On Thu, Apr 07, 2016 at 06:16:06PM +0300, Sagi Grimberg wrote: > >>segment (which results in an extra page). The patch should fix the > >>problem, however in this particular sg list I don't see the violation > >>(counted 127 pages)... > > > >I counted 129 pages. > >I will run some tests in our labs too. > > If you did then that indicates the block layer violates > its obligation to respect max_segments/max_hw_sectors? I spent some time looking over the code and remembered an issue I fixed for NVMf (which might explain why we're not seeing it there). max_segments just means contiguous segments, not nessecarily 4k pages. For that we either need a small enough max_sectors_hw or set a max_segment_size of 4k, otherwise we might get 128 block segments, but due to lacking alignment they'll expand to 129 pages. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html