Re: [RFC PATCH] PM / devfreq: Add policy notifier
On Fri, May 18, 2018 at 08:26:30AM +0900, Chanwoo Choi wrote: > Hi, > > On 2018년 05월 18일 08:07, Matthias Kaehlcke wrote: > > Hi, > > > > On Thu, May 17, 2018 at 11:01:34AM +0900, Chanwoo Choi wrote: > >> Hi, > >> > >> Could you give some use-case of DEVFREQ_POLICY_NOTIFIER > >> or send use-case patch with this patch? > > > > This is a WIP patch that makes use of the DEVFREQ_POLICY_NOTIFIER: > > > > https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/1065122 > > > >> I already knew the CPUFREQ_POLICY_NOTIFIER. > >> But, until now, there are no any requirements of DEVFREQ_POLICY_NOTIFIER. > >> If there are no any use-case, it is not necessary codes. > > > > Sure, I intend to land the above driver upstream if devfreq can > > provide the necessary interfaces. > > I recommend that you should send the patch with the use-case patch. One of the reasons this patch was sent as an RFC was to get an idea whether it would be feasible to add support for DEVFREQ_POLICY_NOTIFIER before spending much time implementing a driver that relies on it, and then possibly hear that it's not going to fly. However since you didn't voice general concerns about the idea I made progress with the driver in the last days. I'll polish it a bit more and send it in a series with related devfreq patches. > >> On 2018년 05월 16일 06:24, Matthias Kaehlcke wrote: > >>> diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c > >>> index fe2af6aa88fc..a7294c056f65 100644 > >>> --- a/drivers/devfreq/devfreq.c > >>> +++ b/drivers/devfreq/devfreq.c > >>> @@ -273,6 +273,9 @@ int update_devfreq(struct devfreq *devfreq) > >>> if (err) > >>> return err; > >>> > >>> + srcu_notifier_call_chain(&devfreq->policy_notifier_list, > >>> + DEVFREQ_ADJUST, &freq); > >> > >> It is not proper to used 'freq' as the passed data. > >> In current step,'freq' is not adjusted and is not final decided > >> frequency. > > > > Right, the next revision will pass a struct devfreq_policy instead, > > where the notifiers can adjust the min/max values, similar to what > > cpufreq does. > > Actually, I don't know the devfreq_policy(?). As I already commented, > it is not proper to discuss it because there is no any real code and patches. > It is difficult to understand for me.
Re: [RFC PATCH] PM / devfreq: Add policy notifier
Hi, On 2018년 05월 18일 08:07, Matthias Kaehlcke wrote: > Hi, > > On Thu, May 17, 2018 at 11:01:34AM +0900, Chanwoo Choi wrote: >> Hi, >> >> Could you give some use-case of DEVFREQ_POLICY_NOTIFIER >> or send use-case patch with this patch? > > This is a WIP patch that makes use of the DEVFREQ_POLICY_NOTIFIER: > > https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/1065122 > >> I already knew the CPUFREQ_POLICY_NOTIFIER. >> But, until now, there are no any requirements of DEVFREQ_POLICY_NOTIFIER. >> If there are no any use-case, it is not necessary codes. > > Sure, I intend to land the above driver upstream if devfreq can > provide the necessary interfaces. I recommend that you should send the patch with the use-case patch. > >> On 2018년 05월 16일 06:24, Matthias Kaehlcke wrote: >>> diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c >>> index fe2af6aa88fc..a7294c056f65 100644 >>> --- a/drivers/devfreq/devfreq.c >>> +++ b/drivers/devfreq/devfreq.c >>> @@ -273,6 +273,9 @@ int update_devfreq(struct devfreq *devfreq) >>> if (err) >>> return err; >>> >>> + srcu_notifier_call_chain(&devfreq->policy_notifier_list, >>> + DEVFREQ_ADJUST, &freq); >> >> It is not proper to used 'freq' as the passed data. >> In current step,'freq' is not adjusted and is not final decided >> frequency. > > Right, the next revision will pass a struct devfreq_policy instead, > where the notifiers can adjust the min/max values, similar to what > cpufreq does. Actually, I don't know the devfreq_policy(?). As I already commented, it is not proper to discuss it because there is no any real code and patches. It is difficult to understand for me. > > Thanks > > Matthias > > > -- Best Regards, Chanwoo Choi Samsung Electronics
Re: [RFC PATCH] PM / devfreq: Add policy notifier
Hi, On Thu, May 17, 2018 at 11:01:34AM +0900, Chanwoo Choi wrote: > Hi, > > Could you give some use-case of DEVFREQ_POLICY_NOTIFIER > or send use-case patch with this patch? This is a WIP patch that makes use of the DEVFREQ_POLICY_NOTIFIER: https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/1065122 > I already knew the CPUFREQ_POLICY_NOTIFIER. > But, until now, there are no any requirements of DEVFREQ_POLICY_NOTIFIER. > If there are no any use-case, it is not necessary codes. Sure, I intend to land the above driver upstream if devfreq can provide the necessary interfaces. > On 2018년 05월 16일 06:24, Matthias Kaehlcke wrote: > > diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c > > index fe2af6aa88fc..a7294c056f65 100644 > > --- a/drivers/devfreq/devfreq.c > > +++ b/drivers/devfreq/devfreq.c > > @@ -273,6 +273,9 @@ int update_devfreq(struct devfreq *devfreq) > > if (err) > > return err; > > > > + srcu_notifier_call_chain(&devfreq->policy_notifier_list, > > + DEVFREQ_ADJUST, &freq); > > It is not proper to used 'freq' as the passed data. > In current step,'freq' is not adjusted and is not final decided > frequency. Right, the next revision will pass a struct devfreq_policy instead, where the notifiers can adjust the min/max values, similar to what cpufreq does. Thanks Matthias
Re: [RFC PATCH] PM / devfreq: Add policy notifier
Hi, Could you give some use-case of DEVFREQ_POLICY_NOTIFIER or send use-case patch with this patch? I already knew the CPUFREQ_POLICY_NOTIFIER. But, until now, there are no any requirements of DEVFREQ_POLICY_NOTIFIER. If there are no any use-case, it is not necessary codes. On 2018년 05월 16일 06:24, Matthias Kaehlcke wrote: > Policy notifiers are called before a frequency change and may adjust the > frequency, similar to CPUFREQ_ADJUST. The purpose of policy notifiers is > to support non-thermal throttling of devfreq devices. > > As of now notifiers may only select a lower frequency, but not increase it. > The reason for this limitation is that devfreq currently doesn't have an > actual policy for frequencies and OPPs > freq might be disabled by a > devfreq cooling device (see drivers/thermal/devfreq_cooling.c). > > Signed-off-by: Matthias Kaehlcke > --- > Marked as RFC since I'm aware that this isn't quite a 'policy' notifier, > but I also didn't want to make it too specific. Maybe devfreq devices should > have a policy similar to cpufreq? > > Should devfreq core be in charge of disabling/enabling OPPs instead of > the thermal cooling device? Typically we don't want to use higher OPPs > than those whitelisted by thermal, so in practice the current situation > shouldn't be much of an issue. > > drivers/devfreq/devfreq.c | 20 > include/linux/devfreq.h | 6 ++ > 2 files changed, 22 insertions(+), 4 deletions(-) > > diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c > index fe2af6aa88fc..a7294c056f65 100644 > --- a/drivers/devfreq/devfreq.c > +++ b/drivers/devfreq/devfreq.c > @@ -273,6 +273,9 @@ int update_devfreq(struct devfreq *devfreq) > if (err) > return err; > > + srcu_notifier_call_chain(&devfreq->policy_notifier_list, > + DEVFREQ_ADJUST, &freq); It is not proper to used 'freq' as the passed data. In current step,'freq' is not adjusted and is not final decided frequency. > + > /* >* Adjust the frequency with user freq, QoS and available freq. >* > @@ -640,6 +643,7 @@ struct devfreq *devfreq_add_device(struct device *dev, > devfreq->last_stat_updated = jiffies; > > srcu_init_notifier_head(&devfreq->transition_notifier_list); > + srcu_init_notifier_head(&devfreq->policy_notifier_list); > > mutex_unlock(&devfreq->lock); > > @@ -1414,7 +1418,7 @@ EXPORT_SYMBOL(devm_devfreq_unregister_opp_notifier); > * devfreq_register_notifier() - Register a driver with devfreq > * @devfreq: The devfreq object. > * @nb: The notifier block to register. > - * @list:DEVFREQ_TRANSITION_NOTIFIER. > + * @list:DEVFREQ_TRANSITION_NOTIFIER or DEVFREQ_POLICY_NOTIFIER. > */ > int devfreq_register_notifier(struct devfreq *devfreq, > struct notifier_block *nb, > @@ -1430,6 +1434,10 @@ int devfreq_register_notifier(struct devfreq *devfreq, > ret = srcu_notifier_chain_register( > &devfreq->transition_notifier_list, nb); > break; > + case DEVFREQ_POLICY_NOTIFIER: > + ret = srcu_notifier_chain_register( > + &devfreq->policy_notifier_list, nb); > + break; > default: > ret = -EINVAL; > } > @@ -1442,7 +1450,7 @@ EXPORT_SYMBOL(devfreq_register_notifier); > * devfreq_unregister_notifier() - Unregister a driver with devfreq > * @devfreq: The devfreq object. > * @nb: The notifier block to be unregistered. > - * @list:DEVFREQ_TRANSITION_NOTIFIER. > + * @list:DEVFREQ_TRANSITION_NOTIFIER or DEVFREQ_POLICY_NOTIFIER. > */ > int devfreq_unregister_notifier(struct devfreq *devfreq, > struct notifier_block *nb, > @@ -1458,6 +1466,10 @@ int devfreq_unregister_notifier(struct devfreq > *devfreq, > ret = srcu_notifier_chain_unregister( > &devfreq->transition_notifier_list, nb); > break; > + case DEVFREQ_POLICY_NOTIFIER: > + ret = srcu_notifier_chain_unregister( > + &devfreq->policy_notifier_list, nb); > + break; > default: > ret = -EINVAL; > } > @@ -1485,7 +1497,7 @@ static void devm_devfreq_notifier_release(struct device > *dev, void *res) > * @dev: The devfreq user device. (parent of devfreq) > * @devfreq: The devfreq object. > * @nb: The notifier block to be unregistered. > - * @list:DEVFREQ_TRANSITION_NOTIFIER. > + * @list:DEVFREQ_TRANSITION_NOTIFIER or DEVFREQ_POLICY_NOTIFIER. > */ > int devm_devfreq_register_notifier(struct device *dev, > struct devfreq *devfreq, > @@ -1521,7 +1533,7 @@ EXPORT_SYMBOL(devm_devfreq_register_notifier); > * @dev: The devfreq user device. (parent of devfreq) > * @devfreq: The devfreq object. > * @nb:
