[BISECTED] Allwinner A33 tablet does not fully power off

  Hi,

After commit 843107498f91e57d1d4b22cd8787112726fdaeb4 (bus: sunxi-rsb: 
Implement suspend/resume/shutdown callbacks) Q8 A33 tablet I have here 
cannot be powered-on after power-off, it needs press-and-hold of the 
power button for 10 seconds (I guess some HW assisted power down 
happens) before it can be powered-on again.

The following patch makes it behave correctly:

  static const struct dev_pm_ops sunxi_rsb_dev_pm_ops = {


I guess the issue comes from the fact that by the time 'power off' 
command to the power management IC has to be send, the bus it lives on 
is already down, so the device is left in semi-powered down state. Ofc 
this is a wild guess, however, preventing the bus being turned off on 
shutdown fixes the issue.

Please LMK if the above is the correct approach so I will send a proper 
patch or something else shall be fixed.

Ivo
  

Hi Ivo,

On 10/29/22 10:23, Ivaylo Dimitrov wrote:
> After commit 843107498f91e57d1d4b22cd8787112726fdaeb4 (bus: sunxi-rsb:
> Implement suspend/resume/shutdown callbacks) Q8 A33 tablet I have here
> cannot be powered-on after power-off, it needs press-and-hold of the
> power button for 10 seconds (I guess some HW assisted power down
> happens) before it can be powered-on again.
> 
> The following patch makes it behave correctly:
> 
> diff --git a/drivers/bus/sunxi-rsb.c b/drivers/bus/sunxi-rsb.c
> index 60b082fe2ed0..30016d62044c 100644
> --- a/drivers/bus/sunxi-rsb.c
> +++ b/drivers/bus/sunxi-rsb.c
> @@ -818,10 +818,7 @@ static int sunxi_rsb_remove(struct platform_device
> *pdev)
> 
>  static void sunxi_rsb_shutdown(struct platform_device *pdev)
>  {
> -       struct sunxi_rsb *rsb = platform_get_drvdata(pdev);
> -
>         pm_runtime_disable(&pdev->dev);
> -       sunxi_rsb_hw_exit(rsb);
>  }
> 
>  static const struct dev_pm_ops sunxi_rsb_dev_pm_ops = {
> 
> 
> I guess the issue comes from the fact that by the time 'power off'
> command to the power management IC has to be send, the bus it lives on
> is already down, so the device is left in semi-powered down state. Ofc
> this is a wild guess, however, preventing the bus being turned off on
> shutdown fixes the issue.

Your guess is correct. The controller gets shut down in

  kernel_power_off()
    kernel_shutdown_prepare()
      device_shutdown()

but the PMIC communication needs to happen later in

  kernel_power_off()
    machine_power_off()
      pm_power_off()

> Please LMK if the above is the correct approach so I will send a proper
> patch or something else shall be fixed.

Yes, this is exactly the right approach. The whole sunxi_rsb_shutdown()
function should be removed. When you send a patch, please add a Fixes:
tag referencing the commit that you bisected to.

