From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Ananyev, Konstantin" Subject: Re: [PATCH v2] ixgbe: Fix disable interrupt twice Date: Tue, 2 Feb 2016 11:03:09 +0000 Message-ID: <2601191342CEEE43887BDE71AB97725836B02321@irsmsx105.ger.corp.intel.com> References: <1454046700-20843-1-git-send-email-michael.qiu@intel.com> <1454047090-21274-1-git-send-email-michael.qiu@intel.com> <6A0DE07E22DDAD4C9103DF62FEBC0909034256DA@shsmsx102.ccr.corp.intel.com> <533710CFB86FA344BFBF2D6802E6028622F28091@SHSMSX101.ccr.corp.intel.com> <6A0DE07E22DDAD4C9103DF62FEBC0909034266D1@shsmsx102.ccr.corp.intel.com> <533710CFB86FA344BFBF2D6802E6028622F28A4D@SHSMSX101.ccr.corp.intel.com> <533710CFB86FA344BFBF2D6802E6028622F28B7B@SHSMSX101.ccr.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable To: "Qiu, Michael" , "Zhang, Helin" , "Lu, Wenzhuo" , "dev@dpdk.org" Return-path: Received: from mga11.intel.com (mga11.intel.com [192.55.52.93]) by dpdk.org (Postfix) with ESMTP id DE79B93F6 for ; Tue, 2 Feb 2016 12:03:12 +0100 (CET) In-Reply-To: <533710CFB86FA344BFBF2D6802E6028622F28B7B@SHSMSX101.ccr.corp.intel.com> Content-Language: en-US List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" > -----Original Message----- > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Qiu, Michael > Sent: Tuesday, February 02, 2016 2:57 AM > To: Zhang, Helin; Lu, Wenzhuo; dev@dpdk.org > Subject: Re: [dpdk-dev] [PATCH v2] ixgbe: Fix disable interrupt twice >=20 > On 2/2/2016 10:14 AM, Zhang, Helin wrote: > > > >> -----Original Message----- > >> From: Qiu, Michael > >> Sent: Tuesday, February 2, 2016 10:07 AM > >> To: Lu, Wenzhuo; dev@dpdk.org > >> Cc: Zhou, Danny; Liu, Yong; Liang, Cunming; Zhang, Helin > >> Subject: Re: [PATCH v2] ixgbe: Fix disable interrupt twice > >> > >> [+cc helin] > >> > >> On 2/2/2016 9:03 AM, Lu, Wenzhuo wrote: > >>> Hi Michael, > >>> > >>>> -----Original Message----- > >>>> From: Qiu, Michael > >>>> Sent: Monday, February 1, 2016 4:05 PM > >>>> To: Lu, Wenzhuo; dev@dpdk.org > >>>> Cc: Zhou, Danny; Liu, Yong; Liang, Cunming > >>>> Subject: Re: [PATCH v2] ixgbe: Fix disable interrupt twice > >>>> > >>>> On 1/29/2016 4:07 PM, Lu, Wenzhuo wrote: > >>>>> Hi Michael, > >>>>> > >>>>>> -----Original Message----- > >>>>>> From: Qiu, Michael > >>>>>> Sent: Friday, January 29, 2016 1:58 PM > >>>>>> To: dev@dpdk.org > >>>>>> Cc: Zhou, Danny; Liu, Yong; Liang, Cunming; Lu, Wenzhuo; Qiu, > >>>>>> Michael > >>>>>> Subject: [PATCH v2] ixgbe: Fix disable interrupt twice > >>>>>> > >>>>>> Currently, ixgbe vf and pf will disable interrupt twice in stop > >>>>>> stage and uninit stage. It will cause an error: > >>>>>> > >>>>>> testpmd> quit > >>>>>> > >>>>>> Shutting down port 0... > >>>>>> Stopping ports... > >>>>>> Done > >>>>>> Closing ports... > >>>>>> EAL: Error disabling MSI-X interrupts for fd 26 > >>>>>> Done > >>>>>> > >>>>>> Becasue the interrupt already been disabled in stop stage. > >>>>>> Since it is enabled in init stage, better remove from stop stage. > >>>>> I'm afraid it's not a good idea to just remove the intr_disable fro= m > >> dev_stop. > >>>>> I think dev_stop have the chance to be used independently with > >>>>> dev_unint. In > >>>> this scenario, we still need intr_disable, right? > >>>>> Maybe what we need is some check before we disable the intr:) > >>>> Yes, indeed we need some check in disable intr, but it need > >>>> additional fields in "struct rte_intr_handle", and it's much saft t= o > >>>> do so, but as I check i40e/fm10k code, only ixgbe disable it in dev_= stop(). > >>> I found fm10k doesn't enable intr in dev_start. So, I think it's OK. = But i40e > >> enables intr in dev_start. > >>> To my opinion, it's more like i40e misses the intr_disable in dev_sto= p. > >> I don't think i40e miss it, because it not the right please to disable= interrupt. > >> because all interrupts are enabled in init stage. > >> > >> Actually, ixgbe enable the interrupt in init stage, but in dev_start, = it disable it > >> first and re-enable, so it just the same with doing nothing about inte= rrupt. > >> > >> Just think below: > >> > >> 1. start the port.(interrupt already enabled in init stage, disable --= > > >> re-enable) > >> 2. stop the port.(disable interrupt) > >> 3. start port again(Try to disable, but failed, already disabled) > >> > >> Would you think the code has issue? > > [Zhang, Helin] in ixgbe PMD, it can be seen that uninit() calls dev_clo= se(), > > which calls dev_stop(). So I think the disabling can be done only in de= v_stop(). > > All others can make use of dev_stop to disable the interrupt. >=20 > As I said, if it is in dev_stop, it will has issue when dev_start --> > dev_stop --> dev_start, this also could applied in i40e and fm10k. If > you want to put it in dev_stop, better to remove enable interrupts in > init stage, and only put it in dev_start. We can't remove enabling interrupt at init stage and put it only in dev_sta= rt(). That means PF couldn't handle interrupts from VF till dev_start() will be e= xecuted on PF - which could never happen. For same reason we can't disable all interrupts in dev_stop(). See: http://dpdk.org/ml/archives/dev/2015-November/027238.html Konstantin >=20 > Thanks, > Michael > > Regards, > > Helin > > > >> Thanks, > >> Michael > >> > >>> Maybe we can follow fm10k's style. > >>> > >>>> On other hand, if we remove it in dev_stop, any side effect? In ixgb= e > >>>> start, it will always disable it first and then re-enable it, so it'= s safe. > >>> I think you mean we can disable intr anyway even if it has been disab= led. > >> Actually, we couldn't, DPDK call VFIO ioctl to kernel to disable inter= rupts, and > >> if we try disable twice, it will return and error. > >> That's why I mean we need a flag to show the interrupts stats. If it a= lready > >> disabled, we do not need call in to kernel. just return and give a war= ning > >> message. > >> > >> Thanks, > >> Michael > >> > >>> Sounds more like why we don't > >>> need this patch :) > >>> > >>>> Thanks, > >>>> Michael > >