[LU-11541] DKMS building the lustre client modules with Mellanox OFED is broken on Debian/Ubuntu

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.12.0
• Environment: Ubuntu with Mellanox OFED 4.4
• Severity: 3
• Assignee: Åke Sandgren
• Reporter: Åke Sandgren

Description

Commit baf83c76f11a28a105e92e59b178a7ca2fb66156 has broken dkms builds with Mellanox OFED based builds on Debian/Ubuntu.

 

config/lustre-build-linux.m4 need to retain the LB_USES_DPKG block so that uses_dpkg remains defined since it is being used in lnet/autoconf/lustre-lnet.m4 to detect how to handle the ofed_info output. And config/lustre-build.m4 should then also be reverted regarding the above commit.

 

There is also another problem when using Mellanox OFED 4.4 DKMS built kernel modules.

/usr/src/ofa_kernel/default doesn't exist in the mlnx-ofed-kernel-dkms package, it is only created (as a link) during package installation. Therefore the

O2IBPATHS=$(eval $OFED_INFO | egrep -w 'mlnx-ofed-kernel-dkms|mlnx-ofa_kernel-devel|compat-rdma-devel|kernel-ib-devel|ofa_kernel-devel' | xargs $LSPKG | grep '(/openib|/ofa_kernel/default)$' | head -n1)

in lnet/autoconf/lustre-lnet.m4 fails.

 

The attached patch fixes the problem (assuming the LB_USES_DPKG block has been put back)

Issue Links

• is related to: LU-10997 Ubuntu 18 support

Comments

[Peter Jones]

Landed for 2.12

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/33702/
Subject: LU-11541 build: Use correct kernel version for DKMS MLNX OFED.
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: cad65d162273daf67ce9106897430faadb3ad5f6

[Peter Jones]

From the timing (I'm lurking on the ticket) I think that he was referring to https://review.whamcloud.com/33702 - he reopened the ticket because an extra patch appeared after the ticket had been marked as resolved.

[Andreas Dilger]

Another fix is needed.

James, could you please clarify? Is another patch needed beyond https://review.whamcloud.com/33702
LU-11541 build: Use correct kernel version for DKMS MLNX OFED that still needs to be written, or were you referring to the previous comment about patch 33396?

[James A Simmons]

Another fix is needed.

[Gerrit Updater]

ake sandgren ([e-mail]) uploaded a new patch: https://review.whamcloud.com/33702
Subject: LU-11541 build: Use correct kernel version for DKMS MLNX OFED.
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: b843a1b2759ec76752c63eece720e0d08227fb35

[Peter Jones]

Make a new patch - 33396 has landed already..

[Åke Sandgren]

I have an adjustment to this. One shouldn't actually use the "default" link, but rather point it to the correct kernel version in /usr/src/ofa_kernel. Found out a bit too late...

Should I make a new patch or update 33396?

[Peter Jones]

Landed for 2.12

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/33396/
Subject: LU-11541 build: Adjust OFED check for DKMS enabled MLNX OFED.
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ae5ca4d6de847a186de7ef0e1121d846a900d4c2