vdpa: dont check vhost_vdpa->suspended when unsupported
PATCH 1/1 vdpa: dont check vhost_vdpa->suspended when unsupported
When read the state of a virtqueue, vhost_vdpa need to check whether the device is suspended.
This commit verifies whether VHOST_BACKEND_F_SUSPEND is negotiated when checking vhost_vdpa->suspended.
--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
hw/virtio/vhost-vdpa.c +1 -1
@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest.
Signature · 1 line
2.39.1
Quoted 7 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.
I'll add: Otherwise, qemu prints XXX error message.
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call virtio_queue_restore_last_avail_idx here? Communicate that the backend cannot suspend so vhost.c can print a better error message?
Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 9 lines · Eugenio Perez Martin
Quoted 6 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.I'll add: Otherwise, qemu prints XXX error message.
Quoted 7 lines
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere? Communicate that the backend cannot suspend sovhost.ccan print a better error message?
I agree it is better not to return an error.
Instead if neither the device does not support _F_SUSPEND nor failed to suspend, I think vhost_vdpa should try to read the last_avail_idx from
the device anyway. We can print an error message here,
like: failed to suspend, the value may not reliable.
Thanks
Quoted 24 lines
Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 20 lines · Lingshan
Quoted 9 lines · Eugenio Perez Martin
Quoted 6 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.I'll add: Otherwise, qemu prints XXX error message.
Quoted 7 lines
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere? Communicate that the backend cannot suspend sovhost.ccan print a better error message?
Quoted 8 lines
I agree it is better not to return an error.
Instead if neither the device does not support
_F_SUSPENDnor failed to suspend,
I thinkvhost_vdpashould try to read thelast_avail_idxfrom the device anyway. We can print an error message here,
like: failed to suspend, the value may not reliable.
We need to drop that value if it is not reliable. If the device keeps using descriptors, used_idx may increase beyond avail_idx, so the usual is to just restore whatever used_idx value the guest had at stop.
Thanks!
Quoted 26 lines
Thanks
Quoted 24 lines
Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 28 lines · Eugenio Perez Martin
Quoted 19 lines · Lingshan
Quoted 9 lines · Eugenio Perez Martin
Quoted 6 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.I'll add: Otherwise, qemu prints XXX error message.
Quoted 7 lines
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere? Communicate that the backend cannot suspend sovhost.ccan print a better error message?Quoted 8 lines
I agree it is better not to return an error.
Instead if neither the device does not support
_F_SUSPENDnor failed to suspend,
I thinkvhost_vdpashould try to read thelast_avail_idxfrom the device anyway. We can print an error message here,
like: failed to suspend, the value may not reliable.
Quoted 4 lines
We need to drop that value if it is not reliable. If the device keeps using descriptors,
used_idxmay increase beyondavail_idx, so the usual is to just restore whateverused_idxvalue the guest had at stop.
This interface is used to read the last_avail_idx, the ideal case is to fetch it after device has been suspended.
If the device is not suspended, the value may be unreliable because the device may keep consuming descriptors.
However at that moment, the guest may be freezed as well, so the guest wouldn't supply more available descriptors to the device.
I think that means the last_avail_idx may not be reliable but still safe, better than read nothing. Because the last_avail_idx always lags behind the in-guest avail_idx.
I am not sure we can restore last_avail_idx with last_used_idx, there is always a delta between them.
Thanks
Quoted 26 lines
Thanks!
Quoted 24 lines
Thanks
Quoted 23 lines
Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 36 lines · Lingshan
Quoted 28 lines · Eugenio Perez Martin
Quoted 19 lines · Lingshan
Quoted 9 lines · Eugenio Perez Martin
Quoted 6 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.I'll add: Otherwise, qemu prints XXX error message.
Quoted 7 lines
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere? Communicate that the backend cannot suspend sovhost.ccan print a better error message?Quoted 8 lines
I agree it is better not to return an error.
Instead if neither the device does not support
_F_SUSPENDnor failed to suspend,
I thinkvhost_vdpashould try to read thelast_avail_idxfrom the device anyway. We can print an error message here,
like: failed to suspend, the value may not reliable.Quoted 4 lines
We need to drop that value if it is not reliable. If the device keeps using descriptors,
used_idxmay increase beyondavail_idx, so the usual is to just restore whateverused_idxvalue the guest had at stop.
