Re: [Xenomai-help] measuring context switch primary<->secondary mode

[Jan Kiszka <jan.kiszka@domain.hid>]

[-- Attachment #1: Type: text/plain, Size: 1767 bytes --]

Markus Franke wrote:
> Jan Kiszka wrote:
>> CONFIG_PREEMPT helps here to reduce the worst case (PREEMPT_RT even
>> more, but it currently collides with patching the kernel). On the other
>> hand, it increases the overhead for your Linux subsystem (check with
>> lmbench e.g.). Still, depending on what you need, you are able to tune
>> Linux'es own preemption model freely when using Xenomai and I-pipe.
> 
> So at the moment it is not possible to use PREEMPT_RT-Patch and
> Xenomai/Adeos together?

[adding the list to CC again]

If you browse the Xenomai and I-pipe code, you will find some hints that
this once worked and Xenomai is basically prepared for it.

To combine both patches today, we would have to adopt I-pipe not only to
genirq (which is almost done for 2.6.19), but also to the full
hrtimer+dyntick (clocksource/clockevent) infrastructure. Moreover, quite
some time has passed since the last I-pipe/-rt combo patch, and both
PREEMPT_RT and I-pipe changed since then. Do there might be more subtle
issues hidden for a combination.


Still, this is technically feasible and will come over the time when
PREEMPT_RT continues to "leak" into mainline. The question is if using
the result (CONFIG_IPIPE+CONFIG_PREEMPT_RT) is also what you want:
Combining both strategies also means adding overhead of I-pipe's
paravirtualisations and PREEMPT_RT's IRQ threading & sleeping locks.

The future development, Xenomai 3, therefore aims at rebasing the
nucleus over the -rt patch while keeping its ability to run
alternatively over vanilla + I-pipe. Depending on your requirements -
full system preemptibility vs. focused RT-subsystem - you will then be
able to switch the model without application or driver rewrites.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]