Re: [PATCH 05/12] libata: xfer_mask is unsigned int not unsigned long

[Jeff Garzik <jeff@garzik.org>]

Tejun Heo wrote:
> Alan Cox wrote:
>> On Tue,  6 Nov 2007 14:39:03 +0900
>> Tejun Heo <htejun@gmail.com> wrote:
>>
>>> xfer_mask is unsigned int not unsigned long.  Change ->mode_filter to
>>> take and return unsigned int.
>>>
>>> While at it, rename @adev of ata_pci_default_filter() to @dev for
>>> consistency.
>>>
>>> Signed-off-by: Tejun Heo <htejun@gmail.com>
>> The filter type was purposefully unsigned long to allow for expansion (eg
>> for SWDMA) without breaking drivers. No problem with it changing but I'd
>> say "unsigned int" was the worst possible choice - its now shorter (no
>> room for expansion) and size dependant on arch. u32 would be shorter and
>> consistent across all platforms, ulong would have more room for expansion.
> 
> I agree it should be one of u* types and am happy to convert.  This
> patch is primarily for consistency as in libata-core, xfer_mask is
> unsigned int.  My first proposal was u32 too but Jeff vetoed it.  IIRC,
> Jeff has affection for machine-independent integral types.  :-)
> 
> Anyways, I think ulong is worse than uint because ulong differs in size
> between 32 and 64 archs.  What's the good of extra 32bits in flags space
> if it's not there on 32bit archs?  Also, we currently only use 20bits of
> xfermask, 12 extra bits seem enough for the time, especially as
> everything is moving over to SATA where xfermode doesn't really matter, no?
> 
> Jeff, are you okay with u32 or 64?

unsigned long is IMO the best choice for bitmaps.

* it is a machine int on all(?) architectures

* it is 32-bit on 32-bit architectures

* it is consistent with test_bit(), set_bit() and lib/bitmap.c interfaces

* conversion to test_bit() and lib/bitmap.c interfaces as a future step 
is trivial

* your structs inflate on 64-bit due to pointers anyway, so I see little 
real consequence to using the lower 32 bits of an unsigned long as a 
portable meme.

	Jeff