[deliverable/linux.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

/*
 * 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;

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++;
}

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;
}

#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_read = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS);
	if (!op_ring_buffer_read)
		goto fail;
	op_ring_buffer_write = ring_buffer_alloc(byte_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->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();
}

/*
 * 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_write, sizeof(struct op_sample) +
		 size * sizeof(entry->sample->data[0]));
	if (entry->event)
		entry->sample = ring_buffer_event_data(entry->event);
	else
		entry->sample = NULL;

	if (!entry->sample)
		return NULL;

	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_write, 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_read, cpu, NULL);
	if (e)
		goto event;
	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)
		goto event;
	return NULL;

event:
	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_read, cpu)
		+ ring_buffer_entries_cpu(op_ring_buffer_write, 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(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 = !user_mode(regs);
	unsigned long pc = profile_pc(regs);

	__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(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(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(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
	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
a5598ca0 CL	57	unsigned long oprofile_get_cpu_buffer_size(void)
a5598ca0 CL	58	{
bd2172f5	59	return oprofile_cpu_buffer_size;
a5598ca0 CL	60	}
	61
	62	void oprofile_cpu_buffer_inc_smpl_lost(void)
	63	{
	64	struct oprofile_cpu_buffer *cpu_buf
	65	= &__get_cpu_var(cpu_buffer);
	66
	67	cpu_buf->sample_lost_overflow++;
	68	}
	69
30015776 RR	70	void free_cpu_buffers(void)
	71	{
	72	if (op_ring_buffer_read)
	73	ring_buffer_free(op_ring_buffer_read);
	74	op_ring_buffer_read = NULL;
	75	if (op_ring_buffer_write)
	76	ring_buffer_free(op_ring_buffer_write);
	77	op_ring_buffer_write = NULL;
	78	}
	79
54f2c841 RR	80	#define RB_EVENT_HDR_SIZE 4
54f2c841 RR	81
1da177e4 LT	82	int alloc_cpu_buffers(void)
	83	{
	84	int i;
6a18037d	85
bd2172f5	86	unsigned long buffer_size = oprofile_cpu_buffer_size;
54f2c841 RR	87	unsigned long byte_size = buffer_size * (sizeof(struct op_sample) +
54f2c841 RR	88	RB_EVENT_HDR_SIZE);
6a18037d	89
54f2c841	90	op_ring_buffer_read = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS);
6dad828b RR	91	if (!op_ring_buffer_read)
6dad828b RR	92	goto fail;
54f2c841	93	op_ring_buffer_write = ring_buffer_alloc(byte_size, OP_BUFFER_FLAGS);
6dad828b RR	94	if (!op_ring_buffer_write)
	95	goto fail;
	96
4bd9b9dc	97	for_each_possible_cpu(i) {
608dfddd	98	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
6a18037d	99
1da177e4 LT	100	b->last_task = NULL;
	101	b->last_is_kernel = -1;
	102	b->tracing = 0;
	103	b->buffer_size = buffer_size;
1da177e4 LT	104	b->sample_received = 0;
1da177e4 LT	105	b->sample_lost_overflow = 0;
df9d177a PE	106	b->backtrace_aborted = 0;
df9d177a PE	107	b->sample_invalid_eip = 0;
1da177e4	108	b->cpu = i;
c4028958	109	INIT_DELAYED_WORK(&b->work, wq_sync_buffer);
1da177e4 LT	110	}
	111	return 0;
	112
	113	fail:
	114	free_cpu_buffers();
	115	return -ENOMEM;
	116	}
1da177e4 LT	117
	118	void start_cpu_work(void)
	119	{
	120	int i;
	121
	122	work_enabled = 1;
	123
	124	for_each_online_cpu(i) {
608dfddd	125	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
