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

/**
 * @file cpu_buffer.c
 *
 * @remark Copyright 2002-2009 OProfile authors
 * @remark Read the file COPYING
 *
 * @author John Levon <levon@movementarian.org>
 * @author Barry Kasindorf <barry.kasindorf@amd.com>
 * @author Robert Richter <robert.richter@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/errno.h>

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

#define OP_BUFFER_FLAGS	0

static struct ring_buffer *op_ring_buffer;
DEFINE_PER_CPU(struct oprofile_cpu_buffer, op_cpu_buffer);

static void wq_sync_buffer(struct work_struct *work);

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

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(op_cpu_buffer);

	cpu_buf->sample_lost_overflow++;
}

void free_cpu_buffers(void)
{
	if (op_ring_buffer)
		ring_buffer_free(op_ring_buffer);
	op_ring_buffer = NULL;
}

#define RB_EVENT_HDR_SIZE 4

int alloc_cpu_buffers(void)
{
	int i;

	unsigned long buffer_size = oprofile_cpu_buffer_size;
	unsigned long byte_size = buffer_size * (sizeof(struct op_sample) +
						 RB_EVENT_HDR_SIZE);

	op_ring_buffer = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS);
	if (!op_ring_buffer)
		goto fail;

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

		b->last_task = NULL;
		b->last_is_kernel = -1;
		b->tracing = 0;
		b->buffer_size = buffer_size;
		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(op_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)
{
	work_enabled = 0;
}

void flush_cpu_work(void)
{
	int i;

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

		/* these works are per-cpu, no need for flush_sync */
		flush_delayed_work(&b->work);
	}
}

/*
 * This function prepares the cpu buffer to write a sample.
 *
 * Struct op_entry is used during operations on the ring buffer while
 * struct op_sample contains the data that is stored in the ring
 * buffer. Struct entry can be uninitialized. The function reserves a
 * data array that is specified by size. Use
 * op_cpu_buffer_write_commit() after preparing the sample. In case of
 * errors a null pointer is returned, otherwise the pointer to the
 * sample.
 *
 */
struct op_sample
*op_cpu_buffer_write_reserve(struct op_entry *entry, unsigned long size)
{
	entry->event = ring_buffer_lock_reserve
		(op_ring_buffer, sizeof(struct op_sample) +
		 size * sizeof(entry->sample->data[0]));
	if (!entry->event)
		return NULL;
	entry->sample = ring_buffer_event_data(entry->event);
	entry->size = size;
	entry->data = entry->sample->data;

	return entry->sample;
}

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

struct op_sample *op_cpu_buffer_read_entry(struct op_entry *entry, int cpu)
{
	struct ring_buffer_event *e;
	e = ring_buffer_consume(op_ring_buffer, cpu, NULL, NULL);
	if (!e)
		return NULL;

	entry->event = e;
	entry->sample = ring_buffer_event_data(e);
	entry->size = (ring_buffer_event_length(e) - sizeof(struct op_sample))
		/ sizeof(entry->sample->data[0]);
	entry->data = entry->sample->data;
	return entry->sample;
}

unsigned long op_cpu_buffer_entries(int cpu)
{
	return ring_buffer_entries_cpu(op_ring_buffer, cpu);
}

static int
op_add_code(struct oprofile_cpu_buffer *cpu_buf, unsigned long backtrace,
	    int is_kernel, struct task_struct *task)
{
	struct op_entry entry;
	struct op_sample *sample;
	unsigned long flags;
	int size;

	flags = 0;

	if (backtrace)
		flags |= TRACE_BEGIN;

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

	/* notice a task switch */
	if (cpu_buf->last_task != task) {
		cpu_buf->last_task = task;
		flags |= USER_CTX_SWITCH;
	}

	if (!flags)
		/* nothing to do */
		return 0;

	if (flags & USER_CTX_SWITCH)
		size = 1;
	else
		size = 0;

	sample = op_cpu_buffer_write_reserve(&entry, size);
	if (!sample)
		return -ENOMEM;

	sample->eip = ESCAPE_CODE;
	sample->event = flags;

	if (size)
		op_cpu_buffer_add_data(&entry, (unsigned long)task);

	op_cpu_buffer_write_commit(&entry);

	return 0;
}

static inline int
op_add_sample(struct oprofile_cpu_buffer *cpu_buf,
	      unsigned long pc, unsigned long event)
{
	struct op_entry entry;
	struct op_sample *sample;

	sample = op_cpu_buffer_write_reserve(&entry, 0);
	if (!sample)
		return -ENOMEM;

