PR/55990 CVS commit: src/sys/arch/m68k/m68k

["Izumi Tsutsui" <tsutsui%netbsd.org@localhost>]

The following reply was made to PR port-m68k/55990; it has been noted by GNATS.

From: "Izumi Tsutsui" <tsutsui%netbsd.org@localhost>
To: gnats-bugs%gnats.NetBSD.org@localhost
Cc: 
Subject: PR/55990 CVS commit: src/sys/arch/m68k/m68k
Date: Sun, 20 Sep 2026 03:58:19 +0000

 Module Name:	src
 Committed By:	tsutsui
 Date:		Sun Sep 20 03:58:19 UTC 2026
 
 Modified Files:
 	src/sys/arch/m68k/m68k: reenter_syscall.s
 
 Log Message:
 m68k: remove invalid diagnostic check inside #ifdef DEBUG
 
 This could happen on the following case:
 ```
  (libc / sigtramp -> trap #0)
  syscall(SYS_setcontext, frame)
   (p->p_md.md_syscall)()
    syscall_plain()
     sy_call(callp, ...)
      (*callp->sy_call)()
       sys_setcontext()
        setucontext()
         cpu_setmcontext()
          (saved mcontext format >= FMT4 && f_stackadj == 0)
          reenter_syscall(frame, sz)
           (relocate trapframe downward by sz and set f_stackadj = sz)
 
           syscall(SYS_setcontext, relocated frame)
            (p->p_md.md_syscall)()
             syscall_plain()
              sy_call(callp, ...)
               (*callp->sy_call)()
                sys_setcontext()
                 setucontext()
                  cpu_setmcontext()
                   (f_stackadj == sz, so restore long exception frame,
                    then f_stackadj = 0)
 
                return EJUSTRETURN from 2nd sys_setcontext()
              return from sy_call()
             return from syscall_plain()
 
            userret()
             ==> PR/55990 triggering case:
                 a pending signal is delivered at this user-return point
             mi_userret()
              lwp_userret()
               (LW_PENDSIG is set;
                issignal() returns a pending signal)
               postsig()
                kpsendsig()
                 (*p->p_emul->e_sendsig)()
                  sendsig()
                   sendsig_siginfo()
                    cpu_getmcontext()
                     (save longframe to mcontext,
                      set f_format = 0, and
                      f_stackadj += sz)
 
            return from userret() in the 2nd syscall()
           return from the 2nd syscall()
          epilogue of reenter_syscall()
           ==> f_stackadj != 0 here;
               reenter_syscall.s rev 1.6 and prior did not handle this case,
               then caused PR/55990
 ```
 so this f_stackadj == 0 check should have been removed in rev 1.7.
 
 
 To generate a diff of this commit:
 cvs rdiff -u -r1.7 -r1.8 src/sys/arch/m68k/m68k/reenter_syscall.s
 
 Please note that diffs are not public domain; they are subject to the
 copyright notices on the relevant files.