From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id C16B6C43219 for ; Thu, 10 Mar 2022 08:29:10 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S240344AbiCJIaJ (ORCPT ); Thu, 10 Mar 2022 03:30:09 -0500 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:42998 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232250AbiCJIaI (ORCPT ); Thu, 10 Mar 2022 03:30:08 -0500 Received: from out30-130.freemail.mail.aliyun.com (out30-130.freemail.mail.aliyun.com [115.124.30.130]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id D8811135701; Thu, 10 Mar 2022 00:29:05 -0800 (PST) X-Alimail-AntiSpam: AC=PASS;BC=-1|-1;BR=01201311R481e4;CH=green;DM=||false|;DS=||;FP=0|-1|-1|-1|0|-1|-1|-1;HT=e01e04357;MF=xuanzhuo@linux.alibaba.com;NM=1;PH=DS;RN=33;SR=0;TI=SMTPD_---0V6neRLx_1646900938; Received: from localhost(mailfrom:xuanzhuo@linux.alibaba.com fp:SMTPD_---0V6neRLx_1646900938) by smtp.aliyun-inc.com(127.0.0.1); Thu, 10 Mar 2022 16:28:59 +0800 Message-ID: <1646900411.6481435-2-xuanzhuo@linux.alibaba.com> Subject: Re: [PATCH v7 17/26] virtio_pci: queue_reset: support VIRTIO_F_RING_RESET Date: Thu, 10 Mar 2022 16:20:11 +0800 From: Xuan Zhuo To: Jason Wang Cc: Jeff Dike , Richard Weinberger , Anton Ivanov , "Michael S. Tsirkin" , "David S. Miller" , Jakub Kicinski , Hans de Goede , Mark Gross , Vadim Pasternak , Bjorn Andersson , Mathieu Poirier , Cornelia Huck , Halil Pasic , Heiko Carstens , Vasily Gorbik , Christian Borntraeger , Alexander Gordeev , Sven Schnelle , Alexei Starovoitov , Daniel Borkmann , Jesper Dangaard Brouer , John Fastabend , Johannes Berg , Vincent Whitchurch , linux-um@lists.infradead.org, platform-driver-x86@vger.kernel.org, linux-remoteproc@vger.kernel.org, linux-s390@vger.kernel.org, kvm@vger.kernel.org, bpf@vger.kernel.org, virtualization@lists.linux-foundation.org, netdev@vger.kernel.org References: <20220308123518.33800-1-xuanzhuo@linux.alibaba.com> <20220308123518.33800-18-xuanzhuo@linux.alibaba.com> <8b9d337d-71c2-07b4-8e65-6f83cf09bf7a@redhat.com> In-Reply-To: <8b9d337d-71c2-07b4-8e65-6f83cf09bf7a@redhat.com> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org On Wed, 9 Mar 2022 16:54:10 +0800, Jason Wang wrote: > > =E5=9C=A8 2022/3/8 =E4=B8=8B=E5=8D=888:35, Xuan Zhuo =E5=86=99=E9=81=93: > > This patch implements virtio pci support for QUEUE RESET. > > > > Performing reset on a queue is divided into these steps: > > > > 1. virtio_reset_vq() - notify the device to reset the qu= eue > > 2. virtqueue_detach_unused_buf() - recycle the buffer submitted > > 3. virtqueue_reset_vring() - reset the vring (may re-alloc) > > 4. virtio_enable_resetq() - mmap vring to device, and enable = the queue > > > > This patch implements virtio_reset_vq(), virtio_enable_resetq() in the > > pci scenario. > > > > Signed-off-by: Xuan Zhuo > > --- > > drivers/virtio/virtio_pci_common.c | 8 +-- > > drivers/virtio/virtio_pci_modern.c | 83 ++++++++++++++++++++++++++++++ > > 2 files changed, 88 insertions(+), 3 deletions(-) > > > > diff --git a/drivers/virtio/virtio_pci_common.c b/drivers/virtio/virtio= _pci_common.c > > index fdbde1db5ec5..863d3a8a0956 100644 > > --- a/drivers/virtio/virtio_pci_common.c > > +++ b/drivers/virtio/virtio_pci_common.c > > @@ -248,9 +248,11 @@ static void vp_del_vq(struct virtqueue *vq) > > struct virtio_pci_vq_info *info =3D vp_dev->vqs[vq->index]; > > unsigned long flags; > > > > - spin_lock_irqsave(&vp_dev->lock, flags); > > - list_del(&info->node); > > - spin_unlock_irqrestore(&vp_dev->lock, flags); > > + if (!vq->reset) { > > + spin_lock_irqsave(&vp_dev->lock, flags); > > + list_del(&info->node); > > + spin_unlock_irqrestore(&vp_dev->lock, flags); > > + } > > > > vp_dev->del_vq(info); > > kfree(info); > > diff --git a/drivers/virtio/virtio_pci_modern.c b/drivers/virtio/virtio= _pci_modern.c > > index 49a4493732cf..3c67d3607802 100644 > > --- a/drivers/virtio/virtio_pci_modern.c > > +++ b/drivers/virtio/virtio_pci_modern.c > > @@ -34,6 +34,9 @@ static void vp_transport_features(struct virtio_devic= e *vdev, u64 features) > > if ((features & BIT_ULL(VIRTIO_F_SR_IOV)) && > > pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV)) > > __virtio_set_bit(vdev, VIRTIO_F_SR_IOV); > > + > > + if (features & BIT_ULL(VIRTIO_F_RING_RESET)) > > + __virtio_set_bit(vdev, VIRTIO_F_RING_RESET); > > } > > > > /* virtio config->finalize_features() implementation */ > > @@ -199,6 +202,82 @@ static int vp_active_vq(struct virtqueue *vq, u16 = msix_vec) > > return 0; > > } > > > > +static int vp_modern_reset_vq(struct virtqueue *vq) > > +{ > > + struct virtio_pci_device *vp_dev =3D to_vp_device(vq->vdev); > > + struct virtio_pci_modern_device *mdev =3D &vp_dev->mdev; > > + struct virtio_pci_vq_info *info; > > + unsigned long flags; > > + unsigned int irq; > > + > > + if (!virtio_has_feature(vq->vdev, VIRTIO_F_RING_RESET)) > > + return -ENOENT; > > + > > + vp_modern_set_queue_reset(mdev, vq->index); > > + > > + info =3D vp_dev->vqs[vq->index]; > > + > > + /* delete vq from irq handler */ > > + spin_lock_irqsave(&vp_dev->lock, flags); > > + list_del(&info->node); > > + spin_unlock_irqrestore(&vp_dev->lock, flags); > > + > > + INIT_LIST_HEAD(&info->node); > > + > > + vq->reset =3D VIRTIO_VQ_RESET_STEP_DEVICE; > > + > > + /* sync irq callback. */ > > + if (vp_dev->intx_enabled) { > > + irq =3D vp_dev->pci_dev->irq; > > + > > + } else { > > + if (info->msix_vector =3D=3D VIRTIO_MSI_NO_VECTOR) > > + return 0; > > + > > + irq =3D pci_irq_vector(vp_dev->pci_dev, info->msix_vector); > > + } > > + > > + synchronize_irq(irq); > > > Synchronize_irq() is not sufficient here since it breaks the effort of > the interrupt hardening which is done by commits: > > 080cd7c3ac87 virtio-pci: harden INTX interrupts > 9e35276a5344 virtio_pci: harden MSI-X interrupts > > Unfortunately=C2=A0 080cd7c3ac87 introduces an issue that disable_irq() w= ere > used for the affinity managed irq but we're discussing a fix. > ok, I think disable_irq() is still used here. I want to determine the solution for this detail first. So I posted the cod= e, I hope Jason can help confirm this point first. There are three situations in which vq corresponds to an interrupt 1. intx 2. msix: per vq vectors 2. msix: share irq Essentially can be divided into two categories: per vq vectors and share ir= q. For share irq is based on virtqueues to find vq, so I think it is safe as l= ong as list_del() is executed under the protection of the lock. In the case of per vq vectors, disable_irq() is used. Thanks. +static int vp_modern_reset_vq(struct virtqueue *vq) +{ + struct virtio_pci_device *vp_dev =3D to_vp_device(vq->vdev); + struct virtio_pci_modern_device *mdev =3D &vp_dev->mdev; + struct virtio_pci_vq_info *info; + unsigned long flags; + unsigned int irq; + + if (!virtio_has_feature(vq->vdev, VIRTIO_F_RING_RESET)) + return -ENOENT; + + vp_modern_set_queue_reset(mdev, vq->index); + + info =3D vp_dev->vqs[vq->index]; + + /* delete vq from irq handler */ + spin_lock_irqsave(&vp_dev->lock, flags); + list_del(&info->node); + vp_modern_set_queue_reset(mdev, vq->index); + + info =3D vp_dev->vqs[vq->index]; + + /* delete vq from irq handler */ + spin_lock_irqsave(&vp_dev->lock, flags); + list_del(&info->node); + spin_unlock_irqrestore(&vp_dev->lock, flags); + + INIT_LIST_HEAD(&info->node); + + /* For the case where vq has an exclusive irq, to prevent the irq f= rom + * being received again and the pending irq, call disable_irq(). + * + * In the scenario based on shared interrupts, vq will be searched = from + * the queue virtqueues. Since the previous list_del() has been del= eted + * from the queue, it is impossible for vq to be called in this cas= e. + * There is no need to close the corresponding interrupt. + */ + if (vp_dev->per_vq_vectors && msix_vec !=3D VIRTIO_MSI_NO_VECTOR) + disable_irq(pci_irq_vector(vp_dev->pci_dev, info->msix_vect= or)); + + vq->reset =3D true; + + return 0; +} + +static int vp_modern_enable_reset_vq(struct virtqueue *vq) +{ + struct virtio_pci_device *vp_dev =3D to_vp_device(vq->vdev); + struct virtio_pci_modern_device *mdev =3D &vp_dev->mdev; + struct virtio_pci_vq_info *info; + unsigned long flags, index; + int err; + + if (!vq->reset) + return -EBUSY; + + index =3D vq->index; + info =3D vp_dev->vqs[index]; + + /* check queue reset status */ + if (vp_modern_get_queue_reset(mdev, index) !=3D 1) + return -EBUSY; + + err =3D vp_active_vq(vq, info->msix_vector); + if (err) + return err; + + if (vq->callback) { + spin_lock_irqsave(&vp_dev->lock, flags); + list_add(&info->node, &vp_dev->virtqueues); + spin_unlock_irqrestore(&vp_dev->lock, flags); + } else { + INIT_LIST_HEAD(&info->node); + } + + vp_modern_set_queue_enable(&vp_dev->mdev, index, true); + vq->reset =3D false; + + if (vp_dev->per_vq_vectors && msix_vec !=3D VIRTIO_MSI_NO_VECTOR) + enable_irq(pci_irq_vector(vp_dev->pci_dev, info->msix_vecto= r)); + + return 0; +}