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

	return op_cpu_buffer_write_commit(&entry);
}

/*
 * This must be safe from any context.
 *
 * 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,
	   unsigned long backtrace, int is_kernel, unsigned long event)
{
	cpu_buf->sample_received++;

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

	if (op_add_code(cpu_buf, backtrace, is_kernel, current))
		goto fail;

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

	return 1;

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

static inline void oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
{
	cpu_buf->tracing = 1;
}

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

static inline 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(op_cpu_buffer);
	unsigned long backtrace = oprofile_backtrace_depth;

	/*
	 * if log_sample() fail we can't backtrace since we lost the
	 * source of this event
	 */
	if (!log_sample(cpu_buf, pc, backtrace, is_kernel, event))
		/* failed */
		return;

	if (!backtrace)
		return;

	oprofile_begin_trace(cpu_buf);
	oprofile_ops.backtrace(regs, backtrace);
	oprofile_end_trace(cpu_buf);
}

void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
			     unsigned long event, int is_kernel)
{
	__oprofile_add_ext_sample(pc, regs, event, is_kernel);
}

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

	if (likely(regs)) {
		is_kernel = !user_mode(regs);
		pc = profile_pc(regs);
	} else {
		is_kernel = 0;    /* This value will not be used */
		pc = ESCAPE_CODE; /* as this causes an early return. */
	}

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

/*
 * Add samples with data to the ring buffer.
 *
 * Use oprofile_add_data(&entry, val) to add data and
 * oprofile_write_commit(&entry) to commit the sample.
 */
void
oprofile_write_reserve(struct op_entry *entry, struct pt_regs * const regs,
		       unsigned long pc, int code, int size)
{
	struct op_sample *sample;
	int is_kernel = !user_mode(regs);
	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_cpu_buffer);

	cpu_buf->sample_received++;

	/* no backtraces for samples with data */
	if (op_add_code(cpu_buf, 0, is_kernel, current))
		goto fail;

	sample = op_cpu_buffer_write_reserve(entry, size + 2);
	if (!sample)
		goto fail;
	sample->eip = ESCAPE_CODE;
	sample->event = 0;		/* no flags */

	op_cpu_buffer_add_data(entry, code);
	op_cpu_buffer_add_data(entry, pc);

	return;

fail:
	entry->event = NULL;
	cpu_buf->sample_lost_overflow++;
}

int oprofile_add_data(struct op_entry *entry, unsigned long val)
{
	if (!entry->event)
		return 0;
	return op_cpu_buffer_add_data(entry, val);
}

int oprofile_add_data64(struct op_entry *entry, u64 val)
{
	if (!entry->event)
		return 0;
	if (op_cpu_buffer_get_size(entry) < 2)
		/*
		 * the function returns 0 to indicate a too small
		 * buffer, even if there is some space left
		 */
		return 0;
	if (!op_cpu_buffer_add_data(entry, (u32)val))
		return 0;
	return op_cpu_buffer_add_data(entry, (u32)(val >> 32));
}

int oprofile_write_commit(struct op_entry *entry)
{
	if (!entry->event)
		return -EINVAL;
	return op_cpu_buffer_write_commit(entry);
}

void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
{
	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_cpu_buffer);
	log_sample(cpu_buf, pc, 0, is_kernel, event);
}

