Marco Elver

#Linux kernel static locking analysis is finally merged! docs.kernel.org/next/dev-tools… This concludes Marco's hard work in the past year, and will enable prevention of locking bugs from entering kernel code base But we still have long road of annotating more code & enabling this on CIs

Clang's -Wthread-safety, named "Context Analysis" for the Linux kernel finally merged after a year's worth of work: git.kernel.org/torvalds/c/092… docs.kernel.org/next/dev-tools… Currently opt-in; go and set 'CONTEXT_ANALYSIS := y' for any kernel code you're developing. Needs Clang 22+.

Compiler-Based Context & Locking Analysis On Deck For Linux 7.0 Paired With Clang 22+ phoronix.com/news/Linux-Com…

Respin of Capability Analysis (aka -Wthread-safety) for the Linux kernel now that Clang gained some new powers: @google.com" target="_blank" rel="nofollow noopener">lkml.kernel.org/r/202509181404… - prior LWN article: lwn.net/Articles/10129…

@Glider Might be due to some link sanitization and/or link protection of LKML. It works if I click on desktop, not on mobile. :-/ Anyway, here is the kasan-dev version: groups.google.com/g/kasan-dev/c/…

@maelver The link doesn't work for me :(

Prototype for type-based partitioning of Linux kernel slab caches: discourse.llvm.org/t/rfc-a-framew… Compiler seems to be doing a good-enough job of inferring allocated types per /proc/slabinfo.

@disruptnhandlr FWIW GWPSan doesn't use DynamoRIO for emulation directly, only uses the instruction decoder. To avoid crazy things like binary rewriting in prod we ended up adding new features to the Linux kernel to do fast emulation from signal handlers with the help of HW breakpoints.

github.com/google/gwpsan/… GWPSan sounds like Valgrind on steroids (enabled with sampling), leveraging DynamoRIO emulation framework.

More about KFENCE: docs.kernel.org/dev-tools/kfen… About the general idea: arxiv.org/pdf/2311.09394

Want to help improve Linux kernel memory safety? KFENCE has been available since Linux 5.12, with several distros enabling KFENCE (chances are your kernel supports it). Though I suspect 1000s of reports are lost. Keeping it simple, I present kfence-cron: gist.github.com/melver/7bf5bdf…

@MayaPosch @dvyukov > [...] IMO testing happens before production For pre-prod testing, sampling-based sanitizers might not be the best fit. For pre-prod testing maybe these can help: - developer.android.com/ndk/guides/hwa… - source.android.com/docs/security/… (Pixel 8 has MTE)

@MayaPosch @dvyukov There's a good chance GWP-ASan is already available for you on Android: source.android.com/docs/security/…

Excited to share bug detection tool we've been working on w/ @maelver: github.com/google/gwpsan It samples unmodified prod binaries with ~1% overhead, can be turned off completely. Detects data races, use-after-return, uninits. Can do more e.g. UAF/OOB in syscalls. Cool tech ...

@MayaPosch @dvyukov That being said, the metadata sections are only needed by specific bug detectors (currently all 3 available detectors need it though). It's possible to develop your own bug detectors in the framework that don't need those ELF sections at all.

@MayaPosch @dvyukov "GWPSan does not modify the executed code" is the main point. Still needs additional compiler flag to generate ELF sections it loads when needed. However, when the tool is completely disabled (or not linked in) it has zero overhead (besides a small binary size increase).

@pramod_bhatotia Congratulations! 🎉

Congrats @kayseesee et al for the #ICSE24 SEIP Distinguished Paper Award! Amazing work!

@_vkaku @debian @nvidia @nickblack @google.com/" target="_blank" rel="nofollow noopener">lore.kernel.org/all/2024011811… May fix it by accident. Should eventually reach stable kernels as well.

@debian @nvidia @nickblack NVIDIA driver specific workaround (re-implements pfn_valid but not really really the RCU way?) github.com/gentoo/gentoo/…

@joel_linux Today I encountered an error when building a module on Debian .... This license linkage business is getting out of hand :-) I just thought of you. What is the preferred Kernel's solution for these modules? The open driver supports Turing, Ampere and Ada ONLY.

@joel_linux @_vkaku @debian @nvidia @nickblack I suspect this will also "fix" it (purely by accident): @google.com/" target="_blank" rel="nofollow noopener">lore.kernel.org/all/2024011811…

@_vkaku @debian @nvidia @nickblack Thanks for bringing it to my attention. This is a tough one. I shall circle with the internal RCU crew and gather opinions. I am not sure if the kernel community cares for proprietary / out of tree drivers though.. but hmm I’m wondering we can liberalize those export license type