From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jason Wang Subject: Re: [PATCH V2 3/6] vDPA: implement IRQ offloading helpers in vDPA core Date: Mon, 20 Jul 2020 17:40:30 +0800 Message-ID: <625c08af-a81f-d834-bb41-538c3dc9acb4@redhat.com> References: <1594898629-18790-1-git-send-email-lingshan.zhu@intel.com> <1594898629-18790-4-git-send-email-lingshan.zhu@intel.com> <45b2cc93-6ae1-47c7-aae6-01afdab1094b@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 8bit Return-path: In-Reply-To: <45b2cc93-6ae1-47c7-aae6-01afdab1094b@intel.com> Content-Language: en-US Sender: kvm-owner@vger.kernel.org To: "Zhu, Lingshan" , mst@redhat.com, alex.williamson@redhat.com, pbonzini@redhat.com, sean.j.christopherson@intel.com, wanpengli@tencent.com Cc: virtualization@lists.linux-foundation.org, netdev@vger.kernel.org, kvm@vger.kernel.org List-Id: virtualization@lists.linuxfoundation.org On 2020/7/20 下午5:07, Zhu, Lingshan wrote: >>> >>> +} >>> + >>> +static void vdpa_unsetup_irq(struct vdpa_device *vdev, int qid) >>> +{ >>> +    struct vdpa_driver *drv = drv_to_vdpa(vdev->dev.driver); >>> + >>> +    if (drv->unsetup_vq_irq) >>> +        drv->unsetup_vq_irq(vdev, qid); >> >> >> Do you need to check the existence of drv before calling unset_vq_irq()? > Yes, we should check this when we take the releasing path into account. >> >> And how can this synchronize with driver releasing and binding? > Will add an vdpa_unsetup_irq() call in vhsot_vdpa_release(). > For binding, I think it is a new dev bound to the the driver, > it should go through the vdpa_setup_irq() routine. or if it is > a device re-bind to vhost_vdpa, I think we have cleaned up > irq_bypass_producer for it as we would call vhdpa_unsetup_irq() > in the release function. I meant can the following things happen? 1) some vDPA device driver probe the hardware and call vdpa_request_irq() in its PCI probe function. 2) vDPA device is probed by vhost-vDPA Then irq bypass can't work since we when vdpa_unsetup_irq() is called, there's no driver bound. Or is there a requirement that vdap_request/free_irq() must be called somewhere (e.g in the set_status bus operations)? If yes, we need document those requirements. Thanks