[v1] PM: domains: Fix return value of API dev_pm_get_subsys_data()

[PATCH RESEND] PM: domains: Fix return value of API dev_pm_get_subsys_data()

Posted by Zijun Hu 3 weeks, 6 days ago

From: Zijun Hu <quic_zijuhu@quicinc.com>

dev_pm_get_subsys_data() has below 2 issues under condition
(@dev->power.subsys_data != NULL):

- it will do unnecessary kzalloc() and kfree().
- it will return -ENOMEM if the kzalloc() fails, that is wrong
  since the kzalloc() is not needed.

Fixed by not doing kzalloc() and returning 0 for the condition.

Fixes: ef27bed1870d ("PM: Reference counting of power.subsys_data")
Cc: stable@vger.kernel.org
Signed-off-by: Zijun Hu <quic_zijuhu@quicinc.com>
---
 drivers/base/power/common.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/base/power/common.c b/drivers/base/power/common.c
index 8c34ae1cd8d5..13cb1f2a06e7 100644
--- a/drivers/base/power/common.c
+++ b/drivers/base/power/common.c
@@ -26,6 +26,14 @@ int dev_pm_get_subsys_data(struct device *dev)
 {
 	struct pm_subsys_data *psd;
 
+	spin_lock_irq(&dev->power.lock);
+	if (dev->power.subsys_data) {
+		dev->power.subsys_data->refcount++;
+		spin_unlock_irq(&dev->power.lock);
+		return 0;
+	}
+	spin_unlock_irq(&dev->power.lock);
+
 	psd = kzalloc(sizeof(*psd), GFP_KERNEL);
 	if (!psd)
 		return -ENOMEM;

---
base-commit: 9852d85ec9d492ebef56dc5f229416c925758edc
change-id: 20241010-fix_dev_pm_get_subsys_data-2478bb200fde

Best regards,
-- 
Zijun Hu <quic_zijuhu@quicinc.com>

Re: [PATCH RESEND] PM: domains: Fix return value of API dev_pm_get_subsys_data()

Posted by Greg Kroah-Hartman 1 week, 5 days ago

On Mon, Oct 28, 2024 at 08:31:11PM +0800, Zijun Hu wrote:
> From: Zijun Hu <quic_zijuhu@quicinc.com>
> 
> dev_pm_get_subsys_data() has below 2 issues under condition
> (@dev->power.subsys_data != NULL):
> 
> - it will do unnecessary kzalloc() and kfree().

But that's ok, everything still works, right?

> - it will return -ENOMEM if the kzalloc() fails, that is wrong
>   since the kzalloc() is not needed.

But it's ok to return the proper error if the system is that broken.

> 
> Fixed by not doing kzalloc() and returning 0 for the condition.
> 
> Fixes: ef27bed1870d ("PM: Reference counting of power.subsys_data")
> Cc: stable@vger.kernel.org

Why is this relevant for stable kernels?

thanks,

greg k-h

Re: [PATCH RESEND] PM: domains: Fix return value of API dev_pm_get_subsys_data()

Posted by Zijun Hu 1 week, 4 days ago

On 2024/11/12 19:46, Greg Kroah-Hartman wrote:
> On Mon, Oct 28, 2024 at 08:31:11PM +0800, Zijun Hu wrote:
>> From: Zijun Hu <quic_zijuhu@quicinc.com>
>>
>> dev_pm_get_subsys_data() has below 2 issues under condition
>> (@dev->power.subsys_data != NULL):
>>
>> - it will do unnecessary kzalloc() and kfree().
> 
> But that's ok, everything still works, right?
> 

i don't think so, as explained below:

under condition (@dev->power.subsys_data != NULL), the API does
not need to do kzalloc() and should always return 0 (success).

but it actually does *unnecessary* kzalloc() and have below impact
shown:

if (kzalloc() is successfully)
	it will degrade the API's performance
else
        it changed the API's return value to -ENOMEM from 0, that
        will impact caller's logic.

>> - it will return -ENOMEM if the kzalloc() fails, that is wrong
>>   since the kzalloc() is not needed.
> 
> But it's ok to return the proper error if the system is that broken.
> >>
>> Fixed by not doing kzalloc() and returning 0 for the condition.
>>
>> Fixes: ef27bed1870d ("PM: Reference counting of power.subsys_data")
>> Cc: stable@vger.kernel.org
> 
> Why is this relevant for stable kernels?
> 

it has impact related to performance and logic as explained above.

> thanks,
> 
> greg k-h

Re: [PATCH RESEND] PM: domains: Fix return value of API dev_pm_get_subsys_data()

Posted by Zijun Hu 1 week, 5 days ago

On 2024/11/12 19:46, Greg Kroah-Hartman wrote:
> On Mon, Oct 28, 2024 at 08:31:11PM +0800, Zijun Hu wrote:
>> From: Zijun Hu <quic_zijuhu@quicinc.com>
>>
>> dev_pm_get_subsys_data() has below 2 issues under condition
>> (@dev->power.subsys_data != NULL):
>>
>> - it will do unnecessary kzalloc() and kfree().
> 
> But that's ok, everything still works, right?

yes.

> 
>> - it will return -ENOMEM if the kzalloc() fails, that is wrong
>>   since the kzalloc() is not needed.
> 
> But it's ok to return the proper error if the system is that broken.

IMO, the API should return 0 (success) instead of -ENOMEM since it does
not need to do kzalloc().

Different return value should impact caller's logic.

> 
>>
>> Fixed by not doing kzalloc() and returning 0 for the condition.
>>
>> Fixes: ef27bed1870d ("PM: Reference counting of power.subsys_data")
>> Cc: stable@vger.kernel.org
> 
> Why is this relevant for stable kernels?

you can remove both Fix and stable tag directly if you like this change.(^^)

> 
> thanks,
> 
> greg k-h