[regression] cone ftbfs on armel/armhf using GCC 4.7

[Matthias Klose]

g++-4.7 -c -g -O[123] myserver.ii
myserver.C: In static member function 'static bool myServer::eventloop(myServer::Callback*)':
myserver.C:168:6: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.

Program received signal SIGSEGV, Segmentation fault.
0x003b90ae in free_scc_vn() ()
(gdb) bt
#0 0x003b90ae in free_scc_vn() ()
#1 0x003aee44 in ?? ()
#2 0x003aee44 in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)

[Doko-v (doko-v)]

Created attachment 28426
preprocessed source

seen building the cone package on arm-linux-gnueabi, works with 4.6 branch, fails with 4.7.

g++-4.7 -c -g -O[123] myserver.ii
myserver.C: In static member function 'static bool myServer::eventloop(myServer::Callback*)':
myserver.C:168:6: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.

Program received signal SIGSEGV, Segmentation fault.
0x003b90ae in free_scc_vn() ()
(gdb) bt
#0 0x003b90ae in free_scc_vn() ()
#1 0x003aee44 in ?? ()
#2 0x003aee44 in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)

[Rguenth (rguenth)]

Please also check trunk.

[Doko-v (doko-v)]

trunk 20120820 and 20121008 do work.

[Mikpe (mikpe)]

I can reproduce the ICE with gcc-4.7.2 on armv5tel-linux-gnueabi, sparc64-linux -m32, x86_64-linux -m32, and m68k-linux. I haven't checked any other targets.

The ICE stopped on trunk with Richard B's "Make gsi_remove return whether EH cleanup is required" patch in r186159:
http://gcc.gnu.org/ml/gcc-patches/2012-04/msg00202.html
http://gcc.gnu.org/ml/gcc-cvs/2012-04/msg00108.html

Backporting r186159 and the follow-up fix in r186164 to gcc-4.7.2 eliminates the ICE there for the four targets mentioned above.

The description of Richard's patch makes it sound more like a cleanup than a bug fix, so I'm not sure if it actually fixes the bug or just masks it.

[Zhenqiang Chen (zhenqiang-chen)]

Thank you for the bug report. I've confirmed this with gcc-linaro-4.7-2012.10 on ARM:

g++ -c [-g] -O[123] myserver.ii
myserver.C: In static member function ‘static bool myServer::eventloop(myServer::Callback*)’:
myserver.C:168:6: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <https://bugs.launchpad.net/gcc-linaro> for instructions.

[Zhenqiang Chen (zhenqiang-chen)]

Created attachment 28512
gdb backtrace

Root cause:

PRE bases on the result of value numbering (run_scc_vn). At the end, it free_scc_vn. But before free_scc_vn, it might call cleanup_tree_cfg ();

  if (do_eh_cleanup || do_ab_cleanup)
    cleanup_tree_cfg ();

cleanup_tree_cfg might call make_ssa_name which might reuse some "name" from the FREE_SSANAMES list. If the VN_INFO of the "name" is NULL, free_scc_vn will "Segmentation fault".

Richard's patch makes "do_eh_cleanup" to false. So cleanup_tree_cfg is not called in PRE. Then no new ssa name will be created.

bt.log shows the gdb backtrace to create a new ssa name.

[Rguenth (rguenth)]

Mine.

[Rguenth (rguenth)]

Created attachment 28517
patch

Patch I am testing.

[Rguenth (rguenth)]

Author: rguenth
Date: Thu Oct 25 11:05:43 2012
New Revision: 192803

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192803
Log:
2012-10-25 Richard Biener <email address hidden>

 PR tree-optimization/54902
 * tree-ssa-pre.c (fini_pre): Return TODO.
 (execute_pre): Adjust.
 * tree-ssa-tailmerge.c (tail_merge_optimize): Delete unreachable
 blocks before computing dominators.

 * g++.dg/torture/pr54902.C: New testcase.

Added:
    branches/gcc-4_7-branch/gcc/testsuite/g++.dg/torture/pr54902.C
Modified:
    branches/gcc-4_7-branch/gcc/ChangeLog
    branches/gcc-4_7-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_7-branch/gcc/tree-ssa-pre.c
    branches/gcc-4_7-branch/gcc/tree-ssa-tail-merge.c

[Rguenth (rguenth)]

Author: rguenth
Date: Thu Oct 25 11:16:12 2012
New Revision: 192804

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192804
Log:
2012-10-25 Richard Biener <email address hidden>

 PR tree-optimization/54902
 * tree-ssa-pre.c (fini_eliminate): Return TODO.
 (do_pre): Adjust.
 (execute_fre): Likewise.
 * tree-ssa-tailmerge.c (tail_merge_optimize): Delete unreachable
 blocks before computing dominators.

 * g++.dg/torture/pr54902.C: New testcase.

Added:
    trunk/gcc/testsuite/g++.dg/torture/pr54902.C
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-pre.c
    trunk/gcc/tree-ssa-tail-merge.c

[Rguenth (rguenth)]

Fixed.

[Launchpad Janitor (janitor)]

This bug was fixed in the package gcc-4.7 - 4.7.2-5ubuntu2

---------------
gcc-4.7 (4.7.2-5ubuntu2) raring; urgency=low

  * Merge with Debian.

gcc-4.7 (4.7.2-5) UNRELEASED; urgency=low

  * Update to SVN 20121025 (r192803) from the gcc-4_7-branch.
    - Fix PR target/54892 (ARM, LP: #1065122), PR rtl-optimization/54870,
      PR rtl-optimization/53701, PR target/53975 (ia64),
      PR tree-optimization/54902 (LP: #1065559), PR middle-end/54945.

  [ Matthias Klose ]
  * Update the Linaro support to the 4.7-2012.10 release.
  * Define MULTIARCH_DIRNAME for arm64 (Wookey).
  * Let the lib*objc-dev packages depend on the lib*gcc-dev packages.
  * Let the libstdc++-dev package depend on the libgcc-dev package.
  * Drop the dependency of the libstdc++-dev package on g++, make
    libstdc++-dev and libstdc++-pic Multi-Arch: same. Closes: #678623.
  * Install override files before calling dh_fixperms.
  * Backport the libffi arm64 port.
  * Build libx32gcc-dev, libx32objc-dev and libx32gfortran-dev packages.
  * Allow conditional building of the x32 multilibs.
  * Fix libmudflap build failure for x32 multilibs.
  * Fix dependency on glibc for triarch builds.
  * Add build-{arch,indep} targets.
  * Fix libquadmath x32 multilib builds on kernels which don't support x32.
  * Fix location of x32 specific C++ header files.

  [ Thibaut Girka ]
  * Split out lib*gcc-dev packages.
  * Split out lib*objc-dev packages.
  * Split out lib*gfortran-dev packages.

  [ Daniel Schepler ]
  * Add support for x32. Closes: #667005.
  * New patch hjl-x32-gcc-4_7-branch.diff to incorporate changes from
    that branch, including --with-abi=mx32 option.

  [ Marcin Juszkiewicz ]
  * lib*-dev packages for cross builds are not Multi-Arch: same. LP: #1070694.
 -- Matthias Klose <email address hidden> Thu, 25 Oct 2012 13:31:07 +0200

[Michael Hope (michaelh1)]

In the release branch. Will be picked up as part of the merge.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Rolf Leggewie (r0lf)]

quantal has seen the end of its life and is no longer receiving any updates. Marking the quantal task for this ticket as "Won't Fix".