On 02/09/2018 09:59 AM, Maxime Coquelin wrote:
Hi Zhihong,

On 02/09/2018 06:44 AM, Wang, Zhihong wrote:
Hi Olivier,

Given the situation that the vec path can be selected silently now once
condition is met. So theoretically speaking this issue impacts the whole
virtio pmd. If you plan to fix it in the next release, do you want to do
a temporary workaround to disable the vec path selection till then?

That may be the less worse solution if we don't fix it quickly.
Reverting the patch isn't trivial, so this is not an option.

I'm trying to reproduce it now, I'll let you know if I find something.

Hmm, I reproduced Tiwei instructions, and in my case, Vhost's testpmd
crashes because Virtio-user makes it doing an out of bound access.

Could you provide a patch to disable vector path selection? I'll
continue to debug, but we can start reviewing it so that it is ready if
we need it.



-----Original Message-----
From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Olivier Matz
Sent: Thursday, February 8, 2018 6:01 AM
To: Xu, Qian Q <qian.q...@intel.com>
Cc: Yao, Lei A <lei.a....@intel.com>; dev@dpdk.org; y...@fridaylinux.org;
maxime.coque...@redhat.com; Thomas Monjalon <tho...@monjalon.net>;
Subject: Re: [dpdk-dev] [PATCH v2 06/10] net/virtio: fix queue setup


It's in my short plans, but unfortunately some other high priority tasks
were inserted before. Honnestly, I'm not sure I'll be able to make it
for the release, but I'll do my best.


On Wed, Feb 07, 2018 at 08:31:07AM +0000, Xu, Qian Q wrote:
Any update, Olivier?
We are near to release, and the bug-fix is important for the virtio vector
path usage. Thanks.

-----Original Message-----
From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Olivier Matz
Sent: Thursday, February 1, 2018 4:28 PM
To: Yao, Lei A <lei.a....@intel.com>
Cc: dev@dpdk.org; y...@fridaylinux.org; maxime.coque...@redhat.com;
Thomas Monjalon <tho...@monjalon.net>; sta...@dpdk.org
Subject: Re: [dpdk-dev] [PATCH v2 06/10] net/virtio: fix queue setup

Hi Lei,

It's on my todo list, I'll check this as soon as possible.


On Thu, Feb 01, 2018 at 03:14:15AM +0000, Yao, Lei A wrote:
Hi, Olivier

This is Lei from DPDK validation team in Intel. During our DPDK
18.02-rc1 test, I find the following patch will cause one serious issue
with virtio
vector path:
the traffic can't resume after stop/start the virtio device.

The step like following:
1. Launch vhost-user port using testpmd at Host 2. Launch VM with
virtio device, mergeable is off 3. Bind the virtio device to pmd
driver, launch testpmd, let the tx/rx use vector path
4. Send traffic to virtio device from vhost side, then stop the virtio
device 5. Start the virtio device again After step 5, the traffic
can't resume.

Could you help check this and give a fix? This issue will impact the
virtio pmd user experience heavily. By the way, this patch is already
included into V17.11. Looks like we need give a patch to this LTS version.
Thanks a lot!

-----Original Message-----
From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Olivier Matz
Sent: Thursday, September 7, 2017 8:14 PM
To: dev@dpdk.org; y...@fridaylinux.org;
Cc: step...@networkplumber.org; sta...@dpdk.org
Subject: [dpdk-dev] [PATCH v2 06/10] net/virtio: fix queue setup

In rx/tx queue setup functions, some code is executed only if
use_simple_rxtx == 1. The value of this variable can change
depending on the offload flags or sse support. If Rx queue setup is
called before Tx queue setup, it can result in an invalid configuration:

- dev_configure is called: use_simple_rxtx is initialized to 0
- rx queue setup is called: queues are initialized without simple path
- tx queue setup is called: use_simple_rxtx switch to 1, and simple
   Rx/Tx handlers are selected

Fix this by postponing a part of Rx/Tx queue initialization in
dev_start(), as it was the case in the initial implementation.

Fixes: 48cec290a3d2 ("net/virtio: move queue configure code to
Cc: sta...@dpdk.org

Signed-off-by: Olivier Matz <olivier.m...@6wind.com>
  drivers/net/virtio/virtio_ethdev.c | 13 +++++++++++++
drivers/net/virtio/virtio_ethdev.h |  6 ++++++
  drivers/net/virtio/virtio_rxtx.c   | 40
  3 files changed, 51 insertions(+), 8 deletions(-)

diff --git a/drivers/net/virtio/virtio_ethdev.c
index 8eee3ff80..c7888f103 100644
--- a/drivers/net/virtio/virtio_ethdev.c
+++ b/drivers/net/virtio/virtio_ethdev.c
@@ -1737,6 +1737,19 @@ virtio_dev_start(struct rte_eth_dev *dev)
      struct virtnet_rx *rxvq;
      struct virtnet_tx *txvq __rte_unused;
      struct virtio_hw *hw = dev->data->dev_private;
+    int ret;
+    /* Finish the initialization of the queues */
+    for (i = 0; i < dev->data->nb_rx_queues; i++) {
+        ret = virtio_dev_rx_queue_setup_finish(dev, i);
+        if (ret < 0)
+            return ret;
+    }
+    for (i = 0; i < dev->data->nb_tx_queues; i++) {
+        ret = virtio_dev_tx_queue_setup_finish(dev, i);
+        if (ret < 0)
+            return ret;
+    }

