Gal Pressman

Does perftest.spec need a change as well?

Hey Ao, Let's take it internally before we open a ticket here, will talk to you offline.

Copy-pasting the same concepts every time a new accelerator is introduced isn't very scalable, I would argue it's time to abstract these and have specific "hardware providers" implementations.

IIRC, the test uses the same completion channel and does: - Send packet - Remote side receives the packet and send another packet - Receive packet from remote And assumes...

It was a real issue, some tests (send_lat for example) do both send and receive, and they might fail without this fix.

@firasj can you please provide the command lines we were running? @sshaulnv what is needed to get this merged? Do you want me to test this on mlx5 hardware?

Hey, can someone please take a look at this patch?

@tmshort thanks for the review! Is anything else needed to get this merged?

I merely suggested it doesn't make sense that the same code is copy-pasted again and again. Since you're already pushing other code that does it, might as well merge this...

https://github.com/linux-rdma/rdma-core/pull/975#issuecomment-824665681