[Intel-gfx] Revised error detection/handling patchset

[Ben Gamari]

Hey everyone,

Here's a new version of the error detection and reporting patchset taking into 
account the changes which were suggested earlier. This drops the code handling
dev_mutex lock contention as this condition really shouldn't be seen if the
rest of the driver honors the locking semantics (although I recall someone
mentioning there was still one locking point in the driver which could cause
issues; this probably ought to be fixed). 

Moreover, we now fire a uevent at the initiation and successful completion of 
the reset procedure. This might be a little excessive so let me know what you
all think.

Anyways, let me know what you think! 

Thanks,
- Ben