[LU-10428] LNet events should generated without resource lock held

• Type: Improvement
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.14.0
• Labels: patch
• Environment: any lustre
• Assignee: Alexey Lyashkov
• Reporter: Alexey Lyashkov

Description

LNet execute an events with resource lock held, but event handler can allocate an memory, so it will block other callers to lockup over hash table or parallel event execution.
I will send a patch son.

Issue Links

• is related to: LU-14425 BUG: unable to handle kernel paging request at ffffffffffffffff

Comments

[James A Simmons]

Neil's patch has landed. If it hasn't resolved the problem feel free to reopen.

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/37068/
Subject: LU-10428 lnet: call event handlers without res_lock
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: d05427a7856e8f89cf6ec47f2731e12c6fa22901

[Gerrit Updater]

Neil Brown ([e-mail]) uploaded a new patch: https://review.whamcloud.com/37068
Subject: LU-10428 lnet: call event handlers without res_lock
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 706e000cd325fb47f2e4b77e3f3afd17fc8b62b8

[Alexey Lyashkov]

It looks not a improvement now.

http://testing.linuxhacker.ru:3333/lustre-reports/1276/testresults/runtests-ssk-ldiskfs-SSK-centos7_x86_64-centos7_x86_64/oleg78-client-console.txt

[Cory Spitz]

For some reason, a comment wasn't added here when https://review.whamcloud.com/#/c/30657 was uploaded for this issue.