On Mon, 29 Sep 2008, Steven Rostedt wrote:
quoted text
>   We do not need to worry about SMP for writes.
>   That is, if you reserve data and atomically push the head forward
>    if the head goes past the end of page, One, you check if your
>    start of head (head - length) is still on the page. If it is
>    you add your padding (you already reserved it), then you atomically
>    push the head forward. Then you start the process again.
>   If the start of the head (head - length) is not on the page, that
>    means that an IRQ or NMI came in and pushed it before you.

I forgot to mention one important detail. The "head" index will stay
on the page frame. That way we do not need to figure out which 
head_page we are on. We grab the head_page, we atomically 
(local_inc_return) the head pointer on that page. If the return value is 
still on the page, we succeeded. We can also increment a value on this 
page frame that will prevent recording if we somehow overflowed the buffer 
before relinquishing the stack.

That is

  reserve_event()

     IRQ->
            reserve_event();

      [...]

     IRQ->reserve_event() came back to original head!

      Here we do not have a big enough buffer, and this is just stupid ;-)
      We would drop packets in this case, and should drop the guy on his
      head who came up with the too small buffer.

-- Steve

--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/