Move headers under include/

[lttng-modules.git] / lttng-context-perf-counters.c

/* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only)
 *
 * lttng-context-perf-counters.c
 *
 * LTTng performance monitoring counters (perf-counters) integration module.
 *
 * Copyright (C) 2009-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 */

#include <linux/module.h>
#include <linux/slab.h>
#include <linux/perf_event.h>
#include <linux/list.h>
#include <linux/string.h>
#include <linux/cpu.h>
#include <linux/mm.h>
#include <lttng/lttng-events.h>
#include <ringbuffer/frontend_types.h>
#include <lttng/lttng-tracer.h>

static
size_t perf_counter_get_size(size_t offset)
{
	size_t size = 0;

	size += lib_ring_buffer_align(offset, lttng_alignof(uint64_t));
	size += sizeof(uint64_t);
	return size;
}

static
void perf_counter_record(struct lttng_ctx_field *field,
			 struct lib_ring_buffer_ctx *ctx,
			 struct lttng_channel *chan)
{
	struct perf_event *event;
	uint64_t value;

	event = field->u.perf_counter->e[ctx->cpu];
	if (likely(event)) {
		if (unlikely(event->state == PERF_EVENT_STATE_ERROR)) {
			value = 0;
		} else {
			event->pmu->read(event);
			value = local64_read(&event->count);
		}
	} else {
		/*
		 * Perf chooses not to be clever and not to support enabling a
		 * perf counter before the cpu is brought up. Therefore, we need
		 * to support having events coming (e.g. scheduler events)
		 * before the counter is setup. Write an arbitrary 0 in this
		 * case.
		 */
		value = 0;
	}
	lib_ring_buffer_align_ctx(ctx, lttng_alignof(value));
	chan->ops->event_write(ctx, &value, sizeof(value));
}

static
void overflow_callback(struct perf_event *event,
		       struct perf_sample_data *data,
		       struct pt_regs *regs)
{
}

static
void lttng_destroy_perf_counter_field(struct lttng_ctx_field *field)
{
	struct perf_event **events = field->u.perf_counter->e;

	{
		int ret;

		ret = cpuhp_state_remove_instance(lttng_hp_online,
			&field->u.perf_counter->cpuhp_online.node);
		WARN_ON(ret);
		ret = cpuhp_state_remove_instance(lttng_hp_prepare,
			&field->u.perf_counter->cpuhp_prepare.node);
		WARN_ON(ret);
	}
	kfree(field->event_field.name);
	kfree(field->u.perf_counter->attr);
	kvfree(events);
	kfree(field->u.perf_counter);
}

int lttng_cpuhp_perf_counter_online(unsigned int cpu,
		struct lttng_cpuhp_node *node)
{
	struct lttng_perf_counter_field *perf_field =
		container_of(node, struct lttng_perf_counter_field,
				cpuhp_online);
	struct perf_event **events = perf_field->e;
	struct perf_event_attr *attr = perf_field->attr;
	struct perf_event *pevent;

	pevent = perf_event_create_kernel_counter(attr,
			cpu, NULL, overflow_callback, NULL);
	if (!pevent || IS_ERR(pevent))
		return -EINVAL;
	if (pevent->state == PERF_EVENT_STATE_ERROR) {
		perf_event_release_kernel(pevent);
		return -EINVAL;
	}
	barrier();	/* Create perf counter before setting event */
	events[cpu] = pevent;
	return 0;
}

int lttng_cpuhp_perf_counter_dead(unsigned int cpu,
		struct lttng_cpuhp_node *node)
{
	struct lttng_perf_counter_field *perf_field =
		container_of(node, struct lttng_perf_counter_field,
				cpuhp_prepare);
	struct perf_event **events = perf_field->e;
	struct perf_event *pevent;

	pevent = events[cpu];
	events[cpu] = NULL;
	barrier();	/* NULLify event before perf counter teardown */
	perf_event_release_kernel(pevent);
	return 0;
}

int lttng_add_perf_counter_to_ctx(uint32_t type,
				  uint64_t config,
				  const char *name,
				  struct lttng_ctx **ctx)
{
	struct lttng_ctx_field *field;
	struct lttng_perf_counter_field *perf_field;
	struct perf_event **events;
	struct perf_event_attr *attr;
	int ret;
	char *name_alloc;

	events = kvzalloc(num_possible_cpus() * sizeof(*events), GFP_KERNEL);
	if (!events)
		return -ENOMEM;

	attr = kzalloc(sizeof(struct perf_event_attr), GFP_KERNEL);
	if (!attr) {
		ret = -ENOMEM;
		goto error_attr;
	}

	attr->type = type;
	attr->config = config;
	attr->size = sizeof(struct perf_event_attr);
	attr->pinned = 1;
	attr->disabled = 0;

	perf_field = kzalloc(sizeof(struct lttng_perf_counter_field), GFP_KERNEL);
	if (!perf_field) {
		ret = -ENOMEM;
		goto error_alloc_perf_field;
	}
	perf_field->e = events;
	perf_field->attr = attr;

	name_alloc = kstrdup(name, GFP_KERNEL);
	if (!name_alloc) {
		ret = -ENOMEM;
		goto name_alloc_error;
	}

	field = lttng_append_context(ctx);
	if (!field) {
		ret = -ENOMEM;
		goto append_context_error;
	}
	if (lttng_find_context(*ctx, name_alloc)) {
		ret = -EEXIST;
		goto find_error;
	}

	perf_field->cpuhp_prepare.component = LTTNG_CONTEXT_PERF_COUNTERS;
	ret = cpuhp_state_add_instance(lttng_hp_prepare,
		&perf_field->cpuhp_prepare.node);
	if (ret)
		goto cpuhp_prepare_error;

	perf_field->cpuhp_online.component = LTTNG_CONTEXT_PERF_COUNTERS;
	ret = cpuhp_state_add_instance(lttng_hp_online,
		&perf_field->cpuhp_online.node);
	if (ret)
		goto cpuhp_online_error;

	field->destroy = lttng_destroy_perf_counter_field;

	field->event_field.name = name_alloc;
	field->event_field.type.atype = atype_integer;
	field->event_field.type.u.integer.size = sizeof(uint64_t) * CHAR_BIT;
	field->event_field.type.u.integer.alignment = lttng_alignof(uint64_t) * CHAR_BIT;
	field->event_field.type.u.integer.signedness = lttng_is_signed_type(uint64_t);
	field->event_field.type.u.integer.reverse_byte_order = 0;
	field->event_field.type.u.integer.base = 10;
	field->event_field.type.u.integer.encoding = lttng_encode_none;
	field->get_size = perf_counter_get_size;
	field->record = perf_counter_record;
	field->u.perf_counter = perf_field;
	lttng_context_update(*ctx);

	return 0;

cpuhp_online_error:
	{
		int remove_ret;

		remove_ret = cpuhp_state_remove_instance(lttng_hp_prepare,
				&perf_field->cpuhp_prepare.node);
		WARN_ON(remove_ret);
	}
cpuhp_prepare_error:
find_error:
	lttng_remove_context_field(ctx, field);
append_context_error:
	kfree(name_alloc);
name_alloc_error:
	kfree(perf_field);
error_alloc_perf_field:
	kfree(attr);
error_attr:
	kvfree(events);
	return ret;
}