Trouble with thread attaching/detaching on Linux

January 20, 2012

[D-runtime] Trouble with thread attaching/detaching on Linux

Posted by Alex

Permalink

Alex

Permalink

Hi,

(Sorry if this is posted in the wrong place.)

Consider the following code:

import core.sys.posix.pthread,
       core.memory,
       core.thread,
       std.stdio;

extern (C) void rt_moduleTlsCtor();
extern (C) void rt_moduleTlsDtor();

__gshared pthread_key_t key;

static this() { writefln("TLS ctor"); }
static ~this() { writefln("TLS dtor"); }

static extern(C) void* threadMain(void *arg)
{
    pthread_setspecific(key, cast(void*)0xbadc0de);

    thread_attachThis();
    rt_moduleTlsCtor();

    writefln("Hello from thread");

    return null;
}

private static extern (C) void threadExit(void *cd)
{
    writefln("Bye from thread");

    GC.disable();
    thread_detachThis();
    rt_moduleTlsDtor();
    GC.enable();

    pthread_setspecific(key, null);
}

int main()
{
    pthread_t thread;
    pthread_key_create(&key, &threadExit);
    pthread_create(&thread, null, &threadMain, null);
    pthread_join(thread, null);

    return 0;
}

Running it gives:

$ rdmd test.d
TLS ctor
TLS ctor
Hello from thread
Bye from thread
Segmentation fault

Valgrind says:

$ valgrind ./test
==3145== Memcheck, a memory error detector
==3145== Copyright (C) 2002-2010, and GNU GPL'd, by Julian Seward et al.
==3145== Using Valgrind-3.6.1-Debian and LibVEX; rerun with -h for
copyright info
==3145== Command: ./test
==3145==
TLS ctor
TLS ctor
Hello from thread
Bye from thread
==3145== Thread 2:
==3145== Invalid read of size 8
==3145==    at 0x452C72:
_D4core6thread6Thread6removeFPS4core6thread6Thread7ContextZv (in
/home/zor/test)
==3145==    by 0x449F8A:
_D4core6thread6Thread6removeFC4core6thread6ThreadZv (in
/home/zor/test)
==3145==    by 0x44A106: thread_detachThis (in /home/zor/test)
==3145==    by 0x44226E: threadExit (in /home/zor/test)
==3145==    by 0x4E36CE2: __nptl_deallocate_tsd (pthread_create.c:155)
==3145==    by 0x4E36F09: start_thread (pthread_create.c:311)
==3145==    by 0x533589C: clone (clone.S:112)
==3145==  Address 0x78 is not stack'd, malloc'd or (recently) free'd
==3145==
==3145==
==3145== Process terminating with default action of signal 11 (SIGSEGV)
==3145==  Access not within mapped region at address 0x78
==3145==    at 0x452C72:
_D4core6thread6Thread6removeFPS4core6thread6Thread7ContextZv (in
/home/zor/test)
==3145==    by 0x449F8A:
_D4core6thread6Thread6removeFC4core6thread6ThreadZv (in
/home/zor/test)
==3145==    by 0x44A106: thread_detachThis (in /home/zor/test)
==3145==    by 0x44226E: threadExit (in /home/zor/test)
==3145==    by 0x4E36CE2: __nptl_deallocate_tsd (pthread_create.c:155)
==3145==    by 0x4E36F09: start_thread (pthread_create.c:311)
==3145==    by 0x533589C: clone (clone.S:112)
==3145==  If you believe this happened as a result of a stack
==3145==  overflow in your program's main thread (unlikely but
==3145==  possible), you can try to increase the size of the
==3145==  main thread stack using the --main-stacksize= flag.
==3145==  The main thread stack size used in this run was 8388608.
==3145==
==3145== HEAP SUMMARY:
==3145==     in use at exit: 50,752 bytes in 16 blocks
==3145==   total heap usage: 23 allocs, 7 frees, 52,811 bytes allocated
==3145==
==3145== LEAK SUMMARY:
==3145==    definitely lost: 0 bytes in 0 blocks
==3145==    indirectly lost: 0 bytes in 0 blocks
==3145==      possibly lost: 288 bytes in 1 blocks
==3145==    still reachable: 50,464 bytes in 15 blocks
==3145==         suppressed: 0 bytes in 0 blocks
==3145== Rerun with --leak-check=full to see details of leaked memory
==3145==
==3145== For counts of detected and suppressed errors, rerun with: -v
==3145== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 4 from 4)
Killed


Does anyone have a clue what's happening here? Am I doing something very wrong, or might there be a bug in druntime?

Thanks,
Alex

This looks roughly correct.  I hadn't thought about adding calls to TlsCtor and TlsDtor, though perhaps I should.  Comments inline.

On Jan 20, 2012, at 6:20 AM, Alex wrote:

> Hi,
> 
> (Sorry if this is posted in the wrong place.)
> 
> Consider the following code:
> 
> import core.sys.posix.pthread,
>       core.memory,
>       core.thread,
>       std.stdio;
> 
> extern (C) void rt_moduleTlsCtor();
> extern (C) void rt_moduleTlsDtor();
> 
> __gshared pthread_key_t key;
> 
> static this() { writefln("TLS ctor"); }
> static ~this() { writefln("TLS dtor"); }
> 
> static extern(C) void* threadMain(void *arg)
> {
>    pthread_setspecific(key, cast(void*)0xbadc0de);
> 
>    thread_attachThis();
>    rt_moduleTlsCtor();
> 
>    writefln("Hello from thread");
> 
>    return null;
> }
> 
> private static extern (C) void threadExit(void *cd)
> {
>    writefln("Bye from thread");
> 
>    GC.disable();
>    thread_detachThis();
>    rt_moduleTlsDtor();

You might want to reverse the detach and TlsDtor calls here, since it's possible that static dtors might want access to the current Thread object.

Forums