Regards,
Samuel
  

Hi Samuel,

On 5.11.22 г. 4:21 ч., Samuel Holland wrote:
> Hi Ivo,
> 
> On 10/29/22 10:23, Ivaylo Dimitrov wrote:
>> After commit 843107498f91e57d1d4b22cd8787112726fdaeb4 (bus: sunxi-rsb:
>> Implement suspend/resume/shutdown callbacks) Q8 A33 tablet I have here
>> cannot be powered-on after power-off, it needs press-and-hold of the
>> power button for 10 seconds (I guess some HW assisted power down
>> happens) before it can be powered-on again.
>>
>> The following patch makes it behave correctly:
>>
>> diff --git a/drivers/bus/sunxi-rsb.c b/drivers/bus/sunxi-rsb.c
>> index 60b082fe2ed0..30016d62044c 100644
>> --- a/drivers/bus/sunxi-rsb.c
>> +++ b/drivers/bus/sunxi-rsb.c
>> @@ -818,10 +818,7 @@ static int sunxi_rsb_remove(struct platform_device
>> *pdev)
>>
>>   static void sunxi_rsb_shutdown(struct platform_device *pdev)
>>   {
>> -       struct sunxi_rsb *rsb = platform_get_drvdata(pdev);
>> -
>>          pm_runtime_disable(&pdev->dev);
>> -       sunxi_rsb_hw_exit(rsb);
>>   }
>>
>>   static const struct dev_pm_ops sunxi_rsb_dev_pm_ops = {
>>
>>
>> I guess the issue comes from the fact that by the time 'power off'
>> command to the power management IC has to be send, the bus it lives on
>> is already down, so the device is left in semi-powered down state. Ofc
>> this is a wild guess, however, preventing the bus being turned off on
>> shutdown fixes the issue.
> 
> Your guess is correct. The controller gets shut down in
> 
>    kernel_power_off()
>      kernel_shutdown_prepare()
>        device_shutdown()
> 
> but the PMIC communication needs to happen later in
> 
>    kernel_power_off()
>      machine_power_off()
>        pm_power_off()
> 
>> Please LMK if the above is the correct approach so I will send a proper
>> patch or something else shall be fixed.
> 
> Yes, this is exactly the right approach. The whole sunxi_rsb_shutdown()

Don't we need pm_runtime_disable() on shutdown? As IIUC, the controller 
might be suspended and we have to resume it to put it in state to accept 
commands later on(in pm_power_off()).

Regards,
Ivo

> function should be removed. When you send a patch, please add a Fixes:
> tag referencing the commit that you bisected to.
> 
> Regards,
> Samuel
>
  

On 11/5/22 03:23, Ivaylo Dimitrov wrote:
> Hi Samuel,
> 
> On 5.11.22 г. 4:21 ч., Samuel Holland wrote:
>> Hi Ivo,
>>
>> On 10/29/22 10:23, Ivaylo Dimitrov wrote:
>>> After commit 843107498f91e57d1d4b22cd8787112726fdaeb4 (bus: sunxi-rsb:
>>> Implement suspend/resume/shutdown callbacks) Q8 A33 tablet I have here
>>> cannot be powered-on after power-off, it needs press-and-hold of the
>>> power button for 10 seconds (I guess some HW assisted power down
>>> happens) before it can be powered-on again.
>>>
>>> The following patch makes it behave correctly:
>>>
>>> diff --git a/drivers/bus/sunxi-rsb.c b/drivers/bus/sunxi-rsb.c
>>> index 60b082fe2ed0..30016d62044c 100644
>>> --- a/drivers/bus/sunxi-rsb.c
>>> +++ b/drivers/bus/sunxi-rsb.c
>>> @@ -818,10 +818,7 @@ static int sunxi_rsb_remove(struct platform_device
>>> *pdev)
>>>
>>>   static void sunxi_rsb_shutdown(struct platform_device *pdev)
>>>   {
>>> -       struct sunxi_rsb *rsb = platform_get_drvdata(pdev);
>>> -
>>>          pm_runtime_disable(&pdev->dev);
>>> -       sunxi_rsb_hw_exit(rsb);
>>>   }
>>>
>>>   static const struct dev_pm_ops sunxi_rsb_dev_pm_ops = {
>>>
>>>
>>> I guess the issue comes from the fact that by the time 'power off'
>>> command to the power management IC has to be send, the bus it lives on
>>> is already down, so the device is left in semi-powered down state. Ofc
>>> this is a wild guess, however, preventing the bus being turned off on
>>> shutdown fixes the issue.
>>
>> Your guess is correct. The controller gets shut down in
>>
>>    kernel_power_off()
>>      kernel_shutdown_prepare()
>>        device_shutdown()
>>
>> but the PMIC communication needs to happen later in
>>
>>    kernel_power_off()
>>      machine_power_off()
>>        pm_power_off()
>>
>>> Please LMK if the above is the correct approach so I will send a proper
>>> patch or something else shall be fixed.
>>
>> Yes, this is exactly the right approach. The whole sunxi_rsb_shutdown()
> 
> Don't we need pm_runtime_disable() on shutdown? As IIUC, the controller
> might be suspended and we have to resume it to put it in state to accept
> commands later on(in pm_power_off()).

sunxi_rsb_write() takes care of resuming the controller, so the
controller being suspended prior to pm_power_off() is fine.
pm_runtime_disable() would actually prevent resuming the controller
later in sunxi_rsb_write().

>> function should be removed. When you send a patch, please add a Fixes:
>> tag referencing the commit that you bisected to.

I found a couple of other issues as well, so I'll send out some fixes
with you CC'd.

Regards,
Samuel
  

On 5.11.22 г. 21:18 ч., Samuel Holland wrote:
> On 11/5/22 03:23, Ivaylo Dimitrov wrote:
>> Hi Samuel,
>>
>> On 5.11.22 г. 4:21 ч., Samuel Holland wrote:
>>> Hi Ivo,
>>>
>>> On 10/29/22 10:23, Ivaylo Dimitrov wrote:
>>>> After commit 843107498f91e57d1d4b22cd8787112726fdaeb4 (bus: sunxi-rsb:
>>>> Implement suspend/resume/shutdown callbacks) Q8 A33 tablet I have here
>>>> cannot be powered-on after power-off, it needs press-and-hold of the
>>>> power button for 10 seconds (I guess some HW assisted power down
>>>> happens) before it can be powered-on again.
>>>>
>>>> The following patch makes it behave correctly:
>>>>
>>>> diff --git a/drivers/bus/sunxi-rsb.c b/drivers/bus/sunxi-rsb.c
>>>> index 60b082fe2ed0..30016d62044c 100644
>>>> --- a/drivers/bus/sunxi-rsb.c
>>>> +++ b/drivers/bus/sunxi-rsb.c
>>>> @@ -818,10 +818,7 @@ static int sunxi_rsb_remove(struct platform_device
>>>> *pdev)
>>>>
>>>>    static void sunxi_rsb_shutdown(struct platform_device *pdev)
>>>>    {
>>>> -       struct sunxi_rsb *rsb = platform_get_drvdata(pdev);
>>>> -
>>>>           pm_runtime_disable(&pdev->dev);
>>>> -       sunxi_rsb_hw_exit(rsb);
>>>>    }
>>>>
>>>>    static const struct dev_pm_ops sunxi_rsb_dev_pm_ops = {
>>>>
>>>>
>>>> I guess the issue comes from the fact that by the time 'power off'
>>>> command to the power management IC has to be send, the bus it lives on
>>>> is already down, so the device is left in semi-powered down state. Ofc
>>>> this is a wild guess, however, preventing the bus being turned off on
>>>> shutdown fixes the issue.
>>>
>>> Your guess is correct. The controller gets shut down in
>>>
>>>     kernel_power_off()
>>>       kernel_shutdown_prepare()
>>>         device_shutdown()
>>>
>>> but the PMIC communication needs to happen later in
>>>
>>>     kernel_power_off()
>>>       machine_power_off()
>>>         pm_power_off()
>>>
>>>> Please LMK if the above is the correct approach so I will send a proper
>>>> patch or something else shall be fixed.
>>>
>>> Yes, this is exactly the right approach. The whole sunxi_rsb_shutdown()
>>
>> Don't we need pm_runtime_disable() on shutdown? As IIUC, the controller
>> might be suspended and we have to resume it to put it in state to accept
>> commands later on(in pm_power_off()).
> 
> sunxi_rsb_write() takes care of resuming the controller, so the
> controller being suspended prior to pm_power_off() is fine.
> pm_runtime_disable() would actually prevent resuming the controller
> later in sunxi_rsb_write().
> 

I see.

>>> function should be removed. When you send a patch, please add a Fixes:
>>> tag referencing the commit that you bisected to.
> 
> I found a couple of other issues as well, so I'll send out some fixes
> with you CC'd.
>

Ok, thanks. What about stable kernels, poweroff is broken since 
843107498f91e57d1d4b22cd8787112726fdaeb4?

Regards,
Ivo

> Regards,
> Samuel
>
  

Dne nedelja, 06. november 2022 ob 08:15:40 CET je Ivaylo Dimitrov napisal(a):
> On 5.11.22 г. 21:18 ч., Samuel Holland wrote:
> > On 11/5/22 03:23, Ivaylo Dimitrov wrote:
> >> Hi Samuel,
> >> 
> >> On 5.11.22 г. 4:21 ч., Samuel Holland wrote:
> >>> Hi Ivo,
> >>> 
> >>> On 10/29/22 10:23, Ivaylo Dimitrov wrote:
> >>>> After commit 843107498f91e57d1d4b22cd8787112726fdaeb4 (bus: sunxi-rsb:
> >>>> Implement suspend/resume/shutdown callbacks) Q8 A33 tablet I have here
> >>>> cannot be powered-on after power-off, it needs press-and-hold of the
> >>>> power button for 10 seconds (I guess some HW assisted power down
> >>>> happens) before it can be powered-on again.
> >>>> 
> >>>> The following patch makes it behave correctly:
> >>>> 
> >>>> diff --git a/drivers/bus/sunxi-rsb.c b/drivers/bus/sunxi-rsb.c
> >>>> index 60b082fe2ed0..30016d62044c 100644
> >>>> --- a/drivers/bus/sunxi-rsb.c
> >>>> +++ b/drivers/bus/sunxi-rsb.c
> >>>> @@ -818,10 +818,7 @@ static int sunxi_rsb_remove(struct platform_device
> >>>> *pdev)
> >>>> 
> >>>>    static void sunxi_rsb_shutdown(struct platform_device *pdev)
> >>>>    {
> >>>> 
> >>>> -       struct sunxi_rsb *rsb = platform_get_drvdata(pdev);
> >>>> -
> >>>> 
> >>>>           pm_runtime_disable(&pdev->dev);
> >>>> 
> >>>> -       sunxi_rsb_hw_exit(rsb);
> >>>> 
> >>>>    }
> >>>>  
> >>>>    static const struct dev_pm_ops sunxi_rsb_dev_pm_ops = {
> >>>> 
> >>>> I guess the issue comes from the fact that by the time 'power off'
> >>>> command to the power management IC has to be send, the bus it lives on
> >>>> is already down, so the device is left in semi-powered down state. Ofc
> >>>> this is a wild guess, however, preventing the bus being turned off on
> >>>> shutdown fixes the issue.
> >>> 
> >>> Your guess is correct. The controller gets shut down in
> >>> 
> >>>     kernel_power_off()
> >>>       kernel_shutdown_prepare()
> >>>         device_shutdown()
> >>> 
> >>> but the PMIC communication needs to happen later in
> >>> 
> >>>     kernel_power_off()
> >>>       machine_power_off()
> >>>         pm_power_off()
> >>>  
> >>>> Please LMK if the above is the correct approach so I will send a proper
> >>>> patch or something else shall be fixed.
> >>> 
> >>> Yes, this is exactly the right approach. The whole sunxi_rsb_shutdown()
> >> 
> >> Don't we need pm_runtime_disable() on shutdown? As IIUC, the controller
> >> might be suspended and we have to resume it to put it in state to accept
> >> commands later on(in pm_power_off()).
> > 
> > sunxi_rsb_write() takes care of resuming the controller, so the
> > controller being suspended prior to pm_power_off() is fine.
> > pm_runtime_disable() would actually prevent resuming the controller
> > later in sunxi_rsb_write().
> 
> I see.
> 
> >>> function should be removed. When you send a patch, please add a Fixes:
> >>> tag referencing the commit that you bisected to.
> > 
> > I found a couple of other issues as well, so I'll send out some fixes
> > with you CC'd.
> 
> Ok, thanks. What about stable kernels, poweroff is broken since
> 843107498f91e57d1d4b22cd8787112726fdaeb4?

All changes must first hit latest development branch. If they are marked with 
fixes tag, they get automatically backported in all currently maintained 
kernels (mostly LTS) if applicable. But this will take a while.

Best regards,
Jernej

> 
> Regards,
> Ivo
> 
> > Regards,
> > Samuel