Re: GCC 5.4 mips64el problems

[Michael <macallan%netbsd.org@localhost>]

Hello,

On Fri, 1 Jul 2016 19:17:59 -0500 (CDT)
"John D. Baker" <jdbaker%mylinuxisp.com@localhost> wrote:

> I decided to try building evbmips-mips64el with GCC 5.4 (-V HAVE_GCC=53).
> Starting with a completely empty OBJDIR/DESTDIR, etc., building eventually
> failed with:
> 
> [...]
> --- ramdiskbin ---
> #      link  ramdisk/ramdiskbin
> /d0/build/current-gcc5/tools/amd64/bin/mips64el--netbsd-gcc    --sysroot=/d0/build/current-gcc5/DEST/evbmips   -static  -o ramdiskbin  ramdiskbin.o  -Wl,-rpath-link,/d0/build/current-gcc5/DEST/evbmips/lib  -L=/lib cat.cro chmod.cro cp.cro date.cro dd.cro df.cro ed.cro ln.cro ls.cro mkdir.cro mv.cro pax.cro ps.cro pwd.cro rm.cro rmdir.cro sh.cro stty.cro sync.cro disklabel.cro fdisk.cro fsck.cro fsck_ext2fs.cro fsck_ffs.cro fsck_msdos.cro ifconfig.cro init.cro mknod.cro mount.cro mount_cd9660.cro mount_ext2fs.cro mount_ffs.cro mount_kernfs.cro mount_msdos.cro mount_nfs.cro mount_tmpfs.cro newfs.cro newfs_ext2fs.cro newfs_msdos.cro ping.cro reboot.cro restore.cro route.cro shutdown.cro slattach.cro swapctl.cro sysctl.cro umount.cro chown.cro ftp.cro gzip.cro less.cro sed.cro tset.cro vmstat.cro chroot.cro flashctl.cro sysinst.cro progress.cro  libhack.o -ledit -lutil -lcurses -lterminfo -lkvm -lrmt -lcrypt -ll -lm -llzma -lbz2 -lz -lprop    
> /d0/build/current-gcc5/DEST/evbmips/usr/lib/libc.a(softfloat.o): In function `__truncdfsf2':
> softfloat.c:(.text+0x35f8): multiple definition of `__truncdfsf2'
> /d0/build/current-gcc5/DEST/evbmips/usr/lib/libgcc.a(truncdfsf2.o):(.text+0x0): first defined here
> collect2: error: ld returned 1 exit status
> *** [ramdiskbin] Error code 1
> nbmake[7]: stopped in /d0/build/current-gcc5/obj/mips64el/distrib/evbmips/instkernel/ramdisk
> 1 error
> [...]

That looks like some softfloat support is messed up. Weird.
I haven't seen that but I didn't build softfloat userlands lately.

> I understand others have been experimentally building evbmips (-mips64eb
> at least) w/GCC 5.4.  Did you run into this and if so, what works around
> it?

I have userlands and kernels built with binutils 2.26 and gcc 5.4
working on Loongson ( LP64 kernel, n32 userland ), sgimips64
( kernel and userland n32 ), both with MKSOFTFLOAT=no, and on CI20
( mips32r2, so everything is o32 ).
Kernels work fine, userland on 64bit has some warts left - trunc.d.*
instructions cause SIGILLs on sgimips64 and I get other, weirder
problems on Loongson, mostly when building stuff from pkgsrc. Also, the
compiler still defaults to softfloat output without -mhard-float for
some reason. I suspect the root cause is the same on both though.

have fun
Michael