[Bug libc/15845] New: glibc-2.18 (32-bit) segfault with i7-3770K CPU

[Bug libc/15845] New: glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

            Bug ID: 15845
           Summary: glibc-2.18 (32-bit) segfault with i7-3770K CPU
           Product: glibc
           Version: 2.18
            Status: NEW
          Severity: normal
          Priority: P2
         Component: libc
          Assignee: unassigned at sourceware dot org
          Reporter: bob.eric.nelson at gmail dot com
                CC: drepper.fsp at gmail dot com

Building a 32-bit glibc-2.18 with the triplet (or quartet) of i686-pc-linux-gnu
produces a segfault on Adobe's acrobat (9.5.5) as well as other 32-bit
applications including a numerically-intensive antenna modeling app (4nec)
running under wine. 

The problem occurs with these Linux distros tested where I have built the
32-bit version of glibc-2.18: Linux from Scratch, Gentoo-2.2, Slackware-14.0
and CentOS-6.4.

1). The problem does NOT occur with a rollback to glibc-2.17. 

2). The problem does NOT occur with the x86_64 (64-bit) build.

3). The problem does NOT occur with glibc-2.18 under VirtualBox which emulates
a core2.

4). The problem does not occur on a modest Supermicro C2SBC-Q mainboard with an
Intel(R) Pentium(R) Dual E2220 @2.40GHz CPU.

5). The problem DOES occur (and is easily reproduced at will) with on a
corei7-avx. Here are the full details of that system:

ASUSTeK COMPUTER INC. P8Z77-V LX (Quad-Core Hyper-Threaded Intel(R) Core(TM)
i7-3770K @3.50GHz)
Memory Error Correction: None
Maximum Memory: 32768MB (32GB)
ChannelA-DIMM0 BANK 0: 8192MB 1333MHz Synchronous DDR3 DIMM, AMI
CMV8GX3M1A1333C9
ChannelA-DIMM1 BANK 1: 8192MB 1333MHz Synchronous DDR3 DIMM, AMI
CMX8GX3M1A1333C9
ChannelB-DIMM0 BANK 2: 8192MB 1333MHz Synchronous DDR3 DIMM, AMI
CMV8GX3M1A1333C9
ChannelB-DIMM1 BANK 3: 8192MB 1333MHz Synchronous DDR3 DIMM, AMI
CMX8GX3M1A1333C9
empty memory sockets: None
total memory = 32768MB (32GB)

6). Here's the rest of pertinent information:

a). gcc-4.8.1 (also fails with gcc-4.7.2)
b). binutils-2.23.2 (also fails with 2.22)
c). Linux kernel 3.10.7 (also fails with 3.8.8)
d). CFLAGS="-O2 -march=native -pipe" (also fails with CFLAGS undefined and with
"-O1 -march=i386")

7). My workaround is to change the triplet from i686-pc-linux to i586-pc-linux
or i486-pc-linux. My guess is one or more of the system-dependent assembly
files in the "i386/i686" hierarchy doesn't play well with the i7-3770K
instruction set. I have not yet toyed with trying to isolated which of the
"i386/i686" sourcefiles triggers the error, opting for the expedient approach
of using the i586-* triplet.

Since I can readily replicate the error, I'll be glad to provide additional
information or perform other tests. Also see the thread here when I initially
thought the problem may have been isolated to Adobe's acroread:

http://www.linuxquestions.org/questions/linux-from-scratch-13/glibc-2-18-and-acroread-4175473288/

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

re_nelson <bob.eric.nelson at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |glibc_2.18
             Target|                            |i686-pc-linux-*
                URL|                            |http://www.linuxquestions.o
                   |                            |rg/questions/linux-from-scr
                   |                            |atch-13/glibc-2-18-and-acro
                   |                            |read-4175473288/

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

re_nelson <bob.eric.nelson at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Build|                            |i686-pc-linux-*, x86_64

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

re_nelson <bob.eric.nelson at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bob.eric.nelson at gmail dot com

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

re_nelson <bob.eric.nelson at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bnelson at nelsonbe dot com

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

--- Comment #1 from re_nelson <bob.eric.nelson at gmail dot com> ---
Created attachment 7148
  --> http://sourceware.org/bugzilla/attachment.cgi?id=7148&action=edit
strace of segfault (32-bit glibc-2.18)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

--- Comment #2 from re_nelson <bob.eric.nelson at gmail dot com> ---
Further exploration of the 32-bit glibc-2.18 segfault, this time using Arch
Linux as the platform on the hardware cited in the initial post, revealed the
following with "--host=i686-pc-linux-gnu" passed to the configure script.

1). Using "-O3" optimization is a fix for the segfault described in the post.
In the attached tarball, see "strace_glibc-2.18-i686.ok" when the 32-bit
acroread application is invoked with the library built with "-O3" optimzation.
The program runs to completion with no errors.

2). With CFLAGS unset, which during the build default to "-O2", the segfault
occurs as shown in "trace_glibc-2.18-i686.fail", also in the attached tarball.
The trace also shows the kernel's general protection error at the end, taken
from dmesg.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[allan at archlinux dot org]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

Allan McRae <allan at archlinux dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |allan at archlinux dot org

--- Comment #3 from Allan McRae <allan at archlinux dot org> ---
Can you run a failing program under gdb?   This sounds like the __strstr_sse42
inlining issue I have seen.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[bob.eric.nelson at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15845

--- Comment #4 from re_nelson <bob.eric.nelson at gmail dot com> ---
Created attachment 7150
  --> http://sourceware.org/bugzilla/attachment.cgi?id=7150&action=edit
inline keyword restored to __m128i_strloadu

Allan: 

I can now confirm with 100% certainty that the __strstr_sse42 issue involving
the inline keyword having been removed upstream is the problem. Your comments
in this thread are spot on:

https://bugs.archlinux.org/task/36556?project=1

Before I even saw that thread, trial and error led me to using an "-O3"
optimization which had the result of inlining the  __m128i_strloadu. After
applying the "glibc-2.18-strstr-hackfix.patch" (attached), which restores the
"inline" keyword to the function definition, I can again use the "-O2" flag and
get a working 32-bit library.

12:26:15 on Sun Aug 18

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[jsm28 at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=15845

Joseph Myers <jsm28 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #5 from Joseph Myers <jsm28 at gcc dot gnu.org> ---
Reported fixed by __strstr_sse42 removal.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15845] glibc-2.18 (32-bit) segfault with i7-3770K CPU

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=15845

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fweimer at redhat dot com
              Flags|                            |security-

-- 
You are receiving this mail because:
You are on the CC list for the bug.