[linux-audio-user] libfst: was: Steinberg The Grand

[Werner Schweer]

On Friday 04 June 2004 14:18, torbenh at gmx.de wrote:
> On Fri, Jun 04, 2004 at 10:16:53AM +0200, Werner Schweer wrote:
> > On Thursday 03 June 2004 20:52, torbenh at gmx.de wrote:
> > > On Thu, Jun 03, 2004 at 08:20:03PM +0200, Werner Schweer wrote:
> > > > On Thursday 03 June 2004 19:33, Mark Knecht wrote:
> > > > > torbenh at gmx.de wrote:
> > > >
> > > > my changed libwinelib.c looks like this:
> > > >
> > > > int
> > > > SharedWineInit (void (*sighandler)(int,siginfo_t*,void*))
> > > > {
> > > >       unsigned char     Error[2024]="";
> > > >       char *WineArguments[] = {"sharedapp", LIBPATH "/libfst.so",
> > > > NULL}; void* stackbase;
> > > >       size_t stacksize;
> > > >       void *ntdll;
> > > >       void *ntso;
> > > >       char ntdllpath[PATH_MAX+1];
> > > >       char* dlerr;
> > > >       char* env[] = { NULL };
> > > >
> > > >       sharedwine_signal_handler = sighandler;
> > > >
> > > >       if (setjmp (jump) == 0) {
> > > >             wine_init (2, WineArguments, env, Error, sizeof (Error));
> > > >       ...
> > >
> > > which wine version are you using ?
> > > i really dont know where this env parameter comes from.
> > > we have copied some code out of wine and libfst uses this.
> > > (its the threading code from wine)
> > >
> > > if this code is expected to behave differently than it does
> > > your pretty much doomed.
> >
> > its from cvs, last updated yesterday (03.06.2004).
>
> hmm... ok... then we will need to adapt libfst to the new wine_init call,
> to be prepared for the next release.
>
> do you already know what #if i need ?
>
> > > please get your wine from www.winehq.org
> >
> > i tried again with wine-20040505. Calling fstconfig produces a core
> > dump. Gdb says:
> >
> > (gdb) bt
> > #0  __pthread_sighandler (signo=11, ctx=
> >       {gs = 0, __gsh = 0, fs = 51, __fsh = 0, es = 123, __esh = 0, ds =
> > 123, __dsh = 0, edi = 131120, esi = 0, ebp = 1080816716, esp =
> > 1080816708, ebx = 1075170628, edx = 0, ecx = 1, eax = 1082129376, trapno
> > = 14, err = 4, eip = 1075232782, cs = 115, __csh = 0, eflags = 2163206,
> > esp_at_signal = 1080816708, ss = 123, __ssh = 0, fpstate = 0x0, oldmask =
> > 2147483648, cr2 = 1082129788})
> >     at sighandler.c:28
> > #1  <signal handler called>
> > #2  0x4016c00e in __pthread_internal_tsd_get (key=1077936127) at
> > specific.c:225
> > #3  0x400a434e in __libc_calloc (n=1, elem_size=1077936127) at
> > malloc.c:3585 #4  0x400d7813 in __opendir (name=0x20000 <Address 0x20000
> > out of bounds>) at ../sysdeps/unix/opendir.c:146
> > #5  0x40315820 in wine_nt_to_unix_file_name (nameW=0x406bfb58,
> >     unix_name_ret=0x406bfb50, disposition=1, check_case=0 '\0')
> >     at directory.c:873
> > #6  0x403164e0 in RtlDoesFileExists_U (file_name=0x40380750)
> >     at directory.c:1215
> > #7  0x40328273 in RtlDosSearchPath_U (paths=0x40380510,
> > search=0x406bfe08, ext=0x0, buffer_size=64, buffer=0x406bfd2c,
> > file_part=0x406bfc68) at path.c:428
> > #8  0x40320e29 in load_dll (load_path=0x40380510, libname=0x406bfe08,
> > flags=0, pwm=0x406bfe00) at loader.c:1359
> > #9  0x40320093 in fixup_imports (wm=0x403806c8, load_path=0x40380510)
> >     at loader.c:384
> > #10 0x40322262 in LdrInitializeThunk (main_file=0x0, unknown2=0,
> > unknown3=0, unknown4=0) at loader.c:1930
> > #11 0x4051296b in start_process (arg=0x0) at process.c:822
> >
> > its essential the same as in the current wine cvs version.
> > My system uses glibc 2.3.2, linux 2.6.7-rc1 (no nptl).
>
> pthread_self() is broken in your case.
> i have seen this behaviour in a signal handler with an alternate stack.
> my pthread_implementation does not use thread local storage and
> therefore pthread_self uses the stack to identify the thread.
>
>
> newer glibc should be using %gs for this stuff.
> but your %gs register is 0.
>
> very strange.
> can you check what the registers contain using breakpoints ?
>
> gdb> info registers
>
> it would be nice if you can find out with what options your glibc is
> configured with.
>
> this must be a glibc issue because its working for marek, who uses the
> same glibc i do.

i do not remember the configure options for my glibc 2.3.3. I replaced it this
weekend with a brand new version 2.3.4 (nptl). And it now works! So i believe
it was a glibc problem on my system. 

Googling around i found the following information about glibc and %gs usage
in the thread implementation:
(http://www-124.ibm.com/developerworks/oss/jikesrvm/userguide/HTML/faq_building.html):

"In order for our m-to-n threading scheme to work, we need to be able to swap 
stacks for a single pthread. This only works if glibc is built with an option 
to use the GS segment register to address certain pthread-specific state. 
Otherwise, glibc attempts to use stack-pointer arithmetic to find 
pthread-specific state, which fails once we bash the stack. 
On Red Hat? Linux and SuSE® Linux, glibc is built with this option by default. 
On Debian® GNU/Linux, as of this writing (6 May 2004), the version of x86 
libc6 
in the Debian Woody distribution does not use GS-segment addressing. 
Debian/Sarge 
will use GS-segment addressing on 2.6 Linux kernels, but will not use it on 
2.4 and 2.2 kernels."

/werner