LLVM Bugzilla is read-only and represents the historical archive of all LLVM issues filled before November 26, 2021. Use github to submit LLVM bugs

Bug 18171 - clang segmentation fault on i386 architecture in -O2
Summary: clang segmentation fault on i386 architecture in -O2
Status: RESOLVED FIXED
Alias: None
Product: clang
Classification: Unclassified
Component: -New Bugs (show other bugs)
Version: 3.3
Hardware: PC FreeBSD
: P normal
Assignee: Unassigned Clang Bugs
URL:
Keywords:
: 18537 (view as bug list)
Depends on:
Blocks:
 
Reported: 2013-12-07 07:55 PST by Ashish SHUKLA
Modified: 2014-03-17 16:28 PDT (History)
6 users (show)

See Also:
Fixed By Commit(s):


Attachments
preprocessed editfns.c (446.05 KB, application/octet-stream)
2013-12-07 07:55 PST, Ashish SHUKLA
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Ashish SHUKLA 2013-12-07 07:55:03 PST
Created attachment 11686 [details]
preprocessed editfns.c

Emacs 24.3 when compiling on FreeBSD ends up with following segmentation fault:

% gdb temacs
GNU gdb 6.1.1 [FreeBSD]
Copyright 2004 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB.  Type "show warranty" for details.
This GDB was configured as "i386-marcel-freebsd"...
Environment variable "DISPLAY" not defined.
TERM = screen-bce
Breakpoint 1 at 0x80cd031: file emacs.c, line 350.
Breakpoint 2 at 0x80e6d08: file sysdep.c, line 850.
(gdb) core-file temacs.core
Core was generated by `temacs'.
Program terminated with signal 11, Segmentation fault.
Reading symbols from /usr/lib/librt.so.1...done.
Loaded symbols for /usr/lib/librt.so.1
Reading symbols from /usr/lib/libexecinfo.so.1...done.
Loaded symbols for /usr/lib/libexecinfo.so.1
Reading symbols from /lib/libutil.so.9...done.
Loaded symbols for /lib/libutil.so.9
Reading symbols from /lib/libncurses.so.8...done.
Loaded symbols for /lib/libncurses.so.8
Reading symbols from /lib/libthr.so.3...done.
Loaded symbols for /lib/libthr.so.3
Reading symbols from /lib/libm.so.5...done.
Loaded symbols for /lib/libm.so.5
Reading symbols from /lib/libz.so.6...done.
Loaded symbols for /lib/libz.so.6
Reading symbols from /lib/libc.so.7...done.
Loaded symbols for /lib/libc.so.7
Reading symbols from /usr/lib/libelf.so.1...done.
Loaded symbols for /usr/lib/libelf.so.1
Reading symbols from /lib/libgcc_s.so.1...done.
Loaded symbols for /lib/libgcc_s.so.1
Reading symbols from /libexec/ld-elf.so.1...done.
Loaded symbols for /libexec/ld-elf.so.1
#0  format_time_string (format=0x0, formatlen=1, t={tv_sec = 0, tv_nsec = 0}) at editfns.c:1787
1787          buf[0] = '\1';
[New Thread 83ab000 (LWP 100088/temacs)]
Current language:  auto; currently minimal
(gdb) bt
#0  format_time_string (format=0x0, formatlen=1, t={tv_sec = 0, tv_nsec = 0}) at editfns.c:1787
#1  0x08133df8 in Fformat_time_string (format_string=<value optimized out>, timeval=-1619017216) at editfns.c:1756
#2  0x0813cfe6 in eval_sub (form=<value optimized out>) at eval.c:2181
#3  0x0813ce05 in eval_sub (form=<value optimized out>) at eval.c:2139
#4  0x0813ded1 in Flet (args=<value optimized out>) at eval.c:936
#5  0x0813cd77 in eval_sub (form=<value optimized out>) at eval.c:2123
#6  0x08140aeb in funcall_lambda (fun=<value optimized out>, nargs=<value optimized out>) at eval.c:458
#7  0x0813f985 in apply_lambda (fun=138488846, args=<value optimized out>) at eval.c:2914
#8  0x0813cc54 in eval_sub (form=<value optimized out>) at eval.c:2220
#9  0x0813f94e in apply_lambda (fun=138781494, args=<value optimized out>) at eval.c:2905
#10 0x0813cc54 in eval_sub (form=<value optimized out>) at eval.c:2220
#11 0x0813ce05 in eval_sub (form=<value optimized out>) at eval.c:2139
#12 0x0813cea5 in eval_sub (form=<value optimized out>) at eval.c:2160
#13 0x0815b708 in readevalloop (readcharfun=138171994, stream=0x283eb34c, sourcename=<value optimized out>, readfun=138064922, start=<value optimized out>, end=<value optimized out>) at lread.c:1887
#14 0x0815a77d in Fload (file=<value optimized out>, noerror=<value optimized out>, nosuffix=<value optimized out>, must_suffix=<value optimized out>) at lread.c:1358
#15 0x0813d045 in eval_sub (form=<value optimized out>) at eval.c:2189
#16 0x0815b708 in readevalloop (readcharfun=138171994, stream=0x283eb264, sourcename=<value optimized out>, readfun=138064922, start=<value optimized out>, end=<value optimized out>) at lread.c:1887
#17 0x0815a77d in Fload (file=<value optimized out>, noerror=<value optimized out>, nosuffix=<value optimized out>, must_suffix=<value optimized out>) at lread.c:1358
#18 0x0813d045 in eval_sub (form=<value optimized out>) at eval.c:2189
#19 0x0813f716 in Feval (form=138093014, lexical=<value optimized out>) at eval.c:1993
#20 0x080df10a in top_level_2 () at keyboard.c:1179
#21 0x0813eb40 in internal_condition_case (bfun=<value optimized out>, handlers=<value optimized out>, hfun=<value optimized out>) at eval.c:1344

