libtracing/relay.h

   1 /*
   2  * linux/include/linux/ltt-relay.h
   3  *
   4  * Copyright (C) 2002, 2003 - Tom Zanussi (zanussi@us.ibm.com), IBM Corp
   5  * Copyright (C) 1999, 2000, 2001, 2002 - Karim Yaghmour (karim@opersys.com)
   6  * Copyright (C) 2008 - Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca)
   7  *
   8  * CONFIG_RELAY definitions and declarations
   9  */
  10
  11 #ifndef _LINUX_LTT_RELAY_H
  12 #define _LINUX_LTT_RELAY_H
  13
  14 //ust// #include <linux/types.h>
  15 //ust// #include <linux/sched.h>
  16 //ust// #include <linux/timer.h>
  17 //ust// #include <linux/wait.h>
  18 //ust// #include <linux/list.h>
  19 //ust// #include <linux/fs.h>
  20 //ust// #include <linux/poll.h>
  21 //ust// #include <linux/kref.h>
  22 //ust// #include <linux/mm.h>
  23 //ust// #include <linux/ltt-core.h>
  24 #include "kref.h"
  25 #include "list.h"
  26
  27 /* Needs a _much_ better name... */
  28 #define FIX_SIZE(x) ((((x) - 1) & PAGE_MASK) + PAGE_SIZE)
  29
  30 /*
  31  * Tracks changes to rchan/rchan_buf structs
  32  */
  33 #define LTT_RELAY_CHANNEL_VERSION               8
  34
  35 struct rchan_buf;
  36
  37 struct buf_page {
  38         struct page *page;
  39         struct rchan_buf *buf;  /* buffer the page belongs to */
  40         size_t offset;          /* page offset in the buffer */
  41         struct list_head list;  /* buffer linked list */
  42 };
  43
  44 /*
  45  * Per-cpu relay channel buffer
  46  */
  47 struct rchan_buf {
  48         struct rchan *chan;             /* associated channel */
  49 //ust// wait_queue_head_t read_wait;    /* reader wait queue */
  50 //ust// struct timer_list timer;        /* reader wake-up timer */
  51 //ust// struct dentry *dentry;          /* channel file dentry */
  52         struct kref kref;               /* channel buffer refcount */
  53 //ust// struct list_head pages;         /* list of buffer pages */
  54         void *buf_data; //ust//
  55         size_t buf_size;
  56 //ust// struct buf_page *wpage;         /* current write page (cache) */
  57 //ust// struct buf_page *hpage[2];      /* current subbuf header page (cache) */
  58 //ust// struct buf_page *rpage;         /* current subbuf read page (cache) */
  59 //ust// unsigned int page_count;        /* number of current buffer pages */
  60         unsigned int finalized;         /* buffer has been finalized */
  61 //ust// unsigned int cpu;               /* this buf's cpu */
  62         int shmid;
  63 } ____cacheline_aligned;
  64
  65 /*
  66  * Relay channel data structure
  67  */
  68 struct rchan {
  69         u32 version;                    /* the version of this struct */
  70         size_t subbuf_size;             /* sub-buffer size */
  71         size_t n_subbufs;               /* number of sub-buffers per buffer */
  72         size_t alloc_size;              /* total buffer size allocated */
  73         struct rchan_callbacks *cb;     /* client callbacks */
  74         struct kref kref;               /* channel refcount */
  75         void *private_data;             /* for user-defined data */
  76 //ust// struct rchan_buf *buf[NR_CPUS]; /* per-cpu channel buffers */
  77         struct rchan_buf *buf;
  78         struct list_head list;          /* for channel list */
  79         struct dentry *parent;          /* parent dentry passed to open */
  80         int subbuf_size_order;          /* order of sub-buffer size */
  81 //ust// char base_filename[NAME_MAX];   /* saved base filename */
  82 };
  83
  84 /*
  85  * Relay channel client callbacks
  86  */
  87 struct rchan_callbacks {
  88         /*
  89          * subbuf_start - called on buffer-switch to a new sub-buffer
  90          * @buf: the channel buffer containing the new sub-buffer
  91          * @subbuf: the start of the new sub-buffer
  92          * @prev_subbuf: the start of the previous sub-buffer
  93          * @prev_padding: unused space at the end of previous sub-buffer
  94          *
  95          * The client should return 1 to continue logging, 0 to stop
  96          * logging.
