When I hot added a CPU, I found 'cpufreq' directory is not created below
/sys/devices/system/cpu/cpuX/. It is because get_cpu_device() failed in
add_cpu_dev_symlink().
cpufreq_add_dev() is the .add_dev callback of a CPU subsys interface. It
will be called when the CPU device registered into the system. The stack
is as follows.
register_cpu()
->device_register()
->device_add()
->bus_probe_device()
->cpufreq_add_dev()
But only after the CPU device has been registered, we can get the CPU
device by get_cpu_device(), otherwise it will return NULL. Since we
already have the CPU device in cpufreq_add_dev(), pass it to
add_cpu_dev_symlink(). I noticed that the 'kobj' of the cpu device has
been added into the system before cpufreq_add_dev().
Signed-off-by: Xiongfeng Wang <[email protected]>
---
drivers/cpufreq/cpufreq.c | 9 ++++-----
1 file changed, 4 insertions(+), 5 deletions(-)
diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index e338d2f010fe..22aa2793e4d2 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -1004,10 +1004,9 @@ static struct kobj_type ktype_cpufreq = {
.release = cpufreq_sysfs_release,
};
-static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
+static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
+ struct device *dev)
{
- struct device *dev = get_cpu_device(cpu);
-
if (unlikely(!dev))
return;
@@ -1391,7 +1390,7 @@ static int cpufreq_online(unsigned int cpu)
if (new_policy) {
for_each_cpu(j, policy->related_cpus) {
per_cpu(cpufreq_cpu_data, j) = policy;
- add_cpu_dev_symlink(policy, j);
+ add_cpu_dev_symlink(policy, j, get_cpu_device(j));
}
policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
@@ -1565,7 +1564,7 @@ static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
/* Create sysfs link on CPU registration */
policy = per_cpu(cpufreq_cpu_data, cpu);
if (policy)
- add_cpu_dev_symlink(policy, cpu);
+ add_cpu_dev_symlink(policy, cpu, dev);
return 0;
}
--
2.20.1
On 29-11-21, 16:02, Xiongfeng Wang wrote:
> When I hot added a CPU, I found 'cpufreq' directory is not created below
> /sys/devices/system/cpu/cpuX/. It is because get_cpu_device() failed in
> add_cpu_dev_symlink().
>
> cpufreq_add_dev() is the .add_dev callback of a CPU subsys interface. It
> will be called when the CPU device registered into the system. The stack
> is as follows.
> register_cpu()
> ->device_register()
> ->device_add()
> ->bus_probe_device()
> ->cpufreq_add_dev()
>
> But only after the CPU device has been registered, we can get the CPU
> device by get_cpu_device(), otherwise it will return NULL. Since we
> already have the CPU device in cpufreq_add_dev(), pass it to
> add_cpu_dev_symlink(). I noticed that the 'kobj' of the cpu device has
> been added into the system before cpufreq_add_dev().
>
> Signed-off-by: Xiongfeng Wang <[email protected]>
> ---
> drivers/cpufreq/cpufreq.c | 9 ++++-----
> 1 file changed, 4 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index e338d2f010fe..22aa2793e4d2 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -1004,10 +1004,9 @@ static struct kobj_type ktype_cpufreq = {
> .release = cpufreq_sysfs_release,
> };
>
> -static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
> +static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
> + struct device *dev)
> {
> - struct device *dev = get_cpu_device(cpu);
> -
> if (unlikely(!dev))
> return;
>
> @@ -1391,7 +1390,7 @@ static int cpufreq_online(unsigned int cpu)
> if (new_policy) {
> for_each_cpu(j, policy->related_cpus) {
> per_cpu(cpufreq_cpu_data, j) = policy;
> - add_cpu_dev_symlink(policy, j);
> + add_cpu_dev_symlink(policy, j, get_cpu_device(j));
> }
>
> policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
> @@ -1565,7 +1564,7 @@ static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
> /* Create sysfs link on CPU registration */
> policy = per_cpu(cpufreq_cpu_data, cpu);
> if (policy)
> - add_cpu_dev_symlink(policy, cpu);
> + add_cpu_dev_symlink(policy, cpu, dev);
>
> return 0;
> }
Interesting that I never hit it earlier despite doing rigorous testing of
hotplug stuff :(
Anyway the patch is okay,
Acked-by: Viresh Kumar <[email protected]>
--
viresh
On Mon, Nov 29, 2021 at 10:10 AM Viresh Kumar <[email protected]> wrote:
>
> On 29-11-21, 16:02, Xiongfeng Wang wrote:
> > When I hot added a CPU, I found 'cpufreq' directory is not created below
> > /sys/devices/system/cpu/cpuX/. It is because get_cpu_device() failed in
> > add_cpu_dev_symlink().
> >
> > cpufreq_add_dev() is the .add_dev callback of a CPU subsys interface. It
> > will be called when the CPU device registered into the system. The stack
> > is as follows.
> > register_cpu()
> > ->device_register()
> > ->device_add()
> > ->bus_probe_device()
> > ->cpufreq_add_dev()
> >
> > But only after the CPU device has been registered, we can get the CPU
> > device by get_cpu_device(), otherwise it will return NULL. Since we
> > already have the CPU device in cpufreq_add_dev(), pass it to
> > add_cpu_dev_symlink(). I noticed that the 'kobj' of the cpu device has
> > been added into the system before cpufreq_add_dev().
> >
> > Signed-off-by: Xiongfeng Wang <[email protected]>
> > ---
> > drivers/cpufreq/cpufreq.c | 9 ++++-----
> > 1 file changed, 4 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> > index e338d2f010fe..22aa2793e4d2 100644
> > --- a/drivers/cpufreq/cpufreq.c
> > +++ b/drivers/cpufreq/cpufreq.c
> > @@ -1004,10 +1004,9 @@ static struct kobj_type ktype_cpufreq = {
> > .release = cpufreq_sysfs_release,
> > };
> >
> > -static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
> > +static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
> > + struct device *dev)
> > {
> > - struct device *dev = get_cpu_device(cpu);
> > -
> > if (unlikely(!dev))
> > return;
> >
> > @@ -1391,7 +1390,7 @@ static int cpufreq_online(unsigned int cpu)
> > if (new_policy) {
> > for_each_cpu(j, policy->related_cpus) {
> > per_cpu(cpufreq_cpu_data, j) = policy;
> > - add_cpu_dev_symlink(policy, j);
> > + add_cpu_dev_symlink(policy, j, get_cpu_device(j));
> > }
> >
> > policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
> > @@ -1565,7 +1564,7 @@ static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
> > /* Create sysfs link on CPU registration */
> > policy = per_cpu(cpufreq_cpu_data, cpu);
> > if (policy)
> > - add_cpu_dev_symlink(policy, cpu);
> > + add_cpu_dev_symlink(policy, cpu, dev);
> >
> > return 0;
> > }
>
> Interesting that I never hit it earlier despite doing rigorous testing of
> hotplug stuff :(
This is the real hot-add path which isn't tested on a regular basis.
> Anyway the patch is okay,
It would be good to add a Fixes: tag to it, though. Any idea about
the commit this should point to?
> Acked-by: Viresh Kumar <[email protected]>
Hi,
On 2021/11/30 19:42, Rafael J. Wysocki wrote:
> On Mon, Nov 29, 2021 at 10:10 AM Viresh Kumar <[email protected]> wrote:
>>
>> On 29-11-21, 16:02, Xiongfeng Wang wrote:
>>> When I hot added a CPU, I found 'cpufreq' directory is not created below
>>> /sys/devices/system/cpu/cpuX/. It is because get_cpu_device() failed in
>>> add_cpu_dev_symlink().
>>>
>>> cpufreq_add_dev() is the .add_dev callback of a CPU subsys interface. It
>>> will be called when the CPU device registered into the system. The stack
>>> is as follows.
