> Going back to 2.6.20 (pre-multicast changes), this area of > the code looks like > it has the same race. Was IPoIB HA testing done on 2.6.20 or > earlier versions > of the code, and if so, were any issues found? (I'm not sure > we've found all of > the problems yet.) No testing was done with 2.6.20 AFAIK. Scott