[v4] net/virtio-user: fix packed ring server mode
Checks
Commit Message
This patch fixes the situation where datapath does not work properly when
vhost reconnects to virtio in server mode with packed ring.
Currently, virtio and vhost share memory of vring. For split ring, vhost
can read the status of discriptors directly from the available ring and
the used ring during reconnection. Therefore, the datapath can continue.
But for packed ring, when reconnecting to virtio, vhost cannot get the
status of discriptors only through the descriptor ring. By resetting Tx
and Rx queues, the datapath can restart from the beginning.
Fixes: 4c3f5822eb214 ("net/virtio: add packed virtqueue defines")
Cc: stable@dpdk.org
Signed-off-by: Xuan Ding <xuan.ding@intel.com>
---
v4:
* Moved change log below '---' marker.
v3:
* Removed an extra asterisk from a comment.
* Renamed device reset function and moved it to virtio_user_ethdev.c.
v2:
* Renamed queue reset functions and moved them to virtqueue.c.
drivers/net/virtio/virtio_ethdev.c | 4 +-
drivers/net/virtio/virtio_user_ethdev.c | 40 ++++++++++++++
drivers/net/virtio/virtqueue.c | 71 +++++++++++++++++++++++++
drivers/net/virtio/virtqueue.h | 4 ++
4 files changed, 117 insertions(+), 2 deletions(-)
Comments
On 12/23/19 8:25 AM, Xuan Ding wrote:
> This patch fixes the situation where datapath does not work properly when
> vhost reconnects to virtio in server mode with packed ring.
>
> Currently, virtio and vhost share memory of vring. For split ring, vhost
> can read the status of discriptors directly from the available ring and
s/discriptors/descriptors/
> the used ring during reconnection. Therefore, the datapath can continue.
>
> But for packed ring, when reconnecting to virtio, vhost cannot get the
> status of discriptors only through the descriptor ring. By resetting Tx
s/discriptor only through/descriptor via/
> and Rx queues, the datapath can restart from the beginning.
>
> Fixes: 4c3f5822eb214 ("net/virtio: add packed virtqueue defines")
> Cc: stable@dpdk.org
>
> Signed-off-by: Xuan Ding <xuan.ding@intel.com>
> ---
>
> v4:
> * Moved change log below '---' marker.
>
> v3:
> * Removed an extra asterisk from a comment.
> * Renamed device reset function and moved it to virtio_user_ethdev.c.
>
> v2:
> * Renamed queue reset functions and moved them to virtqueue.c.
>
> drivers/net/virtio/virtio_ethdev.c | 4 +-
> drivers/net/virtio/virtio_user_ethdev.c | 40 ++++++++++++++
> drivers/net/virtio/virtqueue.c | 71 +++++++++++++++++++++++++
> drivers/net/virtio/virtqueue.h | 4 ++
> 4 files changed, 117 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> index 044eb10a7..f9d0ea70d 100644
> --- a/drivers/net/virtio/virtio_ethdev.c
> +++ b/drivers/net/virtio/virtio_ethdev.c
> @@ -1913,6 +1913,8 @@ eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
> goto err_vtpci_init;
> }
>
> + rte_spinlock_init(&hw->state_lock);
> +
> /* reset device and negotiate default features */
> ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
> if (ret < 0)
> @@ -2155,8 +2157,6 @@ virtio_dev_configure(struct rte_eth_dev *dev)
> return -EBUSY;
> }
>
> - rte_spinlock_init(&hw->state_lock);
> -
> hw->use_simple_rx = 1;
>
> if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
> diff --git a/drivers/net/virtio/virtio_user_ethdev.c b/drivers/net/virtio/virtio_user_ethdev.c
> index 3fc172573..425f48230 100644
> --- a/drivers/net/virtio/virtio_user_ethdev.c
> +++ b/drivers/net/virtio/virtio_user_ethdev.c
> @@ -25,12 +25,48 @@
> #define virtio_user_get_dev(hw) \
> ((struct virtio_user_dev *)(hw)->virtio_user_dev)
>
> +static void
> +virtio_user_reset_queues_packed(struct rte_eth_dev *dev)
> +{
> + struct virtio_hw *hw = dev->data->dev_private;
> + struct virtnet_rx *rxvq;
> + struct virtnet_tx *txvq;
> + uint16_t i;
> +
> + /* Add lock to avoid queue contention. */
> + rte_spinlock_lock(&hw->state_lock);
> + hw->started = 0;
> +
> + /*
> + * Waitting for datapath to complete before resetting queues.
