[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-rdma
Subject:    Re: Data corruption in kernel 5.1+ with iSER attached ramdisk
From:       Sagi Grimberg <sagi () grimberg ! me>
Date:       2020-03-31 1:07:13
Message-ID: 82bbbdf2-cf61-d523-29e0-d756b7f208f4 () grimberg ! me
[Download RAW message or body]


>> Can you try attached patch and see if it solves your issue?
>> WARNING: very lightly tested...
> 
> I have run our tests against this patch and it is working well for our
> "basic" testing as well. The test case that previously failed, now
> passes with this patch. So that's encouraging! Thanks for the quick
> response and quick patch.

Good to know..

> One question we had is regarding the hard coded header length: What
> happens if the initiator sends an extended CDB, like a WRITE32? Are
> there any concerns with an additional header segment (AHS)?

You are absolutely correct! t10-dif is broken with this patch as
32 byte cdb would break into two buffers which is not expected
by the target core...

I take back this patch, I guess we should keep contiguous allocation but
just make the recv wr such that the data is aligned for 16 bytes cdbs,
and for 32-byte cdbs we never support immediate data anyways...
[prev in list] [next in list] [prev in thread] [next in thread]