On Thu, 2004-08-05 at 10:38, Wolfgang Grandegger wrote: > Hi Philippe, > > I'm currently looking into Xenomai, Fusion, Skins and friends and this > raised a few questions especially concerning usability: > > - How usable is this technologie already today, for example, if I want > to port a PSOS or VxWorks application to Linux/Xenomai using the > corresponding skin. How complete, mature and stable is such a "skin". > Is there any experience with real projects. >
o Completeness As a foreword, all skins have been crafted in order to expose the core API of the target RTOS; this means that you should not miss any arch-independent fundamental service, but the hi-level libraries are not emulated (e.g. VxWorks's taskLib, msgQLib, sem*Lib... ok, but no ansi*, usrLib, dhcpLib and so on) . The missing arch-dependent stuff (i.e. BSP-level) is expected to be implemented during the porting effort using ad-hoc code, because it's impossible to generalize it at Xenomai's level anyway. - The pSOS+ core is quite exhaustively emulated with 56 calls. - VxWorks emulates more than 50 calls too: intLib and portions of intArchLib, kernelLib, msgQLib, sem[BCM]Lib, sysLib, taskLib, taskHookLib, taskInfoLib, tickLib and wdLib. - VRTX is simple and complete. The VRTX/sa profile should be too, even if more recent than the VRTX32 one. - the uITRON skin is a prototype; I don't know if it has even run anywhere. It aims at conforming to the 3.0 specs, and at implementing the 'E'-level support, along with a few strictless but convenient calls from levels 'S' and 'R'. - the POSIX skin is in good shape; it aims at conforming to the PSE51 profile. o Use in real projects To my knowledge, the pSOS skin has been used a few times already (3/4) in porting embedded code that used to run on MVME141 boards to Linux. I never saw the code, just heared about the porting effort. The VRTX and VxWorks skins have been used at least once for a first order port to Linux, the plan being to convert all the API calls to POSIX later. Maybe (probably?) the people involved into these ports had to fix rough edges, but if they did, I cannot tell which ones, because I did not receive any patches, so... To sum up, using those skins is still a pioneering effort; this said, the core that underlies all of them (i.e. the nucleus) is constantly refined since 2001 and in use in some significant projects now (e.g. Hyades), so would some problems arise, they should mainly concern the skin(s) (i.e. the "window dressing" part, not the core). Additionally, the bonus is that those skins all run over the simulation engine, so finding bugs there is much more simple and efficient. Of course, one has to remove the arch-dep sections of the original code and possibly provides for stubs here and there, but the former is already a task that needs to be done when migrating to another hw anyway, so there is no loss of time. > - Fusion is based on Linux 2.6, ADEOS and Xenomai. Does it make sense to > use the "old" skins also for Linux 2.4, which are currently available > in the kileau and vesuvio branch. Since they are based on the nucleus interface which has not changed (except for details), they will work the same way in 3.x and fusion, yes. Prefer 3.1 though, it has received more attention wrt backporting fixes applied to fusion than 3.0. > Hope I have not confused terms like Xenomai, Fusion, Skins too much. > Looks ok. > TIA. > > Wolfgang. -- Philippe.
