         return NULL;
1219 }
1220 EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
1221
1222 /**
1223  * ring_buffer_free - free a ring buffer.
1224  * @buffer: the buffer to free.
1225  */
1226 void
1227 ring_buffer_free(struct ring_buffer *buffer)
1228 {
1229         int cpu;
1230
1231         get_online_cpus();
1232
1233 #ifdef CONFIG_HOTPLUG_CPU
1234         unregister_cpu_notifier(&buffer->cpu_notify);
1235 #endif
1236
1237         for_each_buffer_cpu(buffer, cpu)
1238                 rb_free_cpu_buffer(buffer->buffers[cpu]);
1239
1240         put_online_cpus();
1241
1242         kfree(buffer->buffers);
1243         free_cpumask_var(buffer->cpumask);
1244
1245         kfree(buffer);
1246 }
1247 EXPORT_SYMBOL_GPL(ring_buffer_free);
1248
1249 void ring_buffer_set_clock(struct ring_buffer *buffer,
1250                            u64 (*clock)(void))
1251 {
1252         buffer->clock = clock;
1253 }
1254
1255 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
1256
1257 static void
1258 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
1259 {
1260         struct buffer_page *bpage;
1261         struct list_head *p;
1262         unsigned i;
1263
1264         raw_spin_lock_irq(&cpu_buffer->reader_lock);
1265         rb_head_page_deactivate(cpu_buffer);
1266
1267         for (i = 0; i < nr_pages; i++) {
1268                 if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)))
1269                         goto out;
1270                 p = cpu_buffer->pages->next;
1271                 bpage = list_entry(p, struct buffer_page, list);
1272                 list_del_init(&bpage->list);
1273                 free_buffer_page(bpage);
1274         }
1275         if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)))
1276                 goto out;
1277
1278         rb_reset_cpu(cpu_buffer);
1279         rb_check_pages(cpu_buffer);
1280
1281 out:
1282         raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1283 }
1284
1285 static void
1286 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer,
1287                 struct list_head *pages, unsigned nr_pages)
1288 {
1289         struct buffer_page *bpage;
1290         struct list_head *p;
1291         unsigned i;
1292
1293         raw_spin_lock_irq(&cpu_buffer->reader_lock);
1294         rb_head_page_deactivate(cpu_buffer);
1295
1296         for (i = 0; i < nr_pages; i++) {
1297                 if (RB_WARN_ON(cpu_buffer, list_empty(pages)))
1298                         goto out;
1299                 p = pages->next;
1300                 bpage = list_entry(p, struct buffer_page, list);
1301                 list_del_init(&bpage->list);
1302                 list_add_tail(&bpage->list, cpu_buffer->pages);
1303         }
1304         rb_reset_cpu(cpu_buffer);
1305         rb_check_pages(cpu_buffer);
1306
1307 out:
1308         raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1309 }
1310
1311 /**
1312  * ring_buffer_resize - resize the ring buffer
1313  * @buffer: the buffer to resize.
1314  * @size: the new size.
1315  *
1316  * Minimum size is 2 * BUF_PAGE_SIZE.
1317  *
1318  * Returns -1 on failure.
1319  */
1320 int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
1321 {
1322         struct ring_buffer_per_cpu *cpu_buffer;
1323         unsigned nr_pages, rm_pages, new_pages;
1324         struct buffer_page *bpage, *tmp;
1325         unsigned long buffer_size;
1326         LIST_HEAD(pages);
1327         int i, cpu;
1328
1329         /*
1330          * Always succeed at resizing a non-existent buffer:
1331          */
1332         if (!buffer)
1333                 return size;
1334
1335         size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1336         size *= BUF_PAGE_SIZE;
1337         buffer_size = buffer->pages * BUF_PAGE_SIZE;
1338
1339         /* we need a minimum of two pages */
1340         if (size < BUF_PAGE_SIZE * 2)
1341                 size = BUF_PAGE_SIZE * 2;
1342
1343         if (size == buffer_size)
1344                 return size;
1345
1346         atomic_inc(&buffer->record_disabled);
1347
1348         /* Make sure all writers are done with this buffer. */
1349         synchronize_sched();
1350
1351         mutex_lock(&buffer->mutex);
1352         get_online_cpus();
1353
1354         nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1355
1356         if (size < buffer_size) {
1357
1358                 /* easy case, just free pages */
1359                 if (RB_WARN_ON(buffer, nr_pages >= buffer->pages))
1360                         goto out_fail;
1361
1362                 rm_pages = buffer->pages - nr_pages;
1363
1364                 for_each_buffer_cpu(buffer, cpu) {
1365                         cpu_buffer = buffer->buffers[cpu];
1366                         rb_remove_pages(cpu_buffer, rm_pages);
1367                 }
1368                 goto out;
1369         }
1370
1371         /*
1372          * This is a bit more difficult. We only want to add pages
1373          * when we can allocate enough for all CPUs. We do this
1374          * by allocating all the pages and storing them on a local
1375          * link list. If we succeed in our allocation, then we
1376          * add these pages to the cpu_buffers. Otherwise we just free
1377          * them all and return -ENOMEM;
1378          */
1379         if (RB_WARN_ON(buffer, nr_pages <= buffer->pages))
1380                 goto out_fail;
1381
1382         new_pages = nr_pages - buffer->pages;
1383
1384         for_each_buffer_cpu(buffer, cpu) {
1385                 for (i = 0; i < new_pages; i++) {
1386                         struct page *page;
1387                         /*
1388                          * __GFP_NORETRY flag makes sure that the allocation
1389                          * fails gracefully without invoking oom-killer and
1390                          * the system is not destabilized.
1391                          */
1392                         bpage = kzalloc_node(ALIGN(sizeof(*bpage),
1393                                                   cache_line_size()),
1394                                             GFP_KERNEL | __GFP_NORETRY,
1395                                             cpu_to_node(cpu));
1396                         if (!bpage)
1397                                 goto free_pages;
1398                         list_add(&bpage->list, &pages);
1399                         page = alloc_pages_node(cpu_to_node(cpu),
1400                                                 GFP_KERNEL | __GFP_NORETRY, 0);
1401                         if (!page)
1402                                 goto free_pages;
1403                         bpage->page = page_address(page);
1404                         rb_init_page(bpage->page);
1405                 }
1406         }
1407
1408         for_each_buffer_cpu(buffer, cpu) {
1409                 cpu_buffer = buffer->buffers[cpu];
1410                 rb_insert_pages(cpu_buffer, &pages, new_pages);
1411         }
1412
1413         if (RB_WARN_ON(buffer, !list_empty(&pages)))
1414                 goto out_fail;
1415
1416  out:
1417         buffer->pages = nr_pages;
1418         put_online_cpus();
1419         mutex_unlock(&buffer->mutex);
1420
1421         atomic_dec(&buffer->record_disabled);
1422
1423         return size;
1424
1425  free_pages:
1426         list_for_each_entry_safe(bpage, tmp, &pages, list) {
1427                 list_del_init(&bpage->list);
1428                 free_buffer_page(bpage);
1429         }
1430         put_online_cpus();
1431         mutex_unlock(&buffer->mutex);
1432         atomic_dec(&buffer->record_disabled);
1433         return -ENOMEM;
1434
1435         /*
1436          * Something went totally wrong, and we are too paranoid
1437          * to even clean up the mess.
1438          */
1439  out_fail:
1440         put_online_cpus();
1441         mutex_unlock(&buffer->mutex);
1442         atomic_dec(&buffer->record_disabled);
1443         return -1;
1444 }
1445 EXPORT_SYMBOL_GPL(ring_buffer_resize);
1446
1447 void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val)
1448 {
1449         mutex_lock(&buffer->mutex);
1450         if (val)
1451                 buffer->flags |= RB_FL_OVERWRITE;
1452         else
1453                 buffer->flags &= ~RB_FL_OVERWRITE;
1454         mutex_unlock(&buffer->mutex);
1455 }
1456 EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
1457
1458 static inline void *
1459 __rb_data_page_index(struct buffer_data_page *bpage, unsigned index)
1460 {
1461         return bpage->data + index;
1462 }
1463
1464 static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
1465 {
1466         return bpage->page->data + index;
1467 }
1468
1469 static inline struct ring_buffer_event *
1470 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
1471 {
1472         return __rb_page_index(cpu_buffer->reader_page,
1473                                cpu_buffer->reader_page->read);
1474 }
1475
1476 static inline struct ring_buffer_event *
1477 rb_iter_head_event(struct ring_buffer_iter *iter)
1478 {
1479         return __rb_page_index(iter->head_page, iter->head);
1480 }
1481
1482 static inline unsigned long rb_page_write(struct buffer_page *bpage)
1483 {
1484         return local_read(&bpage->write) & RB_WRITE_MASK;
1485 }
1486
1487 static inline unsigned rb_page_commit(struct buffer_page *bpage)
1488 {
1489         return local_read(&bpage->page->commit);
1490 }
1491
1492 static inline unsigned long rb_page_entries(struct buffer_page *bpage)
1493 {
1494         return local_read(&bpage->entries) & RB_WRITE_MASK;
1495 }
1496
1497 /* Size is determined by what has been committed */
1498 static inline unsigned rb_page_size(struct buffer_page *bpage)
1499 {
1500         return rb_page_commit(bpage);
1501 }
1502
1503 static inline unsigned
1504 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
1505 {
1506         return rb_page_commit(cpu_buffer->commit_page);
1507 }
1508
1509 static inline unsigned
1510 rb_event_index(struct ring_buffer_event *event)
1511 {
1512         unsigned long addr = (unsigned long)event;
1513
1514         return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
1515 }
1516
1517 static inline int
1518 rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
1519                    struct ring_buffer_event *event)
1520 {
1521         unsigned long addr = (unsigned long)event;
1522         unsigned long index;
1523
1524         index = rb_event_index(event);
1525         addr &= PAGE_MASK;
1526
1527         return cpu_buffer->commit_page->page == (void *)addr &&
1528                 rb_commit_index(cpu_buffer) == index;
1529 }
1530
1531 static void
1532 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
1533 {
1534         unsigned long max_count;
1535
1536         /*
1537          * We only race with interrupts and NMIs on this CPU.
1538          * If we own the commit event, then we can commit
1539          * all others that interrupted us, since the interruptions
1540          * are in stack format (they finish before they come
1541          * back to us). This allows us to do a simple loop to
1542          * assign the commit to the tail.
1543          */
1544  again:
1545         max_count = cpu_buffer->buffer->pages * 100;
1546
1547         while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
1548                 if (RB_WARN_ON(cpu_buffer, !(--max_count)))
1549                         return;
1550                 if (RB_WARN_ON(cpu_buffer,
1551                                rb_is_reader_page(cpu_buffer->tail_page)))
1552                         return;
1553                 local_set(&cpu_buffer->commit_page->page->commit,
1554                           rb_page_write(cpu_buffer->commit_page));
1555                 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
1556                 cpu_buffer->write_stamp =
1557                         cpu_buffer->commit_page->page->time_stamp;
1558                 /* add barrier to keep gcc from optimizing too much */
1559                 barrier();
1560         }
1561         while (rb_commit_index(cpu_buffer) !=
1562                rb_page_write(cpu_buffer->commit_page)) {
1563
1564                 local_set(&cpu_buffer->commit_page->page->commit,
1565                           rb_page_write(cpu_buffer->commit_page));
1566                 RB_WARN_ON(cpu_buffer,
1567                            local_read(&cpu_buffer->commit_page->page->commit) &
1568                            ~RB_WRITE_MASK);
1569                 barrier();
1570         }
1571
1572         /* again, keep gcc from optimizing */
1573         barrier();
1574
1575         /*
1576          * If an interrupt came in just after the first while loop
1577          * and pushed the tail page forward, we will be left with
1578          * a dangling commit that will never go forward.
