ioremap of PCI resources

[Ian Abbott]

On 2013-04-10 17:16, H Hartley Sweeten wrote:
> On Wednesday, March 06, 2013 4:34 PM, H Hartley Sweeten wrote:
>>
>> Question, when doing an ioremap() of a PCI resource which is more correct?
>>
>> 	iobase = pci_resource_start(pcidev, bar);
>> 	len = pci_resource_len(pcidev, bar);
>> 	mmio = ioremap(iobase, len);
>>
>> or
>>
>> 	mmio = pci_ioremap_bar(pcidev, bar);
>>
>> pci_ioremap_bar() does some extra sanity checking to make sure the bar
>> is actually a memory resource and then does an ioremap_nocache().
>>
>> If the ioremap() is in order to talk to the registers on the PCI device wouldn't
>> the pci_ioremap_bar() be more correct?
>
> Greg / Ian / Dan,
>
> Do you have any ideas/opinions on the question above?

Saves a couple of lines of code, I suppose.  The extra sanity makes 
sense for the function to do for defensive programming reasons, but is 
not particularly useful for the caller unless there are some (presumably 
older) variants of a PCI card that match the driver's PCI ID table, but 
have registers in I/O space instead of memory space (in which case, it's 
better that pci_ioremap_bar() returns the NULL rather than 
ioremap_nocache() attempting to remap a port address (though that would 
probably fail and return NULL anyway).

-- 
-=( Ian Abbott @ MEV Ltd.    E-mail: <abbotti at mev.co.uk>        )=-
-=( Tel: +44 (0)161 477 1898   FAX: +44 (0)161 718 3587         )=-