CFI: Function type mismatches reported by -Wincompatible-function-pointer-types-strict

[samitolvanen]

My work-in-progress -Wincompatible-function-pointer-types-strict Clang warning found ~30 more type mismatches in function pointer assignments, which we should look into fixing.

There are roughly the same number of warnings for both x86_64 and arm64 allmodconfigs, mostly duplicates, except for an arch-specific driver or two. All the warnings seem to be automatic integer <-> enum conversions, similar to issue #1703, each of which could potentially trip CFI checking.

cc @kees @nathanchance @GustavoARSilva

[nathanchance]

That does not seem too bad. I can probably take a look at all of those and send patches for the ones that do not have any outstanding patches, as I think some of those should have been caught in #1703.

[nathanchance]

I have written fixes for all the warnings that I could find with my build testing framework. I just need to add commit messages and fully vet the counter changes, as that one was particularly unfortunate for how large the fix was and how much type information was stripped from the callback functions:

https://gist.github.com/nathanchance/69c392675051e992f8a54c1feaa259ed

Everything else was not bad in my opinion. I pinged a couple patches that appeared to be left behind from Huck's initial blast of patches for #1703 that would be relevant for this warning so that I do not duplicate existing work. I should be able to get everything sent out today or tomorrow.

I would like to get this warning enabled for 6.2 if possible, as it would make catching potential CFI failures more reliable (unfortunately, it would not catch a lot of the sysfs ones that I have found but that is going to be nearly impossible at compile time) and the Intel folks upgrade the tip of tree toolchain weekly so we would get quick access to build testing various patches. In order to guarantee that in a reliable manner, it would make most sense for someone like @kees or Andrew Morton to carry the fixes and the enablement patch in the same tree but it might be early enough in the cycle to get the patches all picked up before the merge window then enable the warning before 6.2-rc1.

[kees]

Yeah, I'm happy to carry whatever. The tradition has been "if someone else does a review/ack but the maintainer ignores it, either Gustavo or I will take it via our tree."