  97          *
  98          * NOTE: subbuf_start will also be invoked when the buffer is
  99          *       created, so that the first sub-buffer can be initialized
 100          *       if necessary.  In this case, prev_subbuf will be NULL.
 101          *
 102          * NOTE: the client can reserve bytes at the beginning of the new
 103          *       sub-buffer by calling subbuf_start_reserve() in this callback.
 104          */
 105         int (*subbuf_start) (struct rchan_buf *buf,
 106                              void *subbuf,
 107                              void *prev_subbuf,
 108                              size_t prev_padding);
 109
 110         /*
 111          * create_buf_file - create file to represent a relay channel buffer
 112          * @filename: the name of the file to create
 113          * @parent: the parent of the file to create
 114          * @mode: the mode of the file to create
 115          * @buf: the channel buffer
 116          *
 117          * Called during relay_open(), once for each per-cpu buffer,
 118          * to allow the client to create a file to be used to
 119          * represent the corresponding channel buffer.  If the file is
 120          * created outside of relay, the parent must also exist in
 121          * that filesystem.
 122          *
 123          * The callback should return the dentry of the file created
 124          * to represent the relay buffer.
 125          *
 126          * Setting the is_global outparam to a non-zero value will
 127          * cause relay_open() to create a single global buffer rather
 128          * than the default set of per-cpu buffers.
 129          *
 130          * See Documentation/filesystems/relayfs.txt for more info.
 131          */
 132         struct dentry *(*create_buf_file)(const char *filename,
 133                                           struct dentry *parent,
 134                                           int mode,
 135                                           struct rchan_buf *buf);
 136
 137         /*
 138          * remove_buf_file - remove file representing a relay channel buffer
 139          * @dentry: the dentry of the file to remove
 140          *
 141          * Called during relay_close(), once for each per-cpu buffer,
 142          * to allow the client to remove a file used to represent a
 143          * channel buffer.
 144          *
 145          * The callback should return 0 if successful, negative if not.
 146          */
 147 //ust// int (*remove_buf_file)(struct rchan_buf *buf);
 148 };
 149
 150 extern struct buf_page *ltt_relay_find_prev_page(struct rchan_buf *buf,
 151         struct buf_page *page, size_t offset, ssize_t diff_offset);
 152
 153 extern struct buf_page *ltt_relay_find_next_page(struct rchan_buf *buf,
 154         struct buf_page *page, size_t offset, ssize_t diff_offset);
 155
 156 extern void _ltt_relay_write(struct rchan_buf *buf, size_t offset,
 157         const void *src, size_t len, ssize_t cpy);
 158
 159 extern int ltt_relay_read(struct rchan_buf *buf, size_t offset,
 160         void *dest, size_t len);
 161
 162 extern struct buf_page *ltt_relay_read_get_page(struct rchan_buf *buf,
 163         size_t offset);
 164
 165 /*
 166  * Return the address where a given offset is located.
 167  * Should be used to get the current subbuffer header pointer. Given we know
 168  * it's never on a page boundary, it's safe to write directly to this address,
 169  * as long as the write is never bigger than a page size.
 170  */
 171 extern void *ltt_relay_offset_address(struct rchan_buf *buf,
 172         size_t offset);
 173
 174 /*
 175  * Find the page containing "offset". Cache it if it is after the currently
 176  * cached page.
