blob: c1bd4ada2a044f9b8eb8971ef41e05b26790b31d [file] [log] [blame]
Jens Axboe2056a782006-03-23 20:00:26 +01001/*
Jens Axboe0fe23472006-09-04 15:41:16 +02002 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
Jens Axboe2056a782006-03-23 20:00:26 +01003 *
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.
7 *
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.
12 *
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
16 *
17 */
Jens Axboe2056a782006-03-23 20:00:26 +010018#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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090024#include <linux/slab.h>
Jens Axboe2056a782006-03-23 20:00:26 +010025#include <linux/debugfs.h>
Paul Gortmaker6e5fdee2011-05-26 16:00:52 -040026#include <linux/export.h>
Olaf Kirchbe1c6342006-12-01 10:39:12 +010027#include <linux/time.h>
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -020028#include <linux/uaccess.h>
Jan Karaa404d552013-09-17 22:30:31 +020029#include <linux/list.h>
Li Zefan55782132009-06-09 13:43:05 +080030
31#include <trace/events/block.h>
32
Frederic Weisbecker2db270a2009-02-07 20:46:45 +010033#include "trace_output.h"
Jens Axboe2056a782006-03-23 20:00:26 +010034
Li Zefan55782132009-06-09 13:43:05 +080035#ifdef CONFIG_BLK_DEV_IO_TRACE
36
Jens Axboe2056a782006-03-23 20:00:26 +010037static unsigned int blktrace_seq __read_mostly = 1;
38
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -020039static struct trace_array *blk_tr;
Li Zefan5006ea72009-03-20 09:48:03 +080040static bool blk_tracer_enabled __read_mostly;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -020041
Jan Karaa404d552013-09-17 22:30:31 +020042static LIST_HEAD(running_trace_list);
43static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
44
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -020045/* Select an alternative, minimalistic output than the original one */
Steven Rostedtef180122009-03-10 14:10:56 -040046#define TRACE_BLK_OPT_CLASSIC 0x1
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -020047
48static struct tracer_opt blk_tracer_opts[] = {
49 /* Default disable the minimalistic output */
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -020050 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -020051 { }
52};
53
54static struct tracer_flags blk_tracer_flags = {
55 .val = 0,
56 .opts = blk_tracer_opts,
57};
58
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010059/* Global reference count of probes */
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010060static atomic_t blk_probes_ref = ATOMIC_INIT(0);
61
Li Zefan3c289ba2009-03-20 09:48:26 +080062static void blk_register_tracepoints(void);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010063static void blk_unregister_tracepoints(void);
64
Jens Axboe2056a782006-03-23 20:00:26 +010065/*
Olaf Kirchbe1c6342006-12-01 10:39:12 +010066 * Send out a notify message.
67 */
Jens Axboea8630552006-12-04 09:30:58 +010068static void trace_note(struct blk_trace *bt, pid_t pid, int action,
69 const void *data, size_t len)
Olaf Kirchbe1c6342006-12-01 10:39:12 +010070{
71 struct blk_io_trace *t;
Li Zefan18cea452009-03-27 10:21:54 +080072 struct ring_buffer_event *event = NULL;
Steven Rostedte77405a2009-09-02 14:17:06 -040073 struct ring_buffer *buffer = NULL;
Li Zefan18cea452009-03-27 10:21:54 +080074 int pc = 0;
75 int cpu = smp_processor_id();
76 bool blk_tracer = blk_tracer_enabled;
77
78 if (blk_tracer) {
Steven Rostedt (Red Hat)12883ef2013-03-05 09:24:35 -050079 buffer = blk_tr->trace_buffer.buffer;
Li Zefan18cea452009-03-27 10:21:54 +080080 pc = preempt_count();
Steven Rostedte77405a2009-09-02 14:17:06 -040081 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
Li Zefan18cea452009-03-27 10:21:54 +080082 sizeof(*t) + len,
83 0, pc);
84 if (!event)
85 return;
86 t = ring_buffer_event_data(event);
87 goto record_it;
88 }
Olaf Kirchbe1c6342006-12-01 10:39:12 +010089
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -020090 if (!bt->rchan)
91 return;
92
Olaf Kirchbe1c6342006-12-01 10:39:12 +010093 t = relay_reserve(bt->rchan, sizeof(*t) + len);
Jens Axboed3d9d2a2006-12-04 09:27:41 +010094 if (t) {
Jens Axboed3d9d2a2006-12-04 09:27:41 +010095 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
Ingo Molnar2997c8c2008-01-11 13:35:54 +010096 t->time = ktime_to_ns(ktime_get());
Li Zefan18cea452009-03-27 10:21:54 +080097record_it:
Jens Axboed3d9d2a2006-12-04 09:27:41 +010098 t->device = bt->dev;
99 t->action = action;
100 t->pid = pid;
101 t->cpu = cpu;
102 t->pdu_len = len;
103 memcpy((void *) t + sizeof(*t), data, len);
Li Zefan18cea452009-03-27 10:21:54 +0800104
105 if (blk_tracer)
Steven Rostedte77405a2009-09-02 14:17:06 -0400106 trace_buffer_unlock_commit(buffer, event, 0, pc);
Jens Axboed3d9d2a2006-12-04 09:27:41 +0100107 }
Olaf Kirchbe1c6342006-12-01 10:39:12 +0100108}
109
110/*
Jens Axboe2056a782006-03-23 20:00:26 +0100111 * Send out a notify for this process, if we haven't done so since a trace
112 * started
113 */
Jan Karaa404d552013-09-17 22:30:31 +0200114static void trace_note_tsk(struct task_struct *tsk)
Jens Axboe2056a782006-03-23 20:00:26 +0100115{
Jan Karaa404d552013-09-17 22:30:31 +0200116 unsigned long flags;
117 struct blk_trace *bt;
118
Jens Axboea8630552006-12-04 09:30:58 +0100119 tsk->btrace_seq = blktrace_seq;
Jan Karaa404d552013-09-17 22:30:31 +0200120 spin_lock_irqsave(&running_trace_lock, flags);
121 list_for_each_entry(bt, &running_trace_list, running_list) {
122 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
123 sizeof(tsk->comm));
124 }
125 spin_unlock_irqrestore(&running_trace_lock, flags);
Olaf Kirchbe1c6342006-12-01 10:39:12 +0100126}
Jens Axboe2056a782006-03-23 20:00:26 +0100127
Olaf Kirchbe1c6342006-12-01 10:39:12 +0100128static void trace_note_time(struct blk_trace *bt)
129{
130 struct timespec now;
131 unsigned long flags;
132 u32 words[2];
133
134 getnstimeofday(&now);
135 words[0] = now.tv_sec;
136 words[1] = now.tv_nsec;
137
138 local_irq_save(flags);
139 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
140 local_irq_restore(flags);
Jens Axboe2056a782006-03-23 20:00:26 +0100141}
142
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +0200143void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
144{
145 int n;
146 va_list args;
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +0200147 unsigned long flags;
Jens Axboe64565912008-05-28 14:45:33 +0200148 char *buf;
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +0200149
Li Zefan18cea452009-03-27 10:21:54 +0800150 if (unlikely(bt->trace_state != Blktrace_running &&
151 !blk_tracer_enabled))
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200152 return;
153
Tao Ma490da402011-01-19 10:51:44 +0800154 /*
155 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
156 * message to the trace.
157 */
158 if (!(bt->act_mask & BLK_TC_NOTIFY))
159 return;
160
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +0200161 local_irq_save(flags);
Shan Weid8a03492012-11-13 09:53:04 +0800162 buf = this_cpu_ptr(bt->msg_data);
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +0200163 va_start(args, fmt);
Jens Axboe64565912008-05-28 14:45:33 +0200164 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +0200165 va_end(args);
166
Jens Axboe64565912008-05-28 14:45:33 +0200167 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +0200168 local_irq_restore(flags);
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +0200169}
170EXPORT_SYMBOL_GPL(__trace_note_message);
171
Jens Axboe2056a782006-03-23 20:00:26 +0100172static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
173 pid_t pid)
174{
175 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
176 return 1;
Shawn Dud0deef52009-04-14 13:58:56 +0800177 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
Jens Axboe2056a782006-03-23 20:00:26 +0100178 return 1;
179 if (bt->pid && pid != bt->pid)
180 return 1;
181
182 return 0;
183}
184
185/*
186 * Data direction bit lookup
187 */
Li Zefane4955c92009-03-24 16:04:37 +0800188static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
189 BLK_TC_ACT(BLK_TC_WRITE) };
Jens Axboe2056a782006-03-23 20:00:26 +0100190
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200191#define BLK_TC_RAHEAD BLK_TC_AHEAD
192
David Woodhouse35ba8f72008-08-10 12:33:00 +0100193/* The ilog2() calls fall out because they're constant */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200194#define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
195 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
Jens Axboe2056a782006-03-23 20:00:26 +0100196
197/*
198 * The worker for the various blk_add_trace*() types. Fills out a
199 * blk_io_trace structure and places it in a per-cpu subbuffer.
