Hi Venkat, We are seeing an issue with rds_iw_send_ack function in Fedora14 OS.
The issue is as follows: RDS protocol requires to send an acknowledgement back to the sender for the data it has received. RDS can send that acknowledgment by two ways: 1. It can send a piggyback ack while sending data 2. It can send only ACK packet without any data. The issue occurs in case 2 above. For sending an ACK only packet RDS takes another path and different variables and calls rds_iw_attempt_ack function. This function forms RDS header by putting ACK number in it. It puts rest of the fields in RDS header as zero. After that it calculates checksum of that header and puts that checksum also in the header. After this it calls ib_post_send. Now the problem is in calculating the checksum. What happens is checksum gets calculated perfectly fine for the "first time". For the second time it calculates the checksum as same as first time even though ACK number is different for the second time. Thus it results into checksum verification failure on the peer side and connection gets torn down and receiver request gets flushed. We see "WC Error: status = 5 opcode = 0" errors in dmesg on the sender side. I suspect here that it is a dma mapping or flushing issue. To be sure that the issue is with rds_iw_send_ack i used the same sg_list for forming work request in rds_iw_send_ack what is being used by the rds_iw_xmit. After this issue is resolved. So, i think something is wrong with the dma mapping in rds_iw_send_ack function. I only changed the following line in rds_iw_recv_init_ack and it started working. - sge->addr = ic->i_ack_dma; + sge->addr = ic->i_send_hdrs_dma; Interestingly, the issue occurs only on Fedora14(2.6.35.6-45) OS. The issue does not occur with both RHEL6.0(2.6.32-71.el6.x86_64) and RHEL6.1(2.6.32-131.el6.x86_64) OS. The RDS module code is similar for both the OSes. Can you please share your thoughts? Thanks, Vipul Pandya -- 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
