On 30/09/2019 17:34, Kevin Hilman wrote:

Neil Armstrong narmstrong@baylibre.com writes:

...

Hi Kevin,

On 25/09/2019 21:31, Kevin Hilman wrote:

...

From: Kevin Hilman khilman@baylibre.com

On some SoCs, the VPU is in a power-domain and needs runtime PM enabled and used in order to keep the power domain on.

Signed-off-by: Kevin Hilman khilman@baylibre.com

drivers/gpu/drm/meson/meson_drv.c | 4 ++++ 1 file changed, 4 insertions(+)

diff --git a/drivers/gpu/drm/meson/meson_drv.c b/drivers/gpu/drm/meson/meson_drv.c index 2310c96fff46..256b6a0e9c6b 100644 --- a/drivers/gpu/drm/meson/meson_drv.c +++ b/drivers/gpu/drm/meson/meson_drv.c @@ -14,6 +14,7 @@ #include <linux/platform_device.h> #include <linux/component.h> #include <linux/of_graph.h> +#include <linux/pm_runtime.h>

#include <drm/drmP.h> #include <drm/drm_atomic.h> @@ -274,6 +275,7 @@ static int meson_drv_bind_master(struct device *dev, bool has_components)

/* Hardware Initialization */

• pm_runtime_get_sync(dev); meson_vpu_init(priv); meson_venc_init(priv); meson_vpp_init(priv);

@@ -416,6 +418,7 @@ static int meson_drv_probe(struct platform_device *pdev) struct device_node *ep, *remote; int count = 0;

• pm_runtime_enable(&pdev->dev); for_each_endpoint_of_node(np, ep) { remote = of_graph_get_remote_port_parent(ep); if (!remote || !of_device_is_available(remote)) {

@@ -440,6 +443,7 @@ static int meson_drv_probe(struct platform_device *pdev) }

/* If no output endpoints were available, simply bail out */

• pm_runtime_disable(&pdev->dev); return 0;

};

I'll rather implement true runtime PM instead,

While this is a minimum implementation, can you explain what you mean by "true" runtime PM?

Implementing runtime_sync()/_put when the display is used/unused.

ARM did a good job in their malidp driver.

Neil

Kevin