Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1540470 > unrolled thread

Re: [PATCH] iio: misc: add a generic regulator driver

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH] iio: misc: add a generic regulator driver Lars-Peter Clausen <lars@metafoo.de> - 2016-12-12 18:20 +0100
    Re: [PATCH] iio: misc: add a generic regulator driver Bartosz Golaszewski <bgolaszewski@baylibre.com> - 2016-12-13 15:30 +0100
      Re: [PATCH] iio: misc: add a generic regulator driver Linus Walleij <linus.walleij@linaro.org> - 2016-12-28 14:10 +0100
    Re: [PATCH] iio: misc: add a generic regulator driver Geert Uytterhoeven <geert@linux-m68k.org> - 2016-12-23 11:10 +0100
      Re: [PATCH] iio: misc: add a generic regulator driver Lars-Peter Clausen <lars@metafoo.de> - 2016-12-23 12:40 +0100
        Re: [PATCH] iio: misc: add a generic regulator driver Geert Uytterhoeven <geert@linux-m68k.org> - 2016-12-23 14:00 +0100
          Re: [PATCH] iio: misc: add a generic regulator driver Jonathan Cameron <jic23@jic23.retrosnub.co.uk> - 2016-12-24 11:50 +0100
        Re: [PATCH] iio: misc: add a generic regulator driver Linus Walleij <linus.walleij@linaro.org> - 2016-12-28 14:10 +0100

#1540470 — Re: [PATCH] iio: misc: add a generic regulator driver

On 12/06/2016 12:12 PM, Bartosz Golaszewski wrote:
> 2016-12-03 10:11 GMT+01:00 Jonathan Cameron <jic23@kernel.org>:
>> On 30/11/16 10:10, Lars-Peter Clausen wrote:
>>> On 11/29/2016 04:35 PM, Bartosz Golaszewski wrote:
>>>> 2016-11-29 16:30 GMT+01:00 Lars-Peter Clausen <lars@metafoo.de>:
>>>>> On 11/29/2016 04:22 PM, Bartosz Golaszewski wrote:
>>>>> [...]
>>>>>> diff --git a/Documentation/devicetree/bindings/iio/misc/iio-regulator.txt b/Documentation/devicetree/bindings/iio/misc/iio-regulator.txt
>>>>>> new file mode 100644
>>>>>> index 0000000..147458f
>>>>>> --- /dev/null
>>>>>> +++ b/Documentation/devicetree/bindings/iio/misc/iio-regulator.txt
>>>>>> @@ -0,0 +1,18 @@
>>>>>> +Industrial IO regulator device driver
>>>>>> +-------------------------------------
>>>>>> +
>>>>>> +This document describes the bindings for the iio-regulator - a dummy device
>>>>>> +driver representing a physical regulator within the iio framework.
>>>>>
>>>>> No bindings for drivers, only for hardware. So this wont work.
>>>>>
>>>>
>>>> What about exporting regulator attributes analogous to the one in this
>>>> patch from the iio-core when a *-supply property is specified for a
>>>> node?
>>>
>>> The problem with exposing direct control to the regulator is that it allows
>>> to modify the hardware state without the drivers knowledge. If you
>>> power-cycle a device all previous configuration that has been written to the
>>> device is reset. The device driver needs to be aware of this otherwise its
>>> assumed state and the actual device state can divert which will result in
>>> undefined behavior. Also access to the device will fail unexpectedly when
>>> the regulator is turned off. So I think generally the driver should
>>> explicitly control the regulator, power-up when needed, power-down when not.
>> I agree with what Lars has said.
>>
>> There 'may' be some argument to ultimately have a bridge driver from
>> regulators to IIO.  That would be for cases where the divide between a regulator
>> and a DAC is blurred.  However it would still have to play nicely with the
>> regulator framework and any other devices registered on that regulator.
>> Ultimately the ideal in that case would then be to describe what the DAC is
>> actually being used to do but that's a more complex issue!
>>
>> That doesn't seem to be what you are targeting here.
>>
>> What it sounds like you need is to have the hardware well enough described that
>> the standard runtime power management can disable the regulator just fine when
>> it is not in use.  This may mean improving the power management in the relevant
>> drivers.
>>
>> Jonathan
>>
>> p.s. If ever proposing to do something 'unusual' with a regulator you should
>> bring in the regulator framework maintainers in the cc list.
>>>
>>> - Lars
>>>
>>
> 
> I wrote the initial patch quickly and didn't give it much of a
> thought. Now I realized I completely missed the point and managed to
> confuse everybody - myself included.
> 
> So the problem we have is not power-cycling the adc - it's
> power-cycling the device connected to a probe on which there's an adc.
> What I was trying to do was adding support for the power-switch on
> baylibre-acme[1] probes.
> 
> For example: we have a USB probe on which the VBUS signal goes through
> a power load switch and than through the adc. The adc (in this case
> ina226) is always powered on, while the fixed regulator I wanted to
> enable/disable actually drives the power switch to cut/restore power
> to the connected USB device i.e. there's no real regulator - just a
> GPIO driving the power switch.
> 
> A typical use case is measuring the power consumption of development
> boards[2]. Rebooting them remotely using acme probes is already done,
> but we're using the obsolete /sys/class/gpio interface.
> 
> We're already using libiio to read the measured data from the power
> monitor, that's why we'd like to use the iio framework for
> power-cycling the devices as well. My question is: would bridging the
> regulator framework be the right solution? Should we look for
> something else? Bridge the GPIO framework instead?

