From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga02.intel.com (mga02.intel.com [134.134.136.20]) by dpdk.org (Postfix) with ESMTP id C60598E58 for ; Thu, 26 Apr 2018 16:45:23 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga003.jf.intel.com ([10.7.209.27]) by orsmga101.jf.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 26 Apr 2018 07:45:22 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.49,330,1520924400"; d="scan'208";a="46253619" Received: from pgsmsx104.gar.corp.intel.com ([10.221.44.91]) by orsmga003.jf.intel.com with ESMTP; 26 Apr 2018 07:45:21 -0700 Received: from pgsmsx112.gar.corp.intel.com ([169.254.3.227]) by PGSMSX104.gar.corp.intel.com ([169.254.3.13]) with mapi id 14.03.0319.002; Thu, 26 Apr 2018 22:45:20 +0800 From: "Dai, Wei" To: "Zhang, Qi Z" , Thomas Monjalon CC: "Yigit, Ferruh" , "dev@dpdk.org" Thread-Topic: [PATCH v4] ethdev: check Rx/Tx offloads Thread-Index: AQHT3I4+3cBS6ct7pk+7G/zmNRdmtKQRL8YAgAD6BgCAAAVWAIAACRQAgADo2aA= Date: Thu, 26 Apr 2018 14:45:19 +0000 Message-ID: <49759EB36A64CF4892C1AFEC9231E8D66CF6F2F5@PGSMSX112.gar.corp.intel.com> References: <20180328085709.28310-1-wei.dai@intel.com> <4bdecf05-3a7b-1e5d-8a3b-e71e0c37a74d@intel.com> <039ED4275CED7440929022BC67E70611531A8BAE@SHSMSX103.ccr.corp.intel.com> <169440796.JqTBcgK5P9@xps> <039ED4275CED7440929022BC67E70611531A8C0D@SHSMSX103.ccr.corp.intel.com> In-Reply-To: <039ED4275CED7440929022BC67E70611531A8C0D@SHSMSX103.ccr.corp.intel.com> Accept-Language: zh-CN, en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-titus-metadata-40: eyJDYXRlZ29yeUxhYmVscyI6IiIsIk1ldGFkYXRhIjp7Im5zIjoiaHR0cDpcL1wvd3d3LnRpdHVzLmNvbVwvbnNcL0ludGVsMyIsImlkIjoiNGU0YTE2ZjItZWZmZS00MDQyLWE5MjQtOWZjY2E2ZWY2OTc1IiwicHJvcHMiOlt7Im4iOiJDVFBDbGFzc2lmaWNhdGlvbiIsInZhbHMiOlt7InZhbHVlIjoiQ1RQX05UIn1dfV19LCJTdWJqZWN0TGFiZWxzIjpbXSwiVE1DVmVyc2lvbiI6IjE2LjUuOS4zIiwiVHJ1c3RlZExhYmVsSGFzaCI6IlgwZjZEb2tySGxmVFZFV3VpVFB3bTFBQjVqQTNcL0c1cnNcL1VcL2FtWkhNQmM9In0= x-ctpclassification: CTP_NT dlp-product: dlpe-windows dlp-version: 11.0.200.100 dlp-reaction: no-action x-originating-ip: [172.30.20.205] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [PATCH v4] ethdev: check Rx/Tx offloads 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: Thu, 26 Apr 2018 14:45:24 -0000 Thanks to Thomas, Ferruh and Zhang Qi for your feedback. I will rework v5 patch to follow your guidance. > -----Original Message----- > From: Zhang, Qi Z > Sent: Thursday, April 26, 2018 4:51 PM > To: Thomas Monjalon > Cc: Yigit, Ferruh ; Dai, Wei ; > dev@dpdk.org > Subject: RE: [PATCH v4] ethdev: check Rx/Tx offloads >=20 >=20 >=20 > > -----Original Message----- > > From: Thomas Monjalon [mailto:thomas@monjalon.net] > > Sent: Thursday, April 26, 2018 4:19 PM > > To: Zhang, Qi Z > > Cc: Yigit, Ferruh ; Dai, Wei > > ; dev@dpdk.org > > Subject: Re: [PATCH v4] ethdev: check Rx/Tx offloads > > > > 26/04/2018 09:59, Zhang, Qi Z: > > > > > > > -----Original Message----- > > > > From: Yigit, Ferruh > > > > Sent: Thursday, April 26, 2018 1:05 AM > > > > To: Dai, Wei ; thomas@monjalon.net; Zhang, Qi Z > > > > > > > > Cc: dev@dpdk.org > > > > Subject: Re: [PATCH v4] ethdev: check Rx/Tx offloads > > > > > > > > On 4/25/2018 12:50 PM, Wei Dai wrote: > > > > > This patch check if a requested offloading is supported in the > > > > > device capability. > > > > > Any offloading is disabled by default if it is not