On Sep 15, 2011, at 12:11 AM, Wolfgang Denk wrote:
Dear Kumar Gala,
In message D144BA6D-B90C-46CD-BE7F-6197015676BB@kernel.crashing.org you wrote:
Now you could say that we should deal with this via #defines. Which is true, however our reference boards allow for running in hundreds if not thousands of various possible HW configurations and we've tried to avoid having different builds for such things unless completely unavoidable.
Understood and agreed.
So this gives the option to the board code to disable things if it knows based on some other information. In the specific case for the P3060QDS board, we have "add-on" cards that provide 4-ports of 1-gig SGMII. We can detect if one of these cards is in a slot. If the card is missing we can 'disable' the ports associated with the slot.
Well, can you not turn this logic around, and come up in a configuration with all ports disabled, and then enable only those ports that are actually present? That appears to make much more sense to me. Eventually it is even more efficient in terms of things likepower cinsumption etc. ?
Unfortuantely not. The HW makes the assumption you turn things off, not on. Once we turn something 'off' in the HW we can't turn it back 'on' as the SoC will gate the power to the block and only a hard reset will bring it back up.
[In the long run (maybe not so long actually) you can even do this device tree based …]
- k