From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga01.intel.com (mga01.intel.com [192.55.52.88]) by dpdk.org (Postfix) with ESMTP id B367D1B1AB for ; Sat, 7 Oct 2017 01:28:13 +0200 (CEST) Received: from orsmga003.jf.intel.com ([10.7.209.27]) by fmsmga101.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 06 Oct 2017 16:28:12 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.42,486,1500966000"; d="scan'208";a="1022591197" Received: from irsmsx110.ger.corp.intel.com ([163.33.3.25]) by orsmga003.jf.intel.com with ESMTP; 06 Oct 2017 16:27:59 -0700 Received: from irsmsx104.ger.corp.intel.com ([169.254.5.248]) by irsmsx110.ger.corp.intel.com ([163.33.3.25]) with mapi id 14.03.0319.002; Sat, 7 Oct 2017 00:27:58 +0100 From: "Mokhtar, Amr" To: Thomas Monjalon CC: "dev@dpdk.org" , "fbl@redhat.com" , "aconole@redhat.com" , "bluca@debian.org" Thread-Topic: [dpdk-dev] [RFC] Wireless Base Band Device (bbdev) Thread-Index: AQHTHaidyFtCgYsXxUqHDbe8mN+d6KK/h2qAgBK8FRCAACX8gIADkwEggAAIbQCAAZ254A== Date: Fri, 6 Oct 2017 23:27:58 +0000 Message-ID: <3D3765A8CDB52A4C8B410430AA19CB236EC358A3@IRSMSX104.ger.corp.intel.com> References: <1503668796-65832-1-git-send-email-amr.mokhtar@intel.com> <1887012.k95aipThBJ@xps> <3D3765A8CDB52A4C8B410430AA19CB236EC35318@IRSMSX104.ger.corp.intel.com> <34263409.OrRmme2Ibo@xps> In-Reply-To: <34263409.OrRmme2Ibo@xps> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-titus-metadata-40: eyJDYXRlZ29yeUxhYmVscyI6IiIsIk1ldGFkYXRhIjp7Im5zIjoiaHR0cDpcL1wvd3d3LnRpdHVzLmNvbVwvbnNcL0ludGVsMyIsImlkIjoiOWRjOGUwMGQtYTk3OS00ZjkwLTgzNDgtNTdlM2Y5ODY2NWY1IiwicHJvcHMiOlt7Im4iOiJDVFBDbGFzc2lmaWNhdGlvbiIsInZhbHMiOlt7InZhbHVlIjoiQ1RQX0lDIn1dfV19LCJTdWJqZWN0TGFiZWxzIjpbXSwiVE1DVmVyc2lvbiI6IjE2LjUuOS4zIiwiVHJ1c3RlZExhYmVsSGFzaCI6ImpkelJWZ3l4NHkzaDI2QTF6MzMzeFpKXC82UWNONDR0VG5LWjVWVFdOeEI0PSJ9 x-ctpclassification: CTP_IC dlp-product: dlpe-windows dlp-version: 11.0.0.116 dlp-reaction: no-action x-originating-ip: [163.33.239.181] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [RFC] Wireless Base Band Device (bbdev) 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, 06 Oct 2017 23:28:14 -0000 > -----Original Message----- > From: Thomas Monjalon [mailto:thomas@monjalon.net] > Sent: Thursday 5 October 2017 23:23 > To: Mokhtar, Amr > Cc: dev@dpdk.org; fbl@redhat.com; aconole@redhat.com; bluca@debian.org > Subject: Re: [dpdk-dev] [RFC] Wireless Base Band Device (bbdev) >=20 > 05/10/2017 23:55, Mokhtar, Amr: > > From: Thomas Monjalon [mailto:thomas@monjalon.net] > > > 03/10/2017 16:29, Mokhtar, Amr: > > > > From: Thomas Monjalon [mailto:thomas@monjalon.net] > > > > > 25/08/2017 15:46, Amr Mokhtar: > > > > > > +int > > > > > > +rte_bbdev_configure(uint8_t dev_id, uint16_t num_queues, > > > > > > + const struct rte_bbdev_conf *conf); > > > > > > > > > > I am not convinced by the "configure all" function in ethdev. > > > > > We break the ABI each time we add a new feature to configure. > > > > > And it does not really help to have all configurations in one str= uct. > > > > > Would you mind to split the struct rte_bbdev_conf and split the > > > > > function accordingly? > > > > > > > > There is nothing to split tbh. The only parameter it has is the soc= ket_id. > > > > And in fact, it's optional, can be null. The only config we need is > num_queues. > > > > > > Indeed, there is nothing in this struct. > > > If you need only to allocate queues, you just have to rename this fun= ction. > > > > > > > I don't see in the near future that we may need to add more config = params. > > > > As a side, in the time of the implementation we were trying to > > > > avoid any diversions from the current design ideology of ethdev and > cryptodev. > > > > > > There is no ideology in ethdev, just some mistakes ;) > > > > > > > Can we leave it for consideration with future releases? > > > > > > No it should be addressed from the beginning. > > > > > > When you will need to add something more to configure port-wise, you > > > should add a new function instead of breaking the ABI of the global c= onf > struct. > > > That's why the configure option should be more specialized. > > > > > > Distro people were complaining about ABI breakage last week. > > > This is exactly an example of how to avoid it from the beginning. > > > > > > > Ok, got your point. I was looking at it