[net-next-2.6.git] / drivers / oprofile / cpu_buffer.c

/**
 * @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 OP_BUFFER_FLAGS	0

/*
 * Read and write access is using spin locking. Thus, writing to the
 * buffer by NMI handler (x86) could occur also during critical
 * sections when reading the buffer. To avoid this, there are 2
 * buffers for independent read and write access. Read access is in
 * process context only, write access only in the NMI handler. If the
 * read buffer runs empty, both buffers are swapped atomically. There
 * is potentially a small window during swapping where the buffers are
 * disabled and samples could be lost.
 *
 * Using 2 buffers is a little bit overhead, but the solution is clear
 * and does not require changes in the ring buffer implementation. It
 * can be changed to a single buffer solution when the ring buffer
 * access is implemented as non-locking atomic code.
 */
static struct ring_buffer *op_ring_buffer_read;
static struct ring_buffer *op_ring_buffer_write;
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)
{
	if (op_ring_buffer_read)
		ring_buffer_free(op_ring_buffer_read);
	op_ring_buffer_read = NULL;
	if (op_ring_buffer_write)
		ring_buffer_free(op_ring_buffer_write);
	op_ring_buffer_write = NULL;
}

unsigned long oprofile_get_cpu_buffer_size(void)
{
	return oprofile_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 = oprofile_cpu_buffer_size;

	op_ring_buffer_read = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
	if (!op_ring_buffer_read)
		goto fail;
	op_ring_buffer_write = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
	if (!op_ring_buffer_write)
		goto fail;

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

		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();
}

int op_cpu_buffer_write_entry(struct op_entry *entry)
{
	entry->event = ring_buffer_lock_reserve(op_ring_buffer_write,
						sizeof(struct op_sample),
						&entry->irq_flags);
	if (entry->event)
		entry->sample = ring_buffer_event_data(entry->event);
	else
		entry->sample = NULL;

	if (!entry->sample)
		return -ENOMEM;

	return 0;
}

int op_cpu_buffer_write_commit(struct op_entry *entry)
{
	return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event,
					 entry->irq_flags);
}

struct op_sample *op_cpu_buffer_read_entry(int cpu)
{
	struct ring_buffer_event *e;
	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	if (e)
		return ring_buffer_event_data(e);
	if (ring_buffer_swap_cpu(op_ring_buffer_read,
				 op_ring_buffer_write,
				 cpu))
		return NULL;
	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	if (e)
		return ring_buffer_event_data(e);
	return NULL;
}

unsigned long op_cpu_buffer_entries(int cpu)
{
	return ring_buffer_entries_cpu(op_ring_buffer_read, cpu)
		+ ring_buffer_entries_cpu(op_ring_buffer_write, cpu);
}

static inline int
add_sample(struct oprofile_cpu_buffer *cpu_buf,
	   unsigned long pc, unsigned long event)
{
	struct op_entry entry;
	int ret;

	ret = op_cpu_buffer_write_entry(&entry);
	if (ret)
		return ret;

	entry.sample->eip = pc;
	entry.sample->event = event;

	ret = op_cpu_buffer_write_commit(&entry);
	if (ret)
		return ret;

	return 0;
}

static inline int
add_code(struct oprofile_cpu_buffer *buffer, unsigned long value)
{
	return 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;
	}

	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;
		if (add_code(cpu_buf, is_kernel))
			goto fail;
	}

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

	if (add_sample(cpu_buf, pc, event))
		goto fail;

	return 1;

fail:
	cpu_buf->sample_lost_overflow++;
	return 0;
}

static int oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
{
	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 (!oprofile_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, oprofile_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;
	int fail = 0;

	cpu_buf->sample_received++;

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

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

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

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

	if (fail)
		goto fail;

	if (oprofile_backtrace_depth)
		oprofile_ops.backtrace(regs, oprofile_backtrace_depth);

	return;

fail:
	cpu_buf->sample_lost_overflow++;
	return;
}

#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;

