Re: [PATCH] Fix race in ring_buffer_consume(): Replace ring_buffer_consume and ring_buffer_peek with ring_buffer_get_event

["Indan Zupancic" <indan@nul.nu>]

On Mon, January 5, 2009 16:09, Steven Rostedt wrote:
> I just hate public functions that have flags that can get confusing.
> Looking at code where I see:
>
> 	ring_buffer_get_event(buffer, cpu, ts, 0);
>
> I find it hard to know if the ",0" is correct or should be a ",1"
> instead.

That's what I didn't like about the original patch either.
Sent a new patch against tip doing the above, as well as doing
the same for rb_iter_peek (ignore the first one, see try 2).

Slightly different topic, there are quite a few unused functions,
e.g. ring_buffer_read_page(). Should those be removed until they
are actually used, or are there things in the pipeline that will
use them?

I'm not sure if rb_event_length() is safe as it is, returning -1
for RINGBUF_TYPE_PADDING, for an unsigned return value. That
easily overflows into a low value in e.g. rb_advance_iter().
Wouldn't it be better to return the maximum possible size?

rb_advance_reader() calls rb_get_reader_page() and rb_reader_event(),
but rb_get_event(), which is the only caller of rb_advance_reader,
does that as well, which seems a bit more of the same.

Further, rb_get_reader_page() takes cpu_buffer->lock, but all callers
of rb_get_event() take the cpu_buffer->reader_lock. As rb_get_event()
always calls rb_get_reader_page(), both locks will be taken for each
of those calls. Doesn't that make the cpu_buffer->lock redundant?
Getting rid of it and moving the locking into rb_get_event() would
simplify things (except that ring_buffer_read_page() is a bit funny).

Greetings,

Indan