Technical Briefing #4 2026: Kernel Memory Leaks, THP Latency, Linked Lists & Android Internals

This technical briefing (Issue 4, 2026) collects and comments on ten in-depth technical articles spanning Linux kernel internals, performance optimization, memory management, and Android systems.

1. ILL Crash Caused by Disappearing Code Segment

References a Juejin post ( juejin.cn/post/7616584171634065414) where the author uses custom tooling to trace an ILL crash to a kernel bug, demonstrating deep systems debugging expertise.

2. Finding and Fixing Ghostty's Largest Memory Leak

Article by Mitchell Hashimoto ( mitchellh.com/writing/ghostty-memory-leak-fix) details a memory leak in the Ghostty terminal, easily triggered by Claude Code. The post explains the root cause, the fix, and the memory leak detection toolchain:

Tagged anonymous memory (similar to Android's approach) made it trivial to identify the leak, associate it with the PageList, and verify the fix by observing tagged memory being properly freed.

Debug builds and unit tests use leak-detecting Zig allocators.

CI runs valgrind on the full unit test suite on every commit to catch leaks and undefined memory usage.

macOS GUI is regularly run via macOS Instruments to find leaks in the Swift codebase.

Every GTK-related PR runs the full GUI under Valgrind to detect leaks in the GTK codepath not covered by unit tests.

3. Linux Performance Counter-Intuitive Cases: THP Latency Trap

First in a series "Linux Performance Counter-Intuitive" (

mp.weixin.qq.com/s?__biz=MzAwODA2Njk2OA==&mid=2247487528&idx=1&sn=b3308ebcb243df444d9575aaf52c0dd4

). The article explains how enabling Transparent Huge Pages (THP) can degrade database performance due to khugepaged page merging and memory compaction overhead causing latency spikes.

4. "Linked Lists Are Dead" in Modern Computer Architecture

References a Zhihu answer (

zhihu.com/question/1997631452192466334/answer/1999052955018629727

) arguing that linked lists are obsolete on modern hardware. The core insight: performance optimization must respect hardware locality principles (spatial and temporal locality). The mapping between logical non-linear structures and physical linear memory is a key concept, analogous to LSM trees or B+ trees in database design.

5. Boosting Android Performance with AutoFDO

Google Android Developers Blog post (

android-developers.googleblog.com/2026/03/BoostingAndroid%20PerformanceIntroducingAutoFDO.html

) describes Honor's deep collaboration with Google on AutoFDO (Auto Feedback-Directed Optimization), yielding significant performance gains.

6. TencentOS Kernel Team: Kunlun 2280 Memory Consistency Debugging

Case study from TencentOS kernel team ( zhuanlan.zhihu.com/p/677178481) detailing a memory consistency issue on Kunlun 2280 hardware. Includes an architecture diagram (image preserved).

7. Linux-Kernel Memory Model

References the formal memory model paper (

www.open-std.org/jtc1/sc22/wg21/docs/papers/2023/p0124r8.html

), noted as a clear explanation of the kernel's memory consistency guarantees.

8. PageCache Detailed Analysis (Read/Write/Writeback) Based on Linux 6.12.32

Deep dive into PageCache internals (

mp.weixin.qq.com/s?__biz=MjM5NTc2MDYxMw==&mid=2458608994&idx=1&sn=4b7864d52ec24a4ac6a4a701a53b2c94

) covering read, write, and writeback paths on Linux 6.12.32.

9. Android DMA-BUF Deep Analysis (Two Parts)

Part 1: Zero-copy sharing architecture and core principles (

mp.weixin.qq.com/s?__biz=MzkxMTcxNDAzNQ==&mid=2247486429&idx=1&sn=b9a7f323871c55397b08e34c3b710d8c

)

Part 2: Synchronization mechanisms and Android system integration (

mp.weixin.qq.com/s?__biz=MzkxMTcxNDAzNQ==&mid=2247486430&idx=1&sn=656c9d94c8720a3c46f6557af167571d

)

10. "No Lock" Still Slow? Linux Mutex, Futex & Priority Inheritance Performance Pitfalls

Article (

mp.weixin.qq.com/s?__biz=Mzk0MTE5MDk0OA==&mid=2247486398&idx=1&sn=0e0034fa9b3eebcefbe1e3ce1047552f

) investigates performance degradation even without lock contention, diving into mutex, futex, and priority inheritance implementation details.