from an API-only standpoint. > > How about modifying it into? > > int > > rte_bbdev_setup_queues(uint16_t dev_id, uint16_t num_queues, int > > socket_id); >=20 > Yes OK >=20 > [...] > > > > > > +struct __rte_cache_aligned rte_bbdev { > > > > > > + rte_bbdev_enqueue_ops_t enqueue_ops; /**< Enqueue > function */ > > > > > > + rte_bbdev_dequeue_ops_t dequeue_ops; /**< Dequeue > function */ > > > > > > + const struct rte_bbdev_ops *dev_ops; /**< Functions > > > > > > +exported by PMD > > > > > */ > > > > > > + struct rte_bbdev_data *data; /**< Pointer to device data */ > > > > > > + bool attached; /**< If device is currently attached or not > > > > > > +*/ > > > > > > > > > > What "attached" means? > > > > > I'm afraid you are trying to manage hotplug in the wrong layer. > > > > > > > > Hotplug is not supported in the current release. > > > > > > It is not answering the question. > > > What is an "attached" device? > > > > "Attached" means that the PCI device was probed and the bbdev device sl= ot is > allocated. > > For software devices, means that a virtual bbdev device (vdev) is alloc= ated for > bbdev. > > Same way the "attached" approach used in cryptodev. >=20 > Not sure to understand. > If "attached" means "allocated", when is it false? Currently in bbdev, it is set to true and never goes false. As I said the Hotplug feature is not fully supported in the current version= . I can remove that flag for now. But generally, it should be cleared to false when rte_pci_driver->remove fu= nction is called. (Hotplug?) >=20 > [...] > > > > > > +/** Structure specifying a single operation */ struct rte_bbde= v_op { > > > > > > + enum rte_bbdev_op_type type; /**< Type of this operation */ > > > > > > + int status; /**< Status of operation that was performed */ > > > > > > + struct rte_mempool *mempool; /**< Mempool which op > instance > > > > > > +is in > > > > > */ > > > > > > + void *opaque_data; /**< Opaque pointer for user data */ > > > > > > + /** > > > > > > + * Anonymous union of operation-type specific parameters. > > > > > > +When > > > > > allocated > > > > > > + * using rte_bbdev_op_pool_create(), space is allocated for t= he > > > > > > + * parameters at the end of each rte_bbdev_op structure, and > the > > > > > > + * pointers here point to it. > > > > > > + */ > > > > > > + RTE_STD_C11 > > > > > > + union { > > > > > > + void *generic; > > > > > > + struct rte_bbdev_op_turbo_dec *turbo_dec; > > > > > > + struct rte_bbdev_op_turbo_enc *turbo_enc; > > > > > > + }; > > > > > > +}; > > > > > > > > > > I am not sure it is a good idea to fit every operations in the > > > > > same struct and the same functions. > > > > > > > > Due to the fact that our design adopts this idea that a device can > > > > support both the encode and decode operations. > > > > Then, at the time of PMD registration, the enqueue functions is all= ocated. > > > > This enqueue() function is common for both operations. > > > > This fitted operation structure is essential for the driver to > > > > decide on the > > > operation. > > > > > > Sorry I do not understand why you must have a "generic operation". > > > Please, could you try again to explain this design to someone not > > > fully understanding how turbo enc/dec works? > > > > Oh, sorry, I was not paying attention that you're referring to "void *g= eneric" > > It is just a place-holder for any other operation types. Can be removed= if you > like. >=20 > No I was not referring to void *generic. > It is the same question as in the RFC. > I don't understand the benefit of grouping different things in an union. There is no benefit, this is a restriction because there is only one functi= on pointer for enq and another one for deq in the ops structure. Again for the same re= ason of trying to keep things in sync with ethdev and cryptodev. I've always wanted to make it as you proposed, that way it is more performa= nt (no checking for the type of operation.) If this is agreed, I will do it wi= th all my pleasure :) The optimum solution though IMHO would be to make the generic enq/deq funct= ion pointers per queue, instead of being per device; that way every enqueue goes straigh= t to the queue-specific function that matches its operation type. Notice that currently we have turbo_enc/turbo_dec, but in the future we may= have more..