kernel/blk-trace-2.6.14-rc1-git-G1

   1 diff --git a/drivers/block/Kconfig b/drivers/block/Kconfig
   2 --- a/drivers/block/Kconfig
   3 +++ b/drivers/block/Kconfig
   4 @@ -419,6 +419,14 @@ config LBD
   5           your machine, or if you want to have a raid or loopback device
   6           bigger than 2TB.  Otherwise say N.
   7
   8 +config BLK_DEV_IO_TRACE
   9 +       bool "Support for tracing block io actions"
  10 +       select RELAYFS_FS
  11 +       help
  12 +         Say Y here, if you want to be able to trace the block layer actions
  13 +         on a given queue.
  14 +
  15 +
  16  config CDROM_PKTCDVD
  17         tristate "Packet writing on CD/DVD media"
  18         depends on !UML
  19 diff --git a/drivers/block/Makefile b/drivers/block/Makefile
  20 --- a/drivers/block/Makefile
  21 +++ b/drivers/block/Makefile
  22 @@ -45,3 +45,5 @@ obj-$(CONFIG_VIODASD)         += viodasd.o
  23  obj-$(CONFIG_BLK_DEV_SX8)      += sx8.o
  24  obj-$(CONFIG_BLK_DEV_UB)       += ub.o
  25
  26 +obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o
  27 +
  28 diff --git a/drivers/block/elevator.c b/drivers/block/elevator.c
  29 --- a/drivers/block/elevator.c
  30 +++ b/drivers/block/elevator.c
  31 @@ -34,6 +34,7 @@
  32  #include <linux/slab.h>
  33  #include <linux/init.h>
  34  #include <linux/compiler.h>
  35 +#include <linux/blktrace.h>
  36
  37  #include <asm/uaccess.h>
  38
  39 @@ -305,6 +306,8 @@ void elv_requeue_request(request_queue_t
  40  void __elv_add_request(request_queue_t *q, struct request *rq, int where,
  41                        int plug)
  42  {
  43 +       blk_add_trace_rq(q, rq, BLK_TA_QUEUE);
  44 +
  45         /*
  46          * barriers implicitly indicate back insertion
  47          */
  48 @@ -371,6 +374,9 @@ struct request *elv_next_request(request
  49         int ret;
  50
  51         while ((rq = __elv_next_request(q)) != NULL) {
  52 +
  53 +               blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
  54 +
  55                 /*
  56                  * just mark as started even if we don't start it, a request
  57                  * that has been delayed should not be passed by new incoming
  58 diff --git a/drivers/block/ioctl.c b/drivers/block/ioctl.c
  59 --- a/drivers/block/ioctl.c
  60 +++ b/drivers/block/ioctl.c
  61 @@ -4,6 +4,7 @@
  62  #include <linux/backing-dev.h>
  63  #include <linux/buffer_head.h>
  64  #include <linux/smp_lock.h>
  65 +#include <linux/blktrace.h>
  66  #include <asm/uaccess.h>
  67
  68  static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
  69 @@ -188,6 +189,10 @@ static int blkdev_locked_ioctl(struct fi
  70                 return put_ulong(arg, bdev->bd_inode->i_size >> 9);
  71         case BLKGETSIZE64:
  72                 return put_u64(arg, bdev->bd_inode->i_size);
  73 +       case BLKSTARTTRACE:
  74 +               return blk_start_trace(bdev, (char __user *) arg);
  75 +       case BLKSTOPTRACE:
  76 +               return blk_stop_trace(bdev);
  77         }
  78         return -ENOIOCTLCMD;
  79  }
  80 diff --git a/drivers/block/ll_rw_blk.c b/drivers/block/ll_rw_blk.c
  81 --- a/drivers/block/ll_rw_blk.c
  82 +++ b/drivers/block/ll_rw_blk.c
  83 @@ -29,6 +29,7 @@
  84  #include <linux/swap.h>
  85  #include <linux/writeback.h>
  86  #include <linux/blkdev.h>
  87 +#include <linux/blktrace.h>
  88
  89  /*
  90   * for max sense size
  91 @@ -1422,8 +1423,10 @@ void blk_plug_device(request_queue_t *q)
  92         if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
  93                 return;
  94
  95 -       if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
  96 +       if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
  97                 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
  98 +               blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
