On Thursday, March 21, 2013 3:46 AM, Ian Abbott wrote: > On 2013-03-20 15:50, H Hartley Sweeten wrote: >> On Wednesday, March 20, 2013 4:46 AM, Ian Abbott wrote: >>> On 2013-03-20 01:07, H Hartley Sweeten wrote: >>>> Each comedi_subdevice uses one of the minors allocated to the >>>> comedi subsystem. The devices used with this driver need a minimum >>>> of 2 and a maximum of 5 subdevices. >>>> >>>> Only allocate and initialize the number of subdevices actually needed >>>> by the device. >>> >>> This will break user-space code using one of these devices that assume >>> the previous subdevice numbers are still valid. >> >> Fair enough. I'll drop this piece. >> >> But, it seems like a waste to allocate the space for the unused subdevices and >> waste the minors. I'm not sure how much space each struct comedi_subevice >> takes but they probably add up. > > It would be possible to add a layer of indirection to effectively pack > the unused subdevices, although that would increase the overhead for > used subdevices and complicate the code. Probably not worth it. No, not worth it. It will just add additional complexity and probably end up being a maintenance nightmare. To be fair, userspace should probably not depend on a given subdevice type being located at a number. That's what the ->type is for... > The subdevice minors are allocated dynamically to those subdevices that > support asynchronous commands, so unused subdevices won't waste the minors. Ah, overlooked that. Only the dev->read_subdev and dev->write_subdev get an additional minor allocated. But, we still waste the memory that is allocated for the unused comedi_subdevice. Regardless, I have dropped this piece of the series. Regards, Hartley _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
