Re: Another multithreading bug (I think)

[Wim Oudshoorn]

Pete French <pete@twisted.org.uk> writes:

>> Thread B on Processor B sees isCreated = 1, but still has a stale
>>             value (nil) for 'theInstance' in its L1/L2 memory cache
>>             and returns the nil.
>
> I can't imagine that this would ever happen - if processors in an SMP
> system had independent caches such that they saw different cached values
> for the same location then it wouldnt work at all. The cache consistency
> mechanism is there to ensure that this never happens, so it shouldnt be an
> issue.
>
Well, I think that consistency is only guarenteed when locking is used
appropriately.  
For example in the paper:

Impact of Java Memory Model on Out-of-Order Multiprocessors

by  Tulika Mitra, Abhik Roychoudhury, Qinghua Shen
from School of Computing National University of Singapore

The examine the old an new proposed Java Memory Model.
They claim that:  [BEGIN QUOTE]

Thread 1                           Thread 2

Write a,1                          read volatile v
Write a,2                          read a
Write volatile v, 1

assuming v and a are initialized to 0, it is possible to read
v=1 and a=1 in the second thread

[END QUOTE]
They are discussing the current Java Memory Model.

But I am trying to find more documentation on multi processor architectures
to really understand what is happening.

Wim Oudshoorn.