drm/i915/gt: Manual rc6 entry upon parking

Now that we rapidly park the GT when the GPU idles, we often find
ourselves idling faster than the RC6 promotion timer. Thus if we tell
the GPU to enter RC6 manually as we park, we can do so quicker (by
around 50ms, half an EI on average) and marginally increase our
powersaving across all execlists platforms.

v2: Now with a selftest to check we can enter RC6 manually
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Andi Shyti <andi.shyti@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Acked-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191127095657.3209854-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gt/intel_gt_pm.c

@@ -64,6 +64,7 @@ static int __gt_unpark(struct intel_wakeref *wf)
 	if (NEEDS_RC6_CTX_CORRUPTION_WA(i915))
 		intel_uncore_forcewake_get(gt->uncore, FORCEWAKE_ALL);
 
+	intel_rc6_unpark(&gt->rc6);
 	intel_rps_unpark(&gt->rps);
 	i915_pmu_gt_unparked(i915);
 
@@ -85,6 +86,7 @@ static int __gt_park(struct intel_wakeref *wf)
 	i915_vma_parked(gt);
 	i915_pmu_gt_parked(i915);
 	intel_rps_park(&gt->rps);
+	intel_rc6_park(&gt->rc6);
 
 	/* Everything switched off, flush any residual interrupt just in case */
 	intel_synchronize_irq(i915);

drivers/gpu/drm/i915/gt/intel_rc6.c

@@ -96,10 +96,10 @@ static void gen11_rc6_enable(struct intel_rc6 *rc6)
 	set(uncore, GEN9_RENDER_PG_IDLE_HYSTERESIS, 60);
 
 	/* 3a: Enable RC6 */
-	set(uncore, GEN6_RC_CONTROL,
-	    GEN6_RC_CTL_HW_ENABLE |
-	    GEN6_RC_CTL_RC6_ENABLE |
-	    GEN6_RC_CTL_EI_MODE(1));
+	rc6->ctl_enable =
+		GEN6_RC_CTL_HW_ENABLE |
+		GEN6_RC_CTL_RC6_ENABLE |
+		GEN6_RC_CTL_EI_MODE(1);
 
 	set(uncore, GEN9_PG_ENABLE,
 	    GEN9_RENDER_PG_ENABLE |
@@ -170,10 +170,10 @@ static void gen9_rc6_enable(struct intel_rc6 *rc6)
 	else
 		rc6_mode = GEN6_RC_CTL_EI_MODE(1);
 
-	set(uncore, GEN6_RC_CONTROL,
-	    GEN6_RC_CTL_HW_ENABLE |
-	    GEN6_RC_CTL_RC6_ENABLE |
-	    rc6_mode);
+	rc6->ctl_enable =
+		GEN6_RC_CTL_HW_ENABLE |
+		GEN6_RC_CTL_RC6_ENABLE |
+		rc6_mode;
 
 	/*
 	 * WaRsDisableCoarsePowerGating:skl,cnl
@@ -200,10 +200,10 @@ static void gen8_rc6_enable(struct intel_rc6 *rc6)
 	set(uncore, GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */
 
 	/* 3: Enable RC6 */
-	set(uncore, GEN6_RC_CONTROL,
+	rc6->ctl_enable =
 	    GEN6_RC_CTL_HW_ENABLE |
 	    GEN7_RC_CTL_TO_MODE |
-	    GEN6_RC_CTL_RC6_ENABLE);
+	    GEN6_RC_CTL_RC6_ENABLE;
 }
 
 static void gen6_rc6_enable(struct intel_rc6 *rc6)
@@ -239,10 +239,10 @@ static void gen6_rc6_enable(struct intel_rc6 *rc6)
 		rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
 	if (HAS_RC6pp(i915))
 		rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
-	set(uncore, GEN6_RC_CONTROL,
+	rc6->ctl_enable =
 	    rc6_mask |
 	    GEN6_RC_CTL_EI_MODE(1) |
-	    GEN6_RC_CTL_HW_ENABLE);
+	    GEN6_RC_CTL_HW_ENABLE;
 
 	rc6vids = 0;
 	ret = sandybridge_pcode_read(i915, GEN6_PCODE_READ_RC6VIDS,
@@ -360,7 +360,7 @@ static void chv_rc6_enable(struct intel_rc6 *rc6)
 			       VLV_RENDER_RC6_COUNT_EN));
 
 	/* 3: Enable RC6 */
-	set(uncore, GEN6_RC_CONTROL, GEN7_RC_CTL_TO_MODE);
+	rc6->ctl_enable = GEN7_RC_CTL_TO_MODE;
 }
 
 static void vlv_rc6_enable(struct intel_rc6 *rc6)
@@ -386,8 +386,8 @@ static void vlv_rc6_enable(struct intel_rc6 *rc6)
 			       VLV_MEDIA_RC6_COUNT_EN |
 			       VLV_RENDER_RC6_COUNT_EN));
 
-	set(uncore, GEN6_RC_CONTROL,
-	    GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL);
+	rc6->ctl_enable =
+	    GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
 }
 
 static bool bxt_check_bios_rc6_setup(struct intel_rc6 *rc6)
@@ -634,6 +634,29 @@ void intel_rc6_enable(struct intel_rc6 *rc6)
 	rc6->enabled = true;
 }
 
