[Intel-gfx] DP MST Stability + Upfront Link Training

[Jim Bride]

We've been looking recently at some changes that we can make to allow for
DP MST to be more reliable on i915.  The things that we are looking at
are the following.

* Change intel_dp_mst_mode_valid() to validate modes against available
  link bandwidth rather than maximum link bandwidth.  This is necessary
  so that we can properly report to user space the modes that are supported
  on the link, taking into account link bandwidth that has already been
  allocated to other virtual channels.  This has been a long-standing TODO
  item in the DP MST code base. (Anusha: New version of patches forthcoming.)
* Modify the callbacks between i915 and audio drivers to supply both the
  port and pipe associated with the audio device, which is required for
  DP MST audio support. (DK: Patches submitted)
* Use the settings from upfront link training to configure the link between
  the source device and the primary branch to a known good configuration.
  The current code just assumes that the maximum legal link configuration will
  work.  (Various folks: Upfront patches submitted, MST implementation coming
  soon.)
* Remove some racy behavior around DP MST link training & first time
  configuration scenarios.  In a situation where multiple monitors are
  connected to a DP MST topology at boot time, we want to make sure that
  we are only doing upfront and normal link training once each (assuming that
  the link does not go bad at some point.)  Additionally, we want to make
  sure that the section of code that is only run when
  intel_dp->active_mst_links == 0 in intel_mst_pre_enable_dp() is also run
  only one time (normal link training for the MST link is done in this
  segment of code.)  We plan to fix this by adding a MST-specific mutex to
  intel_dp that will be taken before we check whether upfront link training is
  needed or before we check if we need to do first-time link setup
  (intel_dp->active_mst_links == 0.)  (Jim: Patches coming soon.)

Feedback on any of the above is welcome, as well as any other ideas that folks
may have to help DP MST be more reliable.  We'd also greatly appreciate some
code reviews for the patch sets discussed above.

Thanks,

Anusha, DK, Jim, and Manasi