>>> register_cpu()
>>> ->device_register()
>>> ->device_add()
>>> ->bus_probe_device()
>>> ->cpufreq_add_dev()
>>>
>>> But only after the CPU device has been registered, we can get the CPU
>>> device by get_cpu_device(), otherwise it will return NULL. Since we
>>> already have the CPU device in cpufreq_add_dev(), pass it to
>>> add_cpu_dev_symlink(). I noticed that the 'kobj' of the cpu device has
>>> been added into the system before cpufreq_add_dev().
>>>
>>> Signed-off-by: Xiongfeng Wang <[email protected]>
>>> ---
>>> drivers/cpufreq/cpufreq.c | 9 ++++-----
>>> 1 file changed, 4 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
>>> index e338d2f010fe..22aa2793e4d2 100644
>>> --- a/drivers/cpufreq/cpufreq.c
>>> +++ b/drivers/cpufreq/cpufreq.c
>>> @@ -1004,10 +1004,9 @@ static struct kobj_type ktype_cpufreq = {
>>> .release = cpufreq_sysfs_release,
>>> };
>>>
>>> -static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
>>> +static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
>>> + struct device *dev)
>>> {
>>> - struct device *dev = get_cpu_device(cpu);
>>> -
>>> if (unlikely(!dev))
>>> return;
>>>
>>> @@ -1391,7 +1390,7 @@ static int cpufreq_online(unsigned int cpu)
>>> if (new_policy) {
>>> for_each_cpu(j, policy->related_cpus) {
>>> per_cpu(cpufreq_cpu_data, j) = policy;
>>> - add_cpu_dev_symlink(policy, j);
>>> + add_cpu_dev_symlink(policy, j, get_cpu_device(j));
>>> }
>>>
>>> policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
>>> @@ -1565,7 +1564,7 @@ static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
>>> /* Create sysfs link on CPU registration */
>>> policy = per_cpu(cpufreq_cpu_data, cpu);
>>> if (policy)
>>> - add_cpu_dev_symlink(policy, cpu);
>>> + add_cpu_dev_symlink(policy, cpu, dev);
>>>
>>> return 0;
>>> }
>>
>> Interesting that I never hit it earlier despite doing rigorous testing of
>> hotplug stuff :(
>
> This is the real hot-add path which isn't tested on a regular basis.
>
>> Anyway the patch is okay,
>
> It would be good to add a Fixes: tag to it, though. Any idea about
> the commit this should point to?
When I look up the commit history, I found this one.
2f0ba790df51 ("cpufreq: Fix creation of symbolic links to policy directories")
Before this commit, the 'dev' is passed to add_cpu_dev_symlink() in
cpufreq_add_dev(). Maybe we can point to this one.
>
>> Acked-by: Viresh Kumar <[email protected]>
> .
>
On 30-11-21, 12:42, Rafael J. Wysocki wrote:
> This is the real hot-add path which isn't tested on a regular basis.
Ahh, I thought this is a simple offline/online thing. Makes sense now.
> > Anyway the patch is okay,
>
> It would be good to add a Fixes: tag to it, though. Any idea about
> the commit this should point to?
This is broken since a very long time then, we need to get this into all stable
kernels we care about.
As Xiongfeng pointed out, 2f0ba790df51 ("cpufreq: Fix creation of symbolic links
to policy directories") looks to be a good candidate.
--
viresh
On Wed, Dec 1, 2021 at 8:23 AM Viresh Kumar <[email protected]> wrote:
>
> On 30-11-21, 12:42, Rafael J. Wysocki wrote:
> > This is the real hot-add path which isn't tested on a regular basis.
>
> Ahh, I thought this is a simple offline/online thing. Makes sense now.
>
> > > Anyway the patch is okay,
> >
> > It would be good to add a Fixes: tag to it, though. Any idea about
> > the commit this should point to?
>
> This is broken since a very long time then, we need to get this into all stable
> kernels we care about.
>
> As Xiongfeng pointed out, 2f0ba790df51 ("cpufreq: Fix creation of symbolic links
> to policy directories") looks to be a good candidate.
Applied as 5.16-rc material, thanks!