200 */
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100201static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
Jens Axboe2056a782006-03-23 20:00:26 +0100202 int rw, u32 what, int error, int pdu_len, void *pdu_data)
203{
204 struct task_struct *tsk = current;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200205 struct ring_buffer_event *event = NULL;
Steven Rostedte77405a2009-09-02 14:17:06 -0400206 struct ring_buffer *buffer = NULL;
Jens Axboe2056a782006-03-23 20:00:26 +0100207 struct blk_io_trace *t;
Arnaldo Carvalho de Melo0a987752009-02-05 16:12:56 -0200208 unsigned long flags = 0;
Jens Axboe2056a782006-03-23 20:00:26 +0100209 unsigned long *sequence;
210 pid_t pid;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200211 int cpu, pc = 0;
Li Zefan18cea452009-03-27 10:21:54 +0800212 bool blk_tracer = blk_tracer_enabled;
Jens Axboe2056a782006-03-23 20:00:26 +0100213
Li Zefan18cea452009-03-27 10:21:54 +0800214 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
Jens Axboe2056a782006-03-23 20:00:26 +0100215 return;
216
217 what |= ddir_act[rw & WRITE];
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200218 what |= MASK_TC_BIT(rw, SYNC);
219 what |= MASK_TC_BIT(rw, RAHEAD);
David Woodhouse35ba8f72008-08-10 12:33:00 +0100220 what |= MASK_TC_BIT(rw, META);
221 what |= MASK_TC_BIT(rw, DISCARD);
Namhyung Kimc09c47c2011-08-11 10:36:05 +0200222 what |= MASK_TC_BIT(rw, FLUSH);
223 what |= MASK_TC_BIT(rw, FUA);
Jens Axboe2056a782006-03-23 20:00:26 +0100224
225 pid = tsk->pid;
Shawn Dud0deef52009-04-14 13:58:56 +0800226 if (act_log_check(bt, what, sector, pid))
Jens Axboe2056a782006-03-23 20:00:26 +0100227 return;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200228 cpu = raw_smp_processor_id();
229
Li Zefan18cea452009-03-27 10:21:54 +0800230 if (blk_tracer) {
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200231 tracing_record_cmdline(current);
232
Steven Rostedt (Red Hat)12883ef2013-03-05 09:24:35 -0500233 buffer = blk_tr->trace_buffer.buffer;
Arnaldo Carvalho de Melo51a763d2009-02-05 16:14:13 -0200234 pc = preempt_count();
Steven Rostedte77405a2009-09-02 14:17:06 -0400235 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
Arnaldo Carvalho de Melo51a763d2009-02-05 16:14:13 -0200236 sizeof(*t) + pdu_len,
237 0, pc);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200238 if (!event)
239 return;
Arnaldo Carvalho de Melo51a763d2009-02-05 16:14:13 -0200240 t = ring_buffer_event_data(event);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200241 goto record_it;
242 }
Jens Axboe2056a782006-03-23 20:00:26 +0100243
Jan Karaa404d552013-09-17 22:30:31 +0200244 if (unlikely(tsk->btrace_seq != blktrace_seq))
245 trace_note_tsk(tsk);
246
Jens Axboe2056a782006-03-23 20:00:26 +0100247 /*
248 * A word about the locking here - we disable interrupts to reserve
249 * some space in the relay per-cpu buffer, to prevent an irq
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +0200250 * from coming in and stepping on our toes.
Jens Axboe2056a782006-03-23 20:00:26 +0100251 */
252 local_irq_save(flags);
Jens Axboe2056a782006-03-23 20:00:26 +0100253 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
254 if (t) {
Jens Axboe2056a782006-03-23 20:00:26 +0100255 sequence = per_cpu_ptr(bt->sequence, cpu);
256
257 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
258 t->sequence = ++(*sequence);
Ingo Molnar2997c8c2008-01-11 13:35:54 +0100259 t->time = ktime_to_ns(ktime_get());
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200260record_it:
Arnaldo Carvalho de Melo08a06b82009-02-02 20:30:40 -0200261 /*
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200262 * These two are not needed in ftrace as they are in the
263 * generic trace_entry, filled by tracing_generic_entry_update,
264 * but for the trace_event->bin() synthesizer benefit we do it
265 * here too.
266 */
267 t->cpu = cpu;
268 t->pid = pid;
Arnaldo Carvalho de Melo08a06b82009-02-02 20:30:40 -0200269
Jens Axboe2056a782006-03-23 20:00:26 +0100270 t->sector = sector;
271 t->bytes = bytes;
272 t->action = what;
Jens Axboe2056a782006-03-23 20:00:26 +0100273 t->device = bt->dev;
Jens Axboe2056a782006-03-23 20:00:26 +0100274 t->error = error;
275 t->pdu_len = pdu_len;
276
277 if (pdu_len)
278 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200279
Li Zefan18cea452009-03-27 10:21:54 +0800280 if (blk_tracer) {
Steven Rostedte77405a2009-09-02 14:17:06 -0400281 trace_buffer_unlock_commit(buffer, event, 0, pc);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -0200282 return;
283 }
Jens Axboe2056a782006-03-23 20:00:26 +0100284 }
285
286 local_irq_restore(flags);
287}
288
Jens Axboe2056a782006-03-23 20:00:26 +0100289static struct dentry *blk_tree_root;
Jens Axboe11a57152008-01-11 13:37:01 +0100290static DEFINE_MUTEX(blk_tree_mutex);
Jens Axboe2056a782006-03-23 20:00:26 +0100291
Li Zefanad5dd542009-03-27 10:20:24 +0800292static void blk_trace_free(struct blk_trace *bt)
Jens Axboe2056a782006-03-23 20:00:26 +0100293{
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200294 debugfs_remove(bt->msg_file);
Jens Axboe2056a782006-03-23 20:00:26 +0100295 debugfs_remove(bt->dropped_file);
Jens Axboef48fc4d2009-01-05 10:17:25 +0100296 relay_close(bt->rchan);
Alan D. Brunelle39cbb602009-08-07 12:01:08 -0400297 debugfs_remove(bt->dir);
Jens Axboe2056a782006-03-23 20:00:26 +0100298 free_percpu(bt->sequence);
Jens Axboe64565912008-05-28 14:45:33 +0200299 free_percpu(bt->msg_data);
Jens Axboe2056a782006-03-23 20:00:26 +0100300 kfree(bt);
Li Zefanad5dd542009-03-27 10:20:24 +0800301}
302
303static void blk_trace_cleanup(struct blk_trace *bt)
304{
305 blk_trace_free(bt);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100306 if (atomic_dec_and_test(&blk_probes_ref))
307 blk_unregister_tracepoints();
Jens Axboe2056a782006-03-23 20:00:26 +0100308}
309
Christof Schmitt6da127a2008-01-11 10:09:43 +0100310int blk_trace_remove(struct request_queue *q)
Jens Axboe2056a782006-03-23 20:00:26 +0100311{
312 struct blk_trace *bt;
313
314 bt = xchg(&q->blk_trace, NULL);
315 if (!bt)
316 return -EINVAL;
317
Li Zefan55547202009-03-25 17:21:26 +0800318 if (bt->trace_state != Blktrace_running)
Jens Axboe2056a782006-03-23 20:00:26 +0100319 blk_trace_cleanup(bt);
320
321 return 0;
322}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100323EXPORT_SYMBOL_GPL(blk_trace_remove);
Jens Axboe2056a782006-03-23 20:00:26 +0100324
Jens Axboe2056a782006-03-23 20:00:26 +0100325static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
326 size_t count, loff_t *ppos)
327{
328 struct blk_trace *bt = filp->private_data;
329 char buf[16];
330
331 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
332
333 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
334}
335
Arjan van de Ven2b8693c2007-02-12 00:55:32 -0800336static const struct file_operations blk_dropped_fops = {
Jens Axboe2056a782006-03-23 20:00:26 +0100337 .owner = THIS_MODULE,
Stephen Boyd234e3402012-04-05 14:25:11 -0700338 .open = simple_open,
Jens Axboe2056a782006-03-23 20:00:26 +0100339 .read = blk_dropped_read,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200340 .llseek = default_llseek,
Jens Axboe2056a782006-03-23 20:00:26 +0100341};
342
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200343static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
344 size_t count, loff_t *ppos)
345{
346 char *msg;
347 struct blk_trace *bt;
348
Li Zefan7635b032009-04-03 15:31:34 +0800349 if (count >= BLK_TN_MAX_MSG)
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200350 return -EINVAL;
351
Carl Henrik Lundea4b3ada2009-04-03 14:27:15 +0200352 msg = kmalloc(count + 1, GFP_KERNEL);
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200353 if (msg == NULL)
354 return -ENOMEM;
355
356 if (copy_from_user(msg, buffer, count)) {
357 kfree(msg);
358 return -EFAULT;
359 }
360
Carl Henrik Lundea4b3ada2009-04-03 14:27:15 +0200361 msg[count] = '\0';
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200362 bt = filp->private_data;
363 __trace_note_message(bt, "%s", msg);
364 kfree(msg);
365
366 return count;
367}
368
369static const struct file_operations blk_msg_fops = {
370 .owner = THIS_MODULE,
Stephen Boyd234e3402012-04-05 14:25:11 -0700371 .open = simple_open,
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200372 .write = blk_msg_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200373 .llseek = noop_llseek,
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200374};
375
Jens Axboe2056a782006-03-23 20:00:26 +0100376/*
377 * Keep track of how many times we encountered a full subbuffer, to aid
378 * the user space app in telling how many lost events there were.
