From: Paul Eggert <eggert@cs.ucla.edu>
To: Ken Brown <kbrown@cornell.edu>
Cc: Wolfgang Jenkner <wjenkner@inode.at>,
Corinna Vinschen <corinna-cygwin@cygwin.com>,
Dmitry Antipov <dmantipov@yandex.ru>,
21004@debbugs.gnu.org
Subject: bug#21004: 25.0.50; Questionable call to getrlimit in handle_sigsegv
Date: Thu, 16 Jul 2015 00:58:10 -0700 [thread overview]
Message-ID: <55A76412.80000@cs.ucla.edu> (raw)
In-Reply-To: <559C3289.9090906@cornell.edu>
[-- Attachment #1: Type: text/plain, Size: 300 bytes --]
Ken, thanks for reporting the problem in <http://bugs.gnu.org/21004>. Please
try the attached patch. Although it avoids getrlimit and it works for me on
both GNU/Linux and Solaris, I can't easily test it on Cygwin. I'll CC: this to
Eli as a heads-up, as he's interested in the MS-Windows case.
[-- Attachment #2: 0001-Better-heuristic-for-C-stack-overflow.patch --]
[-- Type: text/x-diff, Size: 6196 bytes --]
From 4cb3a1e2e3563a1dc0969bce3edd84918067d199 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Thu, 16 Jul 2015 00:48:40 -0700
Subject: [PATCH] Better heuristic for C stack overflow
Improve the heuristic for distinguishing stack overflows from
other SIGSEGV causes (Bug#21004). Corinna Vinschen explained that
the getrlimit method wasn't portable to Cygwin; see:
https://www.cygwin.com/ml/cygwin/2015-07/msg00092.html
Corinna suggested pthread_getattr_np but this also has problems.
Instead, replace the low-level system stuff with a simple
heuristic based on known good stack addresses.
* src/eval.c, src/lisp.h (near_C_stack_top): New function.
* src/sysdep.c: Don't include <sys/resource.h>.
(stack_direction): Remove. All uses removed.
(stack_overflow): New function.
(handle_sigsegv): Use it instead of incorrect getrlimit heuristic.
Make SEGV fatal in non-main threads.
---
src/eval.c | 6 ++++
src/lisp.h | 1 +
src/sysdep.c | 93 ++++++++++++++++++++++++++++++++++++------------------------
3 files changed, 63 insertions(+), 37 deletions(-)
diff --git a/src/eval.c b/src/eval.c
index 4f7f42f..9bdcf4b 100644
--- a/src/eval.c
+++ b/src/eval.c
@@ -200,6 +200,12 @@ backtrace_next (union specbinding *pdl)
return pdl;
}
+/* Return a pointer to somewhere near the top of the C stack. */
+void *
+near_C_stack_top (void)
+{
+ return backtrace_args (backtrace_top ());
+}
void
init_eval_once (void)
diff --git a/src/lisp.h b/src/lisp.h
index c3289c9..341603f 100644
--- a/src/lisp.h
+++ b/src/lisp.h
@@ -4029,6 +4029,7 @@ extern _Noreturn void verror (const char *, va_list)
ATTRIBUTE_FORMAT_PRINTF (1, 0);
extern void un_autoload (Lisp_Object);
extern Lisp_Object call_debugger (Lisp_Object arg);
+extern void *near_C_stack_top (void);
extern void init_eval_once (void);
extern Lisp_Object safe_call (ptrdiff_t, Lisp_Object, ...);
extern Lisp_Object safe_call1 (Lisp_Object, Lisp_Object);
diff --git a/src/sysdep.c b/src/sysdep.c
index 91036f0..30a55f1 100644
--- a/src/sysdep.c
+++ b/src/sysdep.c
@@ -79,9 +79,6 @@ along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */
#include "msdos.h"
#endif
-#ifdef HAVE_SYS_RESOURCE_H
-#include <sys/resource.h>
-#endif
#include <sys/param.h>
#include <sys/file.h>
#include <fcntl.h>
@@ -1625,14 +1622,58 @@ handle_arith_signal (int sig)
#ifdef HAVE_STACK_OVERFLOW_HANDLING
-/* -1 if stack grows down as expected on most OS/ABI variants, 1 otherwise. */
-
-static int stack_direction;
-
/* Alternate stack used by SIGSEGV handler below. */
static unsigned char sigsegv_stack[SIGSTKSZ];
+
+/* Return true if SIGINFO indicates a stack overflow. */
+
+static bool
+stack_overflow (siginfo_t *siginfo)
+{
+ /* In theory, a more-accurate heuristic can be obtained by using
+ GNU/Linux pthread_getattr_np along with POSIX pthread_attr_getstack
+ and pthread_attr_getguardsize to find the location and size of the
+ guard area. In practice, though, these functions are so hard to
+ use reliably that they're not worth bothering with. E.g., see:
+ https://sourceware.org/bugzilla/show_bug.cgi?id=16291
+ Other operating systems also have problems, e.g., Solaris's
+ stack_violation function is tailor-made for this problem, but it
+ doesn't work on Solaris 11.2 x86-64 with a 32-bit executable.