  99 +       }
 100  }
 101
 102  EXPORT_SYMBOL(blk_plug_device);
 103 @@ -1487,14 +1490,27 @@ static void blk_backing_dev_unplug(struc
 104         /*
 105          * devices don't necessarily have an ->unplug_fn defined
 106          */
 107 -       if (q->unplug_fn)
 108 +       if (q->unplug_fn) {
 109 +               if (test_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
 110 +                       int nrq = q->rq.count[READ] + q->rq.count[WRITE];
 111 +
 112 +                       blk_add_trace_pdu_int(q, BLK_TA_UNPLUG, nrq);
 113 +               }
 114 +
 115                 q->unplug_fn(q);
 116 +       }
 117  }
 118
 119  static void blk_unplug_work(void *data)
 120  {
 121         request_queue_t *q = data;
 122
 123 +       if (test_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
 124 +               int nrq = q->rq.count[READ] + q->rq.count[WRITE];
 125 +
 126 +               blk_add_trace_pdu_int(q, BLK_TA_UNPLUG, nrq);
 127 +       }
 128 +
 129         q->unplug_fn(q);
 130  }
 131
 132 @@ -1624,6 +1640,11 @@ void blk_cleanup_queue(request_queue_t *
 133         if (q->queue_tags)
 134                 __blk_queue_free_tags(q);
 135
 136 +       if (q->blk_trace) {
 137 +               blk_cleanup_trace(q->blk_trace);
 138 +               q->blk_trace = NULL;
 139 +       }
 140 +
 141         blk_queue_ordered(q, QUEUE_ORDERED_NONE);
 142
 143         kmem_cache_free(requestq_cachep, q);
 144 @@ -1970,6 +1991,8 @@ rq_starved:
 145
 146         rq_init(q, rq);
 147         rq->rl = rl;
 148 +
 149 +       blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
 150  out:
 151         return rq;
 152  }
 153 @@ -1998,6 +2021,8 @@ static struct request *get_request_wait(
 154                 if (!rq) {
 155                         struct io_context *ioc;
 156
 157 +                       blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
 158 +
 159                         __generic_unplug_device(q);
 160                         spin_unlock_irq(q->queue_lock);
 161                         io_schedule();
 162 @@ -2051,6 +2076,8 @@ EXPORT_SYMBOL(blk_get_request);
 163   */
 164  void blk_requeue_request(request_queue_t *q, struct request *rq)
 165  {
 166 +       blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 167 +
 168         if (blk_rq_tagged(rq))
 169                 blk_queue_end_tag(q, rq);
 170
 171 @@ -2714,6 +2741,8 @@ static int __make_request(request_queue_
 172                         if (!q->back_merge_fn(q, req, bio))
 173                                 break;
 174
 175 +                       blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 176 +
 177                         req->biotail->bi_next = bio;
 178                         req->biotail = bio;
 179                         req->nr_sectors = req->hard_nr_sectors += nr_sectors;
 180 @@ -2729,6 +2758,8 @@ static int __make_request(request_queue_
 181                         if (!q->front_merge_fn(q, req, bio))
 182                                 break;
 183
 184 +                       blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 185 +
 186                         bio->bi_next = req->bio;
 187                         req->bio = bio;
 188
 189 @@ -3030,6 +3061,10 @@ end_io:
 190                 blk_partition_remap(bio);
 191
 192                 ret = q->make_request_fn(q, bio);
 193 +
 194 +               if (ret)
 195 +                       blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 196 +
 197         } while (ret);
 198  }
 199
 200 @@ -3148,6 +3183,8 @@ static int __end_that_request_first(stru
 201         int total_bytes, bio_nbytes, error, next_idx = 0;
 202         struct bio *bio;
 203
 204 +       blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
 205 +
 206         /*
