virtio: skip guest index check on device load
PATCH 1/1 virtio: skip guest index check on device load
QEMU must be careful when loading device state off migration streams to prevent a malicious source from exploiting the emulator. Overdoing these checks has the side effect of allowing a guest to "pin itself" in cloud environments by messing with state which is entirely in its control.
Similarly to what f3081539 achieved in usb_device_post_load(), this commit removes such a check from virtio_load(). Worth noting, the result of a load without this check is the same as if a guest enables a VQ with invalid indexes to begin with. That is, the virtual device is set in a broken state (by the datapath handler) and must be reset.
--- hw/virtio/virtio.c | 12 ------------ 1 file changed, 12 deletions(-)
hw/virtio/virtio.c +0 -12
@@ -3136,8 +3136,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)RCU_READ_LOCK_GUARD ();for (i =0 ; i < num; i++) {if (vdev->vq[i].vring.desc) { -uint16_t nheads; -/* * VIRTIO-1 devices migrate desc, used, and avail ring addresses so * only the region cache needs to be set up. Legacy devices need
@@ -3157,16 +3155,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)continue ; } - nheads =vring_avail_idx (&vdev->vq[i]) - vdev->vq[i].last_avail_idx; -/* Check it isn't doing strange things with descriptor numbers. */ -if (nheads > vdev->vq[i].vring.num) { -error_report ("VQ %d size 0x%x Guest index 0x%x " -"inconsistent with Host index 0x%x: delta 0x%x" , - i, vdev->vq[i].vring.num, -vring_avail_idx (&vdev->vq[i]), - vdev->vq[i].last_avail_idx, nheads); -return -1 ; - } vdev->vq[i].used_idx =vring_used_idx (&vdev->vq[i]); vdev->vq[i].shadow_avail_idx =vring_avail_idx (&vdev->vq[i]);
Signature · 1 line
2.18.4
Quoted 33 lines · Felipe Franciosi
QEMU must be careful when loading device state off migration streams to prevent a malicious source from exploiting the emulator. Overdoing these checks has the side effect of allowing a guest to "pin itself" in cloud environments by messing with state which is entirely in its control.
Similarly to what
f3081539achieved inusb_device_post_load(), this commit removes such a check fromvirtio_load(). Worth noting, the result of a load without this check is the same as if a guest enables a VQ with invalid indexes to begin with. That is, the virtual device is set in a broken state (by the datapath handler) and must be reset.Signed-off-by Felipe Franciosi <[email protected]>--- hw/virtio/virtio.c | 12 ------------ 1 file changed, 12 deletions(-)hw/virtio/virtio.c +0
-2@@ -3136,8 +3136,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)RCU_READ_LOCK_GUARD ();for (i =0 ; i < num; i++) {if (vdev->vq[i].vring.desc) { -uint16_t nheads; -/* * VIRTIO-1 devices migrate desc, used, and avail ring addresses so * only the region cache needs to be set up. Legacy devices need @@ -3157,16 +3155,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)continue ; }
hw/virtio/virtio.c int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) · line 3169
Quoted 10 lines
hw/virtio/virtio.c +0
-10- nheads =vring_avail_idx (&vdev->vq[i]) - vdev->vq[i].last_avail_idx; -/* Check it isn't doing strange things with descriptor numbers. */ -if (nheads > vdev->vq[i].vring.num) { -error_report ("VQ %d size 0x%x Guest index 0x%x " -"inconsistent with Host index 0x%x: delta 0x%x" , - i, vdev->vq[i].vring.num, -vring_avail_idx (&vdev->vq[i]), - vdev->vq[i].last_avail_idx, nheads); -return -1 ; - }
Michael, the commit that introduced this check seems to have been for debugging rather than to prevent a QEMU crash, so this removing the check may be safe:
commit 258dc7c96bb4b7ca71d5bee811e73933310e168c
Author: Michael S. Tsirkin <[email protected]>
Date: Sun Oct 17 20:23:48 2010 +0200
virtio: sanity-check available index
Checking available index upon load instead of only when vm is running makes is easier to debug failures.
Felipe: Did you audit the code to make sure the invalid avail_idx value and the fields it is propagated to (e.g. shadow_avail_idx) are always used in a safe way?
