Incorrect Conversion of Security Identifiers

In a System-On-Chip (SoC), various integrated circuits and hardware engines generate transactions such as to access (reads/writes) assets or perform certain actions (e.g., reset, fetch, compute, etc.). Among various types of message information, a typical transaction is comprised of source identity (to identify the originator of the transaction) and a destination identity (to route the transaction to the respective entity). Sometimes the transactions are qualified with a security identifier. This security identifier helps the destination agent decide on the set of allowed actions (e.g., access an asset for read and writes).

A typical bus connects several leader and follower agents. Some follower agents implement bus protocols differently from leader agents. A protocol conversion happens at a bridge to seamlessly connect different protocols on the bus. One example is a system that implements a leader with the Advanced High-performance Bus (AHB) protocol and a follower with the Open-Core Protocol (OCP). A bridge AHB-to-OCP is needed to translate the transaction from one form to the other.

A common weakness that can exist in this scenario is that this conversion between protocols is implemented incorrectly, whereupon an untrusted agent may gain unauthorized access to an asset.

Consider a system that supports AHB. Let us assume we have a follower agent that only understands OCP. To connect this follower to the leader, a bridge is introduced, i.e., AHB to OCP.

The follower has assets to protect accesses from untrusted leaders, and it employs access controls based on policy, (e.g., AES-Key registers for encryption or decryption). The key is 128 bits implemented as a set of four 32-bit registers. The key registers are assets, and register AES_KEY_ACCESS_POLICY is defined to provide the necessary access controls.

The AES_KEY_ACCESS_POLICY access-policy register defines which agents with a security identifier in the transaction can access the AES-key registers. The implemented AES_KEY_ACCESS_POLICY has 4 bits where each bit when "Set" allows access to the AES-Key registers to the corresponding agent that has the security identifier. The other bits from 31 through 4 are reserved and not used.

During conversion of the AHB-to-OCP transaction, the security identifier information must be preserved and passed on to the follower correctly.