> + * 1 ms should be enough for the ongoing Tx/Rx function to finish.
> + */
> + rte_delay_ms(1);
> +
> + /* Vring reset for each Tx queue and Rx queue. */
> + for (i = 0; i < dev->data->nb_rx_queues; i++) {
> + rxvq = dev->data->rx_queues[i];
> + virtqueue_rxvq_reset_packed(rxvq->vq);
> + virtio_dev_rx_queue_setup_finish(dev, i);
> + }
> +
> + for (i = 0; i < dev->data->nb_tx_queues; i++) {
> + txvq = dev->data->tx_queues[i];
> + virtqueue_txvq_reset_packed(txvq->vq);
> + }
> +
> + hw->started = 1;
> + rte_spinlock_unlock(&hw->state_lock);
> +}
> +
> +
> static int
> virtio_user_server_reconnect(struct virtio_user_dev *dev)
> {
> int ret;
> int connectfd;
> struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
> + struct virtio_hw *hw = eth_dev->data->dev_private;
>
> connectfd = accept(dev->listenfd, NULL, NULL);
> if (connectfd < 0)
> @@ -51,6 +87,10 @@ virtio_user_server_reconnect(struct virtio_user_dev *dev)
>
> dev->features &= dev->device_features;
>
> + /* For packed ring, resetting queues is required in reconnection. */
> + if (vtpci_packed_queue(hw))
> + virtio_user_reset_queues_packed(eth_dev);
I think we should log an error here to state that some packets are
likely to be dropped.
Other than that, I don't think we have other choices than doing that
right now, so I'll ack your patch once commit message is fixed and
error message is printed on vrings reset.
Thanks,
Maxime
Hi Maxime,
Replies are inline.
> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin@redhat.com>
> Sent: Tuesday, January 14, 2020 11:04 PM
> To: Ding, Xuan <xuan.ding@intel.com>; Bie, Tiwei <tiwei.bie@intel.com>; Wang,
> Zhihong <zhihong.wang@intel.com>; Liu, Yong <yong.liu@intel.com>; Ye,
> Xiaolong <xiaolong.ye@intel.com>
> Cc: dev@dpdk.org; stable@dpdk.org
> Subject: Re: [PATCH v4] net/virtio-user: fix packed ring server mode
>
>
>
> On 12/23/19 8:25 AM, Xuan Ding wrote:
> > This patch fixes the situation where datapath does not work properly
> > when vhost reconnects to virtio in server mode with packed ring.
> >
> > Currently, virtio and vhost share memory of vring. For split ring,
> > vhost can read the status of discriptors directly from the available
> > ring and
>
> s/discriptors/descriptors/
>
> > the used ring during reconnection. Therefore, the datapath can continue.
> >
> > But for packed ring, when reconnecting to virtio, vhost cannot get the
> > status of discriptors only through the descriptor ring. By resetting
> > Tx
>
> s/discriptor only through/descriptor via/
>
Thanks, this will be corrected in the new version.
> > and Rx queues, the datapath can restart from the beginning.
> >
> > Fixes: 4c3f5822eb214 ("net/virtio: add packed virtqueue defines")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Xuan Ding <xuan.ding@intel.com>
> > ---
> >
> > v4:
> > * Moved change log below '---' marker.
> >
> > v3:
> > * Removed an extra asterisk from a comment.
> > * Renamed device reset function and moved it to virtio_user_ethdev.c.
> >
> > v2:
> > * Renamed queue reset functions and moved them to virtqueue.c.
