Merge branch 'linus' into tracing/blktrace
authorIngo Molnar <mingo@elte.hu>
Thu, 19 Feb 2009 08:00:35 +0000 (09:00 +0100)
committerIngo Molnar <mingo@elte.hu>
Thu, 19 Feb 2009 08:00:35 +0000 (09:00 +0100)
Conflicts:
block/blktrace.c

Semantic merge:
kernel/trace/blktrace.c

Signed-off-by: Ingo Molnar <mingo@elte.hu>
1  2 
MAINTAINERS
include/linux/blktrace_api.h
kernel/sched.c
kernel/trace/blktrace.c

diff --cc MAINTAINERS
Simple merge
Simple merge
diff --cc kernel/sched.c
Simple merge
index 3b91da064820d9dd16b540f42342ae21eb0bb994,0000000000000000000000000000000000000000..e82cb9e930cc2e328830afe163e89af032c77f41
mode 100644,000000..100644
--- /dev/null
@@@ -1,1538 -1,0 +1,1538 @@@
-       what |= MASK_TC_BIT(rw, SYNC);
 +/*
 + * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
 + *
 + * This program is free software; you can redistribute it and/or modify
 + * it under the terms of the GNU General Public License version 2 as
 + * published by the Free Software Foundation.
 + *
 + * This program is distributed in the hope that it will be useful,
 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 + * GNU General Public License for more details.
 + *
 + * You should have received a copy of the GNU General Public License
 + * along with this program; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 + *
 + */
 +#include <linux/kernel.h>
 +#include <linux/blkdev.h>
 +#include <linux/blktrace_api.h>
 +#include <linux/percpu.h>
 +#include <linux/init.h>
 +#include <linux/mutex.h>
 +#include <linux/debugfs.h>
 +#include <linux/time.h>
 +#include <trace/block.h>
 +#include <linux/uaccess.h>
 +#include "trace_output.h"
 +
 +static unsigned int blktrace_seq __read_mostly = 1;
 +
 +static struct trace_array *blk_tr;
 +static int __read_mostly  blk_tracer_enabled;
 +
 +/* Select an alternative, minimalistic output than the original one */
 +#define TRACE_BLK_OPT_CLASSIC         0x1
 +
 +static struct tracer_opt blk_tracer_opts[] = {
 +      /* Default disable the minimalistic output */
 +      { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
 +      { }
 +};
 +
 +static struct tracer_flags blk_tracer_flags = {
 +      .val  = 0,
 +      .opts = blk_tracer_opts,
 +};
 +
 +/* Global reference count of probes */
 +static DEFINE_MUTEX(blk_probe_mutex);
 +static atomic_t blk_probes_ref = ATOMIC_INIT(0);
 +
 +static int blk_register_tracepoints(void);
 +static void blk_unregister_tracepoints(void);
 +
 +/*
 + * Send out a notify message.
 + */
 +static void trace_note(struct blk_trace *bt, pid_t pid, int action,
 +                     const void *data, size_t len)
 +{
 +      struct blk_io_trace *t;
 +
 +      if (!bt->rchan)
 +              return;
 +
 +      t = relay_reserve(bt->rchan, sizeof(*t) + len);
 +      if (t) {
 +              const int cpu = smp_processor_id();
 +
 +              t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 +              t->time = ktime_to_ns(ktime_get());
 +              t->device = bt->dev;
 +              t->action = action;
 +              t->pid = pid;
 +              t->cpu = cpu;
 +              t->pdu_len = len;
 +              memcpy((void *) t + sizeof(*t), data, len);
 +      }
 +}
 +
 +/*
 + * Send out a notify for this process, if we haven't done so since a trace
 + * started
 + */
 +static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 +{
 +      tsk->btrace_seq = blktrace_seq;
 +      trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 +}
 +
 +static void trace_note_time(struct blk_trace *bt)
 +{
 +      struct timespec now;
 +      unsigned long flags;
 +      u32 words[2];
 +
 +      getnstimeofday(&now);
 +      words[0] = now.tv_sec;
 +      words[1] = now.tv_nsec;
 +
 +      local_irq_save(flags);
 +      trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 +      local_irq_restore(flags);
 +}
 +
 +void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 +{
 +      int n;
 +      va_list args;
 +      unsigned long flags;
 +      char *buf;
 +
 +      if (blk_tr) {
 +              va_start(args, fmt);
 +              ftrace_vprintk(fmt, args);
 +              va_end(args);
 +              return;
 +      }
 +
 +      if (!bt->msg_data)
 +              return;
 +
 +      local_irq_save(flags);
 +      buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 +      va_start(args, fmt);
 +      n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 +      va_end(args);
 +
 +      trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 +      local_irq_restore(flags);
 +}
 +EXPORT_SYMBOL_GPL(__trace_note_message);
 +
 +static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 +                       pid_t pid)
 +{
 +      if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 +              return 1;
 +      if (sector < bt->start_lba || sector > bt->end_lba)
 +              return 1;
 +      if (bt->pid && pid != bt->pid)
 +              return 1;
 +
 +      return 0;
 +}
 +
 +/*
 + * Data direction bit lookup
 + */
 +static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ),
 +                                       BLK_TC_ACT(BLK_TC_WRITE) };
 +
 +/* The ilog2() calls fall out because they're constant */
 +#define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
 +        (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
 +
 +/*
 + * The worker for the various blk_add_trace*() types. Fills out a
 + * blk_io_trace structure and places it in a per-cpu subbuffer.
