[Bug 6254] New: pthread_exit crashes with SIGSEGV when a binary is compiled statically

[bugzilla at busybox.net]

https://bugs.busybox.net/show_bug.cgi?id=6254

           Summary: pthread_exit crashes with SIGSEGV when a binary is
                    compiled statically
           Product: uClibc
           Version: 0.9.33
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: critical
          Priority: P5
         Component: Threads
        AssignedTo: unassigned at uclibc.org
        ReportedBy: cafedetal+uclibcbt at gmail.com
                CC: uclibc-cvs at uclibc.org
   Estimated Hours: 0.0


Created attachment 4898
  --> https://bugs.busybox.net/attachment.cgi?id=4898
Sample code

When compiling one of my applications with -static and uclibc, every detached
thread crashes on exit as long as I use pthread_exit. 

It's reproducible with a minimal sample that I am attaching, and it only
happens in static builds. There is no crash with the sample when using glibc
with static linking. 

When running the sample with a dynamic compilation ( "i586-linux-gcc -pthread
-o test detached.c" ), this is the result:

# ./test 
Main: creating thread
Thread starting...
Thread done.
Main: program completed. Exiting.
# 


When compiling a static binary using "i586-linux-gcc -static -pthread -o test
detached.c", this is the result:

# ./test 
Main: creating thread
Thread starting...
Thread done.
Segmentation fault
# 


Using gdb to backtrace the crash gives this result:

Program received signal SIGSEGV, Segmentation fault.
[Switching to LWP 12596]
0x00000000 in ?? ()
(gdb) bt
#0  0x00000000 in ?? ()
#1  0x0804baf1 in siglongjmp ()
#2  0x0804ab19 in ?? ()
#3  0x08055cd1 in _Unwind_ForcedUnwind_Phase2 (exc=exc at entry=0xf778fdf0,
context=context at entry=0xf778fa14)
    at
/mnt/External/toolchains/buildroot-2013.02/output/toolchain/gcc-4.7.2/libgcc/unwind.inc:163
#4  0x080560fc in _Unwind_ForcedUnwind (exc=0xf778fdf0, stop=0x804aa40,
stop_argument=0xf778fb34)
    at
/mnt/External/toolchains/buildroot-2013.02/output/toolchain/gcc-4.7.2/libgcc/unwind.inc:207
#5  0x0804aba9 in __pthread_unwind ()
#6  0x0804a1ae in pthread_exit ()
#7  0x080489d2 in doWork ()


The siglongjmp seems to be trying to jump to NULL, so my limited knowledge
makes me think there's some function not being added to the final binary in the
static linking and ending up being NULL in the table.

Dynamic builds work as expected. Removing pthread_exit(NULL) from the thread
function makes the thread end gracefully as far as I can tell. According to the
manpage for pthread_exit, "Performing a return from the start function of any
thread other than the main thread results in an implicit call to
pthread_exit()", so I'm not sure if there isn't another bug hidden there.

This should be easily reproduced on linux using buildroot-2013.02 with the
attached .config file. Also tested selecting gcc 4.6 in buildroot, with the
same result.

-- 
Configure bugmail: https://bugs.busybox.net/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.