aboutsummaryrefslogtreecommitdiff
path: root/kernel/profile.c
blob: c2ebddb5e97466a99612adc40a9672249eee9bfc (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
// SPDX-License-Identifier: GPL-2.0-only
/*
 *  linux/kernel/profile.c
 *  Simple profiling. Manages a direct-mapped profile hit count buffer,
 *  with configurable resolution, support for restricting the cpus on
 *  which profiling is done, and switching between cpu time and
 *  schedule() calls via kernel command line parameters passed at boot.
 *
 *  Scheduler profiling support, Arjan van de Ven and Ingo Molnar,
 *	Red Hat, July 2004
 *  Consolidation of architecture support code for profiling,
 *	Nadia Yvette Chambers, Oracle, July 2004
 *  Amortized hit count accounting via per-cpu open-addressed hashtables
 *	to resolve timer interrupt livelocks, Nadia Yvette Chambers,
 *	Oracle, 2004
 */

#include <linux/export.h>
#include <linux/profile.h>
#include <linux/memblock.h>
#include <linux/notifier.h>
#include <linux/mm.h>
#include <linux/cpumask.h>
#include <linux/cpu.h>
#include <linux/highmem.h>
#include <linux/mutex.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/sched/stat.h>

#include <asm/sections.h>
#include <asm/irq_regs.h>
#include <asm/ptrace.h>

struct profile_hit {
	u32 pc, hits;
};
#define PROFILE_GRPSHIFT	3
#define PROFILE_GRPSZ		(1 << PROFILE_GRPSHIFT)
#define NR_PROFILE_HIT		(PAGE_SIZE/sizeof(struct profile_hit))
#define NR_PROFILE_GRP		(NR_PROFILE_HIT/PROFILE_GRPSZ)

static atomic_t *prof_buffer;
static unsigned long prof_len, prof_shift;

int prof_on __read_mostly;
EXPORT_SYMBOL_GPL(prof_on);

static cpumask_var_t prof_cpu_mask;
#if defined(CONFIG_SMP) && defined(CONFIG_PROC_FS)
static DEFINE_PER_CPU(struct profile_hit *[2], cpu_profile_hits);
static DEFINE_PER_CPU(int, cpu_profile_flip);
static DEFINE_MUTEX(profile_flip_mutex);
#endif /* CONFIG_SMP */

int profile_setup(char *str)
{
	static const char schedstr[] = "schedule";
	static const char sleepstr[] = "sleep";
	static const char kvmstr[] = "kvm";
	int par;

	if (!strncmp(str, sleepstr, strlen(sleepstr))) {
#ifdef CONFIG_SCHEDSTATS
		force_schedstat_enabled();
		prof_on = SLEEP_PROFILING;
		if (str[strlen(sleepstr)] == ',')
			str += strlen(sleepstr) + 1;
		if (get_option(&str, &par))
			prof_shift = par;
		pr_info("kernel sleep profiling enabled (shift: %ld)\n",
			prof_shift);
#else
		pr_warn("kernel sleep profiling requires CONFIG_SCHEDSTATS\n");
#endif /* CONFIG_SCHEDSTATS */
	} else if (!strncmp(str, schedstr, strlen(schedstr))) {
		prof_on = SCHED_PROFILING;
		if (str[strlen(schedstr)] == ',')
			str += strlen(schedstr) + 1;
		if (get_option(&str, &par))
			prof_shift = par;
		pr_info("kernel schedule profiling enabled (shift: %ld)\n",
			prof_shift);
	} else if (!strncmp(str, kvmstr, strlen(kvmstr))) {
		prof_on = KVM_PROFILING;
		if (str[strlen(kvmstr)] == ',')
			str += strlen(kvmstr) + 1;
		if (get_option(&str, &par))
			prof_shift = par;
		pr_info("kernel KVM profiling enabled (shift: %ld)\n",
			prof_shift);
	} else if (get_option(&str, &par)) {
		prof_shift = par;
		prof_on = CPU_PROFILING;
		pr_info("kernel profiling enabled (shift: %ld)\n",
			prof_shift);
	}
	return 1;
}
__setup("profile=", profile_setup);


int __ref profile_init(void)
{
	int buffer_bytes;
	if (!prof_on)
		return 0;