I wouldn't necessaries create bridge, but instead just use the GPIO
framework directly.

We now have the GPIO chardev interface which meant to be used to support
application specific logic that control the GPIOs, but where you don't want
to write a kernel driver.

My idea was to add GPIOs and GPIO chips as high level object inside libiio
that can be accessed through the same context as the IIO devices. Similar to
the current IIO API you have a API for gpios that allows to enumerate the
GPIO devices and their pins as well as modify the pin state.

[toc] | [next] | [standalone]

#1541107

2016-12-12 18:15 GMT+01:00 Lars-Peter Clausen <lars@metafoo.de>:
> On 12/06/2016 12:12 PM, Bartosz Golaszewski wrote:

[snip!]

>>
>> So the problem we have is not power-cycling the adc - it's
>> power-cycling the device connected to a probe on which there's an adc.
>> What I was trying to do was adding support for the power-switch on
>> baylibre-acme[1] probes.
>>
>> For example: we have a USB probe on which the VBUS signal goes through
>> a power load switch and than through the adc. The adc (in this case
>> ina226) is always powered on, while the fixed regulator I wanted to
>> enable/disable actually drives the power switch to cut/restore power
>> to the connected USB device i.e. there's no real regulator - just a
>> GPIO driving the power switch.
>>
>> A typical use case is measuring the power consumption of development
>> boards[2]. Rebooting them remotely using acme probes is already done,
>> but we're using the obsolete /sys/class/gpio interface.
>>
>> We're already using libiio to read the measured data from the power
>> monitor, that's why we'd like to use the iio framework for
>> power-cycling the devices as well. My question is: would bridging the
>> regulator framework be the right solution? Should we look for
>> something else? Bridge the GPIO framework instead?
>
> I wouldn't necessaries create bridge, but instead just use the GPIO
> framework directly.
>
> We now have the GPIO chardev interface which meant to be used to support
> application specific logic that control the GPIOs, but where you don't want
> to write a kernel driver.
>
> My idea was to add GPIOs and GPIO chips as high level object inside libiio
> that can be accessed through the same context as the IIO devices. Similar to
> the current IIO API you have a API for gpios that allows to enumerate the
> GPIO devices and their pins as well as modify the pin state.
>

+ Linus

While the new GPIO interface would be very convenient - in our case we
could simply name the lines appropriately in the device tree - I'm not
sure this would be the correct approach.

