[JAUNTY SRU] UBUNTU: SAUCE: added memory barrier to prevent a reference counter leak.

[manoj.iyer at canonical.com]

smb,

The lock used to be there before they used lock bitops to get the desired 
memory ordering. But this memory barrier was removed after 2.6.27 when 
BH_Lock was introduced in the code. smp_mb__before_clear_bit(); was 
removed and smp_mb__after_clear_bit() was kept after clear_bit_unlock() 
call. Not sure is this introduced a regression, which shows up in some 
cases.

The following shaid removed the above mentioned smp_mb__before_clear_bit() 
memory barrier.

SHAID: 51b07fc3c5c830bb49c80fc5eac041e1f66a72e7

Like I mentioned in the patch this seems to be only in the RH errata and 
some custom kernels based on RHEL4, and not mentioned in lkml.

Cheers
--- manjo

On Thu, 23 Jul 2009, Stefan Bader wrote:

> I think this should be save to add without causing regressions. And I think 
> you said some problems were solved by that. Just for general understanding, 
> if this really addresses one of the problems, what is the reason this is only 
> a Redhad erata and not gone upstream?
>