Re: + edac-cpc925-mc-platform-device-setup.patch added to -mm tree

[Harry Ciao <qingtao.cao@windriver.com>]

Michael Ellerman wrote:
> On Thu, 2009-04-16 at 09:57 +0800, Harry Ciao wrote:
>   
>> Kumar Gala wrote:
>>     
>>> On Apr 15, 2009, at 5:27 PM, akpm@linux-foundation.org wrote:
>>>       
>>>> arch/powerpc/kernel/prom_init.c      |   33 +++++++++++++
>>>> arch/powerpc/platforms/maple/setup.c |   59 +++++++++++++++++++++++++
>>>> 2 files changed, 92 insertions(+)
>>>>
>>>> diff -puN 
>>>> arch/powerpc/kernel/prom_init.c~edac-cpc925-mc-platform-device-setup 
>>>> arch/powerpc/kernel/prom_init.c
>>>> --- 
>>>> a/arch/powerpc/kernel/prom_init.c~edac-cpc925-mc-platform-device-setup
>>>> +++ a/arch/powerpc/kernel/prom_init.c
>>>> @@ -1947,8 +1947,40 @@ static void __init fixup_device_tree_map
>>>>     prom_setprop(isa, name, "ranges",
>>>>             isa_ranges, sizeof(isa_ranges));
>>>> }
>>>> +
>>>> +#define CPC925_MC_START        0xf8000000
>>>> +#define CPC925_MC_LENGTH    0x1000000
>>>> +/* The values for memory-controller don't have right number of cells */
>>>> +static void __init fixup_device_tree_maple_memory_controller(void)
>>>> +{
>>>>         
>>> I don't see why this cant be part of the existing 
>>> fixup_device_tree_maple().
>>>
>>> I also find it odd we don't ensure we are running on a maple before we 
>>> apply this fixup.
>>>       
>
>   
>> Hi Kumar,
>>
>> Thanks a lot for your concern.
>>
>> This newly added fixup for memory controller on Maple will be placed 
>> right after fixup_device_tree_maple(), both of them will be controlled 
>> by CONFIG_PPC_MAPLE, so there is no worry that it will be applied 
>> against anything other than Maple.
>>     
>
> Hi Harry,
>
> We regularly build a single kernel with multiple platforms enabled, so
> just having it controlled by a CONFIG symbol is not sufficient. Someone
> might build a kernel for MAPLE & PSERIES & ISERIES & CELL, so the maple
> fixup needs to be careful it doesn't break the other platforms.
>
> The existing maple fixup doesn't check if it's on a maple either, but it
> is a bit more discerning about what it finds before it fixes things up.
>
> Your code already checks that "reg" is 8 bytes long to start with, I
> think if it also checks that #address-cells and #size-cells are == 2,
> then it's pretty safe. Because at that point we know we have a node with
> the right name, the reg property has a known value, and reg is short WRT
> #cells.
>
>   
Many thanks Michael!

That makes a lot of sense to me :) I will integrate the check if both 
its parent #address-cells and #size-cells equal to 2 before fixing 
anything up. The fact that the reg length == 8 bytes whereas parent's 
cells are 2 validate our fixup is necessary.

Best regards,

Harry


>> Meanwhile, it aims at fixup bad cell numbers for the memory controller, 
>> whereas the original fixup_device_tree_maple() aiming at fixing up the 
>> ISA controller on HT channel, we'd better separate them in different 
>> function IMHO.
>>     
>
> I think I agree it's better as a separate routine. We could have a
> firmware that doesn't need the original maple fixup (and so exits from
> that routine early) but does need this one.
>
> cheers
>
>