Re: [RFC PATCH 1/2 v2] Unified trace buffer

[Linus Torvalds <torvalds@linux-foundation.org>]

On Thu, 25 Sep 2008, Steven Rostedt wrote:
> +
> +/**
> + * ring_buffer_event_length - return the length of the event
> + * @event: the event to get the length of
> + *
> + * Note, if the event is bigger than 256 bytes, the length
> + * can not be held in the shifted 5 bits. The length is then
> + * added as a short (unshifted) in the body.

The comment seems stale ;)

> +
> +/**
> + * ring_buffer_peek - peek at the next event to be read
> + * @iter: The ring buffer iterator
> + * @iter_next_cpu: The CPU that the next event belongs on
> + *
> + * This will return the event that will be read next, but does
> + * not increment the iterator.
> + */
> +struct ring_buffer_event *
> +ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
> +{
> +	struct ring_buffer_per_cpu *cpu_buffer;
> +	struct ring_buffer_event *event;
> +	u64 delta;
> +
> +	cpu_buffer = buffer->buffers[cpu];
> +
> + again:
> +	if (ring_buffer_per_cpu_empty(cpu_buffer))
> +		return NULL;
> +
> +	event = ring_buffer_head_event(cpu_buffer);
> +
> +	switch (event->type) {
> +	case RB_TYPE_PADDING:
> +		ring_buffer_inc_page(buffer, &cpu_buffer->head_page);
> +		rb_reset_read_page(cpu_buffer);
> +		goto again;
> +
> +	case RB_TYPE_TIME_EXTENT:
> +		delta = event->data;
> +		delta <<= TS_SHIFT;
> +		delta += event->time_delta;
> +		cpu_buffer->read_stamp += delta;
> +		goto again;
> +
> +	case RB_TYPE_TIME_STAMP:
> +		/* FIXME: not implemented */
> +		goto again;
> +
> +	case RB_TYPE_SMALL_DATA:
> +	case RB_TYPE_LARGE_DATA:
> +	case RB_TYPE_STRING:
> +		if (ts)
> +			*ts = cpu_buffer->read_stamp + event->time_delta;
> +		return event;

Your timestamp handling seems odd. You do it per-event, but I think it 
should happen for all events, ie just do

	*ts += event->time_delta;

_outside_ the case statement, and then in RB_TYPE_TIME_EXTENT you'd do 
either

 - relative:
	*ts += event->data << TS_SHIFT;

 - absolute timestamp events:
	*ts = (event->data << TS_SHIFT) + event->time_delta;

but the bigger issue is that I think the timestamp should be relative to 
the _previous_ event, not relative to the page start. IOW, you really 
should accumulate them. 

IOW, the base timestamp cannot be in the cpu_buffer, it needs to be in the 
iterator data structure, since it updates as you walk over it.

Otherwise the extended TSC format will be _horrible_. You don't want to 
add it in front of every event in the page just because you had a pause at 
the beginning of the page. You want to have a running update, so that you 
only need to add it after there was a pause.

		Linus