On 04/06/20 13:39, Benjamin Gaignard wrote:
> Before start streaming set cpufreq minimum frequency requirement.
> The cpufreq governor will adapt the frequencies and we will have
> no latency for handling interrupts.
> The frequency requirement is retrieved from the device-tree node.
>
> While streaming be notified if the IRQ affinity change thanks to
> irq_affinity_notify callback.
>
> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@xxxxxx>
> ---
> version 3:
> - add a cpumask field to track boosted CPUs
> - add irq_affinity_notify callback
> - protect cpumask field with a mutex
>
> drivers/media/platform/stm32/stm32-dcmi.c | 187 ++++++++++++++++++++++++++++--
> 1 file changed, 179 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/media/platform/stm32/stm32-dcmi.c b/drivers/media/platform/stm32/stm32-dcmi.c
> index b8931490b83b..fb6ab09eaff0 100644
> --- a/drivers/media/platform/stm32/stm32-dcmi.c
> +++ b/drivers/media/platform/stm32/stm32-dcmi.c
> +static void dcmi_irq_notifier_notify(struct irq_affinity_notify *notify,
> + const cpumask_t *mask)
> +{
> + struct stm32_dcmi *dcmi = container_of(notify,
> + struct stm32_dcmi,
> + notify);
> + struct cpufreq_policy *p;
> + int cpu;
> +
> + mutex_lock(&dcmi->freq_lock);
> + /*
> + * For all boosted CPUs check if it is still the case
> + * if not remove the request
> + */
> + for_each_cpu(cpu, dcmi->boosted) {
> + if (cpumask_test_cpu(cpu, mask))
> + continue;
> +
> + p = cpufreq_cpu_get(cpu);
> + if (!p)
> + continue;
> +
> + freq_qos_remove_request(&per_cpu(qos_req, cpu));
> + cpumask_andnot(dcmi->boosted, dcmi->boosted, p->cpus);
> +
> + cpufreq_cpu_put(p);
> + }
> +
> + /*
> + * For CPUs in the mask check if they are boosted if not add
> + * a request
> + */
> + for_each_cpu(cpu, mask) {
> + if (cpumask_test_cpu(cpu, dcmi->boosted))
> + continue;
> +
> + p = cpufreq_cpu_get(cpu);
> + if (!p)
> + continue;
> +
> + freq_qos_add_request(&p->constraints, &per_cpu(qos_req, cpu),
> + FREQ_QOS_MIN, dcmi->min_frequency);
> + cpumask_or(dcmi->boosted, dcmi->boosted, p->cpus);
> + cpufreq_cpu_put(p);
> + }
> +
> + mutex_unlock(&dcmi->freq_lock);
That looks about right.
> +}
> +
> +static void dcmi_irq_notifier_release(struct kref *ref)
> +{
> + /*
> + * This is required by affinity notifier. We don't have anything to
> + * free here.
> + */
> +}
> +
> +static void dcmi_get_cpu_policy(struct stm32_dcmi *dcmi)
> +{
> + struct cpufreq_policy *p;
> + int cpu;
> +
> + if (!alloc_cpumask_var(&dcmi->boosted, GFP_KERNEL))
> + return;
I think you want to actually return i.e. -ENOMEM and do cleanups in the
probe; otherwise you'll silently continue.
> +
> + mutex_lock(&dcmi->freq_lock);
> +
> + for_each_cpu(cpu, irq_get_affinity_mask(dcmi->irq)) {
When I suggested serialization, I was thinking we may want to use the irq's
desc lock to prevent the affinity from moving under our feet. Something
like:
CPU A CPU B
for_each_cpu(cpu, mask)
cpu = cpumask_next(cpu, mask)
// ... cpumask_copy(desc->irq_common_data.affinity, mask)
cpu = cpumask_next(cpu, mask)
Now, should that happen, we would still queue the notifier and run it
shortly after - and since you track which CPUs are boosted, I don't think
we have any loss of information here.
We may have yet another affinity change while the notifier is still queued;
but the notifier boilerplate does grab the desc lock, so I think it's all
good - it wasn't all super obvious so I figured I'd still point it out.
> + if (cpumask_test_cpu(cpu, dcmi->boosted))
> + continue;
> +
> + p = cpufreq_cpu_get(cpu);
> + if (!p)
> + continue;
> +
> + freq_qos_add_request(&p->constraints, &per_cpu(qos_req, cpu),
> + FREQ_QOS_MIN, FREQ_QOS_MIN_DEFAULT_VALUE);
> +
> + cpumask_or(dcmi->boosted, dcmi->boosted, p->cpus);
> +
> + cpufreq_cpu_put(p);
> + }
> +
> + mutex_unlock(&dcmi->freq_lock);
> +}
> +
> +static void dcmi_put_cpu_policy(struct stm32_dcmi *dcmi)
> +{
> + struct cpufreq_policy *p;
> + int cpu;
> +
> + mutex_lock(&dcmi->freq_lock);
> +
> + for_each_cpu(cpu, irq_get_affinity_mask(dcmi->irq)) {
> + if (!cpumask_test_cpu(cpu, dcmi->boosted))
> + continue;
> +
> + p = cpufreq_cpu_get(cpu);
> + if (!p)
> + continue;
> +
> + freq_qos_remove_request(&per_cpu(qos_req, cpu));
> + cpumask_andnot(dcmi->boosted, dcmi->boosted, p->cpus);
> +
> + cpufreq_cpu_put(p);
> + }
> +
> + free_cpumask_var(dcmi->boosted);
> +
> + mutex_unlock(&dcmi->freq_lock);
> +}
> +
> +static void dcmi_set_min_frequency(struct stm32_dcmi *dcmi, s32 freq)
> +{
> + struct irq_affinity_notify *notify = &dcmi->notify;
> + int cpu;
> +
> + mutex_lock(&dcmi->freq_lock);
> +
> + for_each_cpu(cpu, irq_get_affinity_mask(dcmi->irq)) {
> + if (!cpumask_test_cpu(cpu, dcmi->boosted))
> + continue;
> +
If the affinity changed between say .probe() and .start_streaming(), IIUC
you may skip CPUs here - and even if you initialize the notifier earlier in
the function (see below), that won't help you.
I think dcmi_irq_notifier_notify() does almost all you want, if it also did
the QoS update for CPUs that weren't affected by the affinity change, you
may be able to just do:
dcmi_irq_notifier_notify(irq_get_affinity_mask(dcmi->irq));
Or something along those lines.
> + if (!freq_qos_request_active(&per_cpu(qos_req, cpu)))
> + continue;
> +
> + freq_qos_update_request(&per_cpu(qos_req, cpu), freq);
> + }
> +
> + mutex_unlock(&dcmi->freq_lock);
> +
> + if (freq != FREQ_QOS_MIN_DEFAULT_VALUE) {