379 */
380static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
381 void *prev_subbuf, size_t prev_padding)
382{
383 struct blk_trace *bt;
384
385 if (!relay_buf_full(buf))
386 return 1;
387
388 bt = buf->chan->private_data;
389 atomic_inc(&bt->dropped);
390 return 0;
391}
392
393static int blk_remove_buf_file_callback(struct dentry *dentry)
394{
395 debugfs_remove(dentry);
Jens Axboef48fc4d2009-01-05 10:17:25 +0100396
Jens Axboe2056a782006-03-23 20:00:26 +0100397 return 0;
398}
399
400static struct dentry *blk_create_buf_file_callback(const char *filename,
401 struct dentry *parent,
Al Virof4ae40a2011-07-24 04:33:43 -0400402 umode_t mode,
Jens Axboe2056a782006-03-23 20:00:26 +0100403 struct rchan_buf *buf,
404 int *is_global)
405{
406 return debugfs_create_file(filename, mode, parent, buf,
407 &relay_file_operations);
408}
409
410static struct rchan_callbacks blk_relay_callbacks = {
411 .subbuf_start = blk_subbuf_start_callback,
412 .create_buf_file = blk_create_buf_file_callback,
413 .remove_buf_file = blk_remove_buf_file_callback,
414};
415
Li Zefan9908c302009-04-14 13:59:34 +0800416static void blk_trace_setup_lba(struct blk_trace *bt,
417 struct block_device *bdev)
418{
419 struct hd_struct *part = NULL;
420
421 if (bdev)
422 part = bdev->bd_part;
423
424 if (part) {
425 bt->start_lba = part->start_sect;
426 bt->end_lba = part->start_sect + part->nr_sects;
427 } else {
428 bt->start_lba = 0;
429 bt->end_lba = -1ULL;
430 }
431}
432
Jens Axboe2056a782006-03-23 20:00:26 +0100433/*
434 * Setup everything required to start tracing
435 */
Christof Schmitt6da127a2008-01-11 10:09:43 +0100436int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
Shawn Dud0deef52009-04-14 13:58:56 +0800437 struct block_device *bdev,
438 struct blk_user_trace_setup *buts)
Jens Axboe2056a782006-03-23 20:00:26 +0100439{
Jens Axboe2056a782006-03-23 20:00:26 +0100440 struct blk_trace *old_bt, *bt = NULL;
441 struct dentry *dir = NULL;
Jens Axboe2056a782006-03-23 20:00:26 +0100442 int ret, i;
443
Arnd Bergmann171044d42007-10-09 13:23:53 +0200444 if (!buts->buf_size || !buts->buf_nr)
Jens Axboe2056a782006-03-23 20:00:26 +0100445 return -EINVAL;
446
Jens Axboe0497b342008-10-01 16:16:25 +0200447 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
448 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
Jens Axboe2056a782006-03-23 20:00:26 +0100449
450 /*
451 * some device names have larger paths - convert the slashes
452 * to underscores for this to work as expected
453 */
Arnd Bergmann171044d42007-10-09 13:23:53 +0200454 for (i = 0; i < strlen(buts->name); i++)
455 if (buts->name[i] == '/')
456 buts->name[i] = '_';
Jens Axboe2056a782006-03-23 20:00:26 +0100457
Jens Axboe2056a782006-03-23 20:00:26 +0100458 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
459 if (!bt)
Li Zefanad5dd542009-03-27 10:20:24 +0800460 return -ENOMEM;
Jens Axboe2056a782006-03-23 20:00:26 +0100461
Li Zefanad5dd542009-03-27 10:20:24 +0800462 ret = -ENOMEM;
Jens Axboe2056a782006-03-23 20:00:26 +0100463 bt->sequence = alloc_percpu(unsigned long);
464 if (!bt->sequence)
465 goto err;
466
Ingo Molnarf0ef0392009-03-06 16:44:14 +0100467 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
Jens Axboe64565912008-05-28 14:45:33 +0200468 if (!bt->msg_data)
469 goto err;
470
Jens Axboe2056a782006-03-23 20:00:26 +0100471 ret = -ENOENT;
Jens Axboef48fc4d2009-01-05 10:17:25 +0100472
Li Zefanb5230b52009-03-25 17:19:33 +0800473 mutex_lock(&blk_tree_mutex);
Jens Axboef48fc4d2009-01-05 10:17:25 +0100474 if (!blk_tree_root) {
475 blk_tree_root = debugfs_create_dir("block", NULL);
Li Zefanb5230b52009-03-25 17:19:33 +0800476 if (!blk_tree_root) {
477 mutex_unlock(&blk_tree_mutex);
Li Zefan1a176622009-03-20 09:47:30 +0800478 goto err;
Li Zefanb5230b52009-03-25 17:19:33 +0800479 }
Jens Axboef48fc4d2009-01-05 10:17:25 +0100480 }
Li Zefanb5230b52009-03-25 17:19:33 +0800481 mutex_unlock(&blk_tree_mutex);
Jens Axboef48fc4d2009-01-05 10:17:25 +0100482
483 dir = debugfs_create_dir(buts->name, blk_tree_root);
484
Jens Axboe2056a782006-03-23 20:00:26 +0100485 if (!dir)
486 goto err;
487
488 bt->dir = dir;
Christof Schmitt6da127a2008-01-11 10:09:43 +0100489 bt->dev = dev;
Jens Axboe2056a782006-03-23 20:00:26 +0100490 atomic_set(&bt->dropped, 0);
Jan Karaa404d552013-09-17 22:30:31 +0200491 INIT_LIST_HEAD(&bt->running_list);
Jens Axboe2056a782006-03-23 20:00:26 +0100492
493 ret = -EIO;
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200494 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
495 &blk_dropped_fops);
Jens Axboe2056a782006-03-23 20:00:26 +0100496 if (!bt->dropped_file)
497 goto err;
498
Alan D. Brunelle02c62302008-06-11 09:12:52 +0200499 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
500 if (!bt->msg_file)
501 goto err;
502
Arnd Bergmann171044d42007-10-09 13:23:53 +0200503 bt->rchan = relay_open("trace", dir, buts->buf_size,
504 buts->buf_nr, &blk_relay_callbacks, bt);
Jens Axboe2056a782006-03-23 20:00:26 +0100505 if (!bt->rchan)
506 goto err;
Jens Axboe2056a782006-03-23 20:00:26 +0100507
Arnd Bergmann171044d42007-10-09 13:23:53 +0200508 bt->act_mask = buts->act_mask;
Jens Axboe2056a782006-03-23 20:00:26 +0100509 if (!bt->act_mask)
510 bt->act_mask = (u16) -1;
511
Li Zefan9908c302009-04-14 13:59:34 +0800512 blk_trace_setup_lba(bt, bdev);
Jens Axboe2056a782006-03-23 20:00:26 +0100513
Shawn Dud0deef52009-04-14 13:58:56 +0800514 /* overwrite with user settings */
515 if (buts->start_lba)
516 bt->start_lba = buts->start_lba;
517 if (buts->end_lba)
518 bt->end_lba = buts->end_lba;
519
Arnd Bergmann171044d42007-10-09 13:23:53 +0200520 bt->pid = buts->pid;
Jens Axboe2056a782006-03-23 20:00:26 +0100521 bt->trace_state = Blktrace_setup;
522
523 ret = -EBUSY;
524 old_bt = xchg(&q->blk_trace, bt);
525 if (old_bt) {
526 (void) xchg(&q->blk_trace, old_bt);
527 goto err;
528 }
529
Li Zefan17ba97e2009-03-27 10:20:09 +0800530 if (atomic_inc_return(&blk_probes_ref) == 1)
Li Zefancbe28292009-03-20 09:48:47 +0800531 blk_register_tracepoints();
532
Jens Axboe2056a782006-03-23 20:00:26 +0100533 return 0;
534err:
Li Zefanad5dd542009-03-27 10:20:24 +0800535 blk_trace_free(bt);
Jens Axboe2056a782006-03-23 20:00:26 +0100536 return ret;
537}
538
Christof Schmitt6da127a2008-01-11 10:09:43 +0100539int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
Shawn Dud0deef52009-04-14 13:58:56 +0800540 struct block_device *bdev,
Christof Schmitt6da127a2008-01-11 10:09:43 +0100541 char __user *arg)
Arnd Bergmann171044d42007-10-09 13:23:53 +0200542{
543 struct blk_user_trace_setup buts;
544 int ret;
545
546 ret = copy_from_user(&buts, arg, sizeof(buts));
547 if (ret)
548 return -EFAULT;
549
Shawn Dud0deef52009-04-14 13:58:56 +0800550 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
Arnd Bergmann171044d42007-10-09 13:23:53 +0200551 if (ret)
552 return ret;
553
Dmitry Monakhov9a8c28c2010-02-27 20:35:13 +0300554 if (copy_to_user(arg, &buts, sizeof(buts))) {
555 blk_trace_remove(q);
Arnd Bergmann171044d42007-10-09 13:23:53 +0200556 return -EFAULT;
Dmitry Monakhov9a8c28c2010-02-27 20:35:13 +0300557 }
Arnd Bergmann171044d42007-10-09 13:23:53 +0200558 return 0;
559}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100560EXPORT_SYMBOL_GPL(blk_trace_setup);
Arnd Bergmann171044d42007-10-09 13:23:53 +0200561
Arnd Bergmann62c2a7d2010-07-07 16:51:26 +0200562#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
563static int compat_blk_trace_setup(struct request_queue *q, char *name,
564 dev_t dev, struct block_device *bdev,
565 char __user *arg)
566{
567 struct blk_user_trace_setup buts;
568 struct compat_blk_user_trace_setup cbuts;
569 int ret;
570
571 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
572 return -EFAULT;
573
574 buts = (struct blk_user_trace_setup) {
575 .act_mask = cbuts.act_mask,
576 .buf_size = cbuts.buf_size,
