From mboxrd@z Thu Jan 1 00:00:00 1970 From: Thomas Monjalon Subject: Re: i40e queues per VF Date: Thu, 16 Feb 2017 15:55:35 +0100 Message-ID: <6154321.GLCpcWPPOA@xps13> References: <32336171.7h5g7kEJXa@xps13> <9BB6961774997848B5B42BEC655768F810CDE4AF@SHSMSX103.ccr.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7Bit Cc: "Zhang, Helin" , "Mcnamara, John" , dev@dpdk.org To: "Wu, Jingjing" Return-path: Received: from mail-wr0-f179.google.com (mail-wr0-f179.google.com [209.85.128.179]) by dpdk.org (Postfix) with ESMTP id 505AC47CD for ; Thu, 16 Feb 2017 15:55:37 +0100 (CET) Received: by mail-wr0-f179.google.com with SMTP id 89so8833641wrr.3 for ; Thu, 16 Feb 2017 06:55:37 -0800 (PST) In-Reply-To: <9BB6961774997848B5B42BEC655768F810CDE4AF@SHSMSX103.ccr.corp.intel.com> List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" 2017-02-16 13:58, Wu, Jingjing: > From: Thomas Monjalon > > > > Hi, > > > > When reading the documentation, it is not easy to understand the capability of > > i40evf for the number of queues. > > > > First, please could you explain why we need a build-time config option? > > In the doc, there is neither justification nor tuning guidelines: > > > > http://dpdk.org/doc/guides/nics/i40e.html#config-file-options > > " > > CONFIG_RTE_LIBRTE_I40E_QUEUE_NUM_PER_PF (default 64) Number of > > queues reserved for PF. > > CONFIG_RTE_LIBRTE_I40E_QUEUE_NUM_PER_VF (default 4) Number of > > queues reserved for each SR-IOV VF. > > " > > This number is used as initialization time to allocate queue number > for PF/VF for HW's queue pool. Will add more description in i40e.rst. The description "Number of queues reserved for each SR-IOV VF" seems partially wrong. Please explain it is a queue pair. > > I feel these are hard limits and should be some constants in the code, not some > > build configuration options. > > > > The other doc to look at is: > > http://dpdk.org/doc/guides/nics/intel_vf.html#intel-fortville-10-40-gigabit- > > ethernet-controller-vf-infrastructure > > " > > Each VF can have a maximum of 16 queue pairs. > > " > > > > Do we agree that a queue pair is 1 Rx queue / 1 Tx queue? > > Note: the concept of queue pairs in Intel VF should be explained somewhere. > > > Yes. > > Below, a different limitation is given: > > " > > The available queue number(at most 4) per VF depends on the total number of > > pool, which is determined by the max number of VF at PF initialization stage and > > the number of queue specified in config " > > > I think there may be some inconsistent description in doc intel_vf.rst due to > Multiple kinds of NICs. We should correct them. > Thanks for pointing that. > > > So what is the real maximum of queue pairs? 4 or 16? > > The datasheet talks about 16 queues. Is it 8 pairs? > > That's is 16 queue pairs. 16 RX queues and 16 Tx queues. > > > > Is there something to configure the number of queues when creating VF with the > > kernel driver? > > In kernel driver, it seems at most only 4 queues are supported. That's > Why we add build-time config option to make more queues are possible. If we can create 16 queue pairs, why restrict the default configuration to 4? Why is it a build-time config option?