[PATCH] PM / devfreq: Use freq_table for available_frequencies

[PATCH] PM / devfreq: Use freq_table for available_frequencies

[Saravana Kannan]

On 04/14/2014 11:30 PM, ??? wrote:
>> On 04/14/2014 06:36 PM, ??? wrote:
>>>> MyungJoo/Kyungmin,
>>>>
>>>> Bump. Can we accept this patch please?
>>>>
>>>> -Saravana
>>>
>>> Nack.
>>>
>>> Please note that freq_table is also an optional value, which may
>>> be null.
>>
>> Ah, I saw that the max_freq would be zero if freq_table was NULL and I
>> assumed that it can't be NULL. But I see that the max_freq limit is not
>> applied if it's zero. Thanks for catching it.
>>
>>> Besides, please be aware that your code is under rcu_read_lock().
>>
>> Valid point. I was just trying to keep the diff simple. No one's really
>> going to be catting this file often when performance matters.
>>
>>>
>>>
>>> Cheers,
>>> MyungJoo.
>>>
>>> ps. I'll send a related patch (avoid accessing null but not-an-error
>>> pointer at other sysfs nodes). Thank you for letting me catch such bugs anyway.
>>
>> I can go ahead and do this myself if you don't mind.
> 
> No, we don't need it. It was a false alarm.
> Reading again, I've found that we've already made other sysfs nodes
> check if either freq_table is null or its size is 0.
> 
> So, we only need to look at this available_frequencies node now.
> 
> I'll add some notes on the ABI doc for available_frequencies soon.
> 

Ah, I misunderstood your previous email. I thought you Nack-ed my patch
and decided to send your own patch to replace mine. Ok, I'll fix up mine
and send it out.

-Saravana


-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation

[PATCH] PM / devfreq: Use freq_table for available_frequencies

[Saravana Kannan]

On 04/15/2014 11:41 AM, Saravana Kannan wrote:
> Ah, I misunderstood your previous email. I thought you Nack-ed my patch
> and decided to send your own patch to replace mine. Ok, I'll fix up mine
> and send it out.

MyungJoo/Kyungmin,

I sent out an updated patch. Can you please take a look?

Thanks,
Saravana


-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation

[PATCH v2] PM / devfreq: Use freq_table for available_frequencies

[Saravana Kannan]

Some devices use freq_table instead of OPP. For those devices, the
available_frequencies file shows up empty. Fix that by using freq_table to
generate the available_frequencies data when it's available.

OPP find frequency APIs also skips frequencies that have been temporarily
disabled (say, due to thermal, etc). Since available_frequencies is
supposed to show the entire list of available frequencies without taking
temporary limits into consideration, preference is given to freq_table when
available.

Signed-off-by: Saravana Kannan <skannan@codeaurora.org>
---
 drivers/devfreq/devfreq.c | 29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)

diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c
index 2042ec3..527cbe2 100644
--- a/drivers/devfreq/devfreq.c
+++ b/drivers/devfreq/devfreq.c
@@ -912,20 +912,27 @@ static ssize_t available_frequencies_show(struct device *d,
 	struct devfreq *df = to_devfreq(d);
 	struct device *dev = df->dev.parent;
 	struct dev_pm_opp *opp;
+	unsigned int i = 0;
 	ssize_t count = 0;
 	unsigned long freq = 0;
 
-	rcu_read_lock();
-	do {
-		opp = dev_pm_opp_find_freq_ceil(dev, &freq);
-		if (IS_ERR(opp))
-			break;
-
-		count += scnprintf(&buf[count], (PAGE_SIZE - count - 2),
-				   "%lu ", freq);
-		freq++;
-	} while (1);
-	rcu_read_unlock();
+	if (df->profile->freq_table) {
+		for (i = 0; i < df->profile->max_state; i++)
+			count += scnprintf(&buf[count], (PAGE_SIZE - count - 2),
+					   "%u ", df->profile->freq_table[i]);
+	} else {
+		rcu_read_lock();
+		do {
+			opp = dev_pm_opp_find_freq_ceil(dev, &freq);
+			if (IS_ERR(opp))
+				break;
+
+			count += scnprintf(&buf[count], (PAGE_SIZE - count - 2),
+					   "%lu ", freq);
+			freq++;
+		} while (1);
+		rcu_read_unlock();
+	}
 
 	/* Truncate the trailing space */
 	if (count)
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation

[PATCH v2] PM / devfreq: Use freq_table for available_frequencies

[Saravana Kannan]

On 04/27/2014 06:41 PM, MyungJoo Ham wrote:
>> Some devices use freq_table instead of OPP. For those devices, the
>> available_frequencies file shows up empty. Fix that by using freq_table to
>> generate the available_frequencies data when it's available.
>>
>> OPP find frequency APIs also skips frequencies that have been temporarily
>> disabled (say, due to thermal, etc). Since available_frequencies is
>> supposed to show the entire list of available frequencies without taking
>> temporary limits into consideration, preference is given to freq_table when
>> available.
>>
>> Signed-off-by: Saravana Kannan <skannan@codeaurora.org>
>> ---
>>   drivers/devfreq/devfreq.c | 29 ++++++++++++++++++-----------
>>   1 file changed, 18 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c
>> index 2042ec3..527cbe2 100644
>> --- a/drivers/devfreq/devfreq.c
>> +++ b/drivers/devfreq/devfreq.c
>> @@ -912,20 +912,27 @@ static ssize_t available_frequencies_show(struct device *d,
>>   	struct devfreq *df = to_devfreq(d);
>>   	struct device *dev = df->dev.parent;
>>   	struct dev_pm_opp *opp;
>> +	unsigned int i = 0;
>>   	ssize_t count = 0;
>>   	unsigned long freq = 0;
>>   
>> -	rcu_read_lock();
>> -	do {
>> -		opp = dev_pm_opp_find_freq_ceil(dev, &freq);
>> -		if (IS_ERR(opp))
>> -			break;
>> -
>> -		count += scnprintf(&buf[count], (PAGE_SIZE - count - 2),
>> -				   "%lu ", freq);
>> -		freq++;
>> -	} while (1);
>> -	rcu_read_unlock();
>> +	if (df->profile->freq_table) {
>> +		for (i = 0; i < df->profile->max_state; i++)
>> +			count += scnprintf(&buf[count], (PAGE_SIZE - count - 2),
>> +					   "%u ", df->profile->freq_table[i]);
> 
> You are hereby changing the semmantics of the original
> available_frequencies node.
> 
> When a frequency/voltage pair has been disabled (opp_disable), probably
> by opp_disable(), the frequency is no more "available".
> However, when the driver author supplied freq_table as well as OPP
> in order to see the statistics, the node will behave differently.
> 
> Please do not affect the current users as long as it does not give
> additional benefit or fix a bug.

I was actually trying to stick with the semantics as it was documented.
The documentation for this file says it'll show frequencies that are not
allowed by the current min/max settings either. To me, an OPP disable
seems similar to some frequencies "disabled" by min/max settings.

Giving preference to OPP is not a hard change to do, but it seems to go
againsts the documented semantics.

Thoughts?

-Saravana


-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation

[PATCH v2] PM / devfreq: Use freq_table for available_frequencies

[Saravana Kannan]

On 04/29/2014 01:00 PM, Saravana Kannan wrote:
> On 04/27/2014 06:41 PM, MyungJoo Ham wrote:
>> You are hereby changing the semmantics of the original
>> available_frequencies node.
>>
>> When a frequency/voltage pair has been disabled (opp_disable), probably
>> by opp_disable(), the frequency is no more "available".
>> However, when the driver author supplied freq_table as well as OPP
>> in order to see the statistics, the node will behave differently.
>>
>> Please do not affect the current users as long as it does not give
>> additional benefit or fix a bug.
> 
> I was actually trying to stick with the semantics as it was documented.
> The documentation for this file says it'll show frequencies that are not
> allowed by the current min/max settings either. To me, an OPP disable
> seems similar to some frequencies "disabled" by min/max settings.
> 
> Giving preference to OPP is not a hard change to do, but it seems to go
> againsts the documented semantics.
> 
> Thoughts?

I'll send out another patch like you wanted -- with OPP being given
preference over freq_table when listing frequencies.

But I would still like to hear your thoughts. As of today, there's no
clean way to get the complete list of available frequencies that would
give a consistent output irrespective of the temporary limits/conditions
imposed by thermal, current limiting, etc. The round about way is to cat
trans_stat and parse the frequencies from that.

That's why I was trying to give preference to freq_table.

Thanks,
Saravana

-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation

[PATCH v2] PM / devfreq: Use freq_table for available_frequencies

[Saravana Kannan]

On 05/08/2014 03:02 AM, MyungJoo Ham wrote:
>> On 04/29/2014 01:00 PM, Saravana Kannan wrote:
>>> On 04/27/2014 06:41 PM, MyungJoo Ham wrote:
>>>> You are hereby changing the semmantics of the original
>>>> available_frequencies node.
>>>>
>>>> When a frequency/voltage pair has been disabled (opp_disable), probably
>>>> by opp_disable(), the frequency is no more "available".
>>>> However, when the driver author supplied freq_table as well as OPP
>>>> in order to see the statistics, the node will behave differently.
>>>>
>>>> Please do not affect the current users as long as it does not give
>>>> additional benefit or fix a bug.
>>>
>>> I was actually trying to stick with the semantics as it was documented.
>>> The documentation for this file says it'll show frequencies that are not
>>> allowed by the current min/max settings either. To me, an OPP disable
>>> seems similar to some frequencies "disabled" by min/max settings.
>>>
>>> Giving preference to OPP is not a hard change to do, but it seems to go
>>> againsts the documented semantics.
>>>
>>> Thoughts?
>>
>> I'll send out another patch like you wanted -- with OPP being given
>> preference over freq_table when listing frequencies.
>>
>> But I would still like to hear your thoughts. As of today, there's no
>> clean way to get the complete list of available frequencies that would
>> give a consistent output irrespective of the temporary limits/conditions
>> imposed by thermal, current limiting, etc. The round about way is to cat
>> trans_stat and parse the frequencies from that.
>>
>> That's why I was trying to give preference to freq_table.
>>
>> Thanks,
>> Saravana
> 
> The node, available_frequencies, was suggested before freq_table concept.
> At that time, available_frequencies was supposed to show the list of
> available OPP lists for those who use OPP for devfreq device, excluding
> those disabled by OPP. (OPP lists are external to devfreq and devfreq's
> min/max are internal to devfreq)
> 
> Locally, this node has been used to debug the behavior of a devfreq device.
> With min/max nodes, we know the range while we cannot (easily at shell)
> see which OPP points are available at the moment, where we have been able
> to use available_frequencies.
> 
> We do not want to lose such capavility as long as we do not have OPP sysfs
> automatically assigned to any OPP lists. If I remember correctly, we don't
> have it, yet.
> 
> 
> A. I want to minimize semantics changes in sysfs. Adding another without
> interfering with previous usage is ok.
> B. (more importantly) I don't want to lose the debugging capabilities.
> 

Thanks for the response MyungJoo. Makes sense. I was also discussing
this internally and was considering a "possible_frequencies" similar to
available vs possible CPUs.

I'll make such a patch for that and send it out. In that case, I'll
probably leave "available_frequencies" alone.

-Saravana

-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
hosted by The Linux Foundation