Re: [patch 2.6.29-rc2] palm_bk3710 buildfix

[Sergei Shtylyov <sshtylyov@ru.mvista.com>]

Bartlomiej Zolnierkiewicz wrote:

>>>>>>Someone should fix hw_regs_t to neither be a typedef, nor
>>>>>>use "unsigned long" where it should use "void __iomem *".

>>>>>    It cannot use pointers of course -- as the addresses can be I/O 
>>>>>ports.

>>>>It could use the ioread*() calls, which take pointers
>>>>and handle both types of I/O in a type-safe manner.

>>>  Probably they can... though using those would only slow down 
>>>register access on x86.

>>   ... and cause code bloat due to switching to real function calls from 
>>inlines which now should render into in/out isns.

> OTOH it would allow to unify a lot of 'if (hwif->host_flags & IDE_HFLAG_MMIO)'
> cases so the code may even shrink.

    It won't shrink as much as it will grow.

> I also suspect that register access slow
> down is most likely negligable (we can always measure and check it).

    I'd agree here but code bloat won't be as negligible considering the IDE 
code's predominant use of I/O ports on x86.

> The real reason for not using ioread*() and co. is that not all archs support
> them (though I checked some time ago, maybe they do now?) and we don't want to
> break IDE support for them.

    Sure. :-)

> Thanks,
> Bart

MBR, Sergei