Hi Oren, Thanks for your response, this seems in line with my expectations. Still is it possible to add explicit comments in the ib_verbs.h as the current description there is unclear? I think it should be written explicitly that signature MRs should be used in block aligned sizes and offsets and thatwhen used in WR, the offset and size are from network point of view. Thanks, Kobby -----Original Message----- From: linux-rdma-owner@xxxxxxxxxxxxxxx [mailto:linux-rdma-owner@xxxxxxxxxxxxxxx] On Behalf Of Oren Duer Sent: Wednesday, May 16, 2018 2:13 PM To: Carmona, Kobby <Kobby.Carmona@xxxxxxxxxx> Cc: Leon Romanovsky <leon@xxxxxxxxxx>; Jason Gunthorpe <jgg@xxxxxxxxxxxx>; Sagi Grimberg <sagi@xxxxxxxxxxx>; Radzi, Amit <Amit.Radzi@xxxxxxxxxx>; Kalderon, Michal <Michal.Kalderon@xxxxxxxxxx>; linux-rdma@xxxxxxxxxxxxxxx Subject: Re: Clarifications on signature (T10DIF) verbs On Sun, May 13, 2018 at 12:14 PM, Carmona, Kobby <Kobby.Carmona@xxxxxxxxxx> wrote: > 1. In the WR, is the length field in the scatter gather elements the length on host memory and not the length on the network? For example: if T10DIF protection is to be added by the RNIC on the read response and the data on wire does not include DIF. Then the length of the SGE in the read request WR will be the length with the T10 TAGs (I.e. multiple of 520 bytes) while the length of the read request on the wire would be w/o the DIF (multiple of 512 bytes)? The length is always the network (wire) length. > 2. Similar question regarding the addr field in the SGL: Is it the virtual address on host memory? Also network (wire) address. > 3. Signature MR is like a zero-based MR in terms that the first byte of the MR is on VA=0 - right? Correct. > 4. Should the VA in read or write which the peer sends need to consider DIF TAGs insertion or removal? I.e. if there is a MR which includes DIF on host but not on write and block size is 512 bytes. Then if the peer wants to write to the third block, should it send a write to address 1024 or 1040? Again - according to network (wire). > 5. Similar question on length, but I guess than in this case the length field must be the same as the data size on the network, otherwise it will be a violation of the IB spec. Right? Yes, network (wire) length. In all of the above answers, note the limitation which requires va/len to be aligned to the block size (with/without DIF, according to wire). Oren Duer Mellanox Technologies -- 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 ��.n��������+%������w��{.n�����{���fk��ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
