2015-07-03 08:03:57

by Pan Xinhui

Subject: [PATCH] acpi-cpufreq: use same acpi_cpufreq_data if policy is shared

cpufreq core will mark all cpus in policy->cpus as initialized
by setting percpu cpufreq_cpu_data = policy.
However in acpi-cpufreq, we did not set percpu acpi_cpufreq_data for
all cpus in policy->cpus.
That will cause a issue that when we call get_cur_freq_on_cpu for a
cpu, it returns zero instead of a correct value.

So we shared acpi_cpufreq_data for all those cpus in policy->cpus.

Signed-off-by: xinhuix.pan <[email protected]>
---
drivers/cpufreq/acpi-cpufreq.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
index 0136dfc..6c6dc97 100644
--- a/drivers/cpufreq/acpi-cpufreq.c
+++ b/drivers/cpufreq/acpi-cpufreq.c
@@ -673,6 +673,7 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
}

data->acpi_data = per_cpu_ptr(acpi_perf_data, cpu);
+ cpumask_copy(policy->cpus, cpumask_of(cpu));
per_cpu(acfreq_data, cpu) = data;

if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
@@ -712,6 +713,10 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
}
#endif

+ for_each_cpu(i, policy->cpus) {
+ per_cpu(acfreq_data, i) = data;
+ }
+
/* capability check */
if (perf->state_count <= 1) {
pr_debug("No P-States\n");
@@ -843,7 +848,9 @@ err_free_mask:
free_cpumask_var(data->freqdomain_cpus);
err_free:
kfree(data);
- per_cpu(acfreq_data, cpu) = NULL;
+ for_each_cpu(i, policy->cpus) {
+ per_cpu(acfreq_data, i) = NULL;
+ }

return result;
}
@@ -851,11 +858,14 @@ err_free:
static int acpi_cpufreq_cpu_exit(struct cpufreq_policy *policy)
{
struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
+ int i;

pr_debug("acpi_cpufreq_cpu_exit\n");

if (data) {
- per_cpu(acfreq_data, policy->cpu) = NULL;
+ for_each_cpu(i, policy->cpus) {
+ per_cpu(acfreq_data, i) = NULL;
+ }
acpi_processor_unregister_performance(data->acpi_data,
policy->cpu);
free_cpumask_var(data->freqdomain_cpus);
--
1.9.1

2015-07-03 08:47:10

by Viresh Kumar

Subject: Re: [PATCH] acpi-cpufreq: use same acpi_cpufreq_data if policy is shared

On 03-07-15, 16:01, Pan Xinhui wrote:
>
> cpufreq core will mark all cpus in policy->cpus as initialized
> by setting percpu cpufreq_cpu_data = policy.
> However in acpi-cpufreq, we did not set percpu acpi_cpufreq_data for
> all cpus in policy->cpus.
> That will cause a issue that when we call get_cur_freq_on_cpu for a
> cpu, it returns zero instead of a correct value.
>
> So we shared acpi_cpufreq_data for all those cpus in policy->cpus.
>
> Signed-off-by: xinhuix.pan <[email protected]>
> ---
> drivers/cpufreq/acpi-cpufreq.c | 14 ++++++++++++--
> 1 file changed, 12 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
> index 0136dfc..6c6dc97 100644
> --- a/drivers/cpufreq/acpi-cpufreq.c
> +++ b/drivers/cpufreq/acpi-cpufreq.c
> @@ -673,6 +673,7 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
> }
>
> data->acpi_data = per_cpu_ptr(acpi_perf_data, cpu);
> + cpumask_copy(policy->cpus, cpumask_of(cpu));

This is already done by cpufreq-core before calling ->init() and so
isn't required.

> per_cpu(acfreq_data, cpu) = data;
>
> if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
> @@ -712,6 +713,10 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
> }
> #endif
>
> + for_each_cpu(i, policy->cpus) {
> + per_cpu(acfreq_data, i) = data;
> + }
> +
> /* capability check */
> if (perf->state_count <= 1) {
> pr_debug("No P-States\n");
> @@ -843,7 +848,9 @@ err_free_mask:
> free_cpumask_var(data->freqdomain_cpus);
> err_free:
> kfree(data);
> - per_cpu(acfreq_data, cpu) = NULL;
> + for_each_cpu(i, policy->cpus) {
> + per_cpu(acfreq_data, i) = NULL;
> + }

The per-cpu thing is basically wrong to start with. cpufreq-policy has
a field: 'driver_data', which can be used to replace this acfreq_data.
Probably use that and things will get fixed by themselves.

--
viresh