From ca8579839f0ebf0ffe73d1135284363b2155e712 Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Thu, 2 May 2019 13:22:53 -0700
Subject: [PATCH] drm/v3d: Clock V3D down when not in use.

My various attempts at re-enabling runtime PM have failed, so just
crank the clock down when V3D is idle to reduce power consumption.

Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/gpu/drm/v3d/v3d_drv.c | 18 ++++++++++++
 drivers/gpu/drm/v3d/v3d_drv.h |  6 ++++
 drivers/gpu/drm/v3d/v3d_gem.c | 53 +++++++++++++++++++++++++++++++----
 3 files changed, 72 insertions(+), 5 deletions(-)

--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -282,6 +282,21 @@ static int v3d_platform_drm_probe(struct
 		}
 	}
 
+	v3d->clk = devm_clk_get(dev, NULL);
+	if (IS_ERR(v3d->clk)) {
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "Failed to get clock\n");
+		goto dev_free;
+	}
+	v3d->clk_up_rate = clk_get_rate(v3d->clk);
+	/* For downclocking, drop it to the minimum frequency we can get from
+	 * the CPRMAN clock generator dividing off our parent.  The divider is
+	 * 4 bits, but ask for just higher than that so that rounding doesn't
+	 * make cprman reject our rate.
+	 */
+	v3d->clk_down_rate =
+		(clk_get_rate(clk_get_parent(v3d->clk)) / (1 << 4)) + 10000;
+
 	if (v3d->ver < 41) {
 		ret = map_regs(v3d, &v3d->gca_regs, "gca");
 		if (ret)
@@ -316,6 +331,9 @@ static int v3d_platform_drm_probe(struct
 	if (ret)
 		goto irq_disable;
 
+	ret = clk_set_rate(v3d->clk, v3d->clk_down_rate);
+	WARN_ON_ONCE(ret != 0);
+
 	return 0;
 
 irq_disable:
--- a/drivers/gpu/drm/v3d/v3d_drv.h
+++ b/drivers/gpu/drm/v3d/v3d_drv.h
@@ -54,6 +54,12 @@ struct v3d_dev {
 	void __iomem *bridge_regs;
 	void __iomem *gca_regs;
 	struct clk *clk;
+	struct delayed_work clk_down_work;
+	unsigned long clk_up_rate, clk_down_rate;
+	struct mutex clk_lock;
+	u32 clk_refcount;
+	bool clk_up;
+
 	struct reset_control *reset;
 
 	/* Virtual and DMA addresses of the single shared page table. */
--- a/drivers/gpu/drm/v3d/v3d_gem.c
+++ b/drivers/gpu/drm/v3d/v3d_gem.c
@@ -4,6 +4,7 @@
 #include <linux/device.h>
 #include <linux/dma-mapping.h>
 #include <linux/io.h>
+#include <linux/clk.h>
 #include <linux/module.h>
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
@@ -19,6 +20,47 @@
 #include "v3d_trace.h"
 
 static void
+v3d_clock_down_work(struct work_struct *work)
+{
+	struct v3d_dev *v3d =
+		container_of(work, struct v3d_dev, clk_down_work.work);
+	int ret;
+
+	ret = clk_set_rate(v3d->clk, v3d->clk_down_rate);
+	v3d->clk_up = false;
+	WARN_ON_ONCE(ret != 0);
+}
+
+static void
+v3d_clock_up_get(struct v3d_dev *v3d)
+{
+	mutex_lock(&v3d->clk_lock);
+	if (v3d->clk_refcount++ == 0) {
+		cancel_delayed_work_sync(&v3d->clk_down_work);
+		if (!v3d->clk_up)  {
+			int ret;
+
+			ret = clk_set_rate(v3d->clk, v3d->clk_up_rate);
+			WARN_ON_ONCE(ret != 0);
+			v3d->clk_up = true;
+		}
+	}
+	mutex_unlock(&v3d->clk_lock);
+}
+
+static void
+v3d_clock_up_put(struct v3d_dev *v3d)
+{
+	mutex_lock(&v3d->clk_lock);
+	if (--v3d->clk_refcount == 0) {
+		schedule_delayed_work(&v3d->clk_down_work,
+				      msecs_to_jiffies(100));
+	}
+	mutex_unlock(&v3d->clk_lock);
+}
+
+
+static void
 v3d_init_core(struct v3d_dev *v3d, int core)
 {
 	/* Set OVRTMUOUT, which means that the texture sampler uniform
@@ -354,6 +396,7 @@ v3d_job_free(struct kref *ref)
 	struct v3d_job *job = container_of(ref, struct v3d_job, refcount);
 	unsigned long index;
 	struct dma_fence *fence;
+	struct v3d_dev *v3d = job->v3d;
 	int i;
 
 	for (i = 0; i < job->bo_count; i++) {
@@ -367,11 +410,7 @@ v3d_job_free(struct kref *ref)
 	}
 	xa_destroy(&job->deps);
 
-	dma_fence_put(job->irq_fence);
-	dma_fence_put(job->done_fence);
-
-	pm_runtime_mark_last_busy(job->v3d->dev);
-	pm_runtime_put_autosuspend(job->v3d->dev);
+	v3d_clock_up_put(v3d);
 
 	kfree(job);
 }
@@ -453,6 +492,7 @@ v3d_job_init(struct v3d_dev *v3d, struct
 	if (ret)
 		goto fail;
 
+	v3d_clock_up_get(v3d);
 	kref_init(&job->refcount);
 
 	return 0;
@@ -841,6 +881,9 @@ v3d_gem_init(struct drm_device *dev)
 	mutex_init(&v3d->sched_lock);
 	mutex_init(&v3d->cache_clean_lock);
 
+	mutex_init(&v3d->clk_lock);
+	INIT_DELAYED_WORK(&v3d->clk_down_work, v3d_clock_down_work);
+
 	/* Note: We don't allocate address 0.  Various bits of HW
 	 * treat 0 as special, such as the occlusion query counters
 	 * where 0 means "disabled".