Hi Sean, thanks a lot, this is a really nice new feature! On Fri, Jul 13, 2018 at 03:30:06PM +0100, Sean Young wrote: > Once kernel v4.18 is released with IR BPF decoding, this can be merged > to v4l-utils. > > The idea is that IR decoders can be written in C, compiled to BPF relocatable > object file. Any global variables can overriden, so we can supports lots > of variants of similiar protocols (just like in the lircd.conf file). > > The existing rc_keymap file format can't be used for variables, so I've > converted the format to toml. An alternative would be to use the existing > lircd.conf file format, but it's a very awkward file to parse in C and it > contains many features which are irrelevant to us. > > We use libelf to load the bpf relocatable object file. > > After loading our example grundig keymap with bpf decoder, the output of > ir-keytable is: > > Found /sys/class/rc/rc0/ (/dev/input/event8) with: > Name: Winbond CIR > Driver: winbond-cir, table: rc-rc6-mce > LIRC device: /dev/lirc0 > Attached BPF protocols: grundig > Supported kernel protocols: lirc rc-5 rc-5-sz jvc sony nec sanyo mce_kbd rc-6 sharp xmp imon > Enabled protocols: lirc > bus: 25, vendor/product: 10ad:00f1, version: 0x0004 > Repeat delay = 500 ms, repeat period = 125 ms > > Alternatively, you simply specify the path to the object file on the command > line: > > $ ir-keytable -e header_pulse=9000,header_space=4500 -p ./pulse_distance.o > > Derek, please note that you can now convert the dish lircd.conf to toml > and load the keymap; it should just work. It would be great to have your > feedback, thank you. I did a few tests with one of my RC-5 remotes, this lircd.conf file https://github.com/PiSupply/JustBoom/blob/master/LIRC/lircd.conf and kernel 4.18-rc5 on RPi2, with the 32bit ARM kernel and gpio-ir-recv, and on LePotato / aarch64 with meson-ir. lircd2toml.py did a really good job on converting it, the only thing missing was the toggle_bit. When testing the converted toml (with "toggle_bit = 11" added and the obvious volume keycode fixes) I noticed a couple of issues: Buttons seem to be "stuck". The scancode is decoded, key_down event is generated, but after release the key_down events repeat indefinitely - with the built-in rc-5 decoder this works fine. root@upstream:/home/hias/ir-test# ir-keytable -c -w justboom.toml -t Old keytable cleared Wrote 12 keycode(s) to driver Protocols changed to Loaded BPF protocol manchester Testing events. Please, press CTRL-C to abort. 29.065820: lirc protocol(66): scancode = 0x141b 29.065890: event type EV_MSC(0x04): scancode = 0x141b 29.065890: event type EV_KEY(0x01) key_down: KEY_DOWN(0x006c) 29.065890: event type EV_SYN(0x00). 29.570059: event type EV_KEY(0x01) key_down: KEY_DOWN(0x006c) 29.570059: event type EV_SYN(0x00). 29.710062: event type EV_KEY(0x01) key_down: KEY_DOWN(0x006c) 29.710062: event type EV_SYN(0x00). 29.850057: event type EV_KEY(0x01) key_down: KEY_DOWN(0x006c) 29.850057: event type EV_SYN(0x00). 29.990057: event type EV_KEY(0x01) key_down: KEY_DOWN(0x006c) 29.990057: event type EV_SYN(0x00). 30.130055: event type EV_KEY(0x01) key_down: KEY_DOWN(0x006c) 30.130055: event type EV_SYN(0x00). ... Even scancodes, eg KEY_UP / scancode 0x141a, aren't decoded at all, only odd scancodes work. My guess is the manchester decoder could have a problem when the last bit is zero and the message doesn't end with a pulse, but a (rather long) timeout. (Re-)loading a bpf decoder only works 8 times. The 9th attempt gives an error message. # for i in `seq 1 9` ; do ir-keytable -p manchester ; done Protocols changed to Loaded BPF protocol manchester Protocols changed to Loaded BPF protocol manchester Protocols changed to Loaded BPF protocol manchester Protocols changed to Loaded BPF protocol manchester Protocols changed to Loaded BPF protocol manchester Protocols changed to Loaded BPF protocol manchester Protocols changed to Loaded BPF protocol manchester Protocols changed to Loaded BPF protocol manchester Protocols changed to failed to create a map: 1 Operation not permitted Loaded BPF protocol manchester so long, Hias
