More than six years since the Spectre security flaw was first exposed, new research indicates that recent AMD and Intel processors are still vulnerable to speculative execution attacks. Researchers Johannes Wikner and Kaveh Razavi from ETH Zürich have identified a method to bypass the Indirect Branch Predictor Barrier (IBPB) on x86 chips, a crucial safeguard against speculative execution attacks.
Speculative execution optimizes CPU performance by predicting and executing certain instructions out-of-order. When predictions are correct, tasks complete faster, but mispredictions (transient instructions) are declared invalid, and the processor resumes execution with the correct value. Although results of transient instructions are not saved in the architectural state, they may still load sensitive data into the processor cache through forced misprediction, potentially exposing this data to attackers.
Understanding the New Attack
Intel’s IBPB is designed to establish a barrier that blocks indirect branch predictions from being influenced by instructions executed prior to the barrier. This mechanism is intended to protect against Branch Target Injection (BTI), also known as Spectre v2 (CVE-2017-5715). However, ETH Zürich researchers found that Intel’s Golden Cove and Raptor Cove architectures contain a microcode flaw that allows attackers to retain branch predictions even after IBPB clears them, circumventing security boundaries and enabling cross-process Spectre leaks.
AMD’s implementation of IBPB, meanwhile, also proved vulnerable. Due to Linux kernel application inconsistencies, an attack known as Post-Barrier Inception (PB-Inception) can leak privileged memory on AMD Zen 1(+) and Zen 2 processors, enabling unprivileged attackers to access sensitive data.
To address these issues, Intel has released a microcode patch (CVE-2023-38575) with a moderate CVSS score of 5.5, while AMD has tracked the vulnerability under CVE-2022-23824. Users are advised to ensure their microcode and kernel updates are current to mitigate potential exposure.
RowHammer Returns with SpyHammer Variations
Alongside Spectre, ETH Zürich researchers disclosed advancements in RowHammer attacks, specifically ZenHammer and SpyHammer. RowHammer exploits DRAM hardware vulnerabilities by repeatedly accessing memory rows to cause bit flips in adjacent rows, corrupting memory in an attack that now correlates with system temperature.
"RowHammer is extremely sensitive to temperature fluctuations, even within ±1 °C," the ETH Zürich study explains. SpyHammer, a new attack variation, leverages RowHammer’s temperature sensitivity to infer DRAM temperature with high precision. This technique allows attackers to determine system utilization and even deduce when individuals enter or leave rooms based on temperature shifts.
"SpyHammer can monitor temperature in critical systems without requiring modifications or prior knowledge of the target system,” ETH Zürich researchers warn. Until a definitive RowHammer defense is established, this vulnerability could significantly impact both security and privacy, as the RowHammer flaw worsens with further technology scaling.
The discovery of Post-Barrier Inception and SpyHammer highlights ongoing security risks in CPU hardware, especially as these vulnerabilities remain challenging to mitigate fully. With new defenses in development, the research underscores the necessity for users and organizations to stay informed and apply security updates promptly.
