[PATCH v3 00/56] Convert DSI code to use drm_mipi_dsi and drm_panel

[H. Nikolaus Schaller]

> Am 10.11.2020 um 14:49 schrieb H. Nikolaus Schaller <hns at goldelico.com>:
> 
> Hi Tomi,
> 
>> Am 09.11.2020 um 12:33 schrieb Tomi Valkeinen <tomi.valkeinen at ti.com>:
>> 
>> On 09/11/2020 13:09, H. Nikolaus Schaller wrote:
>> 
>>>>> I see.
>>>>> Anyways there is missing some simple thing which makes the driver not prepared/enabled.
>>>>> Or is this related to VC?
>>>> 
>>>> No, that's not related to the VC.
>>> 
>>> Ok, then it is worth searching for that independently. Any idea/hint what could be missing?
>> 
>> Well, if I had to guess, I would go for either 1) some registration or such is missing from the
>> panel driver, or 2) some config value is invalid, which makes the DRM framework or the DSI driver
>> fail before calling prepare or enable.
> 
> I did now some tests based on v5.10-rc3 by adding mre and more printk() and dump_stack().
> And I did blacklist the panel driver so that I could boot and after modprobing it manually
> I could trigger a re-probe by inserting some USB memory stick.
> 
> With this procedure I could trace the problem down to this call sequence:
> 
> 	dsi_probe()
>          ...
> 	  w677l_probe()
>            drm_panel_add()   -- after this, of_drm_find_panel() is successful
>          ...
> 	  component_add()
> 	    try_to_bring_up_master()
> 	      master->ops->bind(master->dev)
> 
> This call to bind() does not return and likely the probing thread is then blocked and
> holds some locks which manifests itself in that the system isn't running stable any
> more (e.g. heartbeat LEDs are sometimes stuck although console still works).
> 
> dbg_info() in try_to_bring_up_master() prints:
> 
> [  102.199362] omapdss_dss 58000000.dss: trying to bring up master
> 
> So I am not sure if this is a panel driver issue at all or the DSI patch set is not
> running stable on OMAP5.
> 
> Is a good next step to trace dss_bind() in drivers/gpu/drm/omapdrm//dss/dss.c?

There is indeed one kernel worker running at 100% CPU load.

top:

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND                                                                   
 3196 root      20   0       0      0      0 R  100.0  0.0   2:58.76 kworker/0:8+events                                                        

More analysis shows that it hangs in drm_client_modeset_probe() in the loop

	for (i = 0; i < connector_count; i++)
		total_modes_count += connectors[i]->funcs->fill_modes(connectors[i], width, height);

Most likely not in the loop because that looks sane, but connectors[i]->funcs->fill_modes().

So I have to find out what function connectors[i]->funcs->fill_modes is...

BTW: connector_count = 2.

BR and thanks,
Nikolaus