Re: [PATCH 0/19] New set of input patches

Re: [PATCH 0/19] New set of input patches

[Dmitry Torokhov]

Vojtech Pavlik wrote:
> On Mon, Jun 28, 2004 at 02:13:58AM -0500, Dmitry Torokhov wrote:
> 
> > > IMO drivers have no bussiness messing with the serio locks. We could
> use
> > > 'plug' and 'unplug' functions like the network driver use, or handle
> it
> > > inside the driver, but taking the lock is the wrong thing to do.
> >
> > OK, I just don't want to introduce another lock just for that...
> 
> I think a bit in flags "PSMOUSE_ENABLED", like we have the
> "ATKBD_ENABLED" bit might be just fine - handle the interrupt, but throw
> away the data during the protocol switch. We aren't interested in the
> data anyway.
>

But the flag will not give you atomicity of resetting other fields, like
pktcount. I guess we can ensure it by carefully rearranging the states
and what is reset at what point but it is too fragile.

Would you accept a pair serio_rx_suspend/serio_rx_resume that would still
take the lock internally but not expose this fact to the driver?
 
> > > > 15-synaptics-passthrough-handling.patch
> > > >         - If data looks like a pass-through packet and tuchpad has
> > > >           pass-through capability do not pass it to the main handler
> > > >           if child port is disconnected.
> > >
> > > I'll have to look closer on this one - I think we want to pass the
> data
> > > to the serio layer even if there is no driver listening on the
> > > passthrough serio.
> >
> > We probably should issue serio_interrupt on child port to force rescan
> but
> > that packet has no business in parent's motion handling routine and
> that's
> > what this patch tries to fix. Anyway, I will look at it more later.
> 
> Indeed, we need it for the rescan. It shouldn't be that hard to fix at
> once.
> 

Ok

> >
> > > > (*) These patches have also been sent to Greg KH.
> > >
> > > Did he accept them already?
> >
> > No, not yet. He promised to take a look at
> platoform_device_register_simple by
> > the end of the week but I guess kernel.bkbits.net troubles might
> intervene...
> > And other 2 I just send out today.
> 
> Ok. I'll wait then.

Sysfs changes should be useable even without platform device changes
and I would like start syncing with you. Would you take patches 2 
through 10 (I will drop the legacy_position stuff)?

--
Dmitry

Re: [PATCH 0/19] New set of input patches

[Vojtech Pavlik]

On Mon, Jun 28, 2004 at 07:54:53AM -0700, Dmitry Torokhov wrote:
 
> But the flag will not give you atomicity of resetting other fields, like
> pktcount. I guess we can ensure it by carefully rearranging the states
> and what is reset at what point but it is too fragile.
> 
> Would you accept a pair serio_rx_suspend/serio_rx_resume that would still
> take the lock internally but not expose this fact to the driver?

Yes, but don't call them suspend/resume. That sounds too much like
powermanagement, which it isn't. Network uses start/stop. Block layer
uses plug/unplug - in the sense that you have a pipe, and if you don't
want any more data, you plug it.

> > > > > (*) These patches have also been sent to Greg KH.
> > > >
> > > > Did he accept them already?
> > >
> > > No, not yet. He promised to take a look at
> > platoform_device_register_simple by
> > > the end of the week but I guess kernel.bkbits.net troubles might
> > intervene...
> > > And other 2 I just send out today.
> > 
> > Ok. I'll wait then.
> 
> Sysfs changes should be useable even without platform device changes
> and I would like start syncing with you. Would you take patches 2 
> through 10 (I will drop the legacy_position stuff)?
 
Yes.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: [PATCH 0/19] New set of input patches

[Dmitry Torokhov]

On Monday 28 June 2004 10:07 am, Vojtech Pavlik wrote:
> On Mon, Jun 28, 2004 at 07:54:53AM -0700, Dmitry Torokhov wrote:
> > 
> > Sysfs changes should be useable even without platform device changes
> > and I would like start syncing with you. Would you take patches 2 
> > through 10 (I will drop the legacy_position stuff)?
>  
> Yes.
> 

Vojtech,

As we discussed I dropped the legacy_position sysfs attribute and moved
patches 2 through 10 to my repository on bkbits.net. I also moved patch
#14 because sa1111ps2, gscps2, ambakmi and pcips2 have already been 
integrated with sysfs so linking their serio ports to their devices
are simple one-liners not depending on anything I sent to Greg.

Please do:

	bk pull bk://dtor.bkbits.net/input

-- 
Dmitry

Re: [PATCH 0/19] New set of input patches

[Vojtech Pavlik]

On Tue, Jun 29, 2004 at 02:01:28AM -0500, Dmitry Torokhov wrote:
> On Monday 28 June 2004 10:07 am, Vojtech Pavlik wrote:
> > On Mon, Jun 28, 2004 at 07:54:53AM -0700, Dmitry Torokhov wrote:
> > > 
> > > Sysfs changes should be useable even without platform device changes
> > > and I would like start syncing with you. Would you take patches 2 
> > > through 10 (I will drop the legacy_position stuff)?
> >  
> > Yes.
> > 
> 
> Vojtech,
> 
> As we discussed I dropped the legacy_position sysfs attribute and moved
> patches 2 through 10 to my repository on bkbits.net. I also moved patch
> #14 because sa1111ps2, gscps2, ambakmi and pcips2 have already been 
> integrated with sysfs so linking their serio ports to their devices
> are simple one-liners not depending on anything I sent to Greg.
> 
> Please do:
> 
> 	bk pull bk://dtor.bkbits.net/input
 
Thanks!

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: [PATCH 0/19] New set of input patches

[Sau Dan Lee]

>>>>> "Vojtech" == Vojtech Pavlik <vojtech@suse.cz> writes:

On Mon, Jun 28, 2004 at 12:08:21AM -0500, Dmitry Torokhov wrote:

...
> 08-serio-rebind.patch
>         - allow user to disconnect or rebind serio port by writing appropriate
>           data to it's sysfs attribute:
>               echo -n "psmouse" > /sys/bus/serio/devices/serio0/driver
>               echo -n "none" > /sys/bus/serio/devices/serio0/driver
>               echo -n "reconnect" > /sys/bus/serio/devices/serio0/driver
>               echo -n "rescan" > /sys/bus/serio/devices/serio0/driver

    Vojtech> Very good idea.

...


    >> 10-serio_raw.patch - raw access to serio data ala 2.4
    >> /dev/psaux

    Vojtech> OK, finally those who insist on /dev/psaux can shut up


Now, guess who originated these ideas!?




-- 
Sau Dan LEE                     李守敦(Big5)                    ~{@nJX6X~}(HZ) 

E-mail: danlee@informatik.uni-freiburg.de
Home page: http://www.informatik.uni-freiburg.de/~danlee

[PATCH 0/19] New set of input patches

[Dmitry Torokhov]

Hi,

Here is the 2nd version of my set of input patches that mostly do serio sysfs
integration. Among the changes - dropped psmouse KVM resync patch as it was
bogus, added platform devices to those of serio providers that don't have
proper parent device (i8042, q40kbd, etc).

01-psmouse-state-locking.patch
        - Acquire underlying serio lock when changing psmouse state to
          prevent interrupt handler running on us

02-serio-connect-mandatory.patch
        - Make serio driviers connect/disconnect methods mandatory as these
          methods open/close serio ports and link ports and drivers together.
          Presently if a driver does not implement connect method it will not
          be able to bind to a port anyway.

03-serio-rename-1.patch
        - Rename serio->driver to serio->port_data as with sysfs integration
          driver is not the best name for arbitrary data

04-serio-rename-2.patch
        - Rename serio_dev to serio_driver as they are drivers in sysfs sense

05-serio-dynamic-alloc.patch
        - Switch from static to dynamic serio port allocation so serio ports
          drivers can be freely unloaded even if not all references to ports
          are dropped (sysfs req.)

06-serio-no-recursion.patch
        - Do not do recursive discovery of children ports, needed for sysfs
          and generally better for stack usage.