> >
> > drivers/net/virtio/virtio_ethdev.c | 4 +-
> > drivers/net/virtio/virtio_user_ethdev.c | 40 ++++++++++++++
> > drivers/net/virtio/virtqueue.c | 71 +++++++++++++++++++++++++
> > drivers/net/virtio/virtqueue.h | 4 ++
> > 4 files changed, 117 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/virtio/virtio_ethdev.c
> > b/drivers/net/virtio/virtio_ethdev.c
> > index 044eb10a7..f9d0ea70d 100644
> > --- a/drivers/net/virtio/virtio_ethdev.c
> > +++ b/drivers/net/virtio/virtio_ethdev.c
> > @@ -1913,6 +1913,8 @@ eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
> > goto err_vtpci_init;
> > }
> >
> > + rte_spinlock_init(&hw->state_lock);
> > +
> > /* reset device and negotiate default features */
> > ret = virtio_init_device(eth_dev,
> VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
> > if (ret < 0)
> > @@ -2155,8 +2157,6 @@ virtio_dev_configure(struct rte_eth_dev *dev)
> > return -EBUSY;
> > }
> >
> > - rte_spinlock_init(&hw->state_lock);
> > -
> > hw->use_simple_rx = 1;
> >
> > if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) { diff --git
> > a/drivers/net/virtio/virtio_user_ethdev.c
> > b/drivers/net/virtio/virtio_user_ethdev.c
> > index 3fc172573..425f48230 100644
> > --- a/drivers/net/virtio/virtio_user_ethdev.c
> > +++ b/drivers/net/virtio/virtio_user_ethdev.c
> > @@ -25,12 +25,48 @@
> > #define virtio_user_get_dev(hw) \
> > ((struct virtio_user_dev *)(hw)->virtio_user_dev)
> >
> > +static void
> > +virtio_user_reset_queues_packed(struct rte_eth_dev *dev) {
> > + struct virtio_hw *hw = dev->data->dev_private;
> > + struct virtnet_rx *rxvq;
> > + struct virtnet_tx *txvq;
> > + uint16_t i;
> > +
> > + /* Add lock to avoid queue contention. */
> > + rte_spinlock_lock(&hw->state_lock);
> > + hw->started = 0;
> > +
> > + /*
> > + * Waitting for datapath to complete before resetting queues.
> > + * 1 ms should be enough for the ongoing Tx/Rx function to finish.
> > + */
> > + rte_delay_ms(1);
> > +
> > + /* Vring reset for each Tx queue and Rx queue. */
> > + for (i = 0; i < dev->data->nb_rx_queues; i++) {
> > + rxvq = dev->data->rx_queues[i];
> > + virtqueue_rxvq_reset_packed(rxvq->vq);
> > + virtio_dev_rx_queue_setup_finish(dev, i);
> > + }
> > +
> > + for (i = 0; i < dev->data->nb_tx_queues; i++) {
> > + txvq = dev->data->tx_queues[i];
> > + virtqueue_txvq_reset_packed(txvq->vq);
> > + }
> > +
> > + hw->started = 1;
> > + rte_spinlock_unlock(&hw->state_lock);
> > +}
> > +
> > +
> > static int
> > virtio_user_server_reconnect(struct virtio_user_dev *dev) {
> > int ret;
> > int connectfd;
> > struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
> > + struct virtio_hw *hw = eth_dev->data->dev_private;
> >
> > connectfd = accept(dev->listenfd, NULL, NULL);
> > if (connectfd < 0)
> > @@ -51,6 +87,10 @@ virtio_user_server_reconnect(struct virtio_user_dev
> > *dev)
> >
> > dev->features &= dev->device_features;
> >
> > + /* For packed ring, resetting queues is required in reconnection. */
> > + if (vtpci_packed_queue(hw))
> > + virtio_user_reset_queues_packed(eth_dev);
>
> I think we should log an error here to state that some packets are likely to be
> dropped.
>
> Other than that, I don't think we have other choices than doing that right now,
> so I'll ack your patch once commit message is fixed and error message is printed
> on vrings reset.
>
Notice message will be added in the new version. This message will remind user that packets on the fly will be dropped when packed ring reconnecting.
Thank you for your comments.
Regards,
Xuan
> Thanks,
> Maxime
@@ -1913,6 +1913,8 @@ eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
goto err_vtpci_init;
}
+ rte_spinlock_init(&hw->state_lock);
+
/* reset device and negotiate default features */
ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
if (ret < 0)
@@ -2155,8 +2157,6 @@ virtio_dev_configure(struct rte_eth_dev *dev)
return -EBUSY;
}
- rte_spinlock_init(&hw->state_lock);
-
hw->use_simple_rx = 1;
if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
@@ -25,12 +25,48 @@
#define virtio_user_get_dev(hw) \
((struct virtio_user_dev *)(hw)->virtio_user_dev)
+static void
+virtio_user_reset_queues_packed(struct rte_eth_dev *dev)
+{
+ struct virtio_hw *hw = dev->data->dev_private;
+ struct virtnet_rx *rxvq;
+ struct virtnet_tx *txvq;
+ uint16_t i;
+
+ /* Add lock to avoid queue contention. */
+ rte_spinlock_lock(&hw->state_lock);
+ hw->started = 0;
+
+ /*
+ * Waitting for datapath to complete before resetting queues.
+ * 1 ms should be enough for the ongoing Tx/Rx function to finish.
