Re: Device Tree updates for xilinx.

[David Gibson <david@gibson.dropbear.id.au>]

On Thu, Dec 13, 2007 at 03:41:16PM -0800, Stephen Neuendorffer wrote:
> These patches synchronize all the in-kernel drivers to use the
> compatible names generated by the UBoot BSP generator.
> (at git://git.xilinx.com/gen-mhs-devtree.git)
> 
> The patches to make this work are coming shortly:
> 
> patches 1-2 are provided for context only...  They're not ready for
> mainline.  They provide 'raw' boot support and port some initialization
> code from ARCH=ppc.
> patches 3-7 are the interesting patches, which I think could be taken
> for 2.6.25.
> 
> I've also pushed the working tree up to git.xilinx.com containing these
> patches.  This tree also includes updates for ps2, gpio, etc.
> The ll_temac currently doesn't work because I haven't gone back and
> fixed the mechanism by which it finds out about the DMA that it is
> connected to.
> 
> If you want output from the boot loader, you'll need something like:
> 		linux,stdout-path = "/plb_v34/opb_v20/serial@40400000"; 
> This seems to work with uartlite, but I wasn't able to get it working
> with a UART design.
> 
> Currently, you'll need to add, by hand the following labels:
> 
> mem_size_cells, on the toplevel #address-cells attribue, e,g. :
> 	mem_size_cells: #address-cells = <1>;

Having a label called *_size_cells which is attached to
#address-cells, rather than #size-cells seems like a bad idea.  Plus,
if I understood correctly, this is used to interpret the 'memsize'
value, which means it should be #size-cells anyway.

Also, if I understood the code in your later patches correctly, this
label is used as a direct pointer to the value you want.  That means
it needs to be on the property value, not on the property, i.e.
	#size-cells = <mem_size_cells: 1>;
instead of
	mem_size_cells: #size-cells = <1>;
(the latter form will put the label on the tag which introduces the
property instead).

> timebase, on the processors timebase-frequency attribute, e.g. :
> 			timebase: timebase-frequency = <11e1a300>;
> 
> memsize, on the memory's size, e.g.:
> 	DDR_256MB_32MX64_rank1_row13_col10_cl2_5: memory@0 {
> 		device_type = "memory";
> 		reg = < 0 memsize:10000000 >;
> 	} ;
> 
> In addition, if you're using uartlite, you'll have to specify
> 'console=ttyUL0' as boot args.
> 
> For reference, below is the device tree for a Virtex2Pro design.  Except
> for the changes noted above, this is entirely automatically generated.
> 
> Steve
> 
> / {
> 	mem_size_cells: #address-cells = <1>;
> 	#size-cells = <1>;
> 	compatible = "xlnx,virtex";
> 	model = "testing";
> 	DDR_256MB_32MX64_rank1_row13_col10_cl2_5: memory@0 {
> 		device_type = "memory";
> 		reg = < 0 memsize:10000000 >;
> 	} ;
> 	chosen {
> 		bootargs = "root=/dev/nfs
> nfsroot=172.19.221.221:/exports/xup/ydl41 ip=dhcp console=ttyUL0";
> 	} ;
> 	cpus {
> 		#address-cells = <1>;
> 		#cpus = <1>;
> 		#size-cells = <0>;
> 		PowerPC,405@0 {

I'm trying to encourage people to move to naming cpu nodes simply
"cpu@XX", with the type of core in "compatible" instead - in keeping
with the generic names convention used elsewhere.  I think this will
make your generator simpler, too.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson