Re: tcp_isn_tick() / dummynet() callout madness ?

Poul-Henning Kamp wrote:

> In message <20050130191410.R62670_at_odysseus.silby.com>, Mike Silbersack writes:
> 
> One of the things I would like to see is for the callout api to gain
> a mutex pointer.
> 
> If the pointer is not null, a mtx_trylock() is atempted, if it fails
> a taskq is used to execute this callout, otherwise the function is used
> as usual.
> 

Taskqueues really are very non-deterministic and a poor choice for
periodic events.  They really are only suitable for events that are
uncommon and don't require any sense whatsoever of urgency.  While the
callout API doesn't have any real-time guarantees, there is an
assumption that assigned callouts will be generated with a reasonable
amount of accuracy and consistency, and not be held up by a task that
has an indefinite run time.  Either a new dedicated kthread-based task
needs to be created for what you propose, or the simplier approach can
be taken of just deferring callouts that fail the trylock test to the
end of the list.

Scott