On Wed 2014-08-27 12:11:55, Jaehoon Chung wrote:

Hi,

On 08/26/2014 07:19 PM, Pavel Machek wrote:

quoted

Hi!

quoted

quoted

quoted

Would you elaborate?

If I have a device like a phone, I may want to put one "slot" inside
phone for basic system, and offer second slot for user expansion
(initially empty).

if multiple slot is supported, then a mmcqd should be processing for multiple slots.
It's too inefficient, and affect the whole performance reduction.

Sorry, Discard this comment. it means dwmci, not mmcqd.

Well, that's a Linux problem, and for many applications, not even
problem at all.

Device tree should describe hardware, and hardware can do multiple
slots per controller, so device tree should describe multiple slots
per controller.

Now, the configuration may be uncommon, but you are moving from good
hardware description to bad hardware description.

Well, i don't think it's bad hardware description. And this policy is suggested by other mmc developers and maintainers.
At first time, I had also suggested same opinion with yours.
Refer to below..

Well, I disagree with them. They want to modify device tree because of
linux limitations.

Plus. I guess that sooner or later someone will wire just the slot 1
(not 0) and not match this description.
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html