 177  */
 178 static inline struct buf_page *ltt_relay_cache_page(struct rchan_buf *buf,
 179                 struct buf_page **page_cache,
 180                 struct buf_page *page, size_t offset)
 181 {
 182         ssize_t diff_offset;
 183         ssize_t half_buf_size = buf->chan->alloc_size >> 1;
 184
 185         /*
 186          * Make sure this is the page we want to write into. The current
 187          * page is changed concurrently by other writers. [wrh]page are
 188          * used as a cache remembering the last page written
 189          * to/read/looked up for header address. No synchronization;
 190          * could have to find the previous page is a nested write
 191          * occured. Finding the right page is done by comparing the
 192          * dest_offset with the buf_page offsets.
 193          * When at the exact opposite of the buffer, bias towards forward search
 194          * because it will be cached.
 195          */
 196
 197         diff_offset = (ssize_t)offset - (ssize_t)page->offset;
 198         if (diff_offset <= -(ssize_t)half_buf_size)
 199                 diff_offset += buf->chan->alloc_size;
 200         else if (diff_offset > half_buf_size)
 201                 diff_offset -= buf->chan->alloc_size;
 202
 203         if (unlikely(diff_offset >= (ssize_t)PAGE_SIZE)) {
 204                 page = ltt_relay_find_next_page(buf, page, offset, diff_offset);
 205                 *page_cache = page;
 206         } else if (unlikely(diff_offset < 0)) {
 207                 page = ltt_relay_find_prev_page(buf, page, offset, diff_offset);
 208         }
 209         return page;
 210 }
 211
 212 //ust// #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
 213  static inline void ltt_relay_do_copy(void *dest, const void *src, size_t len)
 214 {
 215         switch (len) {
 216         case 0: break;
 217         case 1: *(u8 *)dest = *(const u8 *)src;
 218                 break;
 219         case 2: *(u16 *)dest = *(const u16 *)src;
 220                 break;
 221         case 4: *(u32 *)dest = *(const u32 *)src;
 222                 break;
 223 //ust// #if (BITS_PER_LONG == 64)
 224         case 8: *(u64 *)dest = *(const u64 *)src;
 225                 break;
 226 //ust// #endif
 227         default:
 228                 memcpy(dest, src, len);
 229         }
 230 }
 231 //ust// #else
 232 //ust// /*
 233 //ust//  * Returns whether the dest and src addresses are aligned on
 234 //ust//  * min(sizeof(void *), len). Call this with statically known len for efficiency.
 235 //ust//  */
 236 //ust// static inline int addr_aligned(const void *dest, const void *src, size_t len)
 237 //ust// {
 238 //ust//         if (ltt_align((size_t)dest, len))
 239 //ust//                 return 0;
 240 //ust//         if (ltt_align((size_t)src, len))
 241 //ust//                 return 0;
 242 //ust//         return 1;
 243 //ust// }
 244 //ust//
 245 //ust// static inline void ltt_relay_do_copy(void *dest, const void *src, size_t len)
 246 //ust// {
 247 //ust//         switch (len) {
 248 //ust//         case 0: break;
 249 //ust//         case 1: *(u8 *)dest = *(const u8 *)src;
 250 //ust//                 break;
 251 //ust//         case 2: if (unlikely(!addr_aligned(dest, src, 2)))
 252 //ust//                         goto memcpy_fallback;
 253 //ust//                 *(u16 *)dest = *(const u16 *)src;
 254 //ust//                 break;
 255 //ust//         case 4: if (unlikely(!addr_aligned(dest, src, 4)))
 256 //ust//                         goto memcpy_fallback;
 257 //ust//                 *(u32 *)dest = *(const u32 *)src;
 258 //ust//                 break;
 259 //ust// #if (BITS_PER_LONG == 64)
 260 //ust//         case 8: if (unlikely(!addr_aligned(dest, src, 8)))
 261 //ust//                         goto memcpy_fallback;
 262 //ust//                 *(u64 *)dest = *(const u64 *)src;
 263 //ust//                 break;
 264 //ust// #endif
