On Mon, Oct 09, 2017 at 11:14:28AM +0200, Hans Verkuil wrote:
> On 05/10/17 10:45, Sean Young wrote:
-snip-
> > +/*
> > + * struct lirc_scancode - decoded scancode with protocol for use with
> > + * LIRC_MODE_SCANCODE
> > + *
> > + * @timestamp: Timestamp in nanoseconds using CLOCK_MONOTONIC when IR
> > + * was decoded.
> > + * @flags: should be 0 for transmit. When receiving scancodes,
> > + * LIRC_SCANCODE_FLAG_TOGGLE or LIRC_SCANCODE_FLAG_REPEAT can be set
> > + * depending on the protocol
> > + * @target: target for transmit. Unused, set to 0.
> > + * @source: source for receive. Unused, set to 0.
> > + * @unused: set to 0.
> > + * @rc_proto: see enum rc_proto
> > + * @scancode: the scancode received or to be sent
> > + */
> > +struct lirc_scancode {
> > + __u64 timestamp;
> > + __u32 flags;
> > + __u8 target;
> > + __u8 source;
> > + __u8 unused;
> > + __u8 rc_proto;
> > + __u64 scancode;
>
> I'm thinking how this will be implemented using CEC. Some RC commands take arguments
> (up to 4 bytes for the 0x67 (Tune Function) code), so how will they be handled?
>
> See CEC table 6 in the HDMI 1.4 spec.
>
> Should they be part of the scancode, or would it be better to add a '__u8 args[8];'
> field?
>
> I've no idea what makes sense, it's a weird corner case.
I've given it some more thought.
For cec remote control passthrough, you have the tv with the IR receiver (A),
which then transmits CEC_MSG_USER_CONTROL_PRESSED and
CEC_MSG_USER_CONTROL_RELEASED cec messages to the correct target, with
arguments. Then on the target (B), it reads those commands and should execute
them as if it received them itself.
First of all (B) is already implemented in cec using rc-core. If RC
passthrough is enabled, then cec will pass those keycodes to rc-core (which
end up in an input device).
So the problem we are trying to solve here is (A). How I would see this
implemented is:
1) A physical IR receiver exists which has an rc-core driver and a /dev/lircN
device. This is configured using ir-keytable to map to regular input events
2) A process receives input events, and decides that a particular key/command
is not for itself (e.g. tell top set box to tune), so it knows what the
target cec address is and the tune arguments, so it fills out a
cec_msg with the target, CEC_MSG_USER_CONTROL_PRESSED, 0x67, arguments,
and then transmits it using the ioctl CEC_TRANSMIT, followed by
another CEC_MSG_USER_CONTROL_RELEASED cec_msg sent using ioctl CEC_TRANSMIT.
In this way of viewing things, an rc-core device is either cec or lirc, and
thus rc-core lirc devices have a /dev/lircN and rc-core cec devices have a
/dev/cecN.
So, the alternative which is being proposed is that a cec device has both
a /dev/cecN and a /dev/lircN. In this case step 2) would look like:
2) A process receives input events, and decides that a particular key/command
is not for itself (e.g. tell top set box to tune), so it knows what the
target cec address is and the tune arguments, so it fills in a
lirc_scancode with the target, CEC_MSG_USER_CONTROL_PRESSED, 0x67, arguments,
and then transmits it using write() to the /dev/lircN device, which
then passes it on to cec_transmit() in drivers/media/cec/cec-api.c
(without having a cec_fh), and then another lirc_scancode is
filled in CEC_MSG_USER_CONTROL_RELEASED and sent.
Now, I think that this has a number of problems:
- It's a lot of API for simply doing a CEC_TRANSMIT
- and another chardev for a cec device (i.e. /dev/lircN).
- lirc scancode tx deals with scancodes, for cec rc passthrough it isn't
really scancodes.
- Wiring this up is not going to be pretty or easy.
- The lirc chardev has no other function other than sending
CEC_MSG_USER_CONTROL_PRESSED and CEC_MSG_USER_CONTROL_RELEASED cec messages.
So what I am proposing is that we don't use lirc for sending rc passthrough
messages for cec.
I hope this makes sense and where not, please *do* tell me exactly where I
am wrong. I think that I missed something about the scancode tx idea.
>
> > +};
> > +
> > +#define LIRC_SCANCODE_FLAG_TOGGLE 1
> > +#define LIRC_SCANCODE_FLAG_REPEAT 2
>
> These flags need documentation.
They do, fair point.
Thanks,
Sean
