* [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA
@ 2024-07-11 12:44 Srujana Challa
2024-07-11 15:02 ` David Marchand
2024-07-12 12:36 ` [PATCH v3] net/virtio_user: fix cq descriptor conversion with non vDPA backend Srujana Challa
0 siblings, 2 replies; 8+ messages in thread
From: Srujana Challa @ 2024-07-11 12:44 UTC (permalink / raw)
To: dev, maxime.coquelin, chenbox
Cc: david.marchand, jerinj, ndabilpuram, vattunuru, schalla
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")'
Reported-by: David Marchand <david.marchand@redhat.com>
Signed-off-by: Srujana Challa <schalla@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)
{
- 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;
}
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;
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;
/* Update used descriptor */
vring->desc[idx_hdr].id = vring->desc[idx_status].id;
--
2.25.1
^ permalink raw reply [flat|nested] 8+ messages in thread* Re: [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA
2024-07-11 12:44 [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA Srujana Challa
@ 2024-07-11 15:02 ` David Marchand
2024-07-11 17:46 ` [EXTERNAL] " Srujana Challa
2024-07-12 12:36 ` [PATCH v3] net/virtio_user: fix cq descriptor conversion with non vDPA backend Srujana Challa
1 sibling, 1 reply; 8+ messages in thread
From: David Marchand @ 2024-07-11 15:02 UTC (permalink / raw)
To: Srujana Challa
Cc: dev, maxime.coquelin, chenbox, jerinj, ndabilpuram, vattunuru
On Thu, Jul 11, 2024 at 2:44 PM Srujana Challa <schalla@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@redhat.com>
> Signed-off-by: Srujana Challa <schalla@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
^ permalink raw reply [flat|nested] 8+ messages in thread* RE: [EXTERNAL] Re: [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA
2024-07-11 15:02 ` David Marchand
@ 2024-07-11 17:46 ` Srujana Challa
2024-07-12 7:57 ` David Marchand
2024-07-12 11:30 ` David Marchand
0 siblings, 2 replies; 8+ messages in thread
From: Srujana Challa @ 2024-07-11 17:46 UTC (permalink / raw)
To: David Marchand
Cc: dev, maxime.coquelin, chenbox, Jerin Jacob,
Nithin Kumar Dabilpuram, Vamsi Krishna Attunuru
> On Thu, Jul 11, 2024 at 2:44 PM Srujana Challa <schalla@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.
Ack
>
> As for the title, how about: "net/virtio_user: fix cq descriptor conversion with
> non vDPA backend" ?
Ack
>
> >
> > Reported-by: David Marchand <david.marchand@redhat.com>
> > Signed-off-by: Srujana Challa <schalla@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.
Ack
>
> 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?
Made this change to ensure that rte_mem_iova2virt() is not called when the IOVA mode is RTE_IOVA_DC
by any chance.
>
> 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.
It crossed 100 line boundary actually.
>
> >
> > 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
^ permalink raw reply [flat|nested] 8+ messages in thread* Re: [EXTERNAL] Re: [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA
2024-07-11 17:46 ` [EXTERNAL] " Srujana Challa
@ 2024-07-12 7:57 ` David Marchand
2024-07-12 11:30 ` David Marchand
1 sibling, 0 replies; 8+ messages in thread
From: David Marchand @ 2024-07-12 7:57 UTC (permalink / raw)
To: Srujana Challa
Cc: dev, maxime.coquelin, chenbox, Jerin Jacob,
Nithin Kumar Dabilpuram, Vamsi Krishna Attunuru
On Thu, Jul 11, 2024 at 7:46 PM Srujana Challa <schalla@marvell.com> wrote:
> > > +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.
> It crossed 100 line boundary actually.
Ah indeed, it is crossed with "use_va".
But it won't be the case when passing dev.
So please don't touch original indent after applying my suggestion.
--
David Marchand
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [EXTERNAL] Re: [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA
2024-07-11 17:46 ` [EXTERNAL] " Srujana Challa
2024-07-12 7:57 ` David Marchand
@ 2024-07-12 11:30 ` David Marchand
2024-07-12 11:51 ` Srujana Challa
1 sibling, 1 reply; 8+ messages in thread
From: David Marchand @ 2024-07-12 11:30 UTC (permalink / raw)
To: Srujana Challa
Cc: dev, maxime.coquelin, chenbox, Jerin Jacob,
Nithin Kumar Dabilpuram, Vamsi Krishna Attunuru
On Thu, Jul 11, 2024 at 7:46 PM Srujana Challa <schalla@marvell.com> wrote:
> > -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?
> Made this change to ensure that rte_mem_iova2virt() is not called when the IOVA mode is RTE_IOVA_DC
> by any chance.
Just repeating what I replied in the other thread as I see it was
suggested by Jerin.
It is not possible iova mode == RTE_IOVA_DC.