 + */
 +static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 +                   int rw, u32 what, int error, int pdu_len, void *pdu_data)
 +{
 +      struct task_struct *tsk = current;
 +      struct ring_buffer_event *event = NULL;
 +      struct blk_io_trace *t;
 +      unsigned long flags = 0;
 +      unsigned long *sequence;
 +      pid_t pid;
 +      int cpu, pc = 0;
 +
 +      if (unlikely(bt->trace_state != Blktrace_running ||
 +                   !blk_tracer_enabled))
 +              return;
 +
 +      what |= ddir_act[rw & WRITE];
 +      what |= MASK_TC_BIT(rw, BARRIER);
++      what |= MASK_TC_BIT(rw, SYNCIO);
 +      what |= MASK_TC_BIT(rw, AHEAD);
 +      what |= MASK_TC_BIT(rw, META);
 +      what |= MASK_TC_BIT(rw, DISCARD);
 +
 +      pid = tsk->pid;
 +      if (unlikely(act_log_check(bt, what, sector, pid)))
 +              return;
 +      cpu = raw_smp_processor_id();
 +
 +      if (blk_tr) {
 +              tracing_record_cmdline(current);
 +
 +              pc = preempt_count();
 +              event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
 +                                                sizeof(*t) + pdu_len,
 +                                                0, pc);
 +              if (!event)
 +                      return;
 +              t = ring_buffer_event_data(event);
 +              goto record_it;
 +      }
 +
 +      /*
 +       * A word about the locking here - we disable interrupts to reserve
 +       * some space in the relay per-cpu buffer, to prevent an irq
 +       * from coming in and stepping on our toes.
 +       */
 +      local_irq_save(flags);
 +
 +      if (unlikely(tsk->btrace_seq != blktrace_seq))
 +              trace_note_tsk(bt, tsk);
 +
 +      t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 +      if (t) {
 +              sequence = per_cpu_ptr(bt->sequence, cpu);
 +
 +              t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 +              t->sequence = ++(*sequence);
 +              t->time = ktime_to_ns(ktime_get());
 +record_it:
 +              /*
 +               * These two are not needed in ftrace as they are in the
 +               * generic trace_entry, filled by tracing_generic_entry_update,
 +               * but for the trace_event->bin() synthesizer benefit we do it
 +               * here too.
 +               */
 +              t->cpu = cpu;
 +              t->pid = pid;
 +
 +              t->sector = sector;
 +              t->bytes = bytes;
 +              t->action = what;
 +              t->device = bt->dev;
 +              t->error = error;
 +              t->pdu_len = pdu_len;
 +
 +              if (pdu_len)
 +                      memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 +
 +              if (blk_tr) {
 +                      trace_buffer_unlock_commit(blk_tr, event, 0, pc);
 +                      return;
 +              }
 +      }
 +
 +      local_irq_restore(flags);
 +}
 +
 +static struct dentry *blk_tree_root;
 +static DEFINE_MUTEX(blk_tree_mutex);
 +
 +static void blk_trace_cleanup(struct blk_trace *bt)
 +{
 +      debugfs_remove(bt->msg_file);
 +      debugfs_remove(bt->dropped_file);
 +      relay_close(bt->rchan);
 +      free_percpu(bt->sequence);
 +      free_percpu(bt->msg_data);
 +      kfree(bt);
 +      mutex_lock(&blk_probe_mutex);
 +      if (atomic_dec_and_test(&blk_probes_ref))
 +              blk_unregister_tracepoints();
 +      mutex_unlock(&blk_probe_mutex);
 +}
 +
 +int blk_trace_remove(struct request_queue *q)
 +{
 +      struct blk_trace *bt;
 +
 +      bt = xchg(&q->blk_trace, NULL);
 +      if (!bt)
 +              return -EINVAL;
 +
 +      if (bt->trace_state == Blktrace_setup ||
 +          bt->trace_state == Blktrace_stopped)
 +              blk_trace_cleanup(bt);
 +
 +      return 0;
 +}
 +EXPORT_SYMBOL_GPL(blk_trace_remove);
 +
 +static int blk_dropped_open(struct inode *inode, struct file *filp)
 +{
 +      filp->private_data = inode->i_private;
 +
 +      return 0;
 +}
 +
 +static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 +                              size_t count, loff_t *ppos)
 +{
 +      struct blk_trace *bt = filp->private_data;
 +      char buf[16];
 +
 +      snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 +
 +      return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 +}
 +
 +static const struct file_operations blk_dropped_fops = {
 +      .owner =        THIS_MODULE,
 +      .open =         blk_dropped_open,
 +      .read =         blk_dropped_read,
 +};
 +
 +static int blk_msg_open(struct inode *inode, struct file *filp)
 +{
 +      filp->private_data = inode->i_private;
 +
 +      return 0;
 +}
 +
 +static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 +                              size_t count, loff_t *ppos)
 +{
 +      char *msg;
 +      struct blk_trace *bt;
 +
 +      if (count > BLK_TN_MAX_MSG)
 +              return -EINVAL;
 +
 +      msg = kmalloc(count, GFP_KERNEL);
 +      if (msg == NULL)
 +              return -ENOMEM;
 +
 +      if (copy_from_user(msg, buffer, count)) {
 +              kfree(msg);
 +              return -EFAULT;
 +      }
 +
 +      bt = filp->private_data;
 +      __trace_note_message(bt, "%s", msg);
 +      kfree(msg);
 +
 +      return count;
 +}
 +
 +static const struct file_operations blk_msg_fops = {
 +      .owner =        THIS_MODULE,
 +      .open =         blk_msg_open,
 +      .write =        blk_msg_write,
 +};
 +
 +/*
 + * Keep track of how many times we encountered a full subbuffer, to aid
 + * the user space app in telling how many lost events there were.
