[Intel-gfx] [PATCH 00/17] drm/i915: Fix vlv/chv panel power sequencer

[ville.syrjala at linux.intel.com]

From: Ville Syrjälä <ville.syrjala at linux.intel.com>

After weeks or months of beating on the hardware I finally managed
to figure out how to kick the vlv/chv power sequencer in a reasonably
light way after changing the pipe<->port mapping.

Contrary to my earlier comments we don't actually need to kick it
when dealing with regular DP ports. I'm not sure if my previous BSW
was just a bit funky or if I just overlooked something because I clearly
remember having to do that. Anyway now it seems enough to kick eDP
ports only and with the refined kick procedure (which doesn't involve
panel power on+off cycles anymore) it should be a reasonably fast
operation too. The power sequencer on one pipe can control any eDP
port even if that pipe otherwise is driving another non-eDP port.

I've tested this this on BYT eDP+DP, BSW eDP+DP, ILK eDP+DP, IVB DP
and HSW eDP and everything appears to working perfectly now. I've
been beating on it for a few days now trying all kinds combinations
of pipes and ports and swapping them around in various ways.

I also simulated a dual eDP system on the BYT and BSW machines by
making the code pretend that the external DP port is also eDP. That
way I could test that the power sequencer code really would work
in dual eDP machines. Obviously I was unable to test that VDD would
actually get enabled for both ports since one of them was DP but
at least the registers pretend that is, and the power sequencer
didn't cause the port to gets stuck at any point, and the AUX stuff
worked every single time on the actual eDP port so at least there
VDD got applied appropriately.

So I'm quite confident that this vlv/chv power seqeuencer problem
is now solved. Well, until someone breaks it that is. I'm not sure
we can do any kind of satisfactory tests for this stuff since it
depends on eDP and some of trickier interaction even requires dual
eDP. Also some of the failures can manifest as something as benign
as failed OUI reads which doesn't even flag an error in dmesg. So
I don't have particularly good ideas on how we can prevent regressions
to this code.

Ville Syrjälä (17):
  drm/i915: Warn if trying to register eDP on port != B/C on vlv/chv
  drm/i915: Warn if stealing power sequencer from an active eDP port
  drm/i915: Remove high level intel_edp_vdd_{on,off}() from hpd/detect
  drm/i915: Store power sequencer delays in intel_dp
  drm/i915: Don't initialize power seqeuencer delays more than once
  drm/i915: Split power sequencer panel on/off functions to locked and
    unlocked variants
  drm/i915: Hold the pps mutex across the whole panel power enable
    sequence
  drm/i915: Wait for PHY port ready before link training on VLV/CHV
  drm/i915: Fix eDP link training when switching pipes on VLV/CHV
  drm/i915: Kick the power sequencer before AUX transactions
  drm/i915: Make sure DPLL is enabled when kicking the power sequencer
    on VLV/CHV
  drm/i915: Don't kick the power seqeuncer just to check if we have
    vdd/panel power
  drm/i915: Clear PPS port select when giving up the power sequencer
  drm/i915: Warn if stealing non pipe A/B power sequencer
  drm/i915: Steal power sequencer in vlv_power_sequencer_pipe()
  drm/i915: Improve VDD/PPS debugs
  drm/i915: Warn if panel power is already on when enabling it

 drivers/gpu/drm/i915/intel_display.c |  84 +++++----
 drivers/gpu/drm/i915/intel_dp.c      | 347 ++++++++++++++++++++++++-----------
 drivers/gpu/drm/i915/intel_drv.h     |  16 ++
 3 files changed, 304 insertions(+), 143 deletions(-)

-- 
2.0.4