On Fri, May 25, 2018 at 10:17:20AM -0500, Rob Herring wrote: > On Fri, May 25, 2018 at 7:08 AM, Johan Hovold <jo...@kernel.org> wrote: > > On Thu, May 24, 2018 at 11:49:24AM -0500, Rob Herring wrote: > >> On Thu, May 24, 2018 at 7:18 AM, Ricardo Ribalda Delgado > >> <ricardo.riba...@gmail.com> wrote: > >> > Hi Johan, > >> > > >> > On Thu, May 24, 2018 at 2:07 PM Johan Hovold <jo...@kernel.org> wrote: > > > >> >> Serdev currently only supports device tree and ACPI. Using out-of-tree > >> >> code, you could load a device tree fragment during runtime to describe > >> >> your serial bus (or you just amend the device tree). > >> > > >> >> Using device tree overlays would have the benefit of being able to > >> >> describe associated resources (e.g. reset gpios) which a simple > >> >> compatible string (or equivalent) would not. > >> > > >> >> But there are examples where a simple compatible string would do, for > >> >> example an existing CEC device presenting itself as a generic USB CDC > >> >> device (hopefully with a dedicated VID/PID so that no user-space > >> >> configuration is needed at all). > > > >> The fundamental problem here is you need a parent device node to apply > >> a DT overlay to and a USB device hotplugged has no DT device node. The > >> system you are running on may not even have a DT (like a PC). If you > >> have an overlay of the downstream devices, they have to be a child of > >> something for the overlay to apply to. We could just create virtual > >> device nodes for the purposes of applying overlays to. Another option > >> would be allowing multiple DTs. Then you aren't even using overlays > >> (what's the point of an overlay when a system has no real DT to begin > >> with). That also would mean they are completely independent from any > >> real DT or other instances (you may want to plug in multiple of the > >> same device). > > > > Right, there's more (than just a DT overlay loader) that needs to be in > > place before this could be used for the generic hotplug case (and even > > more than that if this was to be usable for ACPI systems). > > > > I think generating DT nodes during enumeration is preferable to having > > detached trees, if only to deal with the case where a loaded overlay do > > overlap with the "real" static tree. > > Generating nodes effectively means we're implementing the full USB > tree as defined for OpenFirmware[1]. I'd prefer to not go there. That > seems a bit pointless as for most of the devices we don't care and > there's really nothing related to USB we care about. We just need to > describe downstream functions. The USB device (or interface) just > happens to be the controller/provider for those functions. > > There should only be overlap with a real tree if devices are soldered > onto a board or use a custom connector and you have other sideband > signals.
Right, we may not want to do it, but having too many mechanisms for doing the same thing can get messy (but so are overlays to begin with, I guess). What we have today (and with the USB serial device tree patches) gets us a long way by covering static setups. And before supporting the generic hotplug case with sideband signals (where the descriptive power of device-tree overlays would be handy) it may be better to focus on the subset where just a compatible string (or equivalent) would do. The USB CEC device I mentioned above would be a good example; and an always-on or DTR/RTS power-controlled GPS could be another. > > Another problem is that we need to deregister any tty device and > > essentially reprobe the underlying serial controller whenever user space > > tells us what can be found on the other end of wire in order to register > > a serdev controller and device instead. IIRC this is something we would > > get for free if using DT overlays (i.e. any device with a modified DT > > node would get removed and readded, or at least notified that something > > changed). > > I did some experiments in this area. Marcel wanted to make all BT > drivers serdev only and then make the BT ldisc just create serdev > devices. And all this would be transparent to existing BT userspace. > As part of this I modified registration to register both serdev ctrlr > and tty char device and allow adding slave devices later. It's up on > my serdev-ldisc-v2 branch in all its hacky glory. Yeah, I remember you mentioning something about that, but I never dared to look at the code. ;) > I don't think we get anything for free with overlays. I think we only > handle platform devices and only new nodes added. There are notifiers > for modifications, but if no one is listening modifications won't have > any effect. Ok, then using DT overlays sounds like too much work for too little gain in any case. The only use case that comes to mind for hotplug + sideband would be to allow people to tinker with evaluation kits (e.g. for a new GPS) on a PC, but they would probably be better off using an embedded system such as the BBB anyway. Johan -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