577 .buf_nr = cbuts.buf_nr,
578 .start_lba = cbuts.start_lba,
579 .end_lba = cbuts.end_lba,
580 .pid = cbuts.pid,
581 };
Arnd Bergmann62c2a7d2010-07-07 16:51:26 +0200582
583 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
584 if (ret)
585 return ret;
586
Chen Gangf8c5e942013-11-03 22:23:39 +0800587 if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
Arnd Bergmann62c2a7d2010-07-07 16:51:26 +0200588 blk_trace_remove(q);
589 return -EFAULT;
590 }
591
592 return 0;
593}
594#endif
595
Christof Schmitt6da127a2008-01-11 10:09:43 +0100596int blk_trace_startstop(struct request_queue *q, int start)
Jens Axboe2056a782006-03-23 20:00:26 +0100597{
Jens Axboe2056a782006-03-23 20:00:26 +0100598 int ret;
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200599 struct blk_trace *bt = q->blk_trace;
Jens Axboe2056a782006-03-23 20:00:26 +0100600
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200601 if (bt == NULL)
Jens Axboe2056a782006-03-23 20:00:26 +0100602 return -EINVAL;
603
604 /*
605 * For starting a trace, we can transition from a setup or stopped
606 * trace. For stopping a trace, the state must be running
607 */
608 ret = -EINVAL;
609 if (start) {
610 if (bt->trace_state == Blktrace_setup ||
611 bt->trace_state == Blktrace_stopped) {
612 blktrace_seq++;
613 smp_mb();
614 bt->trace_state = Blktrace_running;
Jan Karaa404d552013-09-17 22:30:31 +0200615 spin_lock_irq(&running_trace_lock);
616 list_add(&bt->running_list, &running_trace_list);
617 spin_unlock_irq(&running_trace_lock);
Olaf Kirchbe1c6342006-12-01 10:39:12 +0100618
619 trace_note_time(bt);
Jens Axboe2056a782006-03-23 20:00:26 +0100620 ret = 0;
621 }
622 } else {
623 if (bt->trace_state == Blktrace_running) {
624 bt->trace_state = Blktrace_stopped;
Jan Karaa404d552013-09-17 22:30:31 +0200625 spin_lock_irq(&running_trace_lock);
626 list_del_init(&bt->running_list);
627 spin_unlock_irq(&running_trace_lock);
Jens Axboe2056a782006-03-23 20:00:26 +0100628 relay_flush(bt->rchan);
629 ret = 0;
630 }
631 }
632
633 return ret;
634}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100635EXPORT_SYMBOL_GPL(blk_trace_startstop);
Jens Axboe2056a782006-03-23 20:00:26 +0100636
637/**
638 * blk_trace_ioctl: - handle the ioctls associated with tracing
639 * @bdev: the block device
Steven Rostedtef180122009-03-10 14:10:56 -0400640 * @cmd: the ioctl cmd
Jens Axboe2056a782006-03-23 20:00:26 +0100641 * @arg: the argument data, if any
642 *
643 **/
644int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
645{
Jens Axboe165125e2007-07-24 09:28:11 +0200646 struct request_queue *q;
Jens Axboe2056a782006-03-23 20:00:26 +0100647 int ret, start = 0;
Christof Schmitt6da127a2008-01-11 10:09:43 +0100648 char b[BDEVNAME_SIZE];
Jens Axboe2056a782006-03-23 20:00:26 +0100649
650 q = bdev_get_queue(bdev);
651 if (!q)
652 return -ENXIO;
653
654 mutex_lock(&bdev->bd_mutex);
655
656 switch (cmd) {
657 case BLKTRACESETUP:
Jean Delvaref36f21e2008-05-12 14:02:33 -0700658 bdevname(bdev, b);
Shawn Dud0deef52009-04-14 13:58:56 +0800659 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
Jens Axboe2056a782006-03-23 20:00:26 +0100660 break;
Arnd Bergmann62c2a7d2010-07-07 16:51:26 +0200661#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
662 case BLKTRACESETUP32:
663 bdevname(bdev, b);
664 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
665 break;
666#endif
Jens Axboe2056a782006-03-23 20:00:26 +0100667 case BLKTRACESTART:
668 start = 1;
669 case BLKTRACESTOP:
670 ret = blk_trace_startstop(q, start);
671 break;
672 case BLKTRACETEARDOWN:
673 ret = blk_trace_remove(q);
674 break;
675 default:
676 ret = -ENOTTY;
677 break;
678 }
679
680 mutex_unlock(&bdev->bd_mutex);
681 return ret;
682}
683
684/**
685 * blk_trace_shutdown: - stop and cleanup trace structures
686 * @q: the request queue associated with the device
687 *
688 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200689void blk_trace_shutdown(struct request_queue *q)
Jens Axboe2056a782006-03-23 20:00:26 +0100690{
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -0700691 if (q->blk_trace) {
692 blk_trace_startstop(q, 0);
693 blk_trace_remove(q);
694 }
Jens Axboe2056a782006-03-23 20:00:26 +0100695}
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100696
697/*
698 * blktrace probes
699 */
700
701/**
702 * blk_add_trace_rq - Add a trace for a request oriented action
703 * @q: queue the io is for
704 * @rq: the source request
Roman Penaf5040d2014-03-04 23:13:10 +0900705 * @nr_bytes: number of completed bytes
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100706 * @what: the action
707 *
708 * Description:
709 * Records an action against a request. Will log the bio offset + size.
710 *
711 **/
712static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
Roman Penaf5040d2014-03-04 23:13:10 +0900713 unsigned int nr_bytes, u32 what)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100714{
715 struct blk_trace *bt = q->blk_trace;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100716
717 if (likely(!bt))
718 return;
719
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200720 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100721 what |= BLK_TC_ACT(BLK_TC_PC);
Roman Penaf5040d2014-03-04 23:13:10 +0900722 __blk_add_trace(bt, 0, nr_bytes, rq->cmd_flags,
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900723 what, rq->errors, rq->cmd_len, rq->cmd);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100724 } else {
725 what |= BLK_TC_ACT(BLK_TC_FS);
Roman Penaf5040d2014-03-04 23:13:10 +0900726 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes,
Tao Ma805f6b52011-03-11 20:11:59 +0100727 rq->cmd_flags, what, rq->errors, 0, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100728 }
729}
730
Steven Rostedt38516ab2010-04-20 17:04:50 -0400731static void blk_add_trace_rq_abort(void *ignore,
732 struct request_queue *q, struct request *rq)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100733{
Roman Penaf5040d2014-03-04 23:13:10 +0900734 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100735}
736
Steven Rostedt38516ab2010-04-20 17:04:50 -0400737static void blk_add_trace_rq_insert(void *ignore,
738 struct request_queue *q, struct request *rq)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100739{
Roman Penaf5040d2014-03-04 23:13:10 +0900740 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100741}
742
Steven Rostedt38516ab2010-04-20 17:04:50 -0400743static void blk_add_trace_rq_issue(void *ignore,
744 struct request_queue *q, struct request *rq)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100745{
Roman Penaf5040d2014-03-04 23:13:10 +0900746 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100747}
748
Steven Rostedt38516ab2010-04-20 17:04:50 -0400749static void blk_add_trace_rq_requeue(void *ignore,
750 struct request_queue *q,
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200751 struct request *rq)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100752{
Roman Penaf5040d2014-03-04 23:13:10 +0900753 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100754}
755
Steven Rostedt38516ab2010-04-20 17:04:50 -0400756static void blk_add_trace_rq_complete(void *ignore,
757 struct request_queue *q,
Roman Penaf5040d2014-03-04 23:13:10 +0900758 struct request *rq,
759 unsigned int nr_bytes)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100760{
Roman Penaf5040d2014-03-04 23:13:10 +0900761 blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100762}
763
764/**
765 * blk_add_trace_bio - Add a trace for a bio oriented action
766 * @q: queue the io is for
767 * @bio: the source bio
768 * @what: the action
Jens Axboe797a4552011-01-10 09:06:44 +0100769 * @error: error, if any
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100770 *
771 * Description:
772 * Records an action against a bio. Will log the bio offset + size.
