Skip to contentSkip to conversation

vdpa: Fix memory listener deletions of iova tree

PATCH In review Eugenio Pérez 2 participants 3 messages Jul 22, 2022 – Jul 26, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vdpa: Fix memory listener deletions of iova tree

vhost_vdpa_listener_region_del is always deleting the first iova entry of the tree, since it's using the needle iova instead of the result's one.

This was detected using a vga virtual device in the VM using vdpa SVQ. It makes some extra memory adding and deleting, so the wrong one was mapped / unmapped. This was undetected before since all the memory was mappend and unmapped totally without that device, but other conditions could trigger it too:

* mem_region was with .iova = 0, .translated_addr = (correct GPA).
* iova_tree_find_iova returned right result, but does not update mem_region.
* iova_tree_remove always removed region with .iova = 0. Right iova were sent to the device.
* Next map will fill the first region with .iova = 0, causing a mapping with the same iova and device complains, if the next action is a map.
* Next unmap will cause to try to unmap again iova = 0, causing the device to complain that no region was mapped at iova = 0.

Fixes 34e3c94edaef ("vdpa: Add custom IOTLB translations to SVQ")
---
 hw/virtio/vhost-vdpa.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/virtio/vhost-vdpa.c +1 -1
index 291cd19054..00e990ea40 100644
@@ -290,7 +290,7 @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener,
 
         result = vhost_iova_tree_find_iova(v->iova_tree, &mem_region);
         iova = result->iova;
-        vhost_iova_tree_remove(v->iova_tree, &mem_region);
+        vhost_iova_tree_remove(v->iova_tree, result);
     }
     vhost_vdpa_iotlb_batch_begin_once(v);
     ret = vhost_vdpa_dma_unmap(v, iova, int128_get64(llsize));

--
2.31.1

Acked-by on v1 1/1
#1 lore raw
Quoted 21 lines · Eugenio Pérez

vhost_vdpa_listener_region_del is always deleting the first iova entry of the tree, since it's using the needle iova instead of the result's one.

This was detected using a vga virtual device in the VM using vdpa SVQ. It makes some extra memory adding and deleting, so the wrong one was mapped / unmapped. This was undetected before since all the memory was mappend and unmapped totally without that device, but other conditions could trigger it too:

* mem_region was with .iova = 0, .translated_addr = (correct GPA).
* iova_tree_find_iova returned right result, but does not update mem_region.
* iova_tree_remove always removed region with .iova = 0. Right iova were sent to the device.
* Next map will fill the first region with .iova = 0, causing a mapping with the same iova and device complains, if the next action is a map.
* Next unmap will cause to try to unmap again iova = 0, causing the device to complain that no region was mapped at iova = 0.

Quoted 3 lines
Fixes 34e3c94edaef ("vdpa: Add custom IOTLB translations to SVQ")
Quoted 20 lines
---
 hw/virtio/vhost-vdpa.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/virtio/vhost-vdpa.c +1 -2
index 291cd19054..00e990ea40 100644
@@ -290,7 +290,7 @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener,
 
         result = vhost_iova_tree_find_iova(v->iova_tree, &mem_region);
         iova = result->iova;
-        vhost_iova_tree_remove(v->iova_tree, &mem_region);
+        vhost_iova_tree_remove(v->iova_tree, result);
     }
     vhost_vdpa_iotlb_batch_begin_once(v);
     ret = vhost_vdpa_dma_unmap(v, iova, int128_get64(llsize));
--

2.31.1

#2 lore raw
Quoted 23 lines · Jason Wang
Quoted 21 lines · Eugenio Pérez

vhost_vdpa_listener_region_del is always deleting the first iova entry of the tree, since it's using the needle iova instead of the result's one.

This was detected using a vga virtual device in the VM using vdpa SVQ. It makes some extra memory adding and deleting, so the wrong one was mapped / unmapped. This was undetected before since all the memory was mappend and unmapped totally without that device, but other conditions could trigger it too:

* mem_region was with .iova = 0, .translated_addr = (correct GPA).
* iova_tree_find_iova returned right result, but does not update mem_region.
* iova_tree_remove always removed region with .iova = 0. Right iova were sent to the device.
* Next map will fill the first region with .iova = 0, causing a mapping with the same iova and device complains, if the next action is a map.
* Next unmap will cause to try to unmap again iova = 0, causing the device to complain that no region was mapped at iova = 0.

Quoted 5 lines
Quoted 3 lines
Fixes 34e3c94edaef ("vdpa: Add custom IOTLB translations to SVQ")

I've queued this for hard-freeze.

Thanks

Quoted 21 lines
Quoted 21 lines
---
 hw/virtio/vhost-vdpa.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
hw/virtio/vhost-vdpa.c +1 -2
index 291cd19054..00e990ea40 100644
@@ -290,7 +290,7 @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener,
 
         result = vhost_iova_tree_find_iova(v->iova_tree, &mem_region);
         iova = result->iova;
-        vhost_iova_tree_remove(v->iova_tree, &mem_region);
+        vhost_iova_tree_remove(v->iova_tree, result);
     }
     vhost_vdpa_iotlb_batch_begin_once(v);
     ret = vhost_vdpa_dma_unmap(v, iova, int128_get64(llsize));
--

2.31.1