Re: [PATCH net 1/4] Revert "net/smc: don't wait for send buffer space when data was already sent"

On 03/11/2021 04:06, Tony Lu wrote:

I agree with you. I am curious about this deadlock scene. If it was
convenient, could you provide a reproducible test case? We are also
setting up a SMC CI/CD system to find the compatible and performance
fallback problems. Maybe we could do something to make it better.

Run an echo server that uses blocking sockets. First call recv() with an 1MB buffer
and when recv() returns then send all received bytes back to the client, no matter
how many bytes where received. Run this in a loop (recv / send).
On the client side also use only blocking sockets and a send / recv loop. Use
an 1MB data buffer and call send() with the whole 1MB of data.
Now run that with both scenarios (send() returns lesser bytes than requested vs. not).