07-serio-sysfs.patch
        - sysfs integration. Register ports and drivers in driver model, link
          them all together under /sys/bus/serio. Every driver has a default
          attribute "descrip[tion" (from serio_driver->description); serio
          ports have "decription", "driver" and "legacy_position".
          Legacy_position porvides access to serio->phys to allow matching
          with data in /proc/bus/input/devices

08-serio-rebind.patch
        - allow user to disconnect or rebind serio port by writing appropriate
          data to it's sysfs attribute:
              echo -n "psmouse" > /sys/bus/serio/devices/serio0/driver
              echo -n "none" > /sys/bus/serio/devices/serio0/driver
              echo -n "reconnect" > /sys/bus/serio/devices/serio0/driver
              echo -n "rescan" > /sys/bus/serio/devices/serio0/driver

09-serio-manual-bind.patch
        - allow marking some drivers as requiring manual bind (to be used when
          driver does not do automatic HW discovery)

10-serio_raw.patch
        - raw access to serio data ala 2.4 /dev/psaux

(*) 11-platform-device-simple.patch
        - Add platform_device_register_simple to register platform devices
          requiring minimal resource and memory management. The release
          function resides in driver core and that allows drivers registering
          such devices be unloaded without waiting for the last reference to
          be dropped.
	
12-i8042-to-platform-device.patch
        - Convert i8042 to platform device instead of system device so
          its ports have proper parent.

13-serio-add-platform-devices.patch
	- Add platform devices to ct82c710, maceps2, q40kbd and rpckbd.

14-serio-set-up-parents.patch
        - Set up parent devices for serio ports in ambakmi, gscps2,
          pcips2 and sa1111ps2.c

15-synaptics-passthrough-handling.patch
        - If data looks like a pass-through packet and tuchpad has
          pass-through capability do not pass it to the main handler
          if child port is disconnected.

(*) 16-bus-default-drv-attrs.patch
        - Add bus' default driver attributes (similar to defaulr device
          attributes)

17-serio-use-bus-default-attrs.patch
        - Use bus' default driver and device attributes to manage serio
          attributes 

(*) 18-add-driver-find.patch
        - Add driver_find function, similar to device_find, to search for a
          driver by its name.

19-serio-use-driver-find.patch
        - Use driver_find in serio_rebind_driver instead of implementing it
          locally.

(*) These patches have also been sent to Greg KH.

This time I tried compiling the stuff using defconfing for SPACR64 and PPC64
(thanks to Andrew for pointing to me availability of cross-compile
tools), so I should be a bit better now... Alpha failed on cpumask.h...

The patches are against today's Linus tree + recent pull form Vojtech's tree
+ recent pull from Greg KH's tree. I have patches against 2.6.7 that will
bring it to my version of the tree at:

	http://www.geocities.com/dt_or/input/2_6_7/

-- 
Dmitry

Re: [PATCH 0/19] New set of input patches

[Vojtech Pavlik]

On Mon, Jun 28, 2004 at 12:08:21AM -0500, Dmitry Torokhov wrote:
> Hi,
> 
> Here is the 2nd version of my set of input patches that mostly do serio sysfs
> integration. Among the changes - dropped psmouse KVM resync patch as it was
> bogus, added platform devices to those of serio providers that don't have
> proper parent device (i8042, q40kbd, etc).
> 
> 01-psmouse-state-locking.patch
>         - Acquire underlying serio lock when changing psmouse state to
>           prevent interrupt handler running on us

IMO drivers have no bussiness messing with the serio locks. We could use
'plug' and 'unplug' functions like the network driver use, or handle it
inside the driver, but taking the lock is the wrong thing to do.

> 02-serio-connect-mandatory.patch
>         - Make serio driviers connect/disconnect methods mandatory as these
>           methods open/close serio ports and link ports and drivers together.
>           Presently if a driver does not implement connect method it will not
>           be able to bind to a port anyway.

OK

> 03-serio-rename-1.patch
>         - Rename serio->driver to serio->port_data as with sysfs integration
>           driver is not the best name for arbitrary data

OK

> 04-serio-rename-2.patch
>         - Rename serio_dev to serio_driver as they are drivers in sysfs sense

OK

> 05-serio-dynamic-alloc.patch
>         - Switch from static to dynamic serio port allocation so serio ports
>           drivers can be freely unloaded even if not all references to ports
>           are dropped (sysfs req.)

OK

> 06-serio-no-recursion.patch
>         - Do not do recursive discovery of children ports, needed for sysfs
>           and generally better for stack usage.

Yes, this is very good.

> 07-serio-sysfs.patch
>         - sysfs integration. Register ports and drivers in driver model, link
>           them all together under /sys/bus/serio. Every driver has a default
>           attribute "descrip[tion" (from serio_driver->description); serio
>           ports have "decription", "driver" and "legacy_position".
>           Legacy_position porvides access to serio->phys to allow matching
>           with data in /proc/bus/input/devices

I'm not sure if we really need the 'legacy position' thing. We probably
should drop the 'phys' stuff some time after we transition to sysfs, and
exporting it through sysfs will make that harder.

