[Intel-gfx] [PATCH 0/2] RFC: Potential improvements in edid detection timings (v2)

[Eugeni Dodonov]

(Resending with small improvements - also sending to dri-devel for feedback).

This is the the forth iteration of potential fixes for slow edid detection
issues over non-existent outputs
(https://bugs.freedesktop.org/show_bug.cgi?id=41059) - the previous versions
were posted to the bug and were used mostly for debugging the problem.

The problem I attempted to fix here is that some systems would take a very
long time trying to locate all the possible video outputs - in in cases
where such outputs were bogus, this lead to timing out after all the
possible delays were done.

After investigation, I came to think on two different ways to fix the issue:
 - The first patch does a one-line fix in drm_edid.c. I added a check for the
   return value of i2c_transfer - and, if it is -ENXIO, we give up on further
   attempts as the bus is not there. A drawback to this approach is that it
   affects all the devices out there which use drm_get_edid.  From my testing,
   the -ENXIO gave no false positives, but I haven't tested it on non-Intel
   cards. So I'd like to hear what other drm developers think about that.

 - The second patch does a similar procedure within the i915 driver, in case
   the proposed change to drm_edid.c won't be adequate for other drivers. It
   adds a new function - intel_drm_get_valid_edid - which attempts to do a
   simple i2c transfer over the bus prior to calling drm_get_edid. In case such
   transfer fails with -ENXIO, it is a signal that the bus is not there, so we
   shouldn't waste any time trying to communicate with it further.

Note that those patches provide lots of dmesg pollution - for the final
version those printk'ss should probably be removed, but I left them
intentionally with KERN_DEBUG in order to see when the adapters come and go
during the debugging and testing.

According to testing performed at
https://bugs.freedesktop.org/show_bug.cgi?id=41059, the following results
were obtained with those patches:

System1:
 (testing all possible outputs)
 ubuntu 3.0.0-12.19:       840ms
 3.0.0-12.19 +  drm patch: 290ms
 3.0.0-12.19 + i915 patch: 290ms
 (manually ignoring phantom outputs)
 ubuntu 3.0.0-12.19:       690ms
 3.0.0-12.19 +  drm patch: 140ms
 3.0.0-12.19 + i915 patch: 140ms

System2:
 (testing all possible outputs)
 ubuntu 3.0.0-12.19:       315ms
 3.0.0-12.19 +  drm patch: 280ms
 3.0.0-12.19 + i915 patch: 280ms
 (manually ignoring phantom outputs)
 ubuntu 3.0.0-12.19:       175ms
 3.0.0-12.19 +  drm patch: 140ms
 3.0.0-12.19 + i915 patch: 140ms



Eugeni Dodonov (2):
  Give up on edid retries when i2c tells us that bus is not there
  Check if the bus is valid prior to discovering edid.

 drivers/gpu/drm/drm_edid.c         |    5 +++++
 drivers/gpu/drm/i915/intel_dp.c    |    4 ++--
 drivers/gpu/drm/i915/intel_drv.h   |    2 ++
 drivers/gpu/drm/i915/intel_hdmi.c  |    4 ++--
 drivers/gpu/drm/i915/intel_i2c.c   |   33 +++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/intel_lvds.c  |    2 +-
 drivers/gpu/drm/i915/intel_modes.c |    2 +-
 drivers/gpu/drm/i915/intel_sdvo.c  |    4 ++--
 8 files changed, 48 insertions(+), 8 deletions(-)

-- 
1.7.6.3