aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/oprofile/cpu_buffer.c
blob: 01d38e78cde18bd341fca22940249e4a80068217 (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
/**
 * @file cpu_buffer.c
 *
 * @remark Copyright 2002 OProfile authors
 * @remark Read the file COPYING
 *
 * @author John Levon <levon@movementarian.org>
 * @author Barry Kasindorf <barry.kasindorf@amd.com>
 *
 * Each CPU has a local buffer that stores PC value/event
 * pairs. We also log context switches when we notice them.
 * Eventually each CPU's buffer is processed into the global
 * event buffer by sync_buffer().
 *
 * We use a local buffer for two reasons: an NMI or similar
 * interrupt cannot synchronise, and high sampling rates
 * would lead to catastrophic global synchronisation if
 * a global buffer was used.
 */

#include <linux/sched.h>
#include <linux/oprofile.h>
#include <linux/vmalloc.h>
#include <linux/errno.h>

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

DEFINE_PER_CPU(struct oprofile_cpu_buffer, cpu_buffer);

static void wq_sync_buffer(struct work_struct *work);

#define DEFAULT_TIMER_EXPIRE (HZ / 10)
static int work_enabled;

void free_cpu_buffers(void)
{
	int i;

	for_each_possible_cpu(i) {
		vfree(per_cpu(cpu_buffer, i).buffer);
		per_cpu(cpu_buffer, i).buffer = NULL;
	}
}

unsigned long oprofile_get_cpu_buffer_size(void)
{
	return fs_cpu_buffer_size;
}

void oprofile_cpu_buffer_inc_smpl_lost(void)
{
	struct oprofile_cpu_buffer *cpu_buf
		= &__get_cpu_var(cpu_buffer);

	cpu_buf->sample_lost_overflow++;
}

int alloc_cpu_buffers(void)
{
	int i;

	unsigned long buffer_size = fs_cpu_buffer_size;

	for_each_possible_cpu(i) {
		struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);

		b->buffer = vmalloc_node(sizeof(struct op_sample) * buffer_size,
			cpu_to_node(i));
		if (!b->buffer)
			goto fail;

		b->last_task = NULL;
		b->last_is_kernel = -1;
		b->tracing = 0;
		b->buffer_size = buffer_size;
		b->tail_pos = 0;
		b->head_pos = 0;
		b->sample_received = 0;
		b->sample_lost_overflow = 0;
		b->backtrace_aborted = 0;
		b->sample_invalid_eip = 0;
		b->cpu = i;
		INIT_DELAYED_WORK(&b->work, wq_sync_buffer);
	}
	return 0;

fail:
	free_cpu_buffers();
	return -ENOMEM;
}

void start_cpu_work(void)
{
	int i;

	work_enabled = 1;

	for_each_online_cpu(i) {
		struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);

		/*
		 * Spread the work by 1 jiffy per cpu so they dont all
		 * fire at once.
		 */
		schedule_delayed_work_on(i, &b->work, DEFAULT_TIMER_EXPIRE + i);
	}
}

void end_cpu_work(void)
{
	int i;

	work_enabled = 0;

	for_each_online_cpu(i) {
		struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);

		cancel_delayed_work(&b->work);
	}

	flush_scheduled_work();
}

/* Resets the cpu buffer to a sane state. */
void cpu_buffer_reset(struct oprofile_cpu_buffer *cpu_buf)
{
	/* reset these to invalid values; the next sample
	 * collected will populate the buffer with proper
	 * values to initialize the buffer
	 */
	cpu_buf->last_is_kernel = -1;
	cpu_buf->last_task = NULL;
}

/* compute number of available slots in cpu_buffer queue */
static unsigned long nr_available_slots(struct oprofile_cpu_buffer const *b)
{
	unsigned long head = b->head_pos;
	unsigned long tail = b->tail_pos;

	if (tail > head)
		return (tail - head) - 1;

	return tail + (b->buffer_size - head) - 1;
}

static void increment_head(struct oprofile_cpu_buffer *b)
{
	unsigned long new_head = b->head_pos + 1;

	/* Ensure anything written to the slot before we
	 * increment is visible */
	wmb();

	if (new_head < b->buffer_size)
		b->head_pos = new_head;
	else
		b->head_pos = 0;
}

static inline void
add_sample(struct oprofile_cpu_buffer *cpu_buf,
	   unsigned long pc, unsigned long event)
{
	struct op_sample *entry = &cpu_buf->buffer[cpu_buf->head_pos];
	entry->eip = pc;
	entry->event = event;
	increment_head(cpu_buf);
}

static inline void
add_code(struct oprofile_cpu_buffer *buffer, unsigned long value)
{
	add_sample(buffer, ESCAPE_CODE, value);
}

/* This must be safe from any context. It's safe writing here
 * because of the head/tail separation of the writer and reader
 * of the CPU buffer.
 *
 * is_kernel is needed because on some architectures you cannot
 * tell if you are in kernel or user space simply by looking at
 * pc. We tag this in the buffer by generating kernel enter/exit
 * events whenever is_kernel changes
 */
static int log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
		      int is_kernel, unsigned long event)
{
	struct task_struct *task;

	cpu_buf->sample_received++;

	if (pc == ESCAPE_CODE) {
		cpu_buf->sample_invalid_eip++;
		return 0;
	}

	if (nr_available_slots(cpu_buf) < 3) {
		cpu_buf->sample_lost_overflow++;
		return 0;
	}

	is_kernel = !!is_kernel;

	task = current;

