Commit Graph

7 Commits

Author SHA1 Message Date
Luiz Augusto von Dentz e5ed88f328 fifo: Make use of k_queue as implementation
This makes k_fifo functions rely on k_queue and port k_poll to use
k_queue directly.

Once all users of k_fifo migrate to k_queue this should no longer be
needed.

Change-Id: Icf16d580f88d11b2cb89e1abd23ae314f43dbd20
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
2017-02-27 21:20:52 +00:00
Benjamin Walsh 3c1ab5d338 kernel/poll: fix signal.signaled not being set when k_poll() waits
Change-Id: I73d906e4cb4a3d359e1ec193db933a95b4739611
Signed-off-by: Benjamin Walsh <walsh.benj@gmail.com>
2017-02-09 23:54:27 +00:00
Benjamin Walsh 2014ff162e kernel/poll: fix registrations that were not always cleared
Poll events were getting registered even when polling conditions had
already been met, but events with conditions met did not register and
did not increment the number of events registered. This caused a
possible discrepancy between the number of events registered and the
position of the last event registered in the events array.

As soon as one event condition is met, the next ones in the array should
not get registered even if their condition is not met. This is what the
code does now.

Change-Id: Ibcc3b135ec9d3cf463beb9da3f641fec962b34bf
Signed-off-by: Benjamin Walsh <walsh.benj@gmail.com>
2017-02-09 23:54:26 +00:00
Benjamin Walsh 47503e30b2 kernel/poll: refactor is_polling()
It's always called for the current thread.

Change-Id: I6588ae27505e961df5cf82463ca9be90a539685b
Signed-off-by: Benjamin Walsh <walsh.benj@gmail.com>
2017-02-09 23:54:25 +00:00
Benjamin Walsh a304f16773 kernel/poll: add k_poll_signal_init() runtime init
Change-Id: Id5a27f7d25e26a1a71ef87000d35a18777210c19
Signed-off-by: Benjamin Walsh <walsh.benj@gmail.com>
2017-02-03 13:54:01 +00:00
Benjamin Walsh 969d4a7ff1 kernel/poll: add user tag to struct k_poll_event
This will allow users to install a way of finding out what the event and
the objects are used for without looking at the object itself, or to
tag a bunch of objects that belong together.

The runtime init function _does not_ take a tag so that there is no
runtime hit if not needed. The static initializer macro _does_ take the
tag, so that it does not have to be initialized at runtime if needed,
and thus avoids a runtime hit.

Change-Id: I89a36c6f969ff952f9d1673b1bb5136e407535c6
Signed-off-by: Benjamin Walsh <walsh.benj@gmail.com>
2017-02-03 13:53:59 +00:00
Benjamin Walsh acc68c1e59 kernel: add k_poll() API
k_poll() is similar to the POSIX poll() API in spirit in that it allows
a single thread to monitor multiple events without actively polling
them, but rather pending for one or more to become ready. Such events
can be a direct event, or kernel objects (currently only semaphores and
fifos).

When a kernel object being polled on is ready, it is not "given" to the
poller: the poller must then acquire it via the regular API for the
object (e.g. k_sem_take()). Only one thread can poll on a particular
object at one time. These restrictions mean that k_poll() is most
effective when a single thread monitors multiple events that are not
subject for contention. For example, being the sole reader on multiple
fifos, or the only thread being signalled by multiple semaphores, or a
combination of both.

Change-Id: I7035a9baf4aa016fb87afc5f5c0f5f8cb216480f
Signed-off-by: Benjamin Walsh <walsh.benj@gmail.com>
2017-02-02 00:30:00 +00:00