Next: Kernel rewrites itself. Film at 11.

Posted Feb 7, 2026 21:00 UTC (Sat) by smurf (subscriber, #17840)
Parent article: An in-kernel machine-learning library

It's quite obvious that at least some of the code in that patchset has been written with a heap of AI help.

The verbosity of documenting the obvious (yes we know that "open" is used to open a device and the "seek" hook is used to, surprise, seek, thank you very much) while studiously not documenting anything higher-level ("rigorously undocumented" is putting it mildly) is staggering.

to post comments

Next: Kernel rewrites itself. Film at 11.

Posted Feb 9, 2026 19:31 UTC (Mon) by adobriyan (subscriber, #30858) [Link]

For me it was that no self respecting kernel hacker writing a module would document how to make said module built-in.

Also,

* get_random_bytes() is under mutex when it doesn't needs to be,

* dmesg spam on every read/write/ioctl is bad idea,

* copy_from_user/copy_to_user is used instead of put_user/get_user for simple stuff (int).

* missing READ_ONCE annotations, KCSAN should've noticed them (?)

* stuff is both in /proc and /sys (please, leave /proc alone)

* useless debugging stats (who cares how many times device has been opened),

* should probably use devm_() like interfaces or at least investigate,

* /proc, /sys stuff should be removed first before destroying the internals (this one is probably oopsable)

In general, teardown is done from stuff at the userspace boundary towards the center/internal data structures/instances.

* an of course, useless "increment i by 1" comments.