Meltdown: Reading Kernel Memory from User Space
summary
Meltdown demonstrates that out of order execution leaks kernel memory by executing transient instructions that encode secret bytes into the cache and recovering them with a Flush+Reload cache side channel. It reads arbitrary physical memory, achieving about 582 KB/s when data hits L1, ~12.4 KB/s from L3 and only ~10 B/s for uncached lines (an improved method reaches ~3.2 KB/s). Mitigations include KAISER / KPTI; Intel CPUs are broadly affected, AMD mostly not, Apple devices are affected.
AI-written summary · about
chapters (22)
- 0:00Intro
- 0:24Kernel addresses
- 1:20Paging
- 2:39Architecture
- 4:10CPU Cache
- 6:10Out of Order Execution
- 7:27Toy Example
- 8:08Building Blocks
- 9:45Fault Handling
- 10:32Cover Channel
- 10:58Cache Attack
- 11:58Meltdown Attack
- 12:46New Experiment
- 13:23Uncatchable Memory
- 14:04Who is affected
- 14:34Exynos Mongoose
- 15:39Breaking the SLR
- 16:34Locating a victim process
- 17:05Using the physical address
- 17:36Fixing the hardware
- 18:14Kaiser
- 19:56Conclusion
related talks
24mThe Memory of a Meltdown
25mThe Memory of a Meltdown
43mUsing Undocumented CPU Behavior to See Into Kernel Mode and Break KASLR in the Process
50mPractical Ghostbusting: Meltdown, Spectre and You
30mSpectre, Meltdown, & Linux
01h 01mTutorial: Side-Channel Attacks and Transient Execution Vulnerab...
19mKeynote: MDS, Fallout, Zombieland & Linux
41mUsing Microarchitectural Design to Break KASLR and More
