1

In Xenomai's API of Posix skin, I find the following:

POSIX skin.
Clocks and timers services.
Condition variables services.
Interruptions management services.
Message queues services.
Mutex services.
Semaphores services.
Shared memory services.
Signals services.
Threads management services.
Thread cancellation.
Threads scheduling services.
Thread creation attributes.
Thread-specific data.

I can't see anything regarding the file handling and socket programming, so I am guessing that perhaps file handling and sockets are not to be dealt in the real time? Is the guess wrong?

Please guide.

Aquarius_Girl
  • 21,790
  • 65
  • 230
  • 411

2 Answers2

4

Xenomai and its origin, RTAI, both take control of your scheduler, handling the linux kernel itself as a non-real-time thread.

They have provided many services, most of which as you can see is related to threads and synchornization that do NOT call Linux API (in kernel space) or system calls (in user space). As you know, real-time is all about "guaranteeing deadline" and calling Linux violates it (because Linux doesn't guarantee anything).

Since drivers are also important in real-time systems, they have implemented the real-time driver model, or RTDM that helps both implementing and using device drivers in a real-time context.

File handling in kernel is strongly frowned upon. If you are talking about user-space real-time applications, then you can have access to any drivers that is implemented in RTDM. If you don't find one for file handling or sockets, then no you can't use them. Note that even a printf uses Linux system calls and is forbidden.

Note that if you do use them, nothing breaks, you just lose your real-time-ness! I personally do use files for logging, but only call them in case of an error that means real-time is already ruined anyway.

I don't know about Xenomai, but at least in RTAI, if you call a Linux system call, then you get a warning like "RTAI: LXRT changed mode: syscall ..." in your kernel logs.

Shahbaz
  • 46,337
  • 19
  • 116
  • 182
  • That's very helpful. Thanks. Can you point me to a link which talks about dos and don'ts in real time besides what Xenomai says? – Aquarius_Girl Apr 05 '12 at 09:31
  • @AnishaKaul, let me search. Unfortunately, there are not many real-time applications/users for a wide range of information to be available on the internet. – Shahbaz Apr 05 '12 at 09:32
  • Yeah that is the problem. Please if you find something link up here. Thanks. – Aquarius_Girl Apr 05 '12 at 09:33
  • 1
    [This document](https://www.rtai.org/index.php?module=documents&JAS_DocumentManager_op=downloadFile&JAS_File_id=46) is on RTAI 3.4, but it has some good info. Search for `syscall` in the document. Except one part in the middle of some code, all the rest it finds you is worth reading. – Shahbaz Apr 05 '12 at 09:44
  • You said: "Note that even a printf uses Linux system calls and is forbidden." Thanks for the new enlightenment. I think in future if your edit your answers with extra info, it makes sense to inform explicitly though comment. – Aquarius_Girl Apr 05 '12 at 10:03
1

The real-time is a property of the ENTIRE SYSTEM. To achieve property of real-time in the system all its components (including hardware, operating system, drivers, libraries, and applications) should be designed taking into account the requirements applied to real-time systems. Such components (like RTOS) can be used to build a real-time system. But they usage doesn't automatically mean that final system will be a real-time system. Actually, if at least one of the component of your system doesn't support requirements of real time systems, your entire system won't be real-time!

Real-time systems usually has resources significantly exceeding the average requirements of the real-time tasks. Unconsumed resources can be used for performing useful but non-critical background tasks, such as logging, monitoring of the system state, statistics collection and analysis, etc. Applications that will perform this tasks can be designed as non real-time components which run atop of real-time components. This design is safe if you are sure that all components participating in real-time tasks support requirements of real-time. Due to this direct answer to your question:

It completely depends from application. In general, all code, that is not used in handling of real-time tasks, CAN BE written as non real-time. All code which is used in the handling of real-time tasks MUST BE written as real-time.

What the Xenomai is doing is isolation of the non-real time Linux and its activities used for handling of non-real-time tasks in the special container, which is run atop of RTOS kernel and in parallel with RTOS-based real-time tasks. To build real-time system on the Xenomai bases your application should rely only to the Xenomai API and on the other libraries and APIs which are known and are proven to be a real-time. All background activities which can be useful, but completely uncritical can be written as a ordinal Linux applications.

Such systems and services as storage and network services usually are not used in real-time tasks, because the commonly used hardware is very indeterministic and thus doesn't fit well into real-time concept. It is hard to say a priory how much time it will take to send five packets over network or write a file into the HDD. Due to this, interfaces for such systems are not commonplace. But again, the application dictates what real-time services it needs. I can imagine real-time tasks, which involve storage and network actions. In the case of such tasks designer is forced to find such system components, which will provide real-time storage and network services. As you can see, Xenomai is not a candidate.

ZarathustrA
  • 3,434
  • 32
  • 28