Following is the disassembly of the surrounding area:

0x08133ebe <format_time_string+158>:    je     0x8133ed0 <format_time_string+176>
0x08133ec0 <format_time_string+160>:    sub    $0x4,%esp
0x08133ec3 <format_time_string+163>:    lea    0x20(%esi),%eax
0x08133ec6 <format_time_string+166>:    mov    %eax,(%esp)
0x08133ec9 <format_time_string+169>:    call   0x804bdcc <gmtime@plt>
0x08133ece <format_time_string+174>:    jmp    0x8133ede <format_time_string+190>
0x08133ed0 <format_time_string+176>:    sub    $0x4,%esp
0x08133ed3 <format_time_string+179>:    lea    0x20(%esi),%eax
0x08133ed6 <format_time_string+182>:    mov    %eax,(%esp)
0x08133ed9 <format_time_string+185>:    call   0x804b0dc <localtime@plt>
0x08133ede <format_time_string+190>:    add    $0x4,%esp
0x08133ee1 <format_time_string+193>:    mov    %eax,%edx
0x08133ee3 <format_time_string+195>:    test   %edx,%edx
0x08133ee5 <format_time_string+197>:    je     0x8134010 <format_time_string+496>
0x08133eeb <format_time_string+203>:    mov    $0xb,%ecx
0x08133ef0 <format_time_string+208>:    mov    0x14(%ebp),%edi
0x08133ef3 <format_time_string+211>:    mov    %edx,%esi
0x08133ef5 <format_time_string+213>:    rep movsl %ds:(%esi),%es:(%edi)
0x08133ef7 <format_time_string+215>:    mov    0x1c(%esi),%ecx
0x08133efa <format_time_string+218>:    movb   $0x1,(%ecx)
0x08133efd <format_time_string+221>:    sub    $0x14,%esp
0x08133f00 <format_time_string+224>:    mov    0xc(%ebp),%eax

Following is the unoptimized code for the block in question:

.LBB64_6:                               #   in Loop: Header=BB64_1 Depth=1
        movl    $44, %eax
        .loc    1 1785 0                # editfns.c:1785:0
        movl    4104(%esi), %ecx
        movl    80(%esi), %edx
        subl    $12, %esp
        movl    %ecx, (%esp)
        movl    %edx, 4(%esp)
        movl    $44, 8(%esp)
        movl    %eax, 48(%esi)          # 4-byte Spill
        calll   memcpy
        addl    $12, %esp
        .loc    1 1787 0                # editfns.c:1787:0
        movl    100(%esi), %eax
        movb    $1, (%eax)             


Following is the -O2 code for the block in question:

        .loc    1 1785 0                # editfns.c:1785:0
        movl    %edx, %esi              ########### %esi gets clobbered here, and never restored after 'rep movsl'
        rep;movsl
        movl    28(%esi), %ecx          # 4-byte Reload
        .loc    1 1787 0                # editfns.c:1787:0
        movb    $1, (%ecx)              ########### Segmentation Fault happens here

Following is the version of "clang":

FreeBSD clang version 3.3 (tags/RELEASE_33/final 183502) 20130610
Target: i386-unknown-freebsd10.0
Thread model: posix

Let me know if you need more information to debug this.

Thanks
Comment 1 Benjamin Kramer 2013-12-07 08:54:30 PST
This is the unholy conjunction of dynamic alloca and stack realignment again.

- When we reach the memcpy in @format_time_string the check if %esi is used as a base pointer returns false. We emit a rep;movsl
- Then we hit the dynamic alloca in a later BB in the function. We properly set all the flags needed to prevent further allocation of %esi, but it's already too late for the emitted rep;movsl.
Comment 2 Dimitry Andric 2014-01-19 08:25:55 PST
*** Bug 18537 has been marked as a duplicate of this bug. ***
Comment 3 Dimitry Andric 2014-03-17 16:28:11 PDT
I can confirm this is now fixed by r202930 (see bug 19012).  I will commit the (slightly modified) revision to FreeBSD 11.0-CURRENT soon.