Re: [PATCH 3/4] MIPS: pci-ar724x: remove static PCI IO/MEM resources

[Gabor Juhos <juhosg@openwrt.org>]

2013.02.04. 13:13 keltezéssel, Sergei Shtylyov írta:
> Hello.
> 
> On 03-02-2013 16:31, Gabor Juhos wrote:
> 
>>>> @@ -160,6 +163,16 @@ ath79_register_pci_ar724x(int id,
>>>>        res[2].start = irq;
>>>>        res[2].end = irq;
>>>>
>>>> +    res[3].name = "mem_base";
>>>> +    res[3].flags = IORESOURCE_MEM;
>>>> +    res[3].start = mem_base;
>>>> +    res[3].end = mem_base + mem_size - 1;
>>>> +
>>>> +    res[4].name = "io_base";
>>>> +    res[4].flags = IORESOURCE_IO;
>>>> +    res[4].start = io_base;
>>>> +    res[4].end = io_base;
>>>
>>>     One I/O port, hm? What is it good for?
> 
>> Strictly speaking it is not good for anything. This is a PCIe controller and it
>> does not support IO requests at all.
> 
>    Is this the case with every PCIe controller or only this particular one?

It is a limitation of this controller.

> 
>> However the whole PCI code assumes that
>> each PCI controller have an IO resource and uses the hose->io_resource pointer
>> unconditionally.
> 
>> Additionally, this matches with the removed static resource:
> 
>>> -static struct resource ar724x_io_resource = {
>>> -    .name   = "PCI IO space",
>>> -    .start  = 0,
>>> -    .end    = 0,
>>> -    .flags  = IORESOURCE_IO,
>>> -};
>>> -
> 
>    Since you seems to always pass 0, maybe you don't need 'io_base' parameter to
> the function above?

The AR724x SoCs SoCs have one PCIe controller only. However newer chips have
more than one PCIe controllers, and each of those controllers needs a different
io_base value to avoid resource conflicts. I'm preparing patches for newer SoCs
and if everything goes well I will send those today.

-GAbor