Arch Linux

Description:
`clinfo` hangs when intel-oneapi-compiler-shared-opencl-cpu is installed. (This can also affect other programs, like libreoffice.)


Additional info:
* intel-oneapi-compiler-shared-opencl-cpu 2023.0.0-7

Steps to reproduce:
1) install intel-oneapi-compiler-shared-opencl-cpu
2) start clinfo
3) observe that it hangs

---

gdb reveals the following stack trace:

Starting program: /usr/bin/clinfo -l
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib/libthread_db.so.1".
^C
Program received signal SIGINT, Interrupt.
futex_wait (private=0, expected=1, futex_word=0x7ffff7fc3878) at ../sysdeps/nptl/futex-internal.h:146
146 int err = lll_futex_timed_wait (futex_word, expected, NULL, private);
(gdb) bt
#0 futex_wait (private=0, expected=1, futex_word=0x7ffff7fc3878) at ../sysdeps/nptl/futex-internal.h:146
#1 futex_wait_simple (private=0, expected=1, futex_word=0x7ffff7fc3878) at ../sysdeps/nptl/futex-internal.h:177
#2 __pthread_once_slow (once_control=0x7ffff7fc3878, init_routine=0x7ffff7fc0940) at pthread_once.c:105
#3 0x00007ffff7fbd4a3 in clGetExtensionFunctionAddress () from /opt/intel/oneapi/compiler/2023.0.0/linux/lib/libOpenCL.so
#4 0x00007ffff7fbc6e7 in ?? () from /opt/intel/oneapi/compiler/2023.0.0/linux/lib/libOpenCL.so
#5 0x00007ffff7fc0bd9 in ?? () from /opt/intel/oneapi/compiler/2023.0.0/linux/lib/libOpenCL.so
#6 0x00007ffff7aa3b17 in __pthread_once_slow (once_control=0x7ffff7fc3878, init_routine=0x7ffff7fc0940) at pthread_once.c:116
#7 0x00007ffff7fbd4a3 in clGetExtensionFunctionAddress () from /opt/intel/oneapi/compiler/2023.0.0/linux/lib/libOpenCL.so
#8 0x00007ffff7c02fc4 in ?? () from /opt/cuda/lib64/libOpenCL.so.1
#9 0x00007ffff7aa3b17 in __pthread_once_slow (once_control=0x7ffff7e070d0, init_routine=0x7ffff7c02a60) at pthread_once.c:116
#10 0x00007ffff7c048df in clGetPlatformIDs () from /opt/cuda/lib64/libOpenCL.so.1
#11 0x000055555555b1e6 in ?? ()
#12 0x00007ffff7a3c790 in __libc_start_call_main (main=main@entry=0x55555555b020, argc=argc@entry=2, argv=argv@entry=0x7fffffffdf48) at ../sysdeps/nptl/libc_start_call_main.h:58
#13 0x00007ffff7a3c84a in __libc_start_main_impl (main=0x55555555b020, argc=2, argv=0x7fffffffdf48, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffffffdf38)
at ../csu/libc-start.c:360
#14 0x000055555555b8be in ?? ()

which shows that the problem is that the same pthread_once_t object (at 0x7ffff7fc3878) is recursively initialized (frames 2 and 6).

I have observed this on two machines, it happened even when it was the only opencl implementation (no nvidia, no `intel-compute-runtime`, etc.). I have tried to find an upstream bug tracker but the PKGBUILD just downloads binary packages; the closest I found on github was https://github.com/intel/compute-runtime but I am not sure if that is the source. So I decided to report it here.