On Friday, August 2, 2019 8:03:17 PM CEST Jeff Kletsky wrote: > > On 8/2/19 7:46 AM, Adrian Schmutzler wrote: > > This converts all remaining devices to use interrupt-driven > > gpio-keys compatible instead of gpio-keys-polled. > > The poll-interval is removed. > > > > Not that this proposed change makes the situation any different, but > many devices have switches that are poorly handled by the "key-press" > approach. > > One specific case that has bothered me (but not enough to dig into it) > is the Archer C7v2 that has an "rfkill" switch. Not only is it > "backwards" (label "Off" is really "wireless on"), but it only responds > to changes in state, so its state at boot is not respected. You can't, > as I recall, set it for "wireless off", plug in the device, and have the > wireless be off when OpenWrt boots. > > The GL-AR300M series and the GL-AR750S also have a multi-position "mode" > switch. > > Right now, all these switches have to be toggled twice to have their > position be properly respected by the OS if they're not in the > "expected" position. > > It would seem that, at some point, switches like these would be better > served by a driver that can both detect position, as well as transition. > This would likely also require a way to poll the position at > "impacted-service start" and ubus support along with changes in existing > hotplug scripts.
>From playing around with gpio-keys and the openwrt's gpio-button-hotplug.c in the past few weeks, I think I can tell you what's happening here. One (there are more) of the problems is that gpio-keys module gets loaded even before the procd enters its "preinit" phase (the module is part of /etc/modules-boot.d/30-gpio-button-hotplug). And the bad news is that even once procd hits the preinit phase, it intentionally forwards everything to the failsafe button events script: | [ "if", | [ "eq", "SUBSYSTEM", "button" ], | [ "exec", "/etc/rc.button/failsafe" ] | ] <https://github.com/openwrt/openwrt/blob/master/package/system/procd/files/hotplug-preinit.json#L15> /etc/rc.button/failsafe itself is also very telling: |#!/bin/sh | |[ "${TYPE}" = "switch" ] || echo ${BUTTON} > /tmp/failsafe_button | |return 0 The long and short of this is that initial switch state event is generated but it has no change of getting processed properly at the time the driver is loaded as the system isn't ready. Note: If it was loaded later when procd is in the "init" phase, then it works because events are then processed by hotplug.json, which does: [ "if", [ "and", [ "has", "BUTTON" ], [ "eq", "SUBSYSTEM", "button" ] ], [ "button", "/etc/rc.button/%BUTTON%" ] ], <https://github.com/openwrt/openwrt/blob/master/package/system/procd/files/hotplug.json#L58> Then everything would work as you expect. so, it's not the driver that lets you down here, because it can't do much about these userspace antics. (Note: OpenWrt's gpio-button-hotplug.c uses the BUTTON subsystem event type for both EV_KEY (button) and EV_SW (switches) events. So don't let this confuse you). Regards, Christian _______________________________________________ openwrt-devel mailing list [email protected] https://lists.openwrt.org/mailman/listinfo/openwrt-devel
