aboutsummaryrefslogtreecommitdiff
path: root/drivers/oprofile/buffer_sync.c
blob: 37681700b61a6816137cef46ca5ecc5fd666325e (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
/**
 * @file buffer_sync.c
 *
 * @remark Copyright 2002 OProfile authors
 * @remark Read the file COPYING
 *
 * @author John Levon <levon@movementarian.org>
 * @author Barry Kasindorf
 *
 * This is the core of the buffer management. Each
 * CPU buffer is processed and entered into the
 * global event buffer. Such processing is necessary
 * in several circumstances, mentioned below.
 *
 * The processing does the job of converting the
 * transitory EIP value into a persistent dentry/offset
 * value that the profiler can record at its leisure.
 *
 * See fs/dcookies.c for a description of the dentry/offset
 * objects.
 */

#include <linux/mm.h>
#include <linux/workqueue.h>
#include <linux/notifier.h>
#include <linux/dcookies.h>
#include <linux/profile.h>
#include <linux/module.h>
#include <linux/fs.h>
#include <linux/oprofile.h>
#include <linux/sched.h>

#include "oprofile_stats.h"
#include "event_buffer.h"
#include "cpu_buffer.h"
#include "buffer_sync.h"

static LIST_HEAD(dying_tasks);
static LIST_HEAD(dead_tasks);
static cpumask_t marked_cpus = CPU_MASK_NONE;
static DEFINE_SPINLOCK(task_mortuary);
static void process_task_mortuary(void);


/* Take ownership of the task struct and place it on the
 * list for processing. Only after two full buffer syncs
 * does the task eventually get freed, because by then
 * we are sure we will not reference it again.
 * Can be invoked from softirq via RCU callback due to
 * call_rcu() of the task struct, hence the _irqsave.
 */
static int
task_free_notify(struct notifier_block *self, unsigned long val, void *data)
{
	unsigned long flags;
	struct task_struct *task = data;
	spin_lock_irqsave(&task_mortuary, flags);
	list_add(&task->tasks, &dying_tasks);
	spin_unlock_irqrestore(&task_mortuary, flags);
	return NOTIFY_OK;
}


/* The task is on its way out. A sync of the buffer means we can catch
 * any remaining samples for this task.
 */
static int
task_exit_notify(struct notifier_block *self, unsigned long val, void *data)
{
	/* To avoid latency problems, we only process the current CPU,
	 * hoping that most samples for the task are on this CPU
	 */
	sync_buffer(raw_smp_processor_id());
	return 0;
}


/* The task is about to try a do_munmap(). We peek at what it's going to
 * do, and if it's an executable region, process the samples first, so
 * we don't lose any. This does not have to be exact, it's a QoI issue
 * only.
 */
static int
munmap_notify(struct notifier_block *self, unsigned long val, void *data)
{
	unsigned long addr = (unsigned long)data;
	struct mm_struct *mm = current->mm;
	struct vm_area_struct *mpnt;

	down_read(&mm->mmap_sem);

	mpnt = find_vma(mm, addr);
	if (mpnt && mpnt->vm_file && (mpnt->vm_flags & VM_EXEC)) {
		up_read(&mm->mmap_sem);
		/* To avoid latency problems, we only process the current CPU,
		 * hoping that most samples for the task are on this CPU
		 */
		sync_buffer(raw_smp_processor_id());
		return 0;
	}

	up_read(&mm->mmap_sem);
	return 0;
}


/* We need to be told about new modules so we don't attribute to a previously
 * loaded module, or drop the samples on the floor.
 */
static int
module_load_notify(struct notifier_block *self, unsigned long val, void *data)
{
#ifdef CONFIG_MODULES
	if (val != MODULE_STATE_COMING)
		return 0;

	/* FIXME: should we process all CPU buffers ? */
	mutex_lock(&buffer_mutex);
	add_event_entry(ESCAPE_CODE);
	add_event_entry(MODULE_LOADED_CODE);
	mutex_unlock(&buffer_mutex);
#endif
	return 0;
}


static struct notifier_block task_free_nb = {
	.notifier_call	= task_free_notify,
};

static struct notifier_block task_exit_nb = {
	.notifier_call	= task_exit_notify,
};

static struct notifier_block munmap_nb = {
	.notifier_call	= munmap_notify,
};

static struct notifier_block module_load_nb = {
	.notifier_call = module_load_notify,
};


static void end_sync(void)
{
	end_cpu_work();
	/* make sure we don't leak task structs */
	process_task_mortuary();
	process_task_mortuary();
}


int sync_start(void)
{
	int err;

	start_cpu_work();

