[PATCH v3] hwmon: Versatile Express hwmon driver
Guenter Roeck
linux at roeck-us.net
Fri Sep 21 11:54:47 EDT 2012
On Fri, Sep 21, 2012 at 04:38:45PM +0100, Pawel Moll wrote:
> On Fri, 2012-09-21 at 00:57 +0100, Guenter Roeck wrote:
> > Your patch is substantially corrupted and unusable as patch file. Did you send
> > it through outlook ?
>
> No, it's normal git-send-email, but it goes through Exchange mail server
> which can do stupid things with the message body (ie. change it!). Sorry
> about it, I'll send the next version from my private account to avoid
> such problems.
>
> > > +static ssize_t vexpress_hwmon_name_show(struct device *dev,
> > > + struct device_attribute *dev_attr, char *buffer)
> > > +{
> > > + const char *compatible = of_get_property(dev->of_node, "compatible",
> > > + NULL);
> > > +
> > Can compatible be NULL ?
>
> No, compatible value makes the device bound with the driver. Nothing of
> this would happen without correct compatible string. And the device will
> be never successfully probed without DT node "behind it" (the
> of_match_device() below will return NULL).
>
> > > + return sprintf(buffer, "%s\n", compatible);
> > > +}
> > > +
> > > +static ssize_t vexpress_hwmon_label_show(struct device *dev,
> > > + struct device_attribute *dev_attr, char *buffer)
> > > +{
> > > + const char *label = of_get_property(dev->of_node, "label", NULL);
> > > +
> >
> > Can label be NULL (eg if missing in device tree data) ?
> > What happens if it is NULL ?
>
> It is printed as "(null)".
>
> > Seems to me the label attribute should not exist in that case.
>
> I can do that, but I find the lack of label an error - essentially there
> is no other way of understanding the meaning of a value. The order of
> devices can be random, so no label - no idea what the value is.
>
There are lots of unlabeled sensors in hwmon. The context is always board
specific. Following your line of argument, you are declaring each sensor w/o
label to be at error. Besides, one does know what the value is - a voltage in mV,
current in mA, power in uW, and so on. The knowledge _which_ voltage or current
is reflected can be configured in sensors.conf and does not have to be provided
in devicetree. That you want it to be there doesn't make it an error if it
isn't.
I won't accept the lack of an error check and the "(null)" output. Refuse to generate
the device if you like, or don't generate a label, or have the function return an error,
but you can not depend on the printf function fixing your problems.
> I will make the label property "required" in the binding documentation.
>
> > > +#define VEXPRESS_HWMON_ATTR(_name, _show, _divisor) \
> > > +struct vexpress_hwmon_attr vexpress_hwmon_attr_##_name = { \
> > > + .dev_attr = __ATTR(_name, S_IRUGO, _show, NULL), \
> > > + .divisor = _divisor, \
> > > +}
> >
> > You should be able to use SENSOR_DEVICE_ATTR here.
>
> I didn't want to "overload" the index value for other reason, but if you
> think it's better approach I'll do that.
>
> > > +static struct of_device_id vexpress_hwmon_of_match[] = {
> > > +#if !defined(CONFIG_REGULATOR_VEXPRESS)
> > > + {
> > > + .compatible = "arm,vexpress-volt",
> > > + .data = &vexpress_hwmon_group_volt,
> > > + },
> > > +#endif
> >
> > Just wondering - is the hwmon driver mutually exclusive with the regulator
> > driver, or can both coexist ?
>
> The device model will not let one device to be bound to two different
> drivers, so there will be a race for the device, depending on the
> initcall ordering... And the regulator driver must have "higher
> priority" then hwmon.
>
> Having said that, it seems that we could have a "generic" hwmon
> "regulator sensor" driver (as in: in1_input doing
> regulator_get_voltage()), as I think a regulator can be shared between
> consumers under certain conditions (I'd have to check this, though).
>
> > FWIW, the references to "arm,vexpress-volt" I can find on the web all do not
> > include "label", so I think you'll really have to look into what happens if
> > there is no such property.
>
> Yes, the patches you saw were prepared for the previous version of the
> driver. Last night when testing the code I spotted the (null) labels and
> updated the DTS files so the all the "volts" already have labels (I
> haven't reposted the Device Tree patches yet).
>
> > > + {
> > > + .compatible = "arm,vexpress-amp",
> > > + .data = &vexpress_hwmon_group_amp,
> > > + }, {
> > > + .compatible = "arm,vexpress-temp",
> > > + .data = &vexpress_hwmon_group_temp,
> > > + }, {
> > > + .compatible = "arm,vexpress-power",
> > > + .data = &vexpress_hwmon_group_power,
> > > + }, {
> > > + .compatible = "arm,vexpress-energy",
> > > + .data = &vexpress_hwmon_group_energy,
> > > + },
> > > + {}
> > > +};
> > > +MODULE_DEVICE_TABLE(of, vexpress_hwmon_of_match);
> > > +
> >
> > Just wondering .... would the following work ?
> >
> > hwmon at 0 {
> > compatible = "arm,vexpress-hwmon";
> > volt at 0 {
> > /* Logic level voltage */
> > arm,vexpress-sysreg,func = <2 0>;
> > label = "VIO";
> > };
> >
> > amp at 0 {
> > /* Total current for the two cores */
> > arm,vexpress-sysreg,func = <3 0>;
> > label = "Core current";
> > };
> >
> > temp at 0 {
> > /* DCC internal temperature */
> > arm,vexpress-sysreg,func = <4 0>;
> > label = "DCC";
> > };
> >
> > power at 0 {
> > /* Total power */
> > arm,vexpress-sysreg,func = <12 0>;
> > label = "Core power";
> > };
> >
> > energy at 0 {
> > /* Total energy */
> > arm,vexpress-sysreg,func = <13 0>;
> > label = "Core energy";
> > };
> > };
> >
> > I am not a device tree expert, but it seems to me that you would only
> > have a single device node with this approach, and you could parse its
> > children with for_each_child_of_node().
>
> The thing is that there are other config devices, non hwmon relevant. In
> particular: clock generators ("oscillators"), reset controllers, DVI
> multiplexers etc. etc. And they are treated equally, as simple platform
> devices. And the -volt case is a good example why...
>
Hmm ... I don't get the logic, but I'll leave that up to you.
> > > +static int vexpress_hwmon_probe(struct platform_device *pdev)
> > > +{
> > > + int err;
> > > + const struct of_device_id *match;
> > > + struct vexpress_hwmon_data *data;
> > > +
> > > + data = devm_kzalloc(&pdev->dev, sizeof(*data), GFP_KERNEL);
> > > + if (!data) {
> > > + err = -ENOMEM;
> > > + goto error_kzalloc;
> > > + }
> > > +
> > > + match = of_match_device(vexpress_hwmon_of_match, &pdev->dev);
> > > + if (!match) {
> > > + err = -EINVAL;
> >
> > ENODEV ?
>
> Frankly speaking the "/* Invalid argument */" seems more appropriate
> here, but can do, no problem.
>
> > > + goto error_match_device;
> > > + }
> > > +
> > > + data->func = vexpress_config_func_get_by_dev(&pdev->dev);
> > > + if (!data->func) {
> > > + err = -ENXIO;
> >
> > ENODEV ?
>
> Again, it's "/* No such device or address */" vs "/* No such device */".
> Both fine with me, will do.
>
> > > + goto error_get_func;
> > > + }
> > > +
> > > + err = sysfs_create_group(&pdev->dev.kobj, match->data);
> > > + if (err)
> > > + goto error_create_group;
> > > +
> > > + data->hwmon_dev = hwmon_device_register(&pdev->dev);
> > > + if (IS_ERR(data->hwmon_dev)) {
> > > + err = PTR_ERR(data->hwmon_dev);
> > > + goto error_hwmon_device_register;
> > > + }
> > > +
> > > + platform_set_drvdata(pdev, data);
> > > +
> > Must be set before sysfs files are created.
>
> Sure thing, thanks for spotting this.
>
> > > + return 0;
> > > +
> > > +error_hwmon_device_register:
> > > + sysfs_remove_group(&pdev->dev.kobj, match->data);
> > > +error_create_group:
> > > + vexpress_config_func_put(data->func);
> > > +error_get_func:
> > > +error_match_device:
> > > +error_kzalloc:
> >
> > Just return the error instead of using goto. No need for all those goto labels.
>
> I've heard exactly the opposite number of times (usually pointing me at
> CodingStyle ;-) but no problem, will do.
>
Good point. Have a look into CodingStyle, and when the argument comes up again,
point to the example in Chapter 7, which does return immediately if there is
nothing else to do.
Guenter
More information about the linux-arm-kernel
mailing list