On Tue, Apr 19, 2005 at 11:32:59AM -0700, Andy Isaacson wrote: > On Mon, Apr 18, 2005 at 02:48:09PM +0200, Henk wrote: > > > > See section 1.3 on the wiki page: > > http://openwrt.org/Kernel26Firmware > > Feel free to comment here on the list. > > One comment: > > # - Migrate: Should we cluster this with the sibyte stuf? Probably > # there's some shared code.... > > SiByte and the 47xx don't share anything significant beyond both being > MIPS, and both running CFE. And people (including Ralf) actually have > SiByte hardware that they test on. So not breaking SiByte would be a > good thing. > > Because there's no technical connection between SiByte and 47xx I'd lean > towards leaving the SiByte stuff alone, and clean up the 47xx code on > its own. > Point taken, we should have a seperate broadcom branch. I will update this on the wiki. > > General comments on the WRT code: > > The code is full of "Broadcom Proprietary" and "All Rights Reserved" > notices. Does anyone have a clear written statement from Broadcom that > it's redistributable? (If you're depending on the GPL release > requirements to justify relicensing, clear documentation of the chain of > release would be helpful.) Maybe we should create patch sets that will transform the original wrt branch into 2.6 code. Anyway I don't think broadcom is a criminal company, I think it is obliged by law to comply with the requirements of the GPL ;) > > - We should probably make some abstraction/API of the so called Silicon > > Backplane bus that broadcom defined. I see allot of drivers, even in > > the mainline kernel (b44 ethernet driver) that use this. > > The Silicon Backplane bus actually came from another company, it wasn't > defined by Broadcom; google knows all: > http://www.ocpip.org/socket/adoption/sonics > > I think there are other OCP busses supported in the kernel; ISTR seeing > some PPC SoC from IBM that uses OCP... so perhaps this should be brought > up on l-k for general discussion. > > But it's challenging to come up with a useful abstraction that covers > both the b44 scenario and the SoC scenario. > > - for b44, OCP is on the far side of the PCI bus, and is used only to > access a single core (ethernet MAC). > > - for bcm947xx (and ppc SoC, I guess), OCP is the system bus, and is > used to access everything from PCI to DRAM. > > grep grep grep... Take a look at include/asm-ppc/ocp.h and > arch/ppc/platforms/*.c, it looks like the PowerPC people have already > done a bunch of work here. > Also of interest may be the new SOC abstraction on the hh.org branch, see http://handhelds.org/cgi-bin/cvsweb.cgi/linux/kernel26/Documentation/soc.txt Thanks for your comments, greatly appreciated, I will update the OCP stuff on the wiki. The task seems a little more challenging than a week ago, anyway I will focus on the boot code for a while so I can boot test kernels without the need for reflashing... regards, Henk
