On Fri, Sep 30, 2005 at 10:10:52AM -0500, Corey Minyard wrote:

Mark A. Greer wrote:

quoted

On Thu, Sep 29, 2005 at 01:32:21PM -0500, Corey Minyard wrote:

quoted

James looked at an earlier version and said it was ok, and suggested
a few changes.  This has been tested against 2.6.14-rc2 on a Katana.

Note that the bus_clk value from the platform information seems to
be in MHZ, but the actual frequency is generally 133,333,333, not
133,000,000.  I don't think the inaccuracy matters here, but it seems
a little odd.
  

Yes, it makes more sense to me to pass the actual frequency and not do
the '* 1000000'.

The only trouble is if  you go over 4GHZ...  Maybe it should be in KHZ 
instead of MHZ?  or 64-bits?  Or maybe we don't worry about >4GHZ busses?

My vote is to worry about it when the time comes.  IIRC, the fastest a
Marvell bridge part can be clocked right now is 200 MHz so there's some time.

Also, how would one go about changing this?  Would you need something like:

if (bus_clk < 1000)
 bus_clk *= 1000000

for backwards compatability?

When I didn't see anyone passing that value in via platform_data so
there's no compatibility issue.  I could have missed something though...

quoted

Also--I should have thought of this earlier--there should probably be a
patch to add a default platform_data entry in arch/ppc/syslib/mv64x60.c
and a patch for katana.c if the defaults in mv64x60.c aren't correct
for the katana.  At least, that's how things are done now.

I'm not sure I understand completely.  How would you go about setting 
the platform data?

Look near the top of arch/ppc/syslib/mv64x60.c and you will see a bunch
of default platform_data being set up for marvell bridges.  They can then
be tweaked by the platform specific code via the platform_notify() hook
(see arch/ppc/platforms/katana.c:katana_platform_notify(), et. al.).

Mark