2017-11-02 12:10 GMT+01:00 Mark Brown <broonie@xxxxxxxxxx>: > On Thu, Nov 02, 2017 at 11:44:07AM +0100, Greg KH wrote: >> On Tue, Oct 31, 2017 at 07:11:53PM +0000, Mark Brown wrote: > >> > There was a discussion a while ago in the context of I2C/SPI MFDs >> > which concluded that if you need a bus and it's going to be effectively >> > noop then you should just use the platform bus as anything else will >> > consist almost entirely of cut'n'paste from the platform bus with some >> > light sed usage and code duplication is bad. It's not super lovely as >> > it's not actually a memory mapped device but it's the best idea we've >> > got. > >> Ugh, I hate that. What's wrong with using a "virtual" device instead? > > It was the duplication, initially everyone was making buses. > >> I can create a "virtual" bus for things like this if they really want a >> "simple" bus, abusing platform for this is the major reason I hate the >> platform bus code... > > In the MFD case they're physical devices, they're just usually on the > wrong side of an I2C or SPI link. Plus MFD already handles platform > devices for things that are memory mapped so it's a bit of a more > natural fit there. What I can do is to register an ion bus (like cec one for example), add one ion parent device so heaps will appear in /sys/bus/ion/ion* and /sys/devices/ion/ion* Does that could sound good enough ? Benjamin -- To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
