[Letux-kernel] omapdrm: Pandora Blues

[H. Nikolaus Schaller]

> Am 06.11.2017 um 17:00 schrieb Tomi Valkeinen <tomi.valkeinen at ti.com>:
> 
> On 06/11/17 16:04, H. Nikolaus Schaller wrote:
>> Hi,
>> some time after upgrading to 4.14-rc* I tried to boot the OpenPandora.
>> It turned out that the display panel has a bug - it only shows black/blue colors
>> instead of RGB.
>> 
>> Some tests revealed that something happened between 4.13.0 and 4.14-rc1.
>> Here are screen photos:
>> 
>> 4.13.0:   http://download.goldelico.com/letux-kernel/files/thumb_DSC00812_1024.jpeg
>> 4.14-rc1: http://download.goldelico.com/letux-kernel/files/thumb_DSC00813_1024.jpeg
>> 
>> But only for the OpenPandora. For the GTA04 it works.
>> 
>> Well, the GTA04 is using a different panel "toppoly,td028ttec1"
>> and driver instead of "omapdss,tpo,td043mtea1".
>> [BTW: there seems to be some mixup of "compatible" schemes].
> 
> Omapdss boot-time code will prepend the compatibles with "omapdss,". The
> point is that the DTS is generic, but we'll still end up matching with
> the omapdss specific drivers. It's a temporary hack, and gets dropped as
> soon as we can use the common DRM panel model.
> 
>> And there is also another difference between both: the Pandora 600MHz uses an
>> OMAP3530 while the GTA04 uses a DM3730.
>> 
>> So something has become incompatible with *some* DPI panel drivers.
>> 
>> After more than a week of bisecting in parallel to important other tasks
>> (it takes ca. 30-60 minutes for each run to add local patches, compile,
>> install, boot, check results - just to find some "[drm:omap_crtc_error_irq] *ERROR* lcd: errors: 00004000"),
>> I ended up with a specific result:
>> 
>> 
>> iMac:master hns$ git bisect bad
>> d178e034d5653edfbd16d0c71eeeed467e33c96f is the first bad commit
>> commit d178e034d5653edfbd16d0c71eeeed467e33c96f
>> Author: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
>> Date:   Sat Aug 5 01:44:12 2017 +0300
>> 
>>   drm: omapdrm: Move FEAT_DPI_USES_VDDS_DSI feature to dpi code
>> 
>>   The FEAT_DPI_USES_VDDS_DSI feature is specific to the DPI, move it from
>>   the omap_dss_features structure to the dpi code.
> 
> Well, the patch is simple enough... For some reason, the
> soc_device_match(dpi_soc_devices) call there doesn't match Pandora.

Ok, I think I have found it. The pattern in dpi_soc_devices[] is wrong.

Here is what my printk hack() reports on the 600 MHz Pandora (OMAP3530):

[   20.171325] soc_device_match_attr()
[   20.175445] attr
[   20.177398]   machine = OMAP3430/3530
[   20.181274]   family = OMAP3
[   20.184295]   revision = ES3.1
[   20.188232]   soc_id = null
[   20.191192] match
[   20.193237]   machine = null
[   20.196411]   family = OMAP3[456]*
[   20.199981]   revision = null
[   20.203124]   soc_id = null
[   20.182403] dpi_init_regulator() match

Changing the pattern to "OMAP[3456]*" makes it work but I am not sure what exactly is intended.
Should the pattern match the .machine and not the .family?

The DM3730 based GTA04 reports (without patch):

[   18.618621] soc_device_match_attr()
[   18.622314] attr
[   18.624237]   machine = OMAP3630/DM3730
[   18.628509]   family = OMAP3
[   18.631561]   revision = ES1.2
[   18.634765]   soc_id = null
[   18.638000] match
[   18.640014]   machine = null
[   18.643035]   family = OMAP3[456]*
[   18.646697]   revision = null
[   18.649810]   soc_id = null
[   18.652770] soc_device_match_attr()
[   18.656616] attr
[   18.658538]   machine = OMAP3630/DM3730
[   18.662567]   family = OMAP3
[   18.665649]   revision = ES1.2
[   18.668853]   soc_id = null
[   18.671813] match
[   18.673828]   machine = null
[   18.677001]   family = [AD]M37*
[   18.680297]   revision = null
[   18.683410]   soc_id = null
[   18.686614] dpi_init_regulator() no match

Here we have no match and don't enable VDDS_DSI - but it still works (maybe VDDS_DSI is enabled elsewhere).
And I do not see how the second pattern .family = "[AD]M37*" can ever match.

According to the comment in the original patch:

"The DPI uses the DSI VDDS on OMAP34xx, OMAP35xx, OMAP36xx, AM37xx and DM37xx only."

a single match for .family = "OMAP3" should suffice. Shouldn't it?
And there is no family "AM/DM37*". Or is it?

A tiny and subtle flaw which has such a strange result...

But I don't know what would be correct.

BR and thanks,
Nikolaus