Re: Ironclad, the hard-Real Time capable POSIX-like kernel written in SPARK/Ada, received an nlnet grant

[Lawrence D'Oliveiro <ldo@nz.invalid>]

On Sun, 6 Oct 2024 15:53:25 +0200, DrPi wrote:

> Le 06/10/2024 à 01:10, Lawrence D'Oliveiro a écrit :
>>
>> On Sat, 5 Oct 2024 18:24:39 +0200, DrPi wrote:
>> 
>>> You should try QNX.
>> 
>> Was that used in any high-performance situation?
>
> Sure. QNX is designed for hard real time.

Not the same thing. “Hard real time” does not imply “high performance”. 
Ask the folks working on the very subject of this thread.

Is the QNX “microkernel” similar in style to the old Amiga “microkernel”? 
That is, instead of copying message buffers around, it just passed around 
pointers to those message buffers? That would give you speed, but 
sacrifice much of the robustness that a microkernel is supposed to offer.

>> Linux has supported loadable modules for maybe 30 years now.
>
> Yes, they exist but they are some sort of exception.

No they are not. They typically make up the majority of the kernel. 
Looking at my current kernel, the base kernel file is less than 10MB in 
size, whereas the directory containing loadable modules for that kernel is 
about 10× that.

Of course, this is your choice; some prefer to build all the options they 
need into a single monolithic image, for faster loading. Linux is all 
about having the choice.

> And when time comes to debugging a kernel driver...

Hurd should have found that easy to do, if microkernels really make things 
easy to do. But it didn’t.