nvme devices are not showed up in lsscsi

[wxiong38]

[root@ltcrain119-lp2 src]# ./lsscsi
[0:0:1:0] disk AIX VDASD 0001 /dev/sda
[0:0:2:0] cd/dvd AIX VOPTA /dev/sr0
[N:0:65:1] disk 3.2TB NVMe Gen4 U.2 SSD__1 -

We expect to see /dev/nvmxxx in lsscsi.

Thanks,
Wend

[doug-gilbert]

This seems to be a great opportunity to use my new clone_pseudo_fs utility found in this account. By default it will clone /sys to /tmp/sys . Then /tmp/sys can be compressed into a tarball (e.g. 'cd /tmp ; tar cvf sys_nvme_iss3.tar.gz sys') and then email the tarball to me. It is fine to do that as a non-root user. Beware of the security implications: don't do it if there is "secret" hardware on your machine. If that is not a concern, it will allow me to use the --sysfsroot= option on lsscsi to examine what is happening on your machine. My email address can be found in the lsscsi manpage.
Thanks in advance.

[jackeichen]

I had the same problem, one type disk works well, while another doesn't:

[root@localhost tmp]# lsscsi -V
release: 0.32 2021/05/05 [svn: r167]

[root@localhost ~]# lsscsi
[0:0:0:0] disk Micron_5 400_MTFDDAK3T8TG 0 /dev/sda
[N:0:129:1] disk SAMSUNG MZWLO1T9HCJR-00A07__1 -
[N:1:65:1] disk SAMSUNG MZWLJ7T6HALA-00007__1 -
[N:2:6:1] disk SAMSUNG MZQL21T9HCJR-00B7C__1 /dev/nvme2n1
[N:3:65:1] disk SAMSUNG MZWLR3T8HCLS-00A07__1 -

Bellow is the difference:
[root@localhost ~]# find / -name nvme2n1
/dev/nvme2n1
/sys/kernel/debug/block/nvme2n1
/sys/class/block/nvme2n1
/sys/devices/pci0000:c2/0000:c2:04.0/0000:c7:00.0/nvme/nvme2/nvme2n1
/sys/block/nvme2n1
[root@localhost ~]# find / -name nvme0n1
/dev/nvme0n1
/sys/kernel/debug/block/nvme0n1
/sys/class/block/nvme0n1
/sys/devices/virtual/nvme-subsystem/nvme-subsys0/nvme0n1
/sys/block/nvme0n1

[doug-gilbert]

Thanks for the report. The difference seems to be nvme2n1 is a direct connect and nvme0n1 is "virtual" which I take to mean some intermediate device such as a PCIe switch (something similar to a SAS expander) is between that SSD and the CPU. I have the same request for you as the previous reporter: could you use clone_pseudo_ps to clone your /sys and then tar/compress the copy and email it to me. You can use the --exclude=<glob_pattern> option one or more times to prune out anything from your /sys that may comprise your security. I would like the paths that include "nvme" and are in sysfs (e.g. /sys/kernel/debug is another filesystem). Thanks in advance.

[doug-gilbert]

Almost 2 months have passed but no data has been provided to help me find this bug. I have been doing more work on clone_pseudo_fs (found under this github account) to prune only needed parts from /sys so lsscsi -y will work on NVMe devices. Only my laptop /sys contains 52057 nodes (regular files, directories, symlinks, etc) and with the following invocation that number goes down to 348 nodes under /tmp/sys :
clone_pseudo_fs -s /sys -d /tmp/sys -E device -E subsystem -p /sys/class/nvme -S
So hopefully the reporters in this thread will see fit to send me via private email these tar-ed up reduced datasets so I can look into this issue. My email address can be found in the normal places, for example the 'REPORTING BUGS' section in the lsscsi manpage. Thanks in advance.

[ze-tomreid]

I have run into the same issue with a pair of Dell Express Flash PM1725b 1.6TB drives.
I have emailed the clone_pseudo_fs output in case that helps.

Something I noticed:
https://lore.kernel.org/all/[email protected]/t/
If I turn off multipathing in nvme core, lsscsi can see the dev notes properly (they don't appear under virtual).

[doug-gilbert]

Thanks for the report. Still awaiting the email with the clone_pseudo_fs output. Under the examples directory in that repository is a prun4lsscsi_nvme.sh script to reduce the size of that output. Perhaps it might prune too much but you could check that with 'lsscsi --sysfsroot=xxx' .

[ze-tomreid]

Hi Doug,
Did you receive the email? I sent it from [email protected], to the email listed on the man page.
The attachment was a 1.9MB .tar.gz from running:
clone_pseudo_fs -s /sys -d /tmp/sys -E device -E subsystem -p /sys/class/nvme -S
Cheers,
Tom

[doug-gilbert]

Still haven't received any email from you. I also sent you an email, did you receive that? Some email servers don't like the "interlog.com" domain (e.g. FreeBSD newsgroups). Perhaps you might try sending tarball to Martin Wilck [email protected] and ask him to relay it to me. Also I would like '-p /sys/devices/virtual/nvme-subsystem' or '-p /sys/class/nvme-subsystem' added to the clone_pseudo_fs invocation that you have shown above. It seems that lsscsi needs to data-mine that sysfs sub-directory looking for anything not under /sys/class/nvme .

[doug-gilbert]

Svn revision 194 has just been mirrored here. It attempts to address this issue but misses some corner cases. A full fix might involve a partial rewrite. Anyway, feedback would be appreciated.

[jackeichen]

@doug-gilbert Sorry miss your message. The new code works in my test.

If you still need some feedback, show it in this issue. I have watching this project, and should not miss it now : )