1579          */
1580         if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page))
1581                 goto again;
1582 }
1583
1584 static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
1585 {
1586         cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp;
1587         cpu_buffer->reader_page->read = 0;
1588 }
1589
1590 static void rb_inc_iter(struct ring_buffer_iter *iter)
1591 {
1592         struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1593
1594         /*
1595          * The iterator could be on the reader page (it starts there).
1596          * But the head could have moved, since the reader was
1597          * found. Check for this case and assign the iterator
1598          * to the head page instead of next.
1599          */
1600         if (iter->head_page == cpu_buffer->reader_page)
1601                 iter->head_page = rb_set_head_page(cpu_buffer);
1602         else
1603                 rb_inc_page(cpu_buffer, &iter->head_page);
1604
1605         iter->read_stamp = iter->head_page->page->time_stamp;
1606         iter->head = 0;
1607 }
1608
1609 /* Slow path, do not inline */
1610 static noinline struct ring_buffer_event *
1611 rb_add_time_stamp(struct ring_buffer_event *event, u64 delta)
1612 {
1613         event->type_len = RINGBUF_TYPE_TIME_EXTEND;
1614
1615         /* Not the first event on the page? */
1616         if (rb_event_index(event)) {
1617                 event->time_delta = delta & TS_MASK;
1618                 event->array[0] = delta >> TS_SHIFT;
1619         } else {
1620                 /* nope, just zero it */
1621                 event->time_delta = 0;
1622                 event->array[0] = 0;
1623         }
1624
1625         return skip_time_extend(event);
1626 }
1627
1628 /**
1629  * ring_buffer_update_event - update event type and data
1630  * @event: the even to update
1631  * @type: the type of event
1632  * @length: the size of the event field in the ring buffer
1633  *
1634  * Update the type and data fields of the event. The length
1635  * is the actual size that is written to the ring buffer,
1636  * and with this, we can determine what to place into the
1637  * data field.
1638  */
1639 static void
1640 rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
1641                 struct ring_buffer_event *event, unsigned length,
1642                 int add_timestamp, u64 delta)
1643 {
1644         /* Only a commit updates the timestamp */
1645         if (unlikely(!rb_event_is_commit(cpu_buffer, event)))
1646                 delta = 0;
1647
1648         /*
1649          * If we need to add a timestamp, then we
1650          * add it to the start of the resevered space.
1651          */
1652         if (unlikely(add_timestamp)) {
1653                 event = rb_add_time_stamp(event, delta);
1654                 length -= RB_LEN_TIME_EXTEND;
1655                 delta = 0;
1656         }
1657
1658         event->time_delta = delta;
1659         length -= RB_EVNT_HDR_SIZE;
1660         if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) {
1661                 event->type_len = 0;
1662                 event->array[0] = length;
1663         } else
1664                 event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
1665 }
1666
1667 /*
1668  * rb_handle_head_page - writer hit the head page
1669  *
1670  * Returns: +1 to retry page
1671  *           0 to continue
1672  *          -1 on error
1673  */
1674 static int
1675 rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
1676                     struct buffer_page *tail_page,
1677                     struct buffer_page *next_page)
1678 {
1679         struct buffer_page *new_head;
1680         int entries;
1681         int type;
1682         int ret;
1683
1684         entries = rb_page_entries(next_page);
1685
1686         /*
1687          * The hard part is here. We need to move the head
1688          * forward, and protect against both readers on
1689          * other CPUs and writers coming in via interrupts.
1690          */
1691         type = rb_head_page_set_update(cpu_buffer, next_page, tail_page,
1692                                        RB_PAGE_HEAD);
1693
1694         /*
1695          * type can be one of four:
1696          *  NORMAL - an interrupt already moved it for us
1697          *  HEAD   - we are the first to get here.
1698          *  UPDATE - we are the interrupt interrupting
1699          *           a current move.
1700          *  MOVED  - a reader on another CPU moved the next
1701          *           pointer to its reader page. Give up
1702          *           and try again.
1703          */
1704
1705         switch (type) {
1706         case RB_PAGE_HEAD:
1707                 /*
1708                  * We changed the head to UPDATE, thus
1709                  * it is our responsibility to update
1710                  * the counters.
1711                  */
1712                 local_add(entries, &cpu_buffer->overrun);
1713                 local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
1714
1715                 /*
1716                  * The entries will be zeroed out when we move the
1717                  * tail page.
1718                  */
1719
1720                 /* still more to do */
1721                 break;
1722
1723         case RB_PAGE_UPDATE:
1724                 /*
1725                  * This is an interrupt that interrupt the
1726                  * previous update. Still more to do.
1727                  */
1728                 break;
1729         case RB_PAGE_NORMAL:
1730                 /*
1731                  * An interrupt came in before the update
1732                  * and processed this for us.
1733                  * Nothing left to do.
1734                  */
1735                 return 1;
1736         case RB_PAGE_MOVED:
1737                 /*
1738                  * The reader is on another CPU and just did
1739                  * a swap with our next_page.
1740                  * Try again.
1741                  */
1742                 return 1;
1743         default:
1744                 RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
1745                 return -1;
1746         }
1747
1748         /*
1749          * Now that we are here, the old head pointer is
1750          * set to UPDATE. This will keep the reader from
1751          * swapping the head page with the reader page.
1752          * The reader (on another CPU) will spin till
1753          * we are finished.
1754          *
1755          * We just need to protect against interrupts
1756          * doing the job. We will set the next pointer
1757          * to HEAD. After that, we set the old pointer
1758          * to NORMAL, but only if it was HEAD before.
1759          * otherwise we are an interrupt, and only
1760          * want the outer most commit to reset it.
1761          */
1762         new_head = next_page;
1763         rb_inc_page(cpu_buffer, &new_head);
1764
1765         ret = rb_head_page_set_head(cpu_buffer, new_head, next_page,
1766                                     RB_PAGE_NORMAL);
1767
1768         /*
1769          * Valid returns are:
1770          *  HEAD   - an interrupt came in and already set it.
1771          *  NORMAL - One of two things:
1772          *            1) We really set it.
1773          *            2) A bunch of interrupts came in and moved
1774          *               the page forward again.
1775          */
1776         switch (ret) {
1777         case RB_PAGE_HEAD:
1778         case RB_PAGE_NORMAL:
1779                 /* OK */
1780                 break;
1781         default:
1782                 RB_WARN_ON(cpu_buffer, 1);
1783                 return -1;
1784         }
1785
1786         /*
1787          * It is possible that an interrupt came in,
1788          * set the head up, then more interrupts came in
1789          * and moved it again. When we get back here,
1790          * the page would have been set to NORMAL but we
1791          * just set it back to HEAD.
1792          *
1793          * How do you detect this? Well, if that happened
1794          * the tail page would have moved.
1795          */
1796         if (ret == RB_PAGE_NORMAL) {
1797                 /*
1798                  * If the tail had moved passed next, then we need
1799                  * to reset the pointer.
1800                  */
1801                 if (cpu_buffer->tail_page != tail_page &&
1802                     cpu_buffer->tail_page != next_page)
1803                         rb_head_page_set_normal(cpu_buffer, new_head,
1804                                                 next_page,
1805                                                 RB_PAGE_HEAD);
1806         }
1807
1808         /*
1809          * If this was the outer most commit (the one that
1810          * changed the original pointer from HEAD to UPDATE),
1811          * then it is up to us to reset it to NORMAL.
1812          */
1813         if (type == RB_PAGE_HEAD) {
1814                 ret = rb_head_page_set_normal(cpu_buffer, next_page,
1815                                               tail_page,
1816                                               RB_PAGE_UPDATE);
1817                 if (RB_WARN_ON(cpu_buffer,
1818                                ret != RB_PAGE_UPDATE))
1819                         return -1;
1820         }
1821
1822         return 0;
1823 }
1824
1825 static unsigned rb_calculate_event_length(unsigned length)
1826 {
1827         struct ring_buffer_event event; /* Used only for sizeof array */
1828
1829         /* zero length can cause confusions */
1830         if (!length)
1831                 length = 1;
1832
1833         if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT)
1834                 length += sizeof(event.array[0]);
1835
1836         length += RB_EVNT_HDR_SIZE;
1837         length = ALIGN(length, RB_ARCH_ALIGNMENT);
1838
1839         return length;
1840 }
1841
1842 static inline void
1843 rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
1844               struct buffer_page *tail_page,
1845               unsigned long tail, unsigned long length)
1846 {
1847         struct ring_buffer_event *event;
1848
1849         /*
1850          * Only the event that crossed the page boundary
1851          * must fill the old tail_page with padding.
1852          */
1853         if (tail >= BUF_PAGE_SIZE) {
1854                 /*
1855                  * If the page was filled, then we still need
1856                  * to update the real_end. Reset it to zero
1857                  * and the reader will ignore it.
1858                  */
1859                 if (tail == BUF_PAGE_SIZE)
1860                         tail_page->real_end = 0;
1861
1862                 local_sub(length, &tail_page->write);
1863                 return;
1864         }
1865
1866         event = __rb_page_index(tail_page, tail);
1867         kmemcheck_annotate_bitfield(event, bitfield);
1868
1869         /* account for padding bytes */
1870         local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes);
1871
1872         /*
1873          * Save the original length to the meta data.
1874          * This will be used by the reader to add lost event
1875          * counter.
1876          */
1877         tail_page->real_end = tail;
1878
1879         /*
1880          * If this event is bigger than the minimum size, then
1881          * we need to be careful that we don't subtract the
1882          * write counter enough to allow another writer to slip
1883          * in on this page.
1884          * We put in a discarded commit instead, to make sure
1885          * that this space is not used again.
1886          *
1887          * If we are less than the minimum size, we don't need to
1888          * worry about it.
1889          */
1890         if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
1891                 /* No room for any events */
1892
1893                 /* Mark the rest of the page with padding */
1894                 rb_event_set_padding(event);
1895
1896                 /* Set the write back to the previous setting */
1897                 local_sub(length, &tail_page->write);
1898                 return;
1899         }
1900
1901         /* Put in a discarded event */
1902         event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
1903         event->type_len = RINGBUF_TYPE_PADDING;
1904         /* time delta must be non zero */
1905         event->time_delta = 1;
1906
1907         /* Set write to end of buffer */
1908         length = (tail + length) - BUF_PAGE_SIZE;
1909         local_sub(length, &tail_page->write);
1910 }
1911
1912 /*
1913  * This is the slow path, force gcc not to inline it.
1914  */
1915 static noinline struct ring_buffer_event *
1916 rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
1917              unsigned long length, unsigned long tail,
1918              struct buffer_page *tail_page, u64 ts)
1919 {
1920         struct buffer_page *commit_page = cpu_buffer->commit_page;
1921         struct ring_buffer *buffer = cpu_buffer->buffer;
1922         struct buffer_page *next_page;
1923         int ret;
1924
1925         next_page = tail_page;
1926
1927         rb_inc_page(cpu_buffer, &next_page);
1928
1929         /*
1930          * If for some reason, we had an interrupt storm that made
1931          * it all the way around the buffer, bail, and warn
1932          * about it.
1933          */
1934         if (unlikely(next_page == commit_page)) {
1935                 local_inc(&cpu_buffer->commit_overrun);
1936                 goto out_reset;
1937         }
1938
1939         /*
1940          * This is where the fun begins!
1941          *
1942          * We are fighting against races between a reader that
1943          * could be on another CPU trying to swap its reader
1944          * page with the buffer head.