From this year's ELCE in Berlin I remember Linus suggested during his
talk that it's always better to write a kernel driver. Also: this way
the relevant GPIO lines would not be reserved for exclusive use by
power switches.

Linus - do you have any thoughts/suggestions on that subject?

Best regards,
Bartosz Golaszewski

[toc] | [prev] | [next] | [standalone]

#1547947

On Tue, Dec 13, 2016 at 3:28 PM, Bartosz Golaszewski
<bgolaszewski@baylibre.com> wrote:

> + Linus
>
> While the new GPIO interface would be very convenient - in our case we
> could simply name the lines appropriately in the device tree - I'm not
> sure this would be the correct approach.
>
> From this year's ELCE in Berlin I remember Linus suggested during his
> talk that it's always better to write a kernel driver. Also: this way
> the relevant GPIO lines would not be reserved for exclusive use by
> power switches.
>
> Linus - do you have any thoughts/suggestions on that subject?

If the probe you are power cycling has its own DT node and is
described as a device per se in the system, then it should have
a device driver grabbing and toggling its own GPIO line.

If the probe is only really known in userspace, and driven
from userspace, it's GPIO reset line should also be driven
from userspace, using the chardev ABI as you describe.

Whether something should have a userspace or kernelspace
driver is a gray area, admittedly. There are cases for both.
The general consideration would be reuse and deployment.
If you expect all users of this probe to always use libiio and
some other userspace, I guess userspace-only makes sense?

Yours,
Linus Walleij

[toc] | [prev] | [next] | [standalone]

#1546803

Hi Lars,

On Mon, Dec 12, 2016 at 6:15 PM, Lars-Peter Clausen <lars@metafoo.de> wrote:
> On 12/06/2016 12:12 PM, Bartosz Golaszewski wrote:
>> We're already using libiio to read the measured data from the power
>> monitor, that's why we'd like to use the iio framework for
>> power-cycling the devices as well. My question is: would bridging the
>> regulator framework be the right solution? Should we look for
>> something else? Bridge the GPIO framework instead?
>
> I wouldn't necessaries create bridge, but instead just use the GPIO
> framework directly.
>
> We now have the GPIO chardev interface which meant to be used to support
> application specific logic that control the GPIOs, but where you don't want
> to write a kernel driver.
>
> My idea was to add GPIOs and GPIO chips as high level object inside libiio
> that can be accessed through the same context as the IIO devices. Similar to
> the current IIO API you have a API for gpios that allows to enumerate the
> GPIO devices and their pins as well as modify the pin state.

That would mean libiio has access to all GPIOs, allowing a remote person
to not only control through iiod the GPIOs for industrial control, but also the
GPIOs not intended for export, right?

Having a separate GPIO switch driver avoids that, as DT (or some other means)
can be used to specify and label the GPIOs for IIO use.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

[toc] | [prev] | [next] | [standalone]

#1546832

On 12/23/2016 11:00 AM, Geert Uytterhoeven wrote:
> Hi Lars,
> 
> On Mon, Dec 12, 2016 at 6:15 PM, Lars-Peter Clausen <lars@metafoo.de> wrote:
>> On 12/06/2016 12:12 PM, Bartosz Golaszewski wrote:
>>> We're already using libiio to read the measured data from the power
>>> monitor, that's why we'd like to use the iio framework for
>>> power-cycling the devices as well. My question is: would bridging the
>>> regulator framework be the right solution? Should we look for
>>> something else? Bridge the GPIO framework instead?
>>
>> I wouldn't necessaries create bridge, but instead just use the GPIO
>> framework directly.
>>
>> We now have the GPIO chardev interface which meant to be used to support
>> application specific logic that control the GPIOs, but where you don't want
>> to write a kernel driver.
>>
>> My idea was to add GPIOs and GPIO chips as high level object inside libiio
>> that can be accessed through the same context as the IIO devices. Similar to
>> the current IIO API you have a API for gpios that allows to enumerate the
>> GPIO devices and their pins as well as modify the pin state.
> 
> That would mean libiio has access to all GPIOs, allowing a remote person
> to not only control through iiod the GPIOs for industrial control, but also the
> GPIOs not intended for export, right?

