Date:   Fri, 18 Feb 2022 11:48:50 -0800
From:   Won Chung <wonchung@...gle.com>
To:     "Rafael J. Wysocki" <rafael@...nel.org>
Cc:     Heikki Krogerus <heikki.krogerus@...ux.intel.com>,
        Len Brown <lenb@...nel.org>,
        Benson Leung <bleung@...omium.org>,
        Prashant Malani <pmalani@...omium.org>,
        ACPI Devel Maling List <linux-acpi@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Greg Kroah-Hartman <greg@...ah.com>
Subject: Re: [PATCH v6] ACPI: device_sysfs: Add sysfs support for _PLD

On Fri, Feb 18, 2022 at 9:25 AM Rafael J. Wysocki <rafael@...nel.org> wrote:
>
> On Fri, Feb 18, 2022 at 2:15 AM Won Chung <wonchung@...gle.com> wrote:
> >
> > On Wed, Feb 16, 2022 at 8:39 AM Rafael J. Wysocki <rafael@...nel.org> wrote:
> > >
> > > On Wed, Feb 16, 2022 at 12:34 PM Heikki Krogerus
> > > <heikki.krogerus@...ux.intel.com> wrote:
> > > >
> > > > On Tue, Feb 15, 2022 at 02:54:11PM +0100, Rafael J. Wysocki wrote:
> > > > > On Tue, Feb 15, 2022 at 11:14 AM Heikki Krogerus
> > > > > <heikki.krogerus@...ux.intel.com> wrote:
> > > > > >
> > > > > > On Mon, Feb 14, 2022 at 02:58:44PM -0800, Won Chung wrote:
> > > > > > > On Mon, Feb 14, 2022 at 12:30 PM Won Chung <wonchung@...gle.com> wrote:
> > > > > > > >
> > > > > > > > On Mon, Feb 14, 2022 at 11:12 AM Rafael J. Wysocki <rafael@...nel.org> wrote:
> > > > > > > > >
> > > > > > > > > On Fri, Feb 11, 2022 at 3:30 AM Won Chung <wonchung@...gle.com> wrote:
> > > > > > > > > >
> > > > > > > > > > When ACPI table includes _PLD fields for a device, create a new
> > > > > > > > > > directory (pld) in sysfs to share _PLD fields.
> > > > > > > > >
> > > > > > > > > This version of the patch loos better to me, but I'm not sure if it
> > > > > > > > > goes into the right direction overall.
> > > > > > > > >
> > > > > > > > > > Currently without PLD information, when there are multiple of same
> > > > > > > > > > devices, it is hard to distinguish which device corresponds to which
> > > > > > > > > > physical device in which location. For example, when there are two Type
> > > > > > > > > > C connectors, it is hard to find out which connector corresponds to the
> > > > > > > > > > Type C port on the left panel versus the Type C port on the right panel.
> > > > > > > > >
> > > > > > > > > So I think that this is your primary use case and I'm wondering if
> > > > > > > > > this is the best way to address it.
> > > > > > > > >
> > > > > > > > > Namely, by exposing _PLD information under the ACPI device object,
> > > > > > > > > you'll make user space wanting to use that information depend on this
> > > > > > > > > interface, but the problem is not ACPI-specific (inevitably, it will
> > > > > > > > > appear on systems using DT, sooner or later) and making the user space
> > > > > > > > > interface related to it depend on ACPI doesn't look like a perfect
> > > > > > > > > choice.
> > > > > > > > >
> > > > > > > > > IOW, why don't you create a proper ABI for this in the Type C
> > > > > > > > > subsystem and expose the information needed by user space in a generic
> > > > > > > > > way that can be based on the _PLD information on systems with ACPI?
> > > > > > > >
> > > > > > > > Hi Rafael,
> > > > > > > >
> > > > > > > > Thank you for the review.
> > > > > > > >
> > > > > > > > I was thinking that _PLD info is specific to ACPI since it is part of
> > > > > > > > the ACPI table. Could you explain a little bit more on why you think
> > > > > > > > exposing _PLD fields is not an ACPI-specific problem?
> > > > > > >
> > > > > > > Hi Rafael again,
> > > > > > >
> > > > > > > Sorry for the silly question here. I misunderstood your comment a bit,
> > > > > > > but I talked to Benson and Prashant for clarification. I understand
> > > > > > > now what you mean by it is not an ACPI-specific problem and exposing
> > > > > > > PLD would depend on ACPI.
> > > > > > >
> > > > > > > >
> > > > > > > > I gave an example of how _PLD fields can be used for specifying Type C
> > > > > > > > connectors, but it is not Type C specific. For Chrome OS, we plan to
> > > > > > > > initially add PLD to not only Type C connectors but also USB port
> > > > > > > > devices (including Type C and Type A). Also, PLD can be used in the
> > > > > > > > future for describing other types of ports too like HDMI. (Benson and
> > > > > > > > Prashant, please correct or add if I am wrong or missing some
> > > > > > > > information) Maybe my commit message was not detailed enough..
> > > > > > > >
> > > > > > > > I am also curious what Heikki thinks about this. Heikki, can you take
> > > > > > > > a look and share your thoughts?
> > > > > > >
> > > > > > > I am still curious what you and Heikki think about this since it may
> > > > > > > not be a Type C specific issue. We can start from adding generic
> > > > > > > location info to Type C subsystem first, as you suggested, then
> > > > > > > consider how to do the same for USB devices and Type A ports
> > > > > > > afterwards. I would appreciate sharing any thoughts or feedback. Thank
> > > > > > > you very much!
> > > > > >
> > > > > > Like you said, _PLD is not Type-C specific. We can't limit it to any
> > > > > > specific device class. For example, I'm pretty sure that sooner or
> > > > > > later we want to get this information in user space also with camera
> > > > > > sensors, and probable with a few other things as well.
> > > > > >
> > > > > > I think the question here is, can we create a some kind of an
> > > > > > abstraction layer for the user space that exposes the device location
> > > > > > details in generic Linux specific way - so with ACPI it would utilise
> > > > > > the _PLD, and with DT something else (today AFAIK DT does not have
> > > > > > any way to describe locations of the devices). Maybe I'm wrong?
> > > > >
> > > > > No, you aren't.
> > > > >
> > > > > > But if that is the question, then IMO the answer is: maybe one day,
> > > > > > but not today,
> > > > >
> > > > > Why not?
> > > > >
> > > > > > and even if we one day can come up with something like
> > > > > > that, we still should expose the _PLD as ACPI specific information to
> > > > > > the user space as is.
> > > > >
> > > > > Why would it need that information in this particular format?
> > > > >
> > > > > > Even if one day we have common sysfs attributes for all the devices
> > > > > > that contain the location of the device in some form, those attributes
> > > > > > will almost certainly have only a sub-set of the _PLD details, a
> > > > > > sub-set that works also with DT.
> > > > >
> > > > > That doesn't have to be the case.
> > > > >
> > > > > However, things linke cpuidle have been invented to provide user space
> > > > > interfaces for features that previously were only available on systems
> > > > > with ACPI.  Why is _PLD different?
> > > > >
> > > > > > IMO the user space should always have access to all the necessary _PLD
> > > > > > details in their raw form if needed, even if those common device
> > > > > > location attributes exist - duplicated information or not.
> > > > >
> > > > > Again, why would it need that information?
> > > >
> > > > We don't know if we'll need that in the future, and that's the point.
> > >
> > > Well, for me that would be a good enough reason for avoiding to expose it.
> > >
> > > If there is no particular reason for exposing any information to user
> > > space, I don't see why it should be exposed at all.
> > >
> > > There is some cost of exposing things to user space, so why pay it for
> > > no benefit?
> > >
> > > > > > And debugfs
> > > > > > unfortunately is also not OK for that, because the user space needs to
> > > > > > be able to also rely on access to the additional details if needed.
> > > > >
> > > > > What additional details do you mean?
> > > > >
> > > > > > We can limit the _PLD fields that we expose to the ones that we know
> > > > > > we need today (and probable should limit them to those), and we can of
> > > > > > course have a Kconfig option for the _PLD sysfs information if we want
> > > > > > to, but let's not start this by trying to figure out what kind of
> > > > > > abstraction we want for this. Right now we simply can not do that.
> > > > >
> > > > > Why can't we?
> > > >
> > > > Right now we can't say for sure if DT can even supply the details that
> > > > we need from _PLD. I don't think we can at the moment even say are the
> > > > DT guys willing to support this at all.
> > > >
> > > > To play it safe, I would just supply the needed _PLD fields as part of
> > > > the ACPI device nodes (under /sys/bus/acpi).
> > >
> > > That would be suboptimal for a few reasons:
> > >
> > > 1. The interface is potentially confusing.  User space would first
> > > need to locate the ACPI device interface corresponding to the given
> > > "real" device in order to use that information.
> > > 2. It doesn't scale beyond ACPI.
> > > 3. From the ACPI subsystem's perspective the choice of the "relevant"
> > > _PLD fields is arbitrary and exposing all of them is overkill for any
> > > use cases known to me.
> > > 4. The ACPI subsystem doesn't know the devices for which _PLD
> > > information should be exposed and there are some devices for which it
> > > is just not useful.
> > >
> > > > There we can guarantee
> > > > that we'll always be able to supply all the information in the _PDL if
> > > > needed. Since we would add these to the ACPI nodes, it would be
> > > > crystal clear to the userspace that this information is only available
> > > > on ACPI platforms.
> > >
> > > I'm not considering this as a feature.
> > >
> > > > Then if, and only if, we know that DT can supply the same information
> > > > (at least to some of it) I would start thinking about the alternative
> > > > interface to this information that we make part of the actual devices.
> > > > Since at this point we have already the primary ACPI specific
> > > > interface to this same information that guarantees that it can supply
> > > > all the details if necessary, we don't have to worry about having to
> > > > be able to do the same with this new interface. This interface can
> > > > just expose the common details that we know for sure that both ACPI
> > > > and DT can always supply.
> > >
> > > Well, there is another possible approach: Expose the information
> > > needed to address a particular use case in a way that doesn't strictly
> > > depend on ACPI and make this use ACPI as a backend.  Don't worry about
> > > the DT side of things.  If the generic interface is there and it is
> > > suitable enough, DT will be in the receiving end position with much
> > > less of a freedom to introduce a new interface for the same purpose.
> > >
> > > On the other hand, if _PLD information is exposed in an ACPI-specific
> > > way, it is almost guaranteed that there will be a DT-specific
> > > interface for the same thing and utilities wanting to be generic will
> > > need to support both of them which will be extra pain.  Some of them
> > > will choose to support the DT-specific interface only and we'll end up
> > > with utilities that can't be used on ACPI-based systems because of
> > > incompatible interfaces.  Been there already.  Thanks, but no thanks!
> >
> > Hi Rafael,
> >
> > Thank you for the feedback. If we add a generic location to type c
> > connector, would you suggest we do something similar to other devices
> > that would use PLD information? (like USB devices, HDMI ports, and so
> > on).
>
> If there is a specific use case for exposing that information to user
> space, then yes, but it all depends on how user space is going to use
> that information (or how you envision the usage of that information in
> user space).
>
> > Also, I am curious what you think about how to add generic
> > locations for Type A ports which I believe do not have connectors like
> > Type C. I would appreciate it if anyone can share any ideas. Thank you
> > very much!
>
> I'm not sure I understand the question correctly.  Can you clarify,
> please?  Or better, give an example of what exactly you are referring
> to?

Hi Rafael,

For Type C ports, we have Type C connectors at /sys/class/typec in
which we can add generic location information, as you suggested.
However, since Type A ports do not have such connectors, I was curious
what would be a good way to add a generic location, instead of
exposing _PLD directly in the USB-A port's ACPI device.

Now that I think about it again and look through sysfs, I think we can
also add generic location to
/sys/bus/usb/devices/.../<hub_interface>/port<X>, some of which
represents Type A ports. Benson, do you think this could be a good way
for Type A ports? Who would be a good person to get feedback on this?

Heikki, I am also convinced by Rafael's feedback since userspace code
would also be quite ACPI-specific to access _PLD fields from ACPI
device sysfs. Would you agree? Regarding your concerns with DT, we can
look for some ways to have similar location information on systems
with DT. Would it sound okay to you to add generic location in Type C
connectors? If it does, I can start working on it and send patches for
review. If it does not, I would appreciate it if you can share your
thoughts on possible alternative approach.

Thank you very much all for the feedback!
Won

Powered by blists - more mailing lists