cvs commit: src/sys/kern kern_poll.c

[Robert Watson]

On Mon, 5 Sep 2005, Luigi Rizzo wrote:

> Also, what is the overall design - do you want to support multiple 
> polling loops running concurrently (netisr_poll, one poll_idle per CPU, 
> and possibly the poll_in_trap) ?

Issues to be particularly careful about will be priority inversion and 
ordering.  With a single thread associated with any given interface, the 
inversion problem might exist, but not the ordering issue, but if more 
than one thread may process a particular interface, there are potentially 
also ordering issues.

It's been proposed that at the EuroBSDCon developer summit, we set aside a 
chunk of time to talk about the future structuring of polling -- in 
particular, how we can broaden the scope of the polling infrastructure to 
allow non-network device drivers also have their handlers run from a 
polled environment.  A couple of hours of white board session would 
probably be quite useful, since I think there's a lot to discuss here.

Robert N M Watson