spectra2
spectra2 on Hacker News
I think a scheme as simple as running two Linux VMs, where you have isolated the services that have real-time guarantees into their own VM, lets you benefit from seL4.
This is tangential to your comment, but it's worth mentioning that you can often begin securing a (Linux) system with seL4 by putting the entire system into a VM. Then, you iteratively port components to run "natively" over seL4 -- a process called "cyber retrofit" [0, 1].
BTW, it was probably just a typo but I believe "se" in seL4 is not a shortening of "secure", but an initialism for "secure embedded". Which is why it's not "secL4".
[0] https://sel4.systems/About/how-to-use.html
[1] https://trustworthy.systems/publications/abstracts/Klein_AKM...
Sorry! I didn't read "side-channel" as also referring to microarchitectural timing channels, but mostly to refer to (architectural) storage channels. You are right, it's not covered at all until the experimental Time Protection work is completed [0].
[0] https://trustworthy.systems/projects/timeprotection/
I think the danger of a side-channel attack against seL4 is more dangerous than you believe. The security proofs ensure that threads should not be able to read data they do not have permission to, or write to places they do not have permission to etc., through any part of the kernel's interface, which includes the mechanisms for inter-process communication. That means a side-channel, or some gap in this proof, would allow learning about what lives in processes!
While seL4 is definitely not a silver bullet for building a secure and stable system, I would argue that its security guarantees mitigate the risks. seL4's security proofs guarantee that the kernel obeys the information flow policy of your system, derived from the runtime distribution of permissions ("capabilities") in your system. If the interface to your filesystem, for example, is through the judicious granting of capabilities, then you can rest assured there is no side-channel.
Not all caches are cleared. There are some clever techniques to partition off-core caches (i.e. the L3) as part of the isolation boundaries enforced by the kernel. What I think is particularly exciting is that a new RISCV64 instruction to clear microarchitectural state is in the process of being standardised! So I'm hopeful that, while it should not be ignored, the performance penalty will not be too high.