--
David Marchand
^ permalink raw reply [flat|nested] 8+ messages in thread* RE: [EXTERNAL] Re: [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA
2024-07-12 11:30 ` David Marchand
@ 2024-07-12 11:51 ` Srujana Challa
0 siblings, 0 replies; 8+ messages in thread
From: Srujana Challa @ 2024-07-12 11:51 UTC (permalink / raw)
To: David Marchand
Cc: dev, maxime.coquelin, chenbox, Jerin Jacob,
Nithin Kumar Dabilpuram, Vamsi Krishna Attunuru
> On Thu, Jul 11, 2024 at 7:46 PM Srujana Challa <schalla@marvell.com> wrote:
> > > -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?
> > Made this change to ensure that rte_mem_iova2virt() is not called when
> > the IOVA mode is RTE_IOVA_DC by any chance.
>
> Just repeating what I replied in the other thread as I see it was suggested by
> Jerin.
> It is not possible iova mode == RTE_IOVA_DC.
Thank you for the clarification. I’ll incorporate the suggested changes in the next version.
>
>
> --
> David Marchand
^ permalink raw reply [flat|nested] 8+ messages in thread
* [PATCH v3] net/virtio_user: fix cq descriptor conversion with non vDPA backend
2024-07-11 12:44 [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA Srujana Challa
2024-07-11 15:02 ` David Marchand
@ 2024-07-12 12:36 ` Srujana Challa
2024-07-12 14:24 ` David Marchand
1 sibling, 1 reply; 8+ messages in thread
From: Srujana Challa @ 2024-07-12 12:36 UTC (permalink / raw)
To: dev, maxime.coquelin, chenbox
Cc: david.marchand, jerinj, ndabilpuram, vattunuru, schalla
This patch modifies the code to convert descriptor buffer IOVA
addresses to virtual addresses only when use_va flag is false.
This patch fixes segmentation fault with vhost-user backend.
Fixes: 67e9e504dae2 ("net/virtio_user: convert cq descriptor IOVA address to Virtual address")
Reported-by: David Marchand <david.marchand@redhat.com>
Signed-off-by: Srujana Challa <schalla@marvell.com>
---
v3:
- Addressed the review comments from David Marchand.
v2:
- Added Reported-by tag.
.../net/virtio/virtio_user/virtio_user_dev.c | 20 +++++++++----------
1 file changed, 10 insertions(+), 10 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..48b872524a 100644
--- a/drivers/net/virtio/virtio_user/virtio_user_dev.c
+++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c
@@ -905,9 +905,9 @@ 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(struct virtio_user_dev *dev, rte_iova_t iova)
{
- if (rte_eal_iova_mode() == RTE_IOVA_VA)
+ if (rte_eal_iova_mode() == RTE_IOVA_VA || dev->hw.use_va)
return (void *)(uintptr_t)iova;
else
return rte_mem_iova2virt(iova);
@@ -938,18 +938,18 @@ 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(dev, vring->desc[idx_hdr].addr);
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(dev, vring->desc[idx_data].addr);
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(dev, vring->desc[idx_data].addr);
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 +962,7 @@ 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(dev, vring->desc[idx_status].addr) = status;
return n_descs;
}
@@ -1004,18 +1004,18 @@ 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(dev, vring->desc[idx_hdr].addr);
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(dev, vring->desc[idx_data].addr);
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(dev, vring->desc[idx_data].addr);
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 +1028,7 @@ 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(dev, vring->desc[idx_status].addr) = status;
/* Update used descriptor */
vring->desc[idx_hdr].id = vring->desc[idx_status].id;
--
2.25.1
^ permalink raw reply [flat|nested] 8+ messages in thread* Re: [PATCH v3] net/virtio_user: fix cq descriptor conversion with non vDPA backend
2024-07-12 12:36 ` [PATCH v3] net/virtio_user: fix cq descriptor conversion with non vDPA backend Srujana Challa
@ 2024-07-12 14:24 ` David Marchand
0 siblings, 0 replies; 8+ messages in thread
From: David Marchand @ 2024-07-12 14:24 UTC (permalink / raw)
To: Srujana Challa
Cc: dev, maxime.coquelin, chenbox, jerinj, ndabilpuram, vattunuru
On Fri, Jul 12, 2024 at 2:36 PM Srujana Challa <schalla@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 fixes segmentation fault with vhost-user backend.
>
> Fixes: 67e9e504dae2 ("net/virtio_user: convert cq descriptor IOVA address to Virtual address")
This is not the sha1 for the commit in the main repo, I updated it.
>
> Reported-by: David Marchand <david.marchand@redhat.com>
> Signed-off-by: Srujana Challa <schalla@marvell.com>
The reported issue is fixed in OVS unit tests.
Applied, thanks.
--
David Marchand
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2024-07-12 14:25 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-07-11 12:44 [PATCH v2] net/virtio_user: fix issue with converting cq descriptor IOVA address to VA Srujana Challa
2024-07-11 15:02 ` David Marchand
2024-07-11 17:46 ` [EXTERNAL] " Srujana Challa
2024-07-12 7:57 ` David Marchand
2024-07-12 11:30 ` David Marchand
2024-07-12 11:51 ` Srujana Challa
2024-07-12 12:36 ` [PATCH v3] net/virtio_user: fix cq descriptor conversion with non vDPA backend Srujana Challa
2024-07-12 14:24 ` David Marchand
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).