Re: [PATCHv2 0/5] OMAP DSS HWMOD fixes

[Tomi Valkeinen <tomi.valkeinen@ti.com>]

On Mon, 2011-08-08 at 17:51 +0530, Archit Taneja wrote:
> Hi,
> 
> On Monday 08 August 2011 05:33 PM, Valkeinen, Tomi wrote:
> > On Mon, 2011-08-08 at 17:13 +0530, Archit Taneja wrote:
> >> Hi,
> >>
> >> On Monday 08 August 2011 02:45 PM, Valkeinen, Tomi wrote:
> >>> Second try with the DSS HWMODs
> >>>
> >>> This set fixes the DSS clocks in HWMOD data, and implements a new reset
> >>> mechanism for dss_core.
> >>>
> >>> The new dss_reset function doesn't actually do a reset, it just enables all DSS
> >>> clocks and waits for the reset to complete. This should be better approach than
> >>> actually doing a reset, because:
> >>>
> >>> OMAP4 - dss_core HW doesn't contain a SW reset bit so doing a reset is
> >>> impossible. But after power-on we need to enable all DSS clocks and wait for
> >>> the power-on reset to complete.
> >>>
> >>> OMAP2/3 - dss_core does have a SW reset bit, but resetting dss_core also resets
> >>> all the other DSS modules. This means that the other modules could be left
> >>> uninitialized, as the hwmod code handles all modules independently, and in this
> >>> case initializes only dss_core's registers. Thus dss_core's reset shouldn't be
> >>> used, and we should only verify that the power-on reset has completed.
> >>
> >> If the bootloader enables DSS, we need to do a reset so that DSS2 driver
> >> can start with DSS HW in a clean state. With this patch set, how do we
> >> take care of such a scenario?
> >
> > We don't. That should be done in a future patch.
> >
> > I think we should extend this reset function, and disable the LCD
> > outputs and reset the clock switches there.
> >
> > I didn't want to do that yet as I wanted to fix the current problem with
> > the reset first and I don't have an environment where to test it. I hope
> > you can help here =).
> 
> Yes, we can extend over this. But we would still access DISPC registers 
> in the dss_core's custom reset function. Wouldn't this break the "HWMOD 
> model where every DSS module is independent"?

Yes, it won't be as neat as I'd like to, but it shouldn't break
functionally anything.

I thought of adding a new reset function for dss_dispc, but I don't
think that would work, as we need to reset the clock switches (in
dss_core). And if we'd reset the clock switches while dispc output is
enabled, things could again break.

So only solution I can think of is to have dss_core do all those things.
Well, another solution would be to skip the DSS resets totally if DSS
output is enabled. I'm not sure if that's a good solution, though.

 Tomi