773 *
774 **/
775static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
Jens Axboe797a4552011-01-10 09:06:44 +0100776 u32 what, int error)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100777{
778 struct blk_trace *bt = q->blk_trace;
779
780 if (likely(!bt))
781 return;
782
Jens Axboe797a4552011-01-10 09:06:44 +0100783 if (!error && !bio_flagged(bio, BIO_UPTODATE))
784 error = EIO;
785
Kent Overstreet4f024f32013-10-11 15:44:27 -0700786 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
787 bio->bi_rw, what, error, 0, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100788}
789
Steven Rostedt38516ab2010-04-20 17:04:50 -0400790static void blk_add_trace_bio_bounce(void *ignore,
791 struct request_queue *q, struct bio *bio)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100792{
Jens Axboe797a4552011-01-10 09:06:44 +0100793 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100794}
795
Linus Torvalds0a82a8d2013-04-18 09:00:26 -0700796static void blk_add_trace_bio_complete(void *ignore,
797 struct request_queue *q, struct bio *bio,
798 int error)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100799{
Jens Axboe797a4552011-01-10 09:06:44 +0100800 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100801}
802
Steven Rostedt38516ab2010-04-20 17:04:50 -0400803static void blk_add_trace_bio_backmerge(void *ignore,
804 struct request_queue *q,
Tejun Heo8c1cf6b2013-01-11 13:06:34 -0800805 struct request *rq,
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200806 struct bio *bio)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100807{
Jens Axboe797a4552011-01-10 09:06:44 +0100808 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100809}
810
Steven Rostedt38516ab2010-04-20 17:04:50 -0400811static void blk_add_trace_bio_frontmerge(void *ignore,
812 struct request_queue *q,
Tejun Heo8c1cf6b2013-01-11 13:06:34 -0800813 struct request *rq,
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200814 struct bio *bio)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100815{
Jens Axboe797a4552011-01-10 09:06:44 +0100816 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100817}
818
Steven Rostedt38516ab2010-04-20 17:04:50 -0400819static void blk_add_trace_bio_queue(void *ignore,
820 struct request_queue *q, struct bio *bio)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100821{
Jens Axboe797a4552011-01-10 09:06:44 +0100822 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100823}
824
Steven Rostedt38516ab2010-04-20 17:04:50 -0400825static void blk_add_trace_getrq(void *ignore,
826 struct request_queue *q,
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200827 struct bio *bio, int rw)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100828{
829 if (bio)
Jens Axboe797a4552011-01-10 09:06:44 +0100830 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100831 else {
832 struct blk_trace *bt = q->blk_trace;
833
834 if (bt)
835 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
836 }
837}
838
839
Steven Rostedt38516ab2010-04-20 17:04:50 -0400840static void blk_add_trace_sleeprq(void *ignore,
841 struct request_queue *q,
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200842 struct bio *bio, int rw)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100843{
844 if (bio)
Jens Axboe797a4552011-01-10 09:06:44 +0100845 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100846 else {
847 struct blk_trace *bt = q->blk_trace;
848
849 if (bt)
Arnaldo Carvalho de Melo939b3662009-02-03 11:58:29 -0200850 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
851 0, 0, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100852 }
853}
854
Steven Rostedt38516ab2010-04-20 17:04:50 -0400855static void blk_add_trace_plug(void *ignore, struct request_queue *q)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100856{
857 struct blk_trace *bt = q->blk_trace;
858
859 if (bt)
860 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
861}
862
Jens Axboe49cac012011-04-16 13:51:05 +0200863static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
864 unsigned int depth, bool explicit)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100865{
866 struct blk_trace *bt = q->blk_trace;
867
868 if (bt) {
Jens Axboe94b5eb22011-04-12 10:12:19 +0200869 __be64 rpdu = cpu_to_be64(depth);
Jens Axboe49cac012011-04-16 13:51:05 +0200870 u32 what;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100871
Jens Axboe49cac012011-04-16 13:51:05 +0200872 if (explicit)
873 what = BLK_TA_UNPLUG_IO;
874 else
875 what = BLK_TA_UNPLUG_TIMER;
876
877 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100878 }
879}
880
Steven Rostedt38516ab2010-04-20 17:04:50 -0400881static void blk_add_trace_split(void *ignore,
882 struct request_queue *q, struct bio *bio,
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100883 unsigned int pdu)
884{
885 struct blk_trace *bt = q->blk_trace;
886
887 if (bt) {
888 __be64 rpdu = cpu_to_be64(pdu);
889
Kent Overstreet4f024f32013-10-11 15:44:27 -0700890 __blk_add_trace(bt, bio->bi_iter.bi_sector,
891 bio->bi_iter.bi_size, bio->bi_rw, BLK_TA_SPLIT,
892 !bio_flagged(bio, BIO_UPTODATE),
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100893 sizeof(rpdu), &rpdu);
894 }
895}
896
897/**
Mike Snitzerd07335e2010-11-16 12:52:38 +0100898 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
Randy Dunlap546cf442010-05-29 11:45:07 -0700899 * @ignore: trace callback data parameter (not used)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100900 * @q: queue the io is for
901 * @bio: the source bio
902 * @dev: target device
Alan D. Brunellea42aaa32009-05-04 16:27:26 -0400903 * @from: source sector
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100904 *
905 * Description:
906 * Device mapper or raid target sometimes need to split a bio because
907 * it spans a stripe (or similar). Add a trace for that action.
908 *
909 **/
Mike Snitzerd07335e2010-11-16 12:52:38 +0100910static void blk_add_trace_bio_remap(void *ignore,
911 struct request_queue *q, struct bio *bio,
912 dev_t dev, sector_t from)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100913{
914 struct blk_trace *bt = q->blk_trace;
915 struct blk_io_trace_remap r;
916
917 if (likely(!bt))
918 return;
919
Alan D. Brunellea42aaa32009-05-04 16:27:26 -0400920 r.device_from = cpu_to_be32(dev);
921 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
922 r.sector_from = cpu_to_be64(from);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100923
Kent Overstreet4f024f32013-10-11 15:44:27 -0700924 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
925 bio->bi_rw, BLK_TA_REMAP,
926 !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100927}
928
929/**
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +0200930 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
Randy Dunlap546cf442010-05-29 11:45:07 -0700931 * @ignore: trace callback data parameter (not used)
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +0200932 * @q: queue the io is for
933 * @rq: the source request
934 * @dev: target device
935 * @from: source sector
936 *
937 * Description:
938 * Device mapper remaps request to other devices.
939 * Add a trace for that action.
940 *
941 **/
Steven Rostedt38516ab2010-04-20 17:04:50 -0400942static void blk_add_trace_rq_remap(void *ignore,
943 struct request_queue *q,
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +0200944 struct request *rq, dev_t dev,
945 sector_t from)
946{
947 struct blk_trace *bt = q->blk_trace;
948 struct blk_io_trace_remap r;
949
950 if (likely(!bt))
951 return;
952
953 r.device_from = cpu_to_be32(dev);
954 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
955 r.sector_from = cpu_to_be64(from);
956
957 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
958 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
959 sizeof(r), &r);
960}
961
962/**
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100963 * blk_add_driver_data - Add binary message with driver-specific data
964 * @q: queue the io is for
965 * @rq: io request
966 * @data: driver-specific data
967 * @len: length of driver-specific data
968 *
969 * Description:
970 * Some drivers might want to write driver-specific data per request.
971 *
972 **/
973void blk_add_driver_data(struct request_queue *q,
974 struct request *rq,
975 void *data, size_t len)
976{
977 struct blk_trace *bt = q->blk_trace;
978
979 if (likely(!bt))
980 return;
981
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200982 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900983 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
984 BLK_TA_DRV_DATA, rq->errors, len, data);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100985 else
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900986 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
987 BLK_TA_DRV_DATA, rq->errors, len, data);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100988}
989EXPORT_SYMBOL_GPL(blk_add_driver_data);
990
Li Zefan3c289ba2009-03-20 09:48:26 +0800991static void blk_register_tracepoints(void)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100992{
993 int ret;
994
Steven Rostedt38516ab2010-04-20 17:04:50 -0400995 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100996 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -0400997 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100998 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -0400999 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001000 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001001 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001002 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001003 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001004 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001005 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001006 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001007 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001008 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001009 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001010 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001011 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001012 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001013 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001014 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001015 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001016 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001017 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001018 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001019 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001020 WARN_ON(ret);
Jens Axboe49cac012011-04-16 13:51:05 +02001021 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001022 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001023 ret = register_trace_block_split(blk_add_trace_split, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001024 WARN_ON(ret);
Mike Snitzerd07335e2010-11-16 12:52:38 +01001025 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001026 WARN_ON(ret);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001027 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +02001028 WARN_ON(ret);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001029}
1030
1031static void blk_unregister_tracepoints(void)
1032{
Steven Rostedt38516ab2010-04-20 17:04:50 -04001033 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
Mike Snitzerd07335e2010-11-16 12:52:38 +01001034 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001035 unregister_trace_block_split(blk_add_trace_split, NULL);
Jens Axboe49cac012011-04-16 13:51:05 +02001036 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
Steven Rostedt38516ab2010-04-20 17:04:50 -04001037 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1038 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1039 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1040 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1041 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1042 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1043 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1044 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1045 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1046 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1047 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1048 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1049 unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001050
1051 tracepoint_synchronize_unregister();
1052}
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001053
1054/*
1055 * struct blk_io_tracer formatting routines
1056 */
1057
1058static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1059{
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001060 int i = 0;
Li Zefan65796342009-03-24 16:05:06 +08001061 int tc = t->action >> BLK_TC_SHIFT;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001062
Li Zefan18cea452009-03-27 10:21:54 +08001063 if (t->action == BLK_TN_MESSAGE) {
1064 rwbs[i++] = 'N';
1065 goto out;
1066 }
1067
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001068 if (tc & BLK_TC_FLUSH)
1069 rwbs[i++] = 'F';
1070
Li Zefan65796342009-03-24 16:05:06 +08001071 if (tc & BLK_TC_DISCARD)
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001072 rwbs[i++] = 'D';
Li Zefan65796342009-03-24 16:05:06 +08001073 else if (tc & BLK_TC_WRITE)
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001074 rwbs[i++] = 'W';
1075 else if (t->bytes)
1076 rwbs[i++] = 'R';
1077 else
1078 rwbs[i++] = 'N';
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001079
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001080 if (tc & BLK_TC_FUA)
1081 rwbs[i++] = 'F';
Li Zefan65796342009-03-24 16:05:06 +08001082 if (tc & BLK_TC_AHEAD)
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001083 rwbs[i++] = 'A';
Li Zefan65796342009-03-24 16:05:06 +08001084 if (tc & BLK_TC_SYNC)
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001085 rwbs[i++] = 'S';
