[lttng-dev] Bogus openembedded gcc patch

[Mathieu Desnoyers]

----- On Jan 26, 2016, at 11:28 AM, Khem Raj raj.khem at gmail.com wrote:

>> On Jan 25, 2016, at 8:02 PM, Mathieu Desnoyers <mathieu.desnoyers at efficios.com>
>> wrote:
>> 
>> Hi,
>> 
>> We are noticing the presence of the following patch in various
>> openembedded gcc versions:
>> 
>> 0024-PR-target-32219.patch
>> 
>> "From e0d15f4f8bf28c351b9215ca37f1caa24df0e1fd Mon Sep 17 00:00:00 2001
>> From: Khem Raj <raj.khem at gmail.com>
>> Date: Fri, 29 Mar 2013 09:18:54 +0400
>> Subject: [PATCH 24/35] PR target/32219
>> 
>> * varasm.c (default_binds_local_p_1): Weak data is not local.
>> 
>> Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop at gmail.com>
>> Signed-off-by: Khem Raj <raj.khem at gmail.com>
>> 
>> Upstream-Status: Backport"
>> 
>> However, contrarily to its "Backport" status, that patch is
>> not upstream in gcc, and we're starting to wonder if this
>> would be the patch that breaks handling of start/stop automatic
>> hidden symbols we use in lttng-ust.
>> 
>> We are only experiencing problems on the various openembedded
>> compilers, but on no other distro (with same compiler versions),
>> which leads us to suspect a buggy distro-specific gcc patch.
>> 
>> We've been testing with openembedded gcc-4.9.2-r0.
>> 
>> The original gold linker issue that this patch was trying to
>> address seems to have been fixed in the gold linke since
>> then.
>> 
>> Refs:
>> http://lists.lttng.org/pipermail/lttng-dev/2014-May/023112.html
>> https://gcc.gnu.org/ml/gcc-help/2014-05/msg00042.html
>> http://cgit.openembedded.org/openembedded-core/commit/?id=3cb2b003db7371b3a47d02c08352a262e1e419b4
>> https://sourceware.org/bugzilla/show_bug.cgi?id=15435
>> 
>> Thoughts ?
> 
> OE has defaulted to gcc5 since last release, and gold issue is fixed in binutils
> 2.25
> which is what we use in 2.0 as well so it might be OK to unbolt this fix.
> Although I am not sure
> about the original problem reported in PR32219 is fixed for gcc 4.9 if we remove
> this fix.
> Have you narrowed down your issue to this patch ? if not, then it would be good
> to do so

Yes, we have narrowed down our problem to this specific patch by
removing the patch from the OE gcc patches, rebuilding gcc, and
then building our test program with that gcc. The problem goes
away when we do so, which points directly to that patch as a
culprit.

Test program is attached.

Faulty[gcc 4.9-r0 with patch 0024]:

main: myvar addr=0x600c90 __start___test=0x600c90
fct: myvar addr=0x7f8733d1eae8 __start___test=0x600c90
fct2: myvar addr=0x7f8733d1eaec __start___test=0x600c90
fct3: myvar addr=0x600c94 __start___test=0x600c90

Valid[gcc 4.9-r0 without patch 0024]:

main: myvar addr=0x600c60 __start___test=0x600c60
fct: myvar addr=0x7fb62c758a90 __start___test=0x7fb62c758a90
fct2: myvar addr=0x7fb62c758a94 __start___test=0x7fb62c758a90
fct3: myvar addr=0x600c64 __start___test=0x600c60

Thanks,

Mathieu

> 
>> 
>> Thanks,
>> 
>> Mathieu
>> 
>> --
>> Mathieu Desnoyers
>> EfficiOS Inc.
> > http://www.efficios.com

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: yocto-hidden-bug2.tar.gz
Type: application/x-compressed-tar
Size: 681 bytes
Desc: not available
URL: <http://lists.lttng.org/pipermail/lttng-dev/attachments/20160126/19b55666/attachment-0001.bin>