	/*
	 * 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)
		goto fail;

	if (add_sample(cpu_buf, pc, 0))
		goto fail;

	return;
fail:
	cpu_buf->tracing = 0;
	cpu_buf->backtrace_aborted++;
	return;
}

/*
 * 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);
}
Commit	Line	Data
1da177e4 LT	1	/**
	2	* @file cpu_buffer.c
	3	*
	4	* @remark Copyright 2002 OProfile authors
	5	* @remark Read the file COPYING
	6	*
	7	* @author John Levon <levon@movementarian.org>
345c2573	8	* @author Barry Kasindorf <barry.kasindorf@amd.com>
1da177e4 LT	9	*
	10	* Each CPU has a local buffer that stores PC value/event
	11	* pairs. We also log context switches when we notice them.
	12	* Eventually each CPU's buffer is processed into the global
	13	* event buffer by sync_buffer().
	14	*
	15	* We use a local buffer for two reasons: an NMI or similar
	16	* interrupt cannot synchronise, and high sampling rates
	17	* would lead to catastrophic global synchronisation if
	18	* a global buffer was used.
	19	*/
	20
	21	#include <linux/sched.h>
	22	#include <linux/oprofile.h>
	23	#include <linux/vmalloc.h>
	24	#include <linux/errno.h>
6a18037d	25
1da177e4 LT	26	#include "event_buffer.h"
	27	#include "cpu_buffer.h"
	28	#include "buffer_sync.h"
	29	#include "oprof.h"
	30
6dad828b RR	31	#define OP_BUFFER_FLAGS 0
	32
	33	/*
	34	* Read and write access is using spin locking. Thus, writing to the
	35	* buffer by NMI handler (x86) could occur also during critical
	36	* sections when reading the buffer. To avoid this, there are 2
	37	* buffers for independent read and write access. Read access is in
	38	* process context only, write access only in the NMI handler. If the
	39	* read buffer runs empty, both buffers are swapped atomically. There
	40	* is potentially a small window during swapping where the buffers are
	41	* disabled and samples could be lost.
	42	*
	43	* Using 2 buffers is a little bit overhead, but the solution is clear
	44	* and does not require changes in the ring buffer implementation. It
	45	* can be changed to a single buffer solution when the ring buffer
	46	* access is implemented as non-locking atomic code.
	47	*/
9966718d RR	48	static struct ring_buffer *op_ring_buffer_read;
9966718d RR	49	static struct ring_buffer *op_ring_buffer_write;
8b8b4988	50	DEFINE_PER_CPU(struct oprofile_cpu_buffer, cpu_buffer);
1da177e4	51
c4028958	52	static void wq_sync_buffer(struct work_struct *work);
1da177e4 LT	53
	54	#define DEFAULT_TIMER_EXPIRE (HZ / 10)
	55	static int work_enabled;
	56
	57	void free_cpu_buffers(void)
	58	{
6dad828b RR	59	if (op_ring_buffer_read)
	60	ring_buffer_free(op_ring_buffer_read);
	61	op_ring_buffer_read = NULL;
	62	if (op_ring_buffer_write)
	63	ring_buffer_free(op_ring_buffer_write);
	64	op_ring_buffer_write = NULL;
1da177e4	65	}
77933d72	66
a5598ca0 CL	67	unsigned long oprofile_get_cpu_buffer_size(void)
a5598ca0 CL	68	{
bd2172f5	69	return oprofile_cpu_buffer_size;
a5598ca0 CL	70	}
	71
	72	void oprofile_cpu_buffer_inc_smpl_lost(void)
	73	{
	74	struct oprofile_cpu_buffer *cpu_buf
	75	= &__get_cpu_var(cpu_buffer);
	76
	77	cpu_buf->sample_lost_overflow++;
	78	}
	79
1da177e4 LT	80	int alloc_cpu_buffers(void)
	81	{
	82	int i;
6a18037d	83
bd2172f5	84	unsigned long buffer_size = oprofile_cpu_buffer_size;
6a18037d	85
6dad828b RR	86	op_ring_buffer_read = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
	87	if (!op_ring_buffer_read)
	88	goto fail;
	89	op_ring_buffer_write = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
	90	if (!op_ring_buffer_write)
	91	goto fail;
	92
4bd9b9dc	93	for_each_possible_cpu(i) {
608dfddd	94	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
6a18037d	95
1da177e4 LT	96	b->last_task = NULL;
	97	b->last_is_kernel = -1;
	98	b->tracing = 0;
	99	b->buffer_size = buffer_size;
	100	b->tail_pos = 0;
	101	b->head_pos = 0;
	102	b->sample_received = 0;
	103	b->sample_lost_overflow = 0;
df9d177a PE	104	b->backtrace_aborted = 0;
df9d177a PE	105	b->sample_invalid_eip = 0;
1da177e4	106	b->cpu = i;
c4028958	107	INIT_DELAYED_WORK(&b->work, wq_sync_buffer);
1da177e4 LT	108	}
	109	return 0;
	110
	111	fail:
	112	free_cpu_buffers();
	113	return -ENOMEM;
	114	}
1da177e4 LT	115
	116	void start_cpu_work(void)
	117	{
	118	int i;
	119
	120	work_enabled = 1;
	121
	122	for_each_online_cpu(i) {
608dfddd	123	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
1da177e4 LT	124
	125	/*
	126	* Spread the work by 1 jiffy per cpu so they dont all
	127	* fire at once.
	128	*/
	129	schedule_delayed_work_on(i, &b->work, DEFAULT_TIMER_EXPIRE + i);
	130	}
	131	}
	132
1da177e4 LT	133	void end_cpu_work(void)
	134	{
	135	int i;
	136
	137	work_enabled = 0;
	138
	139	for_each_online_cpu(i) {
608dfddd	140	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
1da177e4 LT	141
	142	cancel_delayed_work(&b->work);
	143	}
	144
	145	flush_scheduled_work();
	146	}
	147
9966718d RR	148	int op_cpu_buffer_write_entry(struct op_entry *entry)
	149	{
	150	entry->event = ring_buffer_lock_reserve(op_ring_buffer_write,
	151	sizeof(struct op_sample),
	152	&entry->irq_flags);
	153	if (entry->event)
	154	entry->sample = ring_buffer_event_data(entry->event);
	155	else
	156	entry->sample = NULL;
	157
	158	if (!entry->sample)
	159	return -ENOMEM;
	160
	161	return 0;
	162	}
	163
	164	int op_cpu_buffer_write_commit(struct op_entry *entry)
	165	{
	166	return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event,
	167	entry->irq_flags);
	168	}
	169
	170	struct op_sample *op_cpu_buffer_read_entry(int cpu)
	171	{
	172	struct ring_buffer_event *e;
	173	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	174	if (e)
	175	return ring_buffer_event_data(e);
	176	if (ring_buffer_swap_cpu(op_ring_buffer_read,
	177	op_ring_buffer_write,
	178	cpu))
	179	return NULL;
	180	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	181	if (e)
	182	return ring_buffer_event_data(e);
	183	return NULL;
	184	}
	185
	186	unsigned long op_cpu_buffer_entries(int cpu)
	187	{
	188	return ring_buffer_entries_cpu(op_ring_buffer_read, cpu)
	189	+ ring_buffer_entries_cpu(op_ring_buffer_write, cpu);
	190	}
	191
211117ff	192	static inline int
25ad2913	193	add_sample(struct oprofile_cpu_buffer *cpu_buf,
6a18037d	194	unsigned long pc, unsigned long event)
1da177e4	195	{
6dad828b	196	struct op_entry entry;
211117ff	197	int ret;
6dad828b	198
6d2c53f3	199	ret = op_cpu_buffer_write_entry(&entry);
211117ff RR	200	if (ret)
211117ff RR	201	return ret;
6dad828b RR	202
	203	entry.sample->eip = pc;
	204	entry.sample->event = event;
	205
6d2c53f3	206	ret = op_cpu_buffer_write_commit(&entry);
211117ff RR	207	if (ret)
211117ff RR	208	return ret;
6dad828b	209
211117ff	210	return 0;
1da177e4 LT	211	}
1da177e4 LT	212
211117ff	213	static inline int
25ad2913	214	add_code(struct oprofile_cpu_buffer *buffer, unsigned long value)
1da177e4	215	{
211117ff	216	return add_sample(buffer, ESCAPE_CODE, value);
1da177e4 LT	217	}
1da177e4 LT	218
1da177e4 LT	219	/* This must be safe from any context. It's safe writing here
	220	* because of the head/tail separation of the writer and reader
	221	* of the CPU buffer.
	222	*
	223	* is_kernel is needed because on some architectures you cannot
	224	* tell if you are in kernel or user space simply by looking at
	225	* pc. We tag this in the buffer by generating kernel enter/exit
	226	* events whenever is_kernel changes
	227	*/
25ad2913	228	static int log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
1da177e4 LT	229	int is_kernel, unsigned long event)
1da177e4 LT	230	{
25ad2913	231	struct task_struct *task;
1da177e4 LT	232
	233	cpu_buf->sample_received++;
	234
df9d177a PE	235	if (pc == ESCAPE_CODE) {
	236	cpu_buf->sample_invalid_eip++;
	237	return 0;
	238	}
	239
1da177e4 LT	240	is_kernel = !!is_kernel;
	241
	242	task = current;
	243
	244	/* notice a switch from user->kernel or vice versa */
	245	if (cpu_buf->last_is_kernel != is_kernel) {
	246	cpu_buf->last_is_kernel = is_kernel;
211117ff RR	247	if (add_code(cpu_buf, is_kernel))
211117ff RR	248	goto fail;
1da177e4 LT	249	}
	250
	251	/* notice a task switch */
	252	if (cpu_buf->last_task != task) {
	253	cpu_buf->last_task = task;
211117ff RR	254	if (add_code(cpu_buf, (unsigned long)task))
211117ff RR	255	goto fail;
1da177e4	256	}
6a18037d	257
211117ff RR	258	if (add_sample(cpu_buf, pc, event))
	259	goto fail;
	260
1da177e4	261	return 1;
211117ff RR	262
	263	fail:
	264	cpu_buf->sample_lost_overflow++;
	265	return 0;
1da177e4 LT	266	}
1da177e4 LT	267
345c2573	268	static int oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4	269	{
1da177e4 LT	270	add_code(cpu_buf, CPU_TRACE_BEGIN);
	271	cpu_buf->tracing = 1;
	272	return 1;
	273	}
	274
25ad2913	275	static void oprofile_end_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4 LT	276	{
	277	cpu_buf->tracing = 0;
	278	}
	279
27357716 BR	280	void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
27357716 BR	281	unsigned long event, int is_kernel)
1da177e4	282	{
608dfddd	283	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
1da177e4	284
bd2172f5	285	if (!oprofile_backtrace_depth) {
1da177e4 LT	286	log_sample(cpu_buf, pc, is_kernel, event);
	287	return;
	288	}
	289
	290	if (!oprofile_begin_trace(cpu_buf))
	291	return;
	292
fd13f6c8 RR	293	/*
	294	* if log_sample() fail we can't backtrace since we lost the
	295	* source of this event
	296	*/
1da177e4	297	if (log_sample(cpu_buf, pc, is_kernel, event))
bd2172f5	298	oprofile_ops.backtrace(regs, oprofile_backtrace_depth);
1da177e4 LT	299	oprofile_end_trace(cpu_buf);
	300	}
	301
27357716 BR	302	void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
	303	{
	304	int is_kernel = !user_mode(regs);
	305	unsigned long pc = profile_pc(regs);
	306
	307	oprofile_add_ext_sample(pc, regs, event, is_kernel);
	308	}
	309
852402cc RR	310	#ifdef CONFIG_OPROFILE_IBS
852402cc RR	311
e2fee276 RR	312	#define MAX_IBS_SAMPLE_SIZE 14
e2fee276 RR	313
cdc1834d RR	314	void oprofile_add_ibs_sample(struct pt_regs * const regs,
cdc1834d RR	315	unsigned int * const ibs_sample, int ibs_code)
345c2573	316	{
e2fee276 RR	317	int is_kernel = !user_mode(regs);
e2fee276 RR	318	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
