i386: Don't use static DPCPU and VNET defines in i386 modules

[dchagin]

Summary

As of c84617e8 a similar to 4802a2cb and b6ea4c5a fix should be
applied to i386 too.

This should fix CI

[jhb]

Hmm, does this mean vnet is now subtly broken on i386 as well now?

[dchagin]

In D41195#938325, @jhb wrote:

Hmm, does this mean vnet is now subtly broken on i386 as well now?

yes, https://ci.freebsd.org/view/Test/job/FreeBSD-main-i386-test/lastBuild/console

[jhb]

I mean more in the way that it's very broken on arm64 because VIMAGE/DPCPU depend on some rather dubious behavior in the kernel linker.

[jrtc27]

I think before -fPIC was added you'd end up with text relocations that would mean it could work, whereas after -fPIC was added it started to do the PC-relative thing, or something like that.

But the list of architectures that can't use static should be the list of architectures that use pre-linked kernel modules, i.e. everything except amd64, so at this point probably it should be flipped to say that.

[dchagin]

In D41195#938333, @jrtc27 wrote:

I think before -fPIC was added you'd end up with text relocations that would mean it could work, whereas after -fPIC was added it started to do the PC-relative thing, or something like that.

I think so, that's why I mention -fPIC commit. Prior to -fPIC commit many modules was broken https://bugs.llvm.org/show_bug.cgi?id=51736

But the list of architectures that can't use static should be the list of architectures that use pre-linked kernel modules, i.e. everything except amd64, so at this point probably it should be flipped to say that.

I agree