1945          *
1946          * We are also fighting against interrupts coming in and
1947          * moving the head or tail on us as well.
1948          *
1949          * If the next page is the head page then we have filled
1950          * the buffer, unless the commit page is still on the
1951          * reader page.
1952          */
1953         if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) {
1954
1955                 /*
1956                  * If the commit is not on the reader page, then
1957                  * move the header page.
1958                  */
1959                 if (!rb_is_reader_page(cpu_buffer->commit_page)) {
1960                         /*
1961                          * If we are not in overwrite mode,
1962                          * this is easy, just stop here.
1963                          */
1964                         if (!(buffer->flags & RB_FL_OVERWRITE))
1965                                 goto out_reset;
1966
1967                         ret = rb_handle_head_page(cpu_buffer,
1968                                                   tail_page,
1969                                                   next_page);
1970                         if (ret < 0)
1971                                 goto out_reset;
1972                         if (ret)
1973                                 goto out_again;
1974                 } else {
1975                         /*
1976                          * We need to be careful here too. The
1977                          * commit page could still be on the reader
1978                          * page. We could have a small buffer, and
1979                          * have filled up the buffer with events
1980                          * from interrupts and such, and wrapped.
1981                          *
1982                          * Note, if the tail page is also the on the
1983                          * reader_page, we let it move out.
1984                          */
1985                         if (unlikely((cpu_buffer->commit_page !=
1986                                       cpu_buffer->tail_page) &&
1987                                      (cpu_buffer->commit_page ==
1988                                       cpu_buffer->reader_page))) {
1989                                 local_inc(&cpu_buffer->commit_overrun);
1990                                 goto out_reset;
1991                         }
1992                 }
1993         }
1994
1995         ret = rb_tail_page_update(cpu_buffer, tail_page, next_page);
1996         if (ret) {
1997                 /*
1998                  * Nested commits always have zero deltas, so
1999                  * just reread the time stamp
2000                  */
2001                 ts = rb_time_stamp(buffer);
2002                 next_page->page->time_stamp = ts;
2003         }
2004
2005  out_again:
2006
2007         rb_reset_tail(cpu_buffer, tail_page, tail, length);
2008
2009         /* fail and let the caller try again */
2010         return ERR_PTR(-EAGAIN);
2011
2012  out_reset:
2013         /* reset write */
2014         rb_reset_tail(cpu_buffer, tail_page, tail, length);
2015
2016         return NULL;
2017 }
2018
2019 static struct ring_buffer_event *
2020 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
2021                   unsigned long length, u64 ts,
2022                   u64 delta, int add_timestamp)
2023 {
2024         struct buffer_page *tail_page;
2025         struct ring_buffer_event *event;
2026         unsigned long tail, write;
2027
2028         /*
2029          * If the time delta since the last event is too big to
2030          * hold in the time field of the event, then we append a
2031          * TIME EXTEND event ahead of the data event.
2032          */
2033         if (unlikely(add_timestamp))
2034                 length += RB_LEN_TIME_EXTEND;
2035
2036         tail_page = cpu_buffer->tail_page;
2037         write = local_add_return(length, &tail_page->write);
2038
2039         /* set write to only the index of the write */
2040         write &= RB_WRITE_MASK;
2041         tail = write - length;
2042
2043         /* See if we shot pass the end of this buffer page */
2044         if (unlikely(write > BUF_PAGE_SIZE))
2045                 return rb_move_tail(cpu_buffer, length, tail,
2046                                     tail_page, ts);
2047
2048         /* We reserved something on the buffer */
2049
2050         event = __rb_page_index(tail_page, tail);
2051         kmemcheck_annotate_bitfield(event, bitfield);
2052         rb_update_event(cpu_buffer, event, length, add_timestamp, delta);
2053
2054         local_inc(&tail_page->entries);
2055
2056         /*
2057          * If this is the first commit on the page, then update
2058          * its timestamp.
2059          */
2060         if (!tail)
2061                 tail_page->page->time_stamp = ts;
2062
2063         /* account for these added bytes */
2064         local_add(length, &cpu_buffer->entries_bytes);
2065
2066         return event;
2067 }
2068
2069 static inline int
2070 rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
2071                   struct ring_buffer_event *event)
2072 {
2073         unsigned long new_index, old_index;
2074         struct buffer_page *bpage;
2075         unsigned long index;
2076         unsigned long addr;
2077
2078         new_index = rb_event_index(event);
2079         old_index = new_index + rb_event_ts_length(event);
2080         addr = (unsigned long)event;
2081         addr &= PAGE_MASK;
2082
2083         bpage = cpu_buffer->tail_page;
2084
2085         if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
2086                 unsigned long write_mask =
2087                         local_read(&bpage->write) & ~RB_WRITE_MASK;
2088                 unsigned long event_length = rb_event_length(event);
2089                 /*
2090                  * This is on the tail page. It is possible that
2091                  * a write could come in and move the tail page
2092                  * and write to the next page. That is fine
2093                  * because we just shorten what is on this page.
2094                  */
2095                 old_index += write_mask;
2096                 new_index += write_mask;
2097                 index = local_cmpxchg(&bpage->write, old_index, new_index);
2098                 if (index == old_index) {
2099                         /* update counters */
2100                         local_sub(event_length, &cpu_buffer->entries_bytes);
2101                         return 1;
2102                 }
2103         }
2104
2105         /* could not discard */
2106         return 0;
2107 }
2108
2109 static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
2110 {
2111         local_inc(&cpu_buffer->committing);
2112         local_inc(&cpu_buffer->commits);
2113 }
2114
2115 static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
2116 {
2117         unsigned long commits;
2118
2119         if (RB_WARN_ON(cpu_buffer,
2120                        !local_read(&cpu_buffer->committing)))
2121                 return;
2122
2123  again:
2124         commits = local_read(&cpu_buffer->commits);
2125         /* synchronize with interrupts */
2126         barrier();
2127         if (local_read(&cpu_buffer->committing) == 1)
2128                 rb_set_commit_to_write(cpu_buffer);
2129
2130         local_dec(&cpu_buffer->committing);
2131
2132         /* synchronize with interrupts */
2133         barrier();
2134
2135         /*
2136          * Need to account for interrupts coming in between the
2137          * updating of the commit page and the clearing of the
2138          * committing counter.
2139          */
2140         if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
2141             !local_read(&cpu_buffer->committing)) {
2142                 local_inc(&cpu_buffer->committing);
2143                 goto again;
2144         }
2145 }
2146
2147 static struct ring_buffer_event *
2148 rb_reserve_next_event(struct ring_buffer *buffer,
2149                       struct ring_buffer_per_cpu *cpu_buffer,
2150                       unsigned long length)
2151 {
2152         struct ring_buffer_event *event;
2153         u64 ts, delta;
2154         int nr_loops = 0;
2155         int add_timestamp;
2156         u64 diff;
2157
2158         rb_start_commit(cpu_buffer);
2159
2160 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2161         /*
2162          * Due to the ability to swap a cpu buffer from a buffer
2163          * it is possible it was swapped before we committed.
2164          * (committing stops a swap). We check for it here and
2165          * if it happened, we have to fail the write.
2166          */
2167         barrier();
2168         if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) {
2169                 local_dec(&cpu_buffer->committing);
2170                 local_dec(&cpu_buffer->commits);
2171                 return NULL;
2172         }
2173 #endif
2174
2175         length = rb_calculate_event_length(length);
2176  again:
2177         add_timestamp = 0;
2178         delta = 0;
2179
2180         /*
2181          * We allow for interrupts to reenter here and do a trace.
2182          * If one does, it will cause this original code to loop
2183          * back here. Even with heavy interrupts happening, this
2184          * should only happen a few times in a row. If this happens
2185          * 1000 times in a row, there must be either an interrupt
2186          * storm or we have something buggy.
2187          * Bail!
2188          */
2189         if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
2190                 goto out_fail;
2191
2192         ts = rb_time_stamp(cpu_buffer->buffer);
2193         diff = ts - cpu_buffer->write_stamp;
2194
2195         /* make sure this diff is calculated here */
2196         barrier();
2197
2198         /* Did the write stamp get updated already? */
2199         if (likely(ts >= cpu_buffer->write_stamp)) {
2200                 delta = diff;
2201                 if (unlikely(test_time_stamp(delta))) {
2202                         int local_clock_stable = 1;
2203 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
2204                         local_clock_stable = sched_clock_stable;
2205 #endif
2206                         WARN_ONCE(delta > (1ULL << 59),
2207                                   KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s",
2208                                   (unsigned long long)delta,
2209                                   (unsigned long long)ts,
2210                                   (unsigned long long)cpu_buffer->write_stamp,
2211                                   local_clock_stable ? "" :
2212                                   "If you just came from a suspend/resume,\n"
2213                                   "please switch to the trace global clock:\n"
2214                                   "  echo global > /sys/kernel/debug/tracing/trace_clock\n");
2215                         add_timestamp = 1;
2216                 }
2217         }
2218
2219         event = __rb_reserve_next(cpu_buffer, length, ts,
2220                                   delta, add_timestamp);
2221         if (unlikely(PTR_ERR(event) == -EAGAIN))
2222                 goto again;
2223
2224         if (!event)
2225                 goto out_fail;
2226
2227         return event;
2228
2229  out_fail:
2230         rb_end_commit(cpu_buffer);
2231         return NULL;
2232 }
2233
2234 #ifdef CONFIG_TRACING
2235
2236 #define TRACE_RECURSIVE_DEPTH 16
2237
2238 /* Keep this code out of the fast path cache */
2239 static noinline void trace_recursive_fail(void)
2240 {
2241         /* Disable all tracing before we do anything else */
2242         tracing_off_permanent();
2243
2244         printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:"
2245                     "HC[%lu]:SC[%lu]:NMI[%lu]\n",
2246                     trace_recursion_buffer(),
2247                     hardirq_count() >> HARDIRQ_SHIFT,
2248                     softirq_count() >> SOFTIRQ_SHIFT,
2249                     in_nmi());
2250
2251         WARN_ON_ONCE(1);
2252 }
2253
2254 static inline int trace_recursive_lock(void)
2255 {
2256         trace_recursion_inc();
2257
2258         if (likely(trace_recursion_buffer() < TRACE_RECURSIVE_DEPTH))
2259                 return 0;
2260
2261         trace_recursive_fail();
2262
2263         return -1;
2264 }
2265
2266 static inline void trace_recursive_unlock(void)
2267 {
2268         WARN_ON_ONCE(!trace_recursion_buffer());
2269
2270         trace_recursion_dec();
2271 }
2272
2273 #else
2274
2275 #define trace_recursive_lock()          (0)
2276 #define trace_recursive_unlock()        do { } while (0)
2277
2278 #endif
2279
2280 /**
2281  * ring_buffer_lock_reserve - reserve a part of the buffer
2282  * @buffer: the ring buffer to reserve from
2283  * @length: the length of the data to reserve (excluding event header)
2284  *
2285  * Returns a reseverd event on the ring buffer to copy directly to.
2286  * The user of this interface will need to get the body to write into
2287  * and can use the ring_buffer_event_data() interface.
2288  *
2289  * The length is the length of the data needed, not the event length
2290  * which also includes the event header.
2291  *
2292  * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
2293  * If NULL is returned, then nothing has been allocated or locked.
