| 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 <assert.h> |
| 25 | #include <kcompat/kref.h> |
| 26 | //#include "list.h" |
| 27 | #include "channels.h" |
| 28 | #include "buffer.h" |
| 29 | |
| 30 | /* Needs a _much_ better name... */ |
| 31 | #define FIX_SIZE(x) ((((x) - 1) & PAGE_MASK) + PAGE_SIZE) |
| 32 | |
| 33 | /* |
| 34 | * Tracks changes to rchan/rchan_buf structs |
| 35 | */ |
| 36 | #define LTT_RELAY_CHANNEL_VERSION 8 |
| 37 | |
| 38 | struct rchan_buf; |
| 39 | |
| 40 | struct buf_page { |
| 41 | struct page *page; |
| 42 | struct rchan_buf *buf; /* buffer the page belongs to */ |
| 43 | size_t offset; /* page offset in the buffer */ |
| 44 | struct list_head list; /* buffer linked list */ |
| 45 | }; |
| 46 | |
| 47 | /* |
| 48 | * Per-cpu relay channel buffer |
| 49 | */ |
| 50 | struct rchan_buf { |
| 51 | struct rchan *chan; /* associated channel */ |
| 52 | //ust// wait_queue_head_t read_wait; /* reader wait queue */ |
| 53 | //ust// struct timer_list timer; /* reader wake-up timer */ |
| 54 | //ust// struct dentry *dentry; /* channel file dentry */ |
| 55 | struct kref kref; /* channel buffer refcount */ |
| 56 | //ust// struct list_head pages; /* list of buffer pages */ |
| 57 | void *buf_data; //ust// |
| 58 | size_t buf_size; |
| 59 | //ust// struct buf_page *wpage; /* current write page (cache) */ |
| 60 | //ust// struct buf_page *hpage[2]; /* current subbuf header page (cache) */ |
| 61 | //ust// struct buf_page *rpage; /* current subbuf read page (cache) */ |
| 62 | //ust// unsigned int page_count; /* number of current buffer pages */ |
| 63 | unsigned int finalized; /* buffer has been finalized */ |
| 64 | //ust// unsigned int cpu; /* this buf's cpu */ |
| 65 | int shmid; /* the shmid of the buffer data pages */ |
| 66 | } ____cacheline_aligned; |
| 67 | |
| 68 | /* |
| 69 | * Relay channel data structure |
| 70 | */ |
| 71 | struct rchan { |
| 72 | u32 version; /* the version of this struct */ |
| 73 | size_t subbuf_size; /* sub-buffer size */ |
| 74 | size_t n_subbufs; /* number of sub-buffers per buffer */ |
| 75 | size_t alloc_size; /* total buffer size allocated */ |
| 76 | struct rchan_callbacks *cb; /* client callbacks */ |
| 77 | struct kref kref; /* channel refcount */ |
| 78 | void *private_data; /* for user-defined data */ |
| 79 | //ust// struct rchan_buf *buf[NR_CPUS]; /* per-cpu channel buffers */ |
| 80 | struct rchan_buf *buf; |
| 81 | struct list_head list; /* for channel list */ |
| 82 | struct dentry *parent; /* parent dentry passed to open */ |
| 83 | int subbuf_size_order; /* order of sub-buffer size */ |
| 84 | //ust// char base_filename[NAME_MAX]; /* saved base filename */ |
| 85 | }; |
| 86 | |
| 87 | /* |
| 88 | * Relay channel client callbacks |
| 89 | */ |
| 90 | struct rchan_callbacks { |
| 91 | /* |
| 92 | * subbuf_start - called on buffer-switch to a new sub-buffer |
| 93 | * @buf: the channel buffer containing the new sub-buffer |
| 94 | * @subbuf: the start of the new sub-buffer |
| 95 | * @prev_subbuf: the start of the previous sub-buffer |
| 96 | * @prev_padding: unused space at the end of previous sub-buffer |
| 97 | * |
| 98 | * The client should return 1 to continue logging, 0 to stop |
| 99 | * logging. |
| 100 | * |
| 101 | * NOTE: subbuf_start will also be invoked when the buffer is |
| 102 | * created, so that the first sub-buffer can be initialized |
| 103 | * if necessary. In this case, prev_subbuf will be NULL. |
| 104 | * |
| 105 | * NOTE: the client can reserve bytes at the beginning of the new |
| 106 | * sub-buffer by calling subbuf_start_reserve() in this callback. |
| 107 | */ |
| 108 | int (*subbuf_start) (struct rchan_buf *buf, |
| 109 | void *subbuf, |
| 110 | void *prev_subbuf, |
| 111 | size_t prev_padding); |
| 112 | |
| 113 | /* |