 265 //ust//         default:
 266 //ust//                 goto memcpy_fallback;
 267 //ust//         }
 268 //ust//         return;
 269 //ust// memcpy_fallback:
 270 //ust//         memcpy(dest, src, len);
 271 //ust// }
 272 //ust// #endif
 273
 274 static inline int ltt_relay_write(struct rchan_buf *buf, size_t offset,
 275         const void *src, size_t len)
 276 {
 277 //ust// struct buf_page *page;
 278 //ust// ssize_t pagecpy;
 279 //ust//
 280 //ust// offset &= buf->chan->alloc_size - 1;
 281 //ust// page = buf->wpage;
 282 //ust//
 283 //ust// page = ltt_relay_cache_page(buf, &buf->wpage, page, offset);
 284 //ust// pagecpy = min_t(size_t, len, PAGE_SIZE - (offset & ~PAGE_MASK));
 285 //ust// ltt_relay_do_copy(page_address(page->page)
 286 //ust//         + (offset & ~PAGE_MASK), src, pagecpy);
 287 //ust//
 288 //ust// if (unlikely(len != pagecpy))
 289 //ust//         _ltt_relay_write(buf, offset, src, len, page, pagecpy);
 290 //ust// return len;
 291
 292
 293         size_t cpy;
 294         cpy = min_t(size_t, len, buf->buf_size - offset);
 295         ltt_relay_do_copy(buf->buf_data + offset, src, cpy);
 296
 297         if (unlikely(len != cpy))
 298                 _ltt_relay_write(buf, offset, src, len, cpy);
 299         return len;
 300 }
 301
 302 /*
 303  * CONFIG_LTT_RELAY kernel API, ltt/ltt-relay-alloc.c
 304  */
 305
 306 struct rchan *ltt_relay_open(const char *base_filename,
 307                          struct dentry *parent,
 308                          size_t subbuf_size,
 309                          size_t n_subbufs,
 310                          void *private_data);
 311 extern void ltt_relay_close(struct rchan *chan);
 312
 313 /*
 314  * exported ltt_relay file operations, ltt/ltt-relay-alloc.c
 315  */
 316 extern const struct file_operations ltt_relay_file_operations;
 317
 318
 319 /* LTTng lockless logging buffer info */
 320 struct ltt_channel_buf_struct {
 321         /* First 32 bytes cache-hot cacheline */
 322         local_t offset;                 /* Current offset in the buffer */
 323         local_t *commit_count;          /* Commit count per sub-buffer */
 324         atomic_long_t consumed;         /*
 325                                          * Current offset in the buffer
 326                                          * standard atomic access (shared)
 327                                          */
 328         unsigned long last_tsc;         /*
 329                                          * Last timestamp written in the buffer.
 330                                          */
 331         /* End of first 32 bytes cacheline */
 332         atomic_long_t active_readers;   /*
 333                                          * Active readers count
 334                                          * standard atomic access (shared)
 335                                          */
 336         local_t events_lost;
 337         local_t corrupted_subbuffers;
 338         spinlock_t full_lock;           /*
 339                                          * buffer full condition spinlock, only
 340                                          * for userspace tracing blocking mode
 341                                          * synchronization with reader.
 342                                          */
 343 //ust// wait_queue_head_t write_wait;   /*
 344 //ust//                                  * Wait queue for blocking user space
 345 //ust//                                  * writers
 346 //ust//                                  */
 347         atomic_t wakeup_readers;        /* Boolean : wakeup readers waiting ? */
 348 } ____cacheline_aligned;
 349
 350 int ltt_do_get_subbuf(struct rchan_buf *buf, struct ltt_channel_buf_struct *ltt_buf, long *pconsumed_old);
 351
 352 int ltt_do_put_subbuf(struct rchan_buf *buf, struct ltt_channel_buf_struct *ltt_buf, u32 uconsumed_old);
 353
 354
 355 #endif /* _LINUX_LTT_RELAY_H */
 356