[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;
}

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