 + */
 +static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 +                                   void *prev_subbuf, size_t prev_padding)
 +{
 +      struct blk_trace *bt;
 +
 +      if (!relay_buf_full(buf))
 +              return 1;
 +
 +      bt = buf->chan->private_data;
 +      atomic_inc(&bt->dropped);
 +      return 0;
 +}
 +
 +static int blk_remove_buf_file_callback(struct dentry *dentry)
 +{
 +      struct dentry *parent = dentry->d_parent;
 +      debugfs_remove(dentry);
 +
 +      /*
 +      * this will fail for all but the last file, but that is ok. what we
 +      * care about is the top level buts->name directory going away, when
 +      * the last trace file is gone. Then we don't have to rmdir() that
 +      * manually on trace stop, so it nicely solves the issue with
 +      * force killing of running traces.
 +      */
 +
 +      debugfs_remove(parent);
 +      return 0;
 +}
 +
 +static struct dentry *blk_create_buf_file_callback(const char *filename,
 +                                                 struct dentry *parent,
 +                                                 int mode,
 +                                                 struct rchan_buf *buf,
 +                                                 int *is_global)
 +{
 +      return debugfs_create_file(filename, mode, parent, buf,
 +                                      &relay_file_operations);
 +}
 +
 +static struct rchan_callbacks blk_relay_callbacks = {
 +      .subbuf_start           = blk_subbuf_start_callback,
 +      .create_buf_file        = blk_create_buf_file_callback,
 +      .remove_buf_file        = blk_remove_buf_file_callback,
 +};
 +
 +/*
 + * Setup everything required to start tracing
 + */
 +int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 +                      struct blk_user_trace_setup *buts)
 +{
 +      struct blk_trace *old_bt, *bt = NULL;
 +      struct dentry *dir = NULL;
 +      int ret, i;
 +
 +      if (!buts->buf_size || !buts->buf_nr)
 +              return -EINVAL;
 +
 +      strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 +      buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 +
 +      /*
 +       * some device names have larger paths - convert the slashes
 +       * to underscores for this to work as expected
 +       */
 +      for (i = 0; i < strlen(buts->name); i++)
 +              if (buts->name[i] == '/')
 +                      buts->name[i] = '_';
 +
 +      ret = -ENOMEM;
 +      bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 +      if (!bt)
 +              goto err;
 +
 +      bt->sequence = alloc_percpu(unsigned long);
 +      if (!bt->sequence)
 +              goto err;
 +
 +      bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG);
 +      if (!bt->msg_data)
 +              goto err;
 +
 +      ret = -ENOENT;
 +
 +      if (!blk_tree_root) {
 +              blk_tree_root = debugfs_create_dir("block", NULL);
 +              if (!blk_tree_root)
 +                      return -ENOMEM;
 +      }
 +
 +      dir = debugfs_create_dir(buts->name, blk_tree_root);
 +
 +      if (!dir)
 +              goto err;
 +
 +      bt->dir = dir;
 +      bt->dev = dev;
 +      atomic_set(&bt->dropped, 0);
 +
 +      ret = -EIO;
 +      bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 +                                             &blk_dropped_fops);
 +      if (!bt->dropped_file)
 +              goto err;
 +
 +      bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 +      if (!bt->msg_file)
 +              goto err;
 +
 +      bt->rchan = relay_open("trace", dir, buts->buf_size,
 +                              buts->buf_nr, &blk_relay_callbacks, bt);
 +      if (!bt->rchan)
 +              goto err;
 +
 +      bt->act_mask = buts->act_mask;
 +      if (!bt->act_mask)
 +              bt->act_mask = (u16) -1;
 +
 +      bt->start_lba = buts->start_lba;
 +      bt->end_lba = buts->end_lba;
 +      if (!bt->end_lba)
 +              bt->end_lba = -1ULL;
 +
 +      bt->pid = buts->pid;
 +      bt->trace_state = Blktrace_setup;
 +
 +      mutex_lock(&blk_probe_mutex);
 +      if (atomic_add_return(1, &blk_probes_ref) == 1) {
 +              ret = blk_register_tracepoints();
 +              if (ret)
 +                      goto probe_err;
 +      }
 +      mutex_unlock(&blk_probe_mutex);
 +
 +      ret = -EBUSY;
 +      old_bt = xchg(&q->blk_trace, bt);
 +      if (old_bt) {
 +              (void) xchg(&q->blk_trace, old_bt);
 +              goto err;
 +      }
 +
 +      return 0;
 +probe_err:
 +      atomic_dec(&blk_probes_ref);
 +      mutex_unlock(&blk_probe_mutex);
 +err:
 +      if (bt) {
 +              if (bt->msg_file)
 +                      debugfs_remove(bt->msg_file);
 +              if (bt->dropped_file)
 +                      debugfs_remove(bt->dropped_file);
 +              free_percpu(bt->sequence);
 +              free_percpu(bt->msg_data);
 +              if (bt->rchan)
 +                      relay_close(bt->rchan);
 +              kfree(bt);
 +      }
 +      return ret;
 +}
 +
 +int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 +                  char __user *arg)
 +{
 +      struct blk_user_trace_setup buts;
 +      int ret;
 +
 +      ret = copy_from_user(&buts, arg, sizeof(buts));
 +      if (ret)
 +              return -EFAULT;
 +
 +      ret = do_blk_trace_setup(q, name, dev, &buts);
 +      if (ret)
 +              return ret;
 +
 +      if (copy_to_user(arg, &buts, sizeof(buts)))
 +              return -EFAULT;
 +
 +      return 0;
 +}
 +EXPORT_SYMBOL_GPL(blk_trace_setup);
 +
 +int blk_trace_startstop(struct request_queue *q, int start)
 +{
 +      int ret;
 +      struct blk_trace *bt = q->blk_trace;
 +
 +      if (bt == NULL)
 +              return -EINVAL;
 +
 +      /*
 +       * For starting a trace, we can transition from a setup or stopped
 +       * trace. For stopping a trace, the state must be running
 +       */
 +      ret = -EINVAL;
 +      if (start) {
 +              if (bt->trace_state == Blktrace_setup ||
 +                  bt->trace_state == Blktrace_stopped) {
 +                      blktrace_seq++;
 +                      smp_mb();
 +                      bt->trace_state = Blktrace_running;
 +
 +                      trace_note_time(bt);
 +                      ret = 0;
 +              }
 +      } else {
 +              if (bt->trace_state == Blktrace_running) {
 +                      bt->trace_state = Blktrace_stopped;
 +                      relay_flush(bt->rchan);
 +                      ret = 0;
 +              }
 +      }
 +
 +      return ret;
 +}
 +EXPORT_SYMBOL_GPL(blk_trace_startstop);
 +
 +/**
 + * blk_trace_ioctl: - handle the ioctls associated with tracing
 + * @bdev:     the block device
 + * @cmd:      the ioctl cmd
 + * @arg:      the argument data, if any
 + *
 + **/
 +int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 +{
 +      struct request_queue *q;
 +      int ret, start = 0;
 +      char b[BDEVNAME_SIZE];
 +
 +      q = bdev_get_queue(bdev);
 +      if (!q)
 +              return -ENXIO;
 +
 +      mutex_lock(&bdev->bd_mutex);
 +
 +      switch (cmd) {
 +      case BLKTRACESETUP:
 +              bdevname(bdev, b);
 +              ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 +              break;
 +      case BLKTRACESTART:
 +              start = 1;
 +      case BLKTRACESTOP:
 +              ret = blk_trace_startstop(q, start);
 +              break;
 +      case BLKTRACETEARDOWN:
 +              ret = blk_trace_remove(q);
 +              break;
 +      default:
 +              ret = -ENOTTY;
 +              break;
 +      }
 +
 +      mutex_unlock(&bdev->bd_mutex);
 +      return ret;
 +}
 +
 +/**
 + * blk_trace_shutdown: - stop and cleanup trace structures
 + * @q:    the request queue associated with the device
 + *
 + **/
 +void blk_trace_shutdown(struct request_queue *q)
 +{
 +      if (q->blk_trace) {
 +              blk_trace_startstop(q, 0);
 +              blk_trace_remove(q);
 +      }
 +}
 +
 +/*
 + * blktrace probes
 + */
 +
 +/**
 + * blk_add_trace_rq - Add a trace for a request oriented action
 + * @q:                queue the io is for
 + * @rq:               the source request
 + * @what:     the action
 + *
 + * Description:
 + *     Records an action against a request. Will log the bio offset + size.
 + *
 + **/
 +static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 +                                  u32 what)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +      int rw = rq->cmd_flags & 0x03;
 +
 +      if (likely(!bt))
 +              return;
 +
 +      if (blk_discard_rq(rq))
 +              rw |= (1 << BIO_RW_DISCARD);
 +
 +      if (blk_pc_request(rq)) {
 +              what |= BLK_TC_ACT(BLK_TC_PC);
 +              __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 +                              sizeof(rq->cmd), rq->cmd);
 +      } else  {
 +              what |= BLK_TC_ACT(BLK_TC_FS);
 +              __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 +                              rw, what, rq->errors, 0, NULL);
 +      }
 +}
 +
 +static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 +{
 +      blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 +}
 +
 +static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 +{
 +      blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 +}
 +
 +static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 +{
 +      blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 +}
 +
 +static void blk_add_trace_rq_requeue(struct request_queue *q,
 +                                   struct request *rq)
 +{
 +      blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 +}
 +
 +static void blk_add_trace_rq_complete(struct request_queue *q,
 +                                    struct request *rq)
 +{
 +      blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 +}
 +
 +/**
 + * blk_add_trace_bio - Add a trace for a bio oriented action
 + * @q:                queue the io is for
 + * @bio:      the source bio
 + * @what:     the action
 + *
 + * Description:
 + *     Records an action against a bio. Will log the bio offset + size.
 + *
 + **/
 +static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 +                                   u32 what)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +
 +      if (likely(!bt))
 +              return;
 +
 +      __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 +                      !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 +}
 +
 +static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 +{
 +      blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 +}
 +
 +static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 +{
 +      blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 +}
 +
 +static void blk_add_trace_bio_backmerge(struct request_queue *q,
 +                                      struct bio *bio)
 +{
 +      blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 +}
 +
 +static void blk_add_trace_bio_frontmerge(struct request_queue *q,
 +                                       struct bio *bio)
 +{
 +      blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 +}
 +
 +static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 +{
 +      blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 +}
 +
 +static void blk_add_trace_getrq(struct request_queue *q,
 +                              struct bio *bio, int rw)
 +{
 +      if (bio)
 +              blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 +      else {
 +              struct blk_trace *bt = q->blk_trace;
 +
 +              if (bt)
 +                      __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 +      }
 +}
 +
 +
 +static void blk_add_trace_sleeprq(struct request_queue *q,
 +                                struct bio *bio, int rw)
 +{
 +      if (bio)
 +              blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 +      else {
 +              struct blk_trace *bt = q->blk_trace;
 +
 +              if (bt)
 +                      __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 +                                      0, 0, NULL);
 +      }
 +}
 +
 +static void blk_add_trace_plug(struct request_queue *q)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +
 +      if (bt)
 +              __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 +}
 +
 +static void blk_add_trace_unplug_io(struct request_queue *q)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +
 +      if (bt) {
 +              unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 +              __be64 rpdu = cpu_to_be64(pdu);
 +
 +              __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 +                              sizeof(rpdu), &rpdu);
 +      }
 +}
 +
 +static void blk_add_trace_unplug_timer(struct request_queue *q)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +
 +      if (bt) {
 +              unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 +              __be64 rpdu = cpu_to_be64(pdu);
 +
 +              __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 +                              sizeof(rpdu), &rpdu);
 +      }
 +}
 +
 +static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 +                              unsigned int pdu)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +
 +      if (bt) {
 +              __be64 rpdu = cpu_to_be64(pdu);
 +
 +              __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 +                              BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 +                              sizeof(rpdu), &rpdu);
 +      }
 +}
 +
 +/**
 + * blk_add_trace_remap - Add a trace for a remap operation
 + * @q:                queue the io is for
 + * @bio:      the source bio
 + * @dev:      target device
 + * @from:     source sector
 + * @to:               target sector
 + *
 + * Description:
 + *     Device mapper or raid target sometimes need to split a bio because
 + *     it spans a stripe (or similar). Add a trace for that action.
