From owner-freebsd-hackers  Fri Dec 29 18:02:42 1995
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id SAA03811
          for hackers-outgoing; Fri, 29 Dec 1995 18:02:42 -0800 (PST)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id SAA03804
          for <freebsd-hackers@FreeBSD.ORG>; Fri, 29 Dec 1995 18:02:30 -0800 (PST)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id MAA07930; Sat, 30 Dec 1995 12:58:38 +1100
Date: Sat, 30 Dec 1995 12:58:38 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199512300158.MAA07930@godzilla.zeta.org.au>
To: creilly@maths.tcd.ie, freebsd-hackers@FreeBSD.ORG
Subject: Re: Linux emulator and Mathematica.
Sender: owner-hackers@FreeBSD.ORG
Precedence: bulk

>Using a vanilla 2.1R installation:

>I've been trying to make Linux Mathematica using a network license,run
>under the linux emulator, and after adding a trivial implementation  of
>...

>The problem now is that mathclient crashes. Looking at the syslog output 
>and the system calls, it would appearing to be crashing on a sigreturn:

>   791 mathclient CALL  sigreturn(0xefbfc0e0,0xe,0)
>   791 mathclient RET   sigreturn -1 errno 1 Operation not permitted
>   791 mathclient PSIG  SIGBUS SIG_DFL
>   791 mathclient NAMI  "mathclient.core"

>The signal it is responding to is a sigalarm.

>Now, from my reading of the sigreturn stuff and the Design and Implementation of
>4.3BSD, it would seem that what is happening is that the sigreturn is trying

The Linux sigreturn() isn't fully implemented in the Linux emulator.  It
returns ENOSYS.

The problem here is probably quite different.  Few applications call
sigreturn() directly.  It is normally called as part of returning from
a signal handler.  The kernel builds a signal trampoline with a
sigreturn() syscall at the end of it.  The Linux emulator should
replace the code that builds the trampoline so that the syscall is
the Linux sigreturn() and not the BSD sigreturn().  It should also
convert the BSD sigcontext struct to the Linux sigcontext sigcontext
struct.  It doesn't do this, so the Linux syscall that happens to
have the same number as the BSD sigreturn (#103 = Linux syslog())
is called instead.  This is fudged to work by breaking the Linux
syslog() by pretending that it is the Linux sysreturn().

>to restore some part of the state that it is not permitted to manipulate. Would
>this seem reasonable to the kernel experts out there? One problem with this

Yes, it's quite likely that Mathematica is sophisticated enough to
manipulate the signal context.  Since the sigcontext struct is
incompatible, this is unlikely to work, and could easily attempt
to change a protected register.  However, this should result in
errno EINVAL, not EPERM.

Bruce