	err = task_handoff_register(&task_free_nb);
	if (err)
		goto out1;
	err = profile_event_register(PROFILE_TASK_EXIT, &task_exit_nb);
	if (err)
		goto out2;
	err = profile_event_register(PROFILE_MUNMAP, &munmap_nb);
	if (err)
		goto out3;
	err = register_module_notifier(&module_load_nb);
	if (err)
		goto out4;

out:
	return err;
out4:
	profile_event_unregister(PROFILE_MUNMAP, &munmap_nb);
out3:
	profile_event_unregister(PROFILE_TASK_EXIT, &task_exit_nb);
out2:
	task_handoff_unregister(&task_free_nb);
out1:
	end_sync();
	goto out;
}


void sync_stop(void)
{
	unregister_module_notifier(&module_load_nb);
	profile_event_unregister(PROFILE_MUNMAP, &munmap_nb);
	profile_event_unregister(PROFILE_TASK_EXIT, &task_exit_nb);
	task_handoff_unregister(&task_free_nb);
	end_sync();
}


/* Optimisation. We can manage without taking the dcookie sem
 * because we cannot reach this code without at least one
 * dcookie user still being registered (namely, the reader
 * of the event buffer). */
static inline unsigned long fast_get_dcookie(struct path *path)
{
	unsigned long cookie;

	if (path->dentry->d_cookie)
		return (unsigned long)path->dentry;
	get_dcookie(path, &cookie);
	return cookie;
}


/* Look up the dcookie for the task's first VM_EXECUTABLE mapping,
 * which corresponds loosely to "application name". This is
 * not strictly necessary but allows oprofile to associate
 * shared-library samples with particular applications
 */
static unsigned long get_exec_dcookie(struct mm_struct *mm)
{
	unsigned long cookie = NO_COOKIE;
	struct vm_area_struct *vma;

	if (!mm)
		goto out;

	for (vma = mm->mmap; vma; vma = vma->vm_next) {
		if (!vma->vm_file)
			continue;
		if (!(vma->vm_flags & VM_EXECUTABLE))
			continue;
		cookie = fast_get_dcookie(&vma->vm_file->f_path);
		break;
	}

out:
	return cookie;
}


/* Convert the EIP value of a sample into a persistent dentry/offset
 * pair that can then be added to the global event buffer. We make
 * sure to do this lookup before a mm->mmap modification happens so
 * we don't lose track.
 */
static unsigned long
lookup_dcookie(struct mm_struct *mm, unsigned long addr, off_t *offset)
{
	unsigned long cookie = NO_COOKIE;
	struct vm_area_struct *vma;

	for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {

		if (addr < vma->vm_start || addr >= vma->vm_end)
			continue;

		if (vma->vm_file) {
			cookie = fast_get_dcookie(&vma->vm_file->f_path);
			*offset = (vma->vm_pgoff << PAGE_SHIFT) + addr -
				vma->vm_start;
		} else {
			/* must be an anonymous map */
			*offset = addr;
		}

		break;
	}

	if (!vma)
		cookie = INVALID_COOKIE;

	return cookie;
}

static void increment_tail(struct oprofile_cpu_buffer *b)
{
	unsigned long new_tail = b->tail_pos + 1;

	rmb();	/* be sure fifo pointers are synchromized */

	if (new_tail < b->buffer_size)
		b->tail_pos = new_tail;
	else
		b->tail_pos = 0;
}

static unsigned long last_cookie = INVALID_COOKIE;

static void add_cpu_switch(int i)
{
	add_event_entry(ESCAPE_CODE);
	add_event_entry(CPU_SWITCH_CODE);
	add_event_entry(i);
	last_cookie = INVALID_COOKIE;
}

static void add_kernel_ctx_switch(unsigned int in_kernel)
{
	add_event_entry(ESCAPE_CODE);
	if (in_kernel)
		add_event_entry(KERNEL_ENTER_SWITCH_CODE);
	else
		add_event_entry(KERNEL_EXIT_SWITCH_CODE);
}

static void
add_user_ctx_switch(struct task_struct const *task, unsigned long cookie)
{
	add_event_entry(ESCAPE_CODE);
	add_event_entry(CTX_SWITCH_CODE);
	add_event_entry(task->pid);
	add_event_entry(cookie);
	/* Another code for daemon back-compat */
	add_event_entry(ESCAPE_CODE);
	add_event_entry(CTX_TGID_CODE);
	add_event_entry(task->tgid);
}


static void add_cookie_switch(unsigned long cookie)
{
	add_event_entry(ESCAPE_CODE);
	add_event_entry(COOKIE_SWITCH_CODE);
	add_event_entry(cookie);
}


static void add_trace_begin(void)
{
	add_event_entry(ESCAPE_CODE);
	add_event_entry(TRACE_BEGIN_CODE);
}

