[Intel-gfx] [RFC] Page table sharing and bufferless execbuf

[Jesse Barnes]

I've been carrying something looking rougly like this patchset around
internally for a long time now, and with SKL out there now, I figured
it's time to get it posted and start the process of integration.

David is working on pulling over most of the "driver based PASID
handling" and other code into the IOMMU layer, but the integration with
the rest of the driver needs some careful thought (handling task
referencing along with context referencing, context creation, etc.)

I believe John H has a better version of the sync point patch already;
I'm only posting this one because it was already in my tree.  I can
re-base to the new one once it lands.

A few things to think about:
  - how to signal GPU hangs with the new execbuf (a signal might be more
    natural as the execution appears more CPU-like?  what state do we
    have to worry about restoring for bufferless contexts?
  - page fault re-queue needs to be handled properly for bufferless
    contexts that take a fault and need to stall
  - the new context creation ioctl is needed for something else too I
    believe?

Anyway, comments and feedback welcome.  I have some tests and libdrm
code I'll push to my private fdo repos soon as well.

Thanks,
Jesse