| 114 | * create_buf_file - create file to represent a relay channel buffer |
| 115 | * @filename: the name of the file to create |
| 116 | * @parent: the parent of the file to create |
| 117 | * @mode: the mode of the file to create |
| 118 | * @buf: the channel buffer |
| 119 | * |
| 120 | * Called during relay_open(), once for each per-cpu buffer, |
| 121 | * to allow the client to create a file to be used to |
| 122 | * represent the corresponding channel buffer. If the file is |
| 123 | * created outside of relay, the parent must also exist in |
| 124 | * that filesystem. |
| 125 | * |
| 126 | * The callback should return the dentry of the file created |
| 127 | * to represent the relay buffer. |
| 128 | * |
| 129 | * Setting the is_global outparam to a non-zero value will |
| 130 | * cause relay_open() to create a single global buffer rather |
| 131 | * than the default set of per-cpu buffers. |
| 132 | * |
| 133 | * See Documentation/filesystems/relayfs.txt for more info. |
| 134 | */ |
| 135 | struct dentry *(*create_buf_file)(const char *filename, |
| 136 | struct dentry *parent, |
| 137 | int mode, |
| 138 | struct rchan_buf *buf); |
| 139 | |
| 140 | /* |
| 141 | * remove_buf_file - remove file representing a relay channel buffer |
| 142 | * @dentry: the dentry of the file to remove |
| 143 | * |
| 144 | * Called during relay_close(), once for each per-cpu buffer, |
| 145 | * to allow the client to remove a file used to represent a |
| 146 | * channel buffer. |
| 147 | * |
| 148 | * The callback should return 0 if successful, negative if not. |
| 149 | */ |
| 150 | //ust// int (*remove_buf_file)(struct rchan_buf *buf); |
| 151 | }; |
| 152 | |
| 153 | extern struct buf_page *ltt_relay_find_prev_page(struct rchan_buf *buf, |
| 154 | struct buf_page *page, size_t offset, ssize_t diff_offset); |
| 155 | |
| 156 | extern struct buf_page *ltt_relay_find_next_page(struct rchan_buf *buf, |
| 157 | struct buf_page *page, size_t offset, ssize_t diff_offset); |
| 158 | |
| 159 | extern void _ltt_relay_write(struct rchan_buf *buf, size_t offset, |
| 160 | const void *src, size_t len, ssize_t cpy); |
| 161 | |
| 162 | extern int ltt_relay_read(struct rchan_buf *buf, size_t offset, |
| 163 | void *dest, size_t len); |
| 164 | |
| 165 | extern struct buf_page *ltt_relay_read_get_page(struct rchan_buf *buf, |
| 166 | size_t offset); |
| 167 | |
| 168 | /* |
| 169 | * Return the address where a given offset is located. |
| 170 | * Should be used to get the current subbuffer header pointer. Given we know |
| 171 | * it's never on a page boundary, it's safe to write directly to this address, |
| 172 | * as long as the write is never bigger than a page size. |
| 173 | */ |
| 174 | extern void *ltt_relay_offset_address(struct rchan_buf *buf, |
| 175 | size_t offset); |
| 176 | |
| 177 | /* |
| 178 | * Find the page containing "offset". Cache it if it is after the currently |
| 179 | * cached page. |
| 180 | */ |
| 181 | static inline struct buf_page *ltt_relay_cache_page(struct rchan_buf *buf, |
| 182 | struct buf_page **page_cache, |
| 183 | struct buf_page *page, size_t offset) |
| 184 | { |
| 185 | ssize_t diff_offset; |
| 186 | ssize_t half_buf_size = buf->chan->alloc_size >> 1; |
| 187 | |
| 188 | /* |
| 189 | * Make sure this is the page we want to write into. The current |
| 190 | * page is changed concurrently by other writers. [wrh]page are |
| 191 | * used as a cache remembering the last page written |
| 192 | * to/read/looked up for header address. No synchronization; |
| 193 | * could have to find the previous page is a nested write |
| 194 | * occured. Finding the right page is done by comparing the |
| 195 | * dest_offset with the buf_page offsets. |
| 196 | * When at the exact opposite of the buffer, bias towards forward search |
