[PATCHv11 00/18] nvme: In-kernel TLS support for TCP

[Hannes Reinecke]

On 9/6/23 10:48, Pawel Baldysiak wrote:
> Hi Hannes,
> 
> I'm testing v15 patchset in extensive way and observing that it is not stable.
> I'm running it on two connected machines - one is used as a host, second as target.
> Once connected I'm running I/Os with dd.
> Eventually I got I/O timeouts from nvme layer on host,
> causing connect/disconnect loops and I/O hung.
> Does not matter if the workload is reads or writes.
> 
Do you have the kernel message log here?
It _might_ be that we're overloading the TLS stack, such that processing 
of individual packets is taking too long and triggering a KATO.

> Second test is connect/disconnect loops.
> On disconnect I've got ocasional delays due to nvme commands timeouts.
> The disconnect happens eventually, but first it need to hit timeout.
> 
Again, do you have message logs?

> It is not happening if TLS is disable.
> 
> Are you performing similar tests on you setup? Are you observing the same?
> 
Yes, but probably not as exhaustive as your tests.
Need to create the testcases for blktests, now that the cleanup from 
Daniel is nearing completion.

Cheers,

Hannes