Re: 2.6.17-rc5-mm3: bad unlock ordering (reiser4?)

[Ingo Molnar <mingo@elte.hu>]

* Valdis.Kletnieks@vt.edu <Valdis.Kletnieks@vt.edu> wrote:

> On Sun, 04 Jun 2006 13:33:26 PDT, Andrew Morton said:
> 
> > Why does the locking validator complain about unlocking ordering?
> 
> Presumably, if the lock nesting *should* be "take A, take B, release 
> B, release A", if it sees "Take A, Take B, release A" it means there's 
> potentially a missing 'release B' that got forgotten (most likely an 
> error case that does a 'return;' instead of a 'goto 
> end_of_function_cleanup' like we usually code.
> 
> Having said that, I'm not sure it qualifies as a "BUG".  Certainly 
> would qualify for a "SMELLS_FISHY" though.  But we don't have one of 
> those handy, so maybe BUG is as good as it gets (given that the person 
> who built the kernel *asked* to be nagged about locking funkyness)....

yes. This warning caught a couple of bugs, and documented a couple of 
'fishy' places. Sometimes it's code that is totally correct. I think 
it's worth the extra iteration, there arent that many non-nested 
unlocking places.

straight nested unlocking is also best for performance and scalability: 
the outmost lock should be released last, because that's what the 
waiters are most likely to be blocking/spinning upon.

nevertheless i'll turn that warning into a less scary message.

	Ingo