 + *
 + **/
 +static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 +                                     dev_t dev, sector_t from, sector_t to)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +      struct blk_io_trace_remap r;
 +
 +      if (likely(!bt))
 +              return;
 +
 +      r.device = cpu_to_be32(dev);
 +      r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 +      r.sector = cpu_to_be64(to);
 +
 +      __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 +                      !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 +}
 +
 +/**
 + * blk_add_driver_data - Add binary message with driver-specific data
 + * @q:                queue the io is for
 + * @rq:               io request
 + * @data:     driver-specific data
 + * @len:      length of driver-specific data
 + *
 + * Description:
 + *     Some drivers might want to write driver-specific data per request.
 + *
 + **/
 +void blk_add_driver_data(struct request_queue *q,
 +                       struct request *rq,
 +                       void *data, size_t len)
 +{
 +      struct blk_trace *bt = q->blk_trace;
 +
 +      if (likely(!bt))
 +              return;
 +
 +      if (blk_pc_request(rq))
 +              __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 +                              rq->errors, len, data);
 +      else
 +              __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 +                              0, BLK_TA_DRV_DATA, rq->errors, len, data);
 +}
 +EXPORT_SYMBOL_GPL(blk_add_driver_data);
 +
 +static int blk_register_tracepoints(void)
 +{
 +      int ret;
 +
 +      ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 +      WARN_ON(ret);
 +      ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 +      WARN_ON(ret);
 +      ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 +      WARN_ON(ret);
 +      ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 +      WARN_ON(ret);
 +      ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 +      WARN_ON(ret);
 +      ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 +      WARN_ON(ret);
 +      ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 +      WARN_ON(ret);
 +      ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 +      WARN_ON(ret);
 +      ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 +      WARN_ON(ret);
 +      ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 +      WARN_ON(ret);
 +      ret = register_trace_block_getrq(blk_add_trace_getrq);
 +      WARN_ON(ret);
 +      ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 +      WARN_ON(ret);
 +      ret = register_trace_block_plug(blk_add_trace_plug);
 +      WARN_ON(ret);
 +      ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 +      WARN_ON(ret);
 +      ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 +      WARN_ON(ret);
 +      ret = register_trace_block_split(blk_add_trace_split);
 +      WARN_ON(ret);
 +      ret = register_trace_block_remap(blk_add_trace_remap);
 +      WARN_ON(ret);
 +      return 0;
 +}
 +
 +static void blk_unregister_tracepoints(void)
 +{
 +      unregister_trace_block_remap(blk_add_trace_remap);
 +      unregister_trace_block_split(blk_add_trace_split);
 +      unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 +      unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 +      unregister_trace_block_plug(blk_add_trace_plug);
 +      unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 +      unregister_trace_block_getrq(blk_add_trace_getrq);
 +      unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 +      unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 +      unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 +      unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 +      unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 +      unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 +      unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 +      unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 +      unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 +      unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 +
 +      tracepoint_synchronize_unregister();
 +}
 +
 +/*
 + * struct blk_io_tracer formatting routines
 + */
 +
 +static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 +{
 +      int i = 0;
 +
 +      if (t->action & BLK_TC_DISCARD)
 +              rwbs[i++] = 'D';
 +      else if (t->action & BLK_TC_WRITE)
 +              rwbs[i++] = 'W';
 +      else if (t->bytes)
 +              rwbs[i++] = 'R';
 +      else
 +              rwbs[i++] = 'N';
 +
 +      if (t->action & BLK_TC_AHEAD)
 +              rwbs[i++] = 'A';
 +      if (t->action & BLK_TC_BARRIER)
 +              rwbs[i++] = 'B';
 +      if (t->action & BLK_TC_SYNC)
 +              rwbs[i++] = 'S';
 +      if (t->action & BLK_TC_META)
 +              rwbs[i++] = 'M';
 +
 +      rwbs[i] = '\0';
 +}
 +
 +static inline
 +const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 +{
 +      return (const struct blk_io_trace *)ent;
 +}
 +
 +static inline const void *pdu_start(const struct trace_entry *ent)
 +{
 +      return te_blk_io_trace(ent) + 1;
 +}
 +
 +static inline u32 t_sec(const struct trace_entry *ent)
 +{
 +      return te_blk_io_trace(ent)->bytes >> 9;
 +}
 +
 +static inline unsigned long long t_sector(const struct trace_entry *ent)
 +{
 +      return te_blk_io_trace(ent)->sector;
 +}
 +
 +static inline __u16 t_error(const struct trace_entry *ent)
 +{
 +      return te_blk_io_trace(ent)->sector;
 +}
 +
 +static __u64 get_pdu_int(const struct trace_entry *ent)
 +{
 +      const __u64 *val = pdu_start(ent);
 +      return be64_to_cpu(*val);
 +}
 +
 +static void get_pdu_remap(const struct trace_entry *ent,
 +                        struct blk_io_trace_remap *r)
 +{
 +      const struct blk_io_trace_remap *__r = pdu_start(ent);
 +      __u64 sector = __r->sector;
 +
 +      r->device = be32_to_cpu(__r->device);
 +      r->device_from = be32_to_cpu(__r->device_from);
 +      r->sector = be64_to_cpu(sector);
 +}
 +
 +static int blk_log_action_iter(struct trace_iterator *iter, const char *act)
 +{
 +      char rwbs[6];
 +      unsigned long long ts  = ns2usecs(iter->ts);
 +      unsigned long usec_rem = do_div(ts, USEC_PER_SEC);
 +      unsigned secs          = (unsigned long)ts;
 +      const struct trace_entry *ent = iter->ent;
 +      const struct blk_io_trace *t = (const struct blk_io_trace *)ent;
 +
 +      fill_rwbs(rwbs, t);
 +
 +      return trace_seq_printf(&iter->seq,