Quoted 53 lines · Stefan Hajnoczi
Quoted 33 lines · Felipe Franciosi
QEMU must be careful when loading device state off migration streams to prevent a malicious source from exploiting the emulator. Overdoing these checks has the side effect of allowing a guest to "pin itself" in cloud environments by messing with state which is entirely in its control.
Similarly to what
f3081539achieved inusb_device_post_load(), this commit removes such a check fromvirtio_load(). Worth noting, the result of a load without this check is the same as if a guest enables a VQ with invalid indexes to begin with. That is, the virtual device is set in a broken state (by the datapath handler) and must be reset.Signed-off-by Felipe Franciosi <[email protected]>--- hw/virtio/virtio.c | 12 ------------ 1 file changed, 12 deletions(-)hw/virtio/virtio.c +0
-2@@ -3136,8 +3136,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)RCU_READ_LOCK_GUARD ();for (i =0 ; i < num; i++) {if (vdev->vq[i].vring.desc) { -uint16_t nheads; -/* * VIRTIO-1 devices migrate desc, used, and avail ring addresses so * only the region cache needs to be set up. Legacy devices need @@ -3157,16 +3155,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)continue ; }hw/virtio/virtio.c int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) · line 3169
Quoted 10 lines
hw/virtio/virtio.c +0
-10- nheads =vring_avail_idx (&vdev->vq[i]) - vdev->vq[i].last_avail_idx; -/* Check it isn't doing strange things with descriptor numbers. */ -if (nheads > vdev->vq[i].vring.num) { -error_report ("VQ %d size 0x%x Guest index 0x%x " -"inconsistent with Host index 0x%x: delta 0x%x" , - i, vdev->vq[i].vring.num, -vring_avail_idx (&vdev->vq[i]), - vdev->vq[i].last_avail_idx, nheads); -return -1 ; - }Michael, the commit that introduced this check seems to have been for debugging rather than to prevent a QEMU crash, so this removing the check may be safe:
commit
258dc7c96bb4b7ca71d5bee811e73933310e168c
Author: Michael S. Tsirkin <[email protected]>
Date: Sun Oct 17 20:23:48 2010 +0200
Quoted 5 lines
virtio: sanity-check available index
Checking available index upon load instead of only when vm is running makes is easier to debug failures.
Agreed. Given this, let's keep the message around, just with LOG_GUEST_ERROR ?
Quoted 3 lines
Felipe: Did you audit the code to make sure the invalid
avail_idxvalue and the fields it is propagated to (e.g.shadow_avail_idx) are always used in a safe way?
Quoted 58 lines · Michael S. Tsirkin
Quoted 53 lines · Stefan Hajnoczi
Quoted 33 lines · Felipe Franciosi
QEMU must be careful when loading device state off migration streams to prevent a malicious source from exploiting the emulator. Overdoing these checks has the side effect of allowing a guest to "pin itself" in cloud environments by messing with state which is entirely in its control.
Similarly to what
f3081539achieved inusb_device_post_load(), this commit removes such a check fromvirtio_load(). Worth noting, the result of a load without this check is the same as if a guest enables a VQ with invalid indexes to begin with. That is, the virtual device is set in a broken state (by the datapath handler) and must be reset.Signed-off-by Felipe Franciosi <[email protected]>---
hw/virtio/virtio.c| 12 ------------1 file changed, 12 deletions(-)hw/virtio/virtio.c +0
-2@@ -3136,8 +3136,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)RCU_READ_LOCK_GUARD ();for (i =0 ; i < num; i++) {if (vdev->vq[i].vring.desc) { -uint16_t nheads; -/* * VIRTIO-1 devices migrate desc, used, and avail ring addresses so * only the region cache needs to be set up. Legacy devices need @@ -3157,16 +3155,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)continue ; }hw/virtio/virtio.c int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) · line 3169
Quoted 10 lines
hw/virtio/virtio.c +0
-10- nheads =vring_avail_idx (&vdev->vq[i]) - vdev->vq[i].last_avail_idx; -/* Check it isn't doing strange things with descriptor numbers. */ -if (nheads > vdev->vq[i].vring.num) { -error_report ("VQ %d size 0x%x Guest index 0x%x " -"inconsistent with Host index 0x%x: delta 0x%x" , - i, vdev->vq[i].vring.num, -vring_avail_idx (&vdev->vq[i]), - vdev->vq[i].last_avail_idx, nheads); -return -1 ; - }Michael, the commit that introduced this check seems to have been for debugging rather than to prevent a QEMU crash, so this removing the check may be safe:
commit
258dc7c96bb4b7ca71d5bee811e73933310e168c
Author: Michael S. Tsirkin <[email protected]>
Date: Sun Oct 17 20:23:48 2010 +0200
Quoted 8 lines
Quoted 5 lines
virtio: sanity-check available index
Checking available index upon load instead of only when vm is running makes is easier to debug failures.
