sys: Use armv7 instead of i386 for 32-bit KBI assertions

[kib]

sys/kern/kern_thread.c

108

Since arm is tier2 (I believe), I think this block can be removed.

sys/sys/user.h

342

I do not think that the changes to sys/user.h are right. We support 32bit x86 userspace, in particular the kinfo sysctls.

[emaste]

sys/kern/kern_thread.c

108

IMO there can be a distinction between the guarantees we want to make to FreeBSD users, and what we want to identify as committers. I'm fine with having these assertions fire on arm and having that prompt an explicit decision, or at least being able to provide a heads-up about a KBI break.

[jhb]

sys/kern/kern_thread.c

108

Note we didn't drop them when i386 went to tier 2, and we don't have any assertions for arm64 which is tier 1. I viewed it more as a sample 64-bit and 32-bit?

sys/sys/user.h

342

But we won't compile the static assertions which are part of the kernel (and which are the only place this constant is used). This change ensures we still have a compile time check that these sizes don't change.

[kib]

sys/kern/kern_thread.c

108

Due to the presence of p_md and td_md fields, we cannot make the asserts generic 64- and 32-bit. They must be arch-specific.

sys/sys/user.h

342

Hmm, might be we need to add some asserts to libprocstat?

[jhb]

sys/kern/kern_thread.c

108

Yes, that's what I meant by "sample", though that phrasing is probably obscure. I meant that if we want assertions for a 32-bit platform at all, it has to be armv7 at this point.

sys/sys/user.h

342

We can do that, but in that case we still will want these on for 32-bit arm so that the assertions in the kernel are checked. Possibly we should just define these always and verify the assertions on all architectures?

[kib]

sys/sys/user.h

342

I pondered this quite long after the initial conversation, and I have two thought, one almost trivial, another might be more controversial.

1. We should assert layout for compat32 kinfo structures. (This is a trivial suggestion). At least the size, as we do now.
2. Since compat32 is part of the amd64 host ABI, we should consider compat32 ABI as Tier 1.

Checking the size for compat32 kinfo_file perhaps requires compiling some throw-away object with -m32.