[PATCH v3 0/6] uio_pruss cleanup and platform support

[mporter@ti.com (Matt Porter)]

On Thu, Oct 04, 2012 at 11:11:45AM +0200, Philipp Zabel wrote:
> Hi Matt,
> 
> On 10/3/12, Matt Porter <mporter@ti.com> wrote:
> > This series enables uio_pruss on DA850 and removes use of the
> > private SRAM API by the driver. The driver previously was not
> > enabled by any platform and the private SRAM API was accessing
> > an invalid SRAM bank.
> 
> have you seen my SRAM patch series at https://lkml.org/lkml/2012/9/7/281
> "Add device tree support for on-chip SRAM" ?

Yes.

> I think the generic SRAM/genalloc driver (https://lkml.org/lkml/2012/9/7/282)
> could be useful to map the L3RAM on Davinci.
> With the gen_pool lookup patch (https://lkml.org/lkml/2012/9/7/284) the
> uio_pruss driver could then use the gen_pool_find_by_phys() (or
> of_get_named_gen_pool() for initialization from device tree) to
> retrieve the struct gen_pool*.
> 
> This way you could avoid handing it over via platform data and you could
> get rid of arch/arm/mach-davinci/{sram.c,include/mach/sram.h} completely.

I did miss the gen_pool_find_by_phys() call in that series. That does
look useful. I actually mentioned your series in an earlier posting
since I like it, but since the initialization of the driver was inherently
tied to DT it's not usable for DaVinci that's just starting to convert
to DT and needs !DT support as well.

I do see it moving to your driver exclusively, but I wanted to make this
series focused on only getting rid of the private SRAM API using the
existing pdata framework that's already there. I think once
gen_pool_find_by_phys() goes upstream we can switch to that and get the
address from a resource in the !DT case. I guess we should see if Sekhar
would like to see this happen in two steps or just have us depend on
the gen_pool_find_by_phys() patch now.

BTW, I was going to post a patch for your driver to allow
configurability of the allocation order, but have been busy with other
things.  We'll eventually need that when switching to it as the
hardcoded page size order isn't going to work for all cases.

-Matt