From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dennis Dalessandro Subject: Re: [PATCH 37/37] IB/rdmavt: Add support for new memory registration API Date: Wed, 16 Dec 2015 09:22:25 -0500 Message-ID: <20151216142224.GA28117@phlsvsds.ph.intel.com> References: <20151207204046.8144.18752.stgit@phlsvslse11.ph.intel.com> <20151207204540.8144.94303.stgit@phlsvslse11.ph.intel.com> <5669940D.4040402@dev.mellanox.co.il> <566EEBE8.8020007@dev.mellanox.co.il> <20151214171440.GC23833@phlsvsds.ph.intel.com> <5671653E.40501@dev.mellanox.co.il> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Return-path: Content-Disposition: inline In-Reply-To: <5671653E.40501-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org> Sender: linux-rdma-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org To: Sagi Grimberg Cc: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org, linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Ira Weiny List-Id: linux-rdma@vger.kernel.org On Wed, Dec 16, 2015 at 03:21:02PM +0200, Sagi Grimberg wrote: > >>>>This question is not directly related to this patch, but given that >>>>this is a copy-paste from the qib driver I'll go ahead and take it >>>>anyway. How does qib (and rvt now) do memory key invalidation? I didn't >>>>see any reference to IB_WR_LOCAL_INV anywhere in the qib driver... >>>> >>>>What am I missing? >>> >>>ping? >> >>In short, it doesn't look like qib or hfi1 support this. > >Oh, I'm surprised to learn that. At least I see that >qib is not exposing IB_DEVICE_MEM_MGT_EXTENSIONS. But whats >the point in doing something with a IB_WR_REG_MR at all? >Given that this is not supported anyway, why does this patch >exist? This patch exists to provide parity for what is in qib. Should we not have it? If not, why do we have: commit 38071a461f0a ("IB/qib: Support the new memory registration API") >>That doesn't mean it can't be added to rdmavt as a future enhancement >>though if there is a need. > >Well, given that we're trying to consolidate on post send registration >interface it's kind of a must I'd say. > >>Are you asking because soft-roce will need it? > >I was asking in general, but in specific soft-roce as a consumer will >need to support that yes. I think it makes sense to revisit when soft-roce comes in, since qib/hfi do not need IB_WR_LOCAL_INV. -Denny -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info at http://vger.kernel.org/majordomo-info.html