On Mon, Jun 02, 2014 at 11:47:58AM -0500, Steve Wise wrote: > > > > Steve > > > > I have not checked the code because I am away from my laptop > > But I assume mlx and mthca is using fmr and cxgb4 is using rdma-read-with-invalidate > which > > has implicit fence. > > > > If invalidate is issued before read is completed its a problem. > > > > You're correct. And this bug appears to be in the current upstream code as well. If an > IB_WR_LOCAL_INV wr is used, it must include IB_SEND_FENCE to fence it until the prior read > completes. > > Good catch! I'll post V4 soon. Any chance that can be handled as a separate patch rather than folded in? (Disclaimer: I've been following the discussion only very superficially.) --b. -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
