Open Bug 1049114 Opened 10 years ago Updated 2 years ago

Determine if we should merge orderings of removed entries from DeadlockDetector

Categories

(Core :: XPCOM, defect)

defect

Tracking

()

People

(Reporter: erahm, Unassigned)

References

Details

Chris Jones suggested a follow up to updates to the DeadlockDetector in bug 1027921, comment 41.

With the removal of ordering entries we potentially miss out on detecting deadlock situations. I hypothesize this is okay, but we'd like to get feedback as to whether or not anyone has actually run into such situations and thinks we need to add merging logic in. The basic situation is as follows:

We have ordering: a < b < c
We remove b because it's mutex is destroyed, we no longer have the ordering a < c.

This would get fixed up on the next acquisition chain of a followed by c (if that ever happens), but in the interim there is a potential loss of information.

One "fix" would be to merge b's orderings into a when we remove it, but it's not clear if this is really the correct thing to do and it has a potential impact on performance.

If anyone has run into a situation where this has been helpful we'll probably want to add the logic to retain the a < c ordering.
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.