[intel_gart] Correct ValleyView, Skylake, Broadwell bridge IDs

[waddlesplash]

The comments in the agp_gart bus driver indicate that the ValleyView, Skylake, and Broadwell (mostly missing) drivers have "guessed" bridge IDs, which are probably incorrect.

A more complete list of IDs appears to be found at ​https://xref.landonf.org/source/xref/linux-master/arch/x86/events/intel/uncore_snb.c#6 but I don't know how to map these to the specific devices. korli or KapiX, how is that done?

[pulkomandy]

The list of devices is at ​https://cgit.haiku-os.org/haiku/tree/src/add-ons/kernel/busses/agp_gart/intel_gart.cpp#n197

For each one you have to know:

• Which "model" it is
• If it's mobile or not

You can get the model from te Linux constants names. Sometimes you can learn something about a device from its name in the PCI ID database (​https://pci-ids.ucw.cz/read/PC/8086).

Until now I think devices have been added on-demand when people submit a bugreport and manage to include a full listdev output (and not just the ID of the graphics card itself...).

The "mobile" bit will be used mainly to enable the LVDS panel control. If we improve the code that gets info about the panel from the VBT and make it fully reliable, we could probably disable that. I guess Linux does not have this distinction at all, and makes the decision fully depending on the VBT already?

[korli]

I searched in the ​https://linux-hardware.org/ database and pick the relevant bridge ids for each gpu id. Theorically using the bridge with ID 0 near a supported GPU device should bring the same PCI device.

The 'mobile or not' field is ATM "guessed" from the computer type but I saw it's possible to have one time a GPU in a notebook and another time the same GPU in a NUC-type computer. This probably should be replaced with something else from the VBT (is a eDP port available or such checks).

[korli]

Comet Lake added in hrev56034