Re: amd64 DRMKMS Intel regression - partial success (workaround)

[Robert Swindells <rjs%fdy2.co.uk@localhost>]

Piotr Meyer wrote:
>Regardless, it would be nice to have working EDID - is there any
>debug or test that I can do?

What screen resolution do you get for the console before starting X ?

I think there may be two related bugs here.

Intel GPUs have two ways of driving the i2c bus for each
connector. There is a hardware i2c engine or you can toggle the GPIO
pins in software.

The drmkms driver tries the hardware engine first then falls back to
the software method. The driver in -6 just drove the GPIO pins from
userspace.

I have two systems with intel GPUs. One can read the EDID when running
-current, the other can't. Both worked fine with -6 and with -current
before 20140616.

If I boot a kernel on the working system that has been modified to only
use software i2c then it is unable to read the EDID from the monitor.

I think that the usage of the hardware i2c engine is less reliable
than it was before the last import and that there also seems to be a
problem with the software i2c fallback on my hardware.

The only other thing that changed in the drmkms drivers at around the
same time was to move the EDID probe from after interrupts have been
initialized to before. This changes slightly how the driver uses the
hardware i2c engine.

I see you have raised a PR for this.

Robert Swindells