#1027921 linux-image-5.10.0-20-amd64: aacraid's expose_physicals=-1 is not functional

Package:
src:linux
Source:
src:linux
Submitter:
Tim Düsterhus
Date:
2023-01-05 13:21:03 UTC
Severity:
normal
#1027921#5
Date:
2023-01-04 14:38:32 UTC
From:
To:
Dear Maintainer,

after upgrading my server from Debian Buster to Debian Bullseye, I
noticed that the underlying devices attached to the Adaptec 6405E raid
controller were no longer exposed via the /dev/sgX devices and thus
reading the SMART values was no longer possible:

    $ lsscsi
    [0:0:0:0]    disk    Adaptec  *redacted*       V1.0  /dev/sda
    [0:3:0:0]    enclosu ADAPTEC  Virtual SGPIO          -

Booting into the Linux 4.19 kernel from Debian Buster correctly exposed
the devices with the new userland, indicating a Kernel issue.

I was able to work around the issue by adding the following
configuration to /etc/modprobe.d/:

    options aacraid expose_physicals=1

After setting this configuration and rebooting the server the devices
are properly visible as they were before:

    $ lsscsi
    [0:0:0:0]    disk    Adaptec  *redacted*       V1.0  /dev/sda
    [0:1:0:0]    disk    ATA      *redacted*       xxxx  /dev/sdb
    [0:1:1:0]    disk    ATA      *redacted*       xxxx  /dev/sdc
    [0:3:0:0]    enclosu ADAPTEC  Virtual SGPIO          -

Unfortunately the setting `1` allows direct write access to the devices,
whereas the previous and to my understanding the default setting of `-1`
only allows read access, making this a safer option.

It appears that aacraid's `expose_physicals=-1` option got broken
somewhere between Linux 4.19 and 5.10.

Best regards
#1027921#10
Date:
2023-01-04 14:57:42 UTC
From:
To:
Dear Maintainer,

Forgot to include in my initial report that I found the workaround using
this (German) thread in the Proxmox Support Forum:

https://forum.proxmox.com/threads/adaptec-aacraid-expose_physicals-kernelbug.63256/

The thread is fairly light on details, but it appears to indicate the
the issue already existed with Linux 5.3.

Best regards
Tim Düsterhus
Developer WoltLab GmbH
#1027921#15
Date:
2023-01-04 15:39:33 UTC
From:
To:
5.10 should be EOL by now
#1027921#20
Date:
2023-01-04 15:39:33 UTC
From:
To:
5.10 should be EOL by now
#1027921#25
Date:
2023-01-04 15:50:41 UTC
From:
To:
Please refrain from comments like that.
It doesn't help at all and is also plain false.
#1027921#30
Date:
2023-01-04 15:55:00 UTC
From:
To:
Hi Renato

According to https://www.kernel.org/category/releases.html, Linux 5.10
is supported until 2026.

Whether a fix for this bug qualifies as an "important bugfix" acceptable
for a LTS kernel I can't tell.

In any case, 5.10 is the stock kernel for Debian Bullseye and thus I
wanted to report the issue, if only to ensure it is tracked, even if
the/a fix doesn't make the cut for Bullseye.

Best regards
Tim Düsterhus
Developer WoltLab GmbH
#1027921#35
Date:
2023-01-04 15:55:00 UTC
From:
To:
Hi Renato

According to https://www.kernel.org/category/releases.html, Linux 5.10
is supported until 2026.

Whether a fix for this bug qualifies as an "important bugfix" acceptable
for a LTS kernel I can't tell.

In any case, 5.10 is the stock kernel for Debian Bullseye and thus I
wanted to report the issue, if only to ensure it is tracked, even if
the/a fix doesn't make the cut for Bullseye.

Best regards
Tim Düsterhus
Developer WoltLab GmbH
#1027921#40
Date:
2023-01-04 16:25:44 UTC
From:
To:
relevant, although I do not fully understand it.
One thing that stood out for me is the check `expose_physicals > 0` instead of
`expose_physicals != 0`. But that commit exists since 2.6.35 ...

This very much looks like an upstream kernel issue and therefor it's better to
bring it to the attention of those maintainers.

Those are:
$ scripts/get_maintainer.pl drivers/scsi/aacraid/aachba.c
Adaptec OEM Raid Solutions <aacraid@microsemi.com> (supporter:AACRAID SCSI
RAID DRIVER)
"James E.J. Bottomley" <jejb@linux.ibm.com> (maintainer:SCSI SUBSYSTEM)
"Martin K. Petersen" <martin.petersen@oracle.com> (maintainer:SCSI SUBSYSTEM)
linux-scsi@vger.kernel.org (open list:AACRAID SCSI RAID DRIVER)
linux-kernel@vger.kernel.org (open list)

When you do forward your question, please inform this bug report of the URL.

HTH
#1027921#47
Date:
2023-01-05 13:13:34 UTC
From:
To:
Try revert https://github.com/torvalds/linux/commit/948e922fc44611ee2de0c89583ca958cb5307d36
("scsi: core: map PQ=1, PDT=other values to SCSI_SCAN_TARGET_PRESENT")