void oprofile_add_trace(unsigned long pc)
{
	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_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 (op_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	*
2cc28b9f	4	* @remark Copyright 2002-2009 OProfile authors
1da177e4 LT	5	* @remark Read the file COPYING
	6	*
	7	* @author John Levon <levon@movementarian.org>
345c2573	8	* @author Barry Kasindorf <barry.kasindorf@amd.com>
2cc28b9f	9	* @author Robert Richter <robert.richter@amd.com>
1da177e4 LT	10	*
	11	* Each CPU has a local buffer that stores PC value/event
	12	* pairs. We also log context switches when we notice them.
	13	* Eventually each CPU's buffer is processed into the global
	14	* event buffer by sync_buffer().
	15	*
	16	* We use a local buffer for two reasons: an NMI or similar
	17	* interrupt cannot synchronise, and high sampling rates
	18	* would lead to catastrophic global synchronisation if
	19	* a global buffer was used.
	20	*/
	21
	22	#include <linux/sched.h>
	23	#include <linux/oprofile.h>
1da177e4	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
6dad828b RR	32
cb6e943c	33	static struct ring_buffer *op_ring_buffer;
b3e9f672	34	DEFINE_PER_CPU(struct oprofile_cpu_buffer, op_cpu_buffer);
1da177e4	35
c4028958	36	static void wq_sync_buffer(struct work_struct *work);
1da177e4 LT	37
	38	#define DEFAULT_TIMER_EXPIRE (HZ / 10)
	39	static int work_enabled;
	40
a5598ca0 CL	41	unsigned long oprofile_get_cpu_buffer_size(void)
a5598ca0 CL	42	{
bd2172f5	43	return oprofile_cpu_buffer_size;
a5598ca0 CL	44	}
	45
	46	void oprofile_cpu_buffer_inc_smpl_lost(void)
	47	{
b3e9f672	48	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_cpu_buffer);
a5598ca0 CL	49
	50	cpu_buf->sample_lost_overflow++;
	51	}
	52
30015776 RR	53	void free_cpu_buffers(void)
30015776 RR	54	{
cb6e943c AK	55	if (op_ring_buffer)
	56	ring_buffer_free(op_ring_buffer);
	57	op_ring_buffer = NULL;
30015776 RR	58	}
30015776 RR	59
54f2c841 RR	60	#define RB_EVENT_HDR_SIZE 4
54f2c841 RR	61
1da177e4 LT	62	int alloc_cpu_buffers(void)
	63	{
	64	int i;
6a18037d	65
bd2172f5	66	unsigned long buffer_size = oprofile_cpu_buffer_size;
54f2c841 RR	67	unsigned long byte_size = buffer_size * (sizeof(struct op_sample) +
54f2c841 RR	68	RB_EVENT_HDR_SIZE);
6a18037d	69
cb6e943c AK	70	op_ring_buffer = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS);
cb6e943c AK	71	if (!op_ring_buffer)
6dad828b RR	72	goto fail;
6dad828b RR	73
4bd9b9dc	74	for_each_possible_cpu(i) {
b3e9f672	75	struct oprofile_cpu_buffer *b = &per_cpu(op_cpu_buffer, i);
6a18037d	76
1da177e4 LT	77	b->last_task = NULL;
	78	b->last_is_kernel = -1;
	79	b->tracing = 0;
	80	b->buffer_size = buffer_size;
1da177e4 LT	81	b->sample_received = 0;
1da177e4 LT	82	b->sample_lost_overflow = 0;
df9d177a PE	83	b->backtrace_aborted = 0;
df9d177a PE	84	b->sample_invalid_eip = 0;
1da177e4	85	b->cpu = i;
c4028958	86	INIT_DELAYED_WORK(&b->work, wq_sync_buffer);
1da177e4 LT	87	}
	88	return 0;
	89
	90	fail:
	91	free_cpu_buffers();
	92	return -ENOMEM;
	93	}
1da177e4 LT	94
	95	void start_cpu_work(void)
	96	{
	97	int i;
	98
	99	work_enabled = 1;
	100
	101	for_each_online_cpu(i) {
b3e9f672	102	struct oprofile_cpu_buffer *b = &per_cpu(op_cpu_buffer, i);
1da177e4 LT	103
	104	/*
	105	* Spread the work by 1 jiffy per cpu so they dont all
	106	* fire at once.
	107	*/
	108	schedule_delayed_work_on(i, &b->work, DEFAULT_TIMER_EXPIRE + i);
	109	}
	110	}
	111
1da177e4 LT	112	void end_cpu_work(void)
1da177e4 LT	113	{
1da177e4	114	work_enabled = 0;
3d7851b3 TH	115	}
	116
	117	void flush_cpu_work(void)
	118	{
	119	int i;
1da177e4 LT	120
