397830 - PKIX: eliminate object locks

Status: NEW

(NSS :: Libraries, enhancement, P3)

Description

[Alexei Volkov]

Object locks should be eliminated as they used only to lock object to only
change object reference count. Instead, PR_AtomicIncrement/PR_AtomicDecrement
should be used.

Comment 1

[Nelson Bolyard (seldom reads bugmail)]

I am surprised to learn that the ONLY use of the object locks was for the
ref count.  I agree that if that is the ONLY use of those locks, and if 
there is no other NEED for an object lock, then those locks are superfluous
and they can be eliminated.  

Is there really no need for any locks on any objects?

Comment 2

[Julien Pierre]

No, I don't think that's true. I believe the object lock is also supposed to protect read/write access to other object members, not just the refcount. At least it was supposed to.

Comment 3

[Nelson Bolyard (seldom reads bugmail)]

In today's meeting, Alexei reported that he has found that the object locks
are being used to protect ONLY the object's own ref count.  He says that in 
cases where object CONTENTS need to be protected, ANOTHER lock is typically
being created and used for that purpose.  

It would be nice to fix all that, but time is short, so IF these additional
locks are being used correctly, then they are merely inefficient, and the
effort to eliminate the extra locks and use only the object locks can be 
put off until 3.13 or later, if necessary.

Comment 4

[Nelson Bolyard (seldom reads bugmail)]

This is just a performance enhancement, an efficiency measure.

Comment 5

[Nelson Bolyard (seldom reads bugmail)]

Unsetting target milestone in unresolved bugs whose targets have passed.

Comment 6

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee is inactive on Bugzilla, so the assignee is being reset.