Re: [PATCH v20 2/5] ring-buffer: Introducing ring-buffer mapping functions

[Steven Rostedt <rostedt@goodmis.org>]

On Sat,  6 Apr 2024 18:36:46 +0100
Vincent Donnefort <vdonnefort@google.com> wrote:

> +int ring_buffer_map(struct trace_buffer *buffer, int cpu,
> +		    struct vm_area_struct *vma)
> +{
> +	struct ring_buffer_per_cpu *cpu_buffer;
> +	unsigned long flags, *subbuf_ids;
> +	int err = 0;
> +
> +	if (!cpumask_test_cpu(cpu, buffer->cpumask))
> +		return -EINVAL;
> +
> +	cpu_buffer = buffer->buffers[cpu];
> +
> +	mutex_lock(&cpu_buffer->mapping_lock);
> +
> +	if (cpu_buffer->mapped) {
> +		err = __rb_map_vma(cpu_buffer, vma);
> +		if (!err)
> +			err = __rb_inc_dec_mapped(cpu_buffer, true);
> +		mutex_unlock(&cpu_buffer->mapping_lock);
> +		return err;
> +	}
> +
> +	/* prevent another thread from changing buffer/sub-buffer sizes */
> +	mutex_lock(&buffer->mutex);
> +
> +	err = rb_alloc_meta_page(cpu_buffer);
> +	if (err)
> +		goto unlock;
> +
> +	/* subbuf_ids include the reader while nr_pages does not */
> +	subbuf_ids = kcalloc(cpu_buffer->nr_pages + 1, sizeof(*subbuf_ids), GFP_KERNEL);
> +	if (!subbuf_ids) {
> +		rb_free_meta_page(cpu_buffer);
> +		err = -ENOMEM;
> +		goto unlock;
> +	}
> +
> +	atomic_inc(&cpu_buffer->resize_disabled);
> +
> +	/*
> +	 * Lock all readers to block any subbuf swap until the subbuf IDs are
> +	 * assigned.
> +	 */
> +	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
> +	rb_setup_ids_meta_page(cpu_buffer, subbuf_ids);
> +	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
> +
> +	err = __rb_map_vma(cpu_buffer, vma);
> +	if (!err) {
> +		raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
> +		cpu_buffer->mapped = 1;
> +		raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
> +	} else {
> +		kfree(cpu_buffer->subbuf_ids);
> +		cpu_buffer->subbuf_ids = NULL;
> +		rb_free_meta_page(cpu_buffer);
> +	}
> +unlock:

Nit: For all labels, please add a space before them. Otherwise, diffs will
show "unlock" as the function and not "ring_buffer_map", making it harder
to find where the change is.

Same for the labels below.

-- Steve


> +	mutex_unlock(&buffer->mutex);
> +	mutex_unlock(&cpu_buffer->mapping_lock);
> +
> +	return err;
> +}
> +
> +int ring_buffer_unmap(struct trace_buffer *buffer, int cpu)
> +{
> +	struct ring_buffer_per_cpu *cpu_buffer;
> +	unsigned long flags;
> +	int err = 0;
> +
> +	if (!cpumask_test_cpu(cpu, buffer->cpumask))
> +		return -EINVAL;
> +
> +	cpu_buffer = buffer->buffers[cpu];
> +
> +	mutex_lock(&cpu_buffer->mapping_lock);
> +
> +	if (!cpu_buffer->mapped) {
> +		err = -ENODEV;
> +		goto out;
> +	} else if (cpu_buffer->mapped > 1) {
> +		__rb_inc_dec_mapped(cpu_buffer, false);
> +		goto out;
> +	}
> +
> +	mutex_lock(&buffer->mutex);
> +	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
> +
> +	cpu_buffer->mapped = 0;
> +
> +	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
> +
> +	kfree(cpu_buffer->subbuf_ids);
> +	cpu_buffer->subbuf_ids = NULL;
> +	rb_free_meta_page(cpu_buffer);
> +	atomic_dec(&cpu_buffer->resize_disabled);
> +
> +	mutex_unlock(&buffer->mutex);
> +out:
> +	mutex_unlock(&cpu_buffer->mapping_lock);
> +
> +	return err;
> +}
> +
> +int ring_buffer_map_get_reader(struct trace_buffer *buffer, int cpu)
> +{
> +	struct ring_buffer_per_cpu *cpu_buffer;
> +	unsigned long reader_size;
> +	unsigned long flags;
> +
> +	cpu_buffer = rb_get_mapped_buffer(buffer, cpu);
> +	if (IS_ERR(cpu_buffer))
> +		return (int)PTR_ERR(cpu_buffer);
> +
> +	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
> +consume:
> +	if (rb_per_cpu_empty(cpu_buffer))
> +		goto out;
> +
> +	reader_size = rb_page_size(cpu_buffer->reader_page);
> +
> +	/*
> +	 * There are data to be read on the current reader page, we can
> +	 * return to the caller. But before that, we assume the latter will read
> +	 * everything. Let's update the kernel reader accordingly.
> +	 */
> +	if (cpu_buffer->reader_page->read < reader_size) {
> +		while (cpu_buffer->reader_page->read < reader_size)
> +			rb_advance_reader(cpu_buffer);
> +		goto out;
> +	}
> +
> +	if (WARN_ON(!rb_get_reader_page(cpu_buffer)))
> +		goto out;
> +
> +	goto consume;
> +out:
> +	/* Some archs do not have data cache coherency between kernel and user-space */
> +	flush_dcache_folio(virt_to_folio(cpu_buffer->reader_page->page));
> +
> +	rb_update_meta_page(cpu_buffer);
> +
> +	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
> +	rb_put_mapped_buffer(cpu_buffer);
> +
> +	return 0;
> +}
> +
>  /*
>   * We only allocate new buffers, never free them if the CPU goes down.
>   * If we were to free the buffer, then the user would lose any trace that was in