This suite of tests crashes intermittently on BSW

repeatedly running the single test dEQP-VK.api.object_management.multithreaded_per_thread_device.shader_module
should generate a segfault about 25% of the time.

The stacktrace leaves mesa for libgcrypt at src/util/mesa-sha1.c:190
gcry_control(GCRYCTL_INITIALIZATION_FINISHED, 0);

Mesa initializes libgcrypt on first use, based on a static int.  This is not sufficient to protect the initialization routine from simultaneous execution.  Based on https://gnupg.org/documentation/manuals/gcrypt/Multi_002dThreading.html#Multi_002dThreading there should be a real mutex here.