 +                              "%3d,%-3d %2d %5d.%06lu %5u %2s %3s ",
 +                              MAJOR(t->device), MINOR(t->device), iter->cpu,
 +                              secs, usec_rem, ent->pid, act, rwbs);
 +}
 +
 +static int blk_log_action_seq(struct trace_seq *s, const struct blk_io_trace *t,
 +                            const char *act)
 +{
 +      char rwbs[6];
 +      fill_rwbs(rwbs, t);
 +      return trace_seq_printf(s, "%3d,%-3d %2s %3s ",
 +                              MAJOR(t->device), MINOR(t->device), act, rwbs);
 +}
 +
 +static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
 +{
 +      const char *cmd = trace_find_cmdline(ent->pid);
 +
 +      if (t_sec(ent))
 +              return trace_seq_printf(s, "%llu + %u [%s]\n",
 +                                      t_sector(ent), t_sec(ent), cmd);
 +      return trace_seq_printf(s, "[%s]\n", cmd);
 +}
 +
 +static int blk_log_with_error(struct trace_seq *s,
 +                            const struct trace_entry *ent)
 +{
 +      if (t_sec(ent))
 +              return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
 +                                      t_sec(ent), t_error(ent));
 +      return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
 +}
 +
 +static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
 +{
 +      struct blk_io_trace_remap r = { .device = 0, };
 +
 +      get_pdu_remap(ent, &r);
 +      return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
 +                             t_sector(ent),
 +                             t_sec(ent), MAJOR(r.device), MINOR(r.device),
 +                             (unsigned long long)r.sector);
 +}
 +
 +static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
 +{
 +      return trace_seq_printf(s, "[%s]\n", trace_find_cmdline(ent->pid));
 +}
 +
 +static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
 +{
 +      return trace_seq_printf(s, "[%s] %llu\n", trace_find_cmdline(ent->pid),
 +                              get_pdu_int(ent));
 +}
 +
 +static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
 +{
 +      return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
 +                              get_pdu_int(ent), trace_find_cmdline(ent->pid));
 +}
 +
 +/*
 + * struct tracer operations
 + */
 +
 +static void blk_tracer_print_header(struct seq_file *m)
 +{
 +      if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
 +              return;
 +      seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
 +                  "#  |     |     |           |   |   |\n");
 +}
 +
 +static void blk_tracer_start(struct trace_array *tr)
 +{
 +      mutex_lock(&blk_probe_mutex);
 +      if (atomic_add_return(1, &blk_probes_ref) == 1)
 +              if (blk_register_tracepoints())
 +                      atomic_dec(&blk_probes_ref);
 +      mutex_unlock(&blk_probe_mutex);
 +      trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
 +}
 +
 +static int blk_tracer_init(struct trace_array *tr)
 +{
 +      blk_tr = tr;
 +      blk_tracer_start(tr);
 +      mutex_lock(&blk_probe_mutex);
 +      blk_tracer_enabled++;
 +      mutex_unlock(&blk_probe_mutex);
 +      return 0;
 +}
 +
 +static void blk_tracer_stop(struct trace_array *tr)
 +{
 +      trace_flags |= TRACE_ITER_CONTEXT_INFO;
 +      mutex_lock(&blk_probe_mutex);
 +      if (atomic_dec_and_test(&blk_probes_ref))
 +              blk_unregister_tracepoints();
 +      mutex_unlock(&blk_probe_mutex);
 +}
 +
 +static void blk_tracer_reset(struct trace_array *tr)
 +{
 +      if (!atomic_read(&blk_probes_ref))
 +              return;
 +
 +      mutex_lock(&blk_probe_mutex);
 +      blk_tracer_enabled--;
 +      WARN_ON(blk_tracer_enabled < 0);
 +      mutex_unlock(&blk_probe_mutex);
 +
 +      blk_tracer_stop(tr);
 +}
 +
 +static struct {
 +      const char *act[2];
 +      int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
 +} what2act[] __read_mostly = {
 +      [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
 +      [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
 +      [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
 +      [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
 +      [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
 +      [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
 +      [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
 +      [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
 +      [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
 +      [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
 +      [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
 +      [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
 +      [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
 +      [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
 +      [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
 +};
 +
 +static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
 +                                             int flags)
 +{
 +      struct trace_seq *s = &iter->seq;
 +      const struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
 +      const u16 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
 +      int ret;
 +
 +      if (!trace_print_context(iter))
 +              return TRACE_TYPE_PARTIAL_LINE;
 +
 +      if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
 +              ret = trace_seq_printf(s, "Bad pc action %x\n", what);
 +      else {
 +              const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
 +              ret = blk_log_action_seq(s, t, what2act[what].act[long_act]);
 +              if (ret)
 +                      ret = what2act[what].print(s, iter->ent);
 +      }
 +
 +      return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 +}
 +
 +static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
 +{
 +      struct trace_seq *s = &iter->seq;
 +      struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
 +      const int offset = offsetof(struct blk_io_trace, sector);
 +      struct blk_io_trace old = {
 +              .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
 +              .time     = ns2usecs(iter->ts),
 +      };
 +
 +      if (!trace_seq_putmem(s, &old, offset))
 +              return 0;
 +      return trace_seq_putmem(s, &t->sector,
 +                              sizeof(old) - offset + t->pdu_len);
 +}
 +
 +static enum print_line_t
 +blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
 +{
 +      return blk_trace_synthesize_old_trace(iter) ?
 +                      TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 +}
 +
 +static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
 +{
 +      const struct blk_io_trace *t;
 +      u16 what;
 +      int ret;
 +
 +      if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
 +              return TRACE_TYPE_UNHANDLED;
 +
 +      t = (const struct blk_io_trace *)iter->ent;
 +      what = t->action & ((1 << BLK_TC_SHIFT) - 1);
 +
 +      if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
 +              ret = trace_seq_printf(&iter->seq, "Bad pc action %x\n", what);
 +      else {
 +              const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
 +              ret = blk_log_action_iter(iter, what2act[what].act[long_act]);
 +              if (ret)
 +                      ret = what2act[what].print(&iter->seq, iter->ent);
 +      }
 +
 +      return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 +}
 +
 +static struct tracer blk_tracer __read_mostly = {
 +      .name           = "blk",
 +      .init           = blk_tracer_init,
 +      .reset          = blk_tracer_reset,
 +      .start          = blk_tracer_start,
 +      .stop           = blk_tracer_stop,
 +      .print_header   = blk_tracer_print_header,
 +      .print_line     = blk_tracer_print_line,
 +      .flags          = &blk_tracer_flags,
 +};
 +
 +static struct trace_event trace_blk_event = {
 +      .type           = TRACE_BLK,
 +      .trace          = blk_trace_event_print,
 +      .latency_trace  = blk_trace_event_print,
 +      .binary         = blk_trace_event_print_binary,
 +};
 +
 +static int __init init_blk_tracer(void)
 +{
 +      if (!register_ftrace_event(&trace_blk_event)) {
 +              pr_warning("Warning: could not register block events\n");
 +              return 1;
 +      }
 +
 +      if (register_tracer(&blk_tracer) != 0) {
 +              pr_warning("Warning: could not register the block tracer\n");
 +              unregister_ftrace_event(&trace_blk_event);
 +              return 1;
 +      }
 +
 +      return 0;
 +}
 +
 +device_initcall(init_blk_tracer);
 +
 +static int blk_trace_remove_queue(struct request_queue *q)
 +{
 +      struct blk_trace *bt;
 +
 +      bt = xchg(&q->blk_trace, NULL);
 +      if (bt == NULL)
 +              return -EINVAL;
 +
 +      kfree(bt);
 +      return 0;
 +}
 +
 +/*
 + * Setup everything required to start tracing
 + */
 +static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
 +{
 +      struct blk_trace *old_bt, *bt = NULL;
 +      int ret;
 +
 +      ret = -ENOMEM;
 +      bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 +      if (!bt)
 +              goto err;
 +
 +      bt->dev = dev;
 +      bt->act_mask = (u16)-1;
 +      bt->end_lba = -1ULL;
 +      bt->trace_state = Blktrace_running;
 +
 +      old_bt = xchg(&q->blk_trace, bt);
 +      if (old_bt != NULL) {
 +              (void)xchg(&q->blk_trace, old_bt);
 +              kfree(bt);
 +              ret = -EBUSY;
 +      }
 +      return 0;
 +err:
 +      return ret;
 +}
 +
 +/*
 + * sysfs interface to enable and configure tracing
 + */
 +
 +static ssize_t sysfs_blk_trace_enable_show(struct device *dev,
 +                                         struct device_attribute *attr,
 +                                         char *buf)
 +{
 +      struct hd_struct *p = dev_to_part(dev);
 +      struct block_device *bdev;
 +      ssize_t ret = -ENXIO;
 +
 +      lock_kernel();
 +      bdev = bdget(part_devt(p));
 +      if (bdev != NULL) {
 +              struct request_queue *q = bdev_get_queue(bdev);
 +
 +              if (q != NULL) {
 +                      mutex_lock(&bdev->bd_mutex);
 +                      ret = sprintf(buf, "%u\n", !!q->blk_trace);
 +                      mutex_unlock(&bdev->bd_mutex);
 +              }
 +
 +              bdput(bdev);
 +      }
 +
 +      unlock_kernel();
 +      return ret;
 +}
 +
 +static ssize_t sysfs_blk_trace_enable_store(struct device *dev,
 +                                          struct device_attribute *attr,
 +                                          const char *buf, size_t count)
 +{
 +      struct block_device *bdev;
 +      struct request_queue *q;
 +      struct hd_struct *p;
 +      int value;
 +      ssize_t ret = -ENXIO;
 +
 +      if (count == 0 || sscanf(buf, "%d", &value) != 1)
 +              goto out;
 +
 +      lock_kernel();
 +      p = dev_to_part(dev);
 +      bdev = bdget(part_devt(p));
 +      if (bdev == NULL)
 +              goto out_unlock_kernel;
 +
 +      q = bdev_get_queue(bdev);
 +      if (q == NULL)
 +              goto out_bdput;
 +
 +      mutex_lock(&bdev->bd_mutex);
 +      if (value)
 +              ret = blk_trace_setup_queue(q, bdev->bd_dev);
 +      else
 +              ret = blk_trace_remove_queue(q);
 +      mutex_unlock(&bdev->bd_mutex);
 +
 +      if (ret == 0)
 +              ret = count;
 +out_bdput:
 +      bdput(bdev);
 +out_unlock_kernel:
 +      unlock_kernel();
 +out:
 +      return ret;
 +}
 +
 +static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
 +                                       struct device_attribute *attr,
 +                                       char *buf);
 +static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
 +                                        struct device_attribute *attr,
 +                                        const char *buf, size_t count);
 +#define BLK_TRACE_DEVICE_ATTR(_name) \
