Re: [PATCH V5 libibverbs 0/7] Completion timestamping

[Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

[-- Attachment #1.1: Type: text/plain, Size: 2021 bytes --]

On 6/8/2016 3:14 PM, Jason Gunthorpe wrote:
> On Mon, Jun 06, 2016 at 10:20:12PM -0400, Doug Ledford wrote:
> 
>> Judging by the way yours and mines conversation went in v4 it seemed you
>> were pretty much in favor of it.  If you have specific objections,
>> please raise them, we can do incremental fixes.
> 
> In favor of the general approach, there were still lots of details to
> be delt with, including what QPs can be joined, the exact set of
> accessors and enablement flags,

These items need to be settled before the next official libibverbs release.

> the compat layer,

This also needs to be settled before the next official release.

> minor coding issues
> and micro optimizations.

These can be done any time and does not have to be done for an initial
feature release.

> Fruther, there hasn't seen sign off from any
> other vendors which we said we wanted for these sorts of things.

An official release has not been made, and none of the other vendors
have really shown much interest in moving forward with this on their own
hardware.  If they have an interest, they still have time before
official release to bring up any shortcomings related to their hardware.

> All of these have been raised on the mailing list, but you killed the
> discussion when you said you didn't like the series,

And a discussion ensued.  In that discussion, it became apparent that
while I didn't like the way it was implemented, it was probably in fact
the best of a bad situation.  I didn't kill anything else, and
discussion continued in the v5 patch series.

> so many issues
> never reached a closure.

Indeed, but that's not to say things are set in stone.  There is still
plenty of time to fix things up if there are specific objections.  But
people now have something easy to build/work against, so if other
vendors want to see if they can implement this feature on their
hardware, they can try it out and if it doesn't work, send incremental
patches to the list.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]