OSDev.org

25 years too late, *sighs* well here it goes.

1. On a multi core system, should you allow your kernel to be called in by several instances at the same time on different cores in a system where access to several interfaces is limited? (do not assume the system to be a homogeneous in design)

Flip side,
2. Should hardware if allowed to do so "DMA" while the processor might need the same resources.

Okay let's clarify this topic a bit.

On the first point
ARM has two chips in their portfolio which are suppose to work together.
ARM cortex-A15 and ARM cortex-A7.

Also IBM has it's Cell processor.

And even 15 years ago there were Amiga's with a MC68K version CPU and a PPC co-processor.

Berkus your completely right with your answer but it wasn't what i was looking for.

When programming for a system that can offload general code to a co processor is wise to do so?

CrypticalCode0 wrote:Okay let's clarify this topic a bit.

On the first point
ARM has two chips in their portfolio which are suppose to work together.
ARM cortex-A15 and ARM cortex-A7.

Also IBM has it's Cell processor.

And even 15 years ago there were Amiga's with a MC68K version CPU and a PPC co-processor.

Berkus your completely right with your answer but it wasn't what i was looking for.

When programming for a system that can offload general code to a co processor is wise to do so?

This is a model I'm working on now for my OS. I have so called jobs which similar to the thread pool model. However, a job can be mapped to a completely different CPU cluster. For example if you have PowerPC as main CPU, it is possible to start jobs in specialized slave processors. This was the criteria I was working after in order to make it possible to spread the work load even to alien processors.

Also ARM Cortex A15 and A7 has the same memory architecture. I think they made it on purpose so that it could be easily adapted by Linux multicore support. This is the "big core" and "little core" strategy they have in mind.