@lispi314 Corruption that doesn't immediately result in a crash is indeed a more difficult problem to solve. That said, widespread memory corruption is pretty much guaranteed to eventually cause a fatal error by altering a pointer.

Thanks for the discussion and for confirming that what I had in mind seems feasible. I've added several points to my notes on what I would want a new operating system to do.

@davidrevoy

@lispi314 While it would be great if the kernel could protect itself, I was thinking of a much more simple check limited to the memory address ranges that can easily be checked and disused.

If the kernel crashes, it should automatically run a complete memory test on the next boot to determine if faulty hardware is the likely cause.

@davidrevoy

@lispi314 I didn't mean the kernel should try to completely prevent memory corruption by doing software-based ECC on all pages all the time. I meant the kernel should at least try to detect faulty memory as soon as possible, without significantly impacting performance, instead of doing nothing to address a rare but real problem that affects end users.

@davidrevoy

@lispi314 I'm not sure what you mean. It seems to me that the kernel could regularly check memory pages in the background, stop using memory address ranges that return corrupted data, and of course emit a warning meant to be relayed to the user by its desktop environment. Personal devices rarely max out their hardware capabilities, so there are plenty of times when background checks like this can be run without significantly impacting performance.

@davidrevoy

@davidrevoy It seems to me that operating systems could and should detect bad memory, but sadly a lot of software is built without fully taking into account the fact that hardware fails.

Relevant fact: Linus Torvalds is an advocate of error-correcting memory (https://arstechnica.com/gadgets/2021/01/linus-torvalds-blames-intel-for-lack-of-ecc-ram-in-consumer-pcs/) and uses it on his own machine (https://www.youtube.com/watch?v=mfv0V1SxbNA).