Skip to contentSkip to conversation

vhost: fix vdpa driver multi-queue initialization failure

PATCH NAK Hao Chen 2 participants 5 messages Jan 28, 2023 – Feb 8, 2023

lore
Share & tools
lore

Hao Chen

author
#0 lore raw

PATCH 1/1 vhost: fix vdpa driver multi-queue initialization failure

When 'virtio_is_Ready' returns 1, 'vdpa_dev->ops->dev_conf' will be called, and the vdpa driver called 'rte_vhost_get_vhost_vring' to get the vring addr info from 'vhost_devices->virtqueue[]'. virtio_is_ready's nr_vring is VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY(2), multi-queue's nr_vring is greater than 2.
Only 'vhost_devices->virtqueue[0]' and 'vhost_devices->virtqueue[1]' are obtained in the for loop. Other queues of multiple queues cannot obtain the corresponding 'vhost_devices->virtqueue[i]', which will cause 'vdpa_dev->ops->dev_conf' to obtain QEMU vring addr information from 'vhost_devices->virtqueue[i]' failed.
Here, nr_ving is modified to dev->nr_vring, so that multiple queues can obtain the corresponding 'vhost_devices->virtqueue[i]'.

---
 lib/vhost/vhost_user.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/vhost/vhost_user.c +1 -1
index 9902ae9944..9249eafc06 100644
@@ -1473,7 +1473,7 @@ virtio_is_ready(struct virtio_net *dev)
 	if (dev->nr_vring < nr_vring)
 		return 0;
 
-	for (i = 0; i < nr_vring; i++) {
+	for (i = 0; i < dev->nr_vring; i++) {
 		vq = dev->virtqueue[i];
 
 		if (!vq_is_ready(dev, vq))
Signature · 1 line
2.27.0
author
#1 lore raw

PATCH 1/1 vhost: fix vdpa driver multi-queue initialization failure

When 'virtio_is_Ready' returns 1, 'vdpa_dev->ops->dev_conf' will be called, and the vdpa driver called 'rte_vhost_get_vhost_vring' to get the vring addr info from 'vhost_devices->virtqueue[]'. virtio_is_ready's nr_vring is VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY(2), multi-queue's nr_vring is greater than 2.
Only 'vhost_devices->virtqueue[0]' and 'vhost_devices->virtqueue[1]' are obtained in the for loop. Other queues of multiple queues cannot obtain the corresponding 'vhost_devices->virtqueue[i]', which will cause 'vdpa_dev->ops->dev_conf' to obtain QEMU vring addr information from 'vhost_devices->virtqueue[i]' failed.
Here, nr_ving is modified to dev->nr_vring, so that multiple queues can obtain the corresponding 'vhost_devices->virtqueue[i]'.

---
 lib/vhost/vhost_user.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/vhost/vhost_user.c +1 -1
index 9902ae9944..9249eafc06 100644
@@ -1473,7 +1473,7 @@ virtio_is_ready(struct virtio_net *dev)
 	if (dev->nr_vring < nr_vring)
 		return 0;
 
-	for (i = 0; i < nr_vring; i++) {
+	for (i = 0; i < dev->nr_vring; i++) {
 		vq = dev->virtqueue[i];
 
 		if (!vq_is_ready(dev, vq))
Signature · 1 line
2.27.0
#2 lore raw
Quoted 23 lines · Hao Chen

When 'virtio_is_Ready' returns 1, 'vdpa_dev->ops->dev_conf' will be called, and the vdpa driver called 'rte_vhost_get_vhost_vring' to get the vring addr info from 'vhost_devices->virtqueue[]'. virtio_is_ready's nr_vring is VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY(2), multi-queue's nr_vring is greater than 2.
Only 'vhost_devices->virtqueue[0]' and 'vhost_devices->virtqueue[1]' are obtained in the for loop. Other queues of multiple queues cannot obtain the corresponding 'vhost_devices->virtqueue[i]', which will cause 'vdpa_dev->ops->dev_conf' to obtain QEMU vring addr information from 'vhost_devices->virtqueue[i]' failed.
Here, nr_ving is modified to dev->nr_vring, so that multiple queues can obtain the corresponding 'vhost_devices->virtqueue[i]'.

---
  lib/vhost/vhost_user.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
lib/vhost/vhost_user.c +0 -0
index 9902ae9944..9249eafc06 100644

lib/vhost/vhost_user.c virtio_is_ready(struct virtio_net *dev) · line 1479

Quoted 8 lines
lib/vhost/vhost_user.c +1 -1
@@ -1473,7 +1473,7 @@ virtio_is_ready(struct virtio_net *dev)
  	if (dev->nr_vring < nr_vring)
  		return 0;
 
-	for (i = 0; i < nr_vring; i++) {
+	for (i = 0; i < dev->nr_vring; i++) {
  		vq = dev->virtqueue[i];
 
  		if (!vq_is_ready(dev, vq))

Nack.

We can start processing the vrings as soon as the first queue pair is initialized. I think you may want to rely on vdpa_dev->ops->set_vring_state to initialize other vrings.

What is the vDPA driver/device affected by this issue?

