Re: How to detect ram memory going bad?

[Qu Wenruo <wqu@suse.com>]

在 2025/11/20 01:01, BP25 写道:
> Hello I'm writing to this mailing list as suggested by the btrfs docs. I 
> wanted to ask how to detect and mitigate ram memory going bad when using 
> BTRFS? Because the 'Hardware Considerations' the BTRFS manual suggest in 
> this scenario to run memtest; but this is probs more like right after 
> installing new ram.

You should always do a stress test after hardware modification.
And it's always recommended using things like memtest86+ which is a raw 
UEFI payload, with minimal memory reserved, so that almost all RAM can 
be properly tested.

Memtester can be executed as a user space program but it can not test 
the space reserved by the kernel.
Considering how small the space reserved by the kernel, it's still a 
worthy solution, but it will still take a lot of memory (if you want to 
test as many memory as possible).

For the timing:

If installing new ram, always run a memtest.
If overcloking/changing DIMM timming, always run a memtest.


Finally, backup is always recommended, no matter what.


> Is there any BTRFS tool, perhaps to run 
> periodically, that can help me detect bad ram hence mitigate the 
> consequences? There is a webpage called 'Will ZFS and non-ECC RAM kill 
> your data?' where it's suggested that ZFS scrub effectively detects bad 
> ram

I doubt, because bad ram can easily corrupt your metadata/data, and the 
same bad content is written to disk.

Furthermore the checksum will still match (calculated using the bad 
metadata/data), thus it's impossible to detect no matter what.


And bad RAM can happen at any random byte, if it's some core kernel 
structure, you're doomed anyway.

> (when at least two copies of the same file and/or metadata are 
> stored, and I wonder if there are other assumptions here...), but I'm 
> new to btrfs and I wonder if the reasoning can be applied to btrfs as 
> well, and how effective of a mitigation it would actually provide. OS is 
> GNU (Guix) and I think can't use ECC because I suspect my X200 
> motherboards wouldn't support it?
> 
> Please CC: or BCC: me cause I'm not subscribed.
>