[Intel-gfx] [PATCH v2 4/4] drm/i915/bxt: Tune L3 SQC credits based on actual latencies

[Imre Deak]

BSpec says we need to fine tune these values, so comply. I checked this
with random GPU benchmarks and it does seem to improve things.

Note that I considered to program this from the ring as part of the
context specific workarounds there, I decided against that for the
following reasons:
- It's not a context specific setting, it's part of whatever (power-)
  context the GPU manages regardless of context scheduling to
  save/restore things across power transitions. So it's enough to
  program it once.
- Atm, we don't apply workarounds for engines other than the render
  engine from the ring (although this could be added if needed).
- The same setting is programmed via MMIO for BDW/CHV/VLV and it
  makes sense to program it the same way on BXT too.

v2:
- Specify the actual WA we're implementing. (Ville)

CC: Chris Wilson <chris at chris-wilson.co.uk>
CC: Mika Kuoppala <mika.kuoppala at intel.com>
CC: Ville Syrjälä <ville.syrjala at linux.intel.com>
Signed-off-by: Imre Deak <imre.deak at intel.com>
---
 drivers/gpu/drm/i915/intel_pm.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index b217c44..6853bf8 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -76,6 +76,17 @@ static void bxt_init_clock_gating(struct drm_device *dev)
 	if (IS_BXT_REVID(dev_priv, BXT_REVID_B0, REVID_FOREVER))
 		I915_WRITE(GEN9_CLKGATE_DIS_0, I915_READ(GEN9_CLKGATE_DIS_0) |
 			   PWM1_GATING_DIS | PWM2_GATING_DIS);
+
+	/*
+	 * WaProgramL3SqcReg1DefaultForPerf:bxt
+	 * Note that for dynamic reprogramming we'd need to do a stalling flush
+	 * operation, but we can do away with that here, since the GPU is idle
+	 * at this point.
+	 */
+	if (IS_BXT_REVID(dev_priv, BXT_REVID_A1, REVID_FOREVER))
+		I915_WRITE(GEN8_L3SQCREG1,
+			   L3_GENERAL_PRIO_CREDITS(62) |
+			   L3_HIGH_PRIO_CREDITS(2));
 }
 
 static void i915_pineview_get_mem_freq(struct drm_device *dev)
-- 
2.5.0