345c2573	319	struct task_struct *task;
211117ff	320	int fail = 0;
345c2573 BK	321
	322	cpu_buf->sample_received++;
	323
345c2573 BK	324	/* notice a switch from user->kernel or vice versa */
345c2573 BK	325	if (cpu_buf->last_is_kernel != is_kernel) {
211117ff RR	326	if (add_code(cpu_buf, is_kernel))
211117ff RR	327	goto fail;
345c2573	328	cpu_buf->last_is_kernel = is_kernel;
345c2573 BK	329	}
	330
	331	/* notice a task switch */
	332	if (!is_kernel) {
	333	task = current;
345c2573	334	if (cpu_buf->last_task != task) {
211117ff RR	335	if (add_code(cpu_buf, (unsigned long)task))
211117ff RR	336	goto fail;
345c2573	337	cpu_buf->last_task = task;
345c2573 BK	338	}
	339	}
	340
211117ff RR	341	fail = fail \|\| add_code(cpu_buf, ibs_code);
	342	fail = fail \|\| add_sample(cpu_buf, ibs_sample[0], ibs_sample[1]);
	343	fail = fail \|\| add_sample(cpu_buf, ibs_sample[2], ibs_sample[3]);
	344	fail = fail \|\| add_sample(cpu_buf, ibs_sample[4], ibs_sample[5]);
345c2573 BK	345
345c2573 BK	346	if (ibs_code == IBS_OP_BEGIN) {
211117ff RR	347	fail = fail \|\| add_sample(cpu_buf, ibs_sample[6], ibs_sample[7]);
	348	fail = fail \|\| add_sample(cpu_buf, ibs_sample[8], ibs_sample[9]);
	349	fail = fail \|\| add_sample(cpu_buf, ibs_sample[10], ibs_sample[11]);
345c2573 BK	350	}
345c2573 BK	351
211117ff RR	352	if (fail)
	353	goto fail;
	354
bd2172f5 RR	355	if (oprofile_backtrace_depth)
bd2172f5 RR	356	oprofile_ops.backtrace(regs, oprofile_backtrace_depth);
211117ff RR	357
	358	return;
	359
	360	fail:
	361	cpu_buf->sample_lost_overflow++;
	362	return;
345c2573 BK	363	}
345c2573 BK	364
852402cc RR	365	#endif
852402cc RR	366
1da177e4 LT	367	void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
1da177e4 LT	368	{
608dfddd	369	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
1da177e4 LT	370	log_sample(cpu_buf, pc, is_kernel, event);
	371	}
	372
1da177e4 LT	373	void oprofile_add_trace(unsigned long pc)
1da177e4 LT	374	{
608dfddd	375	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
1da177e4 LT	376
	377	if (!cpu_buf->tracing)
	378	return;
	379
fd13f6c8 RR	380	/*
	381	* broken frame can give an eip with the same value as an
	382	* escape code, abort the trace if we get it
	383	*/
211117ff RR	384	if (pc == ESCAPE_CODE)
	385	goto fail;
	386
	387	if (add_sample(cpu_buf, pc, 0))
	388	goto fail;
1da177e4	389
211117ff RR	390	return;
	391	fail:
	392	cpu_buf->tracing = 0;
	393	cpu_buf->backtrace_aborted++;
	394	return;
1da177e4 LT	395	}
1da177e4 LT	396
1da177e4 LT	397	/*
	398	* This serves to avoid cpu buffer overflow, and makes sure
	399	* the task mortuary progresses
	400	*
	401	* By using schedule_delayed_work_on and then schedule_delayed_work
	402	* we guarantee this will stay on the correct cpu
	403	*/
c4028958	404	static void wq_sync_buffer(struct work_struct *work)
1da177e4	405	{
25ad2913	406	struct oprofile_cpu_buffer *b =
c4028958	407	container_of(work, struct oprofile_cpu_buffer, work.work);
1da177e4	408	if (b->cpu != smp_processor_id()) {
bd17b625	409	printk(KERN_DEBUG "WQ on CPU%d, prefer CPU%d\n",
1da177e4	410	smp_processor_id(), b->cpu);
4bd9b9dc CA	411
	412	if (!cpu_online(b->cpu)) {
	413	cancel_delayed_work(&b->work);
	414	return;
	415	}
1da177e4 LT	416	}
	417	sync_buffer(b->cpu);
	418
	419	/* don't re-add the work if we're shutting down */
	420	if (work_enabled)
	421	schedule_delayed_work(&b->work, DEFAULT_TIMER_EXPIRE);
	422	}