[RFC,v2,1/1] drm/tegra: sor: Fix hang on tegra124 due to NULL clk_out

Message ID 0f776b7500ee0e74b316b9803803b309779d2ff7.1513768618.git.guillaume.tucker@collabora.com
State New
Headers show
Series
  • Fix drm driver hang on tegra124-nyan-big
Related show

Commit Message

Guillaume Tucker Dec. 20, 2017, 11:32 a.m.
When neither HDMI nor DP is supported such as on the tegra124, the
sor->clk_out is not initialised and remains NULL.  In this case, the
parent clock can't be assigned to it so revert to the previous
behaviour of assigning it to the main sor->clk instead.

This fixes a kernel hang on tegra124 and should also affect tegra210
as they both don't support HDMI and DP.  Tested on tegra124 only.

Fixes: e1335e2f0cfc ("drm/tegra: sor: Reimplement pad clock")
Signed-off-by: Guillaume Tucker <guillaume.tucker@collabora.com>
CC: Thierry Reding <treding@nvidia.com>
---
 drivers/gpu/drm/tegra/sor.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Thierry Reding Dec. 20, 2017, 6:15 p.m. | #1
On Wed, Dec 20, 2017 at 11:32:23AM +0000, Guillaume Tucker wrote:
> When neither HDMI nor DP is supported such as on the tegra124, the
> sor->clk_out is not initialised and remains NULL.  In this case, the
> parent clock can't be assigned to it so revert to the previous
> behaviour of assigning it to the main sor->clk instead.
> 
> This fixes a kernel hang on tegra124 and should also affect tegra210
> as they both don't support HDMI and DP.  Tested on tegra124 only.
> 
> Fixes: e1335e2f0cfc ("drm/tegra: sor: Reimplement pad clock")
> Signed-off-by: Guillaume Tucker <guillaume.tucker@collabora.com>
> CC: Thierry Reding <treding@nvidia.com>
> ---
>  drivers/gpu/drm/tegra/sor.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

How about just the below instead? It's one more line than your patch,
but it will automatically handle all occurrences of clk_out properly.

--- >8 ---
diff --git a/drivers/gpu/drm/tegra/sor.c b/drivers/gpu/drm/tegra/sor.c
index f6313c4d612e..4be9edf9c6fe 100644
--- a/drivers/gpu/drm/tegra/sor.c
+++ b/drivers/gpu/drm/tegra/sor.c
@@ -3047,6 +3047,8 @@ static int tegra_sor_probe(struct platform_device *pdev)
                                name, err);
                        goto remove;
                }
+       } else {
+               sor->clk_out = sor->clk;
        }
 
        sor->clk_parent = devm_clk_get(&pdev->dev, "parent");
--- >8 ---

That said, I suspect the SOR might be compatible from a clock point of
view with later versions and perhaps we just didn't implement clocks
correctly back in the Tegra124 timeframe.

Maybe Peter knows.

Thierry
Jon Hunter Jan. 2, 2018, 6:32 p.m. | #2
On 20/12/17 18:15, Thierry Reding wrote:
> On Wed, Dec 20, 2017 at 11:32:23AM +0000, Guillaume Tucker wrote:
>> When neither HDMI nor DP is supported such as on the tegra124, the
>> sor->clk_out is not initialised and remains NULL.  In this case, the
>> parent clock can't be assigned to it so revert to the previous
>> behaviour of assigning it to the main sor->clk instead.
>>
>> This fixes a kernel hang on tegra124 and should also affect tegra210
>> as they both don't support HDMI and DP.  Tested on tegra124 only.
>>
>> Fixes: e1335e2f0cfc ("drm/tegra: sor: Reimplement pad clock")
>> Signed-off-by: Guillaume Tucker <guillaume.tucker@collabora.com>
>> CC: Thierry Reding <treding@nvidia.com>
>> ---
>>  drivers/gpu/drm/tegra/sor.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> How about just the below instead? It's one more line than your patch,
> but it will automatically handle all occurrences of clk_out properly.
> 
> --- >8 ---
> diff --git a/drivers/gpu/drm/tegra/sor.c b/drivers/gpu/drm/tegra/sor.c
> index f6313c4d612e..4be9edf9c6fe 100644
> --- a/drivers/gpu/drm/tegra/sor.c
> +++ b/drivers/gpu/drm/tegra/sor.c
> @@ -3047,6 +3047,8 @@ static int tegra_sor_probe(struct platform_device *pdev)
>                                 name, err);
>                         goto remove;
>                 }
> +       } else {
> +               sor->clk_out = sor->clk;
>         }
>  
>         sor->clk_parent = devm_clk_get(&pdev->dev, "parent");
> --- >8 ---
> 
> That said, I suspect the SOR might be compatible from a clock point of
> view with later versions and perhaps we just didn't implement clocks
> correctly back in the Tegra124 timeframe.
> 
> Maybe Peter knows.

