Indan Zupancic

Also, we don't have that many FPU tests to increase the time significantly. Considering the increase, it seems whatever options we use don't work for Clang and result in software...

Maybe it helps to run `ninja -v` and compare all flags to see what the difference is compared to when it was fast.

Can you guarantee that this isn't necessary on all RISCV implementations supported by seL4? There is usually a difference between theory and practice.

> > Can you guarantee that this isn't necessary on all RISCV implementations supported by seL4? > > There is usually a difference between theory and practice. > > The...

> * things like inequality (>) have odd behaviour when done on types of mix signedness e.g. (int)(-1) < (unsigned int)(1) can return false since -1 becomes 255. But then...

> https://godbolt.org/z/bf8K6Krer > > `enum value { A, B }` is never compatible with its `int` even if the sizeof fits, that wouldn't make sense. Also, two different enum types...

> * deletes isr_el1 from the context since it is read-only. It could be that the user-space can spy the kernel using the register. To elaborate, ISR_EL1 shows the physical...

Requiring an RFC for a clear bugfix like this seems a bit heavy handed, but because the public API is tied to the kernel implementation, it does break the API....

@JorgeMVP I updated the branch with rebase, so it will diverge from your local checkout. Apologies if I caused you any trouble.

I think you should also mention the API changes in CHANGES.md. Are you aware of any other system registers that share state between EL2 and El1?