On Friday 04 May 2012 03:23 PM, Tomi Valkeinen wrote:
On Fri, 2012-05-04 at 14:39 +0530, Archit Taneja wrote:
On Thursday 03 May 2012 07:28 PM, Tomi Valkeinen wrote:
We currently use the id of the dsi platform device (dsidev->id) as the
DSI hardware module ID. This works because we assign the ID manually in
arch/arm/mach-omap2/display.c at boot time.
However, with device tree the platform device IDs are automatically
assigned to an arbitrary number, and we can't use it.
If this number is arbitrary we would need to change the "dsi_pdev_map"
approach of mapping a dsi module and it's corresponding platform device.
Currently dsi_pdev_map is:
static struct platform_device *dsi_pdev_map[MAX_NUM_DSI];
So we either need to increase the array size to take larger arbitrary
numbers, or do something else.
We would also need to fix the usage of dsi_get_dsidev_from_id(), as
right now we manually pass 0 and 1 to it only, for example:
static void dsi1_dump_irqs(struct seq_file *s)
{
struct platform_device *dsidev = dsi_get_dsidev_from_id(0);
dsi_dump_dsidev_irqs(dsidev, s);
}
The immediate solution that comes to mind is to maintain 2 id's, one
which is sequential, and the other which the DT has created, and keep an
array to map these. But this seems messy!
This is only a problem with device tree, and I solved it so that I pass
a DSI module ID in the device tree data. So, with old pdata way I
initialize dsi->module_id from the pdev->id, but with DT I initialize
dsi->module_id from the DT data.
Oh ok, so the code which decides how dsi->module_id is initialised(from
DT or pdata) is not in this series right? And it would come later? Right
now it's just set to dsidev->id in probe.
So basically we remove the use of pdev->id in this patch, and add
dsi->module_id field, which needs to be initialized to 0 or 1, depending
on the corresponding HW module. We just happen to use the pdev->id to
initialize it when using the old pdata method, as we know it tells the
right id. But we could initialize it from any other source.
Right, I get it now.
This allows us to keep the 0 and 1 DSI IDs, and I think we need those
anyway. Some parts of the code could work fine with arbitrary ID, as
long as a pdev can be linked to/from this ID. However, there are things
where we must have the ID, like configuring the clock source settings in
dss_core, where we set a certain bit for DSI module 0, and certain bit
for module 1.
Perhaps even those could be handled without explicit ID of 0 or 1, but
that doesn't sound trivial and I didn't want to start tackling that in
this series.
I wish there was a way to get the module ID from the HW registers
somehow. Then we wouldn't need to pass the ID via SW, which doesn't feel
very correct. At least with DT it's a bit wrong, in my opinion, but best
I could come up with.
We could derive it via a parameter like number of lanes or something
similar through DSI_GNQ, but that doesn't seem very nice, and may not be
usable on future OMAPs.
Archit
Tomi
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
