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

[Linus Torvalds]

On Thu, 13 Jul 2023 at 04:44, Christoph Hellwig <hch at lst.de> wrote:
>
> Please stop putting words in my mouth.  Maybe instead of shouting it
> would help to actually read the text?

I have read the text. You spout unbelievable garbage. Even in this email:

> You said we've never seen devices with reliably IDs, which simply isn't
> true.  And that doesn't mean I'm saying all devices have reliably IDs,
> which somehow you're not trying to make me say.

When I say "devices", I obviously mean in general.

You seem to argue against it.

OF COURSE a _single_ device that has shown to have a UUID is likely to
then repeat that UUID reliably. But that doesn't mean it's true in
general.

The fact is, any driver subsystem that deals with more than one device
manufacturer had better not rely on any serial number for uniqueness.

If some device layer does, and there is a device without serial
numbers, then the bug is squarely on the driver, not the device.
Because the driver author clearly didn't think things through.

THAT is what I mean by "devices do not EVER reliably have uuids".

I just checked. I have on my desk access to three machines with NVMe
drives. Of them

 (a) one didn't have any UUID at all, didn't report namespaces and the
"wwid" exposed in /proc is that

        return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n",
subsys->vendor_id,
                serial_len, subsys->serial, model_len, subsys->model,
                head->ns_id);

     thing.

(b) the other two had a NULL uuid and nguid, but seem to have a eui64,
whatever the hell that is.

So just from my limited test, one in three didn't have any UUID at all.

Deal with it. Stop claiming it's somehow "reliable". Two out of three
out of a random selection is not "reliable".

And it has never been reliable, judging by those quirks. When we have
manufacturers like Samsung, Phison, ADATA, Lexar, Sandisk, Micron, and
Kingston mentioned in the quirk list, we probably have covered most of
the consumer SSD manufacturers.

It's a sad that NVMe - that started out as a "lean mean low-overhead
disk interface definition" results in this kind of mess.

           Linus