multi-threading problems using libcurl on Solaris 10

[anatolzo]

Hi, Solaris gurus. I ran into a problem of incorrect multi-threading performance with libCURL. Here's an exceprt of the code:
////////////////////////////////////////
// DEBUG only (getting desperate)
////////////////////////////////////
curbrp->GetSCV()->Lock();
curbrp->GetSCV()->Unlock();
////////////////////////////////////////
CURLcode rv = curl_easy_perform(m_curlH);
////////////////////////////////////////
// DEBUG only (getting desperate) - blows up after
// curl_easy_perform(m_curlH), if the SCV is created
// as a thread-specific data. Fine if SCV is on the stack
////////////////////////////////////
curbrp->GetSCV()->Lock();
curbrp->GetSCV()->Unlock();
////////////////////////////////////////

The comments say it all; SCV is short for StatefulCondVar - a simple wrapper around a pthread condition variable, which keeps state of the application. The stack trace after SIGABRT looks like this:
#0 0xfffffd7ffee142ea in _lwp_kill () from /usr/lib/64/libc.so.1
#1 0xfffffd7ffedff772 in Abort () from /usr/lib/64/libc.so.1
#2 0xfffffd7ffedff878 in thr_panic () from /usr/lib/64/libc.so.1
#3 0xfffffd7ffee07fbf in _ceil_prio_inherit () from /usr/lib/64/libc.so.1
#4 0xfffffd7ffee09a27 in mutex_lock_internal () from /usr/lib/64/libc.so.1
#5 0xfffffd7ffee09f2e in mutex_lock_impl () from /usr/lib/64/libc.so.1
#6 0xfffffd7ffee09fcb in pthread_mutex_lock () from /usr/lib/64/libc.so.1
#7 0x000000000046ef0f in clsMutex::Lock (this=0x4f7490) at ../../../erokar/common/clsMutex.h:27
#8 0xfffffd7ffe1af2a5 in CStatefulCondVar::Lock (this=0x4f7480) at StatefulCondVar.hh:35
#9 0xfffffd7ffe0f8c9e in CBCThread::main (this=0x2fe96b0, arg=0x0) at BudgetClientThread.cc:96
#10 0xfffffd7ffe0f8de0 in bcthread_main (arg=0x2fe96b0) at BudgetClientThread.cc:8
#11 0xfffffd7ffe0f8376 in BasicThread::startup (arg=0x2fe96b0) at ../../../adn/common/BasicThread.cc:48
#12 0xfffffd7ffee10edb in _thr_setup () from /usr/lib/64/libc.so.1
#13 0xfffffd7ffee11110 in ?? () from /usr/lib/64/libc.so.1
#14 0x0000000000000000 in ?? ()

As the comments say, when the SCV is not thread - specific, but allocated on the stack, the abort like this still happens, but after a couple of hours of load test.

Any ideas? The OS is 64-bit Solaris on Intel:
SunOS 5.10 Generic_120012-14 i86pc i386 i86pc Solaris

Thanks a lot in advance,

--Anatol