Regards,
Maxime

#3 lore raw
In reply to Hao Chen
Quoted 23 lines · Hao Chen

When 'virtio_is_Ready' returns 1, 'vdpa_dev->ops->dev_conf' will be called, and the vdpa driver called 'rte_vhost_get_vhost_vring' to get the vring addr info from 'vhost_devices->virtqueue[]'. virtio_is_ready's nr_vring is VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY(2), multi-queue's nr_vring is greater than 2.
Only 'vhost_devices->virtqueue[0]' and 'vhost_devices->virtqueue[1]' are obtained in the for loop. Other queues of multiple queues cannot obtain the corresponding 'vhost_devices->virtqueue[i]', which will cause 'vdpa_dev->ops->dev_conf' to obtain QEMU vring addr information from 'vhost_devices->virtqueue[i]' failed.
Here, nr_ving is modified to dev->nr_vring, so that multiple queues can obtain the corresponding 'vhost_devices->virtqueue[i]'.

---
  lib/vhost/vhost_user.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
lib/vhost/vhost_user.c +0 -0
index 9902ae9944..9249eafc06 100644

lib/vhost/vhost_user.c virtio_is_ready(struct virtio_net *dev) · line 1479

Quoted 8 lines
lib/vhost/vhost_user.c +1 -1
@@ -1473,7 +1473,7 @@ virtio_is_ready(struct virtio_net *dev)
  	if (dev->nr_vring < nr_vring)
  		return 0;
 
-	for (i = 0; i < nr_vring; i++) {
+	for (i = 0; i < dev->nr_vring; i++) {
  		vq = dev->virtqueue[i];
 
  		if (!vq_is_ready(dev, vq))

(reiterate my reply since the patch was posted twice, and I replied to the one discarded in PW)

Nack.

We can start processing the vrings as soon as the first queue pair is initialized. I think you may want to rely on vdpa_dev->ops->set_vring_state to initialize other vrings.

What is the vDPA driver/device affected by this issue?

Regards,
Maxime

author
#4 lore raw
Quoted 36 lines · Maxime Coquelin
Quoted 23 lines · Hao Chen

When 'virtio_is_Ready' returns 1, 'vdpa_dev->ops->dev_conf' will be called, and the vdpa driver called 'rte_vhost_get_vhost_vring' to get the vring addr info from 'vhost_devices->virtqueue[]'. virtio_is_ready's nr_vring is VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY(2), multi-queue's nr_vring is greater than 2.
Only 'vhost_devices->virtqueue[0]' and 'vhost_devices->virtqueue[1]' are obtained in the for loop. Other queues of multiple queues cannot obtain the corresponding 'vhost_devices->virtqueue[i]', which will cause 'vdpa_dev->ops->dev_conf' to obtain QEMU vring addr information from 'vhost_devices->virtqueue[i]' failed.
Here, nr_ving is modified to dev->nr_vring, so that multiple queues can obtain the corresponding 'vhost_devices->virtqueue[i]'.

---
  lib/vhost/vhost_user.c | 2 +-

  1 file changed, 1 insertion(+), 1 deletion(-)
lib/vhost/vhost_user.c +0 -0
index 9902ae9944..9249eafc06 100644

lib/vhost/vhost_user.c virtio_is_ready(struct virtio_net *dev) · line 1478

Quoted 6 lines
lib/vhost/vhost_user.c +1 -0
@@ -1473,7 +1473,7 @@ virtio_is_ready(struct virtio_net *dev)
       if (dev->nr_vring < nr_vring)
           return 0;
   -    for (i = 0; i < nr_vring; i++) {
+    for (i = 0; i < dev->nr_vring; i++) {
           vq = dev->virtqueue[i];
             if (!vq_is_ready(dev, vq))

(reiterate my reply since the patch was posted twice, and I replied to the one discarded in PW)

Quoted 6 lines

Nack.

We can start processing the vrings as soon as the first queue pair is initialized. I think you may want to rely on vdpa_dev->ops->set_vring_state to initialize other vrings.

Does this mean that the queue should be initialized and enabled in vdpa_dev->ops->set_vring_state, not in vdpa_dev->ops->dev_conf?

Quoted 1 line

What is the vDPA driver/device affected by this issue?

After tracking the code, I found that the dpdk-vdpa example's log shows the calling order of multiple queues as follow:
"
VHOST_USER_SET_VRING_ADDR
VHOST_USER_SET_VRING_KICK
vdpa_dev->ops->set_vring_state  (vq_idx=0)
VHOST_USER_SET_VRING_CALL
vdpa_dev->ops->set_vring_state  (vq_idx=0)
...
VHOST_USER_SET_VRING_ADDR
VHOST_USER_SET_VRING_KICK
vdpa_dev->ops->set_vring_state  (vq_idx=1)
VHOST_USER_SET_VRING_CALL
vdpa_dev->ops->set_vring_state  (vq_idx=1)
"
then virtio_is_ready return 1, vdpa_dev->ops->dev_conf is called.
After that, other queues will call vdpa_dev->ops->set_vring_state (vq_idx=i)

The dev->virtqueue[]->desc/available/used is assigned in VHOST_USER_SET_VRING_KICK's translate_ring_addresses() , so vdpa driver should not initialize all queues in vdpa_dev->ops->dev_conf , otherwise other queue's dev->virtqueue[i]->desc/available/used are NULL , vring's hpa to gpa will fail by rte_vhost_get_vhost_vring(). dev->virtqueue[i]->desc/available/used should be accessed after queue(i)'s VHOST_USER_SET_VRING_KICK is called.

I see that all of ifc/sfc vdpa driver's queues acessed all queues's dev->virtqueue[i]->desc/available/used in vdpa_dev->ops->dev_conf, maybe they should move the code?

Quoted 4 lines

Regards,
Maxime