On Tue, Feb 05, 2013 at 05:06:28PM +0000, Russell King - ARM Linux wrote:
> On Tue, Feb 05, 2013 at 04:47:05PM +0000, Mark Brown wrote:
> > On Tue, Feb 05, 2013 at 05:21:48PM +0100, Linus Walleij wrote:

> > > For IRQ mode, use the completion callback to push each cookie
> > > to NAPI, and thus let the IRQ drive the traffic.

> > The whole purpose of NAPI is to avoid taking interrupts for completion
> > of transfers.  Anything that generates interrupts when NAPI is in
> > polling mode is defeating the point.

> Yes, but you're missing one very important point.  If your DMA engine
> is decoupled from the network device, and the DMA engine relies upon
> interrupts to queue further transfers to move data into RAM, then you
> have two options:

> 1. Receive these interrupts so you can update the DMA engine for
>    further data transfer.
> 2. Don't receive these interrupts, and cause the network device to
>    error out with a FIFO overrun because its DMA requests have not
>    been serviced. (which may raise an interrupt.)

There's still the third option of just implementing a non-NAPI driver,
though I guess we've now built so much infrastructure on top of NAPI
that it's still useful to pretend we can disable interrupts.