RTAI/fusion and Ingo Molnar's extension

[Philippe Gerum]

lanas wrote:
> Hi all,
> 
>   Reading the "Life with Adeos" paper I noticed a mention about Ingo
> Molnar's PREEMPT_RT extension in the paragraph describing the ned of a
> fine-grained Linux kernel.  Does this mean that, starting with a vanilla
> kernel, Ingo's extension must be applied first, then the RTAI/fusion
> code ?
> 

You don't need PREEMPT_RT to run fusion. PREEMPT_RT is mentioned as a 
possible future mean to get better determinism from the Linux kernel, 
when your real-time RTAI tasks operate in "relaxed" (i.e. secondary) 
mode, so that they can invoke vanilla Linux services, aside of calling 
native RTAI ones. If your RTAI tasks never call Linux services during 
time-critical operations, thus remaining in primary mode (i.e. "pure" 
RTAI mode), you basically don't even need to care about this anyway: 
latencies will always be close to the hardware limit, since  the tasks 
are managed by fusion's co-scheduler in this mode, and not by the Linux 
kernel.

For now, the reference kernel for fusion is still vanilla 2.6, even if 
we closely track any progress from the PREEMPT_RT patch.
In any case, we will always provide complete patches for using fusion 
over Adeos+PREEMPT_RT combos when the latter has matured enough, since 
applying them incrementally won't work. There is an experimental combo 
patch already in arch/i386/patches/, which is a proof of concept that 
fusion can work on and get benefit from those combined supports, even if 
rough edges remain to be solved. It's called
adeos-linux-2.6.12-rc2-RT-V0.7.44-03-i386-r10c3.patch.

-- 

Philippe.