AMD last week informed customers that it has updated mitigations for a variant of the Spectre side-channel attack. The update comes in response to research conducted by Intel.
The Meltdown and Spectre attack methods, which can be exploited to obtain potentially sensitive bits of information from a device’s memory by abusing CPUs, were disclosed in January 2018. The most dangerous of the Spectre attacks was dubbed Spectre v2 and Spectre BTI (Branch Target Injection), and it’s tracked as CVE-2017-5715.
The BTI attack, according to Intel, “takes advantage of the indirect branch predictors used by processors to direct what operations are speculatively executed after a near indirect branch instruction. By controlling how indirect branch predictors operate, an attacker can cause certain instructions to be speculatively executed and then use the effects the malicious code has on the processor’s caches to infer data values.”
In response to the discovery of the Meltdown and Spectre attacks, Intel, AMD, Arm and other affected CPU vendors implemented mitigations that should prevent attacks. However, those mitigations can in some cases be bypassed.
One such bypass was demonstrated recently by researchers from the Vrije Universiteit Amsterdam, who identified a new variant — dubbed Spectre-BHB — that can bypass mitigations implemented by Intel and Arm for Spectre v2 attacks.
AMD does not appear to be impacted by Spectre-BHB, but research conducted recently by Intel showed that one of the Spectre v2 mitigations implemented by AMD for BTI is not effective in certain cases. The problematic mitigation is named LFENCE/JMP and is identified as “Mitigation v2-2” in AMD documentation.
“LFENCE/JMP is an existing software mitigation option for BTI and similar transient execution attacks stemming from indirect branch predictions, which is commonly used on AMD processors,” the Intel researchers explained. “However, the effectiveness of this mitigation can be compromised by the inherent race condition between the speculative execution of the predicted target and the architectural resolution of the intended target, since this can create a window in which code can still be transiently executed.”
The researchers released a paper detailing their findings last week. On the same day, AMD confirmed Intel’s findings and provided a list of processors for which LFENCE/JMP may not sufficiently mitigate Spectre v2 attacks.
The chipmaker has advised customers to use the mitigations named Generic Retpoline (Mitigation v2-1) or IBRS (Mitigation v2-4).
“Currently in Linux, users can control which mitigation is used at boot time. Users can choose the generic retpoline at boot time by using the spectre_v2 Linux kernel command for turning on retpoline: spectre_v2=retpoline,generic,” AMD explained. “Alternatively, users can update their version of the Linux kernel that incorporates a patch provided by AMD to the Linux community. The patch includes using generic retpoline, if retpoline is enabled and not explicitly set to the AMD Retpoline (spectre_v2=retpoline,amd).”
AMD last week also provided an update for another speculative execution mitigation, which came in response to research conducted by grsecurity. The issue, tracked as CVE-2021-26341, was detailed in a blog post published on March 8.
AMD pointed out that it’s not aware of any in-the-wild Spectre attacks targeting devices powered by its processors.
Related: AMD Chipset Driver Vulnerability Can Allow Hackers to Obtain Sensitive Data
Related: Voltage Glitching Attack on AMD Chips Poses Risk to Cloud Environments
Related: Researchers Disclose New Side-Channel Attacks Affecting All AMD CPUs
