Re: Scrub: Uncorrectable error due to SSD read error

[Chris Murphy <lists@colorremedies.com>]

On Fri, Jun 4, 2021 at 1:07 AM Martin Steigerwald <martin@lichtvoll.de> wrote:
>
> Mount options look like this.
>
> rw,relatime,lazytime,compress=zstd:3,ssd,discard=async,space_cache,subvolid=257,subvol=/home
>
> I am removing "discard=async". I thought this would be safe with modern
> SSDs.

It should be. So now the question is whether it's some kind of defect
with this particular SSD. Or if it affects make/model. And if it's in
the deny list for mainline. And if upstream is aware of any issues
with this model.

> I always used fstrim manually before, because I was aware of SSDs having
> issue with discard, but when BTRFS introduced queued or batched discard
> I thought I give it a try on a modern SSD.

Completely reasonable.



> > So the drive is not handling queued discard properly is what I
> > suspect. It gets confused. Face plants. Libata then does a hard reset
> > to try to get the drive to come back to reality. And then another
> > discard gets issued and the drive wigs out again.
>
> Thank you. Where did you read this from? I find it challenging to interpret
> the error message the kernel ata layer puts out. They are very detailed and
> I think that is good, but I just do not understand most of it.

Oh man lots of emails on linux-raid@ list over the years. And other
people decipher them. I don't understand most of it either.


-- 
Chris Murphy