port-m68k/50386: O2-level optimization of gcc 4.8.4/4.8.5 broken on m68k

[okuyama%flex.phys.tohoku.ac.jp@localhost]

>Number:         50386
>Category:       port-m68k
>Synopsis:       O2-level optimization of gcc 4.8.4/4.8.5 broken on m68k
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    port-m68k-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Oct 31 22:45:00 +0000 2015
>Originator:     Rin Okuyama
>Release:        7.99.21 and 7.0
>Organization:
Department of Physics, Tohoku University
>Environment:
NetBSD x68k 7.99.21 NetBSD 7.99.21 (GENERIC) #0: Sun Nov  1 01:54:13 JST 2015  root@XXX:/var/build/obj/m68k/sys/arch/x68k/compile/GENERIC x68k
>Description:
On NetBSD/x68k 7.99.21, xterm(1) starts up, and it echoes back keyboard
input. However, shell is not invoked:
  https://twitter.com/LabDrunker/status/660584606123462656

tsutsui@ and isaki@ told me that they observed same phenomenon in
NetBSD 7.0 release on luna68k and x68k, respectively.

Here is output of ps(1) on remote console:
> x68k% ps
> PID TTY   STAT    TIME COMMAND
> 2765 pts/0 S    0:00.94 -ksh
> 3348 pts/0 O+   0:00.43 ps
> 167 ttye0 Z    0:00.00 (xterm)
> 174 ttye0 I    0:10.64 xterm
> 1042 ttye0 I    0:01.09 -ksh
> 3421 ttye0 I+   0:02.62 xinit

You can see a child process of xterm(1) became a zombie before executing
shell.

When libc is replaced by that built with "DBG=-O1", shell succeeds to
run on xterm(1):
  https://twitter.com/LabDrunker/status/660584995619147776

These facts indicate that O2-level optimization of gcc-4.8.[45] is
broken on m68k.
>How-To-Repeat:
Run xterm(1) on NetBSD/m68k 7.99.21 or 7.0.
>Fix:
N/A