Package: emacs;
Reported by: Paul Eggert <eggert <at> cs.ucla.edu>
Date: Mon, 14 Nov 2011 04:56:02 UTC
Severity: normal
Tags: patch
Done: Paul Eggert <eggert <at> cs.ucla.edu>
Bug is archived. No further changes may be made.
Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):
From: Paul Eggert <eggert <at> cs.ucla.edu> To: bug-gnu-emacs <at> gnu.org Subject: VIRT_ADDR_VARIES fix for Emacs under valgrind etc. Date: Sun, 13 Nov 2011 20:55:03 -0800
This is a followup to the thread that began at
<http://lists.gnu.org/archive/html/emacs-devel/2011-11/msg00081.html>.
This patch fixes a bug where Emacs won't run correctly
under valgrind. As Andreas mentioned, without the patch Emacs
probably isn't all that safe even when run outside of valgrind.
# Bazaar merge directive format 2 (Bazaar 0.90)
# revision_id: eggert <at> cs.ucla.edu-20111114043522-l0gmk07ckh964xtp
# target_branch: bzr+ssh://eggert <at> bzr.savannah.gnu.org/emacs/trunk
# testament_sha1: 6e35062324740218b862c3bc27e0aed206fee1ee
# timestamp: 2011-11-13 20:42:27 -0800
# base_revision_id: rgm <at> gnu.org-20111114021401-gnrqc3mar3ra6cjp
#
# Begin patch
=== modified file 'src/ChangeLog'
--- src/ChangeLog 2011-11-12 12:11:54 +0000
+++ src/ChangeLog 2011-11-14 04:35:22 +0000
@@ -1,3 +1,22 @@
+2011-11-14 Paul Eggert <eggert <at> cs.ucla.edu>
+
+ Standardize on VIRT_ADDR_VARIES behavior; otherwise, valgrind
+ does not work on some platforms. Problem reported by Andreas Schwab in
+ <http://lists.gnu.org/archive/html/emacs-devel/2011-11/msg00081.html>.
+ * puresize.h (pure, PURE_P): Always behave as if VIRT_ADDR_VARIES
+ is set, removing the need for VIRT_ADDRESS_VARIES.
+ (PURE_P): Use a more-efficient implementation that needs just one
+ comparison, not two: on x86-64 with GCC 4.6.2, this cut down the
+ number of instructions from 6 (xorl, cmpq, jge, xorl, cmpq, setge)
+ to 4 (xorl, subq, cmpq, setbe).
+ * alloc.c (pure): Always extern now, since that's the
+ VIRT_ADDR_VARIES behavior.
+ (PURE_POINTER_P): Use a single comparison, not two, for
+ consistency with the new puresize.h.
+ * lisp.h (PNTR_COMPARISON_TYPE): Remove; no longer needed.
+ * m/ibms390.h, m/intel386.h, m/template.h, s/cygwin.h, s/hpux10-20.h:
+ Remove VIRT_ADDR_VARIES no longer needed.
+
2011-11-12 Eli Zaretskii <eliz <at> gnu.org>
* makefile.w32-in (HEAPSIZE): New variable, allows to build temacs
=== modified file 'src/alloc.c'
--- src/alloc.c 2011-11-07 05:37:49 +0000
+++ src/alloc.c 2011-11-10 08:14:27 +0000
@@ -203,9 +203,6 @@
remapping on more recent systems because this is less important
nowadays than in the days of small memories and timesharing. */
-#ifndef VIRT_ADDR_VARIES
-static
-#endif
EMACS_INT pure[(PURESIZE + sizeof (EMACS_INT) - 1) / sizeof (EMACS_INT)] = {1,};
#define PUREBEG (char *) pure
@@ -222,10 +219,7 @@
/* Value is non-zero if P points into pure space. */
#define PURE_POINTER_P(P) \
- (((PNTR_COMPARISON_TYPE) (P) \
- < (PNTR_COMPARISON_TYPE) ((char *) purebeg + pure_size)) \
- && ((PNTR_COMPARISON_TYPE) (P) \
- >= (PNTR_COMPARISON_TYPE) purebeg))
+ ((uintptr_t) (P) - (uintptr_t) purebeg <= pure_size)
/* Index in pure at which next pure Lisp object will be allocated.. */
=== modified file 'src/lisp.h'
--- src/lisp.h 2011-11-07 17:04:01 +0000
+++ src/lisp.h 2011-11-10 08:14:27 +0000
@@ -1877,9 +1877,6 @@
CHECK_NATNUM (tmp); \
XSETCDR ((x), tmp); \
} while (0)
-
-/* Cast pointers to this type to compare them. */
-#define PNTR_COMPARISON_TYPE uintptr_t
/* Define a built-in function for calling from Lisp.
`lname' should be the name to give the function in Lisp,
=== modified file 'src/m/ibms390.h'
--- src/m/ibms390.h 2011-02-16 01:35:20 +0000
+++ src/m/ibms390.h 2011-11-10 08:14:27 +0000
@@ -17,11 +17,6 @@
You should have received a copy of the GNU General Public License
along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */
-
-/* Define VIRT_ADDR_VARIES if the virtual addresses of
- pure and impure space as loaded can vary, and even their
- relative order cannot be relied on.
-
- Otherwise Emacs assumes that text space precedes data space,
- numerically. */
-#define VIRT_ADDR_VARIES
+/* This file is a placeholder -- it does not contain any definitions.
+ At some point we should probably fix this by removing the file
+ and removing all uses of it. */
=== modified file 'src/m/intel386.h'
--- src/m/intel386.h 2011-01-31 23:54:50 +0000
+++ src/m/intel386.h 2011-11-10 08:14:27 +0000
@@ -19,7 +19,6 @@
#ifdef WINDOWSNT
-#define VIRT_ADDR_VARIES
#define DATA_START get_data_start ()
#endif
@@ -28,4 +27,3 @@
/* we cannot get the maximum address for brk */
#define ULIMIT_BREAK_VALUE (32*1024*1024)
#endif
-
=== modified file 'src/m/template.h'
--- src/m/template.h 2011-02-16 01:35:20 +0000
+++ src/m/template.h 2011-11-10 08:14:27 +0000
@@ -21,14 +21,6 @@
does not define it automatically.
Ones defined so far include m68k and many others */
-/* Define VIRT_ADDR_VARIES if the virtual addresses of
- pure and impure space as loaded can vary, and even their
- relative order cannot be relied on.
-
- Otherwise Emacs assumes that text space precedes data space,
- numerically. */
-#define VIRT_ADDR_VARIES
-
/* After adding support for a new machine, modify the large case
statement in configure.in to recognize reasonable
configuration names, and add a description of the system to
=== modified file 'src/puresize.h'
--- src/puresize.h 2011-06-09 19:08:29 +0000
+++ src/puresize.h 2011-11-10 08:14:27 +0000
@@ -75,21 +75,7 @@
/* Define PURE_P. */
-#ifdef VIRT_ADDR_VARIES
-/* For machines where text and data can go anywhere
- in virtual memory. */
-
extern EMACS_INT pure[];
#define PURE_P(obj) \
- ((PNTR_COMPARISON_TYPE) XPNTR (obj) < (PNTR_COMPARISON_TYPE) ((char *) pure + PURESIZE) \
- && (PNTR_COMPARISON_TYPE) XPNTR (obj) >= (PNTR_COMPARISON_TYPE) pure)
-
-#else /* not VIRT_ADDR_VARIES */
-
-extern char my_edata[];
-
-#define PURE_P(obj) \
- ((PNTR_COMPARISON_TYPE) XPNTR (obj) < (PNTR_COMPARISON_TYPE) my_edata)
-
-#endif /* VIRT_ADDRESS_VARIES */
+ ((uintptr_t) XPNTR (obj) - (uintptr_t) pure <= PURESIZE)
=== modified file 'src/s/cygwin.h'
--- src/s/cygwin.h 2011-03-17 05:15:08 +0000
+++ src/s/cygwin.h 2011-11-10 08:14:27 +0000
@@ -91,9 +91,6 @@
why it needed to be changed. */
#define GC_MARK_STACK GC_MAKE_GCPROS_NOOPS
-/* Virtual addresses of pure and impure space can vary, as on Windows. */
-#define VIRT_ADDR_VARIES
-
/* Emacs supplies its own malloc, but glib (part of Gtk+) calls
memalign and on Cygwin, that becomes the Cygwin-supplied memalign.
As malloc is not the Cygwin malloc, the Cygwin memalign always
=== modified file 'src/s/hpux10-20.h'
--- src/s/hpux10-20.h 2011-11-11 23:17:16 +0000
+++ src/s/hpux10-20.h 2011-11-14 04:35:22 +0000
@@ -100,14 +100,6 @@
header sections which lose when `static' is defined away, as it is
on HP-UX. (You get duplicate symbol errors on linking). */
#undef _FILE_OFFSET_BITS
-
-/* Define VIRT_ADDR_VARIES if the virtual addresses of
- pure and impure space as loaded can vary, and even their
- relative order cannot be relied on.
-
- Otherwise Emacs assumes that text space precedes data space,
- numerically. */
-#define VIRT_ADDR_VARIES
/* The data segment on this machine always starts at address 0x40000000. */
#define DATA_SEG_BITS 0x40000000
GNU bug tracking system
Copyright (C) 1999 Darren O. Benham,
1997,2003 nCipher Corporation Ltd,
1994-97 Ian Jackson.