[systemd-devel] 60-persistent-storage.rules: add NVMe disks and partitions (again)

[Per Bergqvist]

Hej,

The kernel driver in /sys provide little useful information with current releases.

Even if the nvme driver is extended in future kernels as proposed by Keith Busch this will not solve 
the problem with kernels used today.

NVMe drives are being deployed now (in mass) and it will only increase rapidly due to the supreme performance.

I think the most important issue is to define a naming scheme in /dev/disk/by-id to ensure consistent 
naming and how to resolve this naming is secondary.

If the kernel can provide the more useful information in 4.2 it is great but we need a scheme that is possible 
to create with currently deployed kernels including 2.6.* kernels.

The scsi_id on arch 4.0.1-1 returns a ID_SERIAL of:
ID_SERIAL=SNVMe_INTEL_SSDPEDMD40CVFT504600EE400BGN
or 
ID_VENDOR=NVMe
ID_MODEL=INTEL_SSDPEDMD40
ID_SCSI_SERIAL=CVFT504600EE400BGN

on the same kind of system with SuSE 12 (3.12.28) system what you would get (different unit and serial):

ID_SERIAL=365cd2e4080864356494e000000010000
or
ID_VENDOR=NVMe
ID_MODEL=INTEL_SSDPEDMD40
ID_SCSI_SERIAL=CVFT4324006Z400BGN

A possible scheme to achieve with both is (assuming scsi_id being used):

/dev/disk/by-id/nvme-$ENV{ID_MODEL}_$ENV{ID_SCSI_SERIAL}

Distributions could then add necessary patches to achieve the same result as long as we know what 
we want to achieve.
If additional support is added in future kernel releases, systemd/udev can move away from use of scsi_id.

BR
Per

On 14 May 2015, at 19:07 , Keith Busch <keith.busch at intel.com> wrote:

> Kay Sievers <kay <at> vrfy.org> writes:
>> On Thu, May 14, 2015 at 6:12 PM, Lennart Poettering
>> <lennart <at> poettering.net> wrote:
>>> Well, nothing changed really: we'd like to remove scsi_id from
>>> systemd/udev upstream. Please ask sg_utils to pick it up
>>> instead. We'll continue to support the status quo for a while longer,
>>> but it can't stay this way, and we shouldn't add new features to
>>> it while it's in limbo.
>> 
>> Yeah, not sure how much NVMe has to do with SCSI. It might make more
>> sense to export the primary values to identify these devices by the
>> kernel driver in /sys, instead of excapsulating/emulating the SCSI
>> behavior. The kernel does that for other block device types like mmc
>> too.
> 
> I also like native nvme identification over scsi translation. Is the mmc
> way your preferred method for consuming these? If so, can you provide a
> pointer in that driver for what they're doing? I can probably get it in
> the nvme driver for 4.2.
> 
> Thanks,
> Keith
> 
> _______________________________________________
> systemd-devel mailing list
> systemd-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/systemd-devel

Best Regards
Per Bergqvist

Managing Director
Bergqvist Software Technologies 

Mobile: + 352 691 686 600






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/systemd-devel/attachments/20150514/d70442e7/attachment.html>