Li Zefan65796342009-03-24 16:05:06 +08001086 if (tc & BLK_TC_META)
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001087 rwbs[i++] = 'M';
Li Zefan18cea452009-03-27 10:21:54 +08001088out:
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001089 rwbs[i] = '\0';
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001090}
1091
1092static inline
1093const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1094{
1095 return (const struct blk_io_trace *)ent;
1096}
1097
1098static inline const void *pdu_start(const struct trace_entry *ent)
1099{
1100 return te_blk_io_trace(ent) + 1;
1101}
1102
Li Zefan66de7792009-04-01 16:19:19 +08001103static inline u32 t_action(const struct trace_entry *ent)
1104{
1105 return te_blk_io_trace(ent)->action;
1106}
1107
1108static inline u32 t_bytes(const struct trace_entry *ent)
1109{
1110 return te_blk_io_trace(ent)->bytes;
1111}
1112
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001113static inline u32 t_sec(const struct trace_entry *ent)
1114{
1115 return te_blk_io_trace(ent)->bytes >> 9;
1116}
1117
1118static inline unsigned long long t_sector(const struct trace_entry *ent)
1119{
1120 return te_blk_io_trace(ent)->sector;
1121}
1122
1123static inline __u16 t_error(const struct trace_entry *ent)
1124{
Li Zefane0dc81b2009-03-24 16:05:51 +08001125 return te_blk_io_trace(ent)->error;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001126}
1127
1128static __u64 get_pdu_int(const struct trace_entry *ent)
1129{
1130 const __u64 *val = pdu_start(ent);
1131 return be64_to_cpu(*val);
1132}
1133
1134static void get_pdu_remap(const struct trace_entry *ent,
1135 struct blk_io_trace_remap *r)
1136{
1137 const struct blk_io_trace_remap *__r = pdu_start(ent);
Alan D. Brunellea42aaa32009-05-04 16:27:26 -04001138 __u64 sector_from = __r->sector_from;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001139
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001140 r->device_from = be32_to_cpu(__r->device_from);
Alan D. Brunellea42aaa32009-05-04 16:27:26 -04001141 r->device_to = be32_to_cpu(__r->device_to);
1142 r->sector_from = be64_to_cpu(sector_from);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001143}
1144
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001145typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1146
1147static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001148{
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001149 char rwbs[RWBS_LEN];
Li Zefan35ac51b2009-03-27 10:19:46 +08001150 unsigned long long ts = iter->ts;
1151 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001152 unsigned secs = (unsigned long)ts;
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001153 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001154
1155 fill_rwbs(rwbs, t);
1156
1157 return trace_seq_printf(&iter->seq,
Li Zefan35ac51b2009-03-27 10:19:46 +08001158 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001159 MAJOR(t->device), MINOR(t->device), iter->cpu,
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001160 secs, nsec_rem, iter->ent->pid, act, rwbs);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001161}
1162
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001163static int blk_log_action(struct trace_iterator *iter, const char *act)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001164{
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001165 char rwbs[RWBS_LEN];
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001166 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1167
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001168 fill_rwbs(rwbs, t);
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001169 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001170 MAJOR(t->device), MINOR(t->device), act, rwbs);
1171}
1172
Li Zefan66de7792009-04-01 16:19:19 +08001173static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1174{
Li Zefan04986252009-05-11 14:33:23 +08001175 const unsigned char *pdu_buf;
Li Zefan66de7792009-04-01 16:19:19 +08001176 int pdu_len;
1177 int i, end, ret;
1178
1179 pdu_buf = pdu_start(ent);
1180 pdu_len = te_blk_io_trace(ent)->pdu_len;
1181
1182 if (!pdu_len)
1183 return 1;
1184
1185 /* find the last zero that needs to be printed */
1186 for (end = pdu_len - 1; end >= 0; end--)
1187 if (pdu_buf[end])
1188 break;
1189 end++;
1190
1191 if (!trace_seq_putc(s, '('))
1192 return 0;
1193
1194 for (i = 0; i < pdu_len; i++) {
1195
1196 ret = trace_seq_printf(s, "%s%02x",
1197 i == 0 ? "" : " ", pdu_buf[i]);
1198 if (!ret)
1199 return ret;
1200
1201 /*
1202 * stop when the rest is just zeroes and indicate so
1203 * with a ".." appended
1204 */
1205 if (i == end && end != pdu_len - 1)
1206 return trace_seq_puts(s, " ..) ");
1207 }
1208
1209 return trace_seq_puts(s, ") ");
1210}
1211
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001212static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1213{
Steven Rostedt4ca53082009-03-16 19:20:15 -04001214 char cmd[TASK_COMM_LEN];
1215
1216 trace_find_cmdline(ent->pid, cmd);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001217
Li Zefan66de7792009-04-01 16:19:19 +08001218 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1219 int ret;
1220
1221 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1222 if (!ret)
1223 return 0;
1224 ret = blk_log_dump_pdu(s, ent);
1225 if (!ret)
1226 return 0;
1227 return trace_seq_printf(s, "[%s]\n", cmd);
1228 } else {
1229 if (t_sec(ent))
1230 return trace_seq_printf(s, "%llu + %u [%s]\n",
1231 t_sector(ent), t_sec(ent), cmd);
1232 return trace_seq_printf(s, "[%s]\n", cmd);
1233 }
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001234}
1235
Arnaldo Carvalho de Melo157f9c02009-01-26 15:00:56 -02001236static int blk_log_with_error(struct trace_seq *s,
1237 const struct trace_entry *ent)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001238{
Li Zefan66de7792009-04-01 16:19:19 +08001239 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1240 int ret;
1241
1242 ret = blk_log_dump_pdu(s, ent);
1243 if (ret)
1244 return trace_seq_printf(s, "[%d]\n", t_error(ent));
1245 return 0;
1246 } else {
1247 if (t_sec(ent))
1248 return trace_seq_printf(s, "%llu + %u [%d]\n",
1249 t_sector(ent),
1250 t_sec(ent), t_error(ent));
1251 return trace_seq_printf(s, "%llu [%d]\n",
1252 t_sector(ent), t_error(ent));
1253 }
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001254}
1255
1256static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1257{
Alan D. Brunellea42aaa32009-05-04 16:27:26 -04001258 struct blk_io_trace_remap r = { .device_from = 0, };
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001259
1260 get_pdu_remap(ent, &r);
1261 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
Alan D. Brunellea42aaa32009-05-04 16:27:26 -04001262 t_sector(ent), t_sec(ent),
1263 MAJOR(r.device_from), MINOR(r.device_from),
1264 (unsigned long long)r.sector_from);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001265}
1266
1267static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1268{
Steven Rostedt4ca53082009-03-16 19:20:15 -04001269 char cmd[TASK_COMM_LEN];
1270
1271 trace_find_cmdline(ent->pid, cmd);
1272
1273 return trace_seq_printf(s, "[%s]\n", cmd);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001274}
1275
1276static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1277{
Steven Rostedt4ca53082009-03-16 19:20:15 -04001278 char cmd[TASK_COMM_LEN];
1279
1280 trace_find_cmdline(ent->pid, cmd);
1281
1282 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001283}
1284
1285static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1286{
Steven Rostedt4ca53082009-03-16 19:20:15 -04001287 char cmd[TASK_COMM_LEN];
1288
1289 trace_find_cmdline(ent->pid, cmd);
1290
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001291 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
Steven Rostedt4ca53082009-03-16 19:20:15 -04001292 get_pdu_int(ent), cmd);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001293}
1294
Li Zefan18cea452009-03-27 10:21:54 +08001295static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1296{
1297 int ret;
1298 const struct blk_io_trace *t = te_blk_io_trace(ent);
1299
1300 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1301 if (ret)
1302 return trace_seq_putc(s, '\n');
1303 return ret;
1304}
1305
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001306/*
1307 * struct tracer operations
1308 */
1309
1310static void blk_tracer_print_header(struct seq_file *m)
1311{
1312 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1313 return;
1314 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1315 "# | | | | | |\n");
1316}
1317
1318static void blk_tracer_start(struct trace_array *tr)
1319{
Li Zefanad5dd542009-03-27 10:20:24 +08001320 blk_tracer_enabled = true;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001321}
1322
1323static int blk_tracer_init(struct trace_array *tr)
1324{
1325 blk_tr = tr;
1326 blk_tracer_start(tr);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001327 return 0;
1328}
1329
1330static void blk_tracer_stop(struct trace_array *tr)
1331{
Li Zefanad5dd542009-03-27 10:20:24 +08001332 blk_tracer_enabled = false;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001333}
1334
1335static void blk_tracer_reset(struct trace_array *tr)
1336{
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001337 blk_tracer_stop(tr);
1338}
1339
Li Zefane4955c92009-03-24 16:04:37 +08001340static const struct {
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001341 const char *act[2];
Steven Rostedtef180122009-03-10 14:10:56 -04001342 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
Li Zefane4955c92009-03-24 16:04:37 +08001343} what2act[] = {
Steven Rostedtef180122009-03-10 14:10:56 -04001344 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001345 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1346 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1347 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1348 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1349 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1350 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1351 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1352 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1353 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
Jens Axboe49cac012011-04-16 13:51:05 +02001354 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001355 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1356 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1357 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1358 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1359};
1360
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001361static enum print_line_t print_one_line(struct trace_iterator *iter,
1362 bool classic)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001363{
Arnaldo Carvalho de Melo2c9b2382009-02-02 20:30:12 -02001364 struct trace_seq *s = &iter->seq;
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001365 const struct blk_io_trace *t;
1366 u16 what;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001367 int ret;
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001368 bool long_act;
1369 blk_log_action_t *log_action;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001370
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001371 t = te_blk_io_trace(iter->ent);
1372 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1373 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1374 log_action = classic ? &blk_log_action_classic : &blk_log_action;
Arnaldo Carvalho de Melo08a06b82009-02-02 20:30:40 -02001375
Li Zefan18cea452009-03-27 10:21:54 +08001376 if (t->action == BLK_TN_MESSAGE) {
1377 ret = log_action(iter, long_act ? "message" : "m");
1378 if (ret)
1379 ret = blk_log_msg(s, iter->ent);
1380 goto out;
1381 }
1382
Li Zefaneb08f8e2009-03-24 16:05:27 +08001383 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
Li Zefanb78825d2009-04-01 16:18:53 +08001384 ret = trace_seq_printf(s, "Unknown action %x\n", what);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001385 else {
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001386 ret = log_action(iter, what2act[what].act[long_act]);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001387 if (ret)
Arnaldo Carvalho de Melo2c9b2382009-02-02 20:30:12 -02001388 ret = what2act[what].print(s, iter->ent);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001389 }
Li Zefan18cea452009-03-27 10:21:54 +08001390out:
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001391 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1392}
1393
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001394static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
Steven Rostedta9a57762010-04-22 18:46:14 -04001395 int flags, struct trace_event *event)
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001396{
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001397 return print_one_line(iter, false);
1398}
1399
Arnaldo Carvalho de Melo08a06b82009-02-02 20:30:40 -02001400static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1401{
1402 struct trace_seq *s = &iter->seq;
1403 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1404 const int offset = offsetof(struct blk_io_trace, sector);
1405 struct blk_io_trace old = {
1406 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
Li Zefan6c051ce2009-03-25 17:18:56 +08001407 .time = iter->ts,
Arnaldo Carvalho de Melo08a06b82009-02-02 20:30:40 -02001408 };
1409
1410 if (!trace_seq_putmem(s, &old, offset))
1411 return 0;
1412 return trace_seq_putmem(s, &t->sector,
1413 sizeof(old) - offset + t->pdu_len);
1414}
1415
Arnaldo Carvalho de Meloae7462b2009-02-03 22:05:50 -02001416static enum print_line_t
Steven Rostedta9a57762010-04-22 18:46:14 -04001417blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1418 struct trace_event *event)
Arnaldo Carvalho de Melo08a06b82009-02-02 20:30:40 -02001419{
1420 return blk_trace_synthesize_old_trace(iter) ?
