[linux-2.6-block.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/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;

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;
}

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->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]), &entry->irq_flags);
	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,
					 entry->irq_flags);
}

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 op_cpu_buffer_add_data(&entry, val) to add data and
 * op_cpu_buffer_write_commit(&entry) to commit the sample.
 */
void oprofile_add_data(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:
	cpu_buf->sample_lost_overflow++;
}

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>
	24	#include <linux/vmalloc.h>
	25	#include <linux/errno.h>
6a18037d	26
1da177e4 LT	27	#include "event_buffer.h"
	28	#include "cpu_buffer.h"
	29	#include "buffer_sync.h"
	30	#include "oprof.h"
	31
6dad828b RR	32	#define OP_BUFFER_FLAGS 0
	33
	34	/*
	35	* Read and write access is using spin locking. Thus, writing to the
	36	* buffer by NMI handler (x86) could occur also during critical
	37	* sections when reading the buffer. To avoid this, there are 2
	38	* buffers for independent read and write access. Read access is in
	39	* process context only, write access only in the NMI handler. If the
	40	* read buffer runs empty, both buffers are swapped atomically. There
	41	* is potentially a small window during swapping where the buffers are
	42	* disabled and samples could be lost.
	43	*
	44	* Using 2 buffers is a little bit overhead, but the solution is clear
	45	* and does not require changes in the ring buffer implementation. It
	46	* can be changed to a single buffer solution when the ring buffer
	47	* access is implemented as non-locking atomic code.
	48	*/
9966718d RR	49	static struct ring_buffer *op_ring_buffer_read;
9966718d RR	50	static struct ring_buffer *op_ring_buffer_write;
8b8b4988	51	DEFINE_PER_CPU(struct oprofile_cpu_buffer, cpu_buffer);
1da177e4	52
c4028958	53	static void wq_sync_buffer(struct work_struct *work);
1da177e4 LT	54
	55	#define DEFAULT_TIMER_EXPIRE (HZ / 10)
	56	static int work_enabled;
	57
a5598ca0 CL	58	unsigned long oprofile_get_cpu_buffer_size(void)
a5598ca0 CL	59	{
bd2172f5	60	return oprofile_cpu_buffer_size;
a5598ca0 CL	61	}
	62
	63	void oprofile_cpu_buffer_inc_smpl_lost(void)
	64	{
	65	struct oprofile_cpu_buffer *cpu_buf
	66	= &__get_cpu_var(cpu_buffer);
	67
	68	cpu_buf->sample_lost_overflow++;
	69	}
	70
30015776 RR	71	void free_cpu_buffers(void)
	72	{
	73	if (op_ring_buffer_read)
	74	ring_buffer_free(op_ring_buffer_read);
	75	op_ring_buffer_read = NULL;
	76	if (op_ring_buffer_write)
	77	ring_buffer_free(op_ring_buffer_write);
	78	op_ring_buffer_write = NULL;
	79	}
	80
1da177e4 LT	81	int alloc_cpu_buffers(void)
	82	{
	83	int i;
6a18037d	84
bd2172f5	85	unsigned long buffer_size = oprofile_cpu_buffer_size;
6a18037d	86
6dad828b RR	87	op_ring_buffer_read = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
	88	if (!op_ring_buffer_read)
	89	goto fail;
	90	op_ring_buffer_write = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
	91	if (!op_ring_buffer_write)
	92	goto fail;
	93
4bd9b9dc	94	for_each_possible_cpu(i) {
608dfddd	95	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
6a18037d	96
1da177e4 LT	97	b->last_task = NULL;
	98	b->last_is_kernel = -1;
	99	b->tracing = 0;
	100	b->buffer_size = buffer_size;
1da177e4 LT	101	b->sample_received = 0;
1da177e4 LT	102	b->sample_lost_overflow = 0;
df9d177a PE	103	b->backtrace_aborted = 0;
df9d177a PE	104	b->sample_invalid_eip = 0;
1da177e4	105	b->cpu = i;
c4028958	106	INIT_DELAYED_WORK(&b->work, wq_sync_buffer);
1da177e4 LT	107	}
	108	return 0;
	109
	110	fail:
	111	free_cpu_buffers();
	112	return -ENOMEM;
	113	}
1da177e4 LT	114
	115	void start_cpu_work(void)
	116	{
	117	int i;
	118
	119	work_enabled = 1;
	120
	121	for_each_online_cpu(i) {
608dfddd	122	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
1da177e4 LT	123
	124	/*
	125	* Spread the work by 1 jiffy per cpu so they dont all
	126	* fire at once.
