> Hello,
>
> I'm working on a multi-cpus workstation and I allocate the execution of
> real time task to precise CPUs. As tasks are preempting linux kernel, this
> last one executes itself over non loaded CPUs. I want to strictly separate
> the executions of tasks and linux kernel overs CPUs. The reason is that the
> architecture over which I'm working is a NUMA architecture and I have
> noticed that context migrations are a little bit expensive (NUMA makes the
> cost of moving data from one processor to another more expensive). This may
> lead to increase the latency of RT_tasks. Moreover, as my current
> application runs over multiple VME CPUs, I would like the migration to
> Xenomai environment to keep this ressources cutting (for performance,
> cleanliness and abstraction reasons). I think that being able to allocate
> the execution of linux kernel to a precise CPU would be interesting.
>
> Well, I tried some manipulations in order to reach this goal. I tried to
> boot with NOSMP, MAXCPUS or CPUMASK boot options and this has no effect or
> freezes during boot. I haven't found other solution to force the allocation
> of only one CPU ressource to the linux kernel.
kernel options generally are in small caps. The option you are looking
for is "isolcpus", as described in Documetation/cpusets.txt in Linux
sources.
Xenomai allow setting the default CPU affinity with /proc/affinity. Note
however that this is only a default CPU affinity, tasks may choose
another affinity.
--
Gilles.
_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help