1da177e4 LT	121	for_each_online_cpu(i) {
b3e9f672	122	struct oprofile_cpu_buffer *b = &per_cpu(op_cpu_buffer, i);
1da177e4	123
3d7851b3 TH	124	/* these works are per-cpu, no need for flush_sync */
3d7851b3 TH	125	flush_delayed_work(&b->work);
1da177e4	126	}
1da177e4 LT	127	}
1da177e4 LT	128
2cc28b9f RR	129	/*
	130	* This function prepares the cpu buffer to write a sample.
	131	*
	132	* Struct op_entry is used during operations on the ring buffer while
	133	* struct op_sample contains the data that is stored in the ring
	134	* buffer. Struct entry can be uninitialized. The function reserves a
	135	* data array that is specified by size. Use
	136	* op_cpu_buffer_write_commit() after preparing the sample. In case of
	137	* errors a null pointer is returned, otherwise the pointer to the
	138	* sample.
	139	*
	140	*/
	141	struct op_sample
	142	op_cpu_buffer_write_reserve(struct op_entry entry, unsigned long size)
9966718d	143	{
2cc28b9f	144	entry->event = ring_buffer_lock_reserve
cb6e943c	145	(op_ring_buffer, sizeof(struct op_sample) +
304cc6ae	146	size * sizeof(entry->sample->data[0]));
cb6e943c	147	if (!entry->event)
2cc28b9f	148	return NULL;
cb6e943c	149	entry->sample = ring_buffer_event_data(entry->event);
2cc28b9f RR	150	entry->size = size;
	151	entry->data = entry->sample->data;
	152
	153	return entry->sample;
9966718d RR	154	}
	155
	156	int op_cpu_buffer_write_commit(struct op_entry *entry)
	157	{
cb6e943c	158	return ring_buffer_unlock_commit(op_ring_buffer, entry->event);
9966718d RR	159	}
9966718d RR	160
2d87b14c	161	struct op_sample op_cpu_buffer_read_entry(struct op_entry entry, int cpu)
9966718d RR	162	{
9966718d RR	163	struct ring_buffer_event *e;
b971f061	164	e = ring_buffer_consume(op_ring_buffer, cpu, NULL, NULL);
cb6e943c	165	if (!e)
9966718d	166	return NULL;
2d87b14c	167
2d87b14c RR	168	entry->event = e;
	169	entry->sample = ring_buffer_event_data(e);
	170	entry->size = (ring_buffer_event_length(e) - sizeof(struct op_sample))
	171	/ sizeof(entry->sample->data[0]);
	172	entry->data = entry->sample->data;
	173	return entry->sample;
9966718d RR	174	}
	175
	176	unsigned long op_cpu_buffer_entries(int cpu)
	177	{
cb6e943c	178	return ring_buffer_entries_cpu(op_ring_buffer, cpu);
9966718d RR	179	}
9966718d RR	180
ae735e99 RR	181	static int
	182	op_add_code(struct oprofile_cpu_buffer *cpu_buf, unsigned long backtrace,
	183	int is_kernel, struct task_struct *task)
	184	{
	185	struct op_entry entry;
	186	struct op_sample *sample;
	187	unsigned long flags;
	188	int size;
	189
	190	flags = 0;
	191
	192	if (backtrace)
	193	flags \|= TRACE_BEGIN;
	194
	195	/* notice a switch from user->kernel or vice versa */
	196	is_kernel = !!is_kernel;
	197	if (cpu_buf->last_is_kernel != is_kernel) {
	198	cpu_buf->last_is_kernel = is_kernel;
	199	flags \|= KERNEL_CTX_SWITCH;
	200	if (is_kernel)
	201	flags \|= IS_KERNEL;
	202	}
	203
	204	/* notice a task switch */
	205	if (cpu_buf->last_task != task) {
	206	cpu_buf->last_task = task;
	207	flags \|= USER_CTX_SWITCH;
	208	}
	209
	210	if (!flags)
	211	/* nothing to do */
	212	return 0;
	213
	214	if (flags & USER_CTX_SWITCH)
	215	size = 1;
	216	else
	217	size = 0;
	218
	219	sample = op_cpu_buffer_write_reserve(&entry, size);
	220	if (!sample)
	221	return -ENOMEM;
	222
	223	sample->eip = ESCAPE_CODE;
	224	sample->event = flags;
	225
	226	if (size)
d9928c25	227	op_cpu_buffer_add_data(&entry, (unsigned long)task);
ae735e99 RR	228
	229	op_cpu_buffer_write_commit(&entry);
	230
	231	return 0;
	232	}
	233
211117ff	234	static inline int
d0e23384 RR	235	op_add_sample(struct oprofile_cpu_buffer *cpu_buf,
d0e23384 RR	236	unsigned long pc, unsigned long event)
1da177e4	237	{
6dad828b	238	struct op_entry entry;
2cc28b9f	239	struct op_sample *sample;
6dad828b	240
2cc28b9f RR	241	sample = op_cpu_buffer_write_reserve(&entry, 0);
	242	if (!sample)
	243	return -ENOMEM;
6dad828b	244
2cc28b9f RR	245	sample->eip = pc;
2cc28b9f RR	246	sample->event = event;
6dad828b	247
3967e93e	248	return op_cpu_buffer_write_commit(&entry);
1da177e4 LT	249	}
1da177e4 LT	250
ae735e99 RR	251	/*
ae735e99 RR	252	* This must be safe from any context.
1da177e4 LT	253	*
	254	* is_kernel is needed because on some architectures you cannot
	255	* tell if you are in kernel or user space simply by looking at
	256	* pc. We tag this in the buffer by generating kernel enter/exit
	257	* events whenever is_kernel changes
	258	*/
ae735e99 RR	259	static int
	260	log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
	261	unsigned long backtrace, int is_kernel, unsigned long event)
1da177e4	262	{
1da177e4 LT	263	cpu_buf->sample_received++;
1da177e4 LT	264
df9d177a PE	265	if (pc == ESCAPE_CODE) {
	266	cpu_buf->sample_invalid_eip++;
	267	return 0;
	268	}
	269
ae735e99 RR	270	if (op_add_code(cpu_buf, backtrace, is_kernel, current))
ae735e99 RR	271	goto fail;
6a18037d	272
d0e23384	273	if (op_add_sample(cpu_buf, pc, event))
211117ff RR	274	goto fail;
211117ff RR	275
1da177e4	276	return 1;
211117ff RR	277
	278	fail:
	279	cpu_buf->sample_lost_overflow++;
	280	return 0;
1da177e4 LT	281	}
1da177e4 LT	282
6352d92d	283	static inline void oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4	284	{
1da177e4	285	cpu_buf->tracing = 1;
1da177e4 LT	286	}
1da177e4 LT	287
6352d92d	288	static inline void oprofile_end_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4 LT	289	{
	290	cpu_buf->tracing = 0;
	291	}
	292
d45d23be RR	293	static inline void
	294	__oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
	295	unsigned long event, int is_kernel)
1da177e4	296	{
b3e9f672	297	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_cpu_buffer);
ae735e99	298	unsigned long backtrace = oprofile_backtrace_depth;
1da177e4	299
fd13f6c8 RR	300	/*
	301	* if log_sample() fail we can't backtrace since we lost the
	302	* source of this event
	303	*/
ae735e99 RR	304	if (!log_sample(cpu_buf, pc, backtrace, is_kernel, event))
	305	/* failed */
	306	return;
	307
	308	if (!backtrace)
	309	return;
6352d92d	310
ae735e99 RR	311	oprofile_begin_trace(cpu_buf);
ae735e99 RR	312	oprofile_ops.backtrace(regs, backtrace);
1da177e4 LT	313	oprofile_end_trace(cpu_buf);
	314	}
	315
d45d23be RR	316	void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
	317	unsigned long event, int is_kernel)
	318	{
	319	__oprofile_add_ext_sample(pc, regs, event, is_kernel);
	320	}
	321
27357716 BR	322	void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
27357716 BR	323	{
9414e996 PC	324	int is_kernel;
	325	unsigned long pc;
	326
	327	if (likely(regs)) {
	328	is_kernel = !user_mode(regs);
	329	pc = profile_pc(regs);
	330	} else {
	331	is_kernel = 0; /* This value will not be used */
	332	pc = ESCAPE_CODE; /* as this causes an early return. */
	333	}
27357716	334
d45d23be	335	__oprofile_add_ext_sample(pc, regs, event, is_kernel);
27357716 BR	336	}
27357716 BR	337
1acda878 RR	338	/*
	339	* Add samples with data to the ring buffer.
	340	*
14f0ca8e RR	341	* Use oprofile_add_data(&entry, val) to add data and
14f0ca8e RR	342	* oprofile_write_commit(&entry) to commit the sample.
1acda878	343	*/
14f0ca8e RR	344	void
14f0ca8e RR	345	oprofile_write_reserve(struct op_entry entry, struct pt_regs const regs,
