[LU-9180] Upstream ko2iblnd does not work with map_on_demand <256

• Status: Resolved
• Project: Lustre
• Affects Version/s: Upstream
• Type: Bug
• Priority: Critical
• Reporter: Doug Oucharek (Inactive)
• Assignee: WC Triage
• Resolution: Fixed
• Labels: lnet
• Severity: 3

Description

I tried to set a map_on_demand = 32 (default for OPA) with an upstream Lustre build and run it on an MLX QDR set of system.  When I tried a bulk lnet-selftest test, it failed with this log:

[ 1125.547100] LNetError: 22552:0:(o2iblnd_cb.c:1068:kiblnd_init_rdma()) RDMA is too large for peer 192.168.1.24@o2ib (131072), src size: 1048576 dst size: 1048576

When I went back to a much older upstream build, 4.8rc2, the same test worked.  So, some porting from master to upstream seems to have not worked.  The same test works fine on master.

Issue Links

• is related to: LU-9679 Prepare lustre for adoption into the ...

Comments

[James A Simmons]

Strange it should work. The patch for LU-7650 was never revverted upstream. Instead I expected that once a proper solution for LU-7650 was done the fix could be pushed upstream instead.

[Dmitry Eremin (Inactive)]

It looks this is common issue. I reopened LU-5718 because of I got this error with latest master build after patch "LU-5718 o2iblnd: multiple sges for work request" landed.

 

[Bob Glossman (Inactive)]

Dmitry,
Failure introduced in master by LU-5718 fix may have been fixed by LU-9420, just landed. Maybe you should retry on latest master.

[Doug Oucharek (Inactive)]

This ticket is about upstream staging and not master.  Upstream has it's own set of problems.

[James A Simmons]

This was due to left overs from LU-7650 which was incorrect. All the code has been removed upstream and replaced with what landed during 2.11 development cycle.