Well, it is a policy question. Who gets access to what. Right now it is all
or nothing, a privileged application gets access to all devices/GPIOs, a
unprivileged application gets access to nothing. Same for GPIOs as well as
IIO devices.

iiod at the moment does not have any access control at all, which in itself
is a problem. We need to add support for that at some point. I don't see an
issue with implementing a finer grained access scheme when we do so. E.g.
unprivileged applications only get access to certain pins.

> Having a separate GPIO switch driver avoids that, as DT (or some other means)
> can be used to specify and label the GPIOs for IIO use.

Sure, functionally this would be equivalent, but we have to ask whether this
is the right way to use the DT. Is access policy specification part of the
hardware description? In my opinion the answer is no. At the hardware
description level there is no operating system, there is no userspace or
kernelspace, there is are no access levels. Putting the distinction between
a switch/regulator that can be controlled from userspace or can only be
controlled from kernel space into the DT would be a layering violation. It
is analogous to why we don't have spidev DT bindings. This is an issue that
needs to be solved at a higher level. In my opinion this level is a
cooperation between kernel- and userspace. Kernelspace offering an interface
to export a device for userspace access and userspace making use of that
interface to request access to a device. In a similar way to how vfio is
structured.

[toc] | [prev] | [next] | [standalone]

#1546859

Hi Lars,

On Fri, Dec 23, 2016 at 12:35 PM, Lars-Peter Clausen <lars@metafoo.de> wrote:
> On 12/23/2016 11:00 AM, Geert Uytterhoeven wrote:
>> On Mon, Dec 12, 2016 at 6:15 PM, Lars-Peter Clausen <lars@metafoo.de> wrote:
>>> On 12/06/2016 12:12 PM, Bartosz Golaszewski wrote:
>>>> We're already using libiio to read the measured data from the power
>>>> monitor, that's why we'd like to use the iio framework for
>>>> power-cycling the devices as well. My question is: would bridging the
>>>> regulator framework be the right solution? Should we look for
>>>> something else? Bridge the GPIO framework instead?
>>>
>>> I wouldn't necessaries create bridge, but instead just use the GPIO
>>> framework directly.
>>>
>>> We now have the GPIO chardev interface which meant to be used to support
>>> application specific logic that control the GPIOs, but where you don't want
>>> to write a kernel driver.
>>>
>>> My idea was to add GPIOs and GPIO chips as high level object inside libiio
>>> that can be accessed through the same context as the IIO devices. Similar to
>>> the current IIO API you have a API for gpios that allows to enumerate the
>>> GPIO devices and their pins as well as modify the pin state.
>>
>> That would mean libiio has access to all GPIOs, allowing a remote person
>> to not only control through iiod the GPIOs for industrial control, but also the
>> GPIOs not intended for export, right?
>
> Well, it is a policy question. Who gets access to what. Right now it is all
> or nothing, a privileged application gets access to all devices/GPIOs, a
> unprivileged application gets access to nothing. Same for GPIOs as well as
> IIO devices.
>
> iiod at the moment does not have any access control at all, which in itself
> is a problem. We need to add support for that at some point. I don't see an
> issue with implementing a finer grained access scheme when we do so. E.g.
> unprivileged applications only get access to certain pins.

OK, so that's WIP.

>> Having a separate GPIO switch driver avoids that, as DT (or some other means)
>> can be used to specify and label the GPIOs for IIO use.
>
> Sure, functionally this would be equivalent, but we have to ask whether this
> is the right way to use the DT. Is access policy specification part of the
> hardware description? In my opinion the answer is no. At the hardware
> description level there is no operating system, there is no userspace or
> kernelspace, there is are no access levels. Putting the distinction between
> a switch/regulator that can be controlled from userspace or can only be
> controlled from kernel space into the DT would be a layering violation. It
> is analogous to why we don't have spidev DT bindings. This is an issue that
> needs to be solved at a higher level. In my opinion this level is a
> cooperation between kernel- and userspace. Kernelspace offering an interface
> to export a device for userspace access and userspace making use of that
> interface to request access to a device. In a similar way to how vfio is
> structured.