	127	*/
	128	schedule_delayed_work_on(i, &b->work, DEFAULT_TIMER_EXPIRE + i);
	129	}
	130	}
	131
1da177e4 LT	132	void end_cpu_work(void)
	133	{
	134	int i;
	135
	136	work_enabled = 0;
	137
	138	for_each_online_cpu(i) {
608dfddd	139	struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
1da177e4 LT	140
	141	cancel_delayed_work(&b->work);
	142	}
	143
	144	flush_scheduled_work();
	145	}
	146
2cc28b9f RR	147	/*
	148	* This function prepares the cpu buffer to write a sample.
	149	*
	150	* Struct op_entry is used during operations on the ring buffer while
	151	* struct op_sample contains the data that is stored in the ring
	152	* buffer. Struct entry can be uninitialized. The function reserves a
	153	* data array that is specified by size. Use
	154	* op_cpu_buffer_write_commit() after preparing the sample. In case of
	155	* errors a null pointer is returned, otherwise the pointer to the
	156	* sample.
	157	*
	158	*/
	159	struct op_sample
	160	op_cpu_buffer_write_reserve(struct op_entry entry, unsigned long size)
9966718d	161	{
2cc28b9f RR	162	entry->event = ring_buffer_lock_reserve
	163	(op_ring_buffer_write, sizeof(struct op_sample) +
	164	size * sizeof(entry->sample->data[0]), &entry->irq_flags);
9966718d RR	165	if (entry->event)
	166	entry->sample = ring_buffer_event_data(entry->event);
	167	else
	168	entry->sample = NULL;
	169
	170	if (!entry->sample)
2cc28b9f	171	return NULL;
9966718d	172
2cc28b9f RR	173	entry->size = size;
	174	entry->data = entry->sample->data;
	175
	176	return entry->sample;
9966718d RR	177	}
	178
	179	int op_cpu_buffer_write_commit(struct op_entry *entry)
	180	{
	181	return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event,
	182	entry->irq_flags);
	183	}
	184
2d87b14c	185	struct op_sample op_cpu_buffer_read_entry(struct op_entry entry, int cpu)
9966718d RR	186	{
	187	struct ring_buffer_event *e;
	188	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	189	if (e)
2d87b14c	190	goto event;
9966718d RR	191	if (ring_buffer_swap_cpu(op_ring_buffer_read,
	192	op_ring_buffer_write,
	193	cpu))
	194	return NULL;
	195	e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
	196	if (e)
2d87b14c	197	goto event;
9966718d	198	return NULL;
2d87b14c RR	199
	200	event:
	201	entry->event = e;
	202	entry->sample = ring_buffer_event_data(e);
	203	entry->size = (ring_buffer_event_length(e) - sizeof(struct op_sample))
	204	/ sizeof(entry->sample->data[0]);
	205	entry->data = entry->sample->data;
	206	return entry->sample;
9966718d RR	207	}
	208
	209	unsigned long op_cpu_buffer_entries(int cpu)
	210	{
	211	return ring_buffer_entries_cpu(op_ring_buffer_read, cpu)
	212	+ ring_buffer_entries_cpu(op_ring_buffer_write, cpu);
	213	}
	214
ae735e99 RR	215	static int
	216	op_add_code(struct oprofile_cpu_buffer *cpu_buf, unsigned long backtrace,
	217	int is_kernel, struct task_struct *task)
	218	{
	219	struct op_entry entry;
	220	struct op_sample *sample;
	221	unsigned long flags;
	222	int size;
	223
	224	flags = 0;
	225
	226	if (backtrace)
	227	flags \|= TRACE_BEGIN;
	228
	229	/* notice a switch from user->kernel or vice versa */
	230	is_kernel = !!is_kernel;
	231	if (cpu_buf->last_is_kernel != is_kernel) {
	232	cpu_buf->last_is_kernel = is_kernel;
	233	flags \|= KERNEL_CTX_SWITCH;
	234	if (is_kernel)
	235	flags \|= IS_KERNEL;
	236	}
	237
	238	/* notice a task switch */
	239	if (cpu_buf->last_task != task) {
	240	cpu_buf->last_task = task;
	241	flags \|= USER_CTX_SWITCH;
	242	}
	243
	244	if (!flags)
	245	/* nothing to do */
	246	return 0;
	247
	248	if (flags & USER_CTX_SWITCH)
