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>

drm/v3d: Plug dma_fence leak

The irq_fence and done_fence are given a reference that is never
released. The necessary dma_fence_put()s seem to have been
deleted in error in an earlier commit.

Fixes: 0b73676836b2 ("drm/v3d: Clock V3D down when not in use.")

Signed-off-by: Phil Elwell <phil@raspberrypi.org>

v3d_drv: Handle missing clock more gracefully

Signed-off-by: popcornmix <popcornmix@gmail.com>

v3d_gem: Kick the clock so firmware knows we are using firmware clock interface

Setting the v3d clock to low value allows firmware to handle dvfs in case
where v3d hardware is not being actively used (e.g. console use).

Signed-off-by: popcornmix <popcornmix@gmail.com>

drivers/gpu/drm/v3d/v3d_drv.c

@@ -295,6 +295,21 @@ static int v3d_platform_drm_probe(struct platform_device *pdev)
 		}
 	}
 
+	v3d->clk = devm_clk_get(dev, NULL);
+	if (IS_ERR_OR_NULL(v3d->clk)) {
+		if (PTR_ERR(v3d->clk) != -EPROBE_DEFER)
+			dev_err(dev, "Failed to get clock (%ld)\n", PTR_ERR(v3d->clk));
+		return PTR_ERR(v3d->clk);
+	}
+	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)
@@ -323,6 +338,8 @@ static int v3d_platform_drm_probe(struct platform_device *pdev)
 	ret = v3d_sysfs_init(dev);
 	if (ret)
 		goto drm_unregister;
+	ret = clk_set_rate(v3d->clk, v3d->clk_down_rate);
+	WARN_ON_ONCE(ret != 0);
 
 	return 0;
 

drivers/gpu/drm/v3d/v3d_drv.h

@@ -95,6 +95,12 @@ 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. */
@@ -565,3 +571,4 @@ int v3d_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
 /* v3d_sysfs.c */
 int v3d_sysfs_init(struct device *dev);
 void v3d_sysfs_destroy(struct device *dev);
+void v3d_submit_init(struct drm_device *dev);

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/reset.h>
@@ -268,6 +269,8 @@ v3d_gem_init(struct drm_device *dev)
 	if (ret)
 		return ret;
 
+	v3d_submit_init(dev);
+
 	/* 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".

drivers/gpu/drm/v3d/v3d_submit.c

@@ -5,11 +5,52 @@
  */
 
 #include <drm/drm_syncobj.h>
+#include <linux/clk.h>
 
 #include "v3d_drv.h"
 #include "v3d_regs.h"
 #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);
+}
+
 /* Takes the reservation lock on all the BOs being referenced, so that
  * at queue submit time we can update the reservations.
  *
@@ -87,6 +128,7 @@ static void
 v3d_job_free(struct kref *ref)
 {
 	struct v3d_job *job = container_of(ref, struct v3d_job, refcount);
+	struct v3d_dev *v3d = job->v3d;
 	int i;
 
 	if (job->bo) {
@@ -98,6 +140,8 @@ v3d_job_free(struct kref *ref)
 	dma_fence_put(job->irq_fence);
 	dma_fence_put(job->done_fence);
 
+	v3d_clock_up_put(v3d);
+
 	if (job->perfmon)
 		v3d_perfmon_put(job->perfmon);
 
@@ -199,6 +243,7 @@ v3d_job_init(struct v3d_dev *v3d, struct drm_file *file_priv,
 			goto fail_deps;
 	}
 
+	v3d_clock_up_get(v3d);
 	kref_init(&job->refcount);
 
 	return 0;
@@ -1339,3 +1384,14 @@ fail:
 
 	return ret;
 }
+
+void v3d_submit_init(struct drm_device *dev) {
+	struct v3d_dev *v3d = to_v3d_dev(dev);
+
+	mutex_init(&v3d->clk_lock);
+	INIT_DELAYED_WORK(&v3d->clk_down_work, v3d_clock_down_work);
+
+	/* kick the clock so firmware knows we are using firmware clock interface */
+	v3d_clock_up_get(v3d);
+	v3d_clock_up_put(v3d);
+}