Hi Greg,
On Thu, Feb 03, 2022 at 03:55:19PM +0100, Greg Kroah-Hartman wrote:
> On Thu, Feb 03, 2022 at 05:46:55PM +0300, Heikki Krogerus wrote:
> > +/* These additional details are only available with vSafe5V supplies */
> > +static struct kobj_attribute dual_role_power_attr = __ATTR_RO(dual_role_power);
> > +static struct kobj_attribute usb_suspend_supported_attr = __ATTR_RO(usb_suspend_supported);
> > +static struct kobj_attribute unconstrained_power_attr = __ATTR_RO(unconstrained_power);
> > +static struct kobj_attribute usb_communication_capable_attr = __ATTR_RO(usb_communication_capable);
> > +static struct kobj_attribute dual_role_data_attr = __ATTR_RO(dual_role_data);
> > +static struct kobj_attribute
> > +unchunked_extended_messages_supported_attr = __ATTR_RO(unchunked_extended_messages_supported);
>
> Note, no 'struct device' should ever have a "raw" kobject hanging off of
> it. If so, something went wrong.
>
> If you do this, userspace will never be notified of the attributes and
> any userspace representation of the tree will be messed up.
>
> Please, use an attribute directory with a name, or if you really need to
> go another level deep, use a real 'struct device'. As-is here, I can't
> take it.
OK, got it. I don't think we can avoid the deeper levels, not without
making this really cryptic, and not really usable in all cases. These
objects are trying to represent (parts) of the protocol - the
messages, the objects in those messages, and later the responses to
those messages.
But I'm also trying to avoid having to claim that these objects are
"devices", because honestly, claiming that the packages used in
communication are devices is confusing, and just wrong. If we take
that road, then we really should redefine what struct device is
supposed to represent, and rename it also.
So would it be OK that, instead of registering these objects as
devices, we just introduce a kset where we can group them
(/sys/kernel/usb_power_delivery)?
thanks,
--
heikki
On Fri, Feb 04, 2022 at 12:04:41PM +0200, Heikki Krogerus wrote:
> Hi Greg,
>
> On Thu, Feb 03, 2022 at 03:55:19PM +0100, Greg Kroah-Hartman wrote:
> > On Thu, Feb 03, 2022 at 05:46:55PM +0300, Heikki Krogerus wrote:
> > > +/* These additional details are only available with vSafe5V supplies */
> > > +static struct kobj_attribute dual_role_power_attr = __ATTR_RO(dual_role_power);
> > > +static struct kobj_attribute usb_suspend_supported_attr = __ATTR_RO(usb_suspend_supported);
> > > +static struct kobj_attribute unconstrained_power_attr = __ATTR_RO(unconstrained_power);
> > > +static struct kobj_attribute usb_communication_capable_attr = __ATTR_RO(usb_communication_capable);
> > > +static struct kobj_attribute dual_role_data_attr = __ATTR_RO(dual_role_data);
> > > +static struct kobj_attribute
> > > +unchunked_extended_messages_supported_attr = __ATTR_RO(unchunked_extended_messages_supported);
> >
> > Note, no 'struct device' should ever have a "raw" kobject hanging off of
> > it. If so, something went wrong.
> >
> > If you do this, userspace will never be notified of the attributes and
> > any userspace representation of the tree will be messed up.
> >
> > Please, use an attribute directory with a name, or if you really need to
> > go another level deep, use a real 'struct device'. As-is here, I can't
> > take it.
>
> OK, got it. I don't think we can avoid the deeper levels, not without
> making this really cryptic, and not really usable in all cases. These
> objects are trying to represent (parts) of the protocol - the
> messages, the objects in those messages, and later the responses to
> those messages.
>
> But I'm also trying to avoid having to claim that these objects are
> "devices", because honestly, claiming that the packages used in
> communication are devices is confusing, and just wrong. If we take
> that road, then we really should redefine what struct device is
> supposed to represent, and rename it also.
Fair enough, this isn't really a device, it's an "attribute" of your
device you are wanting to show. It's just that you are really "deep".
You asked for:
/sys/class/typec/port0/usb_power_delivery
|-- revision
|-- sink_capabilities/
| |-- 1:fixed_supply/
| | |-- dual_role_data
| | |-- dual_role_power
| | |-- fast_role_swap_current
| | |-- operational_current
| | |-- unchunked_extended_messages_supported
| | |-- unconstrained_power
| | |-- usb_communication_capable
| | |-- usb_suspend_supported
| | `-- voltage
| |-- 2:variable_supply/
| | |-- maximum_voltage
| | |-- minimum_voltage
| | `-- operational_current
| `-- 3:battery/
| |-- maximum_voltage
| |-- minimum_voltage
| `-- operational_power
`-- source_capabilities/
`-- 1:fixed_supply/
|-- dual_role_data
|-- dual_role_power
|-- maximum_current
|-- unchunked_extended_messages_supported
|-- unconstrained_power
|-- usb_communication_capable
|-- usb_suspend_supported
`-- voltage
To start with, your "attribute" is really "usb_power_delivery" here, so
you can just use an attribute group name to get the "revision" file.
But then the later ones could be flat in that directory as well, using a
':' to split as you did already, and the above could turn into:
/sys/class/typec/port0/usb_power_delivery
|-- revision
|-- sink_capabilites:1:fixed_supply:dual_role_data
|-- sink_capabilites:1:fixed_supply:dual_role_power
|-- sink_capabilites:1:fixed_supply:fase_role_swap_current
....
|-- sink_capabilites:2:variable_supply:maximum_voltage
|-- sink_capabilites:2:variable_supply:minimum_voltage
...
|-- source_capabilities:1:fixed_supply:dual_role_data
|-- source_capabilities:1:fixed_supply:dual_role_power
|-- source_capabilities:1:fixed_supply:maximum_current
...
But ick, that's also a mess as you are now forced to parse filenames in
userspace in a different way than "normal".
Is there anything special about the number here? It's the "position"
which will be unique. So make that position a device, as that's kind of
what it is (like usb endpoints are devices)
Then you could make a bus for the positions and all would be good, and
you could turn this into:
/sys/class/typec/port0/usb_power_delivery
|-- revision
|-- sink_capabilities:1/
| `-- fixed_supply/
| |-- dual_role_data
| |-- dual_role_power
| |-- fast_role_swap_current
| |-- operational_current
| |-- unchunked_extended_messages_supported
| |-- unconstrained_power
| |-- usb_communication_capable
| |-- usb_suspend_supported
| `-- voltage
|-- sink_capabilities:2/
| `-- variable_supply/
| |-- maximum_voltage
| |-- minimum_voltage
| `-- operational_current
|-- sink_capabilities:3/
| `-- battery/
| |-- maximum_voltage
| |-- minimum_voltage
| `-- operational_power
`-- source_capabilities:1/
`-- fixed_supply/
|-- dual_role_data
|-- dual_role_power
|-- maximum_current
|-- unchunked_extended_messages_supported
|-- unconstrained_power
|-- usb_communication_capable
|-- usb_suspend_supported
`-- voltage
Would that work?
> So would it be OK that, instead of registering these objects as
> devices, we just introduce a kset where we can group them
> (/sys/kernel/usb_power_delivery)?
You want to show this as attched to a specific port somehow, so that
location is not going to work.
thanks,
greg k-h