#ifdef CONFIG_OPROFILE_IBS

#define IBS_FETCH_CODE_SIZE	2
#define IBS_OP_CODE_SIZE	5
#define IBS_EIP(offset)				\
	(((struct op_sample *)&cpu_buf->buffer[(offset)])->eip)
#define IBS_EVENT(offset)				\
	(((struct op_sample *)&cpu_buf->buffer[(offset)])->event)

/*
 * Add IBS fetch and op entries to event buffer
 */
static void add_ibs_begin(struct oprofile_cpu_buffer *cpu_buf, int code,
	int in_kernel, struct mm_struct *mm)
{
	unsigned long rip;
	int i, count;
	unsigned long ibs_cookie = 0;
	off_t offset;

	increment_tail(cpu_buf);	/* move to RIP entry */

	rip = IBS_EIP(cpu_buf->tail_pos);

#ifdef __LP64__
	rip += IBS_EVENT(cpu_buf->tail_pos) << 32;
#endif

	if (mm) {
		ibs_cookie = lookup_dcookie(mm, rip, &offset);

		if (ibs_cookie == NO_COOKIE)
			offset = rip;
		if (ibs_cookie == INVALID_COOKIE) {
			atomic_inc(&oprofile_stats.sample_lost_no_mapping);
			offset = rip;
		}
		if (ibs_cookie != last_cookie) {
			add_cookie_switch(ibs_cookie);
			last_cookie = ibs_cookie;
		}
	} else
		offset = rip;

	add_event_entry(ESCAPE_CODE);
	add_event_entry(code);
	add_event_entry(offset);	/* Offset from Dcookie */

	/* we send the Dcookie offset, but send the raw Linear Add also*/
	add_event_entry(IBS_EIP(cpu_buf->tail_pos));
	add_event_entry(IBS_EVENT(cpu_buf->tail_pos));

	if (code == IBS_FETCH_CODE)
		count = IBS_FETCH_CODE_SIZE;	/*IBS FETCH is 2 int64s*/
	else
		count = IBS_OP_CODE_SIZE;	/*IBS OP is 5 int64s*/

	for (i = 0; i < count; i++) {
		increment_tail(cpu_buf);
		add_event_entry(IBS_EIP(cpu_buf->tail_pos));
		add_event_entry(IBS_EVENT(cpu_buf->tail_pos));
	}
}

#endif

static void add_sample_entry(unsigned long offset, unsigned long event)
{
	add_event_entry(offset);
	add_event_entry(event);
}


static int add_us_sample(struct mm_struct *mm, struct op_sample *s)
{
	unsigned long cookie;
	off_t offset;

	cookie = lookup_dcookie(mm, s->eip, &offset);

	if (cookie == INVALID_COOKIE) {
		atomic_inc(&oprofile_stats.sample_lost_no_mapping);
		return 0;
	}

	if (cookie != last_cookie) {
		add_cookie_switch(cookie);
		last_cookie = cookie;
	}

	add_sample_entry(offset, s->event);

	return 1;
}


/* Add a sample to the global event buffer. If possible the
 * sample is converted into a persistent dentry/offset pair
 * for later lookup from userspace.
 */
static int
add_sample(struct mm_struct *mm, struct op_sample *s, int in_kernel)
{
	if (in_kernel) {
		add_sample_entry(s->eip, s->event);
		return 1;
	} else if (mm) {
		return add_us_sample(mm, s);
	} else {
		atomic_inc(&oprofile_stats.sample_lost_no_mm);
	}
	return 0;
}


static void release_mm(struct mm_struct *mm)
{
	if (!mm)
		return;
	up_read(&mm->mmap_sem);
	mmput(mm);
}


static struct mm_struct *take_tasks_mm(struct task_struct *task)
{
	struct mm_struct *mm = get_task_mm(task);
	if (mm)
		down_read(&mm->mmap_sem);
	return mm;
}


static inline int is_code(unsigned long val)
{
	return val == ESCAPE_CODE;
}


/* "acquire" as many cpu buffer slots as we can */
static unsigned long get_slots(struct oprofile_cpu_buffer *b)
{
	unsigned long head = b->head_pos;
	unsigned long tail = b->tail_pos;