Agreed. Given this, let's keep the message around, just with
LOG_GUEST_ERROR?
I thought about it. Happy to send a v2 which keeps the check and logs without throwing an error.
Separately, I'm thinking of hooking up QEMU with VRING_ERR so datapath handlers can notify QEMU that something went broken and NEEDS_RESET can be flipped on the status register, possibly along a configuration interrupt. I can see libvhost-user supports that, but are there any reasons QEMU doesn't do this already?
Quoted 5 lines
Quoted 4 lines
Felipe: Did you audit the code to make sure the invalid
avail_idxvalue and the fields it is propagated to (e.g.shadow_avail_idx) are always used in a safe way?
I did it briefly. I also wrote a mock userspace driver that creates this condition in a very controlled way (so I can step half-way through setting up VQs and trigger a migration, for example). But you know how manual tests are... I may have missed something. Your expert eyes are most welcome. :)
F.
Quoted 67 lines · Felipe Franciosi
Quoted 58 lines · Michael S. Tsirkin
Quoted 53 lines · Stefan Hajnoczi
Quoted 33 lines · Felipe Franciosi
QEMU must be careful when loading device state off migration streams to prevent a malicious source from exploiting the emulator. Overdoing these checks has the side effect of allowing a guest to "pin itself" in cloud environments by messing with state which is entirely in its control.
Similarly to what
f3081539achieved inusb_device_post_load(), this commit removes such a check fromvirtio_load(). Worth noting, the result of a load without this check is the same as if a guest enables a VQ with invalid indexes to begin with. That is, the virtual device is set in a broken state (by the datapath handler) and must be reset.Signed-off-by Felipe Franciosi <[email protected]>---
hw/virtio/virtio.c| 12 ------------1 file changed, 12 deletions(-)hw/virtio/virtio.c +0
-2@@ -3136,8 +3136,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)RCU_READ_LOCK_GUARD ();for (i =0 ; i < num; i++) {if (vdev->vq[i].vring.desc) { -uint16_t nheads; -/* * VIRTIO-1 devices migrate desc, used, and avail ring addresses so * only the region cache needs to be set up. Legacy devices need @@ -3157,16 +3155,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)continue ; }hw/virtio/virtio.c int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) · line 3169
Quoted 10 lines
hw/virtio/virtio.c +0
-10- nheads =vring_avail_idx (&vdev->vq[i]) - vdev->vq[i].last_avail_idx; -/* Check it isn't doing strange things with descriptor numbers. */ -if (nheads > vdev->vq[i].vring.num) { -error_report ("VQ %d size 0x%x Guest index 0x%x " -"inconsistent with Host index 0x%x: delta 0x%x" , - i, vdev->vq[i].vring.num, -vring_avail_idx (&vdev->vq[i]), - vdev->vq[i].last_avail_idx, nheads); -return -1 ; - }Michael, the commit that introduced this check seems to have been for debugging rather than to prevent a QEMU crash, so this removing the check may be safe:
commit
258dc7c96bb4b7ca71d5bee811e73933310e168c
Author: Michael S. Tsirkin <[email protected]>
Date: Sun Oct 17 20:23:48 2010 +0200Quoted 8 lines
Quoted 5 lines
virtio: sanity-check available index
Checking available index upon load instead of only when vm is running makes is easier to debug failures.
Agreed. Given this, let's keep the message around, just with
LOG_GUEST_ERROR?
Quoted 8 lines
I thought about it. Happy to send a v2 which keeps the check and logs without throwing an error.
