Base · Medium

CWE-1243: Sensitive Non-Volatile Information Not Protected During Debug

Access to security-sensitive information stored in fuses is not limited during debug.

CWE-1243 · Base Level ·1 Mitigations

Description

Access to security-sensitive information stored in fuses is not limited during debug.

Several security-sensitive values are programmed into fuses to be used during early-boot flows or later at runtime. Examples of these security-sensitive values include root keys, encryption keys, manufacturing-specific information, chip-manufacturer-specific information, and original-equipment-manufacturer (OEM) data. After the chip is powered on, these values are sensed from fuses and stored in temporary locations such as registers and local memories. These locations are typically access-control protected from untrusted agents capable of accessing them. Even to trusted agents, only read-access is provided.

Potential Impact

Confidentiality, Access Control

Modify Memory, Read Memory, Bypass Protection Mechanism

Demonstrative Examples

Sensitive manufacturing data (such as die information) are stored in fuses. When the chip powers on, these values are read from the fuses and stored in microarchitectural registers. These registers are only given read access to trusted software running on the core. Untrusted software running on the core is not allowed to access these registers.
Bad
All microarchitectural registers in this chip can be accessed through the debug interface. As a result, even an untrusted debugger can access this data and retrieve sensitive manufacturing data.
Good
Registers used to store sensitive values read from fuses should be blocked during debug. These registers should be disconnected from the debug interface.
The example code below is taken from one of the AES cryptographic accelerators of the HACK@DAC'21 buggy OpenPiton SoC [REF-1366]. The operating system (OS) uses three AES keys to encrypt and decrypt sensitive data using this accelerator. These keys are sensitive data stored in fuses. The security of the OS will be compromised if any of these AES keys are leaked. During system bootup, these AES keys are sensed from fuses and stored in temporary hardware registers of the AES peripheral. Access to these temporary registers is disconnected during the debug state to prevent them from leaking through debug access. In this example (see the vulnerable code source), the registers key0, key1, and key2 are used to store the three AES keys (which are accessed through key_big0, key_big1, and key_big2 signals). The OS selects one of these three keys through the key_big signal, which is used by the AES engine.
Bad
...
                        assign key_big0    = debug_mode_i ? 192'b0 : {key0[0],
                        key0[1], key0[2], key0[3], key0[4], key0[5]};
                        
                        assign key_big1    = debug_mode_i ? 192'b0 : {key1[0],
                        key1[1], key1[2], key1[3], key1[4], key1[5]};
                        
                        assign key_big2    = {key2[0], key2[1], key2[2],
                        key2[3], key2[4], key2[5]};
                        ...
                        assign key_big = key_sel[1] ? key_big2 : ( key_sel[0] ?
                        key_big1 : key_big0 );
                        ...
The above code illustrates an instance of a vulnerable implementation for blocking AES key mechanism when the system is in debug mode (i.e., when debug_mode_i is asserted). During debug mode, key accesses through key_big0 and key_big1 are effectively disconnected, as their values are set to zero. However, the key accessed via the key_big2 signal remains accessible, creating a potential pathway for sensitive fuse data leakage, specifically AES key2, during debug mode. Furthermore, even though it is not strictly necessary to disconnect the key_big signal when entering debug mode (since disconnecting key_big0, key_big1, and key_big2 will inherently disconnect key_big), it is advisable, in line with the defense-in-depth strategy, to also sever the connection to key_big. This additional security measure adds an extra layer of protection and safeguards the AES keys against potential future modifications to the key_big logic.
To mitigate this, disconnect access through key_big2 and key_big during debug mode [REF-1367].
Good
...
                    assign key_big0    = debug_mode_i ? 192'b0 : {key0[0],
                    key0[1], key0[2], key0[3], key0[4], key0[5]};
                    
                    assign key_big1    = debug_mode_i ? 192'b0 : {key1[0],
                    key1[1], key1[2], key1[3], key1[4], key1[5]}; 
                    
                    assign key_big2    = debug_mode_i ? 192'b0 : {key2[0],
                    key2[1], key2[2], key2[3], key2[4], key2[5]};
                    ...
                    assign key_big = debug_mode_i ? 192'b0 : (  key_sel[1] ?
                    key_big2 : ( key_sel[0] ? key_big1 : key_big0 )  );
                    ...

Mitigations & Prevention

Architecture and DesignImplementation

Disable access to security-sensitive information stored in fuses directly and also reflected from temporary storage locations when in debug mode.

Frequently Asked Questions

What is CWE-1243?

CWE-1243 (Sensitive Non-Volatile Information Not Protected During Debug) is a software weakness identified by MITRE's Common Weakness Enumeration. It is classified as a Base-level weakness. Access to security-sensitive information stored in fuses is not limited during debug.

How can CWE-1243 be exploited?

Attackers can exploit CWE-1243 (Sensitive Non-Volatile Information Not Protected During Debug) to modify memory, read memory, bypass protection mechanism. This weakness is typically introduced during the Architecture and Design, Implementation phase of software development.

How do I prevent CWE-1243?

Key mitigations include: Disable access to security-sensitive information stored in fuses directly and also reflected from temporary storage locations when in debug mode.

What is the severity of CWE-1243?

CWE-1243 is classified as a Base-level weakness (Medium abstraction). Its actual severity depends on the specific context and how the weakness manifests in your application.