Re: RtlEnterCriticalSection

[Marcus Meissner]

On Thu, Oct 18, 2001 at 12:22:10AM +0200, Michael Marxmeier wrote:
> After encountering timeout messages from
> RtlEnterCriticalSection i had a look at it.
>
> Since interlocked_inc() should never return zero
> the following patch is likely necessary.
> I was unable to test this as now wine crashes with
> a segmentation fault (which also terminates gdb).

That is a sign that your patch is faulty.

Any number of threads can increase this counter, it only succeeds
if the count is really 0 before the call. (0 meaning the critical section
is completely unlocked).

If it is any other number, it is either a recursive enter or a 
different thread.

Ciao, Marcus
> Michael
>
> --- critsection.c.orig  Wed Oct 17 23:24:13 2001
> +++ critsection.c       Thu Oct 18 00:04:32 2001
> @@ -233,7 +233,7 @@
>    */
>   NTSTATUS WINAPI RtlEnterCriticalSection( RTL_CRITICAL_SECTION *crit )
>   {
> - if (interlocked_inc( &crit->LockCount ))
> - if (interlocked_inc( &crit->LockCount ) != 1)
>  {
>      if (crit->OwningThread == GetCurrentThreadId())
>      {

RtlEnterCriticalSection

[Michael Marxmeier]

Marcus Meissner wrote:
> On Thu, Oct 18, 2001 at 12:22:10AM +0200, Michael Marxmeier wrote:
> > After encountering timeout messages from
> > RtlEnterCriticalSection i had a look at it.
> >
> > Since interlocked_inc() should never return zero
> > the following patch is likely necessary.
> > I was unable to test this as now wine crashes with
> > a segmentation fault (which also terminates gdb).
> That is a sign that your patch is faulty.

Yep, looks you're right. The lock count is initialied
to -1 so the code is correct.
Seems the critical sections in ole32 do not become
initialized so the lock count is zero ...
> Ciao, Marcus

Thanks for your correction.

Michael

RtlEnterCriticalSection

[Michael Marxmeier]

Michael Marxmeier wrote:
> Seems the critical sections in ole32 do not become
> initialized so the lock count is zero ...

Found it ...

Seems COMPOBJ_InitProcess() in dlls/ole32/compobj.c
never gets called so the csRegisteredClassList
critical section is uninitialized. After making sure
the critical section gets intialized it works as
expected.

As far as i can see this should get called from
OLE32_DllEntryPoint() in dlls/ole32/ole32_main.c
when the DLL is loaded. No idea why it does not and
where to look next ...

Any clues to share?

Thanks
Michael