Quoted 8 lines
This interface is used to read the
last_avail_idx, the ideal case is to fetch it after device has been suspended.If the device is not suspended, the value may be unreliable because the device may keep consuming descriptors.
However at that moment, the guest may be freezed as well, so the guest wouldn't supply more available descriptors to the device.
Actually, if we cannot suspend the device we reset it, as we cannot afford to modify used_idx.
I'm thinking now that your original idea was way better to be honest. To not call vhost_get_vring_base in case the VM is shutting down and we're not migrating seems to fit the situation way better. We save an ioctl / potential device call for nothing.
Quoted 4 lines
I think that means the
last_avail_idxmay not be reliable but still safe, better than read nothing. Because thelast_avail_idxalways lags behind the in-guestavail_idx.
Assuming we're migrating and we don't either reset or suspend the device
* guest set avail_idx=3
* device fetch last_avail_idx=3
* guest set avail_idx=6
* VM_SUSPEND
* we call unreliable get_vring_base, and obtain device state 3.
* device is not reset, so it reads guest's avail_idx = 6. It can process the descriptors in avail_idx position 3, 4 and 5, and write that used_idx.
* virtio_load detects that inconsistency, as (uint)last_avail_idx - (uint)used_idx > vring size.
So I think we need to keep printing an error and recovery from the guest as a fallback.
Quoted 4 lines
I am not sure we can restore
last_avail_idxwithlast_used_idx, there is always
a delta between them.
Yes, we assume it is safe to process these descriptors again or that they'll be migrated when that is supported.
In any case, if a device does not work with this, we should be more restrictive, not less.
Does it make sense to you?
Thanks!
Quoted 29 lines
Thanks
Quoted 26 lines
Thanks!
Quoted 24 lines
Thanks
Quoted 23 lines
Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 43 lines · Eugenio Perez Martin
Quoted 34 lines · Lingshan
Quoted 27 lines · Eugenio Perez Martin
Quoted 18 lines · Lingshan
Quoted 9 lines · Eugenio Perez Martin
Quoted 6 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.I'll add: Otherwise, qemu prints XXX error message.
Quoted 7 lines
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere? Communicate that the backend cannot suspend sovhost.ccan print a better error message?Quoted 8 lines
I agree it is better not to return an error.
Instead if neither the device does not support
_F_SUSPENDnor failed to suspend,
I thinkvhost_vdpashould try to read thelast_avail_idxfrom the device anyway. We can print an error message here,
like: failed to suspend, the value may not reliable.Quoted 4 lines
We need to drop that value if it is not reliable. If the device keeps using descriptors,
used_idxmay increase beyondavail_idx, so the usual is to just restore whateverused_idxvalue the guest had at stop.Quoted 8 lines
This interface is used to read the
last_avail_idx, the ideal case is to fetch it after device has been suspended.If the device is not suspended, the value may be unreliable because the device may keep consuming descriptors.
However at that moment, the guest may be freezed as well, so the guest wouldn't supply more available descriptors to the device.
Quoted 7 lines
Actually, if we cannot suspend the device we reset it, as we cannot afford to modify
used_idx.I'm thinking now that your original idea was way better to be honest. To not call
vhost_get_vring_basein case the VM is shutting down and we're not migrating seems to fit the situation way better. We save an ioctl / potential device call for nothing.
agree, but I failed to find a code patch indicting the VM is in "shutting down" status, and it may be overkill to add a new status field which only used here.
So maybe a fix like this:
1) if the device does not support _F_SUSPEND:
call virtio_queue_restore_last_avail_idx(dev, state->index)
2) if the device support _F_SUSPEND but failed to suspend:
return -1 and vhost_virtqueue_stop() will call virtio_queue_restore_last_avail_idx() in the original code path.
Does this look good to you?
Quoted 17 lines
Quoted 5 lines
I think that means the
last_avail_idxmay not be reliable but still safe, better than read nothing. Because thelast_avail_idxalways lags behind the in-guestavail_idx.Assuming we're migrating and we don't either reset or suspend the device
* guest set avail_idx=3
* device fetch last_avail_idx=3
* guest set avail_idx=6
*VM_SUSPEND
* we call unreliableget_vring_base, and obtain device state 3.
