On Tue, Jul 21, 2015 at 6:34 AM, Christoph Hellwig [off-list ref] wrote:

On Sun, Jul 19, 2015 at 08:18:00PM -0400, Dan Williams wrote:

quoted

Some archs define the first parameter to ioremap() as unsigned long,
while the balance define it as resource_size_t, similar confusion exists
for the type of the 'size' parameter.  Unify on (resource_size_t,
unsigned long) to enable passing ioremap function pointers.  Also, some
archs use function-like macros for defining ioremap aliases, but
asm-generic/io.h expects object-like macros, unify on the latter.

Move all handling of ioremap aliasing (i.e. ioremap_wt => ioremap) to
include/linux/io.h.  Add a check to lib/devres.c to warn at compile time
if an arch violates type expectations.

I don't think devres really has aything to do with this code.

It's one of the only cross-arch C files that is built with HAS_IOMEM.
It depends on having a common ioremap prototype later in the series.

quoted

Kill ARCH_HAS_IOREMAP_WC and ARCH_HAS_IOREMAP_WT in favor of just
testing for ioremap_wc, and ioremap_wt being defined.  This arrangement
allows drivers to know when ioremap_<foo> are being re-directed to plain
ioremap.  A later patch uses this arrangement to implement support for
strict mappings.

Acked-by: Christoph Hellwig <hch@lst.de>

I only ACKed this as a band-aid to get the pmem code in.  Now that
we got that in with a less invasive hack I don't see any reasons to
do this over doing the proper common prototypes and per-arch runtime checks
of flags variant.

That's not too much of a setback.  I had already started to realize
that nobody really wants the ioremap_<type> variants compared to
memremap.  So rather than introduce this cleanup and deprecate
ioremap_<type> over time, let's just skip a step and start converting
to memremap() with a proper 'flags' parameter for the mapping-type
right away