Tejun Heo wrote:

 Hello, Jeff.

 This patch removes compiler warnings which are caused by using
ioports values (unsigned long) for the address argument of
read/write[bwl]() functions without casting.

Signed-off-by: Tejun Heo <redacted>

NAK...  These warnings exist as a reminder of the remaining 
ioread/iowrite() work that must occur.  Look at the 'iomap' and 
'iomap-step1' branches of libata-dev.git for incomplete examples of this 
work.

One must:

- create a data structure to store a bunch of I/O port values, as 
returned by pci_iomap() and ioport_map()
- update legacy PCI IDE (aka looks like ISA IDE) code path to use 
ioport_map() to obtain the I/O addresses we need
- update native mode PCI IDE (aka looks like PCI) code path to use 
pci_iomap()
- update all the other drivers that use ioremap() to use pci_iomap()
- change struct ata_ioports, s/unsigned long/void __iomem */
- fix up the last few bits in each driver, such as
	* eliminating the 'unsigned long base' variable in each
	  xxx_init_one() function
	* changing the xxx_setup_port() function to indicate
	  void __iomem * rather than unsigned long
- one last check for any last s/long/void iomem */ changes to be made
- install Linus's "sparse" source checker
- run 'make C=1' in the kernel tree, and make sure libata and drivers 
don't spew warnings


So... it's not as simple as just killing the warnings ;-)