Dan Malek wrote:
As a workstation/server BOF, I don't know how much interest exists to
discuss this. The easy answer is a set of orthogonal directory trees,
but I think the highest priority should be on maintaining a set of
common files. Having experienced splitting some processor dependent
files, I would still take the #ifdefs over tracking similar changes
across multiple files.
I'd go with that some of the way, but I think MMU_init() in mm/init.c
goes a bit too far. It feels like every time I port my patches to a new
release kernel somebody has added extra boards and I have to sort switch
statements out by hand.
As an example, looking at 2.2 and 2.3, the actions performed by
xxx_find_end_of_memory() haven't changed very much. There probably
wouldn't be much problem introducing a ppc_md.find_end_of_memory call(),
and patches to add new machines would apply much cleaner.
I don't think the world is ready for my dream of eliminating _machine
completely, but if there's any interest I can attempt a test refactoring
of mm/init.c. This needn't be an all at once change; the idiom I'm
thinking of goes something like:
if (ppc_md.do_whatever())
ppc_md.do_whatever();
else {
#ifndef CONFIG_8xx
switch(_machine)
...
#else
8xx_do_whatever();
#endif
}
- Adrian Cox, AG Electronics
** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/