Fwd: Need NVME QUIRK BOGUS for SAMSUNG MZ1WV480HCGL-000MV (Samsung SM-953 Datacenter SSD)

[Linus Torvalds]

On Tue, 11 Jul 2023 at 05:06, Christoph Hellwig <hch at lst.de> wrote:
>
> As far as I can tell Windows completely ignores the IDs.  Which, looking
> back, I'd love to be able to do as well, but they are already used
> by udev for the /dev/disk/by-id/ links.   Those are usually not used
> on desktop systems, as they use the file system labels and UUIDs, but
> that doesn't work for non-file system uses.

The thing is, the nvme code seems to actively do completely stuipid
things in this area.

> And all this has been working really well with the good old enterprise
> SSDs, it's just that the cheap consumer devices keep f*cking it up.

Christoph, deal with reality, not with what you think things should look like.

Anybody who expected unique ID's is frankly completely incompetent.
People should have *known* not to do this.

 "Those Who Do Not Learn History Are Doomed To Repeat It"
          - Santayana

and we have NEVER EVER seen devices with reliably unique IDs. Really.
We've had these uuid's before (ask Greg about USB devices one day, and
that was *recent*).

We've always known that vendors will fill in a fixed value, and
somebody still decided to make this a correctness issue?

Christoph, don't blame vendors. Somebody did indeed f*ck up.  But it was you.

> If we'd take it away now we'd break existing users, which puts us between
> a rock and a hard place.

Well, here's a suggestion: stop making it worse.

For example, we have this completely unacceptable garbage:

        ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
        if (ret) {
                dev_err(ctrl->device,
                        "globally duplicate IDs for nsid %d\n", info->nsid);
                nvme_print_device_info(ctrl);
                return ret;
        }

iow, the code even checks for and *notices* that there are duplicate
IDs, and what does it do? It then errors out.

Then expecting people TO WAIT FOR A NEW KERNEL VERSION when you
noticed something wrong? What an absolute crock.

So stop blaming anybody else.

I think the code should *default* to "unreliable uuid", and then if
you're sure it's actually ok, then you use it. Then some rare
enterprise user with multipathing  - who is going to be very very
careful about which device to use anyway - can use the "approved
list".

Or "Oh, I noticed a non-unique UUID, let me generate one for you based
on physical location".

But this "my disk doesn't work in v6.0 and later because some clown
added a duplicate check that shouldn't be there" is not a good thing
to then try to make excuses for.

            Linus