Re: fw_devlink breakage on SDM845 / a630 with DSI displays

[Caleb Connolly <caleb@connolly.tech>]

On 20/05/2021 1:06 am, Saravana Kannan wrote:
> On Wed, May 19, 2021 at 3:43 PM Caleb Connolly <caleb@connolly.tech> wrote:
>>
>> Hi Saravana,
>>
>> On 19/05/2021 10:52 pm, Saravana Kannan wrote:
>>> On Wed, May 19, 2021 at 8:36 AM Rob Clark <robdclark@gmail.com> wrote:
>>>>
>>>> + some more folks and msm list..
>>>>
>>>> I suppose I didn't hit this because CONFIG_FBDEV_EMULATION is not
>>>> normally enabled for CrOS.. but I'm not really to familiar with
>>>> fw_devlink
>>>>
>>>> BR,
>>>> -R
>>>>
>>>> On Wed, May 19, 2021 at 8:26 AM Caleb Connolly <caleb@connolly.tech> wrote:
>>>>>
>>>>>
>>>>> Hi,
>>>>>
>>>>> Since -rc1 I've been hit by some DRM breakage in freedreno which happens
>>>>> when fw_devlink=on. It seems to cause some clocks (rcg) to get stuck and
>>>>> break DSI displays, here's a full log from the OnePlus 6:
>>>>> https://paste.ubuntu.com/p/8kPx4nFGF5/ (that is with
>>>>> "deferred_probe_timeout") The PocoPhone F1 also seems to be affected by
>>>>> this.
>>>>>
>>>>> The display will still come up after pressing the power button a few
>>>>> times, although it will be incredibly slow.
>>>>>
>>>>> It's worth noting that the issue only happens with
>>>>> CONFIG_FBDEV_EMULATION is enabled, I've previously required this to see
>>>>> kernel logs during boot and general boot splash with postmarketOS.
>>>>> Without it the display will be stuck on the bootloader splash until I
>>>>> press the power button and cause it to update once UI (like Phosh) has
>>>>> started (though this has been the case for quite some time).
>>>>>
>>>>> I'd appreciate any help with debugging / resolving this issue, I'm
>>>>> afraid I haven't been able to determine much from some brief digging.
>>>>>
>>>
>>> Hi Caleb,
>>>
>>> Is this a device that's supported upstream? If so, can you please
>>> point me to the DTS file that corresponds to this board?
>> The DTS can be found in
>> arch/arm64/boot/dts/qcom/sdm845-oneplus-common.dtsi (the two devices
>> "enchilada" and "fajita" share almost all hardware in common).
>>>
>>> Also, can you please change all the dev_dbg to dev_info in these
>>> functions and give me the full boot log?
>>> device_link_add()
>>> device_links_check_suppliers()
>> I've uploaded a log here: https://paste.ubuntu.com/p/8ynFgRWbYW/
>> For reference, the same but with fw_devlink=permissive:
>> https://paste.ubuntu.com/p/F2853CphHb/
>>>
>>> Can you also tell what device are not probing with fw_devlink=on that
>>> might be probing with fw_devlink=permissive?
>> The devices in question are ae00000.mdss, ae94000.dsi and ae01000.mdp.
>>> You should be able to compare /<debugfs>/devices_deferred to figure that out.
>> devices_deferred is empty for me,
>
> So, ae00000.mdss, ae94000.dsi and ae01000.mdp are probing for =on and
> =permissive?
>
> That's interesting. So maybe fw_devlink is somehow changing the probe
> order that's causing this? AFAIK, fw_devlink shouldn't cause changes
> to probe order (it used to, but I've fixed them all).
Yeah, they always successfully probe, it seems like fw_devlink is
exposing some racy behaviour somewhere.

The issue happens when initialising the display for framebuffer device
emulation - with framebuffer emulation disabled it all mostly works as
expected. With the caveat that there is nothing on the display until the
device has finished booting AND the user causes a refresh by pressing
the power button a few times.

Perhaps this early framebuffer driver is causing confusion as it has
different dependencies than the main driver?
>
>> however device_component contains
>> (only) the following:
>> oneplus6:/home/user# cat /sys/kernel/debug/device_component/ae00000.mdss
>> master name                                            status
>> -------------------------------------------------------------
>> ae00000.mdss                                            bound
>>
>> device name                                            status
>> -------------------------------------------------------------
>> ae01000.mdp                                             bound
>> ae94000.dsi                                             bound
>> 5000000.gpu                                             bound
>
> Sorry, I have no context on this and how to interpret it. I kinda know
> a little about the component framework, but not much. Is this file the
> same for =on and =permissive?
Apologies, I assumed it had something to do with fw_devlink. It's the
same for both cases yeah.
>
> -Saravana
>

--
Kind Regards,
Caleb