Re: [Patch v4 4/7] thermal: Add generic power domain warming device driver.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



[...]

> >> +static struct thermal_cooling_device_ops pd_warming_device_ops = {
> >> +       .get_max_state  = ::pd_wdev_get_max_state,
> >> +       .get_cur_state  = pd_wdev_get_cur_state,
> >> +       .set_cur_state  = pd_wdev_set_cur_state,
> >> +};
> >> +
> >> +struct thermal_cooling_device *
> >> +pwr_domain_warming_register(struct device *parent, char *pd_name, int pd_id)
> >
> > Maybe rename this to: thermal_of_pd_warming_register()
>
> How about pd_of_warming_register? It is consistent with other cooling
> device register like cpuidle_of_cooling_register and
> cpufreq_of_cooling_register.

Well, we actually have the following:
of_devfreq_cooling_register()
of_cpufreq_cooling_register()
cpuidle_of_cooling_register()

So maybe this is the most consistent. :-)
of_pd_warming_register()

>
> > Moreover, I think you could replace the "struct device *parent", with
> > a "struct device_node *node" as in-parameter. That's all you need,
> > right?
>
> You mean pd_wdev->dev.parent need not be populated ? The device
> in this case will be created under /sys/devices which I do not think
> is the correct.

Good point!

> With a parent device specified, the power controller that resides the
> power domain that can act as the warming dev, becomes the parent of the
> warming dev. In case of this patch series, for the mx warming dev,
> 179c0000.rsc/179c0000.rsc\:power-controller/ becomes the parent.(The
> device will be created under
> /sys/devices/platform/soc\@0/179c0000.rsc/179c0000.rsc\:power-controller/)
>
> Other way might be to register the warming device under virtual devices
> as a new category of devices.

No, that sounds wrong.

Another option is to create a specific bus type for these new
pd_warming devices. But I admit that sounds a bit too much, let's
assign a parent.

>
> I prefer to keep it as a child of the power controller device, but I am
> open to explore other options and to re-do this bit. What do you think?

Sure, sorry for the noise.

>
> >
> >> +{
> >> +       struct pd_warming_device *pd_wdev;
> >> +       struct of_phandle_args pd_args;
> >> +       int ret;
> >> +
> >> +       pd_wdev = kzalloc(sizeof(*pd_wdev), GFP_KERNEL);
> >> +       if (!pd_wdev)
> >> +               return ERR_PTR(-ENOMEM);
> >> +
> >> +       dev_set_name(&pd_wdev->dev, "%s_warming_dev", pd_name);
> >
> > Perhaps skip the in-param *pd_name and make use of the suggested
> > "struct device_node *node", the index and something with "warming...",
> > when setting the name.
>
> Won't the index have to be in-param in this case ?

Isn't that already the case?

Huh, long time since I reviewed this.

>
> >
> > Just an idea, as to simplify for the caller.
> >
> >> +       pd_wdev->dev.parent = parent;
> >
> > This isn't needed, I think.

So ignore this comment, as discussed above.

> >
> >> +
> >> +       ret = device_register(&pd_wdev->dev);
> >> +       if (ret)
> >> +               goto error;
> >> +
> >> +       pd_args.np = parent->of_node;
> >> +       pd_args.args[0] = pd_id;
> >> +       pd_args.args_count = 1;
> >> +
> >> +       ret = of_genpd_add_device(&pd_args, &pd_wdev->dev);
> >> +
> >
> > White space.
>
> Will fix it.
>
> >
> >> +       if (ret)
> >> +               goto error;
> >> +
> >> +       ret = dev_pm_genpd_performance_state_count(&pd_wdev->dev);
> >> +       if (ret < 0)
> >> +               goto error;
> >> +
> >> +       pd_wdev->max_state = ret - 1;
> >> +       pm_runtime_enable(&pd_wdev->dev);
> >> +       pd_wdev->runtime_resumed = false;
> >> +
> >> +       pd_wdev->cdev = thermal_of_cooling_device_parent_register
> >> +                                       (NULL, parent, pd_name, pd_wdev,
> >> +                                        &pd_warming_device_ops);
> >
> > As stated in patch3, I don't get it why you need to use this new API
> > for "parents".
>
> I agree with you. I cannot re-collect my thought process for this API.
> I compiled and tested using the regular API and everything works fine.
> I will drop patch 3 and use the thermal_of_cooling_device_register here.

Great, one confusing piece seems to go away then. :-)

>
> >
> >> +       if (IS_ERR(pd_wdev->cdev)) {
> >> +               pr_err("unable to register %s cooling device\n", pd_name);
> >> +               pm_runtime_disable(&pd_wdev->dev);
> >> +               ret = PTR_ERR(pd_wdev->cdev);
> >> +               goto error;
> >> +       }
> >> +
> >> +       return pd_wdev->cdev;
> >> +error:
> >> +       put_device(&pd_wdev->dev);
> >
> > If device_register() succeeds you need to call device_unregister(),
> > rather than put_device() as a part of the error handling.
>
> Will fix this.
>
> >
> >> +       kfree(pd_wdev);
> >
> > You need a ->release() callback to manage kfree(), after you called
> > device_register().
>
> mm?? I did not get this. What release callback? You mean for power
> controller driver to call ?

No, this how life cycle management of devices should be implemented.

Have a look at genpd_release_dev() - and see how that is being used
for genpd's virtual devices, that should explain more.

>
> >
> >> +       return ERR_PTR(ret);
> >
> > Another thing is missing in the error path, which is to remove the
> > device for the genpd. I think calling pm_genpd_remove_device() should
> > work fine here.
>
> I will fix this. I am thinking this will be be needed in
> pwr_domain_warming_unregister as well.

Yep.

Kind regards
Uffe



[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux