Application Level Threading

Contents

Application Level Threading#

Intel® Cryptography Primitives Library API can be used in a multi-threaded application. The library is thread-safe in the sense that you can call its functions from multiple threads. However, the library does not provide any internal synchronization mechanisms for stateful objects. It is the application’s responsibility to ensure that each thread operates on its own independent instance of a state, or that proper synchronization is applied when threads share state objects.

Warning

Every stateful algorithm (for example, XMSS, ML-KEM, ML-DSA, and others) maintains internal mutable state. If multiple threads share the same state object (such as a private key, public key, signature state or any other states) without proper synchronization, a data race may occur, leading to undefined behavior and potentially incorrect or insecure results.

Best Practices#

When using Intel® Cryptography Primitives Library in a multi-threaded application, consider the following best practices:

  • Allocate per-thread state objects. Each thread must have its own instances of all mutable state objects (private keys, public keys, signature states, scratch buffers, etc.). Do not share these objects across threads.

  • Read-only data can be shared. Immutable inputs such as message buffers, algorithm identifiers, and size constants can safely be shared across threads without synchronization.

  • Use thread sanitizers for validation. Tools such as Clang’s ThreadSanitizer (-fsanitize=thread) can detect data races at runtime during the development and testing phases and are strongly recommended for validating multi-threaded usage.

Contents