2 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/blktrace_api.h>
21 #include <linux/percpu.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/debugfs.h>
25 #include <linux/time.h>
26 #include <asm/uaccess.h>
28 static DEFINE_PER_CPU(unsigned long long, blk_trace_cpu_offset) = { 0, };
29 static unsigned int blktrace_seq __read_mostly = 1;
32 * Send out a notify message.
34 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
35 const void *data, size_t len)
37 struct blk_io_trace *t;
39 t = relay_reserve(bt->rchan, sizeof(*t) + len);
41 const int cpu = smp_processor_id();
43 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
44 t->time = sched_clock() - per_cpu(blk_trace_cpu_offset, cpu);
50 memcpy((void *) t + sizeof(*t), data, len);
55 * Send out a notify for this process, if we haven't done so since a trace
58 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
60 tsk->btrace_seq = blktrace_seq;
61 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
64 static void trace_note_time(struct blk_trace *bt)
71 words[0] = now.tv_sec;
72 words[1] = now.tv_nsec;
74 local_irq_save(flags);
75 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
76 local_irq_restore(flags);
79 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
82 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
84 if (sector < bt->start_lba || sector > bt->end_lba)
86 if (bt->pid && pid != bt->pid)
93 * Data direction bit lookup
95 static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ), BLK_TC_ACT(BLK_TC_WRITE) };
98 * Bio action bits of interest
100 static u32 bio_act[9] __read_mostly = { 0, BLK_TC_ACT(BLK_TC_BARRIER), BLK_TC_ACT(BLK_TC_SYNC), 0, BLK_TC_ACT(BLK_TC_AHEAD), 0, 0, 0, BLK_TC_ACT(BLK_TC_META) };
103 * More could be added as needed, taking care to increment the decrementer
104 * to get correct indexing
106 #define trace_barrier_bit(rw) \
107 (((rw) & (1 << BIO_RW_BARRIER)) >> (BIO_RW_BARRIER - 0))
108 #define trace_sync_bit(rw) \
109 (((rw) & (1 << BIO_RW_SYNC)) >> (BIO_RW_SYNC - 1))
110 #define trace_ahead_bit(rw) \
111 (((rw) & (1 << BIO_RW_AHEAD)) << (2 - BIO_RW_AHEAD))
112 #define trace_meta_bit(rw) \
113 (((rw) & (1 << BIO_RW_META)) >> (BIO_RW_META - 3))
116 * The worker for the various blk_add_trace*() types. Fills out a
117 * blk_io_trace structure and places it in a per-cpu subbuffer.
119 void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
120 int rw, u32 what, int error, int pdu_len, void *pdu_data)
122 struct task_struct *tsk = current;
123 struct blk_io_trace *t;
125 unsigned long *sequence;
129 if (unlikely(bt->trace_state != Blktrace_running))
132 what |= ddir_act[rw & WRITE];
133 what |= bio_act[trace_barrier_bit(rw)];
134 what |= bio_act[trace_sync_bit(rw)];
135 what |= bio_act[trace_ahead_bit(rw)];
136 what |= bio_act[trace_meta_bit(rw)];
139 if (unlikely(act_log_check(bt, what, sector, pid)))
143 * A word about the locking here - we disable interrupts to reserve
144 * some space in the relay per-cpu buffer, to prevent an irq
145 * from coming in and stepping on our toes. Once reserved, it's
146 * enough to get preemption disabled to prevent read of this data
147 * before we are through filling it. get_cpu()/put_cpu() does this
150 local_irq_save(flags);
152 if (unlikely(tsk->btrace_seq != blktrace_seq))
153 trace_note_tsk(bt, tsk);
155 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
157 cpu = smp_processor_id();
158 sequence = per_cpu_ptr(bt->sequence, cpu);
160 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
161 t->sequence = ++(*sequence);
162 t->time = sched_clock() - per_cpu(blk_trace_cpu_offset, cpu);
170 t->pdu_len = pdu_len;
173 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
176 local_irq_restore(flags);
179 EXPORT_SYMBOL_GPL(__blk_add_trace);
181 static struct dentry *blk_tree_root;
182 static struct mutex blk_tree_mutex;
183 static unsigned int root_users;
185 static inline void blk_remove_root(void)
188 debugfs_remove(blk_tree_root);
189 blk_tree_root = NULL;
193 static void blk_remove_tree(struct dentry *dir)
195 mutex_lock(&blk_tree_mutex);
197 if (--root_users == 0)
199 mutex_unlock(&blk_tree_mutex);
202 static struct dentry *blk_create_tree(const char *blk_name)
204 struct dentry *dir = NULL;
206 mutex_lock(&blk_tree_mutex);
208 if (!blk_tree_root) {
209 blk_tree_root = debugfs_create_dir("block", NULL);
214 dir = debugfs_create_dir(blk_name, blk_tree_root);
221 mutex_unlock(&blk_tree_mutex);
225 static void blk_trace_cleanup(struct blk_trace *bt)
227 relay_close(bt->rchan);
228 debugfs_remove(bt->dropped_file);
229 blk_remove_tree(bt->dir);
230 free_percpu(bt->sequence);
234 static int blk_trace_remove(request_queue_t *q)
236 struct blk_trace *bt;
238 bt = xchg(&q->blk_trace, NULL);
242 if (bt->trace_state == Blktrace_setup ||
243 bt->trace_state == Blktrace_stopped)
244 blk_trace_cleanup(bt);
249 static int blk_dropped_open(struct inode *inode, struct file *filp)
251 filp->private_data = inode->i_private;
256 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
257 size_t count, loff_t *ppos)
259 struct blk_trace *bt = filp->private_data;
262 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
264 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
267 static struct file_operations blk_dropped_fops = {
268 .owner = THIS_MODULE,
269 .open = blk_dropped_open,
270 .read = blk_dropped_read,
274 * Keep track of how many times we encountered a full subbuffer, to aid
275 * the user space app in telling how many lost events there were.
