So I had some time to go back and look at the 4.3 failures. They fail when srp_fr_pool_get called from srp_map_idb through srp_map_finish_fr fails with -ENOMEM, and go away with register_always=N. Looks like the additional FR for the indirect buffer isn't accounted anywhere.
To me this sounds like another argument to just allocate one FR per request and don't allow non-contiguous SGLs. Also note that 4.4-rc prefer_fr=y register_always=n !register_always still blows up badly with XFS and ext4 due to data integrity errors. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html