TSAN has found a lock-order inversion between some libcrypto internal mutex and another lock. The inversion occurs in one direction when constructing the SSLManagerOpenSSL, and a mongo-initializer (it appears to be this one: https://github.com/10gen/mongo/blob/cbb2dd5bbf0407792b3a808bd8adc131e6fb9b51/src/mongo/util/net/openssl_init.cpp#L204 given that we just logged that we completed fips mode activation before the inversion is detected)