+void intel_rc6_unpark(struct intel_rc6 *rc6)
+{
+	struct intel_uncore *uncore = rc6_to_uncore(rc6);
+
+	if (!rc6->enabled)
+		return;
+
+	/* Restore HW timers for automatic RC6 entry while busy */
+	set(uncore, GEN6_RC_CONTROL, rc6->ctl_enable);
+}
+
+void intel_rc6_park(struct intel_rc6 *rc6)
+{
+	struct intel_uncore *uncore = rc6_to_uncore(rc6);
+
+	if (!rc6->enabled)
+		return;
+
+	/* Turn off the HW timers and go directly to rc6 */
+	set(uncore, GEN6_RC_CONTROL, GEN6_RC_CTL_RC6_ENABLE);
+	set(uncore, GEN6_RC_STATE, 0x4 << RC_SW_TARGET_STATE_SHIFT);
+}
+
 void intel_rc6_disable(struct intel_rc6 *rc6)
 {
 	if (!rc6->enabled)

drivers/gpu/drm/i915/gt/intel_rc6.h

@@ -15,6 +15,9 @@ struct intel_rc6;
 void intel_rc6_init(struct intel_rc6 *rc6);
 void intel_rc6_fini(struct intel_rc6 *rc6);
 
+void intel_rc6_unpark(struct intel_rc6 *rc6);
+void intel_rc6_park(struct intel_rc6 *rc6);
+
 void intel_rc6_sanitize(struct intel_rc6 *rc6);
 void intel_rc6_enable(struct intel_rc6 *rc6);
 void intel_rc6_disable(struct intel_rc6 *rc6);

drivers/gpu/drm/i915/gt/intel_rc6_types.h

@@ -18,6 +18,8 @@ struct intel_rc6 {
 	u64 prev_hw_residency[4];
 	u64 cur_residency[4];
 
+	u32 ctl_enable;
+
 	struct drm_i915_gem_object *pctx;
 
 	bool supported : 1;

drivers/gpu/drm/i915/gt/selftest_gt_pm.c

@@ -51,6 +51,7 @@ static int live_gt_resume(void *arg)
 int intel_gt_pm_live_selftests(struct drm_i915_private *i915)
 {
 	static const struct i915_subtest tests[] = {
+		SUBTEST(live_rc6_manual),
 		SUBTEST(live_gt_resume),
 	};
 

drivers/gpu/drm/i915/gt/selftest_rc6.c

@@ -12,6 +12,64 @@
 
 #include "selftests/i915_random.h"
 
+int live_rc6_manual(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct intel_rc6 *rc6 = &gt->rc6;
+	intel_wakeref_t wakeref;
+	u64 res[2];
+	int err = 0;
+
+	/*
+	 * Our claim is that we can "encourage" the GPU to enter rc6 at will.
+	 * Let's try it!
+	 */
+
+	if (!rc6->enabled)
+		return 0;
+
+	/* bsw/byt use a PCU and decouple RC6 from our manual control */
+	if (IS_VALLEYVIEW(gt->i915) || IS_CHERRYVIEW(gt->i915))
+		return 0;
+
+	wakeref = intel_runtime_pm_get(gt->uncore->rpm);
+
+	/* Force RC6 off for starters */
+	__intel_rc6_disable(rc6);
+	msleep(1); /* wakeup is not immediate, takes about 100us on icl */
+
+	res[0] = intel_rc6_residency_ns(rc6, GEN6_GT_GFX_RC6);
+	msleep(250);
+	res[1] = intel_rc6_residency_ns(rc6, GEN6_GT_GFX_RC6);
+	if ((res[1] - res[0]) >> 10) {
+		pr_err("RC6 residency increased by %lldus while disabled for 250ms!\n",
+		       (res[1] - res[0]) >> 10);
+		err = -EINVAL;
+		goto out_unlock;
+	}
+
+	/* Manually enter RC6 */
+	intel_rc6_park(rc6);
+
+	res[0] = intel_rc6_residency_ns(rc6, GEN6_GT_GFX_RC6);
+	msleep(100);
+	res[1] = intel_rc6_residency_ns(rc6, GEN6_GT_GFX_RC6);
+
+	if (res[1] == res[0]) {
+		pr_err("Did not enter RC6! RC6_STATE=%08x, RC6_CONTROL=%08x\n",
+		       intel_uncore_read_fw(gt->uncore, GEN6_RC_STATE),
+		       intel_uncore_read_fw(gt->uncore, GEN6_RC_CONTROL));
+		err = -EINVAL;
+	}
+
+	/* Restore what should have been the original state! */
+	intel_rc6_unpark(rc6);
+
+out_unlock:
+	intel_runtime_pm_put(gt->uncore->rpm, wakeref);
+	return err;
+}
+
 static const u32 *__live_rc6_ctx(struct intel_context *ce)
 {
 	struct i915_request *rq;

drivers/gpu/drm/i915/gt/selftest_rc6.h

@@ -8,5 +8,6 @@
 #define SELFTEST_RC6_H
 
 int live_rc6_ctx_wa(void *arg);
+int live_rc6_manual(void *arg);
 
 #endif /* SELFTEST_RC6_H */