On Mon, Sep 18, 2017 at 05:26:43PM +0200, Hans Verkuil wrote:
> On 09/18/2017 04:36 PM, Ville Syrjälä wrote:
> > On Mon, Sep 18, 2017 at 04:07:41PM +0200, Hans Verkuil wrote:
> >> Hi Ville,
> >>
> >> On 09/18/2017 03:05 PM, Ville Syrjälä wrote:
> >>> On Sat, Sep 16, 2017 at 04:17:26PM +0200, Hans Verkuil wrote:
> >>>> From: Hans Verkuil <hans.verkuil@xxxxxxxxx>
> >>>>
> >>>> Implement support for this DisplayPort feature.
> >>>>
> >>>> The cec device is created whenever it detects an adapter that
> >>>> has this feature. It is only removed when a new adapter is connected
> >>>> that does not support this. If a new adapter is connected that has
> >>>> different properties than the previous one, then the old cec device is
> >>>> unregistered and a new one is registered to replace the old one.
> >>>>
> >>>> Signed-off-by: Hans Verkuil <hans.verkuil@xxxxxxxxx>
> >>>> Tested-by: Carlos Santa <carlos.santa@xxxxxxxxx>
> >>>> ---
> >>>> drivers/gpu/drm/i915/intel_dp.c | 18 ++++++++++++++----
> >>>> 1 file changed, 14 insertions(+), 4 deletions(-)
> >>>>
> >>>> diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c
> >>>> index 64fa774c855b..fdb853d2c458 100644
> >>>> --- a/drivers/gpu/drm/i915/intel_dp.c
> >>>> +++ b/drivers/gpu/drm/i915/intel_dp.c
> >>>> @@ -32,6 +32,7 @@
> >>>> #include <linux/notifier.h>
> >>>> #include <linux/reboot.h>
> >>>> #include <asm/byteorder.h>
> >>>> +#include <media/cec.h>
> >>>> #include <drm/drmP.h>
> >>>> #include <drm/drm_atomic_helper.h>
> >>>> #include <drm/drm_crtc.h>
> >>>> @@ -1449,6 +1450,7 @@ static void intel_aux_reg_init(struct intel_dp *intel_dp)
> >>>> static void
> >>>> intel_dp_aux_fini(struct intel_dp *intel_dp)
> >>>> {
> >>>> + cec_unregister_adapter(intel_dp->aux.cec_adap);
> >>>> kfree(intel_dp->aux.name);
> >>>> }
> >>>>
> >>>> @@ -4587,6 +4589,7 @@ intel_dp_set_edid(struct intel_dp *intel_dp)
> >>>> intel_connector->detect_edid = edid;
> >>>>
> >>>> intel_dp->has_audio = drm_detect_monitor_audio(edid);
> >>>> + cec_s_phys_addr_from_edid(intel_dp->aux.cec_adap, edid);
> >>>> }
> >>>>
> >>>> static void
> >>>> @@ -4596,6 +4599,7 @@ intel_dp_unset_edid(struct intel_dp *intel_dp)
> >>>>
> >>>> kfree(intel_connector->detect_edid);
> >>>> intel_connector->detect_edid = NULL;
> >>>> + cec_phys_addr_invalidate(intel_dp->aux.cec_adap);
> >>>>
> >>>> intel_dp->has_audio = false;
> >>>> }
> >>>> @@ -4616,13 +4620,17 @@ intel_dp_long_pulse(struct intel_connector *intel_connector)
> >>>> intel_display_power_get(to_i915(dev), intel_dp->aux_power_domain);
> >>>>
> >>>> /* Can't disconnect eDP, but you can close the lid... */
> >>>> - if (is_edp(intel_dp))
> >>>> + if (is_edp(intel_dp)) {
> >>>> status = edp_detect(intel_dp);
> >>>> - else if (intel_digital_port_connected(to_i915(dev),
> >>>> - dp_to_dig_port(intel_dp)))
> >>>> + } else if (intel_digital_port_connected(to_i915(dev),
> >>>> + dp_to_dig_port(intel_dp))) {
> >>>> status = intel_dp_detect_dpcd(intel_dp);
> >>>> - else
> >>>> + if (status == connector_status_connected)
> >>>> + drm_dp_cec_configure_adapter(&intel_dp->aux,
> >>>> + intel_dp->aux.name, dev->dev);
> >>>
> >>> This is cluttering up the code a bit. Maybe do this call somewhere
> >>> around the intel_dp_configure_mst() call instead since that seems to be
> >>> the place where we start to do changes to externally visible state.
> >>>
> >>> Actually, do we want to register cec adapters for MST devices?
> >>>
> >>> And shouldn't we call this regardless of the connector state so that
> >>> the cec adapter gets unregistered when the device is disconnected?
> >>
> >> This hasn't (AFAIK) anything to do with MST. This is in a branch device (i.e.
> >> a DP to HDMI adapter).
