Re: [PATCH] drm/panfrost: fix ref count leak in panfrost_job_hw_submit

[Steven Price]

On 09/07/2020 16:44, Rob Herring wrote:
> On Sun, Jun 14, 2020 at 12:27 AM Navid Emamdoost
> <navid.emamdoost@xxxxxxxxx> wrote:
> > in panfrost_job_hw_submit, pm_runtime_get_sync is called which
> > increments the counter even in case of failure, leading to incorrect
> > ref count. In case of failure, decrement the ref count before returning.
> >
> > Signed-off-by: Navid Emamdoost <navid.emamdoost@xxxxxxxxx>
> > ---
> >   drivers/gpu/drm/panfrost/panfrost_job.c | 8 +++++---
> >   1 file changed, 5 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
> > index 7914b1570841..89ac84667eb1 100644
> > --- a/drivers/gpu/drm/panfrost/panfrost_job.c
> > +++ b/drivers/gpu/drm/panfrost/panfrost_job.c
> > @@ -147,11 +147,10 @@ static void panfrost_job_hw_submit(struct panfrost_job *job, int js)
> >
> >          ret = pm_runtime_get_sync(pfdev->dev);
> >          if (ret < 0)
> > -               return;
> > +               goto out;
>
> If the get failed, I don't think we want to do a put.

The pm_runtime_get_sync() does actually always increment the reference 
(even when returning a failure), but actually I don't think we want it 
here anyway, as I think I explained before[1].
 [1] 
https://lore.kernel.org/dri-devel/272650ba-2c44-9084-7829-b04023eba723@xxxxxxx
> >          if (WARN_ON(job_read(pfdev, JS_COMMAND_NEXT(js)))) {
> > -               pm_runtime_put_sync_autosuspend(pfdev->dev);

This is a correct change - we don't want the put here. (Although this 
should never happen).

> > -               return;
> > +               goto out;
> >          }
> >
> >          cfg = panfrost_mmu_as_get(pfdev, &job->file_priv->mmu);
> > @@ -184,6 +183,9 @@ static void panfrost_job_hw_submit(struct panfrost_job *job, int js)
> >                                  job, js, jc_head);
> >
> >          job_write(pfdev, JS_COMMAND_NEXT(js), JS_COMMAND_START);
>
> So we start the job here and then...
>
> > +out:
> > +       pm_runtime_put_sync_autosuspend(pfdev->dev);
>
> ...turn off clocks/power here. Typically, you'd be fine as autosuspend
> has a delay by default, but userspace is free to change the delay to
> 0.
>
> > +       return;

A return at the end of the function with no argument is pointless.

Steve

> >   }
> >
> >   static void panfrost_acquire_object_fences(struct drm_gem_object **bos,
> > --
> > 2.17.1