gcc 4.2.1 will delete

[Kyle Evans]

(CC'ing emaste@ as well, because this is a bit of a status report)

On Fri, Aug 23, 2019 at 11:01 PM Warner Losh <imp at bsdimp.com> wrote:
>
> On Fri, Aug 23, 2019, 9:54 PM Mori Hiroki <yamori813 at yahoo.co.jp> wrote:
>
> > Hi
> >
> > gcc 4.2.1 will delete.
> >
> > How about go mips architecture ?
> >
> > https://reviews.freebsd.org/rS351429
>
>
> There is good support for an external toolchain today. Clang has some
> support, but it works a lot better in llvm 9.0,  which is due in about a
> month.
>
> Warner
>

I have the following open work for mips32 w/ in-tree llvm (and llvm90
has the same issues for some of them):
- https://reviews.freebsd.org/D21435 (_mcount needs to be explicitly
placed in .text; clang will emit ".section .mdebug.abi32" before
everything else so _mcount ends up in it -- I don't think we can
really blame clang, we probably should've specified)

- https://people.freebsd.org/~kevans/mips-llvm.diff (.eh_frame ends up
with some relocations we can't have in non-writable section; dirty
dirty hack to make .eh_frame writable in clang and making sure the
synthetic section that lld produces is also writable [this part's even
dirtier than the first])

- https://people.freebsd.org/~kevans/mips-opts.diff (fusefs tests
currently hang and I haven't had time to figure out why or try and
reproduce it against llvm90; the rest of the diff is simply enabling
full LLVM/LLD world)

With these three I can make build and install a world w/ MALTA kernel,
all otherwise modified, and things generally work except kldxref:

Building /boot/kernel/linker.hints
kldxref: /boot/kernel/kernel: too many segments

root@:~ # uname -a
FreeBSD  13.0-CURRENT FreeBSD 13.0-CURRENT #0
b101f7f4b1d-c262228(mips-llvm)-dirty: Tue Aug 27 09:42:00 CDT 2019
root at viper.audeuro.com:/usr/obj/usr/src/mips-llvm/sys/MALTA  mips

Next step, I'll fudge this into freebsd-wifi-build and test it on the
hardware in my test lab...

Thanks,

Kyle Evans