Re: [PATCH V3 3/4] ARM: dts: tegra: Correct PCIe entry

[Thierry Reding]

On Tue, Jun 11, 2013 at 10:30:48AM +0300, Peter De Schrijver wrote:
> On Mon, Jun 10, 2013 at 09:55:12PM +0200, Thierry Reding wrote:
> > * PGP Signed by an unknown key
> > 
> > On Wed, Jun 05, 2013 at 12:27:31AM +0530, Jay Agarwal wrote:
> > [...]
> > > @@ -29,7 +29,7 @@
> > >  		ranges = <0x82000000 0 0x00000000 0x00000000 0 0x00001000   /* port 0 configuration space */
> > >  			  0x82000000 0 0x00001000 0x00001000 0 0x00001000   /* port 1 configuration space */
> > >  			  0x82000000 0 0x00004000 0x00004000 0 0x00001000   /* port 2 configuration space */
> > > -			  0x81000000 0 0          0x02000000 0 0x00010000   /* downstream I/O */
> > > +			  0x81000000 0 0          0x02000000 0 0x00100000   /* downstream I/O */
> > >  			  0x82000000 0 0x20000000 0x20000000 0 0x10000000   /* non-prefetchable memory */
> > >  			  0xc2000000 0 0x30000000 0x30000000 0 0x10000000>; /* prefetchable memory */
> > 
> > That increases the I/O region size from 64 KiB to 1 MiB. Why is that
> > necessary? I/O operations can only address 64 KiB, so I don't think
> > adding more makes any sense.
> 
> At least PCI allows 32bit I/O addresses. No idea if anyone uses them though.

I just realized that we are constrained to 64 KiB by the implementation
of pci_ioremap_io(), which assumes each mapping is 64 KiB. Not that it
couldn't be changed, but unless there actually is a use-case where more
than 64 KiB are required I don't think we should worry about it.

Thierry

Attachment: pgp00000.pgp
Description: PGP signature