* device is not reset, so it reads guest'savail_idx= 6. It can process the descriptors inavail_idxposition 3, 4 and 5, and write thatused_idx.
*virtio_loaddetects that inconsistency, as (uint)last_avail_idx - (uint)used_idx > vring size.
Quoted 2 lines
So I think we need to keep printing an error and recovery from the guest as a fallback.
so true, I think that is virtio_queue_restore_last_avail_idx()
Quoted 5 lines
Quoted 5 lines
I am not sure we can restore
last_avail_idxwithlast_used_idx, there is always
a delta between them.
Quoted 7 lines
Yes, we assume it is safe to process these descriptors again or that they'll be migrated when that is supported.
In any case, if a device does not work with this, we should be more restrictive, not less.
Does it make sense to you?
Yes, so if the device doesn't support _F_SUSPEND, we call virtio_queue_restore_last_avail_idx() in vhost_vdpa_get_vring_base(), look good?
Thanks
Quoted 31 lines
Thanks!
Quoted 28 lines
Thanks
Quoted 26 lines
Thanks!
Quoted 24 lines
Thanks
Quoted 23 lines
Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 50 lines · Lingshan
Quoted 46 lines · Eugenio Perez Martin
Quoted 34 lines · Lingshan
Quoted 27 lines · Eugenio Perez Martin
Quoted 18 lines · Lingshan
Quoted 9 lines · Eugenio Perez Martin
Quoted 6 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.I'll add: Otherwise, qemu prints XXX error message.
Quoted 7 lines
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere? Communicate that the backend cannot suspend sovhost.ccan print a better error message?Quoted 8 lines
I agree it is better not to return an error.
Instead if neither the device does not support
_F_SUSPENDnor failed to suspend,
I thinkvhost_vdpashould try to read thelast_avail_idxfrom the device anyway. We can print an error message here,
like: failed to suspend, the value may not reliable.Quoted 4 lines
We need to drop that value if it is not reliable. If the device keeps using descriptors,
used_idxmay increase beyondavail_idx, so the usual is to just restore whateverused_idxvalue the guest had at stop.Quoted 8 lines
This interface is used to read the
last_avail_idx, the ideal case is to fetch it after device has been suspended.If the device is not suspended, the value may be unreliable because the device may keep consuming descriptors.
However at that moment, the guest may be freezed as well, so the guest wouldn't supply more available descriptors to the device.Actually, if we cannot suspend the device we reset it, as we cannot afford to modify
used_idx.
Quoted 8 lines
Quoted 4 lines
I'm thinking now that your original idea was way better to be honest. To not call
vhost_get_vring_basein case the VM is shutting down and we're not migrating seems to fit the situation way better. We save an ioctl / potential device call for nothing.agree, but I failed to find a code patch indicting the VM is in "shutting down" status, and it may be overkill to add a new status field which only used here.
Since vhost is stopped before migration of the net device state, I think runstate_check returns differently depending on the case. If migrating, runstate_check(RUN_STATE_SHUTDOWN) would return false. If shutdown without migration, it would return true.
I didn't test it though.
Thanks!
Quoted 78 lines
So maybe a fix like this:
1) if the device does not support_F_SUSPEND:
callvirtio_queue_restore_last_avail_idx(dev, state->index)
2) if the device support_F_SUSPENDbut failed to suspend:
return -1 andvhost_virtqueue_stop()will callvirtio_queue_restore_last_avail_idx()
in the original code path.Does this look good to you?
Quoted 17 lines
Quoted 5 lines
I think that means the
last_avail_idxmay not be reliable but still safe, better than read nothing. Because thelast_avail_idxalways lags behind the in-guestavail_idx.Assuming we're migrating and we don't either reset or suspend the device
* guest set avail_idx=3
* device fetch last_avail_idx=3
* guest set avail_idx=6
*VM_SUSPEND
* we call unreliableget_vring_base, and obtain device state 3.
