On Thu, Jun 30, 2011 at 06:46:55AM +0200, Wolfram Sang wrote:

Hi Shawn,

quoted

quoted

quoted

I would suggest leave this gpmi specific quirk to gpmi driver to sort
out.  With the following mxs-dma change, it should work if gpmi driver
can pass the valid gpmi irq number for only one gpmi channel, and -1
for all others.

...which brings us right into the 'NO_IRQ is 0' discussion :)

Though I do not know what it means exactly, number 0 is an valid IRQ
on both mx23 and mx28 (see mx23.h and mx28.h).

It could be remapped. It is a looong story. Start here if you are interested:

http://lkml.org/lkml/2005/11/21/221
http://yarchive.net/comp/linux/no_irq.html

Currently, it's a can of worms. Hopefully, irq_desc might help somewhen.

Thanks for the info.  It deserves a good study.  With your comment,
I would use macro NO_IRQ than -1 for the mxs-dma change.

quoted

quoted

Other than that, [thinking loud] this will help if all irq-sharing
channels are handled by the same driver. If not, we would just add
IRQF_SHARED (hopefully this will never be needed). Yup, sounds
reasonable to me. Will give it a second thought later, though.

GPMI is the only mxs-dma user that gets irq-sharing.  So yes, all
irq-sharing channels are handled by the same driver, gpmi-nfc :)

Currently, yes. But we have to make it future-proof.

For the SoCs coming in future, some, e.g. mx6 series, is the same
case, and some, like mx50, has separate irq for each gpmi channel,
which is even better.  So we can survive for quite a long time.

-- 
Regards,
Shawn