On Mon, Aug 30, 2010 at 05:11:01PM -0700, Ralph Campbell wrote: [ add a reference to neighbours inside ipoib ]
> The problem with this solution is that it creates > a reference counting "loop" so that the reference > count never goes to zero. > struct neighbour in the kernel points to struct ipoib_neigh > which points back to struct neighbor. If the "back pointer" > holds a reference, then something besides ipoib_neigh_free() > has to do the neigh_release(neighbour). > > I think the real fix is the patch I sent to linux-rdma: > https://patchwork.kernel.org/patch/120013/ patchwork is keeping this one secret for now: OperationalError at /patch/120013/ (1040, 'Too many connections') But I'll check back later today. We'd hit this corruption at least once a day under load, and we couldn't reboot a rack of machines without half of them exploding as others went down (without CONFIG_PAGE_DEBUG_ALLOC) So I'm a bit relieved that someone else has seen it too ;) -chris -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
