From owner-freebsd-current  Wed Dec  1  3:50:42 1999
Delivered-To: freebsd-current@freebsd.org
Received: from ns.demophon.com (ns.demophon.com [193.65.70.13])
	by hub.freebsd.org (Postfix) with SMTP id 8016314E80
	for <current@FreeBSD.ORG>; Wed,  1 Dec 1999 03:50:39 -0800 (PST)
	(envelope-from will@ns.demophon.com)
Received: (qmail 29385 invoked by uid 1003); 1 Dec 1999 11:48:59 -0000
Date: 1 Dec 1999 11:48:59 -0000
Message-ID: <19991201114859.29382.qmail@ns.demophon.com>
From: Ville-Pertti Keinonen <will@iki.fi>
To: dillon@apollo.backplane.com
Cc: bde@zeta.org.au, marcel@scc.nl, current@FreeBSD.ORG
In-reply-to: <199912010220.SAA28903@apollo.backplane.com> (message from
	Matthew Dillon on Tue, 30 Nov 1999 18:20:24 -0800 (PST))
Subject: Re: kernel: -mpreferred-stack-boundary=2 ??
Sender: owner-freebsd-current@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG


>     It would be better for GCC to force alignment only within those 
>     procedures that need it rather then force all procedures to guarentee
>     alignment.  Then we could have the best of both worlds.

Determining whether it is necessary is non-trivial, you would have to
go through all local variables after stack slots have been allocated.

I'm not certain whether the current code for alignment is useful at
all, since it depends on the alignment being set properly before the
first compiled function is called.  If the exec code in the kernel or
the library startup code does not ensure this, it is useless to
maintain relative alignment in userland programs.

In fact, as far as I can tell, this *is* currently broken (I didn't
look too carefully, though, and am too far from my -current box to
test it right now).

Fixing this in the kernel would be a bit messy as it may vary
depending on the sysentvec.

Perhaps crt1.c should call main like this (I can't immediately figure
out whether the stack pointer should be correctly aligned before the
return address is pushed or after the previous stack frame is pushed,
this is assuming the former, s/subl/addl/ would change it).

    asm volatile("andl $~15,%%esp;"
		 "subl $4,%%esp;"
		 "pushl %3;"
		 "pushl %2;"
		 "pushl %1;"
		 "call main;"
		 "movl %%eax,(%%esp);"
		 "call exit"
		 : "rm" (argc), "rm" (argv), "rm" (env));

Possibly followed by something to indicate to gcc that it shouldn't
generate code to clean up the stack frame (e.g. for (;;) ; ).


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message