perf, trace: Optimize tracepoints by removing IRQ-disable from perf/tracepoint intera...

[linux-block.git] / kernel / trace / trace_event_perf.c

/*
 * trace event based perf event profiling/tracing
 *
 * Copyright (C) 2009 Red Hat Inc, Peter Zijlstra <pzijlstr@redhat.com>
 * Copyright (C) 2009-2010 Frederic Weisbecker <fweisbec@gmail.com>
 */

#include <linux/module.h>
#include <linux/kprobes.h>
#include "trace.h"

EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);

static char *perf_trace_buf[4];

/*
 * Force it to be aligned to unsigned long to avoid misaligned accesses
 * suprises
 */
typedef typeof(unsigned long [PERF_MAX_TRACE_SIZE / sizeof(unsigned long)])
	perf_trace_t;

/* Count the events in use (per event id, not per instance) */
static int	total_ref_count;

static int perf_trace_event_enable(struct ftrace_event_call *event, void *data)
{
	int ret = -ENOMEM;

	if (event->perf_refcount++ > 0) {
		event->perf_data = NULL;
		return 0;
	}

	if (!total_ref_count) {
		char *buf;
		int i;

		for (i = 0; i < 4; i++) {
			buf = (char *)alloc_percpu(perf_trace_t);
			if (!buf)
				goto fail_buf;

			rcu_assign_pointer(perf_trace_buf[i], buf);
		}
	}

	ret = event->perf_event_enable(event);
	if (!ret) {
		event->perf_data = data;
		total_ref_count++;
		return 0;
	}

fail_buf:
	if (!total_ref_count) {
		int i;

		for (i = 0; i < 4; i++) {
			free_percpu(perf_trace_buf[i]);
			perf_trace_buf[i] = NULL;
		}
	}
	event->perf_refcount--;

	return ret;
}

int perf_trace_enable(int event_id, void *data)
{
	struct ftrace_event_call *event;
	int ret = -EINVAL;

	mutex_lock(&event_mutex);
	list_for_each_entry(event, &ftrace_events, list) {
		if (event->id == event_id && event->perf_event_enable &&
		    try_module_get(event->mod)) {
			ret = perf_trace_event_enable(event, data);
			break;
		}
	}
	mutex_unlock(&event_mutex);

	return ret;
}

static void perf_trace_event_disable(struct ftrace_event_call *event)
{
	if (--event->perf_refcount > 0)
		return;

	event->perf_event_disable(event);

	if (!--total_ref_count) {
		char *buf[4];
		int i;

		for (i = 0; i < 4; i++) {
			buf[i] = perf_trace_buf[i];
			rcu_assign_pointer(perf_trace_buf[i], NULL);
		}

		/*
		 * Ensure every events in profiling have finished before
		 * releasing the buffers
		 */
		synchronize_sched();

		for (i = 0; i < 4; i++)
			free_percpu(buf[i]);
	}
}

void perf_trace_disable(int event_id)
{
	struct ftrace_event_call *event;

	mutex_lock(&event_mutex);
	list_for_each_entry(event, &ftrace_events, list) {
		if (event->id == event_id) {
			perf_trace_event_disable(event);
			module_put(event->mod);
			break;
		}
	}
	mutex_unlock(&event_mutex);
}

__kprobes void *perf_trace_buf_prepare(int size, unsigned short type,
				       struct pt_regs *regs, int *rctxp)
{
	struct trace_entry *entry;
	char *trace_buf, *raw_data;
	int pc;

	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(unsigned long));

	pc = preempt_count();

	*rctxp = perf_swevent_get_recursion_context();
	if (*rctxp < 0)
		goto err_recursion;

	trace_buf = rcu_dereference_sched(perf_trace_buf[*rctxp]);
	if (!trace_buf)
		goto err;

	raw_data = per_cpu_ptr(trace_buf, smp_processor_id());

	/* zero the dead bytes from align to not leak stack to user */
	memset(&raw_data[size - sizeof(u64)], 0, sizeof(u64));

	entry = (struct trace_entry *)raw_data;
	tracing_generic_entry_update(entry, regs->flags, pc);
	entry->type = type;

	return raw_data;
err:
	perf_swevent_put_recursion_context(*rctxp);
err_recursion:
	return NULL;
}
EXPORT_SYMBOL_GPL(perf_trace_buf_prepare);