> >
> > You are now potentiall registering the CEC adapter to the immediately
> > upstream MST device (ie. the one that we talk to over the normal AUX stuff),
> > but kms will consider that paticular connector as disconnected, and
> > instead only sinks downstream of that device may have connected connectors
> > associated with them. Presumably the CEC towards that device goes
> > nowhere, and instead we'd have to talk to the remote branch devices
> > somewhere downstream.
> >
> > Thus my question whether we want to potentially register the CEC adapter
> > to the immediately upstream MST device or not. I would imagine not, and
> > thus the call should perhaps be moved past the 'is_mst? -> disconnected'
> > checks.
>
> Ah, now I see what you mean. Sorry, I misunderstood you earlier. I can
> certainly move it down. But an MST device would never set the CEC capability
> in the DPCD, would it?
It might. You don't even have to drive it as an MST device if you don't
want to, or you source device might not even be capable of MST in which
case you have not choice but to drive it in SST mode.
> That makes no sense. So it would never register a
> CEC device in practice. Although I do need to test what happens when you
> first connect a USB-C to HDMI adapter that supports CEC, then disconnect it,
> then connect an MST hub. The CEC device should be unregistered in that case,
> but I'm not sure if that actually happens. I'll have to test that tomorrow.
>
> >
> >>
> >> The CEC adapter should ideally be associated with the branch device (since that
> >> is what implements the CEC tunneling): i.e. when you connect the adapter, then
> >> the CEC device is created, when you disconnect the adapter, then the CEC device
> >> should be unregistered. This is not the same as connecting/disconnecting the
> >> HDMI cable to/from the adapter: that just sets or invalidates the CEC physical
> >> address (which is read from the EDID).
> >>
> >> However, I have not seen any code that tells me when the adapter is plugged in
> >> or is unplugged. So all I have to go on is when the HDMI cable is connected.
> >>
> >> Note that the 'late_register' you mentioned in your 1/3 review isn't called when
> >> connecting the adapter. So that too cannot be used as a trigger to detect if
> >> this protocol is supported.
> >
> > Like I said, you should do the registration directly if the connector
> > has already been registered, otherwise defer to .late_register().
>
> So intel_dp_long_pulse() can be called even if the connector hasn't been fully
> registered yet? Just making sure I really understand this.
Yeah, we enable hotplug processing slightly before registering
everything.
>
> >
> >>
> >> I know doing this here is not ideal, but I have not found another way and I am
> >> not even certain if it is possible at all, it might be intrinsic to how DP works.
> >> I do not consider myself a DP expert, though.
> >>
> >> So this is how it works now, e.g. on my Intel NUC and a USB-C to HDMI adapter:
> >>
> >> 1) I connect the adapter (no HDMI connected yet): nothing happens.
> >> 2) I connect the HDMI cable to the adapter: drm_dp_cec_configure_adapter is called and
> >> it detects the CEC capability in the DPCD. It now creates the CEC device.
> >> 3) I disconnect the HDMI cable: the physical address of the CEC device is invalidated,
> >> but the CEC device is not removed.
> >> 4) I disconnect the adapter: nothing happens.
> >> 5) I reconnect the adapter: nothing happens.
> >> 6) I reconnect the HDMI cable: drm_dp_cec_configure_adapter is called and it checks the
> >> DPCD. If the capabilities are unchanged, then it will continue to use the registered
> >> CEC device. If the capabilities have changed (i.e. the adapter is apparently a
> >> different one), then the CEC device is unregistered and a new one is created, provided
> >> that the new adapter supports CEC, of course.
> >>
> >> The bottom-line is that I cannot tell the difference between disconnecting the adapter
> >> and disconnecting the HDMI cable to the adapter.
> >>
> >> Another consideration is that CEC applications (i.e. HTPCs) do not expect the CEC device
> >> to disappear when you disconnect either the HDMI cable or the adapter. Even though the
> >> CEC device is strictly speaking associated with the adapter, from the point of view of
> >> the user there is no difference between disconnecting an HDMI cable from an adapter, or
> >> disconnecting the adapter itself. So there is a good argument to be made to only unregister
> >> the CEC device when a different (or no) adapter is detected the next time the HPD goes high.
> >
> > Should we just register a CEC adapter always then? Seems rather
> > inconsistent to do it only when a CEC capable device gets plugged in but
> > then leave it lingering around when the device gets disconnected.
>
> I thought about that, but that would clutter /dev with lots of non-functioning CEC
> device nodes. CEC only becomes available if you have an HDMI adapter that actually
> supports this. Most do not. And of course if you connect to a display using a DP
> cable (i.e. without an HDMI adapter), then it is obviously not supported either.
>
> Another problem is that if you switch between adapters with different CEC capabilities,
> then there is no way to signal that to userspace in the CEC API, and I don't think
> that would be a good idea at all anyway.
>
> Having experimented with this for some time now I found that this is a good compromise
> that fits how this is used in practice.
But what you have means userspace will still have to be prepared for the
adapter to disappear at any point, so I don't really see what we gain by
deferring the unregistration.
--
Ville Syrjälä
Intel OTC
