[LU-8571] Lustre 2.8 support for RHEL7.3

• Type: Task
• Resolution: Fixed
• Priority: Minor
• Affects Version/s: Lustre 2.8.0
• Labels: RHEL7.3 llnl
• Assignee: Peter Jones
• Reporter: Christopher Morrone (Inactive)

Description

The RHEL7.3 beta is out. The new kernel (3.10.0-493) introduces a lot of changes in the IB interfaces. We need 2.8 to build against RHEL7.3.

First I needed to backport the following patches (in reverse order):

• 11e4f77 2016-07-11 LU-8056 socklnd: NETIF_F_ALL_CSUM renamed to NETIF_F_CSUM_MASK
• 31d6445 2016-07-05 LU-8056 o2iblnd: ib_query_device removed in 4.5
• ac25785 2016-05-27 LU-6215 lnet: split struct ib_send_wr
• 2366ba2 2016-05-16 LU-6215 o2iblnd: port to new fast reg API introduced in 4.4
• 3e95945 2016-04-28 LU-6215 o2iblnd: rdma_create_id() takes extra parameter
• 868bacf 2016-04-28 LU-6215 o2iblnd: ib_alloc_fast_reg_mr() removed in 4.3

Then I needed to apply two new patches from these tickets:

• LU-8534 RHEL7.3 ldiskfs patch set
• LU-8540 ib_map_mr_sg() with five arguments

Issue Links

• is related to: LU-8675 Lustre 2.8.0 + Centos 6.7 + MLNX_OFED-3.3-1.0.4.0
• is related to: LU-8534 RHEL7.3 ldiskfs patch set
• is related to: LU-8540 ib_map_mr_sg() with five arguments

Comments

[Christopher Morrone (Inactive)]

Let's keep the MLNX discussion in LU-8675.

[Atul Yadav]

Hi Team,

Can we have update on this requirement.

Thank You
Atul Yadav

[Atul Yadav]

HI Team,

If more information required on this issue, please let us know.

Thank You
Atul Yadav

[Atul Yadav]

Hi Team,

For your reference, the issue description is given below:-

Note: checking out '2.8.0'.
..
...
HEAD is now at ea79df5... New tag 2.8.0-RC5
sh autogen.sh
./configure --disable-server --with-o2ib=/usr/src/mlnx-ofa_kernel-3.3
..
..
checking if Linux kernel has tunable backoff TCP support... no
checking whether to use Compat RDMA... yes
checking whether to use any OFED backport headers... no
checking whether to enable OpenIB gen2 support... no
configure: error: can't compile with OpenIB gen2 headers under /usr/src/ofa_kernel-3.3
hca_self_test.ofed
---- Performing Adapter Device Self Test ----
Number of CAs Detected ................. 1
PCI Device Check ....................... PASS
Kernel Arch ............................ x86_64
Host Driver Version .................... MLNX_OFED_LINUX-3.3-1.0.4.0 (OFED-3.3-1.0.4): 2.6.32-573.el6.x86_64
Host Driver RPM Check .................. PASS
Firmware on CA #0 VPI .................. v2.36.5000
Firmware Check on CA #0 (VPI) .......... PASS
Host Driver Initialization ............. PASS
Number of CA Ports Active .............. 1
Port State of Port #1 on CA #0 (VPI)..... UP 4X FDR (InfiniBand)
Error Counter Check on CA #0 (VPI)...... PASS
Kernel Syslog Check .................... PASS
Node GUID on CA #0 (VPI) ............... 7c:fe:90:03:00:17:1e:00
------------------ DONE ---------------------
[root@inblrmaster02 lustre-release]# uname -r
2.6.32-573.el6.x86_64
[root@inblrmaster02 lustre-release]#

Thank You
Atul Yadav

[Atul Yadav]

Thanks Team,

For initiating the activity for linking Lustre 2.8.0 + Centos 6.7 + MLNX_OFED-3.3-1.0.4.0.

Thanks
Atul Yadav

[James A Simmons]

Finished the port, now just to land it.

[James A Simmons]

Chris I started the work of the port to 2.8. I needed it for our Mellanox 3.3 systems as well. So far I have:

http://review.whamcloud.com/#/c/22944
http://review.whamcloud.com/#/c/22945
http://review.whamcloud.com/#/c/22946
http://review.whamcloud.com/#/c/22947

I have more but I like to push the above first.

[Christopher Morrone (Inactive)]

Yes! RH backported an awful lot in this update. My suspicion is that our pressures to get Intel OPA support into the RHEL kernel may have triggered this big packport of rdma-related patches in the RHEL kernel. But I could be wrong.

It took me longer than I expected to get the right patches in place for Lustre 2.8 to work, but it would have been quite a bit worse if I didn't have your new kernel work already in place on Lustre's master branch. So thanks for that work, James!

[James A Simmons]

Yipes, that is a lot of newer kernel stuff being back ported. I didn't expect the work for newer kernels to become so important so quickly.

[Christopher Morrone (Inactive)]

The requirement for this is less specific. Within a few weeks. I already have it building locally. I am more concerned that this brings us up to 44 patches beyond b2_8_fe, and that is more patches than I would like to see us carrying locally.