Incorrect Decoding 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. The security identifier helps the destination agent decide on the set of allowed actions (e.g., access an asset for read and writes). A decoder decodes the bus transactions to map security identifiers into necessary access-controls/protections.

A common weakness that can exist in this scenario is incorrect decoding because an untrusted agent's security identifier is decoded into a trusted agent's security identifier. Thus, an untrusted agent previously without access to an asset can now gain access to the asset.

Consider a system that has four bus masters and a decoder. The decoder is supposed to decode every bus transaction and assign a corresponding security identifier. The security identifier is used to determine accesses to the assets. The bus transaction that contains the security information is Bus_transaction [15:14], and the bits 15 through 14 contain the security identifier information. The table below provides bus masters as well as their security identifiers and trust assumptions:

The assets are the AES-Key registers for encryption or decryption. The key is 128 bits implemented as a set of four 32-bit registers. The AES_KEY_ACCESS_POLICY is used to define which agents with a security identifier in the transaction can access the AES-key registers. The size of the security identifier is 4 bits (i.e., bit 3 through 0). Each bit in these 4 bits defines a security identifier. There are only 4 security identifiers that are allowed accesses to the AES-key registers. The number of the bit when set (i.e., "1") allows respective action from an agent whose identity matches the number of the bit. If clear (i.e., "0"), disallows the respective action to that corresponding agent.