Google says it can mitigate Spectre with "negligible" effect

Two days ago, an industry/academic team released a terrifying alert about a pair of CPU bugs called Spectre and Meltdown that allowed one program to steal data from another, even with the best memory-management and isolation techniques — news that meant that virtually all the mission-critical computers in the world could no longer be trusted to handle sensitive data securely.

Both bugs attack "speculative execution" — a performance-enhancing technique used in microprocessors, in which the processor makes shrewd predictions about operations it will be asked to undertake in the near future and does that work pro-actively, discarding the results when it guesses wrong. Because speculative execution is so important to processor speed, the initial news alerts warned that mitigation would impose up to 30% performance-hits on processors.

Yesterday, the Google Security Blog featured an analysis of two new techniques for guarding against Spectre-style speculative execution attacks: "Retpoline" ("a binary modification technique that protects against 'branch target injection' attacks") and "Kernel Page Table Isolation" ("a general purpose technique for better protecting sensitive information in memory from other software running on a machine"), both of which had been deployed to all of Google's worldwide Linux systems, which power the majority of Google's services.

On these systems, the mitigation techniques imposed "negligible impact on performance" — a far cry from the 30% we were warned of the day before. Google warns that your mileage may vary: since these techniques guard against interference with speculative execution, they may be highly application-specific, so Google advises "thorough testing in your environment before deployment; we cannot guarantee any particular performance or operational impact."

Google's Mitigations Against CPU Speculative Execution Attack Methods [Google Help]

More details about mitigations for the CPU Speculative Execution issue [Matt Linton/Google Security Blog]

(via The Verge)