I'm not advocating using DT for policy, only for hardware description.

We have means (bindings) to describe GPIOs connected to LEDs and switches
(incl. their labels), while you can control LEDs through plain GPIO sysfs
export or chardev, too. It's just more error prone to use the latter.

We do not have bindings to describe GPIOs connected to e.g. relays.

Switching external devices (the internals of those devices not described
itself in DT, like in an industrial context), sounds more like something to
be handled by IIO, doesn't it?

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

[toc] | [prev] | [next] | [standalone]

#1547041

On 23 December 2016 12:56:11 GMT+00:00, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
>Hi Lars,
>
>On Fri, Dec 23, 2016 at 12:35 PM, Lars-Peter Clausen <lars@metafoo.de>
>wrote:
>> On 12/23/2016 11:00 AM, Geert Uytterhoeven wrote:
>>> On Mon, Dec 12, 2016 at 6:15 PM, Lars-Peter Clausen
><lars@metafoo.de> wrote:
>>>> On 12/06/2016 12:12 PM, Bartosz Golaszewski wrote:
>>>>> We're already using libiio to read the measured data from the
>power
>>>>> monitor, that's why we'd like to use the iio framework for
>>>>> power-cycling the devices as well. My question is: would bridging
>the
>>>>> regulator framework be the right solution? Should we look for
>>>>> something else? Bridge the GPIO framework instead?
>>>>
>>>> I wouldn't necessaries create bridge, but instead just use the GPIO
>>>> framework directly.
>>>>
>>>> We now have the GPIO chardev interface which meant to be used to
>support
>>>> application specific logic that control the GPIOs, but where you
>don't want
>>>> to write a kernel driver.
>>>>
>>>> My idea was to add GPIOs and GPIO chips as high level object inside
>libiio
>>>> that can be accessed through the same context as the IIO devices.
>Similar to
>>>> the current IIO API you have a API for gpios that allows to
>enumerate the
>>>> GPIO devices and their pins as well as modify the pin state.
>>>
>>> That would mean libiio has access to all GPIOs, allowing a remote
>person
>>> to not only control through iiod the GPIOs for industrial control,
>but also the
>>> GPIOs not intended for export, right?
>>
>> Well, it is a policy question. Who gets access to what. Right now it
>is all
>> or nothing, a privileged application gets access to all
>devices/GPIOs, a
>> unprivileged application gets access to nothing. Same for GPIOs as
>well as
>> IIO devices.
>>
>> iiod at the moment does not have any access control at all, which in
>itself
>> is a problem. We need to add support for that at some point. I don't
>see an
>> issue with implementing a finer grained access scheme when we do so.
>E.g.
>> unprivileged applications only get access to certain pins.
>
>OK, so that's WIP.
>
>>> Having a separate GPIO switch driver avoids that, as DT (or some
>other means)
>>> can be used to specify and label the GPIOs for IIO use.
>>
>> Sure, functionally this would be equivalent, but we have to ask
>whether this
>> is the right way to use the DT. Is access policy specification part
>of the
>> hardware description? In my opinion the answer is no. At the hardware
>> description level there is no operating system, there is no userspace
>or
>> kernelspace, there is are no access levels. Putting the distinction
>between
>> a switch/regulator that can be controlled from userspace or can only
>be
>> controlled from kernel space into the DT would be a layering
>violation. It
>> is analogous to why we don't have spidev DT bindings. This is an
>issue that
>> needs to be solved at a higher level. In my opinion this level is a
>> cooperation between kernel- and userspace. Kernelspace offering an
>interface
>> to export a device for userspace access and userspace making use of
>that
>> interface to request access to a device. In a similar way to how vfio
>is
>> structured.
>
>I'm not advocating using DT for policy, only for hardware description.
>
>We have means (bindings) to describe GPIOs connected to LEDs and
>switches
>(incl. their labels), while you can control LEDs through plain GPIO
>sysfs
>export or chardev, too. It's just more error prone to use the latter.
>
>We do not have bindings to describe GPIOs connected to e.g. relays.
We should.
>
>Switching external devices (the internals of those devices not
>described
>itself in DT, like in an industrial context), sounds more like
>something to
>be handled by IIO, doesn't it?

