Re: About Figure 6.5 Chapter 6 (v 31.01.2015)

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Sat, May 09, 2015 at 12:01:01PM -0300, Jose Fernando Santos de Carvalho wrote:
> Link: http://ustore1.rnp.br/ustore/faces/publicFile.xhtml?code=35e4c269403136c4a9fdb9227b704409
> 
> On Tue, May 5, 2015 at 11:40 PM, Jose Fernando Santos de Carvalho
> <jfsc@cin.ufpe.br> wrote:
> > Dear all,
> >
> > while writing chapter 6 (Partitioning and Synchronisation Design), Paul has
> > presented five special cases about Left - and Right-Hand Locks(Figure 6.5).
> > However, it is not clear for me the possible problem in the fourth case
> > (link).
> >
> >
> > If possible, could anyone send some feedback about it?
> >
> > Link:
> > http://ustore1.rnp.br/ustore/faces/publicFile.xhtml?code=35e4c269403136c4a9fdb9227b704409--
> >
> > --
> > Jose Fernando

Hello, Jose,

You could indeed claim that the second-to-last row is safe because the two
ends will be operating on different pointers for element 1.  Nevertheless,
the fact remains that both ends need to operate on element 1 simultaneously,
which should give us pause.  Especially given that we are in a concurrent
environment.

For example, suppose that thread A holds the left-hand lock and wants to
remove element 0.  While thread A is comtemplating doing this, some other
thread B might hold the right-hand lock and remove element 2.  We are now
in the state shown in the third row, and thus completely unsafe.

							Thanx, Paul