> From: Nipun Gupta [mailto:nipun.gu...@nxp.com] > Sent: Monday, July 24, 2017 2:07 PM > To: Van Haaren, Harry <harry.van.haa...@intel.com>; Rao, Nikhil > <nikhil....@intel.com>; > jerin.ja...@caviumnetworks.com > Cc: Eads, Gage <gage.e...@intel.com>; dev@dpdk.org; tho...@monjalon.net; > Richardson, Bruce > <bruce.richard...@intel.com>; Hemant Agrawal <hemant.agra...@nxp.com>; > Vangati, Narender > <narender.vang...@intel.com>; Gujjar, Abhinandan S > <abhinandan.guj...@intel.com> > Subject: RE: [PATCH 1/2] eventdev: add event adapter for ethernet Rx queues > > > > > -----Original Message----- > > From: Van Haaren, Harry [mailto:harry.van.haa...@intel.com] > > Sent: Monday, July 24, 2017 16:03 > > To: Nipun Gupta <nipun.gu...@nxp.com>; Rao, Nikhil <nikhil....@intel.com>; > > jerin.ja...@caviumnetworks.com > > Cc: Eads, Gage <gage.e...@intel.com>; dev@dpdk.org; > > tho...@monjalon.net; Richardson, Bruce <bruce.richard...@intel.com>; > > Hemant Agrawal <hemant.agra...@nxp.com>; Vangati, Narender > > <narender.vang...@intel.com>; Gujjar, Abhinandan S > > <abhinandan.guj...@intel.com> > > Subject: RE: [PATCH 1/2] eventdev: add event adapter for ethernet Rx queues > > > > > From: Nipun Gupta [mailto:nipun.gu...@nxp.com] > > > Sent: Monday, July 24, 2017 11:11 AM > > > To: Rao, Nikhil <nikhil....@intel.com>; jerin.ja...@caviumnetworks.com > > > Cc: Eads, Gage <gage.e...@intel.com>; dev@dpdk.org; > > tho...@monjalon.net; Richardson, Bruce > > > <bruce.richard...@intel.com>; Van Haaren, Harry > > <harry.van.haa...@intel.com>; Hemant > > > Agrawal <hemant.agra...@nxp.com>; Vangati, Narender > > <narender.vang...@intel.com>; Gujjar, > > > Abhinandan S <abhinandan.guj...@intel.com> > > > Subject: RE: [PATCH 1/2] eventdev: add event adapter for ethernet Rx > > > queues > > > > > > Hi Nikhil/Edas, > > > > Hi All, > > > > I'll reply on the service-cores topic, and leave the config to those more > > involved > > and familiar with them than I :) > > > > > > -----Original Message----- > > > > From: Nikhil Rao [mailto:nikhil....@intel.com] > > > > Sent: Friday, July 07, 2017 3:23 > > > > To: jerin.ja...@caviumnetworks.com > > > > Cc: gage.e...@intel.com; dev@dpdk.org; tho...@monjalon.net; > > > > bruce.richard...@intel.com; harry.van.haa...@intel.com; Hemant Agrawal > > > > <hemant.agra...@nxp.com>; Nipun Gupta <nipun.gu...@nxp.com>; > > > > narender.vang...@intel.com; Nikhil Rao <nikhil....@intel.com>; > > Abhinandan > > > > Gujjar <abhinandan.guj...@intel.com> > > > > Subject: [PATCH 1/2] eventdev: add event adapter for ethernet Rx queues > > > > > > > > Eventdev-based networking applications require a component to dequeue > > > > packets from NIC Rx queues and inject them into eventdev queues[1]. > > > > While > > > > some platforms (e.g. Cavium Octeontx) do this operation in hardware, > > > > other > > > > platforms use software. > > > > > > > > This patchset introduces an ethernet Rx event adapter that dequeues > > > > packets > > > > from ethernet devices and enqueues them to event devices. It is based on > > > > a previous RFC[2]. > > > > > > > > The adapter is designed to work with the EAL service core[3]. If > > > > an application determines that the adapter is required, it can register > > > > and > > > > launch it on a service core. Alternatively, this adapter can serve as a > > > > template for applications to design customer ethernet Rx event adapters > > > > better suited to their needs. > > > > > > > > The adapter can service multiple ethernet devices and queues. Each > > > > queue is > > > > configured with a servicing weight to control the relative frequency > > > > with > > > > which the adapter polls the queue, and the event fields to use when > > > > constructing packet events. The adapter has two modes for programming an > > > > event's flow ID: use a static per-queue user-specified value or use the > > > > RSS > > > > hash. > > > > > > > > A detailed description of the adapter is contained in the header's > > > > comments. > > > > > > > > [1] http://dpdk.org/ml/archives/dev/2017-May/065341.html > > > > [2] http://dpdk.org/ml/archives/dev/2017-May/065539.html > > > > [3] http://dpdk.org/ml/archives/dev/2017-July/069782.html > > > > > > > > Signed-off-by: Nikhil Rao <nikhil....@intel.com> > > > > Signed-off-by: Gage Eads <gage.e...@intel.com> > > > > Signed-off-by: Abhinandan Gujjar <abhinandan.guj...@intel.com> > > > > --- > > > > > > > > v2: > > > > Thanks Jerin for review - below is a list of changes you > > > > suggested. > > > > > > > > - all public symbols are started with rte_event_. > > > > - Add Doxygen reference with @see. > > > > - Mention setting of ev.event_type. > > > > - Mention adapter to service function mapping. > > > > - Remove rte_eth_rx_event_adapter_dev_add/del(). > > > > - Change rx_queuee_id to int32_t and use -1 to denote all Rx queues. > > > > - Add rte_eth_event_rx_queue_del(). > > > > > > > > Other changes > > > > - Remove adapter's run function (rte_event_eth_rx_adapter_run()) from > > > > the public interface. The adapter internally uses it to create a > > > > service. > > > > - Add a blocked cycle count to stats. Further description is contained > > > > in the header. > > > > - Minor struct renames rte_event_eth_rx_adapter_config -> .._conf > > > > --- > > > > lib/librte_eventdev/rte_event_eth_rx_adapter.h | 301 +++++++++ > > > > lib/librte_eventdev/rte_event_eth_rx_adapter.c | 825 > > > > +++++++++++++++++++++++++ > > > > lib/librte_eventdev/rte_eventdev_version.map | 13 + > > > > lib/Makefile | 2 +- > > > > lib/librte_eventdev/Makefile | 2 + > > > > 5 files changed, 1142 insertions(+), 1 deletion(-) > > > > create mode 100644 lib/librte_eventdev/rte_event_eth_rx_adapter.h > > > > create mode 100644 lib/librte_eventdev/rte_event_eth_rx_adapter.c > > > > > > > > diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h > > > > b/lib/librte_eventdev/rte_event_eth_rx_adapter.h > > > > new file mode 100644 > > > > index 000000000..5ccd0bd24 > > > > --- /dev/null > > > > +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h > > > > @@ -0,0 +1,301 @@ > > > > +/* > > > > + * Copyright(c) 2017 Intel Corporation. All rights reserved. > > > > + * All rights reserved. > > > > + * > > > > + * Redistribution and use in source and binary forms, with or without > > > > + * modification, are permitted provided that the following conditions > > > > + * are met: > > > > + * > > > > + * * Redistributions of source code must retain the above copyright > > > > + * notice, this list of conditions and the following disclaimer. > > > > + * * Redistributions in binary form must reproduce the above > > > > copyright > > > > + * notice, this list of conditions and the following disclaimer > > > > in > > > > + * the documentation and/or other materials provided with the > > > > + * distribution. > > > > + * * Neither the name of Intel Corporation nor the names of its > > > > + * contributors may be used to endorse or promote products > > > > derived > > > > + * from this software without specific prior written permission. > > > > + * > > > > + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND > > > > CONTRIBUTORS > > > > + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT > > NOT > > > > + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND > > > > FITNESS FOR > > > > + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE > > > > COPYRIGHT > > > > + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, > > > > INCIDENTAL, > > > > + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, > > BUT > > > > NOT > > > > + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; > > LOSS > > > > OF USE, > > > > + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED > > AND > > > > ON ANY > > > > + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR > > > > TORT > > > > + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT > > OF > > > > THE USE > > > > + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH > > > > DAMAGE. > > > > + */ > > > > + > > > > +#ifndef _RTE_EVENT_ETH_RX_ADAPTER_ > > > > +#define _RTE_EVENT_ETH_RX_ADAPTER_ > > > > + > > > > +/** > > > > + * @file > > > > + * > > > > + * RTE Event Ethernet Rx Adapter > > > > + * > > > > + * An eventdev-based packet processing application enqueues/dequeues > > mbufs > > > > + * to/from the event device. The ethernet Rx event adapter's role is to > > transfer > > > > + * mbufs from the ethernet receive queues managed by DPDK to an event > > > > device. > > > > + * The application uses the adapter APIs to configure the packet flow > > between > > > > + * the ethernet devices and event devices. The adapter is designed to > > > > work > > with > > > > + * the EAL service cores. The adapter's work can be parallelized by > > > > dividing > > the > > > > + * NIC Rx queues among multiple adapter services that run in parallel. > > > > > > First of all, apologies for commenting late on this. I am going through > > > this > > patchset and > > > have some concerns over this. > > > It is mentioned that "The adapter is designed to work with * the EAL > > > service > > cores" > > > Does this mean that the adapter cannot function without service core? > > > In the case where Ethdev HW is capable of injecting the packet to > > > compatible > > HW eventdev > > > driver, there is no service > > > required. It seems this patchset does not take care of this use-case or > > > maybe I > > am missing > > > something here? > > > > > > The "service cores" is a method of abstracting lcore requirements. In > > short, with > > a SW PMD, a service core is required to perform the event scheduling. With > > the > > HW use case as you describe, no core is required. So in the HW case, no > > service > > cores are allocated in DPDK, and as a result, all your cores are available > > to the > > application. > > > > If I understand correctly, the generic "Event Ethernet RX Adapter" will > > register a > > service which needs to be run (by a service core) in order to perform SW > > polling > > of the NIC, and enqueue to the eventdev. > > > > In order to support the hardware use-case, no service is registered and no > > service lcores required. The HW is allowed to perform the Ethdev -> Eventdev > > enqueue. > > > > Hope that makes sense! -Harry > > With function pointer based approach the adapter calls will be propagated to > eventdev PMD > so > here won't this way the event driver will be responsible for registering the > service? I > don't think the > generic 'Event ethernet Rx adapter' would register a service with this model.
Good point - I should dig into code before replying. Yes, with the function-pointer / PMD model allows the PMD to decide if it requires a service. Thanks for the correction! > Thanks, > Nipun