> 08-serio-rebind.patch
>         - allow user to disconnect or rebind serio port by writing appropriate
>           data to it's sysfs attribute:
>               echo -n "psmouse" > /sys/bus/serio/devices/serio0/driver
>               echo -n "none" > /sys/bus/serio/devices/serio0/driver
>               echo -n "reconnect" > /sys/bus/serio/devices/serio0/driver
>               echo -n "rescan" > /sys/bus/serio/devices/serio0/driver

Very good idea.

> 09-serio-manual-bind.patch
>         - allow marking some drivers as requiring manual bind (to be used when
>           driver does not do automatic HW discovery)

OK

> 10-serio_raw.patch
>         - raw access to serio data ala 2.4 /dev/psaux

OK, finally those who insist on /dev/psaux can shut up

> (*) 11-platform-device-simple.patch
>         - Add platform_device_register_simple to register platform devices
>           requiring minimal resource and memory management. The release
>           function resides in driver core and that allows drivers registering
>           such devices be unloaded without waiting for the last reference to
>           be dropped.
> 	
> 12-i8042-to-platform-device.patch
>         - Convert i8042 to platform device instead of system device so
>           its ports have proper parent.

OK

> 13-serio-add-platform-devices.patch
> 	- Add platform devices to ct82c710, maceps2, q40kbd and rpckbd.

OK

> 14-serio-set-up-parents.patch
>         - Set up parent devices for serio ports in ambakmi, gscps2,
>           pcips2 and sa1111ps2.c

OK

> 15-synaptics-passthrough-handling.patch
>         - If data looks like a pass-through packet and tuchpad has
>           pass-through capability do not pass it to the main handler
>           if child port is disconnected.

I'll have to look closer on this one - I think we want to pass the data
to the serio layer even if there is no driver listening on the
passthrough serio.

> (*) 16-bus-default-drv-attrs.patch
>         - Add bus' default driver attributes (similar to defaulr device
>           attributes)
> 
> 17-serio-use-bus-default-attrs.patch
>         - Use bus' default driver and device attributes to manage serio
>           attributes 
> 
> (*) 18-add-driver-find.patch
>         - Add driver_find function, similar to device_find, to search for a
>           driver by its name.
> 
> 19-serio-use-driver-find.patch
>         - Use driver_find in serio_rebind_driver instead of implementing it
>           locally.

Very good.

> (*) These patches have also been sent to Greg KH.

Did he accept them already?

> This time I tried compiling the stuff using defconfing for SPACR64 and PPC64
> (thanks to Andrew for pointing to me availability of cross-compile
> tools), so I should be a bit better now... Alpha failed on cpumask.h...

Great! ;)

> The patches are against today's Linus tree + recent pull form Vojtech's tree
> + recent pull from Greg KH's tree. I have patches against 2.6.7 that will
> bring it to my version of the tree at:
> 
> 	http://www.geocities.com/dt_or/input/2_6_7/

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: [PATCH 0/19] New set of input patches

[Dmitry Torokhov]

On Monday 28 June 2004 01:52 am, Vojtech Pavlik wrote:
> On Mon, Jun 28, 2004 at 12:08:21AM -0500, Dmitry Torokhov wrote:
> > Hi,
> > 
> > Here is the 2nd version of my set of input patches that mostly do serio sysfs
> > integration. Among the changes - dropped psmouse KVM resync patch as it was
> > bogus, added platform devices to those of serio providers that don't have
> > proper parent device (i8042, q40kbd, etc).
> > 
> > 01-psmouse-state-locking.patch
> >         - Acquire underlying serio lock when changing psmouse state to
> >           prevent interrupt handler running on us
> 
> IMO drivers have no bussiness messing with the serio locks. We could use
> 'plug' and 'unplug' functions like the network driver use, or handle it
> inside the driver, but taking the lock is the wrong thing to do.

OK, I just don't want to introduce another lock just for that...
 
> 
> I'm not sure if we really need the 'legacy position' thing. We probably
> should drop the 'phys' stuff some time after we transition to sysfs, and
> exporting it through sysfs will make that harder.
> 

I'll drop it then...

