AMD SEV OMG: Trusted execution in VMs undone by bad hypervisors' cache meddling

Boffins in Germany and Austria have found a flaw in AMD's SEV trusted execution environment that makes it less than trustworthy. The researchers – Ruiyi Zhang, Lukas Gerlach, Daniel Weber, Lorenz Hetterich, and Michael Schwarz (all with CISPA Helmholtz Center for Information Security), Youheng Lü (independent), Andreas Kogler …

[abend0c4]

This, and the almost simultaneous report of an Intel flaw, do suggest that adding complexity to instruction sets - even in the supposed interests of security - is likely ultimately to be a source of woe.

[Jou (Mxyzptlk)]

"attacker is running a malicious hypervisor"

So this attack only works on those encrypted running machines IF the attacker already took over the hypervisor? Doesn't look that as problematic as the Intel flaw of this month where a guest VM can take down the host.

[TReko]

Re: "attacker is running a malicious hypervisor"

Yes, if you have a malicious hypervisor then it's already game over without needing this complex attack

[Rattus]

Re: "attacker is running a malicious hypervisor"

Most of us run our cloud / semi-hosted services on other people's computers.

Given that the security boundary for such services is supposed to be that the hypervisor (operated by someone other than us) cannot access the VM that we control.

This flaw demonstrates that this security boundary (as provided by SEV and such) is NOT the secure solution that we as VM customers are being told exists (and what we are paying for). Multi hosted VMs were supposed to free us from Dedicated or Co-Lo boxes, enabling our service providers to offer a lower cost shared platform without the security holes.... Once again this "solution" is proven to be flawed.

[Lennart Sorensen]

Re: "attacker is running a malicious hypervisor"

You mean there are people who actually believed this stuff would work and are using it under that impression? And paying extra for it?

[Jou (Mxyzptlk)]

Re: "attacker is running a malicious hypervisor"

Did I say anything like that? No, you just construct something I did not say and did not hint just to play your "I am angry and have to tell why" card. In the end you only throw the general "why cloud is wrong and everything is a lie" around.

Apart from that AMD has a fix in the pipeline for this problem. It is not like they deliberately designed SEV to be leaky, like you clearly hint.

[Rattus]

Re: "attacker is running a malicious hypervisor"

You state "So this attack only works on those encrypted running machines IF the attacker already took over the hypervisor? "

"Did I say anything like that? " Well yes that is exactly what you said...

In this case this is what SEV is supposed to protect against.

I am sure that there will be a fix.

I agree with you that VM to VM is a far larger attack vector, but that was not *this* story.

At no point did I attempt to imply AMD deliberately designed SEV to be leaky any more that you were suggesting an Intel conspiracy