 207          * extend uptodate bool to allow < 0 value to be direct io error
 208          */
 209 diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
 210 --- a/include/linux/blkdev.h
 211 +++ b/include/linux/blkdev.h
 212 @@ -22,6 +22,7 @@ typedef struct request_queue request_que
 213  struct elevator_queue;
 214  typedef struct elevator_queue elevator_t;
 215  struct request_pm_state;
 216 +struct blk_trace;
 217
 218  #define BLKDEV_MIN_RQ  4
 219  #define BLKDEV_MAX_RQ  128     /* Default maximum */
 220 @@ -412,6 +413,8 @@ struct request_queue
 221          */
 222         struct request          *flush_rq;
 223         unsigned char           ordered;
 224 +
 225 +       struct blk_trace        *blk_trace;
 226  };
 227
 228  enum {
 229 diff --git a/include/linux/fs.h b/include/linux/fs.h
 230 --- a/include/linux/fs.h
 231 +++ b/include/linux/fs.h
 232 @@ -196,6 +196,8 @@ extern int dir_notify_enable;
 233  #define BLKBSZGET  _IOR(0x12,112,size_t)
 234  #define BLKBSZSET  _IOW(0x12,113,size_t)
 235  #define BLKGETSIZE64 _IOR(0x12,114,size_t)     /* return device size in bytes (u64 *arg) */
 236 +#define BLKSTARTTRACE _IOWR(0x12,115,struct blk_user_trace_setup)
 237 +#define BLKSTOPTRACE _IO(0x12,116)
 238
 239  #define BMAP_IOCTL 1           /* obsolete - kept for compatibility */
 240  #define FIBMAP    _IO(0x00,1)  /* bmap access */
 241 --- /dev/null   2005-09-09 21:24:12.000000000 +0200
 242 +++ linux-2.6/drivers/block/blktrace.c  2005-09-15 14:13:03.000000000 +0200
 243 @@ -0,0 +1,224 @@
 244 +#include <linux/config.h>
 245 +#include <linux/kernel.h>
 246 +#include <linux/blkdev.h>
 247 +#include <linux/blktrace.h>
 248 +#include <linux/percpu.h>
 249 +#include <linux/init.h>
 250 +#include <asm/uaccess.h>
 251 +
 252 +static DEFINE_PER_CPU(unsigned long long, blk_trace_cpu_offset) = { 0, };
 253 +
 254 +void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 255 +                    int rw, u32 what, int error, int pdu_len, void *pdu_data)
 256 +{
 257 +       struct blk_io_trace t;
 258 +       unsigned long flags;
 259 +       int cpu;
 260 +
 261 +       if (rw & (1 << BIO_RW_BARRIER))
 262 +               what |= BLK_TC_ACT(BLK_TC_BARRIER);
 263 +       if (rw & (1 << BIO_RW_SYNC))
 264 +               what |= BLK_TC_ACT(BLK_TC_SYNC);
 265 +
 266 +       if (rw & WRITE)
 267 +               what |= BLK_TC_ACT(BLK_TC_WRITE);
 268 +       else
 269 +               what |= BLK_TC_ACT(BLK_TC_READ);
 270 +
 271 +       if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 272 +               return;
 273 +
 274 +       t.magic         = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 275 +       t.sequence      = atomic_add_return(1, &bt->sequence);
 276 +
 277 +       cpu = get_cpu();
 278 +       t.cpu           = cpu;
 279 +       t.time          = sched_clock() - per_cpu(blk_trace_cpu_offset, cpu);
 280 +       put_cpu();
 281 +
 282 +       t.device        = bt->dev;
 283 +       t.sector        = sector;
 284 +       t.bytes         = bytes;
 285 +       t.action        = what;
 286 +       t.error         = error;
 287 +       t.pdu_len       = pdu_len;
 288 +
 289 +       t.pid           = current->pid;
 290 +       memcpy(t.comm, current->comm, sizeof(t.comm));
 291 +
 292 +       local_irq_save(flags);
 293 +       __relay_write(bt->rchan, &t, sizeof(t));
 294 +       if (pdu_len)
 295 +               __relay_write(bt->rchan, pdu_data, pdu_len);
 296 +       local_irq_restore(flags);
 297 +}
 298 +
 299 +static struct dentry *blk_tree_root;
 300 +static DECLARE_MUTEX(blk_tree_mutex);
 301 +
 302 +static inline void blk_remove_root(void)
 303 +{
 304 +       if (relayfs_remove_dir(blk_tree_root) != -ENOTEMPTY)
 305 +               blk_tree_root = NULL;
 306 +}
 307 +
 308 +static void blk_remove_tree(struct dentry *dir)
 309 +{
 310 +       down(&blk_tree_mutex);
 311 +       relayfs_remove_dir(dir);
 312 +       blk_remove_root();
 313 +       up(&blk_tree_mutex);
 314 +}
 315 +
 316 +static struct dentry *blk_create_tree(const char *blk_name)
 317 +{
 318 +       struct dentry *dir = NULL;
 319 +
 320 +       down(&blk_tree_mutex);
 321 +
 322 +       if (!blk_tree_root) {
 323 +               blk_tree_root = relayfs_create_dir("block", NULL);
 324 +               if (!blk_tree_root)
 325 +                       goto err;
 326 +       }
 327 +
 328 +       dir = relayfs_create_dir(blk_name, blk_tree_root);
 329 +       if (!dir)
 330 +               blk_remove_root();
 331 +
 332 +err:
 333 +       up(&blk_tree_mutex);
 334 +       return dir;
 335 +}
 336 +
 337 +void blk_cleanup_trace(struct blk_trace *bt)
 338 +{
 339 +       relay_close(bt->rchan);
 340 +       blk_remove_tree(bt->dir);
 341 +       kfree(bt);
 342 +}
 343 +
 344 +int blk_stop_trace(struct block_device *bdev)
 345 +{
 346 +       request_queue_t *q = bdev_get_queue(bdev);
 347 +       struct blk_trace *bt = NULL;
 348 +       int ret = -EINVAL;
 349 +
 350 +       if (!q)
 351 +               return -ENXIO;
 352 +
 353 +       down(&bdev->bd_sem);
 354 +
 355 +       if (q->blk_trace) {
 356 +               bt = q->blk_trace;
 357 +               q->blk_trace = NULL;
 358 +               ret = 0;
 359 +       }
 360 +
 361 +       up(&bdev->bd_sem);
 362 +
 363 +       if (bt)
 364 +               blk_cleanup_trace(bt);
 365 +
 366 +       return ret;
 367 +}
 368 +
 369 +int blk_start_trace(struct block_device *bdev, char __user *arg)
 370 +{
 371 +       request_queue_t *q = bdev_get_queue(bdev);
 372 +       struct blk_user_trace_setup buts;
 373 +       struct blk_trace *bt = NULL;
 374 +       struct dentry *dir = NULL;
 375 +       char b[BDEVNAME_SIZE];
 376 +       int ret;
 377 +
 378 +       if (!q)
 379 +               return -ENXIO;
 380 +
 381 +       if (copy_from_user(&buts, arg, sizeof(buts)))
 382 +               return -EFAULT;
 383 +
 384 +       if (!buts.buf_size || !buts.buf_nr)
 385 +               return -EINVAL;
 386 +
 387 +       strcpy(buts.name, bdevname(bdev, b));
 388 +
 389 +       if (copy_to_user(arg, &buts, sizeof(buts)))
 390 +               return -EFAULT;
 391 +
 392 +       down(&bdev->bd_sem);
 393 +       ret = -EBUSY;
 394 +       if (q->blk_trace)
 395 +               goto err;
 396 +
 397 +       ret = -ENOMEM;
 398 +       bt = kmalloc(sizeof(*bt), GFP_KERNEL);
 399 +       if (!bt)
 400 +               goto err;
 401 +
 402 +       ret = -ENOENT;
 403 +       dir = blk_create_tree(bdevname(bdev, b));
 404 +       if (!dir)
 405 +               goto err;
 406 +
 407 +       bt->dir = dir;
 408 +       bt->dev = bdev->bd_dev;
 409 +       atomic_set(&bt->sequence, 0);
 410 +
 411 +       ret = -EIO;
 412 +       bt->rchan = relay_open("trace", dir, buts.buf_size, buts.buf_nr, NULL);
 413 +       if (!bt->rchan)
 414 +               goto err;
 415 +
 416 +       bt->act_mask = buts.act_mask;
 417 +       if (!bt->act_mask)
 418 +               bt->act_mask = (u16) -1;
 419 +
 420 +       q->blk_trace = bt;
 421 +       up(&bdev->bd_sem);
 422 +       return 0;
 423 +err:
 424 +       up(&bdev->bd_sem);
 425 +       if (dir)
 426 +               blk_remove_tree(dir);
 427 +       if (bt)
 428 +               kfree(bt);
 429 +       return ret;
 430 +}
 431 +
 432 +static void blk_trace_check_cpu_time(void *data)
 433 +{
 434 +       unsigned long long a, b, *t;
 435 +       struct timeval tv;
 436 +       int cpu = get_cpu();
 437 +
 438 +       t = &per_cpu(blk_trace_cpu_offset, cpu);
 439 +
 440 +       a = sched_clock();
 441 +       do_gettimeofday(&tv);
 442 +       b = sched_clock();
 443 +
 444 +       *t = tv.tv_sec * 1000000000 + tv.tv_usec * 1000;
 445 +       *t -= (a + b) / 2;
 446 +       put_cpu();
 447 +}
 448 +
