The Tanenbaum – Torvalds Debate

Years ago, in school, my professor said that the cost of clearing the page table on context switches killed performance on minix (and microkernels in general)

Given that Linux is now implementing certain mitigations against memory side channels, is the performance gain lost? I assume Linux now clears the page table when switching between kernel and user mode?

Perhaps obviously, I don't know many details on this stuff, I'm curious if others know more?

1992:

https://groups.google.com/g/comp.os.minix/c/wlhw16QWltI

I kind of think if we went back 30+ years and said 'lets start on open OS' it would likely be a good idea to start with a more micro approach.

However the whole debate is totally different today with BPF, library kernels, uni-kernels, OS virtualization (containers) and so on. The much more important need to run memory managed languages on top as well.

Of course all of these questions are not totally new, as always in computing it just new spins.

Now machines are a great deal more powerful I think it is high time to revisit the question and introduce these mechanisms to protect the system and terminate ill behaved drivers or tasks.

I literally can't believe Linus mixes up loose and lose multiple times. Come on man!

wish HN had downvotes