On Saturday 19 July 2008 00:59:52 Greg KH wrote:
quoted text
> On Sat, Jul 19, 2008 at 12:38:07AM +0200, Michael Buesch wrote:
> > Thanks for the comments.
> > 
> > On Saturday 19 July 2008 00:10:37 Randy Dunlap wrote:
> > > > +To add a new device, simply echo the configuration string to the "add" file.
> > > > +The config string is composed out of the following elements:
> > > > +
> > > > +DEVNAME DIpin DOpin CLKpin CSpin SPIMODE MAXBUSSPEED NO_SPI_DELAY CSACTIVELOW
> > > > +
> > > > +DEVNAME is a unique name string for the device.
> > > > +DIpin is the SPI DI GPIO pin.
> > > > +DOpin is the SPI DO GPIO pin.
> > > > +CLKpin is the SPI CLOCK GPIO pin.
> > > > +CSpin is the SPI CHIPSELECT GPIO pin.
> > > > +SPIMODE is the hardware mode the device will run at. Can be 0-3.
> > > > +MAXBUSSPEED is the maximum bus speed in Hertz.
> > > > +NO_SPI_DELAY can be 1 or 0. If it is 1, then the lowlevel SPI delay
> > > > +will not be performed. This is not standards compliant, but may be required
> > > > +to gain reasonable speeds on embedded hardware.
> > > > +CSACTIVELOW can be 1 or 0. If it is 1, the chip is considered to be selected, if CS
> > > > +is at a logical 0.
> > > > +
> > > 
> > > Would this be better done via configfs?  sysfs files are supposed to be
> > > single-value files.
> > 
> > 
> > Well, I really want to avoid over-engineering this thing.
> > I thought about using configfs, but that would require to keep lots of state
> > information across the operations.
> > So one would have to allocate a device with mkdir. Then configure the parameters.
> > And then somehow tell the kernel to register it. State has to be maintained over
> > this time and I'm not sure how that "register" operation would look like.
> > Writing a "1" to a "register" file? So why not write all config parameters
> > to an "add" file and be done with all the stuff. ;)
> > 
> > It all depends on how you define "one thing per file", IMO.
> > This "add" file does one thing. It creates a device. We must, of course, pass
> > some configuration parameters, too.
> 
> We have precidence with this for the device id being "added" to a driver
> at run time in PCI and USB.  Not that I really like this, I do think
> configfs would be better to do this with here instead of sysfs.
> 
> That being said, as this is a new sysfs file, please break this
> documentation up into a file in the Documentation/ABI directory instead
> of here.

Ok, let us keep this stuff some time in the experimental -mm tree
and I will think about configfs again. I don't really like that sysfs "add"
file, too. But I can't say if I like configfs more.

I probably need to play with configfs a little bit to find out what's best.
One advantage is that you can easily add more config stuff later without
the risk of breaking stuff. That's quite an advantage, too...
But that "how do we register the device after allocating it?" does worry
me a little bit. Any ideas on how to implement allocate/config/register
in configfs?

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