This problem can be reproduced using netperf and netserver. On server: smc_run netserver On client: smc_run netperf -H 127.0.0.1 -l 1 []# smc_run netperf -H 127.0.0.1 -l 1 MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) ... Recv Send Send Socket ... Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 131072 131072 131072 0.00 1723.79 This is because netperf obtains the smc sndbuf size as 131072 through getsockopt. It calls sendmsg to send 131072 bytes of data at a time, but smc actually returns after only sending 65504 bytes (65536 - sizeof(cdc head)), and then netperf terminates the test. In smc_tx_sendmsg, the processing after the sending ring buf is full is to return directly, so the above phenomenon is caused. I would like to ask if here after the sending ring buf is full, enter smc_tx_wait and wait for the peer to read before continuing to send. When I delete the judgment of send_done in front of smc_tx_wait, it seems to work normally: []# smc_run netperf -H 127.0.0.1 -l 1 MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) ... Recv Send Send Socket ... Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 131072 131072 131072 1.00 11543.80 liqiang (1): Enter smc_tx_wait when the tx length exceeds the available space net/smc/smc_tx.c | 14 ++++++-------- 1 file changed, 6 insertions(+), 8 deletions(-) -- 2.43.0
