2019-01-23 11:15:19

by Jayant Shekhar

[permalink] [raw]
Subject: [v1] drm/msm: Remove clock and bandwidth votes in mdss pm suspend

MDSS PM suspend is dependent on runtime suspend for disabling
clocks and removing bandwidth votes. In case of pm_suspend
triggered, dpm_prepare hold a refcount on power usage of device
and hence runtime suspend is never triggered during pm_suspend.
As runtime suspend is not triggered, clocks and bandwidth votes
remain. Hence explicitly trigger mdss disable in msm_pm_suspend
to disable clocks and remove the votes.

Signed-off-by: Jayant Shekhar <[email protected]>
---
drivers/gpu/drm/msm/msm_drv.c | 8 ++++++++
1 file changed, 8 insertions(+)

diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index 5c60bb3..ffe3a25 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -1068,12 +1068,16 @@ static int msm_pm_suspend(struct device *dev)
{
struct drm_device *ddev = dev_get_drvdata(dev);
struct msm_drm_private *priv = ddev->dev_private;
+ struct msm_mdss *mdss = priv->mdss;

if (!IS_ERR_OR_NULL(priv->pm_state))
return 0;

priv->pm_state = drm_atomic_helper_suspend(ddev);

+ if (mdss && mdss->funcs)
+ mdss->funcs->disable(mdss);
+
return IS_ERR(priv->pm_state) ? PTR_ERR(priv->pm_state) : 0;
}

@@ -1081,11 +1085,15 @@ static int msm_pm_resume(struct device *dev)
{
struct drm_device *ddev = dev_get_drvdata(dev);
struct msm_drm_private *priv = ddev->dev_private;
+ struct msm_mdss *mdss = priv->mdss;
int ret;

if (IS_ERR_OR_NULL(priv->pm_state))
return 0;

+ if (mdss && mdss->funcs)
+ mdss->funcs->enable(mdss);
+
ret = drm_atomic_helper_resume(ddev, priv->pm_state);
if (ret == 0)
priv->pm_state = NULL;
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project



2019-02-17 18:06:26

by Rob Clark

[permalink] [raw]
Subject: Re: [v1] drm/msm: Remove clock and bandwidth votes in mdss pm suspend

On Wed, Jan 23, 2019 at 6:12 AM Jayant Shekhar <[email protected]> wrote:
>
> MDSS PM suspend is dependent on runtime suspend for disabling
> clocks and removing bandwidth votes. In case of pm_suspend
> triggered, dpm_prepare hold a refcount on power usage of device
> and hence runtime suspend is never triggered during pm_suspend.
> As runtime suspend is not triggered, clocks and bandwidth votes
> remain. Hence explicitly trigger mdss disable in msm_pm_suspend
> to disable clocks and remove the votes.
>
> Signed-off-by: Jayant Shekhar <[email protected]>
> ---
> drivers/gpu/drm/msm/msm_drv.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
> index 5c60bb3..ffe3a25 100644
> --- a/drivers/gpu/drm/msm/msm_drv.c
> +++ b/drivers/gpu/drm/msm/msm_drv.c
> @@ -1068,12 +1068,16 @@ static int msm_pm_suspend(struct device *dev)
> {
> struct drm_device *ddev = dev_get_drvdata(dev);
> struct msm_drm_private *priv = ddev->dev_private;
> + struct msm_mdss *mdss = priv->mdss;
>
> if (!IS_ERR_OR_NULL(priv->pm_state))
> return 0;
>
> priv->pm_state = drm_atomic_helper_suspend(ddev);
>
> + if (mdss && mdss->funcs)
> + mdss->funcs->disable(mdss);
> +

This patch doesn't actually seem to apply, I guess it has a dependency
on some other missing patch?

That said, shouldn't the refcount aquired in _prepare() be dropped
somewhere? It seems like if all planes/crtcs/etc are disabled, we
should also disable power (or at least as much as we can while still
having hotplug work), so ideally drm_atomic_helper_suspend() should be
enough.

BR.
-R

> return IS_ERR(priv->pm_state) ? PTR_ERR(priv->pm_state) : 0;
> }
>
> @@ -1081,11 +1085,15 @@ static int msm_pm_resume(struct device *dev)
> {
> struct drm_device *ddev = dev_get_drvdata(dev);
> struct msm_drm_private *priv = ddev->dev_private;
> + struct msm_mdss *mdss = priv->mdss;
> int ret;
>
> if (IS_ERR_OR_NULL(priv->pm_state))
> return 0;
>
> + if (mdss && mdss->funcs)
> + mdss->funcs->enable(mdss);
> +
> ret = drm_atomic_helper_resume(ddev, priv->pm_state);
> if (ret == 0)
> priv->pm_state = NULL;
> --
> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
> a Linux Foundation Collaborative Project
>