Separately, I'm thinking of hooking up QEMU with
VRING_ERRso datapath handlers can notify QEMU that something went broken andNEEDS_RESETcan be flipped on the status register, possibly along a configuration interrupt. I can see libvhost-user supports that, but are there any reasons QEMU doesn't do this already?
Mostly because guest support isn't there. That in turn isn't easy, lots of synchronization is needed within guests.
Quoted 13 lines
Quoted 5 lines
Quoted 4 lines
Felipe: Did you audit the code to make sure the invalid
avail_idxvalue and the fields it is propagated to (e.g.shadow_avail_idx) are always used in a safe way?I did it briefly. I also wrote a mock userspace driver that creates this condition in a very controlled way (so I can step half-way through setting up VQs and trigger a migration, for example). But you know how manual tests are... I may have missed something. Your expert eyes are most welcome. :)
F.
Quoted 75 lines · Michael S. Tsirkin
Quoted 70 lines · Felipe Franciosi
Quoted 58 lines · Michael S. Tsirkin
Quoted 53 lines · Stefan Hajnoczi
Quoted 33 lines · Felipe Franciosi
QEMU must be careful when loading device state off migration streams to prevent a malicious source from exploiting the emulator. Overdoing these checks has the side effect of allowing a guest to "pin itself" in cloud environments by messing with state which is entirely in its control.
Similarly to what
f3081539achieved inusb_device_post_load(), this commit removes such a check fromvirtio_load(). Worth noting, the result of a load without this check is the same as if a guest enables a VQ with invalid indexes to begin with. That is, the virtual device is set in a broken state (by the datapath handler) and must be reset.Signed-off-by Felipe Franciosi <[email protected]>---
hw/virtio/virtio.c| 12 ------------1 file changed, 12 deletions(-)hw/virtio/virtio.c +0
-2@@ -3136,8 +3136,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)RCU_READ_LOCK_GUARD ();for (i =0 ; i < num; i++) {if (vdev->vq[i].vring.desc) { -uint16_t nheads; -/* * VIRTIO-1 devices migrate desc, used, and avail ring addresses so * only the region cache needs to be set up. Legacy devices need @@ -3157,16 +3155,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)continue ; }hw/virtio/virtio.c int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) · line 3169
Quoted 10 lines
hw/virtio/virtio.c +0
-10- nheads =vring_avail_idx (&vdev->vq[i]) - vdev->vq[i].last_avail_idx; -/* Check it isn't doing strange things with descriptor numbers. */ -if (nheads > vdev->vq[i].vring.num) { -error_report ("VQ %d size 0x%x Guest index 0x%x " -"inconsistent with Host index 0x%x: delta 0x%x" , - i, vdev->vq[i].vring.num, -vring_avail_idx (&vdev->vq[i]), - vdev->vq[i].last_avail_idx, nheads); -return -1 ; - }Michael, the commit that introduced this check seems to have been for debugging rather than to prevent a QEMU crash, so this removing the check may be safe:
commit
258dc7c96bb4b7ca71d5bee811e73933310e168cAuthor: Michael S. Tsirkin <[email protected]> Date: Sun Oct 17 20:23:48 2010 +0200Quoted 8 lines
Quoted 5 lines
virtio: sanity-check available index
Checking available index upon load instead of
only when vm is running makes is easier to
debug failures.Agreed. Given this, let's keep the message around, just with
LOG_GUEST_ERROR?I thought about it. Happy to send a v2 which keeps the check and logs without throwing an error.
Quoted 8 lines
Quoted 5 lines
Separately, I'm thinking of hooking up QEMU with
VRING_ERRso datapath handlers can notify QEMU that something went broken andNEEDS_RESETcan be flipped on the status register, possibly along a configuration interrupt. I can see libvhost-user supports that, but are there any reasons QEMU doesn't do this already?Mostly because guest support isn't there. That in turn isn't easy, lots of synchronization is needed within guests.
Do you mean guest support to reset when seeing that bit in status following the configuration interrupt?
It should be safe, though. I can have a stab to see if it breaks Windows/Linux at least, and share an RFC if I get anywhere.
