Christian Schoenebeck wrote on Sat, Jul 16, 2022 at 02:10:05PM +0200:

quoted

OTOH, linux got softiwarp merged in as RDMA_SIW which works perfectly
with my rdma applications, after fixing/working around a couple of bugs
on the server I'm getting hangs that I can't reproduce with debug on
current master so this isn't exactly great, not sure where it goes
wrong :|
At least with debug still enabled I'm not getting any new hang with your
patches, so let's call it ok...?

Well, I would need more info to judge or resolve that, like which patch 
exactly broke RDMA behaviour for you?

I wouldn't have troubles if I knew that, I don't have access to the
hardware I last used 9p/rdma on so it might very well be a softiwarp
compatibility problem, server version, or anything else.

At the very least I'm not getting new errors and the server does receive
everyhing we sent, so as far as these patches are concerned I don't
think we're making anything worse.


I'll get back to you once I hear back from former employer (if they can
have someone run some tests, confirm it works and/or bisect that), I
really spent too much time trying to get the old adapter I got working
already...

All I can say is that there's no error anywhere, I've finally reproduced
it once with debug and I can confirm the server sent the reply and
didn't get any error in ibv_post_send() so the message should have been
sent, but the client just never processed it.
Next step would be to add/enable some logs on the client see if it
actually received something or not and go from there, but I'd like to
see something that works first...

--
Dominique