V4L/DVB: IR: let all protocol decoders have a go at raw data
On Fri, May 28, 2010 at 3:59 PM, Jarod Wilson <jarod@redhat.com> wrote:
> The mceusb driver I'm about to submit handles just about any raw IR you
> can throw at it. The ir-core loads up all protocol decoders, starting
> with NEC, then RC5, then RC6. RUN_DECODER() was trying them in the same
> order, and exiting if any of the decoders didn't like the data. The
> default mceusb remote talks RC6(6A). Well, the RC6 decoder never gets a
> chance to run unless you move the RC6 decoder to the front of the list.
>
> What I believe to be correct is to have RUN_DECODER keep trying all of
> the decoders, even when one triggers an error. I don't think the errors
> matter so much as it matters that at least one was successful -- i.e.,
> that _sumrc is > 0. The following works for me w/my mceusb driver and
> the default decoder ordering -- NEC and RC5 still fail, but RC6 still
> gets a crack at it, and successfully does its job.
>
> Signed-off-by: Jarod Wilson <jarod@redhat.com>
>
> ---
> drivers/media/IR/ir-raw-event.c | 7 ++++---
>
> diff --git a/drivers/media/IR/ir-raw-event.c b/drivers/media/IR/ir-raw-event.c
> index ea68a3f..44162db 100644
> --- a/drivers/media/IR/ir-raw-event.c
> +++ b/drivers/media/IR/ir-raw-event.c
> @@ -36,14 +36,15 @@ static DEFINE_SPINLOCK(ir_raw_handler_lock);
> */
> #define RUN_DECODER(ops, ...) ({ \
> struct ir_raw_handler *_ir_raw_handler; \
> - int _sumrc = 0, _rc; \
> + int _sumrc = 0, _rc, _fail; \
> spin_lock(&ir_raw_handler_lock); \
> list_for_each_entry(_ir_raw_handler, &ir_raw_handler_list, list) { \
> if (_ir_raw_handler->ops) { \
> _rc = _ir_raw_handler->ops(__VA_ARGS__); \
> if (_rc < 0) \
> - break; \
> - _sumrc += _rc; \
> + _fail++; \
> + else \
> + _sumrc += _rc; \
Self-NAK. The only place we actually *care* about the retval from a
RUN_DECODER() call is in __ir_input_register(), and currently, its
looking for retval < 0, which is currently never possible. When we're
running the decoders, either they fail and return -EINVAL or they
succeed and return 0, and in the register case, we get either a
negative error (ex: -ENOMEM from rc6) or 0, so with the above, _sumrc
will *always* be 0 in the two cases I'm looking at. The third place
where RUN_DECODER gets called (decoder unregister) doesn't care about
the retval either. New patch below, including updated comments about
the macro.
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
diff --git a/drivers/media/IR/ir-raw-event.c b/drivers/media/IR/ir-raw-event.c
index ea68a3f..d3bd3f9 100644
--- a/drivers/media/IR/ir-raw-event.c
+++ b/drivers/media/IR/ir-raw-event.c
@@ -31,8 +31,9 @@
*
* Calls ir_raw_handler::ops for all registered IR handlers. It prevents
* new decode addition/removal while running, by locking ir_raw_handler_lock
- * mutex. If an error occurs, it stops the ops. Otherwise, it returns a sum
- * of the return codes.
+ * mutex. If an error occurs, we keep going, as in the decode case, each
+ * decoder must have a crack at decoding the data. We return a sum of the
+ * return codes, which will be either 0 or negative for current callers.
*/
#define RUN_DECODER(ops, ...) ({ \
struct ir_raw_handler *_ir_raw_handler; \
@@ -41,8 +42,6 @@
list_for_each_entry(_ir_raw_handler, &ir_raw_handler_list, list) { \
if (_ir_raw_handler->ops) { \
_rc = _ir_raw_handler->ops(__VA_ARGS__); \
- if (_rc < 0) \
- break; \
_sumrc += _rc; \
} \
} \