Message-ID: <06b3adbb-6777-7022-00d2-beca2b166e10@redhat.com>
Date:   Fri, 11 Mar 2022 13:09:36 +0800
From:   Jason Wang <jasowang@...hat.com>
To:     Xuan Zhuo <xuanzhuo@...ux.alibaba.com>
Cc:     Jeff Dike <jdike@...toit.com>, Richard Weinberger <richard@....at>,
        Anton Ivanov <anton.ivanov@...bridgegreys.com>,
        "Michael S. Tsirkin" <mst@...hat.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Hans de Goede <hdegoede@...hat.com>,
        Mark Gross <markgross@...nel.org>,
        Vadim Pasternak <vadimp@...dia.com>,
        Bjorn Andersson <bjorn.andersson@...aro.org>,
        Mathieu Poirier <mathieu.poirier@...aro.org>,
        Cornelia Huck <cohuck@...hat.com>,
        Halil Pasic <pasic@...ux.ibm.com>,
        Heiko Carstens <hca@...ux.ibm.com>,
        Vasily Gorbik <gor@...ux.ibm.com>,
        Christian Borntraeger <borntraeger@...ux.ibm.com>,
        Alexander Gordeev <agordeev@...ux.ibm.com>,
        Sven Schnelle <svens@...ux.ibm.com>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Jesper Dangaard Brouer <hawk@...nel.org>,
        John Fastabend <john.fastabend@...il.com>,
        Johannes Berg <johannes.berg@...el.com>,
        Vincent Whitchurch <vincent.whitchurch@...s.com>,
        linux-um@...ts.infradead.org, platform-driver-x86@...r.kernel.org,
        linux-remoteproc@...r.kernel.org, linux-s390@...r.kernel.org,
        kvm@...r.kernel.org, bpf@...r.kernel.org,
        virtualization@...ts.linux-foundation.org, netdev@...r.kernel.org
Subject: Re: [PATCH v7 17/26] virtio_pci: queue_reset: support
 VIRTIO_F_RING_RESET


在 2022/3/10 下午4:20, Xuan Zhuo 写道:
> On Wed, 9 Mar 2022 16:54:10 +0800, Jason Wang <jasowang@...hat.com> wrote:
>> 在 2022/3/8 下午8:35, Xuan Zhuo 写道:
>>> 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 queue
>>>    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 <xuanzhuo@...ux.alibaba.com>
>>> ---
>>>    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 = 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_device *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 = to_vp_device(vq->vdev);
>>> +	struct virtio_pci_modern_device *mdev = &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 = 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 = VIRTIO_VQ_RESET_STEP_DEVICE;
>>> +
>>> +	/* sync irq callback. */
>>> +	if (vp_dev->intx_enabled) {
>>> +		irq = vp_dev->pci_dev->irq;
>>> +
>>> +	} else {
>>> +		if (info->msix_vector == VIRTIO_MSI_NO_VECTOR)
>>> +			return 0;
>>> +
>>> +		irq = 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  080cd7c3ac87 introduces an issue that disable_irq() were
>> 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 code, 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 irq.
>
> For share irq is based on virtqueues to find vq, so I think it is safe as long
> as list_del() is executed under the protection of the lock.
>
> In the case of per vq vectors, disable_irq() is used.


See the discussion here[1], disable_irq() could be problematic for the 
block and scsi device that using affinity managed irq. We're waiting for 
the IRQ maintainer to comment on a solution. Other looks sane.

Thanks

[1] https://lkml.org/lkml/2022/3/8/743


>
> Thanks.
>
> +static int vp_modern_reset_vq(struct virtqueue *vq)
> +{
> +       struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
> +       struct virtio_pci_modern_device *mdev = &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 = 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 = 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 from
> +        * 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 deleted
> +        * from the queue, it is impossible for vq to be called in this case.
> +        * There is no need to close the corresponding interrupt.
> +        */
> +       if (vp_dev->per_vq_vectors && msix_vec != VIRTIO_MSI_NO_VECTOR)
> +               disable_irq(pci_irq_vector(vp_dev->pci_dev, info->msix_vector));
> +
> +       vq->reset = true;
> +
> +       return 0;
> +}
> +
> +static int vp_modern_enable_reset_vq(struct virtqueue *vq)
> +{
> +       struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
> +       struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
> +       struct virtio_pci_vq_info *info;
> +       unsigned long flags, index;
> +       int err;
> +
> +       if (!vq->reset)
> +               return -EBUSY;
> +
> +       index = vq->index;
> +       info = vp_dev->vqs[index];
> +
> +       /* check queue reset status */
> +       if (vp_modern_get_queue_reset(mdev, index) != 1)
> +               return -EBUSY;
> +
> +       err = 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 = false;
> +
> +       if (vp_dev->per_vq_vectors && msix_vec != VIRTIO_MSI_NO_VECTOR)
> +               enable_irq(pci_irq_vector(vp_dev->pci_dev, info->msix_vector));
> +
> +       return 0;
> +}
>
>

Powered by blists - more mailing lists