> 
> > 10-serio_raw.patch
> >         - raw access to serio data ala 2.4 /dev/psaux
> 
> OK, finally those who insist on /dev/psaux can shut up

:)

> 
> > 15-synaptics-passthrough-handling.patch
> >         - If data looks like a pass-through packet and tuchpad has
> >           pass-through capability do not pass it to the main handler
> >           if child port is disconnected.
> 
> I'll have to look closer on this one - I think we want to pass the data
> to the serio layer even if there is no driver listening on the
> passthrough serio.

We probably should issue serio_interrupt on child port to force rescan but
that packet has no business in parent's motion handling routine and that's
what this patch tries to fix. Anyway, I will look at it more later.
  
> 
> > (*) These patches have also been sent to Greg KH.
> 
> Did he accept them already?
>
 
No, not yet. He promised to take a look at platoform_device_register_simple by
the end of the week but I guess kernel.bkbits.net troubles might intervene...
And other 2 I just send out today.

> > This time I tried compiling the stuff using defconfing for SPACR64 and PPC64
> > (thanks to Andrew for pointing to me availability of cross-compile
> > tools), so I should be a bit better now... Alpha failed on cpumask.h...
> 
> Great! ;)
> 
> > The patches are against today's Linus tree + recent pull form Vojtech's tree
> > + recent pull from Greg KH's tree. I have patches against 2.6.7 that will
> > bring it to my version of the tree at:
> > 
> > 	http://www.geocities.com/dt_or/input/2_6_7/
> 

-- 
Dmitry

Re: [PATCH 0/19] New set of input patches

[Vojtech Pavlik]

On Mon, Jun 28, 2004 at 02:13:58AM -0500, Dmitry Torokhov wrote:

> > IMO drivers have no bussiness messing with the serio locks. We could use
> > 'plug' and 'unplug' functions like the network driver use, or handle it
> > inside the driver, but taking the lock is the wrong thing to do.
> 
> OK, I just don't want to introduce another lock just for that...

I think a bit in flags "PSMOUSE_ENABLED", like we have the
"ATKBD_ENABLED" bit might be just fine - handle the interrupt, but throw
away the data during the protocol switch. We aren't interested in the
data anyway.

> > > 15-synaptics-passthrough-handling.patch
> > >         - If data looks like a pass-through packet and tuchpad has
> > >           pass-through capability do not pass it to the main handler
> > >           if child port is disconnected.
> > 
> > I'll have to look closer on this one - I think we want to pass the data
> > to the serio layer even if there is no driver listening on the
> > passthrough serio.
> 
> We probably should issue serio_interrupt on child port to force rescan but
> that packet has no business in parent's motion handling routine and that's
> what this patch tries to fix. Anyway, I will look at it more later.

Indeed, we need it for the rescan. It shouldn't be that hard to fix at
once.

>   
> > > (*) These patches have also been sent to Greg KH.
> > 
> > Did he accept them already?
>  
> No, not yet. He promised to take a look at platoform_device_register_simple by
> the end of the week but I guess kernel.bkbits.net troubles might intervene...
> And other 2 I just send out today.

Ok. I'll wait then.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: [PATCH 0/19] New set of input patches

[Neil Brown]

On Monday June 28, vojtech@suse.cz wrote:
> On Mon, Jun 28, 2004 at 12:08:21AM -0500, Dmitry Torokhov wrote:
> 
> > 10-serio_raw.patch
> >         - raw access to serio data ala 2.4 /dev/psaux
> 
> OK, finally those who insist on /dev/psaux can shut up
> 

Alternatively, we could say thank you.  So..

   Thank you, very much!

NeilBrown

Re: [PATCH 0/19] New set of input patches

[Vojtech Pavlik]

On Mon, Jun 28, 2004 at 08:32:36PM +1000, Neil Brown wrote:
> On Monday June 28, vojtech@suse.cz wrote:
> > On Mon, Jun 28, 2004 at 12:08:21AM -0500, Dmitry Torokhov wrote:
> > 
> > > 10-serio_raw.patch
> > >         - raw access to serio data ala 2.4 /dev/psaux
> > 
> > OK, finally those who insist on /dev/psaux can shut up
> > 
> 
> Alternatively, we could say thank you.  So..
> 
>    Thank you, very much!
 
You're welcome.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR