-----Original Message----- > Date: Mon, 24 Jul 2017 10:10:50 +0000 > From: Nipun Gupta <nipun.gu...@nxp.com> > To: Nikhil Rao <nikhil....@intel.com>, "jerin.ja...@caviumnetworks.com" > <jerin.ja...@caviumnetworks.com> > CC: "gage.e...@intel.com" <gage.e...@intel.com>, "dev@dpdk.org" > <dev@dpdk.org>, "tho...@monjalon.net" <tho...@monjalon.net>, > "bruce.richard...@intel.com" <bruce.richard...@intel.com>, > "harry.van.haa...@intel.com" <harry.van.haa...@intel.com>, Hemant Agrawal > <hemant.agra...@nxp.com>, "narender.vang...@intel.com" > <narender.vang...@intel.com>, Abhinandan Gujjar > <abhinandan.guj...@intel.com> > Subject: RE: [PATCH 1/2] eventdev: add event adapter for ethernet Rx queues > > Looking at the rte_event_eth_rx_adapter_queue_add & > event_eth_rx_adapter_service_func > it seems that this indeed will not fit with the cases where ethdev is capable > of enqueing packets > to the eventdev (as was mentioned in Jerin's first RFC). > > In case of hardware based eventdev and queues, these function should also > invoke respective PMD > HW configs. e.g. In queue case - rte_eventdev and rte_ethdev - both PMDs at > hw level shall be configured. > > A typical eventdev hardware will require queues of eth devices will be > configured to directly attach to > eventdev in the hardware. > > Mapping it to NXP eventdev, enabling this functionality requires some > configuration where dev private > information of both the devices (event dev and eth dev) is required at the > same time, > and the final configuration is provided via eth device to H/W. So, this > require inter device communication in DPDK.
Just like dpaa2 bus layer functions used in another subsystems. You can export a function from ethdev dpaa2 to eventdev dpaa2 for ethdev rx queue to eventdev queue connection. > > Jerin, > > I have an impression that Cavium hardware has H/W capability to inject > packets from Ethernet > devices to event devices? If yes, how do you plan to support it? Please check the latest review comments on this patch set http://dpdk.org/dev/patchwork/patch/26534/. Nikhil is changing ops(function pointer) based approach. With that HW support can be added. > > Thanks, > Nipun >
