Re: Question: Linux5.15 - enabling CONFIG_MEMTEST

[daerich <daerich@users.noreply.github.com>]

[-- Attachment #1: Type: text/plain, Size: 1352 bytes --]

New comment by daerich on void-packages repository

https://github.com/void-linux/void-packages/issues/34399#issuecomment-991978539

Comment:
Well that rare usage might be because of its obscurity. 
Otherwise I don't see why anyone would prefer userspace applications to test  (almost) their entire memory.
As mentioned on the memtester website, it relies on the `malloc(3)`-runtime of the underlying libc, which in turn is managed by the Kernel, and/or the `mmap(2)`system call disallowing testing of close to every bit on silicon, since kernel and userspace naturally require at least some memory. That is especially mentioned in the project's _README_.
For memtest86+, on its webpage it seems  ___stable___ support for modern CPU seems immature disqualifying it for production use;
preliminary support for HASWELL CPUs  - Intel's 4th Gen, we are currently on the 12th, Alderlake.
In fact, searching for any ".*lake" in the changelog on the website yields not results.
A perk of CONFIG_MEMTEST would be the feature that the kernel would reserve bad memory as "badmem" ensuring proper functionality of the OS.
Debian seemed to have the discussion a while back. After no real objection they enabled it in 2.6x kernels. But it would be nice if a kernel package maintainer would comment on that, since they, obviously, would have the final say...