[PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA
David Marchand
david.marchand at redhat.com
Thu Jul 11 17:02:43 CEST 2024
On Thu, Jul 11, 2024 at 2:44 PM Srujana Challa <schalla at marvell.com> wrote:
>
> This patch modifies the code to convert descriptor buffer IOVA
> addresses to virtual addresses only when use_va flag is false.
>
> This patch resolves a segmentation fault with the vhost-user backend
> that occurs during the processing of the shadow control queue.
>
> 'Fixes: 67e9e504dae2 ("net/virtio_user: convert cq descriptor IOVA
> address to Virtual address")'
No single quote around the Fixes: tag, and on a single line please.
As for the title, how about: "net/virtio_user: fix cq descriptor
conversion with non vDPA backend" ?
>
> Reported-by: David Marchand <david.marchand at redhat.com>
> Signed-off-by: Srujana Challa <schalla at marvell.com>
> ---
> v2:
> - Added Reported-by tag.
>
> .../net/virtio/virtio_user/virtio_user_dev.c | 28 +++++++++++--------
> 1 file changed, 16 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> index fed66d2ae9..94e0ddcb94 100644
> --- a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> @@ -905,12 +905,12 @@ virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
> #define CVQ_MAX_DATA_DESCS 32
>
> static inline void *
> -virtio_user_iova2virt(rte_iova_t iova)
> +virtio_user_iova2virt(rte_iova_t iova, bool use_va)
All the code in this file passes the virtio_user_dev object, please
keep this convention.
IOW:
-virtio_user_iova2virt(rte_iova_t iova)
+virtio_user_iova2virt(struct virtio_user_dev *dev, rte_iova_t iova)
> {
> - if (rte_eal_iova_mode() == RTE_IOVA_VA)
> - return (void *)(uintptr_t)iova;
> - else
> + if (rte_eal_iova_mode() == RTE_IOVA_PA && !use_va)
> return rte_mem_iova2virt(iova);
> + else
> + return (void *)(uintptr_t)iova;
Why do we need to invert this test?
I would make the change as simple as:
- if (rte_eal_iova_mode() == RTE_IOVA_VA)
+ if (rte_eal_iova_mode() == RTE_IOVA_VA || dev->hw.use_va)
> }
>
> static uint32_t
> @@ -922,6 +922,7 @@ virtio_user_handle_ctrl_msg_split(struct virtio_user_dev *dev, struct vring *vri
> uint16_t i, idx_data, idx_status;
> uint32_t n_descs = 0;
> int dlen[CVQ_MAX_DATA_DESCS], nb_dlen = 0;
> + bool use_va = dev->hw.use_va;
>
> /* locate desc for header, data, and status */
> idx_data = vring->desc[idx_hdr].next;
> @@ -938,18 +939,18 @@ virtio_user_handle_ctrl_msg_split(struct virtio_user_dev *dev, struct vring *vri
> idx_status = i;
> n_descs++;
>
> - hdr = virtio_user_iova2virt(vring->desc[idx_hdr].addr);
> + hdr = virtio_user_iova2virt(vring->desc[idx_hdr].addr, use_va);
> if (hdr->class == VIRTIO_NET_CTRL_MQ &&
> hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
> uint16_t queues, *addr;
>
> - addr = virtio_user_iova2virt(vring->desc[idx_data].addr);
> + addr = virtio_user_iova2virt(vring->desc[idx_data].addr, use_va);
> queues = *addr;
> status = virtio_user_handle_mq(dev, queues);
> } else if (hdr->class == VIRTIO_NET_CTRL_MQ && hdr->cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
> struct virtio_net_ctrl_rss *rss;
>
> - rss = virtio_user_iova2virt(vring->desc[idx_data].addr);
> + rss = virtio_user_iova2virt(vring->desc[idx_data].addr, use_va);
> status = virtio_user_handle_mq(dev, rss->max_tx_vq);
> } else if (hdr->class == VIRTIO_NET_CTRL_RX ||
> hdr->class == VIRTIO_NET_CTRL_MAC ||
> @@ -962,7 +963,8 @@ virtio_user_handle_ctrl_msg_split(struct virtio_user_dev *dev, struct vring *vri
> (struct virtio_pmd_ctrl *)hdr, dlen, nb_dlen);
>
> /* Update status */
> - *(virtio_net_ctrl_ack *)virtio_user_iova2virt(vring->desc[idx_status].addr) = status;
> + *(virtio_net_ctrl_ack *)
> + virtio_user_iova2virt(vring->desc[idx_status].addr, use_va) = status;
Afaics, no need for reindenting.
>
> return n_descs;
> }
> @@ -987,6 +989,7 @@ virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
> /* initialize to one, header is first */
> uint32_t n_descs = 1;
> int dlen[CVQ_MAX_DATA_DESCS], nb_dlen = 0;
> + bool use_va = dev->hw.use_va;
>
> /* locate desc for header, data, and status */
> idx_data = idx_hdr + 1;
> @@ -1004,18 +1007,18 @@ virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
> n_descs++;
> }
>
> - hdr = virtio_user_iova2virt(vring->desc[idx_hdr].addr);
> + hdr = virtio_user_iova2virt(vring->desc[idx_hdr].addr, use_va);
> if (hdr->class == VIRTIO_NET_CTRL_MQ &&
> hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
> uint16_t queues, *addr;
>
> - addr = virtio_user_iova2virt(vring->desc[idx_data].addr);
> + addr = virtio_user_iova2virt(vring->desc[idx_data].addr, use_va);
> queues = *addr;
> status = virtio_user_handle_mq(dev, queues);
> } else if (hdr->class == VIRTIO_NET_CTRL_MQ && hdr->cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
> struct virtio_net_ctrl_rss *rss;
>
> - rss = virtio_user_iova2virt(vring->desc[idx_data].addr);
> + rss = virtio_user_iova2virt(vring->desc[idx_data].addr, use_va);
> status = virtio_user_handle_mq(dev, rss->max_tx_vq);
> } else if (hdr->class == VIRTIO_NET_CTRL_RX ||
> hdr->class == VIRTIO_NET_CTRL_MAC ||
> @@ -1028,7 +1031,8 @@ virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
> (struct virtio_pmd_ctrl *)hdr, dlen, nb_dlen);
>
> /* Update status */
> - *(virtio_net_ctrl_ack *)virtio_user_iova2virt(vring->desc[idx_status].addr) = status;
> + *(virtio_net_ctrl_ack *)
> + virtio_user_iova2virt(vring->desc[idx_status].addr, use_va) = status;
Idem, no reindenting.
>
> /* Update used descriptor */
> vring->desc[idx_hdr].id = vring->desc[idx_status].id;
> --
> 2.25.1
>
--
David Marchand
More information about the dev
mailing list