set in > > > > > rte_eth_dev_configure( ) and rte_eth_[rt]x_queue_setup(). > > > > > A per port offloading can only be enabled in rte_eth_dev_configur= e(). > > > > > If a per port offloading is sent to rte_eth_[rt]x_queue_setup( > > > > > ), return error. > > > > > Only per queue offloading can be sent to > rte_eth_[rt]x_queue_setup( ). > > > > > A per queue offloading is enabled only if it is enabled in > > > > > rte_eth_dev_configure( ) OR if it is enabled in > > > > > rte_eth_[rt]x_queue_setup( ). > > > > > If a per queue offloading is enabled in rte_eth_dev_configure(), > > > > > it can't be disabled in rte_eth_[rt]x_queue_setup( ). > > > > > If a per queue offloading is disabled in rte_eth_dev_configure( > > > > > ), it can be enabled or disabled( ) in rte_eth_[rt]x_queue_setup(= ). > > > > > > > > > > This patch can make such checking in a common way in rte_ethdev > > > > > layer to avoid same checking in underlying PMD. > > > > > > > > Hi Wei, > > > > > > > > For clarification, there is existing API for rc1, and there is a > > > > suggested update in API for rc2. I guess this patch is for > > > > suggested update > > in rc2? > > > > > > > > > Signed-off-by: Wei Dai > > > > > > > > > > --- > > > > > v4: fix a wrong description in git log message. > > > > > > > > > > v3: rework according to dicision of offloading API in community > > > > > > > > > > v2: add offloads checking in rte_eth_dev_configure( ). > > > > > check if a requested offloading is supported. > > > > > --- > > > > > lib/librte_ether/rte_ethdev.c | 76 > > > > > +++++++++++++++++++++++++++++++++++++++++++ > > > > > 1 file changed, 76 insertions(+) > > > > > > > > > > diff --git a/lib/librte_ether/rte_ethdev.c > > > > > b/lib/librte_ether/rte_ethdev.c index f0f53d4..70a7904 100644 > > > > > --- a/lib/librte_ether/rte_ethdev.c > > > > > +++ b/lib/librte_ether/rte_ethdev.c > > > > > @@ -1196,6 +1196,28 @@ rte_eth_dev_configure(uint16_t port_id, > > > > uint16_t nb_rx_q, uint16_t nb_tx_q, > > > > > ETHER_MAX_LEN; > > > > > } > > > > > > > > > > + /* Any requested offload must be within its device capability *= / > > > > > + if ((local_conf.rxmode.offloads & dev_info.rx_offload_capa) != =3D > > > > > + local_conf.rxmode.offloads) { > > > > > + RTE_PMD_DEBUG_TRACE("ethdev port_id=3D%d requested Rx > > > > offloads " > > > > > + "0x%" PRIx64 " doesn't match Rx offloads " > > > > > + "capability 0x%" PRIx64 "\n", > > > > > + port_id, > > > > > + local_conf.rxmode.offloads, > > > > > + dev_info.rx_offload_capa); > > > > > + return -EINVAL; > > > > > + } > > > > > + if ((local_conf.txmode.offloads & dev_info.tx_offload_capa) != =3D > > > > > + local_conf.txmode.offloads) { > > > > > + RTE_PMD_DEBUG_TRACE("ethdev port_id=3D%d requested Tx > > > > offloads " > > > > > + "0x%" PRIx64 " doesn't match Tx offloads " > > > > > + "capability 0x%" PRIx64 "\n", > > > > > + port_id, > > > > > + local_conf.txmode.offloads, > > > > > + dev_info.tx_offload_capa); > > > > > + return -EINVAL; > > > > > + } > > > > +1 having these checks here. > > > > > > > > > + > > > > > /* > > > > > * Setup new number of RX/TX queues and reconfigure device. > > > > > */ > > > > > @@ -1547,6 +1569,33 @@ rte_eth_rx_queue_setup(uint16_t > port_id, > > > > uint16_t rx_queue_id, > > > > > &local_conf.offloads); > > > > > } > > > > > > > > > > + /* > > > > > + * Only per-queue offload can be enabled from application. > > > > > + * If any pure per-port offload is sent to this function, > > > > > +return > > -EINVAL > > > > > + */ > > > > > + if ((local_conf.offloads & dev_info.rx_queue_offload_capa) !