| 197 | * because it will be cached. |
| 198 | */ |
| 199 | |
| 200 | diff_offset = (ssize_t)offset - (ssize_t)page->offset; |
| 201 | if (diff_offset <= -(ssize_t)half_buf_size) |
| 202 | diff_offset += buf->chan->alloc_size; |
| 203 | else if (diff_offset > half_buf_size) |
| 204 | diff_offset -= buf->chan->alloc_size; |
| 205 | |
| 206 | if (unlikely(diff_offset >= (ssize_t)PAGE_SIZE)) { |
| 207 | page = ltt_relay_find_next_page(buf, page, offset, diff_offset); |
| 208 | *page_cache = page; |
| 209 | } else if (unlikely(diff_offset < 0)) { |
| 210 | page = ltt_relay_find_prev_page(buf, page, offset, diff_offset); |
| 211 | } |
| 212 | return page; |
| 213 | } |
| 214 | |
| 215 | //ust// #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS |
| 216 | static inline void ltt_relay_do_copy(void *dest, const void *src, size_t len) |
| 217 | { |
| 218 | union { |
| 219 | const void *src; |
| 220 | const u8 *src8; |
| 221 | const u16 *src16; |
| 222 | const u32 *src32; |
| 223 | const u64 *src64; |
| 224 | } u = { .src = src }; |
| 225 | |
| 226 | switch (len) { |
| 227 | case 0: break; |
| 228 | case 1: *(u8 *)dest = *u.src8; |
| 229 | break; |
| 230 | case 2: *(u16 *)dest = *u.src16; |
| 231 | break; |
| 232 | case 4: *(u32 *)dest = *u.src32; |
| 233 | break; |
| 234 | //ust// #if (BITS_PER_LONG == 64) |
| 235 | case 8: *(u64 *)dest = *u.src64; |
| 236 | break; |
| 237 | //ust// #endif |
| 238 | default: |
| 239 | memcpy(dest, src, len); |
| 240 | } |
| 241 | } |
| 242 | //ust// #else |
| 243 | //ust// /* |
| 244 | //ust// * Returns whether the dest and src addresses are aligned on |
| 245 | //ust// * min(sizeof(void *), len). Call this with statically known len for efficiency. |
| 246 | //ust// */ |
| 247 | //ust// static inline int addr_aligned(const void *dest, const void *src, size_t len) |
| 248 | //ust// { |
| 249 | //ust// if (ltt_align((size_t)dest, len)) |
| 250 | //ust// return 0; |
| 251 | //ust// if (ltt_align((size_t)src, len)) |
| 252 | //ust// return 0; |
| 253 | //ust// return 1; |
| 254 | //ust// } |
| 255 | //ust// |
| 256 | //ust// static inline void ltt_relay_do_copy(void *dest, const void *src, size_t len) |
| 257 | //ust// { |
| 258 | //ust// switch (len) { |
| 259 | //ust// case 0: break; |
| 260 | //ust// case 1: *(u8 *)dest = *(const u8 *)src; |
| 261 | //ust// break; |
| 262 | //ust// case 2: if (unlikely(!addr_aligned(dest, src, 2))) |
| 263 | //ust// goto memcpy_fallback; |
| 264 | //ust// *(u16 *)dest = *(const u16 *)src; |
| 265 | //ust// break; |
| 266 | //ust// case 4: if (unlikely(!addr_aligned(dest, src, 4))) |
| 267 | //ust// goto memcpy_fallback; |
| 268 | //ust// *(u32 *)dest = *(const u32 *)src; |
| 269 | //ust// break; |
| 270 | //ust// #if (BITS_PER_LONG == 64) |
| 271 | //ust// case 8: if (unlikely(!addr_aligned(dest, src, 8))) |
| 272 | //ust// goto memcpy_fallback; |
| 273 | //ust// *(u64 *)dest = *(const u64 *)src; |
| 274 | //ust// break; |
| 275 | //ust// #endif |
| 276 | //ust// default: |
| 277 | //ust// goto memcpy_fallback; |
| 278 | //ust// } |
| 279 | //ust// return; |
| 280 | //ust// memcpy_fallback: |
| 281 | //ust// memcpy(dest, src, len); |
| 282 | //ust// } |
| 283 | //ust// #endif |
| 284 | |
| 285 | static inline int ltt_relay_write(struct rchan_buf *buf, size_t offset, |
| 286 | const void *src, size_t len) |
| 287 | { |
| 288 | //ust// struct buf_page *page; |
| 289 | //ust// ssize_t pagecpy; |
| 290 | //ust// |
| 291 | //ust// offset &= buf->chan->alloc_size - 1; |
| 292 | //ust// page = buf->wpage; |
| 293 | //ust// |
| 294 | //ust// page = ltt_relay_cache_page(buf, &buf->wpage, page, offset); |
| 295 | //ust// pagecpy = min_t(size_t, len, PAGE_SIZE - (offset & ~PAGE_MASK)); |
| 296 | //ust// ltt_relay_do_copy(page_address(page->page) |
| 297 | //ust// + (offset & ~PAGE_MASK), src, pagecpy); |
