Re: [RFC PATCH 1/3] siox: new driver/bus framework for Eckelmann SIOX

From: Greg Kroah-Hartman
Date: Sat Mar 12 2016 - 22:48:01 EST


On Sat, Mar 12, 2016 at 02:43:21PM +0100, Uwe Kleine-König wrote:
> Hello Greg,
>
> On Fri, Mar 11, 2016 at 02:20:39PM -0800, Greg Kroah-Hartman wrote:
> > On Fri, Mar 11, 2016 at 10:52:12PM +0100, Uwe Kleine-König wrote:
> > > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx>
> > > ---
> > > drivers/Kconfig | 2 +
> > > drivers/Makefile | 1 +
> > > drivers/siox/Kconfig | 2 +
> > > drivers/siox/Makefile | 1 +
> > > drivers/siox/siox-core.c | 572 +++++++++++++++++++++++++++++++++++++++++++++++
> > > drivers/siox/siox.h | 48 ++++
> > > include/linux/siox.h | 53 +++++
> > > 7 files changed, 679 insertions(+)
> > > create mode 100644 drivers/siox/Kconfig
> > > create mode 100644 drivers/siox/Makefile
> > > create mode 100644 drivers/siox/siox-core.c
> > > create mode 100644 drivers/siox/siox.h
> > > create mode 100644 include/linux/siox.h
> >
> > What's the relationship between your "master" and "regular" devices
> > here? I think that's getting things confused, you are unregistering
> > your 'master' yet the children it controls don't go away because you
> > never unregister them. You need to fix this heirachy up a bit, and
> > maybe just drop the 'master' logic?
>
> A master is the bus controller (i.e. it drives DOUT, DCLK and DLD). A
> device is a consumer and provides DIN to the master.
>
> The topology looks as follows:
>
> ,------->--DCLK-->---------------+----------------------.
> ^ v v
> ,--------. ,----------------------. ,------
> | | | ,--------------. | |
> | |--->--DOUT-->---|->-|shift register|->-|--->---|
> | | | `--------------' | |
> | master | | device | | device
> | | | ,--------------. | |
> | |---<--DIN---<---|-<-|shift register|-<-|---<---|
> | | | `--------------' | |
> `--------' `----------------------' `------
> v ^ ^
> `----------DLD-------------------+----------------------'
>
> So there are two chains of shift registers, one for letting the master
> write to the devices, and the other to let the master read from the
> devices. DCLK is the clock to shift both chains by a single bit. An edge
> on DLD is used to make the devices load the inputs and sample the
> outputs.
>
> I think I need the following:
>
> void siox_unregister_master(struct siox_master *smaster)
> {
> for (each siox_device d of smaster)
> siox_device_remove(d);
>
> device_unregister(&smaster->dev);
> }

Yes, you have to do this, as that's how USB and other busses do it when
their "controllers" go away.

> But the locking scheme that is necessary isn't obvious for me. For
> example I must prevent that after the for loop is done a new device is
> registered. Do I need to prevent this explicitly (e.g. by doing
> something like:
>
> lock(smaster);
> smaster->is_going_away = true;
> for (each siox_device d of smaster)
> siox_device_remove(d);
> unlock(smaster);
>
> device_unregister(&smaster->dev);
>
> and let registering new devices fail if smaster->is_going_away is true?)
> Or is there something that the kobject stuff/driver core can help me
> with?

Look at what happens when you remove a USB host controller with the
locking there, it shouldn't be that hard.

thanks,

greg k-h