+ */
+ rte_delay_ms(1);
+
+ /* Vring reset for each Tx queue and Rx queue. */
+ for (i = 0; i < dev->data->nb_rx_queues; i++) {
+ rxvq = dev->data->rx_queues[i];
+ virtqueue_rxvq_reset_packed(rxvq->vq);
+ virtio_dev_rx_queue_setup_finish(dev, i);
+ }
+
+ for (i = 0; i < dev->data->nb_tx_queues; i++) {
+ txvq = dev->data->tx_queues[i];
+ virtqueue_txvq_reset_packed(txvq->vq);
+ }
+
+ hw->started = 1;
+ rte_spinlock_unlock(&hw->state_lock);
+}
+
+
static int
virtio_user_server_reconnect(struct virtio_user_dev *dev)
{
int ret;
int connectfd;
struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
+ struct virtio_hw *hw = eth_dev->data->dev_private;
connectfd = accept(dev->listenfd, NULL, NULL);
if (connectfd < 0)
@@ -51,6 +87,10 @@ virtio_user_server_reconnect(struct virtio_user_dev *dev)
dev->features &= dev->device_features;
+ /* For packed ring, resetting queues is required in reconnection. */
+ if (vtpci_packed_queue(hw))
+ virtio_user_reset_queues_packed(eth_dev);
+
ret = virtio_user_start_device(dev);
if (ret < 0)
return -1;
@@ -141,3 +141,74 @@ virtqueue_rxvq_flush(struct virtqueue *vq)
else
virtqueue_rxvq_flush_split(vq);
}
+
+int
+virtqueue_rxvq_reset_packed(struct virtqueue *vq)
+{
+ int size = vq->vq_nentries;
+ struct vq_desc_extra *dxp;
+ struct virtnet_rx *rxvq;
+ uint16_t desc_idx;
+
+ vq->vq_used_cons_idx = 0;
+ vq->vq_desc_head_idx = 0;
+ vq->vq_avail_idx = 0;
+ vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
+ vq->vq_free_cnt = vq->vq_nentries;
+
+ vq->vq_packed.used_wrap_counter = 1;
+ vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
+ vq->vq_packed.event_flags_shadow = 0;
+ vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
+
+ rxvq = &vq->rxq;
+ memset(rxvq->mz->addr, 0, rxvq->mz->len);
+
+ for (desc_idx = 0; desc_idx < vq->vq_nentries; desc_idx++) {
+ dxp = &vq->vq_descx[desc_idx];
+ if (dxp->cookie != NULL) {
+ rte_pktmbuf_free(dxp->cookie);
+ dxp->cookie = NULL;
+ }
+ }
+
+ vring_desc_init_packed(vq, size);
+
+ return 0;
+}
+
+int
+virtqueue_txvq_reset_packed(struct virtqueue *vq)
+{
+ int size = vq->vq_nentries;
+ struct vq_desc_extra *dxp;
+ struct virtnet_tx *txvq;
+ uint16_t desc_idx;
+
+ vq->vq_used_cons_idx = 0;
+ vq->vq_desc_head_idx = 0;
+ vq->vq_avail_idx = 0;
+ vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
+ vq->vq_free_cnt = vq->vq_nentries;
+
+ vq->vq_packed.used_wrap_counter = 1;
+ vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
+ vq->vq_packed.event_flags_shadow = 0;
+
+ txvq = &vq->txq;
+ memset(txvq->mz->addr, 0, txvq->mz->len);
+ memset(txvq->virtio_net_hdr_mz->addr, 0,
+ txvq->virtio_net_hdr_mz->len);
+
+ for (desc_idx = 0; desc_idx < vq->vq_nentries; desc_idx++) {
+ dxp = &vq->vq_descx[desc_idx];
+ if (dxp->cookie != NULL) {
+ rte_pktmbuf_free(dxp->cookie);
+ dxp->cookie = NULL;
+ }
+ }
+
+ vring_desc_init_packed(vq, size);
+
+ return 0;
+}
@@ -443,6 +443,10 @@ struct rte_mbuf *virtqueue_detach_unused(struct virtqueue *vq);
/* Flush the elements in the used ring. */
void virtqueue_rxvq_flush(struct virtqueue *vq);
+int virtqueue_rxvq_reset_packed(struct virtqueue *vq);
+
+int virtqueue_txvq_reset_packed(struct virtqueue *vq);
+
static inline int
virtqueue_full(const struct virtqueue *vq)
{