	/* only text is profiled */
	prof_len = (_etext - _stext) >> prof_shift;
	buffer_bytes = prof_len*sizeof(atomic_t);

	if (!alloc_cpumask_var(&prof_cpu_mask, GFP_KERNEL))
		return -ENOMEM;

	cpumask_copy(prof_cpu_mask, cpu_possible_mask);

	prof_buffer = kzalloc(buffer_bytes, GFP_KERNEL|__GFP_NOWARN);
	if (prof_buffer)
		return 0;

	prof_buffer = alloc_pages_exact(buffer_bytes,
					GFP_KERNEL|__GFP_ZERO|__GFP_NOWARN);
	if (prof_buffer)
		return 0;

	prof_buffer = vzalloc(buffer_bytes);
	if (prof_buffer)
		return 0;

	free_cpumask_var(prof_cpu_mask);
	return -ENOMEM;
}

/* Profile event notifications */

static BLOCKING_NOTIFIER_HEAD(task_exit_notifier);
static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
static BLOCKING_NOTIFIER_HEAD(munmap_notifier);

void profile_task_exit(struct task_struct *task)
{
	blocking_notifier_call_chain(&task_exit_notifier, 0, task);
}

int profile_handoff_task(struct task_struct *task)
{
	int ret;
	ret = atomic_notifier_call_chain(&task_free_notifier, 0, task);
	return (ret == NOTIFY_OK) ? 1 : 0;
}

void profile_munmap(unsigned long addr)
{
	blocking_notifier_call_chain(&munmap_notifier, 0, (void *)addr);
}

int task_handoff_register(struct notifier_block *n)
{
	return atomic_notifier_chain_register(&task_free_notifier, n);
}
EXPORT_SYMBOL_GPL(task_handoff_register);

int task_handoff_unregister(struct notifier_block *n)
{
	return atomic_notifier_chain_unregister(&task_free_notifier, n);
}
EXPORT_SYMBOL_GPL(task_handoff_unregister);

int profile_event_register(enum profile_type type, struct notifier_block *n)
{
	int err = -EINVAL;

	switch (type) {
	case PROFILE_TASK_EXIT:
		err = blocking_notifier_chain_register(
				&task_exit_notifier, n);
		break;
	case PROFILE_MUNMAP:
		err = blocking_notifier_chain_register(
				&munmap_notifier, n);
		break;
	}

	return err;
}
EXPORT_SYMBOL_GPL(profile_event_register);

int profile_event_unregister(enum profile_type type, struct notifier_block *n)
{
	int err = -EINVAL;

	switch (type) {
	case PROFILE_TASK_EXIT:
		err = blocking_notifier_chain_unregister(
				&task_exit_notifier, n);
		break;
	case PROFILE_MUNMAP:
		err = blocking_notifier_chain_unregister(
				&munmap_notifier, n);
		break;
	}

	return err;
}
EXPORT_SYMBOL_GPL(profile_event_unregister);