1421 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1422}
1423
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001424static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1425{
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001426 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1427 return TRACE_TYPE_UNHANDLED;
1428
Li Zefanb6a4b0c2009-03-27 10:21:23 +08001429 return print_one_line(iter, true);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001430}
1431
Steven Rostedt (Red Hat)8c1a49a2014-01-10 11:13:54 -05001432static int
1433blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
Li Zefanf3948f82009-04-15 11:02:56 +08001434{
1435 /* don't output context-info for blk_classic output */
1436 if (bit == TRACE_BLK_OPT_CLASSIC) {
1437 if (set)
1438 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1439 else
1440 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1441 }
1442 return 0;
1443}
1444
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001445static struct tracer blk_tracer __read_mostly = {
1446 .name = "blk",
1447 .init = blk_tracer_init,
1448 .reset = blk_tracer_reset,
1449 .start = blk_tracer_start,
1450 .stop = blk_tracer_stop,
1451 .print_header = blk_tracer_print_header,
1452 .print_line = blk_tracer_print_line,
1453 .flags = &blk_tracer_flags,
Li Zefanf3948f82009-04-15 11:02:56 +08001454 .set_flag = blk_tracer_set_flag,
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001455};
1456
Steven Rostedta9a57762010-04-22 18:46:14 -04001457static struct trace_event_functions trace_blk_event_funcs = {
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001458 .trace = blk_trace_event_print,
Arnaldo Carvalho de Melo08a06b82009-02-02 20:30:40 -02001459 .binary = blk_trace_event_print_binary,
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001460};
1461
Steven Rostedta9a57762010-04-22 18:46:14 -04001462static struct trace_event trace_blk_event = {
1463 .type = TRACE_BLK,
1464 .funcs = &trace_blk_event_funcs,
1465};
1466
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001467static int __init init_blk_tracer(void)
1468{
1469 if (!register_ftrace_event(&trace_blk_event)) {
1470 pr_warning("Warning: could not register block events\n");
1471 return 1;
1472 }
1473
1474 if (register_tracer(&blk_tracer) != 0) {
1475 pr_warning("Warning: could not register the block tracer\n");
1476 unregister_ftrace_event(&trace_blk_event);
1477 return 1;
1478 }
1479
1480 return 0;
1481}
1482
1483device_initcall(init_blk_tracer);
1484
1485static int blk_trace_remove_queue(struct request_queue *q)
1486{
1487 struct blk_trace *bt;
1488
1489 bt = xchg(&q->blk_trace, NULL);
1490 if (bt == NULL)
1491 return -EINVAL;
1492
Li Zefan17ba97e2009-03-27 10:20:09 +08001493 if (atomic_dec_and_test(&blk_probes_ref))
1494 blk_unregister_tracepoints();
1495
Jan Karaa404d552013-09-17 22:30:31 +02001496 spin_lock_irq(&running_trace_lock);
1497 list_del(&bt->running_list);
1498 spin_unlock_irq(&running_trace_lock);
Li Zefanad5dd542009-03-27 10:20:24 +08001499 blk_trace_free(bt);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001500 return 0;
1501}
1502
1503/*
1504 * Setup everything required to start tracing
1505 */
Li Zefan9908c302009-04-14 13:59:34 +08001506static int blk_trace_setup_queue(struct request_queue *q,
1507 struct block_device *bdev)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001508{
1509 struct blk_trace *old_bt, *bt = NULL;
Li Zefan18cea452009-03-27 10:21:54 +08001510 int ret = -ENOMEM;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001511
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001512 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1513 if (!bt)
Li Zefan15152e42009-03-20 09:49:08 +08001514 return -ENOMEM;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001515
Li Zefan18cea452009-03-27 10:21:54 +08001516 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1517 if (!bt->msg_data)
1518 goto free_bt;
1519
Li Zefan9908c302009-04-14 13:59:34 +08001520 bt->dev = bdev->bd_dev;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001521 bt->act_mask = (u16)-1;
Li Zefan9908c302009-04-14 13:59:34 +08001522
1523 blk_trace_setup_lba(bt, bdev);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001524
1525 old_bt = xchg(&q->blk_trace, bt);
1526 if (old_bt != NULL) {
1527 (void)xchg(&q->blk_trace, old_bt);
Li Zefan18cea452009-03-27 10:21:54 +08001528 ret = -EBUSY;
1529 goto free_bt;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001530 }
Li Zefan15152e42009-03-20 09:49:08 +08001531
Li Zefan17ba97e2009-03-27 10:20:09 +08001532 if (atomic_inc_return(&blk_probes_ref) == 1)
1533 blk_register_tracepoints();
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001534 return 0;
Li Zefan18cea452009-03-27 10:21:54 +08001535
1536free_bt:
1537 blk_trace_free(bt);
1538 return ret;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001539}
1540
1541/*
1542 * sysfs interface to enable and configure tracing
1543 */
1544
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001545static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1546 struct device_attribute *attr,
1547 char *buf);
1548static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1549 struct device_attribute *attr,
1550 const char *buf, size_t count);
1551#define BLK_TRACE_DEVICE_ATTR(_name) \
1552 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1553 sysfs_blk_trace_attr_show, \
1554 sysfs_blk_trace_attr_store)
1555
Li Zefancd649b82009-03-20 11:33:55 +08001556static BLK_TRACE_DEVICE_ATTR(enable);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001557static BLK_TRACE_DEVICE_ATTR(act_mask);
1558static BLK_TRACE_DEVICE_ATTR(pid);
1559static BLK_TRACE_DEVICE_ATTR(start_lba);
1560static BLK_TRACE_DEVICE_ATTR(end_lba);
1561
1562static struct attribute *blk_trace_attrs[] = {
1563 &dev_attr_enable.attr,
1564 &dev_attr_act_mask.attr,
1565 &dev_attr_pid.attr,
1566 &dev_attr_start_lba.attr,
1567 &dev_attr_end_lba.attr,
1568 NULL
1569};
1570
1571struct attribute_group blk_trace_attr_group = {
1572 .name = "trace",
1573 .attrs = blk_trace_attrs,
1574};
1575
Li Zefan09341992009-03-24 17:43:30 +08001576static const struct {
1577 int mask;
1578 const char *str;
1579} mask_maps[] = {
1580 { BLK_TC_READ, "read" },
1581 { BLK_TC_WRITE, "write" },
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001582 { BLK_TC_FLUSH, "flush" },
Li Zefan09341992009-03-24 17:43:30 +08001583 { BLK_TC_SYNC, "sync" },
1584 { BLK_TC_QUEUE, "queue" },
1585 { BLK_TC_REQUEUE, "requeue" },
1586 { BLK_TC_ISSUE, "issue" },
1587 { BLK_TC_COMPLETE, "complete" },
1588 { BLK_TC_FS, "fs" },
1589 { BLK_TC_PC, "pc" },
1590 { BLK_TC_AHEAD, "ahead" },
1591 { BLK_TC_META, "meta" },
1592 { BLK_TC_DISCARD, "discard" },
1593 { BLK_TC_DRV_DATA, "drv_data" },
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001594 { BLK_TC_FUA, "fua" },
Li Zefan09341992009-03-24 17:43:30 +08001595};
1596
1597static int blk_trace_str2mask(const char *str)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001598{
Li Zefan09341992009-03-24 17:43:30 +08001599 int i;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001600 int mask = 0;
Li Zefan9eb85122009-04-09 11:19:40 +08001601 char *buf, *s, *token;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001602
Li Zefan9eb85122009-04-09 11:19:40 +08001603 buf = kstrdup(str, GFP_KERNEL);
1604 if (buf == NULL)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001605 return -ENOMEM;
Li Zefan9eb85122009-04-09 11:19:40 +08001606 s = strstrip(buf);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001607
1608 while (1) {
Li Zefan09341992009-03-24 17:43:30 +08001609 token = strsep(&s, ",");
1610 if (token == NULL)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001611 break;
1612
Li Zefan09341992009-03-24 17:43:30 +08001613 if (*token == '\0')
1614 continue;
1615
1616 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1617 if (strcasecmp(token, mask_maps[i].str) == 0) {
1618 mask |= mask_maps[i].mask;
1619 break;
1620 }
1621 }
1622 if (i == ARRAY_SIZE(mask_maps)) {
1623 mask = -EINVAL;
1624 break;
1625 }
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001626 }
Li Zefan9eb85122009-04-09 11:19:40 +08001627 kfree(buf);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001628
1629 return mask;
1630}
1631
Li Zefan09341992009-03-24 17:43:30 +08001632static ssize_t blk_trace_mask2str(char *buf, int mask)
1633{
1634 int i;
1635 char *p = buf;
1636
1637 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1638 if (mask & mask_maps[i].mask) {
1639 p += sprintf(p, "%s%s",
1640 (p == buf) ? "" : ",", mask_maps[i].str);
1641 }
1642 }
1643 *p++ = '\n';
1644
1645 return p - buf;
1646}
1647
Li Zefanb1251302009-03-20 10:34:00 +08001648static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1649{
1650 if (bdev->bd_disk == NULL)
1651 return NULL;
1652
1653 return bdev_get_queue(bdev);
1654}
1655
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001656static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1657 struct device_attribute *attr,
1658 char *buf)
1659{
1660 struct hd_struct *p = dev_to_part(dev);
1661 struct request_queue *q;
1662 struct block_device *bdev;
1663 ssize_t ret = -ENXIO;
1664
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001665 bdev = bdget(part_devt(p));
1666 if (bdev == NULL)
Arnd Bergmann01b284f2010-09-17 20:39:22 +02001667 goto out;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001668
Li Zefanb1251302009-03-20 10:34:00 +08001669 q = blk_trace_get_queue(bdev);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001670 if (q == NULL)
1671 goto out_bdput;
Li Zefanb1251302009-03-20 10:34:00 +08001672
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001673 mutex_lock(&bdev->bd_mutex);
Li Zefancd649b82009-03-20 11:33:55 +08001674
1675 if (attr == &dev_attr_enable) {
1676 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1677 goto out_unlock_bdev;
1678 }
1679
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001680 if (q->blk_trace == NULL)
1681 ret = sprintf(buf, "disabled\n");
1682 else if (attr == &dev_attr_act_mask)
Li Zefan09341992009-03-24 17:43:30 +08001683 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001684 else if (attr == &dev_attr_pid)
1685 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1686 else if (attr == &dev_attr_start_lba)
1687 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1688 else if (attr == &dev_attr_end_lba)
1689 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
Li Zefancd649b82009-03-20 11:33:55 +08001690
1691out_unlock_bdev:
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001692 mutex_unlock(&bdev->bd_mutex);
1693out_bdput:
1694 bdput(bdev);
Arnd Bergmann01b284f2010-09-17 20:39:22 +02001695out:
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001696 return ret;
1697}
1698
1699static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1700 struct device_attribute *attr,
1701 const char *buf, size_t count)
1702{
1703 struct block_device *bdev;
1704 struct request_queue *q;
1705 struct hd_struct *p;
1706 u64 value;
Li Zefan09341992009-03-24 17:43:30 +08001707 ssize_t ret = -EINVAL;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001708
1709 if (count == 0)
1710 goto out;
1711
1712 if (attr == &dev_attr_act_mask) {
1713 if (sscanf(buf, "%llx", &value) != 1) {
1714 /* Assume it is a list of trace category names */
Li Zefan09341992009-03-24 17:43:30 +08001715 ret = blk_trace_str2mask(buf);
1716 if (ret < 0)
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001717 goto out;
Li Zefan09341992009-03-24 17:43:30 +08001718 value = ret;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001719 }
1720 } else if (sscanf(buf, "%llu", &value) != 1)
1721 goto out;
1722
Li Zefan09341992009-03-24 17:43:30 +08001723 ret = -ENXIO;
1724
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001725 p = dev_to_part(dev);
1726 bdev = bdget(part_devt(p));
1727 if (bdev == NULL)
Arnd Bergmann01b284f2010-09-17 20:39:22 +02001728 goto out;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001729
Li Zefanb1251302009-03-20 10:34:00 +08001730 q = blk_trace_get_queue(bdev);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001731 if (q == NULL)
1732 goto out_bdput;
1733
1734 mutex_lock(&bdev->bd_mutex);
Li Zefancd649b82009-03-20 11:33:55 +08001735
1736 if (attr == &dev_attr_enable) {
1737 if (value)
Li Zefan9908c302009-04-14 13:59:34 +08001738 ret = blk_trace_setup_queue(q, bdev);
Li Zefancd649b82009-03-20 11:33:55 +08001739 else
1740 ret = blk_trace_remove_queue(q);
1741 goto out_unlock_bdev;
1742 }
1743
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001744 ret = 0;
1745 if (q->blk_trace == NULL)
Li Zefan9908c302009-04-14 13:59:34 +08001746 ret = blk_trace_setup_queue(q, bdev);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001747
1748 if (ret == 0) {
1749 if (attr == &dev_attr_act_mask)
1750 q->blk_trace->act_mask = value;
1751 else if (attr == &dev_attr_pid)
1752 q->blk_trace->pid = value;
1753 else if (attr == &dev_attr_start_lba)
1754 q->blk_trace->start_lba = value;
1755 else if (attr == &dev_attr_end_lba)
1756 q->blk_trace->end_lba = value;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001757 }
Li Zefancd649b82009-03-20 11:33:55 +08001758
1759out_unlock_bdev:
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001760 mutex_unlock(&bdev->bd_mutex);
1761out_bdput:
1762 bdput(bdev);
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001763out:
Li Zefancd649b82009-03-20 11:33:55 +08001764 return ret ? ret : count;
Arnaldo Carvalho de Meloc71a8962009-01-23 12:06:27 -02001765}
Li Zefancd649b82009-03-20 11:33:55 +08001766
Li Zefan1d54ad62009-04-14 14:00:05 +08001767int blk_trace_init_sysfs(struct device *dev)
1768{
1769 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1770}
1771
Zdenek Kabelac48c0d4d2009-09-25 06:19:26 +02001772void blk_trace_remove_sysfs(struct device *dev)
1773{
1774 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1775}
1776
Li Zefan55782132009-06-09 13:43:05 +08001777#endif /* CONFIG_BLK_DEV_IO_TRACE */
1778
1779#ifdef CONFIG_EVENT_TRACING
1780
1781void blk_dump_cmd(char *buf, struct request *rq)
1782{
1783 int i, end;
1784 int len = rq->cmd_len;
1785 unsigned char *cmd = rq->cmd;
1786
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001787 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
Li Zefan55782132009-06-09 13:43:05 +08001788 buf[0] = '\0';
1789 return;
1790 }
1791
1792 for (end = len - 1; end >= 0; end--)
1793 if (cmd[end])
1794 break;
1795 end++;
1796
1797 for (i = 0; i < len; i++) {
1798 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1799 if (i == end && end != len - 1) {
1800 sprintf(buf, " ..");
1801 break;
1802 }
1803 }
1804}
1805
1806void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1807{
1808 int i = 0;
1809
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001810 if (rw & REQ_FLUSH)
1811 rwbs[i++] = 'F';
1812
Li Zefan55782132009-06-09 13:43:05 +08001813 if (rw & WRITE)
1814 rwbs[i++] = 'W';
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001815 else if (rw & REQ_DISCARD)
Li Zefan55782132009-06-09 13:43:05 +08001816 rwbs[i++] = 'D';
1817 else if (bytes)
1818 rwbs[i++] = 'R';
1819 else
1820 rwbs[i++] = 'N';
1821
Namhyung Kimc09c47c2011-08-11 10:36:05 +02001822 if (rw & REQ_FUA)
1823 rwbs[i++] = 'F';
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001824 if (rw & REQ_RAHEAD)
Li Zefan55782132009-06-09 13:43:05 +08001825 rwbs[i++] = 'A';
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001826 if (rw & REQ_SYNC)
Li Zefan55782132009-06-09 13:43:05 +08001827 rwbs[i++] = 'S';
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001828 if (rw & REQ_META)
Li Zefan55782132009-06-09 13:43:05 +08001829 rwbs[i++] = 'M';
Adrian Hunter8d57a982010-08-11 14:17:49 -07001830 if (rw & REQ_SECURE)
1831 rwbs[i++] = 'E';
Li Zefan55782132009-06-09 13:43:05 +08001832
1833 rwbs[i] = '\0';
1834}
Kent Overstreet9ca8f8e2012-04-13 21:56:43 -04001835EXPORT_SYMBOL_GPL(blk_fill_rwbs);
Li Zefan55782132009-06-09 13:43:05 +08001836
Li Zefan55782132009-06-09 13:43:05 +08001837#endif /* CONFIG_EVENT_TRACING */
1838