Hi Simon,
On Sun, 25 Nov 2018 at 14:09, Simon Goldschmidt simon.k.r.goldschmidt@gmail.com wrote:
On Thu, Nov 22, 2018 at 9:50 PM Simon Glass sjg@chromium.org wrote:
Hi,
On Thu, 22 Nov 2018 at 10:02, Tom Rini trini@konsulko.com wrote:
On Thu, Nov 22, 2018 at 03:44:28PM +0100, Simon Goldschmidt wrote:
Am Do., 22. Nov. 2018, 14:44 hat Tom Rini trini@konsulko.com geschrieben:
On Thu, Nov 22, 2018 at 02:24:49PM +0100, Marek Vasut wrote:
On 11/22/2018 01:52 PM, Tom Rini wrote: > On Thu, Nov 22, 2018 at 10:25:14AM +0100, Christian Gmeiner wrote: > >> Am Mo., 19. Nov. 2018 um 16:56 Uhr schrieb Simon Glass <
sjg@chromium.org>:
>>> >>> This board has not been converted to CONFIG_DM_BLK by the deadline. >>> Remove it. >>> >> >> As the board is still mainted I will NAK it for the moment. Are there >> any hints want needs to be done >> to port thie board over to new DM stuff? > > Yes, as a start you need to switch over to using CONFIG_OF_CONTROL and > selecting/providing a dtb file. I see ot1200 is using DWC_AHSATA which > needs more work, but this is the board-level work that needs doing.
Wasn't there a possibility to use platform data in board file instead of DT ? Or is DT mandatory now , including the libfdt overhead ?
In short, DT for U-Boot and platform data for SPL is what's recommended, yes.
This is a little confusing for me. Socfpga gen5 SPL doesn't do that. And it seems a little strange or outdated overall.
Would there be some kind of reference architecture or mach to look at what's the suggested/up-to-date way to implement SPL? Also regarding code flow?
So, SPL is where things get, ahem, fuzzy. While I don't want to encourage boundless growth in U-Boot proper, we aren't exactly size constrained (but rather, functional/logical constrained). But in SPL, yes, we have many platforms with 32/64/128 kilobyte hard limits (and some smaller) and we can't always shove in a "TPL" before SPL either. So in SPL we do make use of platform data instead. While not the smallest size constraint, am335x_hs_evm is a reasonable thing to look at in this case.
Also 'rock' uses CONFIG_OF_PLATDATA which provides a halfway house - still uses DT, but it gets converted into C structs so saves code space.
firefly-rk3288 is a pretty good DM/DT example, including SPL.
I've currently got an issue on socfpga gen5 that could be solved best by enabling CONFIG_OF_EMBED (mixing const and non-const sections is a problem for CRC calculation). However, it could probably also solve by using platform data (but that doesn't work out of the box, yet). The problem with CONFIG_OF_EMBED is that I think it's OK to enable this for SPL but I don't like enabling it for U-Boot, so:
Would it make sense to duplicate the whole "Provider of DTB for OF control" choice so that it can be OF_EMBED for SPL but different for U-Boot? Or does it make more sense to convert socfpga gen5 to use OF_PLATDATA?
We should not be using OF_EMBED in in-tree boards or production code. Can you please explain the issue a bit more?
Regards, Simon