Rubio, Martial wrote:
>  Hi Jan ...
> 
> ... Thanks a lot for your reply ... And the workaround is OK!
> 
> 2 little last precisions about this topic : 
> 
> 	-is it necessary to rename all the "rt"'s entry points as
> "nrt"'s entry points (close,read,write,ioctl) ?

For close, yes, but the rest should be fine - otherwise the problem is
fatal for you. Changing the rest, which is likely supposed to do the
critical work, will leave you without any improvement over plain Linux.
I said "should" because there is the risk that some other code paths of
PCIe MSI handling triggers the bug report as well.

> 	-is the interrupt handler "vxworks_tick_time_module" (when a HW
> source of PEV1100 will be enable) 
> 	running in Linux (not critical) mode ?  

Is that the handler you register with RTDM? Then no, it will be run over
the primary domain, ie. Xenomai.

> 
> It's very easy (by looking at the name of the interrupt handler) to know
> why I ask these questions :
> vxworks_tick_time_module is waiting for an external pulse at 1 ms (and
> other MSI VME interrupts of course)
> provided by an external PCIe device PEV1100 in order to announce tick
> and so schedule an 'old' VME real time 
> Multi-cpu Application (so I'm using also Xenomai VxWorks skins for the
> migration in multi-core environment)
> ... Only VME slave boards will be stay in the distant crate!
> 
> It's perharps a little bit fresh ... But it must go right !!!
> 
> Regards 
> 
> Martial 
> 

Jan