API changes to message queues

[zephyrbot]

Reported by Benjamin Walsh:

A potpurri of message queue-related enhancements:

• Change "number of items used" API to "amount of space free" API, since the latter is more useful. (Or else support both?)
  

• Figure out how best to track # of slots in a message queue.
  

** Use uint16_t's, since a queue isn't likely to have more that 64K items?
** Compute the usage from the start and end pointers, and avoid the need for extra variables?

• Allow users to specify alignment of message queue items. (Requires an extra argument during initialization.)
  

** Need to decide how to handle cases where alignment would cause a gap between items . (Do we allow this or force the user to ensure that item size is a multiple of alignment?)

(Imported from Jira ZEP-925)

[zephyrbot]

by Peter Mitsis:

1. See https://gerrit.zephyrproject.org/r/#/c/5339/ for the patch that adds k_msgq_freespace_get().

2. Not going to change how the slots are tracked. Using a uint16_t may save a few bytes in the structure, but can cost more in terms of instructions. For example, on the x86 accessing 16-bit operands requires a special opcode prefix.

3. See https://gerrit.zephyrproject.org/r/#/c/5290/ for the patch that adds an alignment parameter to K_MSGQ_DEFINE().

[zephyrbot]

by Sharron LIU:

msg queue verified @x86 @arc @arm on below test points:
/*TESTPOINT: init via K_MSGQ_DEFINE/
/*TESTPOINT: init via k_msgq_init/
/*TESTPOINT: thread-thread data passing via message queue/
/*TESTPOINT: thread-isr data passing via message queue/
/*TESTPOINT: msgq free get/
/*TESTPOINT: msgq used get/
/*TESTPOINT: msgq purge/
/*TESTPOINT: msgq purge while another thread waiting to put msg/
/*TESTPOINT: msgq put returns -ENOMSG/
/*TESTPOINT: msgq put returns -EAGAIN/
/*TESTPOINT: msgq get returns -ENOMSG/
/*TESTPOINT: msgq get returns -EAGAIN/

Closing this. Thanks.