 449 +static int blk_trace_calibrate_offsets(void)
 450 +{
 451 +       unsigned long flags;
 452 +
 453 +       smp_call_function(blk_trace_check_cpu_time, NULL, 1, 1);
 454 +       local_irq_save(flags);
 455 +       blk_trace_check_cpu_time(NULL);
 456 +       local_irq_restore(flags);
 457 +
 458 +       return 0;
 459 +}
 460 +
 461 +static __init int blk_trace_init(void)
 462 +{
 463 +       return blk_trace_calibrate_offsets();
 464 +}
 465 +
 466 +module_init(blk_trace_init);
 467 +
 468 --- /dev/null   2005-09-09 21:24:12.000000000 +0200
 469 +++ linux-2.6/include/linux/blktrace.h  2005-09-15 14:13:12.000000000 +0200
 470 @@ -0,0 +1,169 @@
 471 +#ifndef BLKTRACE_H
 472 +#define BLKTRACE_H
 473 +
 474 +#include <linux/config.h>
 475 +#include <linux/blkdev.h>
 476 +#include <linux/relayfs_fs.h>
 477 +
 478 +/*
 479 + * Trace categories
 480 + */
 481 +enum {
 482 +       BLK_TC_READ     = 1 << 0,       /* reads */
 483 +       BLK_TC_WRITE    = 1 << 1,       /* writes */
 484 +       BLK_TC_BARRIER  = 1 << 2,       /* barrier */
 485 +       BLK_TC_SYNC     = 1 << 3,       /* barrier */
 486 +       BLK_TC_QUEUE    = 1 << 4,       /* queueing/merging */
 487 +       BLK_TC_REQUEUE  = 1 << 5,       /* requeueing */
 488 +       BLK_TC_ISSUE    = 1 << 6,       /* issue */
 489 +       BLK_TC_COMPLETE = 1 << 7,       /* completions */
 490 +       BLK_TC_FS       = 1 << 8,       /* fs requests */
 491 +       BLK_TC_PC       = 1 << 9,       /* pc requests */
 492 +
 493 +       BLK_TC_END      = 1 << 15,      /* only 16-bits, reminder */
 494 +};
 495 +
 496 +#define BLK_TC_SHIFT           (16)
 497 +#define BLK_TC_ACT(act)                ((act) << BLK_TC_SHIFT)
 498 +
 499 +/*
 500 + * Basic trace actions
 501 + */
 502 +enum {
 503 +       __BLK_TA_QUEUE = 1,             /* queued */
 504 +       __BLK_TA_BACKMERGE,             /* back merged to existing rq */
 505 +       __BLK_TA_FRONTMERGE,            /* front merge to existing rq */
 506 +       __BLK_TA_GETRQ,                 /* allocated new request */
 507 +       __BLK_TA_SLEEPRQ,               /* sleeping on rq allocation */
 508 +       __BLK_TA_REQUEUE,               /* request requeued */
 509 +       __BLK_TA_ISSUE,                 /* sent to driver */
 510 +       __BLK_TA_COMPLETE,              /* completed by driver */
 511 +       __BLK_TA_PLUG,                  /* queue was plugged */
 512 +       __BLK_TA_UNPLUG,                /* queue was unplugged */
 513 +};
 514 +
 515 +/*
 516 + * Trace actions in full. Additionally, read or write is masked
 517 + */
 518 +#define BLK_TA_QUEUE           (__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
 519 +#define BLK_TA_BACKMERGE       (__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
 520 +#define BLK_TA_FRONTMERGE      (__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
 521 +#define        BLK_TA_GETRQ            (__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
 522 +#define        BLK_TA_SLEEPRQ          (__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
 523 +#define        BLK_TA_REQUEUE          (__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
 524 +#define BLK_TA_ISSUE           (__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
 525 +#define BLK_TA_COMPLETE                (__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
 526 +#define BLK_TA_PLUG            (__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
 527 +#define BLK_TA_UNPLUG          (__BLK_TA_UNPLUG | BLK_TC_ACT(BLK_TC_QUEUE))
 528 +
 529 +#define BLK_IO_TRACE_MAGIC     0x65617400
 530 +#define BLK_IO_TRACE_VERSION   0x05
 531 +
 532 +/*
 533 + * The trace itself
 534 + */
 535 +struct blk_io_trace {
 536 +       u32 magic;              /* MAGIC << 8 | version */
 537 +       u32 sequence;           /* event number */
 538 +       u64 time;               /* in microseconds */