1acda878	346	unsigned long pc, int code, int size)
345c2573	347	{
1acda878	348	struct op_sample *sample;
e2fee276	349	int is_kernel = !user_mode(regs);
b3e9f672	350	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_cpu_buffer);
345c2573 BK	351
	352	cpu_buf->sample_received++;
	353
1acda878 RR	354	/* no backtraces for samples with data */
	355	if (op_add_code(cpu_buf, 0, is_kernel, current))
	356	goto fail;
	357
	358	sample = op_cpu_buffer_write_reserve(entry, size + 2);
	359	if (!sample)
	360	goto fail;
	361	sample->eip = ESCAPE_CODE;
	362	sample->event = 0; /* no flags */
345c2573	363
1acda878 RR	364	op_cpu_buffer_add_data(entry, code);
1acda878 RR	365	op_cpu_buffer_add_data(entry, pc);
345c2573	366
1acda878	367	return;
345c2573	368
1acda878	369	fail:
fdb6a8f4	370	entry->event = NULL;
1acda878	371	cpu_buf->sample_lost_overflow++;
345c2573 BK	372	}
345c2573 BK	373
14f0ca8e RR	374	int oprofile_add_data(struct op_entry *entry, unsigned long val)
14f0ca8e RR	375	{
fdb6a8f4 RR	376	if (!entry->event)
fdb6a8f4 RR	377	return 0;
14f0ca8e RR	378	return op_cpu_buffer_add_data(entry, val);
	379	}
	380
51563a0e RR	381	int oprofile_add_data64(struct op_entry *entry, u64 val)
	382	{
	383	if (!entry->event)
	384	return 0;
	385	if (op_cpu_buffer_get_size(entry) < 2)
	386	/*
	387	* the function returns 0 to indicate a too small
	388	* buffer, even if there is some space left
	389	*/
	390	return 0;
	391	if (!op_cpu_buffer_add_data(entry, (u32)val))
	392	return 0;
	393	return op_cpu_buffer_add_data(entry, (u32)(val >> 32));
	394	}
	395
14f0ca8e RR	396	int oprofile_write_commit(struct op_entry *entry)
14f0ca8e RR	397	{
fdb6a8f4 RR	398	if (!entry->event)
fdb6a8f4 RR	399	return -EINVAL;
14f0ca8e RR	400	return op_cpu_buffer_write_commit(entry);
	401	}
	402
1da177e4 LT	403	void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
1da177e4 LT	404	{
b3e9f672	405	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_cpu_buffer);
ae735e99	406	log_sample(cpu_buf, pc, 0, is_kernel, event);
1da177e4 LT	407	}
1da177e4 LT	408
1da177e4 LT	409	void oprofile_add_trace(unsigned long pc)
1da177e4 LT	410	{
b3e9f672	411	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(op_cpu_buffer);
1da177e4 LT	412
	413	if (!cpu_buf->tracing)
	414	return;
	415
fd13f6c8 RR	416	/*
	417	* broken frame can give an eip with the same value as an
	418	* escape code, abort the trace if we get it
	419	*/
211117ff RR	420	if (pc == ESCAPE_CODE)
	421	goto fail;
	422
d0e23384	423	if (op_add_sample(cpu_buf, pc, 0))
211117ff	424	goto fail;
1da177e4	425
211117ff RR	426	return;
	427	fail:
	428	cpu_buf->tracing = 0;
	429	cpu_buf->backtrace_aborted++;
	430	return;
1da177e4 LT	431	}
1da177e4 LT	432
1da177e4 LT	433	/*
	434	* This serves to avoid cpu buffer overflow, and makes sure
	435	* the task mortuary progresses
	436	*
	437	* By using schedule_delayed_work_on and then schedule_delayed_work
	438	* we guarantee this will stay on the correct cpu
	439	*/
c4028958	440	static void wq_sync_buffer(struct work_struct *work)
1da177e4	441	{
25ad2913	442	struct oprofile_cpu_buffer *b =
c4028958	443	container_of(work, struct oprofile_cpu_buffer, work.work);
1da177e4	444	if (b->cpu != smp_processor_id()) {
bd17b625	445	printk(KERN_DEBUG "WQ on CPU%d, prefer CPU%d\n",
1da177e4	446	smp_processor_id(), b->cpu);
4bd9b9dc CA	447
	448	if (!cpu_online(b->cpu)) {
	449	cancel_delayed_work(&b->work);
	450	return;
	451	}
1da177e4 LT	452	}
	453	sync_buffer(b->cpu);
	454
	455	/* don't re-add the work if we're shutting down */
	456	if (work_enabled)
	457	schedule_delayed_work(&b->work, DEFAULT_TIMER_EXPIRE);
	458	}