From mboxrd@z Thu Jan  1 00:00:00 1970
From: sagig@dev.mellanox.co.il (Sagi Grimberg)
Date: Sun, 13 Sep 2015 10:53:01 +0300
Subject: [PATCH 3/4] block: Refuse adding appending a gaped integrity page
 to a bio
In-Reply-To: <yq1twr13oeg.fsf@sermon.lab.mkp.net>
References: <1441297703-10753-1-git-send-email-sagig@mellanox.com>
 <1441297703-10753-4-git-send-email-sagig@mellanox.com>
 <yq1a8t2amtj.fsf@sermon.lab.mkp.net> <55EBEF5B.3030608@dev.mellanox.co.il>
 <yq137yo6a78.fsf@sermon.lab.mkp.net> <55F1490C.7050105@dev.mellanox.co.il>
 <yq1twr13oeg.fsf@sermon.lab.mkp.net>
Message-ID: <55F52B5D.5030205@dev.mellanox.co.il>

On 9/11/2015 1:22 AM, Martin K. Petersen wrote:
>>>>>> "Sagi" == Sagi Grimberg <sagig at dev.mellanox.co.il> writes:
>
>>> For some workloads I think bouncing is inevitable by virtue of the PI
>>> payload being teeny tiny by design.
>
> Sagi> I see. I guess we can bounce DIX data, but is it always "right" to
> Sagi> bounce? I don't know, as you say, it is workload dependent.
>
> The right place to make that call, I guess, is in
> bio_integrity_add_page() based on whether we exceed any of the
> constraints set by the LLD.

In case the consumer passed a gappy integrity payload then we can bounce 
but AFAICT, merge flows do not include a call site to
bio_integrity_add_page.

It would be easy enough to bounce gappy IOs and refuse gappy merges,
would that be sufficient?

Sagi.