news
Kernel: Bugs' Age, QEMU 10, and Slop in Linux
-
Bitdefender ☛ Linux Kernel Bugs Can Hide for 20 Years
The Linux developer took a very close look at Linux kernel development, totaling two decades' worth of 125,000 bug-fix commits.
The data revealed that some of the most dangerous bugs won't make themselves known with crashes or obvious failures. Issues in the Linux kernel can linger quietly in cloud infrastructure, enterprise systems, smartphones and billions of connected devices.
-
PC Gamer ☛ Linux researcher and developer says 'there are bugs in your kernel right now that won't be found for years. I know because I analyzed 125,183 of them'
Linux is a beautiful thing. From a completely open-source base, we have probably ten quintillion different distros that treat us like adults, capable of actually owning, controlling, and tinkering with our software. Including, of course, recent boons for gaming such as SteamOS, Bazzite, and Nobara. But that very same free and open foundation might also bring with it some hurdles, such as longstanding bugs.
Researcher and developer Jenny Guanni Qu recently looked at the Linux kernel's bug fixes and found tons of them stick around for a very long time. As someone who feels the allure of Linux distributions and occasionally succumbs to it before quickly scarpering back to Windows when confronted with bugs—usually small and irksome ones, but occasionally catastrophic—this information does not help me. So if you're the same as me: you're welcome.
-
[Old] Linus Heckemann ☛ Make your QEMU 10 times faster with this one weird trick - Linus's blog
What we really want is a structure where we can look up entries by fid more cheaply. We can’t just use an array-based vector, which would consistently give us O(1) lookup, because fids are chosen by the client: we can’t rely on every newly allocated fid just being the smallest unoccupied one, and need to support arbitrary 32-bit integers. I opted for a hash table, as conveniently implemented by glib, which QEMU already depends on. That provides us with O(1) best-case complexity, while keeping the worst case at O(n). The exact real-world performance characteristics are significantly more complex than with the linked list, and there may be marginally more suitable data structures (or hash table implementations) out there, but we’re looking for a big easy win and not micro-optimisation here.
-
PC Gamer ☛ 'There is *zero* point in talking about AI slop. That's just plain stupid': Linus Torvalds weighs in on AI debate in Linux kernel documentation
Where some are sick of AI usage, others are sick of debate about AI usage. And when it comes to mentioning the use of LLMs in kernel documentation, Linux creator Linus Torvalds seems to be very much in the latter camp.
As reported by The Register, Linux kernel engineer for Oracle, Lorenzo Stoakes, recently critiqued Torvalds, arguing that AI tools are not the same as any other tool and need unique documentation and flagging. Stoakes then replied to Dave Hansen, kernel hacker at Intel, saying, "We're noticing a lot more LLM slop than we used to. It is becoming more and more of an issue."
Stoakes argues LLMs have had a negative impact in many areas, "for which you need only take a cursory glance at the world to observe". Furthering this point, Stoakes says, "Thinking LLMs are 'just another tool' is to say effectively that the kernel is immune from this. Which seems to me a silly position."