News for hackers. Stuff that matters.
Re:Side channel that leaked data from Intel CPUs patched by ... by makomk on Tuesday August 06, 2019 @08:26PM (#20630975)
Even though it's supposedly not feasable to exploit on Linux, there are fixes in the upstream kernel now and the changelog entries have some technical details. If I'm understanding this correctly, AMD systems don't speculatively execute GS-based accesses after speculatively excuting SWAPGS like Intel ones do, but that isn't enough to fully fix the problem on Linux because there's a conditional branch over the SWAPGS instruction which can be speculatively executed. Since that execution path doesn't have a SWAPGS all processors will quite happily continue speculatively executing code that fetches data via GS.
Re:Side channel that leaked data from Intel CPUs patched by ... by menzoic on Tuesday August 06, 2019 @10:08PM (#20631428)
What does speculative mean in this context?
Re:Side channel that leaked data from Intel CPUs patched by ... by viraptor on Tuesday August 06, 2019 @10:23PM (#20631504)
CPU speculating that it knows what the state will be after SWAPGS and continuing to load instructions into pipeline and partially executing them. In case of either a different branch bring taken, or SWAPGS causing a fault the pipeline will be rolled back, but side effects of the partial execution (like loads into cache) will not be.