Re: [RFC][PATCH] ring-buffer: Have nested events still record running time stamp

[Steven Rostedt <rostedt@goodmis.org>]

On Fri, 26 Jun 2020 14:58:19 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:

> Second usage:
> 
> 		/* SLOW PATH - Interrupted between A and C */
> 		a_ok = rb_time_read(&cpu_buffer->write_stamp, &after);
> 		ts = rb_time_stamp(cpu_buffer->buffer);
> 		barrier();
>  /*E*/		if (write == (local_read(&tail_page->write) & RB_WRITE_MASK) &&
> 		    a_ok && after < ts) {
> 			/* Nothing came after this event between C and E */
> 			info->delta = ts - after;
> 			(void)rb_time_cmpxchg(&cpu_buffer->write_stamp, after, info->ts);
> 			info->ts = ts;
> 		} else {
> 			info->delta = 0;

Actually, I don't think a_ok can every be false here. An uninterrupted
event will leave with both before_stamp and write_stamp valid. As an
uninterrupted event will write to both (and a rb_time_t is only invalid
from reading an interrupted event).

To get to this path we have:

	w = local_read(write_tail);

	<--- Interrupt event (makes write_stamp valid!)

	write = local_add_return(write_tail, length);

	w != write - length;


	[..]

	a_ok = rb_time_read(write_stamp);

Must always be valid!

-- Steve