Phoenix Attack Exposes Ongoing DDR5 Memory Security Risks

Just when it seemed modern DDR5 memory was secure, researchers have demonstrated that Rowhammer-style vulnerabilities are still very much alive. The newly revealed Phoenix attack, developed by the Computer Security Group (COMSEC) at ETH Zürich in collaboration with Google, exposes critical flaws in DDR5 memory, allowing attackers to escalate privileges and even gain root access. This proof-of-concept attack shows that existing countermeasures like TRR (Target Row Refresh), ECC, and on-die ECC (ODECC) are insufficient to stop determined adversaries.

How Phoenix Bypasses DDR5 Defenses

The Phoenix attack is a sophisticated evolution of Rowhammer, a technique that induces bit-flips in memory by rapidly and repeatedly accessing specific rows. By carefully orchestrating these access patterns, attackers can manipulate memory contents, potentially extracting sensitive data or altering critical code. 

Even though DDR5 introduced advanced mechanisms like TRR and ODECC, Phoenix sidesteps them. The core issue is that these defenses aren’t deterministic: TRR doesn’t precisely track row accesses, and ODECC is limited to correcting errors only during certain operations or after lengthy intervals, making persistent attacks viable.

Key Takeaways for System Administrators and Users

• Current DDR5 memory is still vulnerable to sophisticated Rowhammer attacks like Phoenix, even with advanced error correction and refresh mechanisms.
  
  
• Mitigations come with trade-offs: Increasing refresh rates reduces attack viability but impacts system performance.
  
  
• New standards like PRAC are on the horizon, but immediate, comprehensive fixes are not yet available for most DDR5 systems.

Real-World Testing and Impact

COMSEC’s team focused their tests on an AMD Zen 4 platform using 15 SK hynix DDR5 DIMMs manufactured between 2021 and 2024. The results were alarming. The attack achieved:

• 100% success rate in manipulating Page Table Entries (PTE) to access protected memory
• 73% chance of extracting SSH login keys from a virtual machine on the same server
• 33% probability of gaining root access by modifying the in-memory binary of the sudo utility

In one scenario, privilege escalation was achieved in just over five minutes. These results highlight the attack’s effectiveness, especially in multi-tenant server environments where thousands of clients may be affected.

Mitigation Attempts and Their Drawbacks

So far, there’s no bulletproof solution for Phoenix, particularly for SK hynix DDR5 modules. Researchers found that tripling the row refresh rate (tREFI) in UEFI settings significantly reduces the attack’s success, but at a steep cost, an 8.4% hit to system performance based on SPEC CPU2017 benchmarks. 

While AMD is reportedly preparing a BIOS update to address the issue, its effectiveness remains unproven. Meanwhile, Google notes that none of the current memory error correction or refresh mechanisms can guarantee full protection due to their inherent limitations.

Looking Ahead: PRAC and Future Standards

The persistent threat of Rowhammer has driven industry groups to devise new standards. JEDEC, the consortium behind memory technologies, introduced the Per-Row Activation Counting (PRAC) standard in April 2024. PRAC aims to precisely count row activations and alert the host system when a threshold is crossed, prompting mitigation measures. 

This innovation is slated for future DDR5 revisions and will be included from the outset in the upcoming LPDDR6 standard, offering hope for more robust protection against Rowhammer-style exploits.

For now, administrators should stay vigilant, apply BIOS/firmware updates promptly, and monitor for emerging memory security standards. As research continues and standards like PRAC gain adoption, the industry moves closer to closing one of memory’s most persistent security gaps.

Source: Tom's Hardware