Skip to contentSkip to conversation

qemu-kvm: fix crash on reboot with vhost-net

PATCH Applied Michael S. Tsirkin 3 participants 4 messages Apr 28, 2010 KVM

lore
Share & tools
lore

Michael S. Tsirkin

author
#0 lore raw

PATCH v2 1/1 qemu-kvm: fix crash on reboot with vhost-net

When vhost-net is disabled on reboot, we set msix mask notifier to NULL to disable further mask/unmask notifications. Code currently tries to pass this NULL to notifier, leading to a crash. The right thing to do is
to add explicit APIs to enable/disable notifications. Now when disabling notifications:
- if vector is masked, we don't need to notify backend, just disable future notifications
- if vector is unmasked, invoke callback to unassign backend, then disable future notifications

This patch also polls notifier before closing it, to make sure we don't lose events if poll callback didn't have time to run.

Signed-off-by Michael S. Tsirkin <[email protected]>

---

Changes from v1:
Separate APIs to set and unset notifiers Test and clear notifier before destroying it

 hw/msix.c       |   40 +++++++++++++++++++++++++++++++++++-----
 hw/msix.h       |    1 +
 hw/virtio-pci.c |    7 +++++--
 3 files changed, 41 insertions(+), 7 deletions(-)
hw/msix.c +35 -5
index 3ec8805..8f9a621 100644
@@ -609,14 +609,44 @@ void msix_unuse_all_vectors(PCIDevice *dev)
 
 int msix_set_mask_notifier(PCIDevice *dev, unsigned vector, void *opaque)
 {
+    int r;
+    if (vector >= dev->msix_entries_nr || !dev->msix_entry_used[vector])
+        return 0;
+
+    assert(dev->msix_mask_notifier);
+    assert(opaque);
+    assert(!dev->msix_mask_notifier_opaque[vector]);
+
+    if (msix_is_masked(dev, vector)) {
+        return 0;
+    }
+    r = dev->msix_mask_notifier(dev, vector, opaque,
+                                msix_is_masked(dev, vector));
+    if (r < 0) {
+        return r;
+    }
+    dev->msix_mask_notifier_opaque[vector] = opaque;
+    return r;
+}
+
+int msix_unset_mask_notifier(PCIDevice *dev, unsigned vector)
+{
     int r = 0;
     if (vector >= dev->msix_entries_nr || !dev->msix_entry_used[vector])
         return 0;
 
-    if (dev->msix_mask_notifier)
-        r = dev->msix_mask_notifier(dev, vector, opaque,
-                                    msix_is_masked(dev, vector));
-    if (r >= 0)
-        dev->msix_mask_notifier_opaque[vector] = opaque;
+    assert(dev->msix_mask_notifier);
+    assert(dev->msix_mask_notifier_opaque[vector]);
+
+    if (msix_is_masked(dev, vector)) {
+        return 0;
+    }
+    r = dev->msix_mask_notifier(dev, vector,
+                                dev->msix_mask_notifier_opaque[vector],
+                                msix_is_masked(dev, vector));
+    if (r < 0) {
+        return r;
+    }
+    dev->msix_mask_notifier_opaque[vector] = NULL;
     return r;
 }
hw/msix.h +1 -0
index f167231..6b21ffb 100644
@@ -34,4 +34,5 @@ void msix_reset(PCIDevice *dev);
 extern int msix_supported;
 
 int msix_set_mask_notifier(PCIDevice *dev, unsigned vector, void *opaque);
+int msix_unset_mask_notifier(PCIDevice *dev, unsigned vector);
 #endif
hw/virtio-pci.c +5 -2
index 99a588c..c4bc633 100644
@@ -462,10 +462,13 @@ static int virtio_pci_set_guest_notifier(void *opaque, int n, bool assign)
         msix_set_mask_notifier(&proxy->pci_dev,
                                virtio_queue_vector(proxy->vdev, n), vq);
     } else {
-        msix_set_mask_notifier(&proxy->pci_dev,
-                               virtio_queue_vector(proxy->vdev, n), NULL);
+        msix_unset_mask_notifier(&proxy->pci_dev,
+				 virtio_queue_vector(proxy->vdev, n));
         qemu_set_fd_handler(event_notifier_get_fd(notifier),
                             NULL, NULL, NULL);
+        /* Test and clear notifier before closing it,
+         * in case poll callback didn't have time to run. */
+        virtio_pci_guest_notifier_read(vq);
         event_notifier_cleanup(notifier);
     }
Signature · 1 line
1.7.1.rc1.22.g3163
Acked-by on v2 1/1
#1 lore raw
Quoted 18 lines · Michael S. Tsirkin

When vhost-net is disabled on reboot, we set msix mask notifier to NULL to disable further mask/unmask notifications. Code currently tries to pass this NULL to notifier, leading to a crash. The right thing to do is
to add explicit APIs to enable/disable notifications. Now when disabling notifications:
- if vector is masked, we don't need to notify backend, just disable future notifications
- if vector is unmasked, invoke callback to unassign backend, then disable future notifications

This patch also polls notifier before closing it, to make sure we don't lose events if poll callback didn't have time to run.

Signed-off-by Michael S. Tsirkin <[email protected]>

---

Quoted 8 lines

Changes from v1:
Separate APIs to set and unset notifiers Test and clear notifier before destroying it

 hw/msix.c       |   40 +++++++++++++++++++++++++++++++++++-----
 hw/msix.h       |    1 +
 hw/virtio-pci.c |    7 +++++--
 3 files changed, 41 insertions(+), 7 deletions(-)
Acked-by Juan Quintela <[email protected]>

This patch addresses the issues that I pointed in previous mail and fixes the problem at hand.

Creating a better API for interaction between virtio-pci <-> msix still needs more work/thought, already discussed that with mst.

Later, Juan.

Applied
#2 lore raw
In reply to Michael S. Tsirkin
Quoted 11 lines · Michael S. Tsirkin

When vhost-net is disabled on reboot, we set msix mask notifier to NULL to disable further mask/unmask notifications. Code currently tries to pass this NULL to notifier, leading to a crash. The right thing to do is
to add explicit APIs to enable/disable notifications. Now when disabling notifications:
- if vector is masked, we don't need to notify backend, just disable future notifications
- if vector is unmasked, invoke callback to unassign backend, then disable future notifications

Quoted 5 lines

This patch also polls notifier before closing it, to make sure we don't lose events if poll callback didn't have time to run.

Signed-off-by Michael S. Tsirkin <[email protected]>

Applied, thanks.