From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga06.intel.com (mga06.intel.com [134.134.136.31]) by dpdk.org (Postfix) with ESMTP id 16353201 for ; Fri, 27 Jan 2017 11:03:38 +0100 (CET) Received: from orsmga001.jf.intel.com ([10.7.209.18]) by orsmga104.jf.intel.com with ESMTP; 27 Jan 2017 02:03:37 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.33,294,1477983600"; d="scan'208";a="1087900161" Received: from bricha3-mobl3.ger.corp.intel.com ([10.237.221.61]) by orsmga001.jf.intel.com with SMTP; 27 Jan 2017 02:03:35 -0800 Received: by (sSMTP sendmail emulation); Fri, 27 Jan 2017 10:03:34 +0000 Date: Fri, 27 Jan 2017 10:03:34 +0000 From: Bruce Richardson To: "Eads, Gage" Cc: Jerin Jacob , "'dev@dpdk.org'" , "'thomas.monjalon@6wind.com'" , "'hemant.agrawal@nxp.com'" , "Van Haaren, Harry" , "McDaniel, Timothy" Message-ID: <20170127100334.GA61696@bricha3-MOBL3.ger.corp.intel.com> References: <1480996340-29871-1-git-send-email-jerin.jacob@caviumnetworks.com> <1482312326-2589-1-git-send-email-jerin.jacob@caviumnetworks.com> <1482312326-2589-2-git-send-email-jerin.jacob@caviumnetworks.com> <9184057F7FC11744A2107296B6B8EB1E01E5EF2C@FMSMSX108.amr.corp.intel.com> <59AF69C657FD0841A61C55336867B5B035B8EBB9@IRSMSX103.ger.corp.intel.com> <9184057F7FC11744A2107296B6B8EB1E01E5F0AC@FMSMSX108.amr.corp.intel.com> <9184057F7FC11744A2107296B6B8EB1E01E5F3D3@FMSMSX108.amr.corp.intel.com> <20170126093924.GA5276@localhost.localdomain> <9184057F7FC11744A2107296B6B8EB1E01E5F76C@FMSMSX108.amr.corp.intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <9184057F7FC11744A2107296B6B8EB1E01E5F76C@FMSMSX108.amr.corp.intel.com> Organization: Intel Research and =?iso-8859-1?Q?De=ACvel?= =?iso-8859-1?Q?opment?= Ireland Ltd. User-Agent: Mutt/1.7.1 (2016-10-04) Subject: Re: [dpdk-dev] [PATCH v4 1/6] eventdev: introduce event driven programming model X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 27 Jan 2017 10:03:39 -0000 On Thu, Jan 26, 2017 at 08:39:57PM +0000, Eads, Gage wrote: > > > > -----Original Message----- > > From: Jerin Jacob [mailto:jerin.jacob@caviumnetworks.com] > > Sent: Thursday, January 26, 2017 3:39 AM > > To: Eads, Gage > > Cc: Richardson, Bruce ; 'dev@dpdk.org' > > ; 'thomas.monjalon@6wind.com' > > ; 'hemant.agrawal@nxp.com' > > ; Van Haaren, Harry > > ; McDaniel, Timothy > > > > Subject: Re: [dpdk-dev] [PATCH v4 1/6] eventdev: introduce event driven > > programming model > > > > On Wed, Jan 25, 2017 at 10:36:21PM +0000, Eads, Gage wrote: > > > > > > > Subject: [dpdk-dev] [PATCH > > > > v4 > 1/6] eventdev: introduce event driven > > programming model > > > > > > > > > +/** > > + * Enqueue > > > > a burst > of events objects or an event object supplied > > in > > > > > > > *rte_event* > > + * structure on an event device designated > > > > by its > *dev_id* > > through the event + * port specified by > > > > *port_id*. Each > event > > object specifies the event queue on + > > > > * which it will be > enqueued. > > > > > > > + * > > > > > > > + * The *nb_events* parameter is the number of event > > > > objects to > > > enqueue which are + * supplied in the *ev* array > > > > of *rte_event* > > > structure. > > > > > > > + * > > > > > > > + * The rte_event_enqueue_burst() function returns the > > > > number of > + > > * events objects it actually enqueued. A return > > > > value equal to > > > *nb_events* + * means that all event objects have > > been enqueued. > > > > > > > + * > > > > > > > + * @param dev_id > > > > > > > + * The identifier of the device. > > > > > > > + * @param port_id > > > > > > > + * The identifier of the event port. > > > > > > > + * @param ev > > > > > > > + * Points to an array of *nb_events* objects of type *rte_event* > > > > > > structure > > > > > > > + * which contain the event object enqueue operations to be > > > > > > processed. > > > > > > > + * @param nb_events > > > > > > > + * The number of event objects to enqueue, typically number of > > > > > > > + * rte_event_port_enqueue_depth() available for this port. > > > > > > > + * > > > > > > > + * @return > > > > > > > + * The number of event objects actually enqueued on the event > > > > > > device. The > > > > > > > + * return value can be less than the value of the *nb_events* > > > > > > parameter > > > > > > > when > > > > > > > + * the event devices queue is full or if invalid parameters are > > > > > > specified in a > > > > > > > + * *rte_event*. If return value is less than *nb_events*, the > > > > > > remaining events > > > > > > > + * at the end of ev[] are not consumed,and the caller has to take > > > > > > care of > > > > > > > them > > > > > > > + * > > > > > > > + * @see rte_event_port_enqueue_depth() + */ +uint16_t > > > > > > > +rte_event_enqueue_burst(uint8_t dev_id, uint8_t port_id, > > > > > > > + const struct rte_event ev[], uint16_t > > nb_events); > > > > > > > > > > > > There are a number of reasons this operation could fail to > > > > enqueue > all > the events, including: > > > > > > - Backpressure > > > > > > - Invalid port ID > > > > > > - Invalid queue ID > > > > > > - Invalid sched type when a queue is configured for > > > > ATOMIC_ONLY, > > ORDERED_ONLY, or PARALLEL_ONLY > - ... > > > > > > > > > > > > The current API doesn't provide a straightforward way to > > > > determine > the > cause of a failure. This is a particular issue > > > > on event PMDs > that can > backpressure, where the app may want to > > > > treat that case > differently > than the other failure cases. > > > > > > > > > > > > Could we change the return type to int16_t, and define a set > > > > of > error > cases (e.g. -ENOSPC for backpressure, -EINVAL for an > > > > invalid argument)? > > > > > > (With corresponding changes needed in the PMD API) Similarly > > > > we > could > change rte_event_dequeue_burst() to return an > > > > int16_t, with > -EINVAL as > a possible error case. > > > > > > > > > > Use rte_errno instead, I suggest. That's what it's there for. > > > > > > > > > > /Bruce > > > > > > > > That makes sense. In that case, the API comment could be tweaked like so: > > > > > > > > * If the return value is less than *nb_events*, the remaining events at the > > > > * end of ev[] are not consumed and the caller has to take care of them, > > and > > > > * rte_errno is set accordingly. Possible errno values include: > > > > * - EINVAL - The port ID is invalid, an event's queue ID is invalid, or an > > > > * event's sched type doesn't match the capabilities of the > > > > * destination queue. > > > > * - ENOSPC - The event port was backpressured and unable to enqueue > > one or > > > > * more events. > > > > > > However it seems better to use a signed integer for the dequeue burst return > > value, if it is to use rte_errno. Application code could be simplified: > > > > > > (signed return value) > > > ret = rte_event_dequeue_burst(...); > > > if (ret < 0) > > > rte_panic("Dequeued returned errno %d\n", rte_errno); > > > > > > vs. > > > > > > (unsigned return value) > > > ret = rte_event_dequeue_burst(...); > > > if (ret == 0 && rte_errno != 0) > > > rte_panic("Dequeued returned errno %d\n", rte_errno); > > > > > > And with an unsigned return value, all dequeue implementations would have > > to clear rte_errno when no events are dequeued. > > After some internal discussion, I don't think the signed return value is necessary for burst dequeue. Burst enqueue is the more interesting case... > > > > > Gage, > > > > Just to understand, what is the expected application behavior if the > > implementation returns -ENOSPC > > It's application-dependent -- depending on the importance of the event, the application could decide to retry the enqueue some number of times or decide to drop the event. > > > > > Apart for the above SW driver behavior, I think, HW implementation has two > > more different behavior > > a) Implementation make sure that it never returns -ENOSPC by allocating more > > space on the fly or any other scheme > > b) Tail drop > > > > By "tail drop," do you mean the hardware drops the event (and presumably frees any memory it points to)? Or the enqueue is unsuccessful and the application drops the event? > > > Considering different implementation has different behaviors, How about > > enumerating the overflow policy at the port configuration time? and let > > implementation act accordingly to avoid fast-patch change in > > application(effects in all implementation irrespective of the capability) > > > > possible enumerating value at the port configuration time, > > - PANIC or similar scheme to denote it cannot proceed > > - TAIL DROP > > or any expected application behavior you want to add > > I wonder if that's necessary? Hardware behavior a) means the function will always return nb_events. If hardware drops the event(s), I assume enqueue_burst would still return nb_events and the app behaves as if all events were sent. If the enqueue fails (ret < nb_events), app software could check rte_errno and take the action it deems necessary. So all fast-path enqueue code could look like: > > ret = rte_event_enqueue_burst(..., nb_events); > if (ret < nb_events) { > .... > } I would agree with that. I think both enqueue and dequeue should have unsigned return values. Both should set rte_errno on unsuccessful or partially successful operation i.e.: enqueue: sets errno where ret < nb_events dequeue: sets errno where ret == 0 (errno may be set to no-error if queue is just empty) /Bruce