[LU-11064] o2iblnd fast reg gaps case is determined incompletely

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.11.0
• Fix Version/s: Lustre 2.12.0
• Type: Bug
• Priority: Major
• Reporter: Amir Shehata (Inactive)
• Assignee: Amir Shehata (Inactive)
• Resolution: Fixed
• Severity: 3

Description

We're allowed to start at a non-aligned page offset in the first fragment and end at a non-aligned page offset in the last fragment.

Currently the first fragment is not considered. So it's possible to start on a non-aligned page boundary and consider that there is a gap in the transmit buffer, which is not correct.

Issue Links

• is duplicated by: LU-11308 LustreError: 2742:0:(events.c:199:cl...
• is duplicated by: LU-11105 Seeing "Using FastReg with no GAPS su...
• is related to: LU-11350 Review of NASA LNet patches

Comments

[Gerrit Updater]

Amir Shehata ([e-mail]) uploaded a new patch: https://review.whamcloud.com/32586
Subject: LU-11064 lnd: determine gaps correctly
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: d2105fede63b081a61fb8c77683d79d7991693f5

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/32586/
Subject: LU-11064 lnd: determine gaps correctly
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: e40ea6fd4494b86d26a9d459f7bf620d816af4e1

[Peter Jones]

Landed for 2.12

[Mahmoud Hanafi]

Does this LU impact 2.10.x ?

[Amir Shehata (Inactive)]

Plane 2.10.x doesn't include any of the GAPs changes. These were landed in 2.11.

However, if you applied these patches on your own branch based on 2.10.x, then you'll need that patch as well.

patches I'm talking about are:

LU-9943 lnd: correct WR fast reg accounting
LU-9810 lnd: use less CQ entries for each connection
LU-10129 lnd: rework map_on_demand behavior
LU-10129 lnd: set device capabilities

[Jay Lan (Inactive)]

We have cherry-picked "LU-9810 lnd: use less CQ entries for each connection". Not the other three.

 

[Mahmoud Hanafi]

FYI, We cherry-picked LU-9810 into 2.11.

[Peter Jones]

Mahmoud

My recommendation is to open a ticket to ask whether it is ok before you include any patch into your distribution - sometimes there are non-obvious pre-requisites.

Peter

[Amir Shehata (Inactive)]

I think 9810 should be ok. There's just been a lot of rework in that area regarding map_on_demand and usage of gaps in 2.11. So I wouldn't recommend pulling that in to a tree based on 2.10.x, unless you pull in the entire set of changes.

Based on LU-11308, I think you guys might be using trees based on 2.10.X as well as 2.11. That's why I was making sure you guys are aware of the dependencies there.

[Jay Lan (Inactive)]

Well, we actually have
LU-9810: lnd: use less CQ entries for each connection, and
LU-9810 lnet: fix build with M-OFED 4.1
in our 2.10.3 and 2.10.5 branches.

LU-9810 was included in Whamcloud's 2.11.0 release. I cherry-picked LU-11064 into our 2.11.0.

Mahmoud has decided to run 2.10.x. So what do you recommend us to do? I do not recall why I cherry-picked LU-9810, do you Mahmoud? Probably trying to fix a problem we encountered?

[Mahmoud Hanafi]

LU-9810 added mofed4.x support. I don't think we cherry-picked that one.

[Amir Shehata (Inactive)]

Would it be possible to point me to the 2.10.X git repo that you guys are using now, so I can take a look at the commits. This way I'm clear on which patches you have.

[Peter Jones]

How about we transfer this discussion to a new ticket so we can identify it as a NASA support issue?

[Jay Lan (Inactive)]

Admir,
https://github.com/jlan/lustre-nas