Unless you think it's a waste of time. Ideally guests shouldn't find themselves in this situation to begin with, and if they did, resetting would arguably just lead them into corruption again (for example). But it does provide a mechanism for QEMU to find out that the vhost backend stopped. That would help in the context of this patch.
F.
Quoted 16 lines
Quoted 15 lines
Quoted 7 lines
Quoted 6 lines
Felipe: Did you audit the code to make sure the invalid
avail_idxvalue and the fields it is propagated to (e.g.shadow_avail_idx) are always used in a safe way?I did it briefly. I also wrote a mock userspace driver that creates this condition in a very controlled way (so I can step half-way through setting up VQs and trigger a migration, for example). But you know how manual tests are... I may have missed something. Your expert eyes are most welcome. :)
F.
Quoted 90 lines · Felipe Franciosi
Quoted 75 lines · Michael S. Tsirkin
Quoted 70 lines · Felipe Franciosi
Quoted 58 lines · Michael S. Tsirkin
Quoted 53 lines · Stefan Hajnoczi
Quoted 33 lines · Felipe Franciosi
QEMU must be careful when loading device state off migration streams to prevent a malicious source from exploiting the emulator. Overdoing these checks has the side effect of allowing a guest to "pin itself" in cloud environments by messing with state which is entirely in its control.
Similarly to what
f3081539achieved inusb_device_post_load(), this commit removes such a check fromvirtio_load(). Worth noting, the result of a load without this check is the same as if a guest enables a VQ with invalid indexes to begin with. That is, the virtual device is set in a broken state (by the datapath handler) and must be reset.Signed-off-by Felipe Franciosi <[email protected]>---
hw/virtio/virtio.c| 12 ------------1 file changed, 12 deletions(-)hw/virtio/virtio.c +0
-2@@ -3136,8 +3136,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)RCU_READ_LOCK_GUARD ();for (i =0 ; i < num; i++) {if (vdev->vq[i].vring.desc) { -uint16_t nheads; -/* * VIRTIO-1 devices migrate desc, used, and avail ring addresses so * only the region cache needs to be set up. Legacy devices need @@ -3157,16 +3155,6 @@ int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)continue ; }hw/virtio/virtio.c int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id) · line 3169
Quoted 10 lines
hw/virtio/virtio.c +0
-10- nheads =vring_avail_idx (&vdev->vq[i]) - vdev->vq[i].last_avail_idx; -/* Check it isn't doing strange things with descriptor numbers. */ -if (nheads > vdev->vq[i].vring.num) { -error_report ("VQ %d size 0x%x Guest index 0x%x " -"inconsistent with Host index 0x%x: delta 0x%x" , - i, vdev->vq[i].vring.num, -vring_avail_idx (&vdev->vq[i]), - vdev->vq[i].last_avail_idx, nheads); -return -1 ; - }Michael, the commit that introduced this check seems to have been for debugging rather than to prevent a QEMU crash, so this removing the check may be safe:
commit
258dc7c96bb4b7ca71d5bee811e73933310e168cAuthor: Michael S. Tsirkin <[email protected]> Date: Sun Oct 17 20:23:48 2010 +0200Quoted 8 lines
Quoted 5 lines
virtio: sanity-check available index
Checking available index upon load instead of
only when vm is running makes is easier to
debug failures.Agreed. Given this, let's keep the message around, just with
LOG_GUEST_ERROR?I thought about it. Happy to send a v2 which keeps the check and logs without throwing an error.
Quoted 8 lines
Quoted 5 lines
Separately, I'm thinking of hooking up QEMU with
VRING_ERRso datapath handlers can notify QEMU that something went broken andNEEDS_RESETcan be flipped on the status register, possibly along a configuration interrupt. I can see libvhost-user supports that, but are there any reasons QEMU doesn't do this already?Mostly because guest support isn't there. That in turn isn't easy, lots of synchronization is needed within guests.
Do you mean guest support to reset when seeing that bit in status following the configuration interrupt?
It should be safe, though. I can have a stab to see if it breaks Windows/Linux at least, and share an RFC if I get anywhere.
Quoted 7 lines
Unless you think it's a waste of time. Ideally guests shouldn't find themselves in this situation to begin with, and if they did, resetting would arguably just lead them into corruption again (for example). But it does provide a mechanism for QEMU to find out that the vhost backend stopped. That would help in the context of this patch.
