[PATCH 00/15] PCI io.h cleanups

On 07/09/2012 11:28 AM, Arnd Bergmann wrote:

On Friday 06 July 2012, Rob Herring wrote:

quoted

From: Rob Herring <redacted>

I've finally gotten back to this after several months. This is the 2nd
part of mach/io.h removals. This series removes io.h on platforms with
PCI by creating a fixed virtual I/O mapping and a common __io() macro.

As several platforms have 2 1MB I/O windows, I've changed the fixed
mapping to 2MB (vs. 1MB in the prior posting) and allow platforms to
control the size. This also allows using a single 1MB section mapping
rather than 4KB pages.

Looks really good, feel free to add my Reviewed-by on patches 2-15, since
I've done a line-by-line walkthrough and could not find anything besides the
trivial comments I gave.

Thanks for the review. It's going to change a bit with using
ioremap_page_range. Hope to have new version out soon.

quoted

I've only tested versatilepb under qemu, so any testing is appreciated.
iop3xx and mv78xx0 have some risk of breaking as the PCI bus addresses
are moved to 0 from matching the cpu host bus addesss.

Did you try running any devices that actually use I/O space in qemu?
I think there are plenty of devices that you could use for testing that.

I had with sym53c8xx, but it defaulted to memory accesses. Now that I
enable i/o access mode, it fails due to pcibios_min_io = 0x44000000.
With that removed it gets into the driver setup but then fails (with or
without my changes). Might be a qemu model issue. I'll try some other
devices.

Rob

	Arnd