[PATCH] PM / devfreq: passive: Use non-devm notifiers

[PATCH] PM / devfreq: passive: Use non-devm notifiers

[Leonard Crestez]

The devfreq passive governor registers and unregisters devfreq
transition notifiers on DEVFREQ_GOV_START/GOV_STOP using devm wrappers.

If devfreq itself is registered with devm then a warning is triggered on
rmmod from devm_devfreq_unregister_notifier. Call stack looks like this:

	devm_devfreq_unregister_notifier+0x30/0x40
	devfreq_passive_event_handler+0x4c/0x88
	devfreq_remove_device.part.8+0x6c/0x9c
	devm_devfreq_dev_release+0x18/0x20
	release_nodes+0x1b0/0x220
	devres_release_all+0x78/0x84
	device_release_driver_internal+0x100/0x1c0
	driver_detach+0x4c/0x90
	bus_remove_driver+0x7c/0xd0
	driver_unregister+0x2c/0x58
	platform_driver_unregister+0x10/0x18
	imx_devfreq_platdrv_exit+0x14/0xd40 [imx_devfreq]

This happens because devres_release_all will first remove all the nodes
into a separate todo list so the nested devres_release from
devm_devfreq_unregister_notifier won't find anything.

Fix the warning by calling the non-devm APIS for frequency notification.
Using devm wrappers is not actually useful for a governor anyway: it
relies on the devfreq core to correctly match the GOV_START/GOV_STOP
notifications.

Signed-off-by: Leonard Crestez <leonard.crestez@nxp.com>

---
 drivers/devfreq/governor_passive.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

The only current user of passive governor is exynos-bus; does rmmod work
for you? Maybe I'm missing something.

It also seems that no attempt is made to increase the ref count of
the passive "parent" so in theory devices can be removed while still
referenced. However that would be a separate issue.

diff --git a/drivers/devfreq/governor_passive.c b/drivers/devfreq/governor_passive.c
index 58308948b863..da485477065c 100644
--- a/drivers/devfreq/governor_passive.c
+++ b/drivers/devfreq/governor_passive.c
@@ -163,16 +163,16 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
 	case DEVFREQ_GOV_START:
 		if (!p_data->this)
 			p_data->this = devfreq;
 
 		nb->notifier_call = devfreq_passive_notifier_call;
-		ret = devm_devfreq_register_notifier(dev, parent, nb,
+		ret = devfreq_register_notifier(parent, nb,
 					DEVFREQ_TRANSITION_NOTIFIER);
 		break;
 	case DEVFREQ_GOV_STOP:
-		devm_devfreq_unregister_notifier(dev, parent, nb,
-					DEVFREQ_TRANSITION_NOTIFIER);
+		WARN_ON(devfreq_unregister_notifier(parent, nb,
+					DEVFREQ_TRANSITION_NOTIFIER));
 		break;
 	default:
 		break;
 	}
 
-- 
2.17.1

Re: [PATCH] PM / devfreq: passive: Use non-devm notifiers

[Chanwoo Choi]

Hi,

On 19. 8. 9. 오전 1:54, Leonard Crestez wrote:
> The devfreq passive governor registers and unregisters devfreq
> transition notifiers on DEVFREQ_GOV_START/GOV_STOP using devm wrappers.
> 
> If devfreq itself is registered with devm then a warning is triggered on
> rmmod from devm_devfreq_unregister_notifier. Call stack looks like this:
> 
> 	devm_devfreq_unregister_notifier+0x30/0x40
> 	devfreq_passive_event_handler+0x4c/0x88
> 	devfreq_remove_device.part.8+0x6c/0x9c
> 	devm_devfreq_dev_release+0x18/0x20
> 	release_nodes+0x1b0/0x220
> 	devres_release_all+0x78/0x84
> 	device_release_driver_internal+0x100/0x1c0
> 	driver_detach+0x4c/0x90
> 	bus_remove_driver+0x7c/0xd0
> 	driver_unregister+0x2c/0x58
> 	platform_driver_unregister+0x10/0x18
> 	imx_devfreq_platdrv_exit+0x14/0xd40 [imx_devfreq]
> 
> This happens because devres_release_all will first remove all the nodes
> into a separate todo list so the nested devres_release from
> devm_devfreq_unregister_notifier won't find anything.
> 
> Fix the warning by calling the non-devm APIS for frequency notification.
> Using devm wrappers is not actually useful for a governor anyway: it
> relies on the devfreq core to correctly match the GOV_START/GOV_STOP
> notifications.
> 
> Signed-off-by: Leonard Crestez <leonard.crestez@nxp.com>
> 
> ---
>  drivers/devfreq/governor_passive.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> The only current user of passive governor is exynos-bus; does rmmod work
> for you? Maybe I'm missing something.
> 
> It also seems that no attempt is made to increase the ref count of
> the passive "parent" so in theory devices can be removed while still
> referenced. However that would be a separate issue.
> 
> diff --git a/drivers/devfreq/governor_passive.c b/drivers/devfreq/governor_passive.c
> index 58308948b863..da485477065c 100644
> --- a/drivers/devfreq/governor_passive.c
> +++ b/drivers/devfreq/governor_passive.c
> @@ -163,16 +163,16 @@ static int devfreq_passive_event_handler(struct devfreq *devfreq,
>  	case DEVFREQ_GOV_START:
>  		if (!p_data->this)
>  			p_data->this = devfreq;
>  
>  		nb->notifier_call = devfreq_passive_notifier_call;
> -		ret = devm_devfreq_register_notifier(dev, parent, nb,
> +		ret = devfreq_register_notifier(parent, nb,
>  					DEVFREQ_TRANSITION_NOTIFIER);
>  		break;
>  	case DEVFREQ_GOV_STOP:
> -		devm_devfreq_unregister_notifier(dev, parent, nb,
> -					DEVFREQ_TRANSITION_NOTIFIER);
> +		WARN_ON(devfreq_unregister_notifier(parent, nb,
> +					DEVFREQ_TRANSITION_NOTIFIER));
>  		break;
>  	default:
>  		break;
>  	}
>  
> 

Looks good to me. But, you have to add the following fixes tag
and send it to stable mailing list to fix the bug.
- Fixes: 996133119f57 ("PM / devfreq: Add new passive governor")

Acked-by: Chanwoo Choi <cw00.choi@samsung.com> 

-- 
Best Regards,
Chanwoo Choi
Samsung Electronics