On Wed, 26 Mar 2008 12:43:50 +0100 Stefan Roese sr@denx.de wrote:
On Wednesday 26 March 2008, Haavard Skinnemoen wrote:
Yes, same SoC ethernet interfaces. Currently PHY address etc is configured via config options.
Judging by the number of #ifdefs in ppc_4xx_eth_initialize, that function contains quite a bit of board-specific code.
Most of it is not board-specific, but platform-specific. This driver handles all 4xx EMAC variants and there are quite a lot. But basically you are correct.
Yeah, but the platform-specific bits are normal for a platform-specific driver. I just think that ideally, the board-specific bits should be split out and placed in board_eth_init() for each board.
Can't you just add a weak definition of board_eth_init() in the CPU code then?
No. The weak definition is already in net/eth.c.
Make it non-weak then.
In net/eth.c? It can't get overwritten then.
I mean the cpu-specific one. The generic "fallback" function should stay weak.
If a platform wants to implement board-specific ethernet init code as #ifdef hell, it should be free to do so. I don't think we should add workarounds in the generic code for such platforms though.
I could use board_eth_init() as a common PPC4xx platform code. But I would loose the chance being able to implement a "real" board specific board_eth_init() for some boards in the future then.
Yeah...it would be bad to prevent future boards from doing the right thing...
How about you do an internal split in cpu/ppc4xx/4xx_enet.c first, placing the truly platform-specfic bits in a separate function that can be called by "new-style" boards, and the board-specific bits in board_eth_init(), serving as a wrapper around the platform-specific init function, with as many #ifdefs as needed in order to handle the existing differences between the boards. This function could be conditional on CONFIG_PPC4XX_LEGACY_BOARD_ETH_INIT or something so that boards can be migrated to the new API one by one.
U-Boot could really use a Janitor team ;-)
Haavard