2294  */
2295 struct ring_buffer_event *
2296 ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length)
2297 {
2298         struct ring_buffer_per_cpu *cpu_buffer;
2299         struct ring_buffer_event *event;
2300         int cpu;
2301
2302         if (ring_buffer_flags != RB_BUFFERS_ON)
2303                 return NULL;
2304
2305         /* If we are tracing schedule, we don't want to recurse */
2306         preempt_disable_notrace();
2307
2308         if (atomic_read(&buffer->record_disabled))
2309                 goto out_nocheck;
2310
2311         if (trace_recursive_lock())
2312                 goto out_nocheck;
2313
2314         cpu = raw_smp_processor_id();
2315
2316         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2317                 goto out;
2318
2319         cpu_buffer = buffer->buffers[cpu];
2320
2321         if (atomic_read(&cpu_buffer->record_disabled))
2322                 goto out;
2323
2324         if (length > BUF_MAX_DATA_SIZE)
2325                 goto out;
2326
2327         event = rb_reserve_next_event(buffer, cpu_buffer, length);
2328         if (!event)
2329                 goto out;
2330
2331         return event;
2332
2333  out:
2334         trace_recursive_unlock();
2335
2336  out_nocheck:
2337         preempt_enable_notrace();
2338         return NULL;
2339 }
2340 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
2341
2342 static void
2343 rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer,
2344                       struct ring_buffer_event *event)
2345 {
2346         u64 delta;
2347
2348         /*
2349          * The event first in the commit queue updates the
2350          * time stamp.
2351          */
2352         if (rb_event_is_commit(cpu_buffer, event)) {
2353                 /*
2354                  * A commit event that is first on a page
2355                  * updates the write timestamp with the page stamp
2356                  */
2357                 if (!rb_event_index(event))
2358                         cpu_buffer->write_stamp =
2359                                 cpu_buffer->commit_page->page->time_stamp;
2360                 else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) {
2361                         delta = event->array[0];
2362                         delta <<= TS_SHIFT;
2363                         delta += event->time_delta;
2364                         cpu_buffer->write_stamp += delta;
2365                 } else
2366                         cpu_buffer->write_stamp += event->time_delta;
2367         }
2368 }
2369
2370 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
2371                       struct ring_buffer_event *event)
2372 {
2373         local_inc(&cpu_buffer->entries);
2374         rb_update_write_stamp(cpu_buffer, event);
2375         rb_end_commit(cpu_buffer);
2376 }
2377
2378 /**
2379  * ring_buffer_unlock_commit - commit a reserved
2380  * @buffer: The buffer to commit to
2381  * @event: The event pointer to commit.
2382  *
2383  * This commits the data to the ring buffer, and releases any locks held.
2384  *
2385  * Must be paired with ring_buffer_lock_reserve.
2386  */
2387 int ring_buffer_unlock_commit(struct ring_buffer *buffer,
2388                               struct ring_buffer_event *event)
2389 {
2390         struct ring_buffer_per_cpu *cpu_buffer;
2391         int cpu = raw_smp_processor_id();
2392
2393         cpu_buffer = buffer->buffers[cpu];
2394
2395         rb_commit(cpu_buffer, event);
2396
2397         trace_recursive_unlock();
2398
2399         preempt_enable_notrace();
2400
2401         return 0;
2402 }
2403 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
2404
2405 static inline void rb_event_discard(struct ring_buffer_event *event)
2406 {
2407         if (event->type_len == RINGBUF_TYPE_TIME_EXTEND)
2408                 event = skip_time_extend(event);
2409
2410         /* array[0] holds the actual length for the discarded event */
2411         event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
2412         event->type_len = RINGBUF_TYPE_PADDING;
2413         /* time delta must be non zero */
2414         if (!event->time_delta)
2415                 event->time_delta = 1;
2416 }
2417
2418 /*
2419  * Decrement the entries to the page that an event is on.
2420  * The event does not even need to exist, only the pointer
2421  * to the page it is on. This may only be called before the commit
2422  * takes place.
2423  */
2424 static inline void
2425 rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
2426                    struct ring_buffer_event *event)
2427 {
2428         unsigned long addr = (unsigned long)event;
2429         struct buffer_page *bpage = cpu_buffer->commit_page;
2430         struct buffer_page *start;
2431
2432         addr &= PAGE_MASK;
2433
2434         /* Do the likely case first */
2435         if (likely(bpage->page == (void *)addr)) {
2436                 local_dec(&bpage->entries);
2437                 return;
2438         }
2439
2440         /*
2441          * Because the commit page may be on the reader page we
2442          * start with the next page and check the end loop there.
2443          */
2444         rb_inc_page(cpu_buffer, &bpage);
2445         start = bpage;
2446         do {
2447                 if (bpage->page == (void *)addr) {
2448                         local_dec(&bpage->entries);
2449                         return;
2450                 }
2451                 rb_inc_page(cpu_buffer, &bpage);
2452         } while (bpage != start);
2453
2454         /* commit not part of this buffer?? */
2455         RB_WARN_ON(cpu_buffer, 1);
2456 }
2457
2458 /**
2459  * ring_buffer_commit_discard - discard an event that has not been committed
2460  * @buffer: the ring buffer
2461  * @event: non committed event to discard
2462  *
2463  * Sometimes an event that is in the ring buffer needs to be ignored.
2464  * This function lets the user discard an event in the ring buffer
2465  * and then that event will not be read later.
2466  *
2467  * This function only works if it is called before the the item has been
2468  * committed. It will try to free the event from the ring buffer
2469  * if another event has not been added behind it.
2470  *
2471  * If another event has been added behind it, it will set the event
2472  * up as discarded, and perform the commit.
2473  *
2474  * If this function is called, do not call ring_buffer_unlock_commit on
2475  * the event.
2476  */
2477 void ring_buffer_discard_commit(struct ring_buffer *buffer,
2478                                 struct ring_buffer_event *event)
2479 {
2480         struct ring_buffer_per_cpu *cpu_buffer;
2481         int cpu;
2482
2483         /* The event is discarded regardless */
2484         rb_event_discard(event);
2485
2486         cpu = smp_processor_id();
2487         cpu_buffer = buffer->buffers[cpu];
2488
2489         /*
2490          * This must only be called if the event has not been
2491          * committed yet. Thus we can assume that preemption
2492          * is still disabled.
2493          */
2494         RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
2495
2496         rb_decrement_entry(cpu_buffer, event);
2497         if (rb_try_to_discard(cpu_buffer, event))
2498                 goto out;
2499
2500         /*
2501          * The commit is still visible by the reader, so we
2502          * must still update the timestamp.
2503          */
2504         rb_update_write_stamp(cpu_buffer, event);
2505  out:
2506         rb_end_commit(cpu_buffer);
2507
2508         trace_recursive_unlock();
2509
2510         preempt_enable_notrace();
2511
2512 }
2513 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
2514
2515 /**
2516  * ring_buffer_write - write data to the buffer without reserving
2517  * @buffer: The ring buffer to write to.
2518  * @length: The length of the data being written (excluding the event header)
2519  * @data: The data to write to the buffer.
2520  *
2521  * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
2522  * one function. If you already have the data to write to the buffer, it
2523  * may be easier to simply call this function.
2524  *
2525  * Note, like ring_buffer_lock_reserve, the length is the length of the data
2526  * and not the length of the event which would hold the header.
2527  */
2528 int ring_buffer_write(struct ring_buffer *buffer,
2529                         unsigned long length,
2530                         void *data)
2531 {
2532         struct ring_buffer_per_cpu *cpu_buffer;
2533         struct ring_buffer_event *event;
2534         void *body;
2535         int ret = -EBUSY;
2536         int cpu;
2537
2538         if (ring_buffer_flags != RB_BUFFERS_ON)
2539                 return -EBUSY;
2540
2541         preempt_disable_notrace();
2542
2543         if (atomic_read(&buffer->record_disabled))
2544                 goto out;
2545
2546         cpu = raw_smp_processor_id();
2547
2548         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2549                 goto out;
2550
2551         cpu_buffer = buffer->buffers[cpu];
2552
2553         if (atomic_read(&cpu_buffer->record_disabled))
2554                 goto out;
2555
2556         if (length > BUF_MAX_DATA_SIZE)
2557                 goto out;
2558
2559         event = rb_reserve_next_event(buffer, cpu_buffer, length);
2560         if (!event)
2561                 goto out;
2562
2563         body = rb_event_data(event);
2564
2565         memcpy(body, data, length);
2566
2567         rb_commit(cpu_buffer, event);
2568
2569         ret = 0;
2570  out:
2571         preempt_enable_notrace();
2572
2573         return ret;
2574 }
2575 EXPORT_SYMBOL_GPL(ring_buffer_write);
2576
2577 static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
2578 {
2579         struct buffer_page *reader = cpu_buffer->reader_page;
2580         struct buffer_page *head = rb_set_head_page(cpu_buffer);
2581         struct buffer_page *commit = cpu_buffer->commit_page;
2582
2583         /* In case of error, head will be NULL */
2584         if (unlikely(!head))
2585                 return 1;
2586
2587         return reader->read == rb_page_commit(reader) &&
2588                 (commit == reader ||
2589                  (commit == head &&
2590                   head->read == rb_page_commit(commit)));
2591 }
2592
2593 /**
2594  * ring_buffer_record_disable - stop all writes into the buffer
2595  * @buffer: The ring buffer to stop writes to.
2596  *
2597  * This prevents all writes to the buffer. Any attempt to write
2598  * to the buffer after this will fail and return NULL.
2599  *
2600  * The caller should call synchronize_sched() after this.
2601  */
2602 void ring_buffer_record_disable(struct ring_buffer *buffer)
2603 {
2604         atomic_inc(&buffer->record_disabled);
2605 }
2606 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
2607
2608 /**
2609  * ring_buffer_record_enable - enable writes to the buffer
2610  * @buffer: The ring buffer to enable writes
2611  *
2612  * Note, multiple disables will need the same number of enables
2613  * to truly enable the writing (much like preempt_disable).
2614  */
2615 void ring_buffer_record_enable(struct ring_buffer *buffer)
2616 {
2617         atomic_dec(&buffer->record_disabled);
2618 }
2619 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
2620
2621 /**
2622  * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
2623  * @buffer: The ring buffer to stop writes to.
2624  * @cpu: The CPU buffer to stop
2625  *
2626  * This prevents all writes to the buffer. Any attempt to write
2627  * to the buffer after this will fail and return NULL.
2628  *
2629  * The caller should call synchronize_sched() after this.
2630  */
2631 void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
2632 {
2633         struct ring_buffer_per_cpu *cpu_buffer;
2634
2635         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2636                 return;
2637
2638         cpu_buffer = buffer->buffers[cpu];
2639         atomic_inc(&cpu_buffer->record_disabled);
2640 }
2641 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
2642
2643 /**
2644  * ring_buffer_record_enable_cpu - enable writes to the buffer
2645  * @buffer: The ring buffer to enable writes
2646  * @cpu: The CPU to enable.
2647  *
2648  * Note, multiple disables will need the same number of enables
2649  * to truly enable the writing (much like preempt_disable).
2650  */
2651 void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
2652 {
2653         struct ring_buffer_per_cpu *cpu_buffer;
2654
2655         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2656                 return;
2657
2658         cpu_buffer = buffer->buffers[cpu];
2659         atomic_dec(&cpu_buffer->record_disabled);
2660 }
2661 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
2662
2663 /*
2664  * The total entries in the ring buffer is the running counter
2665  * of entries entered into the ring buffer, minus the sum of
2666  * the entries read from the ring buffer and the number of
2667  * entries that were overwritten.
2668  */
2669 static inline unsigned long
2670 rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
2671 {
2672         return local_read(&cpu_buffer->entries) -
2673                 (local_read(&cpu_buffer->overrun) + cpu_buffer->read);
2674 }
2675
2676 /**
2677  * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
2678  * @buffer: The ring buffer
2679  * @cpu: The per CPU buffer to read from.
2680  */
2681 unsigned long ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu)
2682 {
2683         unsigned long flags;
2684         struct ring_buffer_per_cpu *cpu_buffer;
2685         struct buffer_page *bpage;
2686         unsigned long ret;
2687
2688         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2689                 return 0;
2690
2691         cpu_buffer = buffer->buffers[cpu];
2692         raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2693         /*
2694          * if the tail is on reader_page, oldest time stamp is on the reader
2695          * page
2696          */
2697         if (cpu_buffer->tail_page == cpu_buffer->reader_page)
2698                 bpage = cpu_buffer->reader_page;
2699         else
2700                 bpage = rb_set_head_page(cpu_buffer);
2701         ret = bpage->page->time_stamp;
2702         raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2703
2704         return ret;
2705 }
2706 EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
2707
2708 /**
2709  * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
2710  * @buffer: The ring buffer
2711  * @cpu: The per CPU buffer to read from.
2712  */
2713 unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu)
2714 {
2715         struct ring_buffer_per_cpu *cpu_buffer;
2716         unsigned long ret;
2717
2718         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2719                 return 0;
2720
2721         cpu_buffer = buffer->buffers[cpu];
2722         ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
2723
2724         return ret;
2725 }
2726 EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
2727
2728 /**
2729  * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
2730  * @buffer: The ring buffer
2731  * @cpu: The per CPU buffer to get the entries from.
2732  */
2733 unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
2734 {
2735         struct ring_buffer_per_cpu *cpu_buffer;
2736
2737         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2738                 return 0;
2739
2740         cpu_buffer = buffer->buffers[cpu];
2741
2742         return rb_num_of_entries(cpu_buffer);
2743 }
2744 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
2745
2746 /**
2747  * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
2748  * @buffer: The ring buffer
2749  * @cpu: The per CPU buffer to get the number of overruns from
2750  */
2751 unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
2752 {
2753         struct ring_buffer_per_cpu *cpu_buffer;
2754         unsigned long ret;
2755
2756         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2757                 return 0;
2758
2759         cpu_buffer = buffer->buffers[cpu];
2760         ret = local_read(&cpu_buffer->overrun);
2761
2762         return ret;
2763 }
2764 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
2765
2766 /**
2767  * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits
2768  * @buffer: The ring buffer
2769  * @cpu: The per CPU buffer to get the number of overruns from
2770  */
2771 unsigned long
2772 ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu)
2773 {
2774         struct ring_buffer_per_cpu *cpu_buffer;
2775         unsigned long ret;
2776
2777         if (!cpumask_test_cpu(cpu, buffer->cpumask))
2778                 return 0;
2779
2780         cpu_buffer = buffer->buffers[cpu];
2781         ret = local_read(&cpu_buffer->commit_overrun);
2782
2783         return ret;
2784 }
2785 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
2786
2787 /**
2788  * ring_buffer_entries - get the number of entries in a buffer
2789  * @buffer: The ring buffer
2790  *
2791  * Returns the total number of entries in the ring buffer
2792  * (all CPU entries)
2793  */
2794 unsigned long ring_buffer_entries(struct ring_buffer *buffer)
2795 {
2796         struct ring_buffer_per_cpu *cpu_buffer;
2797         unsigned long entries = 0;
2798         int cpu;
2799
2800         /* if you care about this being correct, lock the buffer */
2801         for_each_buffer_cpu(buffer, cpu) {
2802                 cpu_buffer = buffer->buffers[cpu];
2803                 entries += rb_num_of_entries(cpu_buffer);
2804         }
2805
2806         return entries;
2807 }
2808 EXPORT_SYMBOL_GPL(ring_buffer_entries);
2809
2810 /**
2811  * ring_buffer_overruns - get the number of overruns in buffer
2812  * @buffer: The ring buffer
2813  *
2814  * Returns the total number of overruns in the ring buffer
2815  * (all CPU entries)
2816  */
2817 unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
2818 {
2819         struct ring_buffer_per_cpu *cpu_buffer;
2820         unsigned long overruns = 0;
2821         int cpu;
2822
2823         /* if you care about this being correct, lock the buffer */
2824         for_each_buffer_cpu(buffer, cpu) {
2825                 cpu_buffer = buffer->buffers[cpu];
2826                 overruns += local_read(&cpu_buffer->overrun);
2827         }
2828
2829         return overruns;
2830 }
2831 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
2832
2833 static void rb_iter_reset(struct ring_buffer_iter *iter)
2834 {
2835         struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2836
2837         /* Iterator usage is expected to have record disabled */
2838         if (list_empty(&cpu_buffer->reader_page->list)) {
2839                 iter->head_page = rb_set_head_page(cpu_buffer);
2840                 if (unlikely(!iter->head_page))
2841                         return;
2842                 iter->head = iter->head_page->read;
2843         } else {
2844                 iter->head_page = cpu_buffer->reader_page;
2845                 iter->head = cpu_buffer->reader_page->read;
2846         }
2847         if (iter->head)
2848                 iter->read_stamp = cpu_buffer->read_stamp;
2849         else
2850                 iter->read_stamp = iter->head_page->page->time_stamp;
2851         iter->cache_reader_page = cpu_buffer->reader_page;
2852         iter->cache_read = cpu_buffer->read;
2853 }
2854
2855 /**
2856  * ring_buffer_iter_reset - reset an iterator
2857  * @iter: The iterator to reset
2858  *
2859  * Resets the iterator, so that it will start from the beginning
2860  * again.
2861  */
2862 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
2863 {
2864         struct ring_buffer_per_cpu *cpu_buffer;
2865         unsigned long flags;
2866
2867         if (!iter)
2868                 return;
2869
2870         cpu_buffer = iter->cpu_buffer;
2871
2872         raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2873         rb_iter_reset(iter);
2874         raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2875 }
2876 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
2877
2878 /**
2879  * ring_buffer_iter_empty - check if an iterator has no more to read
2880  * @iter: The iterator to check
2881  */
2882 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
2883 {
2884         struct ring_buffer_per_cpu *cpu_buffer;
2885
2886         cpu_buffer = iter->cpu_buffer;
2887
2888         return iter->head_page == cpu_buffer->commit_page &&
2889                 iter->head == rb_commit_index(cpu_buffer);
2890 }
2891 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
2892
2893 static void
2894 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
2895                      struct ring_buffer_event *event)
2896 {
2897         u64 delta;
2898
2899         switch (event->type_len) {
2900         case RINGBUF_TYPE_PADDING:
2901                 return;
2902
2903         case RINGBUF_TYPE_TIME_EXTEND:
2904                 delta = event->array[0];
2905                 delta <<= TS_SHIFT;
2906                 delta += event->time_delta;
2907                 cpu_buffer->read_stamp += delta;
2908                 return;
2909
2910         case RINGBUF_TYPE_TIME_STAMP:
2911                 /* FIXME: not implemented */
2912                 return;
2913
2914         case RINGBUF_TYPE_DATA:
2915                 cpu_buffer->read_stamp += event->time_delta;
2916                 return;
2917
2918         default:
2919                 BUG();
2920         }
2921         return;
2922 }
2923
2924 static void
2925 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
2926                           struct ring_buffer_event *event)
2927 {
2928         u64 delta;
2929
2930         switch (event->type_len) {
2931         case RINGBUF_TYPE_PADDING:
2932                 return;
2933
2934         case RINGBUF_TYPE_TIME_EXTEND:
2935                 delta = event->array[0];
2936                 delta <<= TS_SHIFT;
2937                 delta += event->time_delta;
2938                 iter->read_stamp += delta;
2939                 return;
2940
2941         case RINGBUF_TYPE_TIME_STAMP:
2942                 /* FIXME: not implemented */
2943                 return;
2944
2945         case RINGBUF_TYPE_DATA:
2946                 iter->read_stamp += event->time_delta;
2947                 return;
2948
2949         default:
2950                 BUG();
2951         }
2952         return;
2953 }
2954
2955 static struct buffer_page *
2956 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
2957 {
2958         struct buffer_page *reader = NULL;
2959         unsigned long overwrite;
2960         unsigned long flags;
2961         int nr_loops = 0;
2962         int ret;
2963
2964         local_irq_save(flags);
2965         arch_spin_lock(&cpu_buffer->lock);
2966
2967  again:
2968         /*
2969          * This should normally only loop twice. But because the
2970          * start of the reader inserts an empty page, it causes
2971          * a case where we will loop three times. There should be no
2972          * reason to loop four times (that I know of).
2973          */
2974         if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
2975                 reader = NULL;
2976                 goto out;
2977         }
2978
2979         reader = cpu_buffer->reader_page;
2980
2981         /* If there's more to read, return this page */
2982         if (cpu_buffer->reader_page->read < rb_page_size(reader))
2983                 goto out;
2984
2985         /* Never should we have an index greater than the size */
2986         if (RB_WARN_ON(cpu_buffer,
2987                        cpu_buffer->reader_page->read > rb_page_size(reader)))
2988                 goto out;
2989
2990         /* check if we caught up to the tail */
2991         reader = NULL;
2992         if (cpu_buffer->commit_page == cpu_buffer->reader_page)
2993                 goto out;
2994
2995         /*
2996          * Reset the reader page to size zero.
2997          */
2998         local_set(&cpu_buffer->reader_page->write, 0);
2999         local_set(&cpu_buffer->reader_page->entries, 0);
3000         local_set(&cpu_buffer->reader_page->page->commit, 0);
3001         cpu_buffer->reader_page->real_end = 0;
3002
3003  spin:
3004         /*
3005          * Splice the empty reader page into the list around the head.
3006          */
3007         reader = rb_set_head_page(cpu_buffer);
3008         cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
3009         cpu_buffer->reader_page->list.prev = reader->list.prev;
3010
3011         /*
3012          * cpu_buffer->pages just needs to point to the buffer, it
3013          *  has no specific buffer page to point to. Lets move it out
3014          *  of our way so we don't accidentally swap it.
3015          */
3016         cpu_buffer->pages = reader->list.prev;
3017
3018         /* The reader page will be pointing to the new head */
3019         rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list);
3020
3021         /*
3022          * We want to make sure we read the overruns after we set up our
3023          * pointers to the next object. The writer side does a
3024          * cmpxchg to cross pages which acts as the mb on the writer
3025          * side. Note, the reader will constantly fail the swap
3026          * while the writer is updating the pointers, so this
3027          * guarantees that the overwrite recorded here is the one we
3028          * want to compare with the last_overrun.
3029          */
3030         smp_mb();
3031         overwrite = local_read(&(cpu_buffer->overrun));
3032
3033         /*
3034          * Here's the tricky part.
3035          *
3036          * We need to move the pointer past the header page.
3037          * But we can only do that if a writer is not currently
3038          * moving it. The page before the header page has the
3039          * flag bit '1' set if it is pointing to the page we want.
3040          * but if the writer is in the process of moving it
3041          * than it will be '2' or already moved '0'.
3042          */
3043
3044         ret = rb_head_page_replace(reader, cpu_buffer->reader_page);
3045
3046         /*
3047          * If we did not convert it, then we must try again.
3048          */
3049         if (!ret)
3050                 goto spin;
3051
3052         /*
3053          * Yeah! We succeeded in replacing the page.
3054          *
3055          * Now make the new head point back to the reader page.
