The µITRON specification defines a highly flexible operating system architecture designed specifically for application in embedded systems. The specification addresses features which are common to the majority of processor architectures and deliberately avoids virtualization which would adversely impact real-time performance. The µITRON specification may be implemented on many hardware platforms and provides significant advantages by reducing the effort involved in understanding and porting application software to new processor architectures.
Several revisions of the µITRON specification exist. In this release, eCos supports the µITRON version 3.02 specification, with complete “Standard functionality” (level S), plus many “Extended” (level E) functions. The definitive reference on µITRON is Dr. Sakamura book µITRON 3.0, An Open and Portable Real-Time Operating System for Embedded Systems. If you have purchased the eCos Developer's Kit, you will have received a copy of this book. Otherwise, the book can be purchased from the IEEE Press, and an ASCII version of the standard can be found online at http://tron.um.u-tokyo.ac.jp/TRON/ITRON/ .
The eCos kernel implements the functionality used by the µITRON compatibility subsystem. The configuration of the kernel influences the behavior of µITRON programs.
In particular, the default configuration has time slicing (also known as round-robin scheduling) switched on; this means that a task can be moved from RUN state to READY state at any time, in order that one of its peers may run. This is not strictly conformant to the µITRON specification, which states that timeslicing may be implemented by periodically issuing a rot_rdq(0) call from within a periodic task or cyclic handler; otherwise it is expected that a task runs until it is pre-empted in consequence of synchronization or communications calls it makes, or the effects of an interrupt or other external event on a higher priority task cause that task to become READY. To disable timeslicing functionality in the kernel and µITRON compatibility environment, please disable the CYGSEM_KERNEL_SCHED_TIMESLICE configuration option in the kernel package. A description of kernel scheduling is in the section called Thread operations in Chapter 5, and the configuration options for scheduling are detailed in the section called Component: Kernel schedulers in Chapter 14.
For another example, the semantics of task queueing when waiting on a synchronization object depend solely on the way the underlying kernel is configured. As discussed above, the multi-level queue scheduler is the only one which is µITRON compliant, and it queues waiting tasks in FIFO order. Future releases of that scheduler might be configurable to support priority ordering of task queues. Other schedulers might be different again: for example the bitmap scheduler can be used with the µITRON compatibility layer, even though it only allows one task at each priority and as such is not µITRON compliant, but it supports only priority ordering of task queues. So which queueing scheme is supported is not really a property of the µITRON compatibility layer; it depends on the kernel.
In this version of the µITRON compatibility layer, the calls to disable and enable scheduling and interrupts (dis_dsp(), ena_dsp(), loc_cpu() and unl_cpu()) call underlying kernel functions; in particular, the xxx_dsp() functions lock the scheduler entirely, which prevents dispatching of DSRs; functions implemented by DSRs include clock counters and alarm timers. Thus time "stops" whilst dispatching is disabled with dis_dsp().
Like all parts of the eCos system, the detailed semantics of the µITRON layer are dependent on its configuration and the configuration of other components that it uses. The µITRON configuration options are all defined in the file pkgconf/uitron.h, and can be set using the configuration tool or editing this file by hand.
An important configuration option for the µITRON compatibility layer is CYGSEM_UITRON_BAD_PARAMS_RETURN_ERRORS (see the section called Option: Return Error Codes for Bad Params in Chapter 15), which allows a lot of the error checking code in the µITRON compatibility layer to be removed; of course this leaves a program open to undetected errors, so it should only be used once an application is fully debugged and tested. Its benefits include reduced code size and faster execution. However, it affects the API significantly, in that with this option enabled, bad calls do not return errors, but either cause an assert failure (if that is itself enabled: see the section called Component: Use asserts in Chapter 13) or malfunction internally. There is discussion in more detail about this in each section below.
We now give a brief description of the µITRON functions which are implemented in this release. Note that all C and C++ source files should have the following #include statement:
#include <cyg/compat/uitron/uit_func.h> |
The following functions are fully supported in this release:
ER sta_tsk(ID tskid, INT stacd);void ext_tsk(void);void exd_tsk(void);ER dis_dsp(void);ER ena_dsp(void);ER chg_pri(ID tskid, PRI tskpri);ER rot_rdq(PRI tskpri);ER get_tid(ID *p_tskid);ER ref_tsk(T_RTSK *pk_rtsk, ID tskid);ER ter_tsk(ID tskid);ER rel_wai(ID tskid);The following two functions are supported in this release, if enabled with the configuration option CYGPKG_UITRON_TASKS_CREATE_DELETE (see the section called Component: Support create and delete in Chapter 15), with some restrictions:
ER cre_tsk(ID tskid, T_CTSK *pk_ctsk);ER del_tsk(ID tskid);These functions are restricted as follows:
Because of the static initialization facilities provided for system objects, a task is allocated stack space statically in the configuration (see the section called Component: Tasks in Chapter 15). So while tasks can be created and deleted, the same stack space is used for that task (task ID number) each time. Thus the stack size (pk_ctsk->stksz) requested in cre_tsk() is checked for being less than that which was statically allocated, and otherwise ignored. This ensures that the new task will have enough stack to run. For this reason del_tsk() does not in any sense free the memory that was in use for the task's stack.
The task attributes (pk_ctsk->tskatr) are ignored; current versions of eCos do not need to know whether a task is written in assembler or C/C++ so long as the procedure call standard appropriate to the CPU is followed.
Extended information (pk_ctsk->exinf) is ignored.
For all these calls, an invalid task id (tskid) (less than 1 or greater than the number of configured tasks, or a non-existent task) only returns E_ID or E_NOEXS if bad params return errors (CYGSEM_UITRON_BAD_PARAMS_RETURN_ERRORS is enabled, see above).
Similarly, the following conditions are only checked for, and only return errors if CYGSEM_UITRON_BAD_PARAMS_RETURN_ERRORS is enabled:
pk_crtk in cre_tsk() is a valid pointer, otherwise return E_PAR
ter_tsk() or rel_wai() on the calling task returns E_OBJ
the CPU is not locked already in dis_dsp() and ena_dsp(); returns E_CTX
priority level in chg_pri() and rot_rdq() is checked for validity, E_PAR
return value pointer in get_tid() and ref_tsk() is a valid pointer, or E_PAR