=3D > > > > > + local_conf.offloads) { > > > > > + RTE_PMD_DEBUG_TRACE("Ethdev port_id=3D%d > > rx_queue_id=3D%d " > > > > > + "Requested offload 0x%" PRIx64 "doesn't " > > > > > + "match per-queue capability 0x%" PRIx64 > > > > > + " in rte_eth_rx_queue_setup( )\n", > > > > > + port_id, > > > > > + rx_queue_id, > > > > > + local_conf.offloads, > > > > > + dev_info.rx_queue_offload_capa); > > > > > + return -EINVAL; > > > > > + } > > > > > > > > There is a change here. If requested offload is already enabled in > > > > port level, instead of returning error, ignore it. > > > > So this removes the restriction for apps that "only an offload > > > > from queue capabilities can be send for queue_setup() functions". > > > > This is not requirement for application as it has been before, but > > > > this is allowed for app now. > > > > > > > > If app tried to enable a port offload in queue level that is not > > > > already enabled, it should still return error. > > > > > > > > > + > > > > > + /* > > > > > + * If a per-queue offload is enabled in rte_eth_dev_configure( = ), > > > > > + * it is also enabled on all queues and can't be disabled here. > > > > > + * If it is diabled in rte_eth_dev_configure( ), it can be enab= led > > > > > + * or disabled here. > > > > > + * If a per-port offload is enabled in rte_eth_dev_configure( )= , > > > > > + * it is also enabled for all queues here. > > > > > + */ > > > > > + local_conf.offloads |=3D dev->data->dev_conf.rxmode.offloads; > > > > > > > > I didn't get this one, why add rxmode.offloads into queue offloads? > > > > > > > > Based on above change Thomas has an suggestion [1]: > > > > > > > > " > > > > In the case of offload already enabled at port level and repeated > > > > in queue setup, ethdev can avoid passing it to the PMD queue setup > > function. > > > > " > > > > > > > > So almost reverse of what you are doing, strip rxmode.offloads > > > > from local_conf.offloads for PMDs. What do you think? > > > > > > Should we do like below > > > local_conf.offloads |=3D dev->data->dev_conf.rxmode.offloads; > > > local_conf.offloads &=3D dev_info.rx_queue_offload_capa > > > > > > I thinks it's better to only strip port offloads. But keep all queue > > > offload, since this is exact we going to configure the queue and > > > during device start, it can simply iterate on each bit on > > > local_conf.offloads > > to turn on queue offload and don't need to worry about rxmode.offloads. > > > > No > > The offloads which are already enabled at port level does not need to > > be enabled again at queue level. > > But the PMD can decide to not configure the offload at port level for > > real, and configure the port offloads in every queue setups. > > It is an implementation choice, and can be different per-offload. >=20 > OK, got your point, that make sense. >=20 > > So it is simpler to filter such request for queue setups. > > This way, we will be sure that all offloads, requested in queue setup > > PMD function, must be setup for the queue. > > The PMD implementation will need to setup all the requested offloads > > in queue setup, plus the port offloads which were deferred to all queue= s. > > > > Hope it's clear. > > > >