3056          */
3057         rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
3058         rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
3059
3060         /* Finally update the reader page to the new head */
3061         cpu_buffer->reader_page = reader;
3062         rb_reset_reader_page(cpu_buffer);
3063
3064         if (overwrite != cpu_buffer->last_overrun) {
3065                 cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
3066                 cpu_buffer->last_overrun = overwrite;
3067         }
3068
3069         goto again;
3070
3071  out:
3072         arch_spin_unlock(&cpu_buffer->lock);
3073         local_irq_restore(flags);
3074
3075         return reader;
3076 }
3077
3078 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
3079 {
3080         struct ring_buffer_event *event;
3081         struct buffer_page *reader;
3082         unsigned length;
3083
3084         reader = rb_get_reader_page(cpu_buffer);
3085
3086         /* This function should not be called when buffer is empty */
3087         if (RB_WARN_ON(cpu_buffer, !reader))
3088                 return;
3089
3090         event = rb_reader_event(cpu_buffer);
3091
3092         if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
3093                 cpu_buffer->read++;
3094
3095         rb_update_read_stamp(cpu_buffer, event);
3096
3097         length = rb_event_length(event);
3098         cpu_buffer->reader_page->read += length;
3099 }
3100
3101 static void rb_advance_iter(struct ring_buffer_iter *iter)
3102 {
3103         struct ring_buffer_per_cpu *cpu_buffer;
3104         struct ring_buffer_event *event;
3105         unsigned length;
3106
3107         cpu_buffer = iter->cpu_buffer;
3108
3109         /*
3110          * Check if we are at the end of the buffer.
3111          */
3112         if (iter->head >= rb_page_size(iter->head_page)) {
3113                 /* discarded commits can make the page empty */
3114                 if (iter->head_page == cpu_buffer->commit_page)
3115                         return;
3116                 rb_inc_iter(iter);
3117                 return;
3118         }
3119
3120         event = rb_iter_head_event(iter);
3121
3122         length = rb_event_length(event);
3123
3124         /*
3125          * This should not be called to advance the header if we are
3126          * at the tail of the buffer.
3127          */
3128         if (RB_WARN_ON(cpu_buffer,
3129                        (iter->head_page == cpu_buffer->commit_page) &&
3130                        (iter->head + length > rb_commit_index(cpu_buffer))))
3131                 return;
3132
3133         rb_update_iter_read_stamp(iter, event);
3134
3135         iter->head += length;
3136
3137         /* check for end of page padding */
3138         if ((iter->head >= rb_page_size(iter->head_page)) &&
3139             (iter->head_page != cpu_buffer->commit_page))
3140                 rb_advance_iter(iter);
3141 }
3142
3143 static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
3144 {
3145         return cpu_buffer->lost_events;
3146 }
3147
3148 static struct ring_buffer_event *
3149 rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
3150                unsigned long *lost_events)
3151 {
3152         struct ring_buffer_event *event;
3153         struct buffer_page *reader;
3154         int nr_loops = 0;
3155
3156  again:
3157         /*
3158          * We repeat when a time extend is encountered.
3159          * Since the time extend is always attached to a data event,
3160          * we should never loop more than once.
3161          * (We never hit the following condition more than twice).
3162          */
3163         if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
3164                 return NULL;
3165
3166         reader = rb_get_reader_page(cpu_buffer);
3167         if (!reader)
3168                 return NULL;
3169
3170         event = rb_reader_event(cpu_buffer);
3171
3172         switch (event->type_len) {
3173         case RINGBUF_TYPE_PADDING:
3174                 if (rb_null_event(event))
3175                         RB_WARN_ON(cpu_buffer, 1);
3176                 /*
3177                  * Because the writer could be discarding every
3178                  * event it creates (which would probably be bad)
3179                  * if we were to go back to "again" then we may never
3180                  * catch up, and will trigger the warn on, or lock
3181                  * the box. Return the padding, and we will release
3182                  * the current locks, and try again.
3183                  */
3184                 return event;
3185
3186         case RINGBUF_TYPE_TIME_EXTEND:
3187                 /* Internal data, OK to advance */
3188                 rb_advance_reader(cpu_buffer);
3189                 goto again;
3190
3191         case RINGBUF_TYPE_TIME_STAMP:
3192                 /* FIXME: not implemented */
3193                 rb_advance_reader(cpu_buffer);
3194                 goto again;
3195
3196         case RINGBUF_TYPE_DATA:
3197                 if (ts) {
3198                         *ts = cpu_buffer->read_stamp + event->time_delta;
3199                         ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
3200                                                          cpu_buffer->cpu, ts);
3201                 }
3202                 if (lost_events)
3203                         *lost_events = rb_lost_events(cpu_buffer);
3204                 return event;
3205
3206         default:
3207                 BUG();
3208         }
3209
3210         return NULL;
3211 }
3212 EXPORT_SYMBOL_GPL(ring_buffer_peek);
3213
3214 static struct ring_buffer_event *
3215 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
3216 {
3217         struct ring_buffer *buffer;
3218         struct ring_buffer_per_cpu *cpu_buffer;
3219         struct ring_buffer_event *event;
3220         int nr_loops = 0;
3221
3222         cpu_buffer = iter->cpu_buffer;
3223         buffer = cpu_buffer->buffer;
3224
3225         /*
3226          * Check if someone performed a consuming read to
3227          * the buffer. A consuming read invalidates the iterator
3228          * and we need to reset the iterator in this case.
3229          */
3230         if (unlikely(iter->cache_read != cpu_buffer->read ||
3231                      iter->cache_reader_page != cpu_buffer->reader_page))
3232                 rb_iter_reset(iter);
3233
3234  again:
3235         if (ring_buffer_iter_empty(iter))
3236                 return NULL;
3237
3238         /*
3239          * We repeat when a time extend is encountered.
3240          * Since the time extend is always attached to a data event,
3241          * we should never loop more than once.
3242          * (We never hit the following condition more than twice).
3243          */
3244         if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
3245                 return NULL;
3246
3247         if (rb_per_cpu_empty(cpu_buffer))
3248                 return NULL;
3249
3250         if (iter->head >= local_read(&iter->head_page->page->commit)) {
3251                 rb_inc_iter(iter);
3252                 goto again;
3253         }
3254
3255         event = rb_iter_head_event(iter);
3256
3257         switch (event->type_len) {
3258         case RINGBUF_TYPE_PADDING:
3259                 if (rb_null_event(event)) {
3260                         rb_inc_iter(iter);
3261                         goto again;
3262                 }
3263                 rb_advance_iter(iter);
3264                 return event;
3265
3266         case RINGBUF_TYPE_TIME_EXTEND:
3267                 /* Internal data, OK to advance */
3268                 rb_advance_iter(iter);
3269                 goto again;
3270
3271         case RINGBUF_TYPE_TIME_STAMP:
3272                 /* FIXME: not implemented */
3273                 rb_advance_iter(iter);
3274                 goto again;
3275
3276         case RINGBUF_TYPE_DATA:
3277                 if (ts) {
3278                         *ts = iter->read_stamp + event->time_delta;
3279                         ring_buffer_normalize_time_stamp(buffer,
3280                                                          cpu_buffer->cpu, ts);
3281                 }
3282                 return event;
3283
3284         default:
3285                 BUG();
3286         }
3287
3288         return NULL;
3289 }
3290 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
3291
3292 static inline int rb_ok_to_lock(void)
3293 {
3294         /*
3295          * If an NMI die dumps out the content of the ring buffer
3296          * do not grab locks. We also permanently disable the ring
3297          * buffer too. A one time deal is all you get from reading
3298          * the ring buffer from an NMI.
3299          */
3300         if (likely(!in_nmi()))
3301                 return 1;
3302
3303         tracing_off_permanent();
3304         return 0;
3305 }
3306
3307 /**
3308  * ring_buffer_peek - peek at the next event to be read
3309  * @buffer: The ring buffer to read
3310  * @cpu: The cpu to peak at
3311  * @ts: The timestamp counter of this event.
3312  * @lost_events: a variable to store if events were lost (may be NULL)
3313  *
3314  * This will return the event that will be read next, but does
3315  * not consume the data.
3316  */
3317 struct ring_buffer_event *
3318 ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts,
3319                  unsigned long *lost_events)
3320 {
3321         struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
3322         struct ring_buffer_event *event;
3323         unsigned long flags;
3324         int dolock;
3325
3326         if (!cpumask_test_cpu(cpu, buffer->cpumask))
3327                 return NULL;
3328
3329         dolock = rb_ok_to_lock();
3330  again:
3331         local_irq_save(flags);
3332         if (dolock)
3333                 raw_spin_lock(&cpu_buffer->reader_lock);
3334         event = rb_buffer_peek(cpu_buffer, ts, lost_events);
3335         if (event && event->type_len == RINGBUF_TYPE_PADDING)
3336                 rb_advance_reader(cpu_buffer);
3337         if (dolock)
3338                 raw_spin_unlock(&cpu_buffer->reader_lock);
3339         local_irq_restore(flags);
3340
3341         if (event && event->type_len == RINGBUF_TYPE_PADDING)
3342                 goto again;
3343
3344         return event;
3345 }
3346
3347 /**
3348  * ring_buffer_iter_peek - peek at the next event to be read
3349  * @iter: The ring buffer iterator
3350  * @ts: The timestamp counter of this event.
3351  *
3352  * This will return the event that will be read next, but does
3353  * not increment the iterator.
3354  */
3355 struct ring_buffer_event *
3356 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
3357 {
3358         struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3359         struct ring_buffer_event *event;
3360         unsigned long flags;
3361
3362  again:
3363         raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3364         event = rb_iter_peek(iter, ts);
3365         raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3366
3367         if (event && event->type_len == RINGBUF_TYPE_PADDING)
3368                 goto again;
3369
3370         return event;
3371 }
3372
3373 /**
3374  * ring_buffer_consume - return an event and consume it
3375  * @buffer: The ring buffer to get the next event from
3376  * @cpu: the cpu to read the buffer from
3377  * @ts: a variable to store the timestamp (may be NULL)
3378  * @lost_events: a variable to store if events were lost (may be NULL)
3379  *
3380  * Returns the next event in the ring buffer, and that event is consumed.
3381  * Meaning, that sequential reads will keep returning a different event,
3382  * and eventually empty the ring buffer if the producer is slower.
3383  */
3384 struct ring_buffer_event *
3385 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts,
3386                     unsigned long *lost_events)
3387 {
3388         struct ring_buffer_per_cpu *cpu_buffer;
3389         struct ring_buffer_event *event = NULL;
3390         unsigned long flags;
3391         int dolock;
3392
3393         dolock = rb_ok_to_lock();
3394
3395  again:
3396         /* might be called in atomic */
3397         preempt_disable();
3398
3399         if (!cpumask_test_cpu(cpu, buffer->cpumask))
3400                 goto out;
3401
3402         cpu_buffer = buffer->buffers[cpu];
3403         local_irq_save(flags);
3404         if (dolock)
3405                 raw_spin_lock(&cpu_buffer->reader_lock);
3406
3407         event = rb_buffer_peek(cpu_buffer, ts, lost_events);
3408         if (event) {
3409                 cpu_buffer->lost_events = 0;
3410                 rb_advance_reader(cpu_buffer);
3411         }
3412
3413         if (dolock)
3414                 raw_spin_unlock(&cpu_buffer->reader_lock);
3415         local_irq_restore(flags);
3416
3417  out:
3418         preempt_enable();
3419
3420         if (event && event->type_len == RINGBUF_TYPE_PADDING)
3421                 goto again;
3422
3423         return event;
3424 }
3425 EXPORT_SYMBOL_GPL(ring_buffer_consume);
3426
3427 /**
3428  * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
3429  * @buffer: The ring buffer to read from
3430  * @cpu: The cpu buffer to iterate over
3431  *
3432  * This performs the initial preparations necessary to iterate
3433  * through the buffer.  Memory is allocated, buffer recording
3434  * is disabled, and the iterator pointer is returned to the caller.
3435  *
3436  * Disabling buffer recordng prevents the reading from being
3437  * corrupted. This is not a consuming read, so a producer is not
3438  * expected.
3439  *
3440  * After a sequence of ring_buffer_read_prepare calls, the user is
3441  * expected to make at least one call to ring_buffer_prepare_sync.
3442  * Afterwards, ring_buffer_read_start is invoked to get things going
3443  * for real.
3444  *
3445  * This overall must be paired with ring_buffer_finish.
3446  */
3447 struct ring_buffer_iter *
3448 ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu)
3449 {
3450         struct ring_buffer_per_cpu *cpu_buffer;
3451         struct ring_buffer_iter *iter;
3452
3453         if (!cpumask_test_cpu(cpu, buffer->cpumask))
3454                 return NULL;
3455
3456         iter = kmalloc(sizeof(*iter), GFP_KERNEL);
3457         if (!iter)
3458                 return NULL;
3459
3460         cpu_buffer = buffer->buffers[cpu];
3461
3462         iter->cpu_buffer = cpu_buffer;
3463
3464         atomic_inc(&cpu_buffer->record_disabled);
3465
3466         return iter;
3467 }
3468 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare);
3469
3470 /**
3471  * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
3472  *
3473  * All previously invoked ring_buffer_read_prepare calls to prepare
3474  * iterators will be synchronized.  Afterwards, read_buffer_read_start
3475  * calls on those iterators are allowed.
3476  */
3477 void
3478 ring_buffer_read_prepare_sync(void)
3479 {
3480         synchronize_sched();
3481 }
3482 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync);
3483
3484 /**
3485  * ring_buffer_read_start - start a non consuming read of the buffer
3486  * @iter: The iterator returned by ring_buffer_read_prepare
3487  *
3488  * This finalizes the startup of an iteration through the buffer.
3489  * The iterator comes from a call to ring_buffer_read_prepare and
3490  * an intervening ring_buffer_read_prepare_sync must have been
3491  * performed.
3492  *
3493  * Must be paired with ring_buffer_finish.
3494  */
3495 void
3496 ring_buffer_read_start(struct ring_buffer_iter *iter)
3497 {
3498         struct ring_buffer_per_cpu *cpu_buffer;
3499         unsigned long flags;
3500
3501         if (!iter)
3502                 return;
3503
3504         cpu_buffer = iter->cpu_buffer;
3505
3506         raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3507         arch_spin_lock(&cpu_buffer->lock);
3508         rb_iter_reset(iter);
3509         arch_spin_unlock(&cpu_buffer->lock);
3510         raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3511 }
3512 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
3513
3514 /**
3515  * ring_buffer_finish - finish reading the iterator of the buffer
3516  * @iter: The iterator retrieved by ring_buffer_start
3517  *
3518  * This re-enables the recording to the buffer, and frees the
3519  * iterator.
3520  */
3521 void
3522 ring_buffer_read_finish(struct ring_buffer_iter *iter)
3523 {
3524         struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3525
3526         atomic_dec(&cpu_buffer->record_disabled);
3527         kfree(iter);
3528 }
3529 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
3530
3531 /**
3532  * ring_buffer_read - read the next item in the ring buffer by the iterator
3533  * @iter: The ring buffer iterator
3534  * @ts: The time stamp of the event read.
3535  *
3536  * This reads the next event in the ring buffer and increments the iterator.
3537  */
3538 struct ring_buffer_event *
3539 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
3540 {
3541         struct ring_buffer_event *event;
3542         struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3543         unsigned long flags;
3544
3545         raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3546  again:
3547         event = rb_iter_peek(iter, ts);
3548         if (!event)
3549                 goto out;
3550
3551         if (event->type_len == RINGBUF_TYPE_PADDING)
3552                 goto again;
3553
3554         rb_advance_iter(iter);
3555  out:
3556         raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3557
3558         return event;
3559 }
3560 EXPORT_SYMBOL_GPL(ring_buffer_read);
3561
3562 /**
3563  * ring_buffer_size - return the size of the ring buffer (in bytes)
3564  * @buffer: The ring buffer.
3565  */
3566 unsigned long ring_buffer_size(struct ring_buffer *buffer)
3567 {
3568         return BUF_PAGE_SIZE * buffer->pages;
3569 }
3570 EXPORT_SYMBOL_GPL(ring_buffer_size);
3571
3572 static void
3573 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
3574 {
3575         rb_head_page_deactivate(cpu_buffer);
3576
3577         cpu_buffer->head_page
3578                 = list_entry(cpu_buffer->pages, struct buffer_page, list);
3579         local_set(&cpu_buffer->head_page->write, 0);
3580         local_set(&cpu_buffer->head_page->entries, 0);
3581         local_set(&cpu_buffer->head_page->page->commit, 0);
3582
3583         cpu_buffer->head_page->read = 0;
3584
3585         cpu_buffer->tail_page = cpu_buffer->head_page;
3586         cpu_buffer->commit_page = cpu_buffer->head_page;
3587
3588         INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
3589         local_set(&cpu_buffer->reader_page->write, 0);
3590         local_set(&cpu_buffer->reader_page->entries, 0);
3591         local_set(&cpu_buffer->reader_page->page->commit, 0);
3592         cpu_buffer->reader_page->read = 0;
3593
3594         local_set(&cpu_buffer->commit_overrun, 0);
3595         local_set(&cpu_buffer->entries_bytes, 0);
3596         local_set(&cpu_buffer->overrun, 0);
3597         local_set(&cpu_buffer->entries, 0);
3598         local_set(&cpu_buffer->committing, 0);
3599         local_set(&cpu_buffer->commits, 0);
3600         cpu_buffer->read = 0;
3601         cpu_buffer->read_bytes = 0;
3602
3603         cpu_buffer->write_stamp = 0;
3604         cpu_buffer->read_stamp = 0;
3605
3606         cpu_buffer->lost_events = 0;
3607         cpu_buffer->last_overrun = 0;
3608
3609         rb_head_page_activate(cpu_buffer);
3610 }
3611
3612 /**
3613  * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
3614  * @buffer: The ring buffer to reset a per cpu buffer of
3615  * @cpu: The CPU buffer to be reset
3616  */
3617 void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
3618 {
3619         struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
3620         unsigned long flags;
3621
3622         if (!cpumask_test_cpu(cpu, buffer->cpumask))
3623                 return;
3624
3625         atomic_inc(&cpu_buffer->record_disabled);
3626
3627         raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3628
3629         if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
3630                 goto out;
3631
3632         arch_spin_lock(&cpu_buffer->lock);
3633
3634         rb_reset_cpu(cpu_buffer);
3635
3636         arch_spin_unlock(&cpu_buffer->lock);
3637
3638  out:
3639         raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3640
3641         atomic_dec(&cpu_buffer->record_disabled);
3642 }
3643 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
3644
3645 /**
3646  * ring_buffer_reset - reset a ring buffer
3647  * @buffer: The ring buffer to reset all cpu buffers
3648  */
3649 void ring_buffer_reset(struct ring_buffer *buffer)
3650 {
3651         int cpu;
3652
3653         for_each_buffer_cpu(buffer, cpu)
3654                 ring_buffer_reset_cpu(buffer, cpu);
3655 }
3656 EXPORT_SYMBOL_GPL(ring_buffer_reset);
3657
3658 /**
3659  * rind_buffer_empty - is the ring buffer empty?
3660  * @buffer: The ring buffer to test
3661  */
3662 int ring_buffer_empty(struct ring_buffer *buffer)
3663 {
3664         struct ring_buffer_per_cpu *cpu_buffer;
3665         unsigned long flags;
3666         int dolock;
3667         int cpu;
3668         int ret;
3669
3670         dolock = rb_ok_to_lock();
3671
3672         /* yes this is racy, but if you don't like the race, lock the buffer */
3673         for_each_buffer_cpu(buffer, cpu) {
3674                 cpu_buffer = buffer->buffers[cpu];
3675                 local_irq_save(flags);
3676                 if (dolock)
3677                         raw_spin_lock(&cpu_buffer->reader_lock);
3678                 ret = rb_per_cpu_empty(cpu_buffer);
3679                 if (dolock)
3680                         raw_spin_unlock(&cpu_buffer->reader_lock);
3681                 local_irq_restore(flags);
3682
3683                 if (!ret)
3684                         return 0;
3685         }
3686
3687         return 1;
3688 }
3689 EXPORT_SYMBOL_GPL(ring_buffer_empty);
3690
3691 /**
3692  * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
3693  * @buffer: The ring buffer
3694  * @cpu: The CPU buffer to test
3695  */
3696 int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
3697 {
3698         struct ring_buffer_per_cpu *cpu_buffer;
3699         unsigned long flags;
3700         int dolock;
3701         int ret;
3702
3703         if (!cpumask_test_cpu(cpu, buffer->cpumask))
3704                 return 1;
3705
3706         dolock = rb_ok_to_lock();
3707
3708         cpu_buffer = buffer->buffers[cpu];
3709         local_irq_save(flags);
3710         if (dolock)
3711                 raw_spin_lock(&cpu_buffer->reader_lock);
3712         ret = rb_per_cpu_empty(cpu_buffer);
3713         if (dolock)
3714                 raw_spin_unlock(&cpu_buffer->reader_lock);
3715         local_irq_restore(flags);
3716
3717         return ret;
3718 }
3719 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
3720
3721 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
3722 /**
3723  * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
3724  * @buffer_a: One buffer to swap with
3725  * @buffer_b: The other buffer to swap with
3726  *
3727  * This function is useful for tracers that want to take a "snapshot"
3728  * of a CPU buffer and has another back up buffer lying around.
3729  * it is expected that the tracer handles the cpu buffer not being
3730  * used at the moment.
3731  */
3732 int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
3733                          struct ring_buffer *buffer_b, int cpu)
3734 {
3735         struct ring_buffer_per_cpu *cpu_buffer_a;
3736         struct ring_buffer_per_cpu *cpu_buffer_b;
3737         int ret = -EINVAL;
3738
3739         if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
3740             !cpumask_test_cpu(cpu, buffer_b->cpumask))
3741                 goto out;
3742
3743         /* At least make sure the two buffers are somewhat the same */
3744         if (buffer_a->pages != buffer_b->pages)
3745                 goto out;
3746
3747         ret = -EAGAIN;
3748
3749         if (ring_buffer_flags != RB_BUFFERS_ON)
3750                 goto out;
3751
3752         if (atomic_read(&buffer_a->record_disabled))
3753                 goto out;
3754
3755         if (atomic_read(&buffer_b->record_disabled))
3756                 goto out;
3757
3758         cpu_buffer_a = buffer_a->buffers[cpu];
3759         cpu_buffer_b = buffer_b->buffers[cpu];
3760
3761         if (atomic_read(&cpu_buffer_a->record_disabled))
3762                 goto out;
3763
3764         if (atomic_read(&cpu_buffer_b->record_disabled))
3765                 goto out;
3766
3767         /*
3768          * We can't do a synchronize_sched here because this
3769          * function can be called in atomic context.
3770          * Normally this will be called from the same CPU as cpu.
3771          * If not it's up to the caller to protect this.
3772          */
3773         atomic_inc(&cpu_buffer_a->record_disabled);
3774         atomic_inc(&cpu_buffer_b->record_disabled);
3775
3776         ret = -EBUSY;
3777         if (local_read(&cpu_buffer_a->committing))
3778                 goto out_dec;
3779         if (local_read(&cpu_buffer_b->committing))
3780                 goto out_dec;
3781
3782         buffer_a->buffers[cpu] = cpu_buffer_b;
3783         buffer_b->buffers[cpu] = cpu_buffer_a;
3784
3785         cpu_buffer_b->buffer = buffer_a;
3786         cpu_buffer_a->buffer = buffer_b;
3787
3788         ret = 0;
3789
3790 out_dec:
3791         atomic_dec(&cpu_buffer_a->record_disabled);
3792         atomic_dec(&cpu_buffer_b->record_disabled);
3793 out:
3794         return ret;
3795 }
3796 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
3797 #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
3798
3799 /**
3800  * ring_buffer_alloc_read_page - allocate a page to read from buffer
3801  * @buffer: the buffer to allocate for.
3802  *
3803  * This function is used in conjunction with ring_buffer_read_page.
3804  * When reading a full page from the ring buffer, these functions
3805  * can be used to speed up the process. The calling function should
3806  * allocate a few pages first with this function. Then when it
3807  * needs to get pages from the ring buffer, it passes the result
3808  * of this function into ring_buffer_read_page, which will swap
3809  * the page that was allocated, with the read page of the buffer.
3810  *
3811  * Returns:
3812  *  The page allocated, or NULL on error.
3813  */
3814 void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu)
3815 {
3816         struct buffer_data_page *bpage;
3817         struct page *page;
3818
3819         page = alloc_pages_node(cpu_to_node(cpu),
3820                                 GFP_KERNEL | __GFP_NORETRY, 0);
3821         if (!page)
3822                 return NULL;
3823
3824         bpage = page_address(page);
3825
3826         rb_init_page(bpage);
3827
3828         return bpage;
3829 }
3830 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
3831
3832 /**
3833  * ring_buffer_free_read_page - free an allocated read page
3834  * @buffer: the buffer the page was allocate for
3835  * @data: the page to free
3836  *
3837  * Free a page allocated from ring_buffer_alloc_read_page.
3838  */
3839 void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
3840 {
3841         free_page((unsigned long)data);
3842 }
3843 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
3844
3845 /**
3846  * ring_buffer_read_page - extract a page from the ring buffer
3847  * @buffer: buffer to extract from
3848  * @data_page: the page to use allocated from ring_buffer_alloc_read_page
3849  * @len: amount to extract
3850  * @cpu: the cpu of the buffer to extract
3851  * @full: should the extraction only happen when the page is full.
3852  *
3853  * This function will pull out a page from the ring buffer and consume it.
3854  * @data_page must be the address of the variable that was returned
3855  * from ring_buffer_alloc_read_page. This is because the page might be used
3856  * to swap with a page in the ring buffer.
3857  *
3858  * for example:
3859  *      rpage = ring_buffer_alloc_read_page(buffer);
3860  *      if (!rpage)
3861  *              return error;
3862  *      ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
3863  *      if (ret >= 0)
3864  *              process_page(rpage, ret);
3865  *
3866  * When @full is set, the function will not return true unless
3867  * the writer is off the reader page.
3868  *
3869  * Note: it is up to the calling functions to handle sleeps and wakeups.
3870  *  The ring buffer can be used anywhere in the kernel and can not
3871  *  blindly call wake_up. The layer that uses the ring buffer must be
3872  *  responsible for that.
3873  *
3874  * Returns:
3875  *  >=0 if data has been transferred, returns the offset of consumed data.
3876  *  <0 if no data has been transferred.
3877  */
3878 int ring_buffer_read_page(struct ring_buffer *buffer,
3879                           void **data_page, size_t len, int cpu, int full)
3880 {
3881         struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
3882         struct ring_buffer_event *event;
3883         struct buffer_data_page *bpage;
3884         struct buffer_page *reader;
3885         unsigned long missed_events;
3886         unsigned long flags;
3887         unsigned int commit;
3888         unsigned int read;
3889         u64 save_timestamp;
3890         int ret = -1;
3891
3892         if (!cpumask_test_cpu(cpu, buffer->cpumask))
3893                 goto out;
3894
3895         /*
3896          * If len is not big enough to hold the page header, then
3897          * we can not copy anything.
3898          */
3899         if (len <= BUF_PAGE_HDR_SIZE)
3900                 goto out;
3901
3902         len -= BUF_PAGE_HDR_SIZE;
3903
3904         if (!data_page)
3905                 goto out;
3906
3907         bpage = *data_page;
3908         if (!bpage)
3909                 goto out;
3910
3911         raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3912
3913         reader = rb_get_reader_page(cpu_buffer);
3914         if (!reader)
3915                 goto out_unlock;
3916
3917         event = rb_reader_event(cpu_buffer);
3918
3919         read = reader->read;
3920         commit = rb_page_commit(reader);
3921
3922         /* Check if any events were dropped */
3923         missed_events = cpu_buffer->lost_events;
3924
3925         /*
3926          * If this page has been partially read or
3927          * if len is not big enough to read the rest of the page or
3928          * a writer is still on the page, then
3929          * we must copy the data from the page to the buffer.
3930          * Otherwise, we can simply swap the page with the one passed in.
3931          */
3932         if (read || (len < (commit - read)) ||
3933             cpu_buffer->reader_page == cpu_buffer->commit_page) {
3934                 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
3935                 unsigned int rpos = read;
3936                 unsigned int pos = 0;
3937                 unsigned int size;
3938
3939                 if (full)
3940                         goto out_unlock;
3941
3942                 if (len > (commit - read))
3943                         len = (commit - read);
3944
3945                 /* Always keep the time extend and data together */
3946                 size = rb_event_ts_length(event);
3947
3948                 if (len < size)
3949                         goto out_unlock;
3950
3951                 /* save the current timestamp, since the user will need it */
3952                 save_timestamp = cpu_buffer->read_stamp;
3953
3954                 /* Need to copy one event at a time */
3955                 do {
3956                         /* We need the size of one event, because
3957                          * rb_advance_reader only advances by one event,
3958                          * whereas rb_event_ts_length may include the size of
3959                          * one or two events.
3960                          * We have already ensured there's enough space if this
3961                          * is a time extend. */
3962                         size = rb_event_length(event);
3963                         memcpy(bpage->data + pos, rpage->data + rpos, size);
3964
3965                         len -= size;
3966
3967                         rb_advance_reader(cpu_buffer);
3968                         rpos = reader->read;
3969                         pos += size;
3970
3971                         if (rpos >= commit)
3972                                 break;
3973
3974                         event = rb_reader_event(cpu_buffer);
3975                         /* Always keep the time extend and data together */
3976                         size = rb_event_ts_length(event);
3977                 } while (len >= size);
3978
3979                 /* update bpage */
3980                 local_set(&bpage->commit, pos);
3981                 bpage->time_stamp = save_timestamp;
3982
3983                 /* we copied everything to the beginning */
3984                 read = 0;
3985         } else {
3986                 /* update the entry counter */
3987                 cpu_buffer->read += rb_page_entries(reader);
3988                 cpu_buffer->read_bytes += BUF_PAGE_SIZE;
3989
3990                 /* swap the pages */
3991                 rb_init_page(bpage);
3992                 bpage = reader->page;
3993                 reader->page = *data_page;
3994                 local_set(&reader->write, 0);
3995                 local_set(&reader->entries, 0);
3996                 reader->read = 0;
3997                 *data_page = bpage;
3998
3999                 /*
4000                  * Use the real_end for the data size,
4001                  * This gives us a chance to store the lost events
4002                  * on the page.
4003                  */
4004                 if (reader->real_end)
4005                         local_set(&bpage->commit, reader->real_end);
4006         }
4007         ret = read;
4008
4009         cpu_buffer->lost_events = 0;
4010
4011         commit = local_read(&bpage->commit);
4012         /*
4013          * Set a flag in the commit field if we lost events
4014          */
4015         if (missed_events) {
4016                 /* If there is room at the end of the page to save the
4017                  * missed events, then record it there.
4018                  */
4019                 if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) {
4020                         memcpy(&bpage->data[commit], &missed_events,
4021                                sizeof(missed_events));
4022                         local_add(RB_MISSED_STORED, &bpage->commit);
4023                         commit += sizeof(missed_events);
4024                 }
4025                 local_add(RB_MISSED_EVENTS, &bpage->commit);
4026         }
4027
4028         /*
4029          * This page may be off to user land. Zero it out here.
4030          */
4031         if (commit < BUF_PAGE_SIZE)
4032                 memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit);
4033
4034  out_unlock:
4035         raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4036
4037  out:
4038         return ret;
4039 }
4040 EXPORT_SYMBOL_GPL(ring_buffer_read_page);
4041
4042 #ifdef CONFIG_TRACING
4043 static ssize_t
4044 rb_simple_read(struct file *filp, char __user *ubuf,
4045                size_t cnt, loff_t *ppos)
4046 {
4047         unsigned long *p = filp->private_data;
4048         char buf[64];
4049         int r;
4050
4051         if (test_bit(RB_BUFFERS_DISABLED_BIT, p))
4052                 r = sprintf(buf, "permanently disabled\n");
4053         else
4054                 r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p));
4055
4056         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4057 }
4058
4059 static ssize_t
4060 rb_simple_write(struct file *filp, const char __user *ubuf,
4061                 size_t cnt, loff_t *ppos)
4062 {
4063         unsigned long *p = filp->private_data;
4064         unsigned long val;
4065         int ret;
4066
4067         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4068         if (ret)
4069                 return ret;
4070
4071         if (val)
4072                 set_bit(RB_BUFFERS_ON_BIT, p);
4073         else
4074                 clear_bit(RB_BUFFERS_ON_BIT, p);
4075
4076         (*ppos)++;
4077
4078         return cnt;
4079 }
4080
4081 static const struct file_operations rb_simple_fops = {
4082         .open           = tracing_open_generic,
4083         .read           = rb_simple_read,
4084         .write          = rb_simple_write,
4085         .llseek         = default_llseek,
4086 };
4087
4088
4089 static __init int rb_init_debugfs(void)
4090 {
4091         struct dentry *d_tracer;
4092
4093         d_tracer = tracing_init_dentry();
4094
4095         trace_create_file("tracing_on", 0644, d_tracer,
4096                             &ring_buffer_flags, &rb_simple_fops);
4097
4098         return 0;
4099 }
4100
4101 fs_initcall(rb_init_debugfs);
4102 #endif
4103
4104 #ifdef CONFIG_HOTPLUG_CPU
4105 static int rb_cpu_notify(struct notifier_block *self,
4106                          unsigned long action, void *hcpu)
4107 {
4108         struct ring_buffer *buffer =
4109                 container_of(self, struct ring_buffer, cpu_notify);
4110         long cpu = (long)hcpu;
4111
4112         switch (action) {
4113         case CPU_UP_PREPARE:
4114         case CPU_UP_PREPARE_FROZEN:
4115                 if (cpumask_test_cpu(cpu, buffer->cpumask))
4116                         return NOTIFY_OK;
4117
4118                 buffer->buffers[cpu] =
4119                         rb_allocate_cpu_buffer(buffer, cpu);
4120                 if (!buffer->buffers[cpu]) {
4121                         WARN(1, "failed to allocate ring buffer on CPU %ld\n",
4122                              cpu);
4123                         return NOTIFY_OK;
4124                 }
4125                 smp_wmb();
4126                 cpumask_set_cpu(cpu, buffer->cpumask);
4127                 break;
4128         case CPU_DOWN_PREPARE:
4129         case CPU_DOWN_PREPARE_FROZEN:
4130                 /*
4131                  * Do nothing.
4132                  *  If we were to free the buffer, then the user would
4133                  *  lose any trace that was in the buffer.
4134                  */
4135                 break;
4136         default:
4137                 break;
4138         }
4139         return NOTIFY_OK;
4140 }
4141 #endif