TheaterFire

New Intel CPU Flaw Bypasses Spectre v2 Defenses to Leak Kernel Memory

Posted by ElementII5@reddit | hardware | View on Reddit | 7 comments

Reply to Post

7 Comments

RuinousRubric@reddit

With the ever-increasing number of performance-impacting security mitigations, I have to wonder if we'll eventually see cores specialize into secure and insecure variants. That's performance being left on the table after all, and there are use cases where data security either doesn't matter or can be achieved to an acceptable degree by airgapping. General purpose chips could have a mix, and just throw all of the sensitive tasks at the S cores.
View on Reddit #56262705

einmaldrin_alleshin@reddit

From what I understand, the vulnerability allows access to protected memory. So having a secure core on the same CPU wouldn't help, if the malware runs on an insecure core. So instead, you just don't want security critical stuff run on a shared system in the cloud.
View on Reddit #56283465

hardware-ModTeam@reddit

Thank you for your submission! Unfortunately, your submission has been removed for the following reason: * It is a duplicate post, or it contains the same information as a previous post. This also includes re-hosting news from the past as new information.
View on Reddit #56263149

AreYouAWiiizard@reddit

>This update mitigates the flaw with relatively low performance impact (up to 2.7% in lmbench tests). Lol such a terrible article, probably just AI summary, looks like they quoted the mistype from the original article that was supposed to represent Alder Lake. . These are the actual figures from the research paper: Mitigation overhead in UnixBench / lmbench. CPU | PRED_DIS_S | Retpoline ---|---|---- Sapphire Rapids | 1.7% / 6.4% | 2.4% / 8.0% Raptor Lake | 1.1% / 6.3% | 2.2% / 6.6% Alder Lake | 1.5% / 4.7% | 2.4% / 8.2% Rocket Lake | - | 3.1% / 8.3% Comet Lake | - | 0.5% / 2.4% Coffee Lake Refresh | - | 0.5% / 1.6%
View on Reddit #56255896

TheRealBurritoJ@reddit

The table shows the non-microcode mitigation overhead, tested by the researchers in the original paper, which is not the same thing. 2.7% was the highest measured overhead from the microcode update.
View on Reddit #56258161

AreYouAWiiizard@reddit

Ah you're right, I misread.
View on Reddit #56260526

ElementII5@reddit (OP)

Not good especially that eIBRS needs to be disabled.
View on Reddit #56253514