2009-01-29 15:32:33

by Mark McLoughlin

Subject: [PATCH] virtio-pci: do not oops on config change if driver not loaded

The host really shouldn't be notifying us of config changes
before the device status is VIRTIO_CONFIG_S_DRIVER or
VIRTIO_CONFIG_S_DRIVER_OK.

However, if we do happen to be interrupted while we're not
attached to a driver, we really shouldn't oops. Prevent
this simply by checking that device->driver is non-NULL
before trying to notify the driver of config changes.

Problem observed by doing a "set_link virtio.0 down" with
QEMU before the net driver had been loaded.

Signed-off-by: Mark McLoughlin <[email protected]>
---
drivers/virtio/virtio_pci.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/virtio/virtio_pci.c b/drivers/virtio/virtio_pci.c
index bef6b45..330aacb 100644
--- a/drivers/virtio/virtio_pci.c
+++ b/drivers/virtio/virtio_pci.c
@@ -192,7 +192,7 @@ static irqreturn_t vp_interrupt(int irq, void *opaque)
drv = container_of(vp_dev->vdev.dev.driver,
struct virtio_driver, driver);

- if (drv->config_changed)
+ if (drv && drv->config_changed)
drv->config_changed(&vp_dev->vdev);
}

--
1.6.0.6

2009-01-30 01:45:55

by Rusty Russell

Subject: Re: [PATCH] virtio-pci: do not oops on config change if driver not loaded

On Friday 30 January 2009 02:02:16 Mark McLoughlin wrote:
> The host really shouldn't be notifying us of config changes
> before the device status is VIRTIO_CONFIG_S_DRIVER or
> VIRTIO_CONFIG_S_DRIVER_OK.
>
> However, if we do happen to be interrupted while we're not
> attached to a driver, we really shouldn't oops.

Thanks, applied!

Rusty.