James Le Cuirot reports that glibc-2.28 for m68k-unknown-linux-gnu when built with --enable-stack-protector=all setjmp()/longjmp() breaks 'dash -c exit'.

Here is the minimal reproducer or a failure (dash's 'exit' uses 'longjmp'):

  // cat m68k-longjmp-bug.c 
  #include <setjmp.h>
  #include <stdio.h>

  int main() {
    jmp_buf jb;
    volatile register int r asm ("a2");
    r = 0x1234;
    if (setjmp(jb) == 0)
        longjmp(jb, 1);
    printf ("r = %x\n", r);
  }

  $ m68k-unknown-linux-gnu-gcc m68k-longjmp-bug.c -fstack-protector-all -o a && ./a
  *** stack smashing detected ***: <unknown> terminated

The bug is very similar in nature to https://sourceware.org/PR22624 where setjmp() was written in C and inserted by compiler canary clobbered to-be-saved register.

This fix is enough to make the example work:

--- a/sysdeps/m68k/setjmp.c
+++ b/sysdeps/m68k/setjmp.c
@@ -17,10 +17,11 @@

 #include <setjmp.h>

 /* Save the current program position in ENV and return 0.  */
 int
+inhibit_stack_protector
 #if defined BSD_SETJMP
 # undef setjmp
 # define savemask 1
 setjmp (jmp_buf env)
 #elif defined BSD__SETJMP