[PATCH] devfreq: simple_ondemand: Update devfreq stats when governor started

[PATCH] devfreq: simple_ondemand: Update devfreq stats when governor started

[Jeffy Chen]

If governor suspends soon after started, it may not have the chance to
update devfreq stats, which leaves devfreq stats' current frequence be
zero.

So when the thermal core tries to throttle the power, it would failed
to get the correct static power of current frequence and print these
warnings:
[    2.024735] core: dev_pm_opp_get_voltage: Invalid parameters
[    2.024743] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
-34
...
[    2.026320] core: dev_pm_opp_get_voltage: Invalid parameters
[    2.026327] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
-34

Update devfreq stats when governor started to avoid that.

Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>
---

 drivers/devfreq/governor_simpleondemand.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/devfreq/governor_simpleondemand.c b/drivers/devfreq/governor_simpleondemand.c
index ae72ba5..26e0e58 100644
--- a/drivers/devfreq/governor_simpleondemand.c
+++ b/drivers/devfreq/governor_simpleondemand.c
@@ -98,6 +98,7 @@ static int devfreq_simple_ondemand_handler(struct devfreq *devfreq,
 {
 	switch (event) {
 	case DEVFREQ_GOV_START:
+		devfreq_update_stats(devfreq);
 		devfreq_monitor_start(devfreq);
 		break;
 
-- 
2.1.4

RE: [PATCH] devfreq: simple_ondemand: Update devfreq stats when governor started

[MyungJoo Ham]

> If governor suspends soon after started, it may not have the chance to
> update devfreq stats, which leaves devfreq stats' current frequence be
> zero.
> 
> So when the thermal core tries to throttle the power, it would failed
> to get the correct static power of current frequence and print these
> warnings:
> [    2.024735] core: dev_pm_opp_get_voltage: Invalid parameters
> [    2.024743] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
> -34
> ...
> [    2.026320] core: dev_pm_opp_get_voltage: Invalid parameters
> [    2.026327] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
> -34
> 
> Update devfreq stats when governor started to avoid that.
> 
> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>

How are you calling suspend/resume routins in the mali drivers?

It appears that your mali device driver's suspend/resume routines are
accessing "struct devfreq_dev_status" directly. Is this correct?


Plus, for MALI (if that assumption is correct), this issue might occur
with other governors.

If so, would it be better to put the first initializing call of
devfreq_update_stats() at devfreq_add_device()?


Cheers,
MyungJoo

ps. Where can I look at the mali-devfreq device driver? (I couldn't find it)

Re: [PATCH] devfreq: simple_ondemand: Update devfreq stats when governor started

[jeffy]

hi Ham,

Thanks for your reply.

On 07/18/2017 06:35 PM, MyungJoo Ham wrote:
>> If governor suspends soon after started, it may not have the chance to
>> update devfreq stats, which leaves devfreq stats' current frequence be
>> zero.
>>
>> So when the thermal core tries to throttle the power, it would failed
>> to get the correct static power of current frequence and print these
>> warnings:
>> [    2.024735] core: dev_pm_opp_get_voltage: Invalid parameters
>> [    2.024743] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
>> -34
>> ...
>> [    2.026320] core: dev_pm_opp_get_voltage: Invalid parameters
>> [    2.026327] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
>> -34
>>
>> Update devfreq stats when governor started to avoid that.
>>
>> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>
>
> How are you calling suspend/resume routins in the mali drivers?
>
i was porting mali driver from here:
https://chromium.googlesource.com/chromiumos/third_party/kernel/+/chromeos-4.4/drivers/gpu/arm/midgard/mali_kbase_core_linux.c#4079
> It appears that your mali device driver's suspend/resume routines are
> accessing "struct devfreq_dev_status" directly. Is this correct?
>
no, but the suspend/resume would suspend/resume the governor.
>
> Plus, for MALI (if that assumption is correct), this issue might occur
> with other governors.
>
> If so, would it be better to put the first initializing call of
> devfreq_update_stats() at devfreq_add_device()?
>
right, that make sense. i put it here because the comment of this 
function said it should be called by governor, but you're right, it 
would be better this way, will send new version soon.
>
> Cheers,
> MyungJoo
>
> ps. Where can I look at the mali-devfreq device driver? (I couldn't find it)
sorry, i should mention that, please check above link :)
>
>
>
>

Re: [PATCH] devfreq: simple_ondemand: Update devfreq stats when governor started

[jeffy]

Hi Ham,

oops, sorry, it turns out the devfreq would already tried to handle this 
case(by using profile's initial_freq):

8d39fc0 PM / devfreq: fix initialization of current frequency in last status

and my local kernel didn't contains this commit.

so this patch is not needed, please ignore it.

On 07/18/2017 07:27 PM, jeffy wrote:
> hi Ham,
>
> Thanks for your reply.
>
> On 07/18/2017 06:35 PM, MyungJoo Ham wrote:
>>> If governor suspends soon after started, it may not have the chance to
>>> update devfreq stats, which leaves devfreq stats' current frequence be
>>> zero.
>>>
>>> So when the thermal core tries to throttle the power, it would failed
>>> to get the correct static power of current frequence and print these
>>> warnings:
>>> [    2.024735] core: dev_pm_opp_get_voltage: Invalid parameters
>>> [    2.024743] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
>>> -34
>>> ...
>>> [    2.026320] core: dev_pm_opp_get_voltage: Invalid parameters
>>> [    2.026327] mali ff9a0000.gpu: Failed to get voltage for frequency 0:
>>> -34
>>>
>>> Update devfreq stats when governor started to avoid that.
>>>
>>> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>
>>
>> How are you calling suspend/resume routins in the mali drivers?
>>
> i was porting mali driver from here:
> https://chromium.googlesource.com/chromiumos/third_party/kernel/+/chromeos-4.4/drivers/gpu/arm/midgard/mali_kbase_core_linux.c#4079
>
>> It appears that your mali device driver's suspend/resume routines are
>> accessing "struct devfreq_dev_status" directly. Is this correct?
>>
> no, but the suspend/resume would suspend/resume the governor.
>>
>> Plus, for MALI (if that assumption is correct), this issue might occur
>> with other governors.
>>
>> If so, would it be better to put the first initializing call of
>> devfreq_update_stats() at devfreq_add_device()?
>>
> right, that make sense. i put it here because the comment of this
> function said it should be called by governor, but you're right, it
> would be better this way, will send new version soon.
>>
>> Cheers,
>> MyungJoo
>>
>> ps. Where can I look at the mali-devfreq device driver? (I couldn't
>> find it)
> sorry, i should mention that, please check above link :)
>>
>>
>>
>>
>