Re: Fwd: Re: [PATCHv5 0/3] drm/i915: add DisplayPort CEC-Tunneling-over-AUX support

[Hans Verkuil <hverkuil@xs4all.nl>]

On 01/12/2018 07:12 PM, Hans Verkuil wrote:
> On 01/12/2018 06:52 PM, Ville Syrjälä wrote:
>> On Fri, Jan 12, 2018 at 06:14:53PM +0100, Hans Verkuil wrote:
>>> On 01/12/2018 05:30 PM, Ville Syrjälä wrote:
>>>> On Fri, Jan 12, 2018 at 05:19:44PM +0100, Hans Verkuil wrote:
>>>>> Hi Ville,
>>>>>
>>>>> For some strange reason your email disappeared from the Cc list. Perhaps it's the
>>>>> ä that confuses something somewhere.
>>>>>
>>>>> So I'll just forward this directly to you.
>>>>>
>>>>> Can you please take a look? This patch series has been in limbo for too long.
>>>>
>>>> IIRC last I looked we still had some ragistration race to deal with.
>>>> Was that fixed?
>>>
>>> That was fixed in v5.
>>>
>>>>
>>>> Also I think we got stuck on leaving the zombie device lingering around
>>>> when the display is disconnected. I couldn't understand why that is
>>>> at all useful since you anyway remove the device eventually.
>>>
>>> It's not a zombie device. If you disconnect and reconnect the display then the
>>> application using the CEC device will see the display disappear and reappear
>>> as expected.
>>>
>>> It helps if you think of the normal situation (as is present in most ARM SoCs)
>>> where CEC is integral to the HDMI transmitter. I.e. it is not functionality that
>>> can be removed. So the cec device is always there and an application opens the
>>> device and can use it, regardless of whether a display is connected or not.
>>>
>>> If a display is detected, the EDID will be read and the CEC physical address is
>>> set. The application is informed of that through an event and the CEC adapter
>>> can be used. If the HPD disappears the physical address is reset to f.f.f.f and
>>> again the application is informed. And in fact it still has to be able to use
>>> the CEC adapter even if there is no HPD since some displays turn off the HPD when
>>> in standby, but CEC can still be used to power them up again.
>>
>> Hmm. So you're saying there are DP devices out there that release HPD
>> but still respond to DPCD accesses? That sounds... wrong.
> 
> Not quite. To be precise: there are HDMI displays that release HPD when in standby
> but still respond to CEC commands.
> 
> Such displays are still being made today so if you are building a product like
> a media streaming box, then this is something to take into account.
> 
> However, for this specific case (CEC tunneling) it is a non-issue since the
> DP CEC protocol simply doesn't support sending CEC commands without HPD.
> 
>> In general I don't think we can assume that a device has retained its
>> state if it has deasserted HPD, thus we have to reconfigure everything
>> again anyway.
>>
>>>
>>> Now consider a future Intel NUC with an HDMI connector on the backplane and
>>> working DP CEC-Tunneling-over-AUX support (e.g. the Megachips MCDP2900): the
>>> CEC support is always there (it's built in), but only becomes visible to the
>>> kernel when you connect a display. You don't want the cec device to disappear
>>> whenever you unplug the display, that makes no sense. Applications would
>>> loose the CEC configuration and have to close and reopen (when it reappears)
>>> the cec device for no good reason since it is built in.
>>
>> If the application can't remember its settings across a disconnect it
>> sounds broken anwyay. This would anyway happen when you disconenct the
>> entire dongle.
> 
> Huh?
> 
>>
>>>
>>> The same situation is valid when using a USB-C to HDMI adapter: disconnecting
>>> or reconnecting a display should not lead to the removal of the CEC device.
>>> Only when an adapter with different CEC capabilities is detected is there a
>>> need to actually unregister the CEC device.
>>>
>>> All this is really a workaround of the fact that when the HPD disappears the
>>> DP-to-HDMI adapter (either external or built-in) also disappears from the
>>> topology, even though it is physically still there.
>>
>> The dongles I've seen do not disappear. The downstream hpd is
>> signalled with short hpd pulses + SINK_COUNT instead.
>>
>> But I've not actually seen a dongle that implements the
>> BRANCH_DEVICE_CTRL DPCD register, so not quite sure what those would
>> actually do. The spec does say they should default to using long
>> hpd for downstream hpd handling.
> 
> I did a few more experiments and it appears that someone somewhere keeps
> track of DP branch devices. I.e. after disconnecting my usb-c to hdmi adapter
> it still appears in i915_display_info. At least until something else is
> connected. I basically need to hook into the DP branch detection.
> 
> Something to look at this weekend.

I found a better place to do the CEC (un)registration: a long HPD pulse
indicates that the DPCD registers have changed, so that's when I should
detect whether there is a new branch device with CEC capabilities.

Regards,

	Hans