On 6/24/24 15:24, Jonathan Cameron wrote:
> On Mon, 24 Jun 2024 11:47:39 -0700
> Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>
>> On 6/24/24 10:46, Sean Anderson wrote:
>> > Add labels from IIO channels to our channels. This allows userspace to
>> > display more meaningful names instead of "in0" or "temp5".
>> >
>> > Although lm-sensors gracefully handles errors when reading channel
>> > labels, the ABI says the label attribute
>> >
>> >> Should only be created if the driver has hints about what this voltage
>> >> channel is being used for, and user-space doesn't.
>> >
>> > Therefore, we test to see if the channel has a label before
>> > creating the attribute.
>> >
>>
>> FWIW, complaining about an ABI really does not belong into a commit
>> message. Maybe you and lm-sensors don't care about error returns when
>> reading a label, but there are other userspace applications which may
>> expect drivers to follow the ABI. Last time I checked, the basic rule
>> was still "Don't break userspace", and that doesn't mean "it's ok to
>> violate / break an ABI as long as no one notices".
>>
>> > Signed-off-by: Sean Anderson <sean.anderson@xxxxxxxxx>
>> > ---
>> >
>> > Changes in v2:
>> > - Check if the label exists before creating the attribute
>> >
>> > drivers/hwmon/iio_hwmon.c | 45 ++++++++++++++++++++++++++++++++++-----
>> > 1 file changed, 40 insertions(+), 5 deletions(-)
>> >
>> > diff --git a/drivers/hwmon/iio_hwmon.c b/drivers/hwmon/iio_hwmon.c
>> > index 4c8a80847891..5722cb9d81f9 100644
>> > --- a/drivers/hwmon/iio_hwmon.c
>> > +++ b/drivers/hwmon/iio_hwmon.c
>> > @@ -33,6 +33,17 @@ struct iio_hwmon_state {
>> > struct attribute **attrs;
>> > };
>> >
>> > +static ssize_t iio_hwmon_read_label(struct device *dev,
>> > + struct device_attribute *attr,
>> > + char *buf)
>> > +{
>> > + struct sensor_device_attribute *sattr = to_sensor_dev_attr(attr);
>> > + struct iio_hwmon_state *state = dev_get_drvdata(dev);
>> > + struct iio_channel *chan = &state->channels[sattr->index];
>> > +
>> > + return iio_read_channel_label(chan, buf);
>> > +}
>> > +
>>
>> I personally find it a bit kludgy that an in-kernel API would do a
>> sysfs write like this and expect a page-aligned buffer as parameter,
>> but since Jonathan is fine with it:
>
> That's a good point that I'd not picked up on and it probably makes sense
> to address that before it bites us on some other subsystem.
>
> It was more reasonable when the only path was to a light wrapper that went
> directly around the sysfs callback. Now we are wrapping these up for more
> general use we should avoid that restriction.
>
> Two approaches to that occur to me.
> 1) Fix up read_label() everywhere to not use sysfs_emit and take a size
> of the buffer to print into. There are only 11 implementations so
> far so this should be straight forward.
This API is the same as the existing iio_read_channel_ext_info. It is
used for the same purpose: forwarding sysfs reads/writes from one
device to another (see e.g. iio-mux and iio-rescale). ext_info is used
by around 85 drivers, so it is not so trivial to change the API. While I
agree that the current API is unusual, it's not too bad given that we
get the same guarantees from device_attribute.show.
--Sean
> 2) Add a bounce buffer so we emit into a suitable size for sysfs_emit()
> then reprint from there into a buffer provided via this interface with
> the appropriate size provided. This one is clunky and given the relatively
> few call sits I think fixing it via option 1 is the better route forwards.