So the above change from Thierry works for me and we need this for v4.15
(otherwise nyan-big does not boot) so you can have my ...

Tested-by: Jon Hunter <jonathanh@nvidia.com>

However, would be good to have Peter's ACK, especially seeing that
Tegra210 sor0 does not support HDMI and DP. So we need to make sure this
is correct for Tegra210 as well (although I have not seen any
regressions for Tegra210).

Cheers
Jon
Thierry Reding Jan. 10, 2018, 12:13 p.m. | #3
On Tue, Jan 02, 2018 at 06:32:11PM +0000, Jon Hunter wrote:
> 
> On 20/12/17 18:15, Thierry Reding wrote:
> > On Wed, Dec 20, 2017 at 11:32:23AM +0000, Guillaume Tucker wrote:
> >> When neither HDMI nor DP is supported such as on the tegra124, the
> >> sor->clk_out is not initialised and remains NULL.  In this case, the
> >> parent clock can't be assigned to it so revert to the previous
> >> behaviour of assigning it to the main sor->clk instead.
> >>
> >> This fixes a kernel hang on tegra124 and should also affect tegra210
> >> as they both don't support HDMI and DP.  Tested on tegra124 only.
> >>
> >> Fixes: e1335e2f0cfc ("drm/tegra: sor: Reimplement pad clock")
> >> Signed-off-by: Guillaume Tucker <guillaume.tucker@collabora.com>
> >> CC: Thierry Reding <treding@nvidia.com>
> >> ---
> >>  drivers/gpu/drm/tegra/sor.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > How about just the below instead? It's one more line than your patch,
> > but it will automatically handle all occurrences of clk_out properly.
> > 
> > --- >8 ---
> > diff --git a/drivers/gpu/drm/tegra/sor.c b/drivers/gpu/drm/tegra/sor.c
> > index f6313c4d612e..4be9edf9c6fe 100644
> > --- a/drivers/gpu/drm/tegra/sor.c
> > +++ b/drivers/gpu/drm/tegra/sor.c
> > @@ -3047,6 +3047,8 @@ static int tegra_sor_probe(struct platform_device *pdev)
> >                                 name, err);
> >                         goto remove;
> >                 }
> > +       } else {
> > +               sor->clk_out = sor->clk;
> >         }
> >  
> >         sor->clk_parent = devm_clk_get(&pdev->dev, "parent");
> > --- >8 ---
> > 
> > That said, I suspect the SOR might be compatible from a clock point of
> > view with later versions and perhaps we just didn't implement clocks
> > correctly back in the Tegra124 timeframe.
> > 
> > Maybe Peter knows.
> 
> So the above change from Thierry works for me and we need this for v4.15
> (otherwise nyan-big does not boot) so you can have my ...
> 
> Tested-by: Jon Hunter <jonathanh@nvidia.com>
> 
> However, would be good to have Peter's ACK, especially seeing that
> Tegra210 sor0 does not support HDMI and DP. So we need to make sure this
> is correct for Tegra210 as well (although I have not seen any
> regressions for Tegra210).

There are no Tegra210 boards with eDP or LVDS support, which are the
only setups that are impacted by this.

I've sent out the above as a proper patch. I'll submit this for a late
fixes pull request.

Thierry

Patch

diff --git a/drivers/gpu/drm/tegra/sor.c b/drivers/gpu/drm/tegra/sor.c
index b0a1dedac802..a344f66d4ecb 100644
--- a/drivers/gpu/drm/tegra/sor.c
+++ b/drivers/gpu/drm/tegra/sor.c
@@ -255,7 +255,7 @@  static int tegra_sor_set_parent_clock(struct tegra_sor *sor, struct clk *parent)
 
 	clk_disable_unprepare(sor->clk);
 
-	err = clk_set_parent(sor->clk_out, parent);
+	err = clk_set_parent(sor->clk_out ? sor->clk_out : sor->clk, parent);
 	if (err < 0)
 		return err;