	249	size = 1;
	250	else
	251	size = 0;
	252
	253	sample = op_cpu_buffer_write_reserve(&entry, size);
	254	if (!sample)
	255	return -ENOMEM;
	256
	257	sample->eip = ESCAPE_CODE;
	258	sample->event = flags;
	259
	260	if (size)
d9928c25	261	op_cpu_buffer_add_data(&entry, (unsigned long)task);
ae735e99 RR	262
	263	op_cpu_buffer_write_commit(&entry);
	264
	265	return 0;
	266	}
	267
211117ff	268	static inline int
d0e23384 RR	269	op_add_sample(struct oprofile_cpu_buffer *cpu_buf,
d0e23384 RR	270	unsigned long pc, unsigned long event)
1da177e4	271	{
6dad828b	272	struct op_entry entry;
2cc28b9f	273	struct op_sample *sample;
6dad828b	274
2cc28b9f RR	275	sample = op_cpu_buffer_write_reserve(&entry, 0);
	276	if (!sample)
	277	return -ENOMEM;
6dad828b	278
2cc28b9f RR	279	sample->eip = pc;
2cc28b9f RR	280	sample->event = event;
6dad828b	281
3967e93e	282	return op_cpu_buffer_write_commit(&entry);
1da177e4 LT	283	}
1da177e4 LT	284
ae735e99 RR	285	/*
ae735e99 RR	286	* This must be safe from any context.
1da177e4 LT	287	*
	288	* is_kernel is needed because on some architectures you cannot
	289	* tell if you are in kernel or user space simply by looking at
	290	* pc. We tag this in the buffer by generating kernel enter/exit
	291	* events whenever is_kernel changes
	292	*/
ae735e99 RR	293	static int
	294	log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
	295	unsigned long backtrace, int is_kernel, unsigned long event)
1da177e4	296	{
1da177e4 LT	297	cpu_buf->sample_received++;
1da177e4 LT	298
df9d177a PE	299	if (pc == ESCAPE_CODE) {
	300	cpu_buf->sample_invalid_eip++;
	301	return 0;
	302	}
	303
ae735e99 RR	304	if (op_add_code(cpu_buf, backtrace, is_kernel, current))
ae735e99 RR	305	goto fail;
6a18037d	306
d0e23384	307	if (op_add_sample(cpu_buf, pc, event))
211117ff RR	308	goto fail;
211117ff RR	309
1da177e4	310	return 1;
211117ff RR	311
	312	fail:
	313	cpu_buf->sample_lost_overflow++;
	314	return 0;
1da177e4 LT	315	}
1da177e4 LT	316
6352d92d	317	static inline void oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4	318	{
1da177e4	319	cpu_buf->tracing = 1;
1da177e4 LT	320	}
1da177e4 LT	321
6352d92d	322	static inline void oprofile_end_trace(struct oprofile_cpu_buffer *cpu_buf)
1da177e4 LT	323	{
	324	cpu_buf->tracing = 0;
	325	}
	326
d45d23be RR	327	static inline void
	328	__oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
	329	unsigned long event, int is_kernel)
1da177e4	330	{
608dfddd	331	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
ae735e99	332	unsigned long backtrace = oprofile_backtrace_depth;
1da177e4	333
fd13f6c8 RR	334	/*
	335	* if log_sample() fail we can't backtrace since we lost the
	336	* source of this event
	337	*/
ae735e99 RR	338	if (!log_sample(cpu_buf, pc, backtrace, is_kernel, event))
	339	/* failed */
	340	return;
	341
	342	if (!backtrace)
	343	return;
6352d92d	344
ae735e99 RR	345	oprofile_begin_trace(cpu_buf);
ae735e99 RR	346	oprofile_ops.backtrace(regs, backtrace);
1da177e4 LT	347	oprofile_end_trace(cpu_buf);
	348	}
	349
d45d23be RR	350	void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
	351	unsigned long event, int is_kernel)
	352	{
	353	__oprofile_add_ext_sample(pc, regs, event, is_kernel);
	354	}
	355
27357716 BR	356	void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
	357	{
	358	int is_kernel = !user_mode(regs);
	359	unsigned long pc = profile_pc(regs);
	360
d45d23be	361	__oprofile_add_ext_sample(pc, regs, event, is_kernel);
27357716 BR	362	}
27357716 BR	363
1acda878 RR	364	/*
	365	* Add samples with data to the ring buffer.
