[pok-devel] RE: [pok] Two questions about Pok.

[Herve.Peignelin@xxxxxxxxxxxxxxxxxxx]

Hi,

I can't help you on the first question,
but only to the second one:

For questions of determinism, SMP (Symmetric
Multi Processor) model is generally not used in the safety applications
that need spatial and temporal partitioning. AMP (Asymmetric Multi-Processor)
model is generally preferred, which means that each core has its own set
of partitions, and that each partition "belongs" to only one
core.

Running such an AMP software model on
a SMP hardware architecture is sometimes called ASMP.

I think, as you suggest, that running
one scheduler on each core could be sufficient to make POK multi-core friendly.
What I don't know (due to my lack of
POK-specific knowledge) is the impact it will have on the configuration
part: will it be a simple attribute to each partition telling which core
runs the partition, or should each core have its own, independent configuration?
The former case makes the time partitioning less straightforward, the latter
case makes the space partitioning less straightforward. In both cases,
you could need more work to adapt the way POK is configured and configuration
data is checked, than to do the run-time code modification.

Best regards,

ZODIAC AEROTECHNICS AIRCRAFT SYSTEMS Zodiac Sensing & System Managament
Hervé Peignelin Software Manager
61, rue Pierre Curie - CS20001 - 78373 Plaisir Cedex - France Tel: +33 1 61 34 15 38 - Fax: +33 1 61 34 13 01 Herve.Peignelin@xxxxxxxxxxxxxxxxxxx http://www.zodiacaerospace.com

De :      
 Zhen Zhang <zhen.zhang@xxxxxxxx>
A :      
 pok@xxxxxxxxxxxxxxxxxxx,
pok-devel@xxxxxxxxxxxxxxxxxxx
Date :      
 26/09/2014 17:03
Objet :    
   [pok] Two questions
about Pok.

---

Hi,

   I'm working to port Pok to a SMP multicore
system,  two questions troubles me:

1) I try to replace tsim by qemu-sysem-sparc to simulate
the example of partition-scheduling, however that is 
blocked at beginning of execution, do you know the reason?
What should I do?

2) In order to support SMP with a minimal cost, I just
want to create one scheduler per core. 
Do you think that is sufficient? 

Any comment will be welcome.

Best regards
Zhen ZHANG