1da177e4 LT	126
	127	/*
	128	* Spread the work by 1 jiffy per cpu so they dont all
	129	* fire at once.
	130	*/
	131	schedule_delayed_work_on(i, &b->work, DEFAULT_TIMER_EXPIRE + i);
	132	}
	133	}
	134
1da177e4 LT	135	void end_cpu_work(void)
	136	{
	137	int i;
	138
	139	work_enabled = 0;
	140
	141	for_each_online_cpu(i) {
608dfddd	142	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
1da177e4 LT	143
	144	cancel_delayed_work(&b->work);
	145	}
	146
	147	flush_scheduled_work();
	148	}
	149
2cc28b9f RR	150	/*
	151	* This function prepares the cpu buffer to write a sample.
	152	*
	153	* Struct op_entry is used during operations on the ring buffer while
	154	* struct op_sample contains the data that is stored in the ring
	155	* buffer. Struct entry can be uninitialized. The function reserves a
	156	* data array that is specified by size. Use
	157	* op_cpu_buffer_write_commit() after preparing the sample. In case of
	158	* errors a null pointer is returned, otherwise the pointer to the
	159	* sample.
	160	*
	161	*/
	162	struct op_sample
	163	op_cpu_buffer_write_reserve(struct op_entry entry, unsigned long size)
9966718d	164	{
2cc28b9f RR	165	entry->event = ring_buffer_lock_reserve
2cc28b9f RR	166	(op_ring_buffer_write, sizeof(struct op_sample) +
304cc6ae	167	size * sizeof(entry->sample->data[0]));
9966718d RR	168	if (entry->event)
	169	entry->sample = ring_buffer_event_data(entry->event);
	170	else
	171	entry->sample = NULL;
	172
	173	if (!entry->sample)
2cc28b9f	174	return NULL;
9966718d	175
2cc28b9f RR	176	entry->size = size;
	177	entry->data = entry->sample->data;
	178
	179	return entry->sample;
9966718d RR	180	}
	181
	182	int op_cpu_buffer_write_commit(struct op_entry *entry)
	183	{
304cc6ae	184	return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event);
9966718d RR	185	}
9966718d RR	186
2d87b14c	187	struct op_sample op_cpu_buffer_read_entry(struct op_entry entry, int cpu)
9966718d RR	188	{
	189	struct ring_buffer_event *e;
	190	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	191	if (e)
2d87b14c	192	goto event;
9966718d RR	193	if (ring_buffer_swap_cpu(op_ring_buffer_read,
	194	op_ring_buffer_write,
	195	cpu))
	196	return NULL;
	197	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	198	if (e)
2d87b14c	199	goto event;
9966718d	200	return NULL;
2d87b14c RR	201
	202	event:
	203	entry->event = e;
	204	entry->sample = ring_buffer_event_data(e);
	205	entry->size = (ring_buffer_event_length(e) - sizeof(struct op_sample))
	206	/ sizeof(entry->sample->data[0]);
	207	entry->data = entry->sample->data;
	208	return entry->sample;
9966718d RR	209	}
	210
	211	unsigned long op_cpu_buffer_entries(int cpu)
	212	{
	213	return ring_buffer_entries_cpu(op_ring_buffer_read, cpu)
	214	+ ring_buffer_entries_cpu(op_ring_buffer_write, cpu);
	215	}
	216
ae735e99 RR	217	static int
	218	op_add_code(struct oprofile_cpu_buffer *cpu_buf, unsigned long backtrace,
	219	int is_kernel, struct task_struct *task)
	220	{
	221	struct op_entry entry;
	222	struct op_sample *sample;
	223	unsigned long flags;
	224	int size;
	225
	226	flags = 0;
	227
	228	if (backtrace)
	229	flags \|= TRACE_BEGIN;
	230
	231	/* notice a switch from user->kernel or vice versa */
	232	is_kernel = !!is_kernel;
	233	if (cpu_buf->last_is_kernel != is_kernel) {
	234	cpu_buf->last_is_kernel = is_kernel;
	235	flags \|= KERNEL_CTX_SWITCH;
	236	if (is_kernel)
	237	flags \|= IS_KERNEL;
	238	}
	239
	240	/* notice a task switch */
	241	if (cpu_buf->last_task != task) {
	242	cpu_buf->last_task = task;
	243	flags \|= USER_CTX_SWITCH;
	244	}
	245
	246	if (!flags)
	247	/* nothing to do */
	248	return 0;
	249
	250	if (flags & USER_CTX_SWITCH)
	251	size = 1;
	252	else
	253	size = 0;
	254
	255	sample = op_cpu_buffer_write_reserve(&entry, size);
	256	if (!sample)
	257	return -ENOMEM;
	258
	259	sample->eip = ESCAPE_CODE;
	260	sample->event = flags;
	261
	262	if (size)
d9928c25	263	op_cpu_buffer_add_data(&entry, (unsigned long)task);
ae735e99 RR	264
	265	op_cpu_buffer_write_commit(&entry);
	266
	267	return 0;
	268	}
	269
211117ff	270	static inline int
d0e23384 RR	271	op_add_sample(struct oprofile_cpu_buffer *cpu_buf,
d0e23384 RR	272	unsigned long pc, unsigned long event)
1da177e4	273	{
6dad828b	274	struct op_entry entry;
2cc28b9f	275	struct op_sample *sample;
6dad828b	276
2cc28b9f RR	277	sample = op_cpu_buffer_write_reserve(&entry, 0);
	278	if (!sample)
	279	return -ENOMEM;
6dad828b	280
2cc28b9f RR	281	sample->eip = pc;
2cc28b9f RR	282	sample->event = event;
6dad828b	283
3967e93e	284	return op_cpu_buffer_write_commit(&entry);
1da177e4 LT	285	}
1da177e4 LT	286
ae735e99 RR	287	/*
ae735e99 RR	288	* This must be safe from any context.
1da177e4 LT	289	*
	290	* is_kernel is needed because on some architectures you cannot
	291	* tell if you are in kernel or user space simply by looking at
	292	* pc. We tag this in the buffer by generating kernel enter/exit
	293	* events whenever is_kernel changes
	294	*/
ae735e99 RR	295	static int
	296	log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
	297	unsigned long backtrace, int is_kernel, unsigned long event)
1da177e4	298	{
1da177e4 LT	299	cpu_buf->sample_received++;
1da177e4 LT	300
df9d177a PE	301	if (pc == ESCAPE_CODE) {
	302	cpu_buf->sample_invalid_eip++;
	303	return 0;
	304	}
	305
ae735e99 RR	306	if (op_add_code(cpu_buf, backtrace, is_kernel, current))
ae735e99 RR	307	goto fail;
6a18037d	308
d0e23384	309	if (op_add_sample(cpu_buf, pc, event))
211117ff RR	310	goto fail;
211117ff RR	311
1da177e4	312	return 1;
211117ff RR	313
	314	fail:
	315	cpu_buf->sample_lost_overflow++;
	316	return 0;
1da177e4 LT	317	}
1da177e4 LT	318
6352d92d	319	static inline void oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4	320	{
1da177e4	321	cpu_buf->tracing = 1;
1da177e4 LT	322	}
1da177e4 LT	323
6352d92d	324	static inline void oprofile_end_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4 LT	325	{
	326	cpu_buf->tracing = 0;
	327	}
	328
d45d23be RR	329	static inline void
	330	__oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
	331	unsigned long event, int is_kernel)
1da177e4	332	{
608dfddd	333	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
ae735e99	334	unsigned long backtrace = oprofile_backtrace_depth;
1da177e4	335
fd13f6c8 RR	336	/*
	337	* if log_sample() fail we can't backtrace since we lost the
	338	* source of this event
	339	*/
ae735e99 RR	340	if (!log_sample(cpu_buf, pc, backtrace, is_kernel, event))
	341	/* failed */
	342	return;
	343
	344	if (!backtrace)
	345	return;
6352d92d	346
ae735e99 RR	347	oprofile_begin_trace(cpu_buf);
ae735e99 RR	348	oprofile_ops.backtrace(regs, backtrace);
1da177e4 LT	349	oprofile_end_trace(cpu_buf);
	350	}
	351
d45d23be RR	352	void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
	353	unsigned long event, int is_kernel)
	354	{
	355	__oprofile_add_ext_sample(pc, regs, event, is_kernel);
	356	}
	357
27357716 BR	358	void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
	359	{
	360	int is_kernel = !user_mode(regs);
	361	unsigned long pc = profile_pc(regs);
	362
d45d23be	363	__oprofile_add_ext_sample(pc, regs, event, is_kernel);
27357716 BR	364	}
27357716 BR	365
1acda878 RR	366	/*
	367	* Add samples with data to the ring buffer.
	368	*
14f0ca8e RR	369	* Use oprofile_add_data(&entry, val) to add data and
14f0ca8e RR	370	* oprofile_write_commit(&entry) to commit the sample.
1acda878	371	*/
