On 19/11/16 22:08, Peter Rosin wrote: > On 2016-11-19 16:49, Jonathan Cameron wrote: >> On 17/11/16 21:48, Peter Rosin wrote: >>> When a multiplexer changes how an iio device behaves (for example >>> by feeding different signals to an ADC), this driver can be used >>> create one virtual iio channel for each multiplexer state. >>> >>> Depends on the generic multiplexer driver. >> I'm not really following what all the ext info stuff in here is about. >> Could you add a little more description of that? > > Certainly. I have two needs for this series. The first one is simple > when it comes to the iio part and complex because the mux is shared > between three 8-way muxes on three of the inputs to an ADS-1015 ADC. > The forth ADC line to the ADS-1015 is not muxed. Those three muxes > are of course GPIO-controlled and share GPIO pins. And the GPIO > pins also control an i2c bus that is muxed 8-ways as well. There are > eight (possible) batteries, and we monitor voltage/current/temp with > the 3 muxed ADC lines, and 8 chargers sit on i2c behind the i2c mux. > I guess it felt natural for the HW designer to select battery with > the GPIO lines, but that do not fit too well with the code as it > is without this series... > > For this first need, the iio mux does not need ext_info. > > The second need is simple in the mux part but worse in the iio > department. It's another 8-way mux that simply muxes an ADC line, > so that is simple. However, the ADC line is the envelope detector > that just got added to linux-next, and it has two ext_info > attributes that needs to be set differently for different mux > states. Two of the states need "invert" to be false, the rest need > "invert" to be true. And it is also preferable to have different > values for "compare_interval" for different mux states since the > signals on the diffrent mux states have the different frequency > characteristics. > > True, I could have the ext-info attributes go straight through > the mux, and just start by writing values to "invert" > and "compare_interval", and only then read a sample. But then I > would need to lock out other users during the duration of this > transaction. I believe that the best place to put that lock is > in the iio mux (when it locks its control-mux) and not leave it > to userspace to solve this in some brittle cooperative manner. > >> Perhaps an example of how it is used and what the resulting interface >> looks like? > > The resulting interface is just a copy of the (ext_info) interface > exposed by the parent channel (with a cache that is rewritten to > the parent on every iio mux state change). I have plans to add code > to not rewrite ext_info attributes that have never been changed in > any mux state. > > Below I have an example file listing. > > device0 is a dpot (mcp4561). > device1 is a dac (dpot-dac, wrapping the above dpot). > device2 is an adc (envelope-detector, wrapping the above dac) > device3 is a mux (iio-mux, wrapping the above adc) > > The 8-way iio-mux have no signals attached on mux states 0 and 1, which > is why the first channel for device 3 is in_altvoltage2. > > Ultimately, I would like some knob to hide devices 0, 1 and 2 from > userspace. They need/should only be visible to in-kernel users. Or > is there such a knob already? > There isn't and this feeds into what Lars was suggesting with a fabric driver. The complexity of the description in device tree is getting really very nasty indeed, perhaps we are better off allowing for complex cases to have a driver that directly hooks into all the relevant elements and can do magic channel remapping etc. That could then register the 3 muxes and acquire all the chanenls required to build a single many channel device that hides all the complexity from userspace. I've considered working out how to do an IIO multiplexer before in software which would allow us to make 'fake' devices that wrap multiple physical devices. In the general case it has always felt to complex as we'd want it to handle triggered and buffered data flows. That brings all sorts of nasty data alignment problems with it as there is no guarantee the devices are producing aligned data at all. In the specific case though a driver can bundle up everything it needs to create a pseudo device (for triggered flows we'd probably need a little magic to hold off the trigger but that's not hard). With sysfs only access it would be simple to do but hard to describe. Hence bringing us back to fabric drivers. Thanks for the description. Good to understand what you are trying to handle. Jonathan > Cheers, > Peter > > $ ls /sys/bus/iio/devices/iio\:device* > /sys/bus/iio/devices/iio:device0: > dev out_resistance_raw_available > name out_resistance_scale > of_node power > out_resistance0_raw subsystem > out_resistance1_raw uevent > > /sys/bus/iio/devices/iio:device1: > dev out_voltage0_scale > name power > of_node subsystem > out_voltage0_raw uevent > out_voltage0_raw_available > > /sys/bus/iio/devices/iio:device2: > dev name > in_altvoltage0_compare_interval of_node > in_altvoltage0_invert power > in_altvoltage0_raw subsystem > in_altvoltage0_scale uevent > > /sys/bus/iio/devices/iio:device3: > dev in_altvoltage5_raw > in_altvoltage2_compare_interval in_altvoltage5_scale > in_altvoltage2_invert in_altvoltage6_compare_interval > in_altvoltage2_raw in_altvoltage6_invert > in_altvoltage2_scale in_altvoltage6_raw > in_altvoltage3_compare_interval in_altvoltage6_scale > in_altvoltage3_invert in_altvoltage7_compare_interval > in_altvoltage3_raw in_altvoltage7_invert > in_altvoltage3_scale in_altvoltage7_raw > in_altvoltage4_compare_interval in_altvoltage7_scale > in_altvoltage4_invert name > in_altvoltage4_raw of_node > in_altvoltage4_scale power > in_altvoltage5_compare_interval subsystem > in_altvoltage5_invert uevent > -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html