Ideas/suggestions to avoid repeated locking and reducing too many lists with dmaengine?

[Russell King - ARM Linux]

On Mon, Feb 24, 2014 at 04:38:48PM -0600, Joel Fernandes wrote:
> Actually I did do some tracing earlier before I posted this thread- and
> notice there was excessive traces of locking/unlocking. It is very light
> though as you pointed and lighter without debug options. The only other
> notable difference is the fact that we are now going through the dmaengine
> framework in the newer kernel vs the faster one.

Okay, for OMAP, there's additional latency caused by the tasklet which
starts a transaction when there's no previous transaction running -
that will be your biggest problem.

I do want to move omap-dma away from having a fixed binding between
claimed channels and the logical channels since that's too wasteful,
and that's what the tasklet will eventually be doing - but with
omap-dma being divorsed from the old code, this tasklet may not be
necessary anymore (since we no longer have to operate in process
context for that bit.)

There's a huge pile of patches I have queued up for the next merge
window in linux-next - I'd rather get these into mainline before
trying to do any restructuring, otherwise we're going to get into
a hell of a problem with merge conflicts.

-- 
FTTC broadband for 0.8mile line: now at 9.7Mbps down 460kbps up... slowly
improving, and getting towards what was expected from it.