Some Intel 4th generation GPUs are being parsed as OTHER

[Ara Pulido]

Example of a submission where this is happening:

https://certification.canonical.com/hardware/201410-15951/submission/101027/

[Daniel Manrique (roadmr)]

OK, We've seen this kind of problem before so I'll be succint with the explanation, pretty much copying it verbatim from udevadm.py:

                # Not all DISPLAY devices are display adapters. The ones with
                # subclass OTHER are usually uninteresting devices. As an
                # exception, some vendors have recently begun to use the
                # 0x80 (Pci.CLASS_DISPLAY_OTHER) subclass identifier. In order
                # to correctly identify them special heuristics are needed,
                # these are encapsulated in the known_to_be_video_device method
                # (further below).

The Intel integrated adapter in this system is such a case, as further explained in the method mentioned here:

    # devices. This method encapsulates heuristics to decide if a device is a
    # valid video adapter, based on product/vendor and pci class/subclass
    # information.
    if vendor_id == Pci.VENDOR_ID_INTEL:
        # Intel recently (2014) started using subclass OTHER erratically, some
        # older GPUs have subdevices with OTHER which are uninteresting. If
        # Intel, we only consider OTHER devices as VIDEO if they are in this
        # explicit list

Note the above comment about intel. Unfortunately, with them we have no way of finding all offending devices: I did try to identify all Intel DISPLAY-class devices as VIDEO but that regressed some device detections for older systems (thanks Sylvain for the test suite). So we will need to continue adding to the explicit list mentioned there.

I added the PCI IDs for this device and things are working correctly now.

Merge request coming up in a second.

[Daniel Manrique (roadmr)]

I uploaded the original offending submission and the device in question now shows up as VIDEO. So the fix as proposed in hexr/certification staging is good to go. Once this is in production we can find, then reparse borked submissions, which can be done by hitting the API.

[Daniel Manrique (roadmr)]

https://certification.staging.canonical.com/hardware/201112-10259/submission/96859/ shows this device.

[Daniel Manrique (roadmr)]

Interestingly, I only needed to reparse one submission to update the device's category, then all other affected systems got fixed magically.

So, all records in production are reparsed and OK now.