2022-03-25 19:11:26

by Kant Fan

[permalink] [raw]
Subject: [PATCH v2] thermal: devfreq_cooling: use local ops instead of global ops

Fix access illegal address problem in following condition:
There are muti devfreq cooling devices in system, some of them has
em model but other does not, energy model ops such as state2power will
append to global devfreq_cooling_ops when the cooling device with
em model register. It makes the cooling device without em model
also use devfreq_cooling_ops after appending when register later by
of_devfreq_cooling_register_power() or of_devfreq_cooling_register().

IPA governor regards the cooling devices without em model as a power actor
because they also have energy model ops, and will access illegal address
at dfc->em_pd when execute cdev->ops->get_requested_power,
cdev->ops->state2power or cdev->ops->power2state.

Fixes: 615510fe13bd2 ("thermal: devfreq_cooling: remove old power model and use EM")
Cc: [email protected] # 5.13+
Signed-off-by: Kant Fan <[email protected]>
---
drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------
1 file changed, 18 insertions(+), 7 deletions(-)

diff --git a/drivers/thermal/devfreq_cooling.c b/drivers/thermal/devfreq_cooling.c
index 4310cb342a9f..d38a80adec73 100644
--- a/drivers/thermal/devfreq_cooling.c
+++ b/drivers/thermal/devfreq_cooling.c
@@ -358,21 +358,28 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
struct thermal_cooling_device *cdev;
struct device *dev = df->dev.parent;
struct devfreq_cooling_device *dfc;
+ struct thermal_cooling_device_ops *ops;
char *name;
int err, num_opps;

- dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
- if (!dfc)
+ ops = kmemdup(&devfreq_cooling_ops, sizeof(*ops), GFP_KERNEL);
+ if (!ops)
return ERR_PTR(-ENOMEM);

+ dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
+ if (!dfc) {
+ err = -ENOMEM;
+ goto free_ops;
+ }
+
dfc->devfreq = df;

dfc->em_pd = em_pd_get(dev);
if (dfc->em_pd) {
- devfreq_cooling_ops.get_requested_power =
+ ops->get_requested_power =
devfreq_cooling_get_requested_power;
- devfreq_cooling_ops.state2power = devfreq_cooling_state2power;
- devfreq_cooling_ops.power2state = devfreq_cooling_power2state;
+ ops->state2power = devfreq_cooling_state2power;
+ ops->power2state = devfreq_cooling_power2state;

dfc->power_ops = dfc_power;

@@ -407,8 +414,7 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
if (!name)
goto remove_qos_req;

- cdev = thermal_of_cooling_device_register(np, name, dfc,
- &devfreq_cooling_ops);
+ cdev = thermal_of_cooling_device_register(np, name, dfc, ops);
kfree(name);

if (IS_ERR(cdev)) {
@@ -429,6 +435,8 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
kfree(dfc->freq_table);
free_dfc:
kfree(dfc);
+free_ops:
+ kfree(ops);

return ERR_PTR(err);
}
@@ -510,11 +518,13 @@ EXPORT_SYMBOL_GPL(devfreq_cooling_em_register);
void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)
{
struct devfreq_cooling_device *dfc;
+ const struct thermal_cooling_device_ops *ops;
struct device *dev;

if (IS_ERR_OR_NULL(cdev))
return;

+ ops = cdev->ops;
dfc = cdev->devdata;
dev = dfc->devfreq->dev.parent;

@@ -525,5 +535,6 @@ void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)

kfree(dfc->freq_table);
kfree(dfc);
+ kfree(ops);
}
EXPORT_SYMBOL_GPL(devfreq_cooling_unregister);
--
2.29.0


2022-04-13 21:58:45

by Lukasz Luba

[permalink] [raw]
Subject: Re: [PATCH v2] thermal: devfreq_cooling: use local ops instead of global ops