Certainly, if there is known hardware to describe, we should endeavour to describe it.
Userspace interfaces are needed wherever we hit the boundary of what we can describe,
 whether because we are measuring things not in our control (e.g. what key is pressed on a
 keyboard) or because the next bit of hardware is interchangeable (e.g. your relay example, or
 this power switch).

The challenge is to structure the device model for the interchangeable edge case to be the
 same, more or less, as it would be if we knew what was hanging off the switch.
Hence, we either cut out early (gpio) or we attempt to put an appropriate consumer in place
 for the gpio (or possibly the power switch if we describe that). No problem at all in doing that
 last chunk with IIO or GPIO userspace as appropriate...

The challenge is that we are representing the fact the hardware is unknown in device tree.
Perhaps we need a way to make that explicit? Is there one already? Things like extcon do similar
 things I guess.
Same is true for regulators, when they are at the edge of the device...

On the binary channel types in IIO we have discussed this a fair bit in the past. There
Is a non trivial amount of work needed to do triggered input (demuxing to multiple consumers
In particular). 
Sysfs stuff would be simple but then it would really be gpio interface wrapped up a bit.
What IIO would bring to the mix ultimately is synchronized triggering of input and output.
(Speaking of which, Lars any progress on output buffers? Perhaps if we post that someone else might pick it up and run with it?)

One could argue the relay case is more of a mux than anything else so perhaps the ongoing
 generic mux subsystem discussion would be a good place to talk about that?

Interesting discussion, sorry it took me until my Christmas train journey to join in).

Linus, if you get a chance, you have probably thought more about gpio IIO interactions than I have!

Jonathan
>
>Gr{oetje,eeting}s,
>
>                        Geert
>
>--
>Geert Uytterhoeven -- There's lots of Linux beyond ia32 --
>geert@linux-m68k.org
>
>In personal conversations with technical people, I call myself a
>hacker. But
>when I'm talking to journalists I just say "programmer" or something
>like that.
>                                -- Linus Torvalds
>--
>To unsubscribe from this list: send the line "unsubscribe linux-iio" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.

[toc] | [prev] | [next] | [standalone]

#1547948

On Fri, Dec 23, 2016 at 12:35 PM, Lars-Peter Clausen <lars@metafoo.de> wrote:
> On 12/23/2016 11:00 AM, Geert Uytterhoeven wrote:

> Well, it is a policy question. Who gets access to what. Right now it is all
> or nothing, a privileged application gets access to all devices/GPIOs, a
> unprivileged application gets access to nothing. Same for GPIOs as well as
> IIO devices.
>
> iiod at the moment does not have any access control at all, which in itself
> is a problem. We need to add support for that at some point. I don't see an
> issue with implementing a finer grained access scheme when we do so. E.g.
> unprivileged applications only get access to certain pins.

I don't know why this is percieved as such a big practical problem.

It seems to me as more of a theoretical exploit path than a practical one.
(Famous last words...)

We have per-device and not per-line GPIO access restrictions.
/dev/gpiochip0
/dev/gpiochip1
etc
can all have per-device access restrictions.

This is no different from /dev/sda for example. You do not have
per-sector control of the block device, because it doesn't make sense.
Either you access all of the device, or nothing.
The same goes for IIO devices.

This pattern is very clear. You get access to a whole device or none
of it.

As with disks and IIO devices, if you want more granular access
restrictions, that is policy, and should reside in userspace.

Yours,
Linus Walleij

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web