On Wed, Jun 20, 2007 at 10:54:14PM +0100, Andrew Doran wrote:
> What I mean is that from a hardware ISR that does not have thread context
> (like at IPL_NET) it wouldn't be possible to call callout_reset() or similar
> functions. That would need to be deferred to a soft interrupt or a kernel
> thread.

Uggg...  wouldn't it be possible to be able to specify a PL level at
which a callout function will run?
That would allow such callouts to be used from WH interrupts.
SPL locking the callout table shouldn't be a problem.

> That may well make sense. There is the potential to cause a deadlock by
> maniuplating a callout when holding a lock that the callout routine wants
> to grab:

I know ...

> resetit()
> {
> 	mutex_enter(&lock);
> 	callout_reset(&ch, 1, mycallout, myarg);
> 		^ blocks waiting for mycallout()
> 	mutex_exit(&lock);
> }
> 
> mycallout()
> {

If the ISR happens here, then try_callout_reset() would fail, but the ISR
could set state.

> 	mutex_enter(&lock);
> 	/* do stuff */
> 	mutex_enter(&lock);

If the ISR happens here, the 'do stuff' above should have marked the
callout as 'not pending'.

> }

	David

-- 
David Laight: david@l8s.co.uk