	/*
	 * Subtle. This resets the persistent last_task
	 * and in_kernel values used for switching notes.
	 * BUT, there is a small window between reading
	 * head_pos, and this call, that means samples
	 * can appear at the new head position, but not
	 * be prefixed with the notes for switching
	 * kernel mode or a task switch. This small hole
	 * can lead to mis-attribution or samples where
	 * we don't know if it's in the kernel or not,
	 * at the start of an event buffer.
	 */
	cpu_buffer_reset(b);

	if (head >= tail)
		return head - tail;

	return head + (b->buffer_size - tail);
}


/* Move tasks along towards death. Any tasks on dead_tasks
 * will definitely have no remaining references in any
 * CPU buffers at this point, because we use two lists,
 * and to have reached the list, it must have gone through
 * one full sync already.
 */
static void process_task_mortuary(void)
{
	unsigned long flags;
	LIST_HEAD(local_dead_tasks);
	struct task_struct *task;
	struct task_struct *ttask;

	spin_lock_irqsave(&task_mortuary, flags);

	list_splice_init(&dead_tasks, &local_dead_tasks);
	list_splice_init(&dying_tasks, &dead_tasks);

	spin_unlock_irqrestore(&task_mortuary, flags);

	list_for_each_entry_safe(task, ttask, &local_dead_tasks, tasks) {
		list_del(&task->tasks);
		free_task(task);
	}
}


static void mark_done(int cpu)
{
	int i;

	cpu_set(cpu, marked_cpus);

	for_each_online_cpu(i) {
		if (!cpu_isset(i, marked_cpus))
			return;
	}

	/* All CPUs have been processed at least once,
	 * we can process the mortuary once
	 */
	process_task_mortuary();

	cpus_clear(marked_cpus);
}


/* FIXME: this is not sufficient if we implement syscall barrier backtrace
 * traversal, the code switch to sb_sample_start at first kernel enter/exit
 * switch so we need a fifth state and some special handling in sync_buffer()
 */
typedef enum {
	sb_bt_ignore = -2,
	sb_buffer_start,
	sb_bt_start,
	sb_sample_start,
} sync_buffer_state;

/* Sync one of the CPU's buffers into the global event buffer.
 * Here we need to go through each batch of samples punctuated
 * by context switch notes, taking the task's mmap_sem and doing
 * lookup in task->mm->mmap to convert EIP into dcookie/offset
 * value.
 */
void sync_buffer(int cpu)
{
	struct oprofile_cpu_buffer *cpu_buf = &per_cpu(cpu_buffer, cpu);
	struct mm_struct *mm = NULL;
	struct task_struct *new;
	unsigned long cookie = 0;
	int in_kernel = 1;
	unsigned int i;
	sync_buffer_state state = sb_buffer_start;
	unsigned long available;

	mutex_lock(&buffer_mutex);

	add_cpu_switch(cpu);

	/* Remember, only we can modify tail_pos */

	available = get_slots(cpu_buf);

	for (i = 0; i < available; ++i) {
		struct op_sample *s = &cpu_buf->buffer[cpu_buf->tail_pos];

		if (is_code(s->eip)) {
			if (s->event <= CPU_IS_KERNEL) {
				/* kernel/userspace switch */
				in_kernel = s->event;
				if (state == sb_buffer_start)
					state = sb_sample_start;
				add_kernel_ctx_switch(s->event);
			} else if (s->event == CPU_TRACE_BEGIN) {
				state = sb_bt_start;
				add_trace_begin();
#ifdef CONFIG_OPROFILE_IBS
			} else if (s->event == IBS_FETCH_BEGIN) {
				state = sb_bt_start;
				add_ibs_begin(cpu_buf,
					IBS_FETCH_CODE, in_kernel, mm);
			} else if (s->event == IBS_OP_BEGIN) {
				state = sb_bt_start;
				add_ibs_begin(cpu_buf,
					IBS_OP_CODE, in_kernel, mm);
#endif
			} else {
				struct mm_struct *oldmm = mm;

				/* userspace context switch */
				new = (struct task_struct *)s->event;

				release_mm(oldmm);
				mm = take_tasks_mm(new);
				if (mm != oldmm)
					cookie = get_exec_dcookie(mm);
				add_user_ctx_switch(new, cookie);
			}
		} else if (state >= sb_bt_start &&
			   !add_sample(mm, s, in_kernel)) {
			if (state == sb_bt_start) {
				state = sb_bt_ignore;
				atomic_inc(&oprofile_stats.bt_lost_no_mapping);
			}
		}

		increment_tail(cpu_buf);
	}
	release_mm(mm);

	mark_done(cpu);

	mutex_unlock(&buffer_mutex);
}

/* The function can be used to add a buffer worth of data directly to
 * the kernel buffer. The buffer is assumed to be a circular buffer.
 * Take the entries from index start and end at index end, wrapping
 * at max_entries.
 */
void oprofile_put_buff(unsigned long *buf, unsigned int start,
		       unsigned int stop, unsigned int max)
{
	int i;

	i = start;

	mutex_lock(&buffer_mutex);
	while (i != stop) {
		add_event_entry(buf[i++]);

		if (i >= max)
			i = 0;
	}

	mutex_unlock(&buffer_mutex);
}