Re: [Freedreno] [PATCH] drm/msm/dp: make eDP panel as the first connected connector

[Abhinav Kumar]

On 7/4/2022 11:14 AM, Dmitry Baryshkov wrote:
> On 30/06/2022 09:14, Dmitry Baryshkov wrote:
> > On 30 June 2022 04:57:35 GMT+03:00, Rob Clark <robdclark@xxxxxxxxx> 
> > wrote:
> > > On Wed, Jun 29, 2022 at 5:36 PM Doug Anderson <dianders@xxxxxxxxxxxx> 
> > > wrote:
> > > > Hi,
> > > >
> > > > On Tue, Jun 28, 2022 at 1:14 PM Dmitry Baryshkov
> > > > <dmitry.baryshkov@xxxxxxxxxx> wrote:
> > > > > On 28 June 2022 18:20:06 GMT+03:00, Kuogee Hsieh 
> > > > > <quic_khsieh@xxxxxxxxxxx> wrote:
> > > > > > Some userspace presumes that the first connected connector is the 
> > > > > > main
> > > > > > display, where it's supposed to display e.g. the login screen. For
> > > > > > laptops, this should be the main panel.
> > > > > >
> > > > > > This patch call drm_helper_move_panel_connectors_to_head() after
> > > > > > drm_bridge_connector_init() to make sure eDP stay at head of
> > > > > > connected connector list. This fixes unexpected corruption happen
> > > > > > at eDP panel if eDP is not placed at head of connected connector
> > > > > > list.
> > > > >
> > > > > The change itself is a good fix anyway. (And I'd ack it.) However I 
> > > > > would like to understand why does it fix the corruption issue. What 
> > > > > is we have eDP and DSI, with DSI ending up before the eDP? Would we 
> > > > > see the issue?
> > > > > Also could you please describe the mind of corruption you are 
> > > > > observing?
> > > >
> > > > I've spent a whole bunch of time poking at this and in the end my
> > > > conclusion is this:
> > > >
> > > > 1. The glitchyness seems to be a result of the Chrome OS userspace
> > > > somehow telling the kernel to do something wrong.
> > > >
> > > > 2. I believe (though I have no proof other than Kuogee's patch fixing
> > > > things) that the Chrome OS userspace is simply confused by the eDP
> > > > connector being second. This would imply that Kuogee's patch is
> > > > actually the right one.
> > > >
> > > > 3. It would be ideal if the Chrome OS userspace were fixed to handle
> > > > this, but it's an area of code that I've never looked at. It also
> > > > seems terribly low priority to fix since apparently other OSes have
> > > > similar problems (seems like this code was originally added by
> > > > RedHat?)
> > > >
> > > >
> > > > Specifically, I tested with a similar but "persistent" glitch that I
> > > > reproduced. The glitch Kuogee was digging into was a transitory glitch
> > > > on the eDP (internal) display when you plugged in a DP (external)
> > > > display. It would show up for a frame or two and then be fixed. I can
> > > > get a similar-looking glitch (vertical black and white bars) that
> > > > persists by doing these steps on a Chrome OS device (and Chrome OS
> > > > kernel):
> > > >
> > > > a) Observe screen looks good.
> > > > b) Observe DP not connected.
> > > > c) Plug in DP
> > > > d) See transitory glitch on screen, then it all looks fine.
> > > > e) set_power_policy --ac_screen_dim_delay=5 --ac_screen_off_delay=10
> > > > f) Wait for screen to turn off
> > > > g) Unplug DP
> > > > h) Hit key on keyboard to wake device.
> > > > i) See glitchy.
> > > > j) Within 5 seconds: set_power_policy --ac_screen_dim_delay=5000
> > > > --ac_screen_off_delay=10000
> > > >
> > > > Once I'm in the persistent glitch:
> > > >
> > > > * The "screenshot" command in Chrome OS shows corruption. Not exactly
> > > > black and white bars, but the image produced has distinct bands of
> > > > garbage.
> > > >
> > > > * I can actually toggle between VT2 and the main screen (VT1). Note
> > > > that VT1/VT2 are not quite the normal Linux managed solution--I
> > > > believe they're handled by frecon. In any case, when I switch to VT2
> > > > it looks normal (I can see the login prompt). Then back to VT1 and the
> > > > vertical bars glitch. Back to VT2 and it's normal. Back to VT1 and the
> > > > glitch again. This implies (especially with the extra evidence of
> > > > screenshot) that the display controller hardware is all fine and that
> > > > it's the underlying data that's somehow messed up.
> > >
> > > fwiw, from looking at this a bit w/ Doug, I think the "glitch" is
> > > simply just an un-renderered buffer being interpreted by the display
> > > controller as UBWC (because userspace tells it to)
> >
> > Thanks for the description. I think the userspace code should be fixed 
> > too, but this patch can go in on its own.
> >
> > Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
>
> After some time (please excuse me), musing with the code and even 
> picking up the commit for the merge branch, I understood the fact that I 
> did not like about this change. It moves all panel connectors (generic 
> code) from the DP-specific driver.
>
> I'd like to retract my R-b. Please move this call to the msm_drm_init(). 
> Calling this function somewhere after the ->kms_init() would make sure 
> that all panel connectors are close to the top of the list, whichever 
> MDP/DPU driver is used and whichever actual interface is bound to this 
> panel.
Ah. True, but just to add. It should be after kms_init() but before 
drm_dev_register().

> > > BR,
> > > -R
> > >
> > > > When I pick Kuogee's patch then this "persistent" glitch goes away
> > > > just like the transitory one does.
> > > >
> > > > I'm going to go ahead and do:
> > > >
> > > > Reviewed-by: Douglas Anderson <dianders@xxxxxxxxxxxx>
> > > > Tested-by: Douglas Anderson <dianders@xxxxxxxxxxxx>