Re: [PATCH] media: stm32-dcmi: Set minimum cpufreq requirement

[Vincent Guittot]

On Wed, 3 Jun 2020 at 09:34, Benjamin GAIGNARD <benjamin.gaignard@xxxxxx> wrote:
>
>
>
> On 6/2/20 3:35 PM, Valentin Schneider wrote:
> > On 02/06/20 12:37, Benjamin GAIGNARD wrote:
> >> On 6/2/20 11:31 AM, Valentin Schneider wrote:
> >>>> @@ -99,6 +100,8 @@ enum state {
> >>>>
> >>>>    #define OVERRUN_ERROR_THRESHOLD 3
> >>>>
> >>>> +#define DCMI_MIN_FREQ     650000 /* in KHz */
> >>>> +
> >>> This assumes the handling part is guaranteed to always run on the same CPU
> >>> with the same performance profile (regardless of the platform). If that's
> >>> not guaranteed, it feels like you'd want this to be configurable in some
> >>> way.
> >> Yes I could add a st,stm32-dcmi-min-frequency (in KHz) parameter the
> >> device tree node.
> >>
> > Something like that - I'm not sure how well this fits with the DT
> > landscape, as you could argue it isn't really a description of the
> > hardware, more of a description of the performance expectations of the
> > software. I won't really argue here.
> >
> >>>>    struct dcmi_graph_entity {
> >>>>         struct v4l2_async_subdev asd;
> >>>>
> >>> [...]
> >>>> @@ -2020,6 +2042,8 @@ static int dcmi_probe(struct platform_device *pdev)
> >>>>                 goto err_cleanup;
> >>>>         }
> >>>>
> >>>> +  dcmi->policy = cpufreq_cpu_get(0);
> >>>> +
> >>> Ideally you'd want to fetch the policy of the CPU your IRQ (and handling
> >>> thread) is affined to; The only compatible DTS I found describes a single
> >>> A7, which is somewhat limited in the affinity area...
> >> If I move this code just before start streaming and use get_cpu(), would
> >> it works ?
> >>
> > AFAIA streaming_start() is not necessarily executing on the same CPU as the
> > one that will handle the interrupt. I was thinking you could use the IRQ's
> > effective affinity as a hint of which CPU(s) to boost, i.e. something like:
> >
> > ---
> >      struct cpumask_var_t visited;
> >      struct irq_data *d = irq_get_irq_data(irq);
> >
> >      err = alloc_cpumask_var(visited, GFP_KERNEL);
> >      /* ... */
> >      for_each_cpu(cpu, irq_data_get_effective_affinity_mask(d)) {
> >              /* check if not already spanned */
> >              if (cpumask_test_cpu(cpu, visited))
> >                      continue;
> >
> >              policy = cpufreq_cpu_get(cpu);
> >              cpumask_or(visited, visited, policy->cpus);
> >              /* do the boost for that policy here */
> >              /* ... */
> >              cpufreq_cpu_put(policy);
> >      }
> > ---
> >
> > That of course falls apart when hotplug gets involved, and the effective
> > affinity changes... There's irq_set_affinity_notifier() out there, but it
> > seems it's only about the affinity, not the effective_affinity, I'm not
> > sure how valid it would be to query the effective_affinity in that
> > notifier.
> If I wait to be in the irq it will be too late so I think I will do a
> loop over all possible CPUs
> before start the streaming to change the policies.

Can't you use irq_get_affinity_mask  and loop over it ?

Also You should better use freq_qos_add/remove_request during probe
and remove of the driver and use freq_qos_update_request in
dcmi_start/stop_streaming to set/unset your constraint.

>
> >
> >> Benjamin
> >>>>         dev_info(&pdev->dev, "Probe done\n");
> >>>>
> >>>>         platform_set_drvdata(pdev, dcmi);
> >>>> @@ -2049,6 +2073,9 @@ static int dcmi_remove(struct platform_device *pdev)
> >>>>
> >>>>         pm_runtime_disable(&pdev->dev);
> >>>>
> >>>> +  if (dcmi->policy)
> >>>> +          cpufreq_cpu_put(dcmi->policy);
> >>>> +
> >>>>         v4l2_async_notifier_unregister(&dcmi->notifier);
> >>>>         v4l2_async_notifier_cleanup(&dcmi->notifier);
> >>>>         media_entity_cleanup(&dcmi->vdev->entity);