ir-core.h has the kABI to be used by the bridge drivers, when needing to register
IR protocols and pass IR events. However, the same file also contains IR subsystem
internal calls, meant to be used inside ir-core and between ir-core and the raw
decoders.
Better to move those functions to an internal header, for some reasons:
1) Header will be a little more cleaner;
2) It avoids the need of recompile everything (bridge/hardware drivers, etc),
just because a new decoder were added, or some other internal change were needed;
3) Better organize the ir-core API, splitting the functions that are internal to
IR core and the ancillary drivers (decoders, lirc_dev) from the features that
should be exported to IR subsystem clients.
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
drivers/media/IR/ir-raw-event.c is currently written with the assumption
that all "raw" hardware will generate events only on state change (i.e.
when a pulse or space starts).
However, some hardware (like mceusb, probably the most popular IR receiver
out there) only generates duration data (and that data is buffered so using
any kind of timing on the data is futile).
Furthermore, using signed int's to represent pulse/space durations is a
well-known approach when writing ir decoders.
With this patch:
- s64 int's are used to represent pulse/space durations in ns
- a workqueue is used to decode the ir protocols outside of interrupt context
- #defines are added to make decoders clearer
- decoder reset is implemented by passing a zero duration to the kfifo queue
and decoders are updated accordingly
Signed-off-by: David Härdeman <david@hardeman.nu>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Some spinlocks are not properly initialized on ir core:
[ 471.714132] BUG: spinlock bad magic on CPU#0, modprobe/1899
[ 471.719838] lock: f92a08ac, .magic: 00000000, .owner: <none>/-1, .owner_cpu: 0
[ 471.727301] Pid: 1899, comm: modprobe Not tainted 2.6.33 #36
[ 471.733062] Call Trace:
[ 471.735537] [<c1498793>] ? printk+0x1d/0x22
[ 471.739866] [<c12694e3>] spin_bug+0xa3/0xf0
[ 471.744224] [<c126962d>] do_raw_spin_lock+0x7d/0x160
[ 471.749364] [<f92a01ff>] ? ir_rc5_register+0x6f/0xf0 [ir_rc5_decoder]
So, use static initialization for the static spinlocks, instead of the
dynamic ones (currently used), as proposed by David Härdeman on one
of his RFC patches.
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Reimplement the RC-5 decoder state machine. Code is now clear, and works
properly. It is also simpler than the previous implementations.
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
A previous cleanup patch removed more than needed. Re-add the logic that
disable the decoders.
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Now that the decoders are state machine, there's no need to create
an ancillary buffer while decoding the protocol. Just call the decoders
code directly, event by event.
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
This decoder is also based on a state machine, just like the NEC protocol
decoder. It is pedantic in the sense that accepts only 14 bits. As there
are some variants that outputs less bits, it needs to be improved to also
handle those.
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
