On 09:16 Tue 10 Jun , Ben Warren wrote:
Shinya Kuribayashi wrote:
Stefan Roese wrote:
On Tuesday 10 June 2008, Shinya Kuribayashi wrote:
Shouldn't this be the other way around?
if (board_eth_init(bis) < 0)eth_eth_init(bis);So that the board init routine can "overwrite" the cpu init version.
Yeah, I think you're right. If board_eth_init() exists, it gets highest priority.
Just wondered, does that mean we could only have either cpu_eth_init or board_eth_init at a time?
Not really. board_eth_init() could call cpu_eth_init() if necessary.
Hm. What is cpu_eth_init for then? Just
board_eth_init(bis);seems to be enough for me. I also wonder where is the best place to have cpu_eth_init?
The cpu_init() was suggested by Stefan in our original discussion, when I only had the board function. His perspective is ppc_4xx, where tons of CPUs and boards share the EMAC driver, and he didn't want to modify each board. As you'll see in the discussion with JDL, it can probably apply to TSEC as well.
I'm not going to argue with you, I'm just thinking about my targets. One of my targets has internal ethernet MAC, and its evaluation board has an on-board external PCI NIC. Another target has internal MAC, but doesn't have PCI NIC.
I thought it'll be something like
cpu_eth_init(bis); board_eth_init(bis);The idea is that cpu_eth_init is a default for a CPU family, and board_eth_init is a board override, which can of course call cpu_eth_init.
What about a section to declare the netdev?
In this case we can have more than 2 eth and it more generic.
Best Regards, J.