drm/msm/gpu: Fix crash on devices without devfreq support

From: Rob Clark <robdclark@chromium.org>

Avoid going down devfreq paths on devices where devfreq is not
initialized.

Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
Reported-by: Anders Roxell <anders.roxell@linaro.org>
Signed-off-by: Rob Clark <robdclark@chromium.org>
---
 drivers/gpu/drm/msm/msm_gpu_devfreq.c | 31 +++++++++++++++++++++------
 1 file changed, 25 insertions(+), 6 deletions(-)

On Sat, Feb 19, 2022 at 10:32 AM Rob Clark <robdclark@gmail.com> wrote:
>
> From: Rob Clark <robdclark@chromium.org>
>
> Avoid going down devfreq paths on devices where devfreq is not
> initialized.
>
> Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
> Reported-by: Anders Roxell <anders.roxell@linaro.org>

forgot to add:

Fixes: 6aa89ae1fb04 ("drm/msm/gpu: Cancel idle/boost work on suspend")

> Signed-off-by: Rob Clark <robdclark@chromium.org>
> ---
>  drivers/gpu/drm/msm/msm_gpu_devfreq.c | 31 +++++++++++++++++++++------
>  1 file changed, 25 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/msm/msm_gpu_devfreq.c b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> index 9bf319be11f6..26a3669a97b3 100644
> --- a/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> +++ b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> @@ -83,12 +83,17 @@ static struct devfreq_dev_profile msm_devfreq_profile = {
>  static void msm_devfreq_boost_work(struct kthread_work *work);
>  static void msm_devfreq_idle_work(struct kthread_work *work);
>
> +static bool has_devfreq(struct msm_gpu *gpu)
> +{
> +       return !!gpu->funcs->gpu_busy;
> +}
> +
>  void msm_devfreq_init(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
>         /* We need target support to do devfreq */
> -       if (!gpu->funcs->gpu_busy)
> +       if (!has_devfreq(gpu))
>                 return;
>
>         dev_pm_qos_add_request(&gpu->pdev->dev, &df->idle_freq,
> @@ -149,6 +154,9 @@ void msm_devfreq_cleanup(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> +       if (!has_devfreq(gpu))
> +               return;
> +
>         devfreq_cooling_unregister(gpu->cooling);
>         dev_pm_qos_remove_request(&df->boost_freq);
>         dev_pm_qos_remove_request(&df->idle_freq);
> @@ -156,16 +164,24 @@ void msm_devfreq_cleanup(struct msm_gpu *gpu)
>
>  void msm_devfreq_resume(struct msm_gpu *gpu)
>  {
> -       gpu->devfreq.busy_cycles = 0;
> -       gpu->devfreq.time = ktime_get();
> +       struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> -       devfreq_resume_device(gpu->devfreq.devfreq);
> +       if (!has_devfreq(gpu))
> +               return;
> +
> +       df->busy_cycles = 0;
> +       df->time = ktime_get();
> +
> +       devfreq_resume_device(df->devfreq);
>  }
>
>  void msm_devfreq_suspend(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> +       if (!has_devfreq(gpu))
> +               return;
> +
>         devfreq_suspend_device(df->devfreq);
>
>         cancel_idle_work(df);
> @@ -185,6 +201,9 @@ void msm_devfreq_boost(struct msm_gpu *gpu, unsigned factor)
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>         uint64_t freq;
>
> +       if (!has_devfreq(gpu))
> +               return;
> +
>         freq = get_freq(gpu);
>         freq *= factor;
>
> @@ -207,7 +226,7 @@ void msm_devfreq_active(struct msm_gpu *gpu)
>         struct devfreq_dev_status status;
>         unsigned int idle_time;
>
> -       if (!df->devfreq)
> +       if (!has_devfreq(gpu))
>                 return;
>
>         /*
> @@ -253,7 +272,7 @@ void msm_devfreq_idle(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> -       if (!df->devfreq)
> +       if (!has_devfreq(gpu))
>                 return;
>
>         msm_hrtimer_queue_work(&df->idle_work, ms_to_ktime(1),
> --
> 2.34.1
>