 539 +       u64 sector;             /* disk offset */
 540 +       u32 bytes;              /* transfer length */
 541 +       u32 action;             /* what happened */
 542 +       u32 pid;                /* who did it */
 543 +       u32 cpu;                /* on what cpu did it happen */
 544 +       u16 error;              /* completion error */
 545 +       u16 pdu_len;            /* length of data after this trace */
 546 +       u32 device;             /* device number */
 547 +       char comm[16];          /* task command name (TASK_COMM_LEN) */
 548 +};
 549 +
 550 +struct blk_trace {
 551 +       struct dentry *dir;
 552 +       struct rchan *rchan;
 553 +       atomic_t sequence;
 554 +       u32 dev;
 555 +       u16 act_mask;
 556 +};
 557 +
 558 +/*
 559 + * User setup structure passed with BLKSTARTTRACE
 560 + */
 561 +struct blk_user_trace_setup {
 562 +       char name[BDEVNAME_SIZE];       /* output */
 563 +       u16 act_mask;                   /* input */
 564 +       u32 buf_size;                   /* input */
 565 +       u32 buf_nr;                     /* input */
 566 +};
 567 +
 568 +#if defined(CONFIG_BLK_DEV_IO_TRACE)
 569 +extern int blk_start_trace(struct block_device *, char __user *);
 570 +extern int blk_stop_trace(struct block_device *);
 571 +extern void blk_cleanup_trace(struct blk_trace *);
 572 +extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
 573 +
 574 +static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 575 +                                   u32 what)
 576 +{
 577 +       struct blk_trace *bt = q->blk_trace;
 578 +       int rw = rq->flags & 0x07;
 579 +
 580 +       if (likely(!bt))
 581 +               return;
 582 +
 583 +       if (blk_pc_request(rq)) {
 584 +               what |= BLK_TC_ACT(BLK_TC_PC);
 585 +               __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
 586 +       } else  {
 587 +               what |= BLK_TC_ACT(BLK_TC_FS);
 588 +               __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
 589 +       }
 590 +}
 591 +
 592 +static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 593 +                                    u32 what)
 594 +{
 595 +       struct blk_trace *bt = q->blk_trace;
 596 +
 597 +       if (likely(!bt))
 598 +               return;
 599 +
 600 +       __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 601 +}
 602 +
 603 +static inline void blk_add_trace_generic(struct request_queue *q,
 604 +                                        struct bio *bio, int rw, u32 what)
 605 +{
 606 +       struct blk_trace *bt = q->blk_trace;
 607 +
 608 +       if (likely(!bt))
 609 +               return;
 610 +
 611 +       if (bio)
 612 +               blk_add_trace_bio(q, bio, what);
 613 +       else
 614 +               __blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
 615 +}
 616 +
 617 +static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
 618 +                                        unsigned int pdu)
 619 +{
 620 +       struct blk_trace *bt = q->blk_trace;
 621 +       u64 rpdu = pdu;
 622 +
 623 +       if (likely(!bt))
 624 +               return;
 625 +
 626 +       __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 627 +}
 628 +
 629 +#else /* !CONFIG_BLK_DEV_IO_TRACE */
 630 +#define blk_start_trace(bdev, arg)             (-EINVAL)
 631 +#define blk_stop_trace(bdev)                   (-EINVAL)
 632 +#define blk_cleanup_trace(bt)                  do { } while (0)
 633 +#define blk_add_trace_rq(q, rq, what)          do { } while (0)
 634 +#define blk_add_trace_bio(q, rq, what)         do { } while (0)
 635 +#define blk_add_trace_generic(q, rq, rw, what) do { } while (0)
 636 +#define blk_add_trace_pdu_int(q, what, pdul, pdu)      do { } while (0)
 637 +#endif /* CONFIG_BLK_DEV_IO_TRACE */
 638 +
 639 +#endif