F.
It's not a waste of time, it's just a lot of work
within guests.
Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
Quoted 7 lines
Quoted 6 lines
Felipe: Did you audit the code to make sure the invalid
avail_idxvalue and the fields it is propagated to (e.g.shadow_avail_idx) are always used in a safe way?I did it briefly. I also wrote a mock userspace driver that creates this condition in a very controlled way (so I can step half-way through setting up VQs and trigger a migration, for example). But you know how manual tests are... I may have missed something. Your expert eyes are most welcome. :)
F.
Quoted 2 lines · Michael S. Tsirkin
It's not a waste of time, it's just a lot of work
within guests.
Luckily it does no harm to set the NEEDS_RESET bit even if the guest doesn't handle it.
If the guest driver is unaware it may continue to submit requests to the device for a while. The device emulation code stops accepting new requests though. This means the device will become unresponsive until reset, which is not ideal but okay in the case where the device was put into an invalid state.
I agree that supporting NEEDS_RESET transparently inside guests is difficult. The driver needs to reset and resume the device without reporting errors to applications. In some cases drivers may not have enough state in order to do that. It's also tricky to test all code paths. I guess this is why no one has done it: drivers shouldn't enter the NEEDS_RESET state anyway and handling it is complex.
Stefan
Quoted 4 lines · Stefan Hajnoczi
Quoted 2 lines · Michael S. Tsirkin
It's not a waste of time, it's just a lot of work
within guests.
Quoted 8 lines
Luckily it does no harm to set the
NEEDS_RESETbit even if the guest doesn't handle it.If the guest driver is unaware it may continue to submit requests to the device for a while. The device emulation code stops accepting new requests though. This means the device will become unresponsive until reset, which is not ideal but okay in the case where the device was put into an invalid state.
There is no actual rule that device must stop processing requests. Driver can only assume that is the case after it started the actual reset.
Quoted 8 lines
I agree that supporting
NEEDS_RESETtransparently inside guests is difficult. The driver needs to reset and resume the device without reporting errors to applications. In some cases drivers may not have enough state in order to do that. It's also tricky to test all code paths. I guess this is why no one has done it: drivers shouldn't enter theNEEDS_RESETstate anyway and handling it is complex.Stefan
Quoted 14 lines · Michael S. Tsirkin
Quoted 4 lines · Stefan Hajnoczi
Quoted 2 lines · Michael S. Tsirkin
It's not a waste of time, it's just a lot of work
within guests.Quoted 8 lines
Luckily it does no harm to set the
NEEDS_RESETbit even if the guest doesn't handle it.If the guest driver is unaware it may continue to submit requests to the device for a while. The device emulation code stops accepting new requests though. This means the device will become unresponsive until reset, which is not ideal but okay in the case where the device was put into an invalid state.
Quoted 3 lines
There is no actual rule that device must stop processing requests. Driver can only assume that is the case after it started the actual reset.
Right, I mean QEMU's implementation stops processing new requests once the device has been marked "broken":
void *virtqueue_pop(VirtQueue *vq, size_t sz)
{
if (virtio_device_disabled(vq->vdev)) { <---
return NULL;Quoted 13 lines · * Stefan Hajnoczi ([email protected]) wrote:
Quoted 2 lines · Michael S. Tsirkin
It's not a waste of time, it's just a lot of work
within guests.Luckily it does no harm to set the
NEEDS_RESETbit even if the guest doesn't handle it.If the guest driver is unaware it may continue to submit requests to the device for a while. The device emulation code stops accepting new requests though. This means the device will become unresponsive until reset, which is not ideal but okay in the case where the device was put into an invalid state.
Quoted 3 lines
I agree that supporting
NEEDS_RESETtransparently inside guests is difficult. The driver needs to reset and resume the device without reporting errors to applications.
Is that required? I mean, what are the semantics of NEEDS_RESET - is it assuming that you must be able to do a silent recovery?
Dave
Quoted 6 lines
In some cases drivers may not have
enough state in order to do that. It's also tricky to test all code paths. I guess this is why no one has done it: drivers shouldn't enter theNEEDS_RESETstate anyway and handling it is complex.Stefan
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK