Re: [PATCH V4 8/8] cpufreq: Add Rust based cpufreq-dt driver

[Danilo Krummrich <dakr@redhat.com>]

On Thu, Jul 11, 2024 at 07:21:59PM +0200, Greg KH wrote:
> On Thu, Jul 11, 2024 at 06:41:29PM +0200, Greg KH wrote:
> > On Thu, Jul 11, 2024 at 06:34:22PM +0200, Greg KH wrote:
> > > On Thu, Jul 11, 2024 at 06:12:08PM +0200, Danilo Krummrich wrote:
> > > > On Thu, Jul 11, 2024 at 04:37:50PM +0200, Greg KH wrote:
> > > > > On Thu, Jul 11, 2024 at 03:21:31PM +0200, Danilo Krummrich wrote:
> > > > > > (2) You require drivers to always implement a "dummy" struct platform_device,
> > > > > > there is platform_device_register_simple() for that purpose.
> > > > > 
> > > > > No, NEVER do that.  platform devices are only for real platform devices,
> > > > > do not abuse that interface any more than it already is.
> > > > 
> > > > I thought we're talking about cases like [1] or [2], but please correct me if
> > > > those are considered abusing the platform bus as well.
> > > > 
> > > > (Those drivers read the CPU OF nodes, instead of OF nodes that represent a
> > > > separate device.)
> > > > 
> > > > [1] https://elixir.bootlin.com/linux/latest/source/drivers/cpuidle/cpuidle-riscv-sbi.c#L586
> > > > [2] https://elixir.bootlin.com/linux/latest/source/drivers/cpuidle/cpuidle-psci.c#L441
> > > 
> > > Yes, these are abuses of that and should be virtual devices as they have
> > > nothing to do with the platform bus.
> > > 
> > > > > > I think (2) is the preferred option.
> > > > > 
> > > > > No, not at all, sorry.
> > > > > 
> > > > > Use a real device, you have one somewhere that relates to this hardware,
> > > > > otherwise you aren't controlling anything and then you can use a virtual
> > > > > device.
> > > > 
> > > > Of course we should stick to a real device if there is one, I didn't meant to
> > > > say anything else.
> > > > 
> > > > But since it came up now, some virtual drivers still require a parent device.
> > > 
> > > Great, use the default one that the kernel gives you :)
> > > 
> > > > For instance, in DRM we have vGEM [3] and vKMS [4], that use
> > > > platform_device_register_simple() for this purpose.
> > > 
> > > Again, abuse, please do not do so.
> > > 
> > > > What should they use instead? I'm happy to fix things up if required.
> > > > 
> > > > [3] https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/vgem/vgem_drv.c
> > > > [4] https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/vkms/vkms_drv.c
> > > 
> > > Use the virtual device interface please, that's what it is there for.
> > 
> > To be specific, look at the devices in /sys/devices/virtual/ that's
> > where yours should be showing up in, not in the root of /sys/devices/
> > like they are by creating a "fake" platform device at the root of the
> > device tree.
> 
> Ok, at first glance this seems a little bit more complex than what the
> platform api gives you, let me knock something up next week during the
> merge window to make this more simple and then let some interns at it to
> sweep the kernel tree and fix up this proliferation of platform device
> abuse.

Yeah, I stared at this for the last 30 minutes and was just about to reply.

I think that we probably want to get rid of this abuse, as there are quite a lot
of examples for this.

And considering that I wasn't able to find a rather straight forward replacement
that makes it go into /sys/devices/virtual/ I think it's not super unexpected
that this spreads.

It looks like we probably want some kind virtual device API for that purpose?

> 
> thanks,
> 
> greg k-h
>