+
+ GNU libsigsegv is overkill for Emacs; otherwise it might be a
+ candidate here. */
+
+ if (!siginfo)
+ return false;
+
+ /* The faulting address. */
+ char *addr = siginfo->si_addr;
+ if (!addr)
+ return false;
+
+ /* The known top and bottom of the stack. The actual stack may
+ extend a bit beyond these boundaries. */
+ char *bot = stack_bottom;
+ char *top = near_C_stack_top ();
+
+ /* Log base 2 of the stack heuristic ratio. This ratio is the size
+ of the known stack divided by the size of the guard area past the
+ end of the stack top. The heuristic is that a bad address is
+ considered to be a stack overflow if it occurs within
+ stacksize>>LG_STACK_HEURISTIC bytes above the top of the known
+ stack. This heuristic is not exactly correct but it's good
+ enough in practice. */
+ enum { LG_STACK_HEURISTIC = 8 };
+
+ if (bot < top)
+ return 0 <= addr - top && addr - top < (top - bot) >> LG_STACK_HEURISTIC;
+ else
+ return 0 <= top - addr && top - addr < (bot - top) >> LG_STACK_HEURISTIC;
+}
+
+
/* Attempt to recover from SIGSEGV caused by C stack overflow. */
static void
@@ -1640,35 +1681,15 @@ handle_sigsegv (int sig, siginfo_t *siginfo, void *arg)
{
/* Hard GC error may lead to stack overflow caused by
too nested calls to mark_object. No way to survive. */
- if (!gc_in_progress)
- {
- struct rlimit rlim;
+ bool fatal = gc_in_progress;
- if (!getrlimit (RLIMIT_STACK, &rlim))
- {
- /* STACK_DANGER_ZONE has to be bigger than 16K on Cygwin, for
- reasons explained in
- https://www.cygwin.com/ml/cygwin/2015-06/msg00381.html. */
-#ifdef CYGWIN
- enum { STACK_DANGER_ZONE = 32 * 1024 };
-#else
- enum { STACK_DANGER_ZONE = 16 * 1024 };
-#endif
- char *beg, *end, *addr;
-
- beg = stack_bottom;
- end = stack_bottom + stack_direction * rlim.rlim_cur;
- if (beg > end)
- addr = beg, beg = end, end = addr;
- addr = (char *) siginfo->si_addr;
- /* If we're somewhere on stack and too close to
- one of its boundaries, most likely this is it. */
- if (beg < addr && addr < end
- && (addr - beg < STACK_DANGER_ZONE
- || end - addr < STACK_DANGER_ZONE))
- siglongjmp (return_to_command_loop, 1);
- }
- }
+#ifdef FORWARD_SIGNAL_TO_MAIN_THREAD
+ if (!fatal && !pthread_equal (pthread_self (), main_thread))
+ fatal = true;
+#endif
+
+ if (!fatal && stack_overflow (siginfo))
+ siglongjmp (return_to_command_loop, 1);
/* Otherwise we can't do anything with this. */
deliver_fatal_thread_signal (sig);
@@ -1683,8 +1704,6 @@ init_sigsegv (void)
struct sigaction sa;
stack_t ss;
- stack_direction = ((char *) &ss < stack_bottom) ? -1 : 1;
-
ss.ss_sp = sigsegv_stack;
ss.ss_size = sizeof (sigsegv_stack);
ss.ss_flags = 0;
--
2.1.0
next prev parent reply other threads:[~2015-07-16 7:58 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-07-07 20:11 bug#21004: 25.0.50; Questionable call to getrlimit in handle_sigsegv Ken Brown
2015-07-08 13:33 ` Wolfgang Jenkner
2015-07-08 14:20 ` Ken Brown
2015-07-16 7:58 ` Paul Eggert [this message]
2015-07-16 12:15 ` Ken Brown
2015-07-16 14:34 ` Eli Zaretskii
2015-07-16 14:38 ` Paul Eggert
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=55A76412.80000@cs.ucla.edu \
--to=eggert@cs.ucla.edu \
--cc=21004@debbugs.gnu.org \
--cc=corinna-cygwin@cygwin.com \
--cc=dmantipov@yandex.ru \
--cc=kbrown@cornell.edu \
--cc=wjenkner@inode.at \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this external index
https://git.savannah.gnu.org/cgit/emacs.git
https://git.savannah.gnu.org/cgit/emacs/org-mode.git
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.