Michael Trimarchi wrote:
> Little clean up and move the do_irq on a deferrable work

Hmm, did the current solution cause any problems ? A workqueue is
a heavy thing, so that's generally not what you want to invoke on
a per packet basis.

Besides, please don't call a workqueue "tasklet". That's an
entirely different mechanism. Tasklets are orders of magnitude
more light-weight than workqueues, and are in fact appropriate
for network traffic.

Did you measure what your change does to throughput, maximum
latency, and CPU load during transfers ? I think you may find
some surprises, particularly when it comes to latency :-)

- Werner

Reply via email to