Breaking and Fixing the Kernel: Crashes, Debugging & Performance Tun...
summary
Covers four concrete kernel crash classes—null pointer dereference, stack overflow, use after free and divide by zero—reproduced on Azure Linux VMs and diagnosed with gdb and KASAN traces from dmesg and kernel crash reports. Shows a stack overflow module and a use after free module that KASAN flags, demonstrates live backtraces with gdb, and quantifies a real-world cost: enabling the SRSO mitigation in newer kernels causes about a 12% performance drop versus 5.15. Explains srso_untrained_return_alias/srso_safe_return_alias and the security versus performance tradeoffs of disabling mitigations.
AI-written summary · about
chapters (15)
- 0:00Kernel crashes and debugging
- 1:09Three stages of kernel breakage
- 2:35Stack overflow debugging
- 6:07KASAN for memory errors
- 8:14Runtime debugging with GDB
- 9:41Use-after-free debugging
- 12:53Null pointer dereference impact
- 15:19Divide-by-zero errors
- 17:28Kernel crash prevention
- 19:13Kernel command line and performance
- 23:01Kernel mitigations explained
- 24:03Speculative execution flaws
- 26:13Mitigating SRSO with RSB
- 30:08Performance impact of mitigations
- 31:34Q&A
related talks
52mLinux Kernel Debugging: Going Beyond Printk Messages
37mPracticing Linux Crash/Panic Issue on Production and Cloud Server: Using Kdump + Crash
59mSanitizing the Linux Kernel — On KASAN and other Dynamic Bug-finding Tools
27mBreaking Boundaries & Demystifying Kernel SU 4 Root Access In Azure Cloud Shell
06mQ&A
18mExploiting Uses of Uninitialized Stack Variables in Linux Kernels to Leak Kernel Pointers
46mThe State of Kernel Self Protection Project
30mSpectre, Meltdown, & Linux