[PATCH v7 17/26] virtio_pci: queue_reset: support VIRTIO_F_RING_RESET

[Jason Wang]

在 2022/3/9 下午5:32, Xuan Zhuo 写道:
> On Wed, 9 Mar 2022 16:54:10 +0800, Jason Wang <jasowang at redhat.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 at linux.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.
> I need to understand it first.
>
>>
>>> +
>>> +	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 != VIRTIO_VQ_RESET_STEP_VRING_ATTACH)
>>> +		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);
>>
>> Any reason we need to check queue_enable() here?
> The purpose of this function is to enable a reset vq, so call queue_enable() to
> activate it.


Ok, this is what spec mandate.

Thanks


>
> Thanks.
>
>> Thanks
>>
>>
>>> +	vq->reset = VIRTIO_VQ_RESET_STEP_NONE;
>>> +
>>> +	return 0;
>>> +}
>>> +
>>>    static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
>>>    {
>>>    	return vp_modern_config_vector(&vp_dev->mdev, vector);
>>> @@ -407,6 +486,8 @@ static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
>>>    	.set_vq_affinity = vp_set_vq_affinity,
>>>    	.get_vq_affinity = vp_get_vq_affinity,
>>>    	.get_shm_region  = vp_get_shm_region,
>>> +	.reset_vq	 = vp_modern_reset_vq,
>>> +	.enable_reset_vq = vp_modern_enable_reset_vq,
>>>    };
>>>
>>>    static const struct virtio_config_ops virtio_pci_config_ops = {
>>> @@ -425,6 +506,8 @@ static const struct virtio_config_ops virtio_pci_config_ops = {
>>>    	.set_vq_affinity = vp_set_vq_affinity,
>>>    	.get_vq_affinity = vp_get_vq_affinity,
>>>    	.get_shm_region  = vp_get_shm_region,
>>> +	.reset_vq	 = vp_modern_reset_vq,
>>> +	.enable_reset_vq = vp_modern_enable_reset_vq,
>>>    };
>>>
>>>    /* the PCI probing function */