2018-03-28 09:33:57

by Shunyong Yang

Subject: [PATCH] cpufreq: cppc_cpufreq: Initialize shared cpu's perf capabilities

When multiple cpus are related in one cpufreq policy, the first online
cpu will be chosen by default to handle cpufreq operations. Let's take
cpu0 and cpu1 as an example.

When cpu0 is offline, policy->cpu will be shifted to cpu1. Cpu1's should
be initialized. Otherwise, perf capabilities are 0s and speed change can
not take effect.

This patch copies perf capabilities of the first online cpu to other
shared cpus when policy shared type is CPUFREQ_SHARED_TYPE_ANY.

Cc: Joey Zheng <[email protected]>
Signed-off-by: Shunyong Yang <[email protected]>
---

The original RFC link,
https://patchwork.kernel.org/patch/10299055/.

This patch solves same issue as RFC above.

Patch name is changed as code is too much different with RFC above.

Remove extra init() per Viresh Kumar's comments and only handle
CPPC CPUFREQ_SHARED_TYPE_ANY case.

---
drivers/cpufreq/cppc_cpufreq.c | 12 +++++++++++-
1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
index 8f7b21a4d537..dc625a93a58e 100644
--- a/drivers/cpufreq/cppc_cpufreq.c
+++ b/drivers/cpufreq/cppc_cpufreq.c
@@ -164,8 +164,18 @@ static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy)
policy->cpuinfo.transition_latency = cppc_get_transition_latency(cpu_num);
policy->shared_type = cpu->shared_type;

- if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
+ if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
+ int i;
+
cpumask_copy(policy->cpus, cpu->shared_cpu_map);
+
+ for_each_cpu(i, policy->cpus) {
+ if (i != policy->cpu)
+ memcpy(&all_cpu_data[i]->perf_caps,
+ &cpu->perf_caps,
+ sizeof(cpu->perf_caps));
+ }
+ }
else if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL) {
/* Support only SW_ANY for now. */
pr_debug("Unsupported CPU co-ord type\n");
--
1.8.3.1

2018-04-02 07:09:28

by Viresh Kumar

Subject: Re: [PATCH] cpufreq: cppc_cpufreq: Initialize shared cpu's perf capabilities

On 28-03-18, 17:31, Shunyong Yang wrote:
> When multiple cpus are related in one cpufreq policy, the first online
> cpu will be chosen by default to handle cpufreq operations. Let's take
> cpu0 and cpu1 as an example.
>
> When cpu0 is offline, policy->cpu will be shifted to cpu1. Cpu1's should
> be initialized. Otherwise, perf capabilities are 0s and speed change can
> not take effect.
>
> This patch copies perf capabilities of the first online cpu to other
> shared cpus when policy shared type is CPUFREQ_SHARED_TYPE_ANY.
>
> Cc: Joey Zheng <[email protected]>
> Signed-off-by: Shunyong Yang <[email protected]>
> ---
>
> The original RFC link,
> https://patchwork.kernel.org/patch/10299055/.
>
> This patch solves same issue as RFC above.
>
> Patch name is changed as code is too much different with RFC above.
>
> Remove extra init() per Viresh Kumar's comments and only handle
> CPPC CPUFREQ_SHARED_TYPE_ANY case.
>
> ---
> drivers/cpufreq/cppc_cpufreq.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
> index 8f7b21a4d537..dc625a93a58e 100644
> --- a/drivers/cpufreq/cppc_cpufreq.c
> +++ b/drivers/cpufreq/cppc_cpufreq.c
> @@ -164,8 +164,18 @@ static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy)
> policy->cpuinfo.transition_latency = cppc_get_transition_latency(cpu_num);
> policy->shared_type = cpu->shared_type;
>
> - if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
> + if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
> + int i;
> +
> cpumask_copy(policy->cpus, cpu->shared_cpu_map);
> +
> + for_each_cpu(i, policy->cpus) {
> + if (i != policy->cpu)

I would rather do:

if (unlikely(i == policy->cpu))
continue;

> + memcpy(&all_cpu_data[i]->perf_caps,
> + &cpu->perf_caps,
> + sizeof(cpu->perf_caps));
> + }
> + }
> else if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL) {

It should be:

} else if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL) {

> /* Support only SW_ANY for now. */
> pr_debug("Unsupported CPU co-ord type\n");

And thanks for making it work this way.

--
viresh

2018-04-02 10:02:18

by Shunyong Yang

Subject: Re: [PATCH] cpufreq: cppc_cpufreq: Initialize shared cpu's perf capabilities

Hi, Kumar,

On Mon, 2018-04-02 at 12:38 +0530, Viresh Kumar wrote:
> On 28-03-18, 17:31, Shunyong Yang wrote:
> >
> > When multiple cpus are related in one cpufreq policy, the first
> > online
> > cpu will be chosen by default to handle cpufreq operations. Let's
> > take
> > cpu0 and cpu1 as an example.
> >
> > When cpu0 is offline, policy->cpu will be shifted to cpu1. Cpu1's
> > should
> > be initialized. Otherwise, perf capabilities are 0s and speed
> > change can
> > not take effect.
> >
> > This patch copies perf capabilities of the first online cpu to
> > other
> > shared cpus when policy shared type is CPUFREQ_SHARED_TYPE_ANY.
> >
> > Cc: Joey Zheng <[email protected]>
> > Signed-off-by: Shunyong Yang <[email protected]>
> > ---
> >
> > The original RFC link,
> > ? https://patchwork.kernel.org/patch/10299055/.
> >
> > This patch solves same issue as RFC above.
> >
> > Patch name is changed as code is too much different with RFC above.
> >
> > Remove extra init() per Viresh Kumar's comments and only handle
> > CPPC CPUFREQ_SHARED_TYPE_ANY case.
> >
> > ---
> > ?drivers/cpufreq/cppc_cpufreq.c | 12 +++++++++++-
> > ?1 file changed, 11 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/cpufreq/cppc_cpufreq.c
> > b/drivers/cpufreq/cppc_cpufreq.c
> > index 8f7b21a4d537..dc625a93a58e 100644
> > --- a/drivers/cpufreq/cppc_cpufreq.c
> > +++ b/drivers/cpufreq/cppc_cpufreq.c
> > @@ -164,8 +164,18 @@ static int cppc_cpufreq_cpu_init(struct
> > cpufreq_policy *policy)
> > ? policy->cpuinfo.transition_latency =
> > cppc_get_transition_latency(cpu_num);
> > ? policy->shared_type = cpu->shared_type;
> > ?
> > - if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
> > + if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
> > + int i;
> > +
> > ? cpumask_copy(policy->cpus, cpu->shared_cpu_map);
> > +
> > + for_each_cpu(i, policy->cpus) {
> > + if (i != policy->cpu)
> I would rather do:
>
> ????????????????????????if (unlikely(i == policy->cpu))
> ????????????????????????????????continue;

Thanks. Will change in v2.

> >
> > + memcpy(&all_cpu_data[i]-
> > >perf_caps,
> > + ???????&cpu->perf_caps,
> > + ???????sizeof(cpu->perf_caps));
> > + }
> > + }
> > ? else if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL) {
> It should be:
>
> ?? } else if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL)
> {
>

Thanks. Will change in v2. I ignored this as checkpatch.pl reports
nothing here.

Thanks.
Shunyong.


> >
> > ? /* Support only SW_ANY for now. */
> > ? pr_debug("Unsupported CPU co-ord type\n");
> And thanks for making it work this way.
>