	366	*
	367	* Use op_cpu_buffer_add_data(&entry, val) to add data and
	368	* op_cpu_buffer_write_commit(&entry) to commit the sample.
	369	*/
	370	void oprofile_add_data(struct op_entry entry, struct pt_regs const regs,
	371	unsigned long pc, int code, int size)
345c2573	372	{
1acda878	373	struct op_sample *sample;
e2fee276 RR	374	int is_kernel = !user_mode(regs);
e2fee276 RR	375	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
345c2573 BK	376
	377	cpu_buf->sample_received++;
	378
1acda878 RR	379	/* no backtraces for samples with data */
	380	if (op_add_code(cpu_buf, 0, is_kernel, current))
	381	goto fail;
	382
	383	sample = op_cpu_buffer_write_reserve(entry, size + 2);
	384	if (!sample)
	385	goto fail;
	386	sample->eip = ESCAPE_CODE;
	387	sample->event = 0; /* no flags */
345c2573	388
1acda878 RR	389	op_cpu_buffer_add_data(entry, code);
1acda878 RR	390	op_cpu_buffer_add_data(entry, pc);
345c2573	391
1acda878	392	return;
345c2573	393
1acda878 RR	394	fail:
1acda878 RR	395	cpu_buf->sample_lost_overflow++;
345c2573 BK	396	}
345c2573 BK	397
1da177e4 LT	398	void oprofile_add_pc(unsigned long pc, int is_kernel, unsigned long event)
1da177e4 LT	399	{
608dfddd	400	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
ae735e99	401	log_sample(cpu_buf, pc, 0, is_kernel, event);
1da177e4 LT	402	}
1da177e4 LT	403
1da177e4 LT	404	void oprofile_add_trace(unsigned long pc)
1da177e4 LT	405	{
608dfddd	406	struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
1da177e4 LT	407
	408	if (!cpu_buf->tracing)
	409	return;
	410
fd13f6c8 RR	411	/*
	412	* broken frame can give an eip with the same value as an
	413	* escape code, abort the trace if we get it
	414	*/
211117ff RR	415	if (pc == ESCAPE_CODE)
	416	goto fail;
	417
d0e23384	418	if (op_add_sample(cpu_buf, pc, 0))
211117ff	419	goto fail;
1da177e4	420
211117ff RR	421	return;
	422	fail:
	423	cpu_buf->tracing = 0;
	424	cpu_buf->backtrace_aborted++;
	425	return;
1da177e4 LT	426	}
1da177e4 LT	427
1da177e4 LT	428	/*
	429	* This serves to avoid cpu buffer overflow, and makes sure
	430	* the task mortuary progresses
	431	*
	432	* By using schedule_delayed_work_on and then schedule_delayed_work
	433	* we guarantee this will stay on the correct cpu
	434	*/
c4028958	435	static void wq_sync_buffer(struct work_struct *work)
1da177e4	436	{
25ad2913	437	struct oprofile_cpu_buffer *b =
c4028958	438	container_of(work, struct oprofile_cpu_buffer, work.work);
1da177e4	439	if (b->cpu != smp_processor_id()) {
bd17b625	440	printk(KERN_DEBUG "WQ on CPU%d, prefer CPU%d\n",
1da177e4	441	smp_processor_id(), b->cpu);
4bd9b9dc CA	442
	443	if (!cpu_online(b->cpu)) {
	444	cancel_delayed_work(&b->work);
	445	return;
	446	}
1da177e4 LT	447	}
	448	sync_buffer(b->cpu);
	449
	450	/* don't re-add the work if we're shutting down */
	451	if (work_enabled)
	452	schedule_delayed_work(&b->work, DEFAULT_TIMER_EXPIRE);
	453	}