14f0ca8e RR	372	void
14f0ca8e RR	373	oprofile_write_reserve(struct op_entry entry, struct pt_regs const regs,
1acda878	374	unsigned long pc, int code, int size)
345c2573	375	{
1acda878	376	struct op_sample *sample;
e2fee276 RR	377	int is_kernel = !user_mode(regs);
e2fee276 RR	378	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
345c2573 BK	379
	380	cpu_buf->sample_received++;
	381
1acda878 RR	382	/* no backtraces for samples with data */
	383	if (op_add_code(cpu_buf, 0, is_kernel, current))
	384	goto fail;
	385
	386	sample = op_cpu_buffer_write_reserve(entry, size + 2);
	387	if (!sample)
	388	goto fail;
	389	sample->eip = ESCAPE_CODE;
	390	sample->event = 0; /* no flags */
345c2573	391
1acda878 RR	392	op_cpu_buffer_add_data(entry, code);
1acda878 RR	393	op_cpu_buffer_add_data(entry, pc);
345c2573	394
1acda878	395	return;
345c2573	396
1acda878	397	fail:
fdb6a8f4	398	entry->event = NULL;
1acda878	399	cpu_buf->sample_lost_overflow++;
345c2573 BK	400	}
345c2573 BK	401
14f0ca8e RR	402	int oprofile_add_data(struct op_entry *entry, unsigned long val)
14f0ca8e RR	403	{
fdb6a8f4 RR	404	if (!entry->event)
fdb6a8f4 RR	405	return 0;
14f0ca8e RR	406	return op_cpu_buffer_add_data(entry, val);
	407	}
	408
51563a0e RR	409	int oprofile_add_data64(struct op_entry *entry, u64 val)
	410	{
	411	if (!entry->event)
	412	return 0;
	413	if (op_cpu_buffer_get_size(entry) < 2)
	414	/*
	415	* the function returns 0 to indicate a too small
	416	* buffer, even if there is some space left
	417	*/
	418	return 0;
	419	if (!op_cpu_buffer_add_data(entry, (u32)val))
	420	return 0;
	421	return op_cpu_buffer_add_data(entry, (u32)(val >> 32));
	422	}
	423
14f0ca8e RR	424	int oprofile_write_commit(struct op_entry *entry)
14f0ca8e RR	425	{
fdb6a8f4 RR	426	if (!entry->event)
fdb6a8f4 RR	427	return -EINVAL;
14f0ca8e RR	428	return op_cpu_buffer_write_commit(entry);
	429	}
	430
1da177e4 LT	431	void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
1da177e4 LT	432	{
608dfddd	433	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
ae735e99	434	log_sample(cpu_buf, pc, 0, is_kernel, event);
1da177e4 LT	435	}
1da177e4 LT	436
1da177e4 LT	437	void oprofile_add_trace(unsigned long pc)
1da177e4 LT	438	{
608dfddd	439	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
1da177e4 LT	440
	441	if (!cpu_buf->tracing)
	442	return;
	443
fd13f6c8 RR	444	/*
	445	* broken frame can give an eip with the same value as an
	446	* escape code, abort the trace if we get it
	447	*/
211117ff RR	448	if (pc == ESCAPE_CODE)
	449	goto fail;
	450
d0e23384	451	if (op_add_sample(cpu_buf, pc, 0))
211117ff	452	goto fail;
1da177e4	453
211117ff RR	454	return;
	455	fail:
	456	cpu_buf->tracing = 0;
	457	cpu_buf->backtrace_aborted++;
	458	return;
1da177e4 LT	459	}
1da177e4 LT	460
1da177e4 LT	461	/*
	462	* This serves to avoid cpu buffer overflow, and makes sure
	463	* the task mortuary progresses
	464	*
	465	* By using schedule_delayed_work_on and then schedule_delayed_work
	466	* we guarantee this will stay on the correct cpu
	467	*/
c4028958	468	static void wq_sync_buffer(struct work_struct *work)
1da177e4	469	{
25ad2913	470	struct oprofile_cpu_buffer *b =
c4028958	471	container_of(work, struct oprofile_cpu_buffer, work.work);
1da177e4	472	if (b->cpu != smp_processor_id()) {
bd17b625	473	printk(KERN_DEBUG "WQ on CPU%d, prefer CPU%d\n",
1da177e4	474	smp_processor_id(), b->cpu);
4bd9b9dc CA	475
	476	if (!cpu_online(b->cpu)) {
	477	cancel_delayed_work(&b->work);
	478	return;
	479	}
1da177e4 LT	480	}
	481	sync_buffer(b->cpu);
	482
	483	/* don't re-add the work if we're shutting down */
	484	if (work_enabled)
	485	schedule_delayed_work(&b->work, DEFAULT_TIMER_EXPIRE);
	486	}