On Wed, 23 Feb 2022 at 18:46, Rob Clark <robdclark@gmail.com> wrote:
>
> On Tue, Feb 22, 2022 at 7:11 PM Dmitry Baryshkov
> <dmitry.baryshkov@linaro.org> wrote:
> >
> > On 19/02/2022 21:33, Rob Clark wrote:
> > > From: Rob Clark <robdclark@chromium.org>
> > >
> > > Avoid going down devfreq paths on devices where devfreq is not
> > > initialized.
> > >
> > > Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
> > > Reported-by: Anders Roxell <anders.roxell@linaro.org>
> > > Signed-off-by: Rob Clark <robdclark@chromium.org>
> > > ---
> > >   drivers/gpu/drm/msm/msm_gpu_devfreq.c | 31 +++++++++++++++++++++------
> > >   1 file changed, 25 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/msm/msm_gpu_devfreq.c b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> > > index 9bf319be11f6..26a3669a97b3 100644
> > > --- a/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> > > +++ b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> > > @@ -83,12 +83,17 @@ static struct devfreq_dev_profile msm_devfreq_profile = {
> > >   static void msm_devfreq_boost_work(struct kthread_work *work);
> > >   static void msm_devfreq_idle_work(struct kthread_work *work);
> > >
> > > +static bool has_devfreq(struct msm_gpu *gpu)
> > > +{
> > > +     return !!gpu->funcs->gpu_busy;
> >
> > I see that devfreq init will be skipped if gpu_busy is NULL.
> > Can we use gpu->devfreq instead of this condition?
>
> We could, but then we couldn't also use the same has_devfreq() helper
> in msm_devfreq_init().  I thought it was clearer to use the same
> helper everywhere.

Well... It is not clear at first glance how gpu_busy is related to
devfreq. On the other hand, if gpu->devfreq is NULL, it's obvious that
devfreq is not initialized.

I'd propose to use if (gpu->funcs->gpu_busy) to check if we can init
devfreq and after that to check (gpu->devfreq) as a way to know
whether the devfreq is available.

>
> > I noticed that you have replaced some of gpu->devfreq checks with
> > has_devreq() calls. Is there any difference?
>
> It amounts to the same thing because if you don't have gpu_busy, then
> devfreq is never initialized.  I just thought it clearer to use the
> same check in all places.

See my comment above.

>
> BR,
> -R
>
> > > +}
> > > +
> > >   void msm_devfreq_init(struct msm_gpu *gpu)
> > >   {
> > >       struct msm_gpu_devfreq *df = &gpu->devfreq;
> > >
> > >       /* We need target support to do devfreq */
> > > -     if (!gpu->funcs->gpu_busy)
> > > +     if (!has_devfreq(gpu))
> > >               return;
> > >
> > >       dev_pm_qos_add_request(&gpu->pdev->dev, &df->idle_freq,
> > > @@ -149,6 +154,9 @@ void msm_devfreq_cleanup(struct msm_gpu *gpu)
> > >   {
> > >       struct msm_gpu_devfreq *df = &gpu->devfreq;
> > >
> > > +     if (!has_devfreq(gpu))
> > > +             return;
> > > +
> > >       devfreq_cooling_unregister(gpu->cooling);
> > >       dev_pm_qos_remove_request(&df->boost_freq);
> > >       dev_pm_qos_remove_request(&df->idle_freq);
> > > @@ -156,16 +164,24 @@ void msm_devfreq_cleanup(struct msm_gpu *gpu)
> > >
> > >   void msm_devfreq_resume(struct msm_gpu *gpu)
> > >   {
> > > -     gpu->devfreq.busy_cycles = 0;
> > > -     gpu->devfreq.time = ktime_get();
> > > +     struct msm_gpu_devfreq *df = &gpu->devfreq;
> > >
> > > -     devfreq_resume_device(gpu->devfreq.devfreq);
> > > +     if (!has_devfreq(gpu))
> > > +             return;
> > > +
> > > +     df->busy_cycles = 0;
> > > +     df->time = ktime_get();
> > > +
> > > +     devfreq_resume_device(df->devfreq);
> > >   }
> > >
> > >   void msm_devfreq_suspend(struct msm_gpu *gpu)
> > >   {
> > >       struct msm_gpu_devfreq *df = &gpu->devfreq;
> > >
> > > +     if (!has_devfreq(gpu))
> > > +             return;
> > > +
> > >       devfreq_suspend_device(df->devfreq);
> > >
> > >       cancel_idle_work(df);
> > > @@ -185,6 +201,9 @@ void msm_devfreq_boost(struct msm_gpu *gpu, unsigned factor)
> > >       struct msm_gpu_devfreq *df = &gpu->devfreq;
> > >       uint64_t freq;
> > >
> > > +     if (!has_devfreq(gpu))
> > > +             return;
> > > +
> > >       freq = get_freq(gpu);
> > >       freq *= factor;
> > >
> > > @@ -207,7 +226,7 @@ void msm_devfreq_active(struct msm_gpu *gpu)
> > >       struct devfreq_dev_status status;
> > >       unsigned int idle_time;
> > >
> > > -     if (!df->devfreq)
> > > +     if (!has_devfreq(gpu))
> > >               return;
> > >
> > >       /*
> > > @@ -253,7 +272,7 @@ void msm_devfreq_idle(struct msm_gpu *gpu)
> > >   {
> > >       struct msm_gpu_devfreq *df = &gpu->devfreq;
> > >
> > > -     if (!df->devfreq)
> > > +     if (!has_devfreq(gpu))
> > >               return;
> > >
> > >       msm_hrtimer_queue_work(&df->idle_work, ms_to_ktime(1),
> >
> >
> > --
> > With best wishes
> > Dmitry

Hi Rob,

On Sun, 20 Feb 2022 at 00:02, Rob Clark <robdclark@gmail.com> wrote:
>
> From: Rob Clark <robdclark@chromium.org>
>
> Avoid going down devfreq paths on devices where devfreq is not
> initialized.
>
> Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
> Reported-by: Anders Roxell <anders.roxell@linaro.org>
> Signed-off-by: Rob Clark <robdclark@chromium.org>

I have tested this patch and the reported kernel crash is fixed [1].

Tested-by: Linux Kernel Functional Testing <lkft@linaro.org>

> ---
>  drivers/gpu/drm/msm/msm_gpu_devfreq.c | 31 +++++++++++++++++++++------
>  1 file changed, 25 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/msm/msm_gpu_devfreq.c b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> index 9bf319be11f6..26a3669a97b3 100644
> --- a/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> +++ b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
> @@ -83,12 +83,17 @@ static struct devfreq_dev_profile msm_devfreq_profile = {
>  static void msm_devfreq_boost_work(struct kthread_work *work);
>  static void msm_devfreq_idle_work(struct kthread_work *work);
>
> +static bool has_devfreq(struct msm_gpu *gpu)
> +{
> +       return !!gpu->funcs->gpu_busy;
> +}
> +
>  void msm_devfreq_init(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
>         /* We need target support to do devfreq */
> -       if (!gpu->funcs->gpu_busy)
> +       if (!has_devfreq(gpu))
>                 return;
>
>         dev_pm_qos_add_request(&gpu->pdev->dev, &df->idle_freq,
> @@ -149,6 +154,9 @@ void msm_devfreq_cleanup(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> +       if (!has_devfreq(gpu))
> +               return;
> +
>         devfreq_cooling_unregister(gpu->cooling);
>         dev_pm_qos_remove_request(&df->boost_freq);
>         dev_pm_qos_remove_request(&df->idle_freq);
> @@ -156,16 +164,24 @@ void msm_devfreq_cleanup(struct msm_gpu *gpu)
>
>  void msm_devfreq_resume(struct msm_gpu *gpu)
>  {
> -       gpu->devfreq.busy_cycles = 0;
> -       gpu->devfreq.time = ktime_get();
> +       struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> -       devfreq_resume_device(gpu->devfreq.devfreq);
> +       if (!has_devfreq(gpu))
> +               return;
> +
> +       df->busy_cycles = 0;
> +       df->time = ktime_get();
> +
> +       devfreq_resume_device(df->devfreq);
>  }
>
>  void msm_devfreq_suspend(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> +       if (!has_devfreq(gpu))
> +               return;
> +
>         devfreq_suspend_device(df->devfreq);
>
>         cancel_idle_work(df);
> @@ -185,6 +201,9 @@ void msm_devfreq_boost(struct msm_gpu *gpu, unsigned factor)
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>         uint64_t freq;
>
> +       if (!has_devfreq(gpu))
> +               return;
> +
>         freq = get_freq(gpu);
>         freq *= factor;
>
> @@ -207,7 +226,7 @@ void msm_devfreq_active(struct msm_gpu *gpu)
>         struct devfreq_dev_status status;
>         unsigned int idle_time;
>
> -       if (!df->devfreq)
> +       if (!has_devfreq(gpu))
>                 return;
>
>         /*
> @@ -253,7 +272,7 @@ void msm_devfreq_idle(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_devfreq *df = &gpu->devfreq;
>
> -       if (!df->devfreq)
> +       if (!has_devfreq(gpu))
>                 return;
>
>         msm_hrtimer_queue_work(&df->idle_work, ms_to_ktime(1),
> --
> 2.34.1


--
Linaro LKFT
https://lkft.linaro.org

[1] https://lkft.validation.linaro.org/scheduler/job/4664600#L1894