[PATCH v6 1/5] drivers: memory: Introduce Marvell EBU Device Bus driver

[ezequiel.garcia@free-electrons.com (Ezequiel Garcia)]

Hi Jason,

On Wed, Apr 10, 2013 at 10:57:50AM -0600, Jason Gunthorpe wrote:
> On Wed, Apr 10, 2013 at 07:19:57AM -0300, Ezequiel Garcia wrote:
> > Hi Jason,
> > 
> > On Tue, Apr 09, 2013 at 02:38:55PM -0600, Jason Gunthorpe wrote:
> > > On Tue, Apr 09, 2013 at 05:26:17PM -0300, Ezequiel Garcia wrote:
> > > > +	/*
> > > > +	 * Since this is only part of the workaround,
> > > > +	 * we can do this dirty 'ranges' property parsing.
> > > > +	 * Of course, this will be removed once the address windows
> > > > +	 * are declared in the device tree.
> > > > +	 */
> > > > +	err = of_property_read_u32_array(node, "ranges", addr_region,
> > > > +					 ARRAY_SIZE(addr_region));
> > > 
> > > Oh, this is a dangerous way to parse ranges.. 
> > > 
> > > Try something like:
> > > 
> > > const __be32 *prop;
> > > u64 base;
> > > u64 size;
> > > 
> > > prop = of_get_property("ranges",&psize);
> > > if (prop == NULL || psize != of_n_addr_cells(node) + of_n_size_cells(node))
> > >    err;
> > > base = of_translate_address(node,prop);
> > > if (base == OF_BAD_ADDR)
> > >    err;
> > > size = of_read_number(prop + of_n_addr_cells(node),of_n_size_cells(node));
> > > 
> > 
> > Thanks for giving me these hints!
> 
> NP, the motivation for doing this properly is to support cases where
> p_addr_cells/addr_cells/size_cells is != 1, and to support cases where
> the parent bus address is not the CPU address. The proposed bindings
> for mbus will exercise both of these conditions..
> 
> > Following your suggestion, here's a working piece of the ranges parsing code,
> > how does this look?
> > 
> > const __be32 *ranges;
> > int addr_cells, p_addr_cells, size_cells;
> > int ranges_len, tuple_len;
> > u32 base, size;
> > 
> > p_addr_cells = of_n_addr_cells(node->parent);
> 
> I wonder if of_get_parent / of_node_put is technically needed here?
>

Yes, that sounds definitely better.

> > addr_cells = of_n_addr_cells(node);
> > size_cells = of_n_size_cells(node);
> > tuple_len = (p_addr_cells + addr_cells + size_cells) * sizeof(__be32);
> > 
> > ranges = of_get_property(node, "ranges", &ranges_len);
> > if (ranges == NULL || ranges_len != tuple_len)
> > 	return -EINVAL;
> > 
> > base = of_translate_address(node, ranges + addr_cells);
> > if (base == 0 || base == OF_BAD_ADDR)
> > 	return -EINVAL;
> 
> Why the check for base == 0?
> 

Well, when writing this code my first experiments where failing
(because the code was wrong) and I was getting base=0.
So I added this to catch that error.

Thinking it twice, it doesn't seem correct to check that.

> > size = of_read_number(ranges + addr_cells + p_addr_cells, size_cells);
> 
> Everything basically looks right. I definitely forgot the include the
> p_addr in my quick sample.
> 

Great, I'll send the v7, with these two fixes.

-- 
Ezequiel Garc?a, Free Electrons
Embedded Linux, Kernel and Android Engineering
http://free-electrons.com