Re: [PATCH 2/2] Remove input_call_hotplug

[Dmitry Torokhov]

On Wed, 19 Jan 2005 15:16:08 +0100, Hannes Reinecke <hare@xxxxxxx> wrote:
> Dmitry Torokhov wrote:
> > Hi Hannes,
> >
> > On Wed, 19 Jan 2005 10:59:30 +0100, Hannes Reinecke <hare@xxxxxxx> wrote:
> >
> >>Dmitry Torokhov wrote:
> >>
> >>>But the real question is whether we really need class devices have
> >>>unique names or we could do with inputX thus leaving individual
> >>>drivers intact and only modifying the input core. As far as I
> >>>understand userspace should be concerned only with device
> >>>capabilities, not particular name, besides, it gets PRODUCT string
> >>>which has all needed data encoded.
> >>>
> >>
> >>Indeed. What about using 'phys' (with all '/' replaced by '-') as the
> >>class_id? This way we'll retain compability with /proc/bus/input/devices
> >>and do not have to touch every single driver.
> >>
> >
> >
> > I want to kill phys at some point - we have topology information
> > already present in sysfs in much better form. Can we have a new
> > hotplug variable HWDEV= which is kobject_path(input_dev->dev). If
> > input_dev is not set then we can just dump phys in it. And the class
> > id will still be inputX. Will this work?
> >  
> Sure. And we don't need a special HWDEV variable, as there is already a
> PHYSDEVPATH variable providing exactly this information.

Oh right! Even better.

> I'm not too happy about this 'inputX' thing (as this doesn't carry any
> information, whereas 'phys' gives you at least a rough guess what this
> device's about), but if phys is to go it would be the logical choice.
> 

The problem with encoding phys in class ID is the following: you have
to guarantee that the moment you destroy underlying hw device yur
input_device has to be gone too. Imagine you have input_device for
your PS/2 mouse and you decide to unload psmouse module. You also have
one of user processes holding any of the class device attributes open.
This causes input_device to be pinned into memory so when you load
psmouse module back again it will not be able to create new
input_device and mouse will be dead. With monotonicaly increasing
inputX name you will never have this issue.

You had a workaround for this problem with your original path but when
you fold it all in core you lost it.

-- 
Dmitry
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/