#if defined(CONFIG_SMP) && defined(CONFIG_PROC_FS)
/*
 * Each cpu has a pair of open-addressed hashtables for pending
 * profile hits. read_profile() IPI's all cpus to request them
 * to flip buffers and flushes their contents to prof_buffer itself.
 * Flip requests are serialized by the profile_flip_mutex. The sole
 * use of having a second hashtable is for avoiding cacheline
 * contention that would otherwise happen during flushes of pending
 * profile hits required for the accuracy of reported profile hits
 * and so resurrect the interrupt livelock issue.
 *
 * The open-addressed hashtables are indexed by profile buffer slot
 * and hold the number of pending hits to that profile buffer slot on
 * a cpu in an entry. When the hashtable overflows, all pending hits
 * are accounted to their corresponding profile buffer slots with
 * atomic_add() and the hashtable emptied. As numerous pending hits
 * may be accounted to a profile buffer slot in a hashtable entry,
 * this amortizes a number of atomic profile buffer increments likely
 * to be far larger than the number of entries in the hashtable,
 * particularly given that the number of distinct profile buffer
 * positions to which hits are accounted during short intervals (e.g.
 * several seconds) is usually very small. Exclusion from buffer
 * flipping is provided by interrupt disablement (note that for
 * SCHED_PROFILING or SLEEP_PROFILING profile_hit() may be called from
 * process context).
 * The hash function is meant to be lightweight as opposed to strong,
 * and was vaguely inspired by ppc64 firmware-supported inverted
 * pagetable hash functions, but uses a full hashtable full of finite
 * collision chains, not just pairs of them.
 *
 * -- nyc
 */
static void __profile_flip_buffers(void *unused)
{
	int cpu = smp_processor_id();

	per_cpu(cpu_profile_flip, cpu) = !per_cpu(cpu_profile_flip, cpu);
}

static void profile_flip_buffers(void)
{
	int i, j, cpu;

	mutex_lock(&profile_flip_mutex);
	j = per_cpu(cpu_profile_flip, get_cpu());
	put_cpu();
	on_each_cpu(__profile_flip_buffers, NULL, 1);
	for_each_online_cpu(cpu) {
		struct profile_hit *hits = per_cpu(cpu_profile_hits, cpu)[j];
		for (i = 0; i < NR_PROFILE_HIT; ++i) {
			if (!hits[i].hits) {
				if (hits[i].pc)
					hits[i].pc = 0;
				continue;
			}
			atomic_add(hits[i].hits, &prof_buffer[hits[i].pc]);
			hits[i].hits = hits[i].pc = 0;
		}
	}
	mutex_unlock(&profile_flip_mutex);
}

static void profile_discard_flip_buffers(void)
{
	int i, cpu;

	mutex_lock(&profile_flip_mutex);
	i = per_cpu(cpu_profile_flip, get_cpu());
	put_cpu();
	on_each_cpu(__profile_flip_buffers, NULL, 1);
	for_each_online_cpu(cpu) {
		struct profile_hit *hits = per_cpu(cpu_profile_hits, cpu)[i];
		memset(hits, 0, NR_PROFILE_HIT*sizeof(struct profile_hit));
	}
	mutex_unlock(&profile_flip_mutex);
}

static void do_profile_hits(int type, void *__pc, unsigned int nr_hits)
{
	unsigned long primary, secondary, flags, pc = (unsigned long)__pc;
	int i, j, cpu;
	struct profile_hit *hits;

	pc = min((pc - (unsigned long)_stext) >> prof_shift, prof_len - 1);
	i = primary = (pc & (NR_PROFILE_GRP - 1)) << PROFILE_GRPSHIFT;
	secondary = (~(pc << 1) & (NR_PROFILE_GRP - 1)) << PROFILE_GRPSHIFT;
	cpu = get_cpu();
	hits = per_cpu(cpu_profile_hits, cpu)[per_cpu(cpu_profile_flip, cpu)];
	if (!hits) {
		put_cpu();
		return;
	}
	/*
	 * We buffer the global profiler buffer into a per-CPU
	 * queue and thus reduce the number of global (and possibly
	 * NUMA-alien) accesses. The write-queue is self-coalescing:
	 */
	local_irq_save(flags);
	do {
		for (j = 0; j < PROFILE_GRPSZ; ++j) {
			if (hits[i + j].pc == pc) {
				hits[i + j].hits += nr_hits;
				goto out;
			} else if (!hits[i + j].hits) {
				hits[i + j].pc = pc;
				hits[i + j].hits = nr_hits;
				goto out;
			}
		}
		i = (i + secondary) & (NR_PROFILE_HIT - 1);
	} while (i != primary);

