Message-ID: <3d27029b-caef-45a1-88bf-98f78e6d6a61@redhat.com>
Date: Mon, 5 Feb 2024 12:43:54 +0100
From: Hans de Goede <hdegoede@...hat.com>
To: Thomas Weißschuh <linux@...ssschuh.net>,
 Sebastian Reichel <sre@...nel.org>, Konrad Dybcio <konradybcio@...nel.org>
Cc: linux-pm@...r.kernel.org, linux-kernel@...r.kernel.org,
 Sebastian Reichel <sebastian.reichel@...labora.com>
Subject: Re: [PATCH 1/4] power: supply: core: fix charge_behaviour formatting

Hi,

On 2/5/24 10:52, Hans de Goede wrote:
> Hi Thomas,
> 
> Thank you for your patches for this.
> 
> On 2/4/24 18:26, Thomas Weißschuh wrote:
>> This property is documented to have a special format which exposes all
>> available behaviours and the currently active one at the same time.
>> For this special format some helpers are provided.
>>
>> However the default property logic in power_supply_sysfs.c is not using
>> the helper and the default logic only prints the currently active
>> behaviour.
>>
>> Adjust power_supply_sysfs.c to follow the documented format.
>>
>> There are currently two in-tree drivers exposing charge behaviours:
>> thinkpad_acpi and mm8013.
>> thinkpad_acpi is not affected by the change, as it directly uses the
>> helpers and does not use the power_supply_sysfs.c logic.
>>
>> As mm8013 does not implement POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR_AVAILABLE
>> the new logic will preserve the simple output format in this case.
>>
>> Fixes: 1b0b6cc8030d ("power: supply: add charge_behaviour attributes")
>> Signed-off-by: Thomas Weißschuh <linux@...ssschuh.net>
>> ---
>>  drivers/power/supply/power_supply_sysfs.c | 32 +++++++++++++++++++++++++++++++
>>  include/linux/power_supply.h              |  1 +
>>  2 files changed, 33 insertions(+)
>>
>> diff --git a/drivers/power/supply/power_supply_sysfs.c b/drivers/power/supply/power_supply_sysfs.c
>> index 977611e16373..3680cfc2e908 100644
>> --- a/drivers/power/supply/power_supply_sysfs.c
>> +++ b/drivers/power/supply/power_supply_sysfs.c
>> @@ -271,6 +271,32 @@ static ssize_t power_supply_show_usb_type(struct device *dev,
>>  	return count;
>>  }
>>  
>> +static ssize_t power_supply_show_charge_behaviour(struct device *dev,
>> +						  struct power_supply *psy,
>> +						  struct power_supply_attr *ps_attr,
>> +						  union power_supply_propval *value,
>> +						  char *buf)
>> +{
>> +	union power_supply_propval available;
>> +	int ret;
>> +
>> +	ret = power_supply_get_property(psy,
>> +					POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR,
>> +					value);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	ret = power_supply_get_property(psy,
>> +					POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR_AVAILABLE,
>> +					&available);
>> +	if (ret == -EINVAL)
>> +		return sysfs_emit(buf, "%s\n", ps_attr->text_values[value->intval]);
>> +	else if (ret < 0)
> 
> No need for "else if" here since the if above does a return. So you can just do:
> 
> 	if (ret < 0)
> 		return ret;
> 
> 
>> +		return ret;
>> +
>> +	return power_supply_charge_behaviour_show(dev, available.intval, value->intval, buf);
>> +}
>> +
>>  static ssize_t power_supply_show_property(struct device *dev,
>>  					  struct device_attribute *attr,
>>  					  char *buf) {
>> @@ -282,6 +308,8 @@ static ssize_t power_supply_show_property(struct device *dev,
>>  
>>  	if (psp == POWER_SUPPLY_PROP_TYPE) {
>>  		value.intval = psy->desc->type;
>> +	} else if (psp == POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR) {
>> +		value.intval = -1;
>>  	} else {
>>  		ret = power_supply_get_property(psy, psp, &value);
>>  
> 
> I'm not a fan of this, I guess that you are doing this because you do not
> want to enter this if:
> 
>         if (ps_attr->text_values_len > 0 &&
>             value.intval < ps_attr->text_values_len && value.intval >= 0) {
>                 return sysfs_emit(buf, "%s\n", ps_attr->text_values[value.intval]);
>         }
> 
> But by doing this you add both the special case of setting value.intval = -1
> here and you now need to do the power_supply_get_property() which is in the else
> manually in power_supply_show_charge_behaviour() and the error handling / logging
> of power_supply_get_property() in power_supply_show_charge_behaviour() is different.
> 
> What I think you can (and should) do here instead is move:
> 
>         if (ps_attr->text_values_len > 0 &&
>             value.intval < ps_attr->text_values_len && value.intval >= 0) {
>                 return sysfs_emit(buf, "%s\n", ps_attr->text_values[value.intval]);
>         }
> 
> into the default case of the switch (psp) {} below it in a preparation patch.
> 
> This if is never entered for the 2 non default cases in that switch, so it
> is safe to move it into the default case, e.g. something like this:
> 
> 	default:
> 	        if (ps_attr->text_values_len > 0 &&
>         	    value.intval < ps_attr->text_values_len && value.intval >= 0)
>                 	ret = sysfs_emit(buf, "%s\n", ps_attr->text_values[value.intval]);
> 		else
> 			ret = sysfs_emit(buf, "%d\n", value.intval);
> 
> I think that also actually makes things a bit cleaner then the current
> early-exit for printing enum values.
> 
> And then in the next patch you can just add:
> 
> 	case POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR:
> 		ret = power_supply_show_charge_behaviour(dev, psy, ps_attr,
> 							 &value, buf);
> 		break;
> 
> Without needing to set intval = -1 and without needing to get the value
> inside power_supply_show_charge_behaviour() since that will already
> be done for you then.

One more note on this. With the suggested preparation patch to move
the if checking for ps_attr->text_values into the default case,
you can then also properly turn POWER_SUPPLY_PROP_USB_TYPE into an
enum too:

diff --git a/drivers/power/supply/power_supply_sysfs.c b/drivers/power/supply/power_supply_sysfs.c
index 977611e16373..768df64330f4 100644
--- a/drivers/power/supply/power_supply_sysfs.c
+++ b/drivers/power/supply/power_supply_sysfs.c
@@ -209,7 +209,7 @@ static struct power_supply_attr power_supply_attrs[] = {
 	POWER_SUPPLY_ATTR(TIME_TO_FULL_NOW),
 	POWER_SUPPLY_ATTR(TIME_TO_FULL_AVG),
 	POWER_SUPPLY_ENUM_ATTR(TYPE),
-	POWER_SUPPLY_ATTR(USB_TYPE),
+	POWER_SUPPLY_ENUM_ATTR(USB_TYPE),
 	POWER_SUPPLY_ENUM_ATTR(SCOPE),
 	POWER_SUPPLY_ATTR(PRECHARGE_CURRENT),
 	POWER_SUPPLY_ATTR(CHARGE_TERM_CURRENT),

IOW you can now set ps_attr->text_values* for POWER_SUPPLY_PROP_USB_TYPE
too, without needing to worry this causing power_supply_show_usb_type()
no longer to get called.

The reason I'm mentioning this is because power_supply_show_usb_type()
and power_supply_charge_behaviour_show() show a lot of code-duplication.

And I think that we can have one generic function replacing both
by using ps_attr->text_values* instead of hardcoding POWER_SUPPLY_USB_TYPE_TEXT
resp. power_supply_charge_behaviour_show (and yes this contradicts
my earlier comment on patch 4/4).

Although at a closer look I see now that the usb-types code uses
a list of supported type enum values in the power_supply_desc,
where as the charge_behavior code uses a bitmask which you retrieve
through a new property.

Thinking more about this, I think that adding a fake
POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR_AVAILABLE property as you do here
might actually not be the best idea. All the other properties
are visible in sysfs, so this one is a bit weird. I think it might
be better to add this info to power_supply_desc like how this is
done for the usb-types.

And:
        const enum power_supply_usb_type *usb_types;
        size_t num_usb_types;

Should probably be converted into a bitmask too. I checked
and there are not that many users of this.

Once we have that as a bitmask too, we can refactor
power_supply_show_usb_type() and power_supply_charge_behaviour_show()
into a single new helper.

But I believe that refactoring:

        const enum power_supply_usb_type *usb_types;
        size_t num_usb_types;

into a bitmask is out of scope for this series. So I guess
that for now just add the bitmask of available charge behaviors
to power_supply_desc rather then making it a fake property
and then in the future we can do the refactor of usb-type
to a bitmask and remove the code duplication between
power_supply_show_usb_type() and power_supply_charge_behaviour_show()

Sebastian, any comments ?

Regards,

Hans









>> @@ -308,6 +336,10 @@ static ssize_t power_supply_show_property(struct device *dev,
>>  		ret = power_supply_show_usb_type(dev, psy->desc,
>>  						&value, buf);
>>  		break;
>> +	case POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR:
>> +		ret = power_supply_show_charge_behaviour(dev, psy, ps_attr,
>> +							 &value, buf);
>> +		break;
>>  	case POWER_SUPPLY_PROP_MODEL_NAME ... POWER_SUPPLY_PROP_SERIAL_NUMBER:
>>  		ret = sysfs_emit(buf, "%s\n", value.strval);
>>  		break;
>> diff --git a/include/linux/power_supply.h b/include/linux/power_supply.h
>> index c0992a77feea..9a6e6b488164 100644
>> --- a/include/linux/power_supply.h
>> +++ b/include/linux/power_supply.h
>> @@ -135,6 +135,7 @@ enum power_supply_property {
>>  	POWER_SUPPLY_PROP_CHARGE_CONTROL_START_THRESHOLD, /* in percents! */
>>  	POWER_SUPPLY_PROP_CHARGE_CONTROL_END_THRESHOLD, /* in percents! */
>>  	POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR,
>> +	POWER_SUPPLY_PROP_CHARGE_BEHAVIOUR_AVAILABLE,
>>  	POWER_SUPPLY_PROP_INPUT_CURRENT_LIMIT,
>>  	POWER_SUPPLY_PROP_INPUT_VOLTAGE_LIMIT,
>>  	POWER_SUPPLY_PROP_INPUT_POWER_LIMIT,
>>
> 

Powered by blists - more mailing lists