Comment 50 for bug 1275794

Revision history for this message
In , Consume-noise-7 (consume-noise-7) wrote :

Created attachment 90151
printk to the rescue

As mentioned in earlier comments I've missed the link training messages in case of the bug, whereas they show up when it works.

So, I've followed the link training path and sprinkled more and more DRM_DEBUG_KMS along the way and finally found a place where adding such a message seem to masquerade the problem. It may not workaround the problem if I decrease the verbosity level - haven't tested.

It's the for(;;) loop in intel_dp_aux_native_write(), in case of ((ack & AUX_NATIVE_REPLY_MASK) == AUX_NATIVE_REPLY_DEFER) it adds an udelay(100). Having a DRM_DEBUG_KMS before the udelay() (as done in this attachment) the loop seems to be throttled enough to masquerade the problem.

With this hack applied plugging in the monitor results in its activation - the console gets cloned to it. Even replugging works.
When plugging in the monitor for the first time the "throttle message" showed up 6k to 7k times, when replugging it there've been roughly 2k messages.

I haven't tried to simply raise the udelay().