* device is not reset, so it reads guest'savail_idx= 6. It can process the descriptors inavail_idxposition 3, 4 and 5, and write thatused_idx.
*virtio_loaddetects that inconsistency, as (uint)last_avail_idx - (uint)used_idx > vring size.Quoted 2 lines
So I think we need to keep printing an error and recovery from the guest as a fallback.
so true, I think that is
virtio_queue_restore_last_avail_idx()Quoted 5 lines
Quoted 5 lines
I am not sure we can restore
last_avail_idxwithlast_used_idx, there is always
a delta between them.Quoted 7 lines
Yes, we assume it is safe to process these descriptors again or that they'll be migrated when that is supported.
In any case, if a device does not work with this, we should be more restrictive, not less.
Does it make sense to you?
Yes, so if the device doesn't support
_F_SUSPEND, we callvirtio_queue_restore_last_avail_idx()
invhost_vdpa_get_vring_base(), look good?Thanks
Quoted 31 lines
Thanks!
Quoted 28 lines
Thanks
Quoted 26 lines
Thanks!
Quoted 24 lines
Thanks
Quoted 23 lines
Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 10 lines · Eugenio Perez Martin
Quoted 7 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.
Quoted 8 lines
I'll add: Otherwise, qemu prints XXX error message.
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere?
It should be the duty of the caller:
E.g in vhost_virtqueue_stop() we had:
=> r = dev->vhost_ops->vhost_get_vring_base(dev, &state);
if (r < 0) {
VHOST_OPS_DEBUG(r, "vhost VQ %u ring restore failed: %d", idx, r);
/* Connection to the backend is broken, so let's sync internal
* last avail idx to the device used idx.
*/
virtio_queue_restore_last_avail_idx(vdev, idx);
} else {
virtio_queue_set_last_avail_idx(vdev, idx, state.num);Thansk
Quoted 27 lines
Communicate that the backend
cannot suspend sovhost.ccan print a better error message?Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1
Quoted 21 lines · Jason Wang
Quoted 7 lines · Eugenio Perez Martin
Quoted 6 lines · Zhu Lingshan
When read the state of a virtqueue,
vhost_vdpaneed to check whether the device is suspended.This commit verifies whether
VHOST_BACKEND_F_SUSPENDis negotiated when checkingvhost_vdpa->suspended.Quoted 8 lines
I'll add: Otherwise, qemu prints XXX error message.
On second thought, not returning an error prevents the caller (vhost.c:vhost_virtqueue_stop) from recovering used idx from the guest.
I'm not sure about the right fix for this, should we call
virtio_queue_restore_last_avail_idxhere?It should be the duty of the caller:
E.g in
vhost_virtqueue_stop()we had:
Quoted 9 lines
=> r = dev->vhost_ops->vhost_get_vring_base(dev, &state);
if (r < 0) {
VHOST_OPS_DEBUG(r, "vhost VQ %u ring restore failed: %d", idx, r);
/* Connection to the backend is broken, so let's sync internal
* last avail idx to the device used idx.
*/
virtio_queue_restore_last_avail_idx(vdev, idx);
} else {
virtio_queue_set_last_avail_idx(vdev, idx, state.num);I agree call virtio_queue_restore_last_avail_idx here in the caller is better. However I found this debug message is confused:
_get_vring_base() read the queue state, it does not restore anything,
and if failed to read the queue state, it invokes virtio_queue_restore_last_avail_idx() to recover, still works, handled the problem internally.
So how about we remove this debug message or change it to:
restore last_avail_idx of vhost vq N from vring used_idx.
Thanks
Quoted 29 lines
Thansk
Quoted 26 lines
Communicate that the backend
cannot suspend sovhost.ccan print a better error message?Thanks!
Quoted 21 lines
Signed-off-by Zhu Lingshan <[email protected]>--- hw/virtio/vhost-vdpa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)hw/virtio/vhost-vdpa.c +1
-2@@ -1397,7 +1397,7 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,return 0 ; } -if (!v->suspended) { +if ((dev->backend_cap &BIT_ULL (VHOST_BACKEND_F_SUSPEND )) && (!v->suspended)) {/* * Cannot trust in value returned by device, let vhost recover used * idx from guest. -- 2.39.1