 +      DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
 +                  sysfs_blk_trace_attr_show, \
 +                  sysfs_blk_trace_attr_store)
 +
 +static DEVICE_ATTR(enable, S_IRUGO | S_IWUSR,
 +                 sysfs_blk_trace_enable_show, sysfs_blk_trace_enable_store);
 +static BLK_TRACE_DEVICE_ATTR(act_mask);
 +static BLK_TRACE_DEVICE_ATTR(pid);
 +static BLK_TRACE_DEVICE_ATTR(start_lba);
 +static BLK_TRACE_DEVICE_ATTR(end_lba);
 +
 +static struct attribute *blk_trace_attrs[] = {
 +      &dev_attr_enable.attr,
 +      &dev_attr_act_mask.attr,
 +      &dev_attr_pid.attr,
 +      &dev_attr_start_lba.attr,
 +      &dev_attr_end_lba.attr,
 +      NULL
 +};
 +
 +struct attribute_group blk_trace_attr_group = {
 +      .name  = "trace",
 +      .attrs = blk_trace_attrs,
 +};
 +
 +static int blk_str2act_mask(const char *str)
 +{
 +      int mask = 0;
 +      char *copy = kstrdup(str, GFP_KERNEL), *s;
 +
 +      if (copy == NULL)
 +              return -ENOMEM;
 +
 +      s = strstrip(copy);
 +
 +      while (1) {
 +              char *sep = strchr(s, ',');
 +
 +              if (sep != NULL)
 +                      *sep = '\0';
 +
 +              if (strcasecmp(s, "barrier") == 0)
 +                      mask |= BLK_TC_BARRIER;
 +              else if (strcasecmp(s, "complete") == 0)
 +                      mask |= BLK_TC_COMPLETE;
 +              else if (strcasecmp(s, "fs") == 0)
 +                      mask |= BLK_TC_FS;
 +              else if (strcasecmp(s, "issue") == 0)
 +                      mask |= BLK_TC_ISSUE;
 +              else if (strcasecmp(s, "pc") == 0)
 +                      mask |= BLK_TC_PC;
 +              else if (strcasecmp(s, "queue") == 0)
 +                      mask |= BLK_TC_QUEUE;
 +              else if (strcasecmp(s, "read") == 0)
 +                      mask |= BLK_TC_READ;
 +              else if (strcasecmp(s, "requeue") == 0)
 +                      mask |= BLK_TC_REQUEUE;
 +              else if (strcasecmp(s, "sync") == 0)
 +                      mask |= BLK_TC_SYNC;
 +              else if (strcasecmp(s, "write") == 0)
 +                      mask |= BLK_TC_WRITE;
 +
 +              if (sep == NULL)
 +                      break;
 +
 +              s = sep + 1;
 +      }
 +      kfree(copy);
 +
 +      return mask;
 +}
 +
 +static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
 +                                       struct device_attribute *attr,
 +                                       char *buf)
 +{
 +      struct hd_struct *p = dev_to_part(dev);
 +      struct request_queue *q;
 +      struct block_device *bdev;
 +      ssize_t ret = -ENXIO;
 +
 +      lock_kernel();
 +      bdev = bdget(part_devt(p));
 +      if (bdev == NULL)
 +              goto out_unlock_kernel;
 +
 +      q = bdev_get_queue(bdev);
 +      if (q == NULL)
 +              goto out_bdput;
 +      mutex_lock(&bdev->bd_mutex);
 +      if (q->blk_trace == NULL)
 +              ret = sprintf(buf, "disabled\n");
 +      else if (attr == &dev_attr_act_mask)
 +              ret = sprintf(buf, "%#x\n", q->blk_trace->act_mask);
 +      else if (attr == &dev_attr_pid)
 +              ret = sprintf(buf, "%u\n", q->blk_trace->pid);
 +      else if (attr == &dev_attr_start_lba)
 +              ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
 +      else if (attr == &dev_attr_end_lba)
 +              ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
 +      mutex_unlock(&bdev->bd_mutex);
 +out_bdput:
 +      bdput(bdev);
 +out_unlock_kernel:
 +      unlock_kernel();
 +      return ret;
 +}
 +
 +static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
 +                                        struct device_attribute *attr,
 +                                        const char *buf, size_t count)
 +{
 +      struct block_device *bdev;
 +      struct request_queue *q;
 +      struct hd_struct *p;
 +      u64 value;
 +      ssize_t ret = -ENXIO;
 +
 +      if (count == 0)
 +              goto out;
 +
 +      if (attr == &dev_attr_act_mask) {
 +              if (sscanf(buf, "%llx", &value) != 1) {
 +                      /* Assume it is a list of trace category names */
 +                      value = blk_str2act_mask(buf);
 +                      if (value < 0)
 +                              goto out;
 +              }
 +      } else if (sscanf(buf, "%llu", &value) != 1)
 +              goto out;
 +
 +      lock_kernel();
 +      p = dev_to_part(dev);
 +      bdev = bdget(part_devt(p));
 +      if (bdev == NULL)
 +              goto out_unlock_kernel;
 +
 +      q = bdev_get_queue(bdev);
 +      if (q == NULL)
 +              goto out_bdput;
 +
 +      mutex_lock(&bdev->bd_mutex);
 +      ret = 0;
 +      if (q->blk_trace == NULL)
 +              ret = blk_trace_setup_queue(q, bdev->bd_dev);
 +
 +      if (ret == 0) {
 +              if (attr == &dev_attr_act_mask)
 +                      q->blk_trace->act_mask = value;
 +              else if (attr == &dev_attr_pid)
 +                      q->blk_trace->pid = value;
 +              else if (attr == &dev_attr_start_lba)
 +                      q->blk_trace->start_lba = value;
 +              else if (attr == &dev_attr_end_lba)
 +                      q->blk_trace->end_lba = value;
 +              ret = count;
 +      }
 +      mutex_unlock(&bdev->bd_mutex);
 +out_bdput:
 +      bdput(bdev);
 +out_unlock_kernel:
 +      unlock_kernel();
 +out:
 +      return ret;
 +}