	/*
	 * Add the current hit(s) and flush the write-queue out
	 * to the global buffer:
	 */
	atomic_add(nr_hits, &prof_buffer[pc]);
	for (i = 0; i < NR_PROFILE_HIT; ++i) {
		atomic_add(hits[i].hits, &prof_buffer[hits[i].pc]);
		hits[i].pc = hits[i].hits = 0;
	}
out:
	local_irq_restore(flags);
	put_cpu();
}

static int profile_dead_cpu(unsigned int cpu)
{
	struct page *page;
	int i;

	if (cpumask_available(prof_cpu_mask))
		cpumask_clear_cpu(cpu, prof_cpu_mask);

	for (i = 0; i < 2; i++) {
		if (per_cpu(cpu_profile_hits, cpu)[i]) {
			page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[i]);
			per_cpu(cpu_profile_hits, cpu)[i] = NULL;
			__free_page(page);
		}
	}
	return 0;
}

static int profile_prepare_cpu(unsigned int cpu)
{
	int i, node = cpu_to_mem(cpu);
	struct page *page;

	per_cpu(cpu_profile_flip, cpu) = 0;

	for (i = 0; i < 2; i++) {
		if (per_cpu(cpu_profile_hits, cpu)[i])
			continue;

		page = __alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
		if (!page) {
			profile_dead_cpu(cpu);
			return -ENOMEM;
		}
		per_cpu(cpu_profile_hits, cpu)[i] = page_address(page);

	}
	return 0;
}

static int profile_online_cpu(unsigned int cpu)
{
	if (cpumask_available(prof_cpu_mask))
		cpumask_set_cpu(cpu, prof_cpu_mask);

	return 0;
}

#else /* !CONFIG_SMP */
#define profile_flip_buffers()		do { } while (0)
#define profile_discard_flip_buffers()	do { } while (0)

static void do_profile_hits(int type, void *__pc, unsigned int nr_hits)
{
	unsigned long pc;
	pc = ((unsigned long)__pc - (unsigned long)_stext) >> prof_shift;
	atomic_add(nr_hits, &prof_buffer[min(pc, prof_len - 1)]);
}
#endif /* !CONFIG_SMP */

void profile_hits(int type, void *__pc, unsigned int nr_hits)
{
	if (prof_on != type || !prof_buffer)
		return;
	do_profile_hits(type, __pc, nr_hits);
}
EXPORT_SYMBOL_GPL(profile_hits);

void profile_tick(int type)
{
	struct pt_regs *regs = get_irq_regs();

	if (!user_mode(regs) && cpumask_available(prof_cpu_mask) &&
	    cpumask_test_cpu(smp_processor_id(), prof_cpu_mask))
		profile_hit(type, (void *)profile_pc(regs));
}

#ifdef CONFIG_PROC_FS
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
#include <linux/uaccess.h>

static int prof_cpu_mask_proc_show(struct seq_file *m, void *v)
{
	seq_printf(m, "%*pb\n", cpumask_pr_args(prof_cpu_mask));
	return 0;
}

static int prof_cpu_mask_proc_open(struct inode *inode, struct file *file)
{
	return single_open(file, prof_cpu_mask_proc_show, NULL);
}

static ssize_t prof_cpu_mask_proc_write(struct file *file,
	const char __user *buffer, size_t count, loff_t *pos)
{
	cpumask_var_t new_value;
	int err;

	if (!zalloc_cpumask_var(&new_value, GFP_KERNEL))
		return -ENOMEM;

	err = cpumask_parse_user(buffer, count, new_value);
	if (!err) {
		cpumask_copy(prof_cpu_mask, new_value);
		err = count;
	}
	free_cpumask_var(new_value);
	return err;
}

static const struct proc_ops prof_cpu_mask_proc_ops = {
	.proc_open	= prof_cpu_mask_proc_open,
	.proc_read	= seq_read,
	.proc_lseek	= seq_lseek,
	.proc_release	= single_release,
	.proc_write	= prof_cpu_mask_proc_write,
};

void create_prof_cpu_mask(void)
{
	/* create /proc/irq/prof_cpu_mask */
	proc_create("irq/prof_cpu_mask", 0600, NULL, &prof_cpu_mask_proc_ops);
}

