Researchers attack AMD's Infinity Fabric to bypass hardware security protections with 'Fabricked' — flaw lets malicious cloud hosts silently read

“Researchers attack AMD's Infinity Fabric to bypass hardware security protections with 'Fabricked' — flaw lets malicious cloud hosts silently read confidential VM memory and forge attestation reports​

Researchers at ETH Zurich disclosed a software-only vulnerability in April that silently undermines AMD SEV-SNP confidential computing protections on AMD's EPYC platforms, giving a malicious cloud host full read and write access to supposedly protected virtual machine memory. The technique, dubbed “Fabricked,” exploits flaws in how the CPU's Infinity Fabric interconnect handles memory routing during boot — and can forge the cryptographic attestation reports tenants rely on to verify their environment hasn't been tampered with.“

Source: https://www.tomshardware.com/pc-com...ntial-vm-memory-and-forge-attestation-reports

I mean now that AMD has some presence in the Datacenters and Enterprise, researchers were bound to start looking at their hardware eventually.

Intel got a lot of shit for all their flaws, but it's not like AMD has fewer, just different, but before a few years ago, there wasn't enough of it out there to be worth digging into; that is no longer the case.

Lakados said:

I mean now that AMD has some presence in the Datacenters and Enterprise, researchers were bound to start looking at their hardware eventually.

Intel got a lot of shit for all their flaws, but it's not like AMD has fewer, just different, but before a few years ago, there wasn't enough of it out there to be worth digging into; that is no longer the case.

Click to expand...

I seem to remember a flaw that allowed an attacker to embed malicious code inside the chip responsible for AMDs security essentially making the CPU compromised at a hardware level. Am I misremembering? If I'm not, seems to me AMD flaws have been found consistently for last couple of years. It just feels most of the flaws like to be brushed aside as not worth the worry because AMD could do not wrong as usual.

Surprising nobody.

Protecting the guest from a malicious host is incredibly difficult. Still a solid effort.

ZeroBarrier said:

I seem to remember a flaw that allowed an attacker to embed malicious code inside the chip responsible for AMDs security essentially making the CPU compromised at a hardware level. Am I misremembering? If I'm not, seems to me AMD flaws have been found consistently for last couple of years. It just feels most of the flaws like to be brushed aside as not worth the worry because AMD could do not wrong as usual.

Click to expand...

You are remembering correctly. There were 2 different ones.
Sinkclose, which attacked System Management Mode and Microcode Signature Verification Vulnerability, which attacked the Hardware Security Platform Processor.


They were overshadowed by the Intel ones at the time because the Intel "fixes" had noticeable performance impacts, and that was deserving of more attention.

Glad that the firmware update has been pushed and its relatively small scope of attack (EPYC is enterprise to begin with, and SEV-SNP fills a particular niche usage etc). While there will always be potential vulnerabilities, I wonder if this will encourage AMD to focus on OpenSIL, a (Coreboot compatible) FOSS alternative firmware to AGESA and typical proprietary UEFI elements. Years ago I remember reading that AMD was going to be implementing it widely in 2026, so I hope that continues to come to fruition, especially if the focus isn't exclusively on EPYC to start and instead is implemented across the range.

Lakados said:

You are remembering correctly. There were 2 different ones.
Sinkclose, which attacked System Management Mode and Microcode Signature Verification Vulnerability, which attacked the Hardware Security Platform Processor.


They were overshadowed by the Intel ones at the time because the Intel "fixes" had noticeable performance impacts, and that was deserving of more attention.

Click to expand...

Performance impacts vs compromised hardware. It's wild to me that people glossed over the potential for a CPU itself to carry malicious code allowing it to infect any system it is installed in (regardless of formatting and fresh install of the OS) over performance impacts.

Also, I was unaware there were 2 such vulnerabilities. I only remember hearing of one. Thank you for the clarification.