      /* check if lsc interrupt feature is enabled */
      if (dev->data->dev_conf.intr_conf.lsc) { diff --git
index c3413c6d9..2039bc547 100644
--- a/drivers/net/virtio/virtio_ethdev.h
+++ b/drivers/net/virtio/virtio_ethdev.h
@@ -92,10 +92,16 @@ int  virtio_dev_rx_queue_setup(struct
rte_eth_dev *dev, uint16_t rx_queue_id,
          const struct rte_eth_rxconf *rx_conf,
          struct rte_mempool *mb_pool);

+int virtio_dev_rx_queue_setup_finish(struct rte_eth_dev *dev,
+                uint16_t rx_queue_id);
  int  virtio_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t
          uint16_t nb_tx_desc, unsigned int socket_id,
          const struct rte_eth_txconf *tx_conf);

+int virtio_dev_tx_queue_setup_finish(struct rte_eth_dev *dev,
+                uint16_t tx_queue_id);
  uint16_t virtio_recv_pkts(void *rx_queue, struct rte_mbuf
          uint16_t nb_pkts);

diff --git a/drivers/net/virtio/virtio_rxtx.c
index e30377c51..a32e3229f 100644
--- a/drivers/net/virtio/virtio_rxtx.c
+++ b/drivers/net/virtio/virtio_rxtx.c
@@ -421,9 +421,6 @@ virtio_dev_rx_queue_setup(struct
rte_eth_dev *dev,
      struct virtio_hw *hw = dev->data->dev_private;
      struct virtqueue *vq = hw->vqs[vtpci_queue_idx];
      struct virtnet_rx *rxvq;
-    int error, nbufs;
-    struct rte_mbuf *m;
-    uint16_t desc_idx;


@@ -440,10 +437,24 @@ virtio_dev_rx_queue_setup(struct
      dev->data->rx_queues[queue_idx] = rxvq;

+    return 0;
+virtio_dev_rx_queue_setup_finish(struct rte_eth_dev *dev,
+    uint16_t vtpci_queue_idx = 2 * queue_idx +
+    struct virtio_hw *hw = dev->data->dev_private;
+    struct virtqueue *vq = hw->vqs[vtpci_queue_idx];
+    struct virtnet_rx *rxvq = &vq->rxq;
+    struct rte_mbuf *m;
+    uint16_t desc_idx;
+    int error, nbufs;

      /* Allocate blank mbufs for the each rx descriptor */
      nbufs = 0;
-    error = ENOSPC;

      if (hw->use_simple_rxtx) {
          for (desc_idx = 0; desc_idx < vq->vq_nentries; @@ -
@@ virtio_dev_tx_queue_setup(struct rte_eth_dev *dev,
      struct virtqueue *vq = hw->vqs[vtpci_queue_idx];
      struct virtnet_tx *txvq;
      uint16_t tx_free_thresh;
-    uint16_t desc_idx;


@@ -563,9 +573,24 @@ virtio_dev_tx_queue_setup(struct

      vq->vq_free_thresh = tx_free_thresh;

-    if (hw->use_simple_rxtx) {
-        uint16_t mid_idx  = vq->vq_nentries >> 1;
+    dev->data->tx_queues[queue_idx] = txvq;
+    return 0;
+virtio_dev_tx_queue_setup_finish(struct rte_eth_dev *dev,
+                uint16_t queue_idx)
+    uint8_t vtpci_queue_idx = 2 * queue_idx +
+    struct virtio_hw *hw = dev->data->dev_private;
+    struct virtqueue *vq = hw->vqs[vtpci_queue_idx];
+    uint16_t mid_idx = vq->vq_nentries >> 1;
+    struct virtnet_tx *txvq = &vq->txq;
+    uint16_t desc_idx;

+    if (hw->use_simple_rxtx) {
          for (desc_idx = 0; desc_idx < mid_idx; desc_idx++) {
              vq->vq_ring.avail->ring[desc_idx] =
                  desc_idx + mid_idx;
@@ -587,7 +612,6 @@ virtio_dev_tx_queue_setup(struct


-    dev->data->tx_queues[queue_idx] = txvq;
      return 0;


Reply via email to