/*
 * This function accesses profiling information. The returned data is
 * binary: the sampling step and the actual contents of the profile
 * buffer. Use of the program readprofile is recommended in order to
 * get meaningful info out of these data.
 */
static ssize_t
read_profile(struct file *file, char __user *buf, size_t count, loff_t *ppos)
{
	unsigned long p = *ppos;
	ssize_t read;
	char *pnt;
	unsigned int sample_step = 1 << prof_shift;

	profile_flip_buffers();
	if (p >= (prof_len+1)*sizeof(unsigned int))
		return 0;
	if (count > (prof_len+1)*sizeof(unsigned int) - p)
		count = (prof_len+1)*sizeof(unsigned int) - p;
	read = 0;

	while (p < sizeof(unsigned int) && count > 0) {
		if (put_user(*((char *)(&sample_step)+p), buf))
			return -EFAULT;
		buf++; p++; count--; read++;
	}
	pnt = (char *)prof_buffer + p - sizeof(atomic_t);
	if (copy_to_user(buf, (void *)pnt, count))
		return -EFAULT;
	read += count;
	*ppos += read;
	return read;
}

/*
 * Writing to /proc/profile resets the counters
 *
 * Writing a 'profiling multiplier' value into it also re-sets the profiling
 * interrupt frequency, on architectures that support this.
 */
static ssize_t write_profile(struct file *file, const char __user *buf,
			     size_t count, loff_t *ppos)
{
#ifdef CONFIG_SMP
	extern int setup_profiling_timer(unsigned int multiplier);

	if (count == sizeof(int)) {
		unsigned int multiplier;

		if (copy_from_user(&multiplier, buf, sizeof(int)))
			return -EFAULT;

		if (setup_profiling_timer(multiplier))
			return -EINVAL;
	}
#endif
	profile_discard_flip_buffers();
	memset(prof_buffer, 0, prof_len * sizeof(atomic_t));
	return count;
}

static const struct proc_ops profile_proc_ops = {
	.proc_read	= read_profile,
	.proc_write	= write_profile,
	.proc_lseek	= default_llseek,
};

int __ref create_proc_profile(void)
{
	struct proc_dir_entry *entry;
#ifdef CONFIG_SMP
	enum cpuhp_state online_state;
#endif

	int err = 0;

	if (!prof_on)
		return 0;
#ifdef CONFIG_SMP
	err = cpuhp_setup_state(CPUHP_PROFILE_PREPARE, "PROFILE_PREPARE",
				profile_prepare_cpu, profile_dead_cpu);
	if (err)
		return err;

	err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "AP_PROFILE_ONLINE",
				profile_online_cpu, NULL);
	if (err < 0)
		goto err_state_prep;
	online_state = err;
	err = 0;
#endif
	entry = proc_create("profile", S_IWUSR | S_IRUGO,
			    NULL, &profile_proc_ops);
	if (!entry)
		goto err_state_onl;
	proc_set_size(entry, (1 + prof_len) * sizeof(atomic_t));

	return err;
err_state_onl:
#ifdef CONFIG_SMP
	cpuhp_remove_state(online_state);
err_state_prep:
	cpuhp_remove_state(CPUHP_PROFILE_PREPARE);
#endif
	return err;
}
subsys_initcall(create_proc_profile);
#endif /* CONFIG_PROC_FS */