277 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
278 void *prev_subbuf, size_t prev_padding)
280 struct blk_trace *bt;
282 if (!relay_buf_full(buf))
285 bt = buf->chan->private_data;
286 atomic_inc(&bt->dropped);
290 static int blk_remove_buf_file_callback(struct dentry *dentry)
292 debugfs_remove(dentry);
296 static struct dentry *blk_create_buf_file_callback(const char *filename,
297 struct dentry *parent,
299 struct rchan_buf *buf,
302 return debugfs_create_file(filename, mode, parent, buf,
303 &relay_file_operations);
306 static struct rchan_callbacks blk_relay_callbacks = {
307 .subbuf_start = blk_subbuf_start_callback,
308 .create_buf_file = blk_create_buf_file_callback,
309 .remove_buf_file = blk_remove_buf_file_callback,
313 * Setup everything required to start tracing
315 static int blk_trace_setup(request_queue_t *q, struct block_device *bdev,
318 struct blk_user_trace_setup buts;
319 struct blk_trace *old_bt, *bt = NULL;
320 struct dentry *dir = NULL;
321 char b[BDEVNAME_SIZE];
324 if (copy_from_user(&buts, arg, sizeof(buts)))
327 if (!buts.buf_size || !buts.buf_nr)
330 strcpy(buts.name, bdevname(bdev, b));
333 * some device names have larger paths - convert the slashes
334 * to underscores for this to work as expected
336 for (i = 0; i < strlen(buts.name); i++)
337 if (buts.name[i] == '/')
340 if (copy_to_user(arg, &buts, sizeof(buts)))
344 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
348 bt->sequence = alloc_percpu(unsigned long);
353 dir = blk_create_tree(buts.name);
358 bt->dev = bdev->bd_dev;
359 atomic_set(&bt->dropped, 0);
362 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
363 if (!bt->dropped_file)
366 bt->rchan = relay_open("trace", dir, buts.buf_size, buts.buf_nr, &blk_relay_callbacks);
369 bt->rchan->private_data = bt;
371 bt->act_mask = buts.act_mask;
373 bt->act_mask = (u16) -1;
375 bt->start_lba = buts.start_lba;
376 bt->end_lba = buts.end_lba;
381 bt->trace_state = Blktrace_setup;
384 old_bt = xchg(&q->blk_trace, bt);
386 (void) xchg(&q->blk_trace, old_bt);
393 blk_remove_tree(dir);
395 if (bt->dropped_file)
396 debugfs_remove(bt->dropped_file);
397 free_percpu(bt->sequence);
399 relay_close(bt->rchan);
405 static int blk_trace_startstop(request_queue_t *q, int start)
407 struct blk_trace *bt;
410 if ((bt = q->blk_trace) == NULL)
414 * For starting a trace, we can transition from a setup or stopped
415 * trace. For stopping a trace, the state must be running
419 if (bt->trace_state == Blktrace_setup ||
420 bt->trace_state == Blktrace_stopped) {
423 bt->trace_state = Blktrace_running;
429 if (bt->trace_state == Blktrace_running) {
430 bt->trace_state = Blktrace_stopped;
431 relay_flush(bt->rchan);
440 * blk_trace_ioctl: - handle the ioctls associated with tracing
441 * @bdev: the block device
442 * @cmd: the ioctl cmd
443 * @arg: the argument data, if any
446 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
451 q = bdev_get_queue(bdev);
455 mutex_lock(&bdev->bd_mutex);
459 ret = blk_trace_setup(q, bdev, arg);
464 ret = blk_trace_startstop(q, start);
466 case BLKTRACETEARDOWN:
467 ret = blk_trace_remove(q);
474 mutex_unlock(&bdev->bd_mutex);
479 * blk_trace_shutdown: - stop and cleanup trace structures
480 * @q: the request queue associated with the device
483 void blk_trace_shutdown(request_queue_t *q)
486 blk_trace_startstop(q, 0);
492 * Average offset over two calls to sched_clock() with a gettimeofday()
495 static void blk_check_time(unsigned long long *t)
497 unsigned long long a, b;
501 do_gettimeofday(&tv);
504 *t = tv.tv_sec * 1000000000 + tv.tv_usec * 1000;
509 * calibrate our inter-CPU timings
511 static void blk_trace_check_cpu_time(void *data)
513 unsigned long long *t;
516 t = &per_cpu(blk_trace_cpu_offset, cpu);
519 * Just call it twice, hopefully the second call will be cache hot
520 * and a little more precise
528 static void blk_trace_set_ht_offsets(void)
530 #if defined(CONFIG_SCHED_SMT)
534 * now make sure HT siblings have the same time offset
537 for_each_online_cpu(cpu) {
538 unsigned long long *cpu_off, *sibling_off;
540 for_each_cpu_mask(i, cpu_sibling_map[cpu]) {
544 cpu_off = &per_cpu(blk_trace_cpu_offset, cpu);
545 sibling_off = &per_cpu(blk_trace_cpu_offset, i);
546 *sibling_off = *cpu_off;
553 static __init int blk_trace_init(void)
555 mutex_init(&blk_tree_mutex);
556 on_each_cpu(blk_trace_check_cpu_time, NULL, 1, 1);
557 blk_trace_set_ht_offsets();
562 module_init(blk_trace_init);