| 298 | //ust// |
| 299 | //ust// if (unlikely(len != pagecpy)) |
| 300 | //ust// _ltt_relay_write(buf, offset, src, len, page, pagecpy); |
| 301 | //ust// return len; |
| 302 | |
| 303 | size_t cpy; |
| 304 | size_t buf_offset = BUFFER_OFFSET(offset, buf->chan); |
| 305 | |
| 306 | assert(buf_offset < buf->chan->subbuf_size*buf->chan->n_subbufs); |
| 307 | |
| 308 | cpy = min_t(size_t, len, buf->buf_size - buf_offset); |
| 309 | ltt_relay_do_copy(buf->buf_data + buf_offset, src, cpy); |
| 310 | |
| 311 | if (unlikely(len != cpy)) |
| 312 | _ltt_relay_write(buf, buf_offset, src, len, cpy); |
| 313 | return len; |
| 314 | } |
| 315 | |
| 316 | /* |
| 317 | * CONFIG_LTT_RELAY kernel API, ltt/ltt-relay-alloc.c |
| 318 | */ |
| 319 | |
| 320 | struct rchan *ltt_relay_open(const char *base_filename, |
| 321 | struct dentry *parent, |
| 322 | size_t subbuf_size, |
| 323 | size_t n_subbufs, |
| 324 | void *private_data); |
| 325 | extern void ltt_relay_close(struct rchan *chan); |
| 326 | |
| 327 | /* |
| 328 | * exported ltt_relay file operations, ltt/ltt-relay-alloc.c |
| 329 | */ |
| 330 | extern const struct file_operations ltt_relay_file_operations; |
| 331 | |
| 332 | |
| 333 | /* LTTng lockless logging buffer info */ |
| 334 | struct ltt_channel_buf_struct { |
| 335 | /* First 32 bytes cache-hot cacheline */ |
| 336 | local_t offset; /* Current offset in the buffer */ |
| 337 | local_t *commit_count; /* Commit count per sub-buffer */ |
| 338 | atomic_long_t consumed; /* |
| 339 | * Current offset in the buffer |
| 340 | * standard atomic access (shared) |
| 341 | */ |
| 342 | unsigned long last_tsc; /* |
| 343 | * Last timestamp written in the buffer. |
| 344 | */ |
| 345 | /* End of first 32 bytes cacheline */ |
| 346 | //ust// #ifdef CONFIG_LTT_VMCORE |
| 347 | //ust// local_t *commit_seq; /* Consecutive commits */ |
| 348 | //ust// #endif |
| 349 | atomic_long_t active_readers; /* |
| 350 | * Active readers count |
| 351 | * standard atomic access (shared) |
| 352 | */ |
| 353 | local_t events_lost; |
| 354 | local_t corrupted_subbuffers; |
| 355 | //ust// spinlock_t full_lock; /* |
| 356 | //ust// * buffer full condition spinlock, only |
| 357 | //ust// * for userspace tracing blocking mode |
| 358 | //ust// * synchronization with reader. |
| 359 | //ust// */ |
| 360 | //ust// wait_queue_head_t write_wait; /* |
| 361 | //ust// * Wait queue for blocking user space |
| 362 | //ust// * writers |
| 363 | //ust// */ |
| 364 | //ust// atomic_t wakeup_readers; /* Boolean : wakeup readers waiting ? */ |
| 365 | /* one byte is written to this pipe when data is available, in order |
| 366 | to wake the consumer */ |
| 367 | /* portability: Single byte writes must be as quick as possible. The kernel-side |
| 368 | buffer must be large enough so the writer doesn't block. From the pipe(7) |
| 369 | man page: Since linux 2.6.11, the pipe capacity is 65536 bytes. */ |
| 370 | int data_ready_fd_write; |
| 371 | /* the reading end of the pipe */ |
| 372 | int data_ready_fd_read; |
| 373 | |
| 374 | /* commit count per subbuffer; must be at end of struct */ |
| 375 | local_t commit_seq[0] ____cacheline_aligned; |
| 376 | } ____cacheline_aligned; |
| 377 | |
| 378 | int ltt_do_get_subbuf(struct rchan_buf *buf, struct ltt_channel_buf_struct *ltt_buf, long *pconsumed_old); |
| 379 | |
| 380 | int ltt_do_put_subbuf(struct rchan_buf *buf, struct ltt_channel_buf_struct *ltt_buf, u32 uconsumed_old); |
| 381 | |
| 382 | void init_ustrelay_transport(void); |
| 383 | |
| 384 | /*static*/ /* inline */ notrace void ltt_commit_slot( |
| 385 | struct ltt_channel_struct *ltt_channel, |
| 386 | void **transport_data, long buf_offset, |
| 387 | size_t data_size, size_t slot_size); |
| 388 | |
| 389 | #endif /* _LINUX_LTT_RELAY_H */ |