[LU-17999] LNet: relax use of msg_rx_ready_delay

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Serguei Smirnov
• Reporter: Serguei Smirnov

Description

When processing a message, lnet_post_routed_recv_locked currently asserts that msg_rx_ready_delay be set when detecting negative peer or router credits. 

It appears to be possible though for msg_rx_ready_delay to actually be not set at that point because it only gets set if the credit counts were non-positive some time before, and even though cpt lock was being (mostly?) held during this time, lpni lock, for example, was not. This means that the credit counts could have been decremented since.

Comments

[Peter Jones]

Merged for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/55620/
Subject: LU-17999 lnet: prevent race in access to peer rtrcredits count
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: f3ee8ec7c820979f278c3bcbf36789d9b9b24ae6

[Gerrit Updater]

"Serguei Smirnov <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/55620
Subject: LU-17999 lnet: relax use of msg_rx_ready_delay when routing
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: b66c72e565a1fd7c205aa8b98e45d36facead1c3