diff mbox

Enable debug_frame code by default on x86*

Message ID 1379324871-20826-1-git-send-email-jean.pihet@linaro.org
State Rejected
Headers show

Commit Message

Jean Pihet Sept. 16, 2013, 9:47 a.m. UTC
The dwarf unwinding feature is being used on ARM and supported on
x86, cf. [1], [2]. Enable it by default on those platforms.

[1] http://lwn.net/Articles/499116/
[2] http://www.spinics.net/lists/kernel/msg1603264.html

Signed-off-by: Jean Pihet <jean.pihet@linaro.org>
---
 configure.ac | 1 +
 1 file changed, 1 insertion(+)

Comments

Chris January Sept. 16, 2013, 9:59 a.m. UTC | #1
On Mon, 2013-09-16 at 11:47 +0200, Jean Pihet wrote:
> The dwarf unwinding feature is being used on ARM and supported on
> x86, cf. [1], [2]. Enable it by default on those platforms.

The .debug_frame parsing code uses code that is less async-signal-safe
than the rest of libunwind - which makes it troublesome to use from a
signal handler. I would caution against enabling it by default for that
reason.

Regards,
Chris January - VP Engineering - Allinea Software Ltd.
Jean Pihet Sept. 16, 2013, 10:01 a.m. UTC | #2
Hi!

On 16 September 2013 11:59, Chris January <chris.january@allinea.com> wrote:
> On Mon, 2013-09-16 at 11:47 +0200, Jean Pihet wrote:
>> The dwarf unwinding feature is being used on ARM and supported on
>> x86, cf. [1], [2]. Enable it by default on those platforms.
>
> The .debug_frame parsing code uses code that is less async-signal-safe
> than the rest of libunwind - which makes it troublesome to use from a
> signal handler. I would caution against enabling it by default for that
> reason.
Ok my main concern is about not breaking anything.
In this case isn't it better to make that code more robust? Any
pointer is appreciated.

Thanks,
Jean

>
> Regards,
> Chris January - VP Engineering - Allinea Software Ltd.
>
>
Arun Sharma Sept. 16, 2013, 10:29 a.m. UTC | #3
On Mon, Sep 16, 2013 at 3:31 PM, Jean Pihet <jean.pihet@linaro.org> wrote:

> Ok my main concern is about not breaking anything.
> In this case isn't it better to make that code more robust? Any
> pointer is appreciated.
>

Avoid calls to malloc and anything in stdio that might indirectly call malloc.

We should try to use fixed memory pools (mi/mempool.c) and use raw
syscalls where possible.

 -Arun
diff mbox

Patch

diff --git a/configure.ac b/configure.ac
index 4e2fd23..4bead56 100644
--- a/configure.ac
+++ b/configure.ac
@@ -221,6 +221,7 @@  AC_ARG_ENABLE(debug_frame,
 AS_HELP_STRING([--enable-debug-frame],[Load the ".debug_frame" section if available]),, [
 case "${target_arch}" in
   (arm) enable_debug_frame=yes;;
+  (x86*) enable_debug_frame=yes;;
   (*)   enable_debug_frame=no;;
 esac])
 if test x$enable_debug_frame = xyes; then