This file documents externally visible changes of the native API. Make sure you read it carefully when upgrading versions of this software. ___________________________________________________________________ 0.6.9 -> 0.7 o mlockall() is no longer implicitely issued by the rt_task_create() service in user-space. Applications should make sure to lock the memory pages appropriately by themselves (--rpm 2005-02-22). o The skin now uses an increasing priority scale for tasks which is the same as the Linux one for the SCHED_FIFO scheduling class, i.e. 1 - 99 with 1 being the lowest effective priority level (--rpm 2005-03-13). o The default task stack size in user-space is now PTHREAD_STACK_MIN * 4. A mininum size of PTHREAD_STACK_MIN is enforced in rt_task_create(). ___________________________________________________________________ 0.7.2 -> 0.7.3 o Added basic message-passing support: rt_task_send(), rt_task_receive(), rt_task_reply(). Those work the QNX way. ___________________________________________________________________ 0.7.3 -> 0.7.4 o Renames: rt_pipe_open() -> rt_pipe_create() rt_pipe_close() -> rt_pipe_delete() rt_pipe_read() -> rt_pipe_receive() rt_pipe_write() -> rt_pipe_send() o Added "name" parameter to rt_pipe_create() as 2nd argument, so that message pipe objects can be registered for naming support. o New implementations of rt_pipe_write() and rt_pipe_read() taking raw data buffers instead of canned message buffers. ___________________________________________________________________ 0.8.2 -> 0.8.3 o Added "iack" parameter to rt_intr_create() from the kernel-space API, so that a specific acknowledge routine for the interrupt can be installed. o Fixed rt_timer_ns2ticks and rt_timer_ticks2ns behaviour, so that calling rt_task_sleep(rt_timer_ns2ticks(2000000000)) will sleep two seconds whatever the timer mode is. o Added rt_timer_ns2tsc and rt_timer_tsc2ns to convert between nanoseconds and values returned by rt_timer_tsc. ___________________________________________________________________ 0.8.3 -> 0.9 o Added a timeout parameter to rt_mutex_lock(). Calling rt_mutex_lock(mutex,TM_INFINITE) is equivalent to the former single-parameter syntax. o Added a timeout parameter to all rt_*_bind() services. Calling rt_*_bind(descp,name,TM_INFINITE) is equivalent to the former syntax. o rt_task_set_periodic requires the value passed as a start time to be obtained using rt_timer_read(). o the timer rt_timer_info structure has a "tsc" member, whose value is synchronized with the value of the date member. ___________________________________________________________________ 2.0 -> 2.1-rc1 o Added a parameter to rt_pipe_create(), specifying the size of a local memory pool to associate to the message pipe. Passing zero is equivalent to the former syntax. Additionally, rt_pipe_alloc() and rt_pipe_free() now need to be passed the address of the message pipe descriptor of reference. ___________________________________________________________________ 2.1-rc2 -> 2.1-rc3 o rt_event_wait() has been fixed so that the satisfied event bits are not auto-cleared upon receipt anymore. An explicit call to rt_event_clear() is therefore needed to zero the received bits into the event group. o rt_heap_create() called from user-space with a null "mode" argument now creates a regular mappable heap supporting multiple blocks of memory, instead of a shared memory segment managed as a single-block area. Passing H_SHARED is equivalent to the former behaviour. o rt_timer_start() and rt_timer_stop() have been deprecated. The Xenomai nucleus now automatically starts the system timer according to the configured policy and period when a real-time skin is loaded (see CONFIG_XENO_OPT_TIMING_PERIOD). The new rt_timer_set_mode() service can be used to change the current timing mode and clock tick rate dynamically. ___________________________________________________________________ 2.1-rc3 -> 2.1-rc4 o rt_task_wait_period() now takes a pointer-to-variable parameter which is used to collect the number of pending overruns. Passing NULL tells the API to discard this count. Once collected or discarded, the count of past overruns is cleared, and the timeline is resynchronized to the current period. This update is basically a fix, which allows applications to tell their overrun recovery procedure about the exact number of pending overruns at once.