On 4/13/22 16:06, Lukasz Luba wrote:
>
>
> On 4/13/22 15:58, Rafael J. Wysocki wrote:
>> On Fri, Mar 25, 2022 at 10:02 AM Lukasz Luba <[email protected]> wrote:
>>>
>>> Hi Kant,
>>>
>>> On 3/25/22 07:30, Kant Fan wrote:
>>>> Fix access illegal address problem in following condition:
>>>> There are muti devfreq cooling devices in system, some of them has
>>>> em model but other does not, energy model ops such as state2power will
>>>> append to global devfreq_cooling_ops when the cooling device with
>>>> em model register. It makes the cooling device without em model
>>>> also use devfreq_cooling_ops after appending when register later by
>>>> of_devfreq_cooling_register_power() or of_devfreq_cooling_register().
>>>>
>>>> IPA governor regards the cooling devices without em model as a power
>>>> actor
>>>> because they also have energy model ops, and will access illegal
>>>> address
>>>> at dfc->em_pd when execute cdev->ops->get_requested_power,
>>>> cdev->ops->state2power or cdev->ops->power2state.
>>>>
>>>> Fixes: 615510fe13bd2 ("thermal: devfreq_cooling: remove old power
>>>> model and use EM")
>>>> Cc: [email protected] # 5.13+
>>>> Signed-off-by: Kant Fan <[email protected]>
>>>> ---
>>>>    drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------
>>>>    1 file changed, 18 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/drivers/thermal/devfreq_cooling.c
>>>> b/drivers/thermal/devfreq_cooling.c
>>>> index 4310cb342a9f..d38a80adec73 100644
>>>> --- a/drivers/thermal/devfreq_cooling.c
>>>> +++ b/drivers/thermal/devfreq_cooling.c
>>>> @@ -358,21 +358,28 @@ of_devfreq_cooling_register_power(struct
>>>> device_node *np, struct devfreq *df,
>>>>        struct thermal_cooling_device *cdev;
>>>>        struct device *dev = df->dev.parent;
>>>>        struct devfreq_cooling_device *dfc;
>>>> +     struct thermal_cooling_device_ops *ops;
>>>>        char *name;
>>>>        int err, num_opps;
>>>>
>>>> -     dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
>>>> -     if (!dfc)
>>>> +     ops = kmemdup(&devfreq_cooling_ops, sizeof(*ops), GFP_KERNEL);
>>>> +     if (!ops)
>>>>                return ERR_PTR(-ENOMEM);
>>>>
>>>> +     dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
>>>> +     if (!dfc) {
>>>> +             err = -ENOMEM;
>>>> +             goto free_ops;
>>>> +     }
>>>> +
>>>>        dfc->devfreq = df;
>>>>
>>>>        dfc->em_pd = em_pd_get(dev);
>>>>        if (dfc->em_pd) {
>>>> -             devfreq_cooling_ops.get_requested_power =
>>>> +             ops->get_requested_power =
>>>>                        devfreq_cooling_get_requested_power;
>>>> -             devfreq_cooling_ops.state2power =
>>>> devfreq_cooling_state2power;
>>>> -             devfreq_cooling_ops.power2state =
>>>> devfreq_cooling_power2state;
>>>> +             ops->state2power = devfreq_cooling_state2power;
>>>> +             ops->power2state = devfreq_cooling_power2state;
>>>>
>>>>                dfc->power_ops = dfc_power;
>>>>
>>>> @@ -407,8 +414,7 @@ of_devfreq_cooling_register_power(struct
>>>> device_node *np, struct devfreq *df,
>>>>        if (!name)
>>>>                goto remove_qos_req;
>>>>
>>>> -     cdev = thermal_of_cooling_device_register(np, name, dfc,
>>>> -                                               &devfreq_cooling_ops);
>>>> +     cdev = thermal_of_cooling_device_register(np, name, dfc, ops);
>>>>        kfree(name);
>>>>
>>>>        if (IS_ERR(cdev)) {
>>>> @@ -429,6 +435,8 @@ of_devfreq_cooling_register_power(struct
>>>> device_node *np, struct devfreq *df,
>>>>        kfree(dfc->freq_table);
>>>>    free_dfc:
>>>>        kfree(dfc);
>>>> +free_ops:
>>>> +     kfree(ops);
>>>>
>>>>        return ERR_PTR(err);
>>>>    }
>>>> @@ -510,11 +518,13 @@ EXPORT_SYMBOL_GPL(devfreq_cooling_em_register);
>>>>    void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)
>>>>    {
>>>>        struct devfreq_cooling_device *dfc;
>>>> +     const struct thermal_cooling_device_ops *ops;
>>>>        struct device *dev;
>>>>
>>>>        if (IS_ERR_OR_NULL(cdev))
>>>>                return;
>>>>
>>>> +     ops = cdev->ops;
>>>>        dfc = cdev->devdata;
>>>>        dev = dfc->devfreq->dev.parent;
>>>>
>>>> @@ -525,5 +535,6 @@ void devfreq_cooling_unregister(struct
>>>> thermal_cooling_device *cdev)
>>>>
>>>>        kfree(dfc->freq_table);
>>>>        kfree(dfc);
>>>> +     kfree(ops);
>>>>    }
>>>>    EXPORT_SYMBOL_GPL(devfreq_cooling_unregister);
>>>
>>>
>>> Thank you for updating it, LGTM
>>>
>>> Reviewed-by: Lukasz Luba <[email protected]>
>>
>> Applied as 5.19 material.
>>
>> Lukasz, this had a conflict with your EM series, please double check
>> if my resolution in the bleeding-edge branch is correct.
>
> OK, I'll let you know after I fetch and build that branch.

I've read the code and confirm you've do this correctly.
I've also built that branch with ENERGY_MODEL and DEVFREQ_COOLING
configs set - no issues observed.
Later this week I would use it for some other development
so I will test it as well.

Thank you for solving this!

Regards,
Lukasz

2022-04-14 07:08:19

by Lukasz Luba

[permalink] [raw]
Subject: Re: [PATCH v2] thermal: devfreq_cooling: use local ops instead of global ops



On 4/13/22 15:58, Rafael J. Wysocki wrote:
> On Fri, Mar 25, 2022 at 10:02 AM Lukasz Luba <[email protected]> wrote:
>>
>> Hi Kant,
>>
>> On 3/25/22 07:30, Kant Fan wrote:
>>> Fix access illegal address problem in following condition:
>>> There are muti devfreq cooling devices in system, some of them has
>>> em model but other does not, energy model ops such as state2power will
>>> append to global devfreq_cooling_ops when the cooling device with
>>> em model register. It makes the cooling device without em model
>>> also use devfreq_cooling_ops after appending when register later by
>>> of_devfreq_cooling_register_power() or of_devfreq_cooling_register().
>>>
>>> IPA governor regards the cooling devices without em model as a power actor
>>> because they also have energy model ops, and will access illegal address
>>> at dfc->em_pd when execute cdev->ops->get_requested_power,
>>> cdev->ops->state2power or cdev->ops->power2state.
>>>
>>> Fixes: 615510fe13bd2 ("thermal: devfreq_cooling: remove old power model and use EM")
>>> Cc: [email protected] # 5.13+
>>> Signed-off-by: Kant Fan <[email protected]>
>>> ---
>>> drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------
>>> 1 file changed, 18 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/drivers/thermal/devfreq_cooling.c b/drivers/thermal/devfreq_cooling.c
>>> index 4310cb342a9f..d38a80adec73 100644
>>> --- a/drivers/thermal/devfreq_cooling.c
>>> +++ b/drivers/thermal/devfreq_cooling.c
>>> @@ -358,21 +358,28 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
>>> struct thermal_cooling_device *cdev;
>>> struct device *dev = df->dev.parent;
>>> struct devfreq_cooling_device *dfc;
>>> + struct thermal_cooling_device_ops *ops;
>>> char *name;
>>> int err, num_opps;
>>>
>>> - dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
>>> - if (!dfc)
>>> + ops = kmemdup(&devfreq_cooling_ops, sizeof(*ops), GFP_KERNEL);
>>> + if (!ops)
>>> return ERR_PTR(-ENOMEM);
>>>
>>> + dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
>>> + if (!dfc) {
>>> + err = -ENOMEM;
>>> + goto free_ops;
>>> + }
>>> +
>>> dfc->devfreq = df;
>>>
>>> dfc->em_pd = em_pd_get(dev);
>>> if (dfc->em_pd) {
>>> - devfreq_cooling_ops.get_requested_power =
>>> + ops->get_requested_power =
>>> devfreq_cooling_get_requested_power;
>>> - devfreq_cooling_ops.state2power = devfreq_cooling_state2power;
>>> - devfreq_cooling_ops.power2state = devfreq_cooling_power2state;
>>> + ops->state2power = devfreq_cooling_state2power;
>>> + ops->power2state = devfreq_cooling_power2state;
>>>
>>> dfc->power_ops = dfc_power;
>>>
>>> @@ -407,8 +414,7 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
>>> if (!name)
>>> goto remove_qos_req;
>>>
>>> - cdev = thermal_of_cooling_device_register(np, name, dfc,
>>> - &devfreq_cooling_ops);
>>> + cdev = thermal_of_cooling_device_register(np, name, dfc, ops);
>>> kfree(name);
>>>
>>> if (IS_ERR(cdev)) {
>>> @@ -429,6 +435,8 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
>>> kfree(dfc->freq_table);
>>> free_dfc:
>>> kfree(dfc);
>>> +free_ops:
>>> + kfree(ops);
>>>
>>> return ERR_PTR(err);
>>> }
>>> @@ -510,11 +518,13 @@ EXPORT_SYMBOL_GPL(devfreq_cooling_em_register);
>>> void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)
>>> {
>>> struct devfreq_cooling_device *dfc;
>>> + const struct thermal_cooling_device_ops *ops;
>>> struct device *dev;
>>>
>>> if (IS_ERR_OR_NULL(cdev))
>>> return;
>>>
>>> + ops = cdev->ops;
>>> dfc = cdev->devdata;
>>> dev = dfc->devfreq->dev.parent;
>>>
>>> @@ -525,5 +535,6 @@ void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)
>>>
>>> kfree(dfc->freq_table);
>>> kfree(dfc);
>>> + kfree(ops);
>>> }
>>> EXPORT_SYMBOL_GPL(devfreq_cooling_unregister);
>>
>>
>> Thank you for updating it, LGTM
>>
>> Reviewed-by: Lukasz Luba <[email protected]>
>
> Applied as 5.19 material.
>
> Lukasz, this had a conflict with your EM series, please double check
> if my resolution in the bleeding-edge branch is correct.

OK, I'll let you know after I fetch and build that branch.

>
> Thanks!