Michael S (already5chosen.delete@this.yahoo.com) on March 18, 2021 3:32 am wrote:
> NvaxPlus (anon.delete.delete@this.this.example.com) on March 17, 2021 11:48 am wrote:
> > Linus Torvalds (torvalds.delete@this.linux-foundation.org) on March 16, 2021 7:25 pm wrote:
> > > Why is this even a discussion any more? Microkernels failed.
> > > Give them up. You want a monolithic kernel. End of story.
> > >
> >
> > What possible, useful sense is this true? There are billions
> > and billions of devices running L4. Stop trafficking
> > in tropes that were stale 20 years ago. Exactly nothing about microkernels notionally implies share-nothing
> > message passing between address space isolated processes
> > as the sole "IPC" (taken broadly) mechanism. In fact,
> > most don't even work that way. Amazingly, kernel development has actually progressed since Mach.
> >
>
> Is L4 in Qualcomm modems really used as microkernel, with all services running in separate
> HW-protected memory domains and with each service having at least one thread of its own?


I have no idea what Qualcomm does, but I also doubt very much that those small microcontrollers have any other kind of memory protection except for the kernel and maybe not even that.

Your doubt is extremely relevant, because in the absence of multiple separate HW-protected memory domains there is no difference in the security properties of a microkernel and of a monolithic kernel, so the claim that there are billions of devices running microkernels would be false.

However, without HW-memory protection, i.e. without paging and/or segmentation, and likely also without caches, the context switches would be very fast so there will also be no differences in performance between a kernel structured like a microkernel and a kernel structured as a monolithic kernel.


My guess is that the choice of L4 in Qualcomm devices has nothing to do with the fact that on CPUs with memory protection it is possible to use L4 as the base of a microkernel, but everything to do with the fact that L4 is a small and bug-free kernel (formally proved) and for the reduced needs of an embedded microcontroller only relatively little code must be added to the kernel, so the additional custom code can hopefully be also bug-free.