	/* notice a switch from user->kernel or vice versa */
	if (cpu_buf->last_is_kernel != is_kernel) {
		cpu_buf->last_is_kernel = is_kernel;
		add_code(cpu_buf, is_kernel);
	}

	/* notice a task switch */
	if (cpu_buf->last_task != task) {
		cpu_buf->last_task = task;
		add_code(cpu_buf, (unsigned long)task);
	}

	add_sample(cpu_buf, pc, event);
	return 1;
}

static int oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
{
	if (nr_available_slots(cpu_buf) < 4) {
		cpu_buf->sample_lost_overflow++;
		return 0;
	}

	add_code(cpu_buf, CPU_TRACE_BEGIN);
	cpu_buf->tracing = 1;
	return 1;
}

static void oprofile_end_trace(struct oprofile_cpu_buffer *cpu_buf)
{
	cpu_buf->tracing = 0;
}

void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
				unsigned long event, int is_kernel)
{
	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);

	if (!backtrace_depth) {
		log_sample(cpu_buf, pc, is_kernel, event);
		return;
	}

	if (!oprofile_begin_trace(cpu_buf))
		return;

	/* if log_sample() fail we can't backtrace since we lost the source
	 * of this event */
	if (log_sample(cpu_buf, pc, is_kernel, event))
		oprofile_ops.backtrace(regs, backtrace_depth);
	oprofile_end_trace(cpu_buf);
}

void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
{
	int is_kernel = !user_mode(regs);
	unsigned long pc = profile_pc(regs);

	oprofile_add_ext_sample(pc, regs, event, is_kernel);
}

#ifdef CONFIG_OPROFILE_IBS

#define MAX_IBS_SAMPLE_SIZE 14

void oprofile_add_ibs_sample(struct pt_regs *const regs,
			     unsigned int *const ibs_sample, int ibs_code)
{
	int is_kernel = !user_mode(regs);
	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
	struct task_struct *task;

	cpu_buf->sample_received++;

	if (nr_available_slots(cpu_buf) < MAX_IBS_SAMPLE_SIZE) {
		/* we can't backtrace since we lost the source of this event */
		cpu_buf->sample_lost_overflow++;
		return;
	}

	/* notice a switch from user->kernel or vice versa */
	if (cpu_buf->last_is_kernel != is_kernel) {
		cpu_buf->last_is_kernel = is_kernel;
		add_code(cpu_buf, is_kernel);
	}

	/* notice a task switch */
	if (!is_kernel) {
		task = current;
		if (cpu_buf->last_task != task) {
			cpu_buf->last_task = task;
			add_code(cpu_buf, (unsigned long)task);
		}
	}

	add_code(cpu_buf, ibs_code);
	add_sample(cpu_buf, ibs_sample[0], ibs_sample[1]);
	add_sample(cpu_buf, ibs_sample[2], ibs_sample[3]);
	add_sample(cpu_buf, ibs_sample[4], ibs_sample[5]);

	if (ibs_code == IBS_OP_BEGIN) {
		add_sample(cpu_buf, ibs_sample[6], ibs_sample[7]);
		add_sample(cpu_buf, ibs_sample[8], ibs_sample[9]);
		add_sample(cpu_buf, ibs_sample[10], ibs_sample[11]);
	}

	if (backtrace_depth)
		oprofile_ops.backtrace(regs, backtrace_depth);
}

#endif

void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
{
	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
	log_sample(cpu_buf, pc, is_kernel, event);
}

void oprofile_add_trace(unsigned long pc)
{
	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);

	if (!cpu_buf->tracing)
		return;

	if (nr_available_slots(cpu_buf) < 1) {
		cpu_buf->tracing = 0;
		cpu_buf->sample_lost_overflow++;
		return;
	}

	/* broken frame can give an eip with the same value as an escape code,
	 * abort the trace if we get it */
	if (pc == ESCAPE_CODE) {
		cpu_buf->tracing = 0;
		cpu_buf->backtrace_aborted++;
		return;
	}

	add_sample(cpu_buf, pc, 0);
}

/*
 * This serves to avoid cpu buffer overflow, and makes sure
 * the task mortuary progresses
 *
 * By using schedule_delayed_work_on and then schedule_delayed_work
 * we guarantee this will stay on the correct cpu
 */
static void wq_sync_buffer(struct work_struct *work)
{
	struct oprofile_cpu_buffer *b =
		container_of(work, struct oprofile_cpu_buffer, work.work);
	if (b->cpu != smp_processor_id()) {
		printk(KERN_DEBUG "WQ on CPU%d, prefer CPU%d\n",
		       smp_processor_id(), b->cpu);

		if (!cpu_online(b->cpu)) {
			cancel_delayed_work(&b->work);
			return;
		}
	}
	sync_buffer(b->cpu);

	/* don't re-add the work if we're shutting down */
	if (work_enabled)
		schedule_delayed_work(&b->work, DEFAULT_TIMER_EXPIRE);
}