drm/i915: CHV DDR DVFS support and another watermark rewrite
authorVille Syrjälä <ville.syrjala@linux.intel.com>
Wed, 24 Jun 2015 19:00:04 +0000 (22:00 +0300)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Mon, 29 Jun 2015 08:55:13 +0000 (10:55 +0200)
commit262cd2e154c29dc3a235f68cc91e13d8f48e8002
treea9fe90af6493f1fb062dec12b63ddd262449055a
parent6eb1a6817246f1a67de4d6959a84d09efead5329
drm/i915: CHV DDR DVFS support and another watermark rewrite

Turns out the VLV/CHV system agent doesn't understand memory
latencies, so trying to rely on the PND deadline mechanism is not
going to fly especially when DDR DVFS is enabled. Currently we try to
avoid the problems by lying to the system agent about the deadlines
and setting the FIFO watermarks to 8 cachelines. This however leads to
bad memory self refresh residency.

So in order to satosfy everyone we'll just give up on the deadline
scheme and program the watermarks old school based on the worst case
memory latency.

I've modelled this a bit on the ILK+ approach where we compute multiple
sets of watermarks for each pipe (PM2,PM5,DDR DVFS) and when merge thet
appropriate one later with the watermarks from other pipes. There isn't
too much to merge actually since each pipe has a totally independent
FIFO (well apart from the mess with the partially shared DSPARB
registers), but still decopuling the pipes from each other seems like a
good idea.

Eventually we'll want to perform the watermark update in two phases
around the plane update to avoid underruns due to the single buffered
watermark registers. But that's still in limbo for ILK+ too, so I've not
gone that far yet for VLV/CHV either.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_drv.h
drivers/gpu/drm/i915/intel_display.c
drivers/gpu/drm/i915/intel_drv.h
drivers/gpu/drm/i915/intel_pm.c
This page took 0.071002 seconds and 5 git commands to generate.