GNU bug report logs - #13332
[PATCH] unbreak Emacs on Linux/M68K

Previous Next

To add a comment to this bug, you must first unarchive it, by sending
a message to control AT debbugs.gnu.org, with unarchive 13332 in the body.
You can then email your comments to 13332 AT debbugs.gnu.org in the normal way.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Mikael Pettersson <mikpe <at> it.uu.se>
To: bug-gnu-emacs <at> gnu.org
Cc: Mikael Pettersson <mikpe <at> it.uu.se>
Subject: [PATCH] unbreak Emacs on Linux/M68K
Date: Tue, 1 Jan 2013 22:51:12 +0100

Attempting to bootstrap Emacs 23.x or 24.x on Linux/M68K fails with
a SEGV as soon as temacs is run.  Inspecting the core dump shows:

Core was generated by `/home/mikpe/emacs-24.2/build-nox/src/temacs --batch --load loadu'.
Program terminated with signal 11, Segmentation fault.
#0  0x800d5444 in mark_object (arg=1174437910) at /home/mikpe/emacs-24.2/src/alloc.c:5720
5720		if (CONS_MARKED_P (ptr))
(gdb) bt
#0  0x800d5444 in mark_object (arg=1174437910) at /home/mikpe/emacs-24.2/src/alloc.c:5720
#1  0x800d55ba in mark_object (arg=-2143958762) at /home/mikpe/emacs-24.2/src/alloc.c:5731
#2  0x800d5c3a in mark_vectorlike (ptr=0x8035b120) at /home/mikpe/emacs-24.2/src/alloc.c:5420
#3  0x800d5d62 in Fgarbage_collect () at /home/mikpe/emacs-24.2/src/alloc.c:5128
#4  0x800eb7be in Ffuncall (nargs=3, args=0xefcdfad4) at /home/mikpe/emacs-24.2/src/eval.c:2938
#5  0x800ebc46 in call2 (fn=0, arg1=0, arg2=-2143972138) at /home/mikpe/emacs-24.2/src/eval.c:2785
#6  0x800ebf82 in Fsignal (error_symbol=0, data=-2143972138) at /home/mikpe/emacs-24.2/src/eval.c:1710
#7  0x800ec412 in xsignal (error_symbol=0, data=-2143972138) at /home/mikpe/emacs-24.2/src/eval.c:1785
#8  0x800ecdc8 in xsignal1 (error_symbol=0, arg=-2143955495) at /home/mikpe/emacs-24.2/src/eval.c:1800
#9  0x800ece32 in verror (m=0x8015f90d "Bad data in guts of obarray", ap=0xefce0b24) at /home/mikpe/emacs-24.2/src/eval.c:1998
#10 0x800ece44 in error (m=0x8015f90d "Bad data in guts of obarray") at /home/mikpe/emacs-24.2/src/eval.c:2010
#11 0x80108eec in oblookup (obarray=<optimized out>, ptr=0x80156383 "char-table-extra-slots", size=22, size_byte=22) at /home/mikpe/emacs-24.2/src/lread.c:3912
#12 0x8010c578 in intern_c_string (str=0x80156383 "char-table-extra-slots") at /home/mikpe/emacs-24.2/src/lread.c:3728
#13 0x8005ce76 in init_category_once () at /home/mikpe/emacs-24.2/src/category.c:461
#14 0x80006558 in main (argc=<optimized out>, argv=0xefce0ef4) at /home/mikpe/emacs-24.2/src/emacs.c:1265
(gdb) q

This turns out to be caused by the well-known issue of data alignment
on Linux/M68K: the ABI only aligns 32-bit (and larger) primitive types
to 16-bit boundaries, which doesn't provide enough known-zero bits in
object pointers for Emacs' tagging scheme, causing major confusion and
breakage at runtime.

This patch introduces two new integer types in src/m/m68k.h that are
exactly like int and unsigned int, except they are explicitly aligned
to 32-bit boundaries; gcc's attribute aligned is used for this purpose.
It then defines EMACS_INT and EMACS_UINT to these types to override the
default choices in src/lisp.h.

With this patch in place bootstrap succeeds and the final emacs
executable works fine.  Tested with Emacs-24.2 on Linux/M68K.

/Mikael Pettersson

[I consider this 6-line patch (not counting the comment) to be trivial so
I hope it will be acceptable even without a formal copyright assignment.]

src/ChangeLog:

2013-01-01  Mikael Pettersson  <mikpe <at> it.uu.se>

	* m/m68k.h [GNU_LINUX] (m68k_aligned_int_t, m68k_aligned_uint_t):
	New typedefs.
	(EMACS_INT, EMACS_UINT, BITS_PER_EMACS_INT, pI): Define.

--- emacs-24.2/src/m/m68k.h.~1~	2012-08-23 07:33:42.000000000 +0200
+++ emacs-24.2/src/m/m68k.h	2013-01-01 14:26:10.000000000 +0100
@@ -28,5 +28,19 @@ along with GNU Emacs.  If not, see <http
 #define DATA_SEG_BITS 0x80000000
 #endif
 
+/* Define the type to use.
+   Emacs cannot use the default type (plain int) since the ABI on
+   Linux/M68K only aligns 32-bit (and larger) primitive types to
+   16-bit boundaries.  Define new explicitly aligned integer types
+   and use them for EMACS_INT and EMACS_UINT.  Compiling Emacs
+   with -malign-int does not work since it changes data layout in
+   external interfaces (system calls, libraries).  */
+typedef int m68k_aligned_int_t __attribute__ ((__aligned__ (4)));
+typedef unsigned int m68k_aligned_uint_t __attribute__ ((__aligned__ (4)));
+#define EMACS_INT m68k_aligned_int_t
+#define EMACS_UINT m68k_aligned_uint_t
+#define BITS_PER_EMACS_INT 32
+#define pI ""
+
 #endif
 

Message #8 received at 13332 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: Mikael Pettersson <mikpe <at> it.uu.se>
Cc: 13332 <at> debbugs.gnu.org
Subject: Re: [PATCH] unbreak Emacs on Linux/M68K
Date: Tue, 01 Jan 2013 19:12:46 -0800

This part of Emacs has changed in the trunk -- there's no longer
an m/m68k.h file.  But more important, can you explain more why the
patch is needed, and how it works?  First, Emacs shouldn't require
any particular alignment for EMACS_INT: the alignment requirements
are on the larger objects that Emacs allocates, not on EMACS_INT
per se.  Second, even if there's a bug in Emacs where it's
incorrectly assuming that EMACS_INT is aligned, I would expect it
to require an alignment of 8, not of 4, since Emacs uses 3 tag
bits, not 2.

Message #11 received at 13332 <at> debbugs.gnu.org (full text, mbox):

From: Andreas Schwab <schwab <at> linux-m68k.org>
To: Mikael Pettersson <mikpe <at> it.uu.se>
Cc: 13332 <at> debbugs.gnu.org
Subject: Re: bug#13332: [PATCH] unbreak Emacs on Linux/M68K
Date: Wed, 02 Jan 2013 11:12:58 +0100

The last time I tested emacs on m68k (about 8 months ago, before 24.2)
it worked fine after I've fixed an alignment bug, see commit d2220b5.

Andreas.

-- 
Andreas Schwab, schwab <at> linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Message #14 received at 13332 <at> debbugs.gnu.org (full text, mbox):

From: Andreas Schwab <schwab <at> linux-m68k.org>
To: Mikael Pettersson <mikpe <at> it.uu.se>
Cc: 13332 <at> debbugs.gnu.org
Subject: Re: bug#13332: [PATCH] unbreak Emacs on Linux/M68K
Date: Wed, 02 Jan 2013 17:10:59 +0100

Both emacs-24 and trunk are working fine for me.

Andreas.

-- 
Andreas Schwab, schwab <at> linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Message #19 received at 13332-done <at> debbugs.gnu.org (full text, mbox):

From: Andreas Schwab <schwab <at> linux-m68k.org>
To: Mikael Pettersson <mikpe <at> it.uu.se>
Cc: 13332-done <at> debbugs.gnu.org
Subject: Re: bug#13332: [PATCH] unbreak Emacs on Linux/M68K
Date: Wed, 02 Jan 2013 17:33:24 +0100

I just noticed that the commit didn't make it into 24.2.  But it should
be easy enough to backport for any interested parties.

Andreas.

-- 
Andreas Schwab, schwab <at> linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Message #22 received at 13332-done <at> debbugs.gnu.org (full text, mbox):

From: Mikael Pettersson <mikpe <at> it.uu.se>
To: Andreas Schwab <schwab <at> linux-m68k.org>
Cc: Mikael Pettersson <mikpe <at> it.uu.se>, 13332-done <at> debbugs.gnu.org
Subject: Re: bug#13332: [PATCH] unbreak Emacs on Linux/M68K
Date: Wed, 2 Jan 2013 23:38:22 +0100

Andreas Schwab writes:
 > I just noticed that the commit didn't make it into 24.2.  But it should
 > be easy enough to backport for any interested parties.

I spotted three alignment-related trunk commits around that time:
f17846357a5b00985131f1766dc705efa995191f
d2220b5dc1122f7ee597d26f5ce874a768be9b08
68be74b0a5c9cdac93f4f1d19760925dd3d264ed 

I can confirm that applying all three to 24.2 unbreaks it on m68k.

Case closed.

/Mikael

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.