virtio: remove explicit pci ids from virtio_pci.c

Qumranet let us use their PCI vendor ID, with device ids >= 0x1000. We can specify that we accept all of them in the device ID table, and then return -ENODEV in the probe routine.

modprobe will load all the modules which match a given alias so we might be loaded when we're not needed, but that's the worst that can happen.

Signed-off-by Rusty Russell <rusty-8n+1lVoiYb80n/[email protected]>

---
 drivers/virtio/virtio_pci.c |    9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff -r 4edff5e24614 drivers/virtio/virtio_pci.c

drivers/virtio/virtio_pci.c +2 -3

@@ -62,10 +62,9 @@ struct virtio_pci_vq_info
 	struct list_head node;
 };
 
-/* We have to enumerate here all virtio PCI devices. */
+/* Qumranet donated their vendor ID and devices >= 0x1000 */
 static struct pci_device_id virtio_pci_id_table[] = {
-	{ 0x1af4, 0x1000, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, /* virtio net */
-	{ 0x1af4, 0x1001, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, /* virtio blk */
+	{ 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
 	{ 0 },
 };
 

Qumranet let us use their PCI vendor ID, with device ids >= 0x1000. We can specify that we accept all of them in the device ID table, and then return -ENODEV in the probe routine.

modprobe will load all the modules which match a given alias so we might be loaded when we're not needed, but that's the worst that can happen.

Signed-off-by Rusty Russell <[email protected]>

---
 drivers/virtio/virtio_pci.c |    9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff -r 4edff5e24614 drivers/virtio/virtio_pci.c

drivers/virtio/virtio_pci.c +6 -3

@@ -62,10 +62,9 @@ struct virtio_pci_vq_info
 	struct list_head node;
 };
 
-/* We have to enumerate here all virtio PCI devices. */
+/* Qumranet donated their vendor ID and devices >= 0x1000 */
 static struct pci_device_id virtio_pci_id_table[] = {
-	{ 0x1af4, 0x1000, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, /* virtio net */
-	{ 0x1af4, 0x1001, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, /* virtio blk */
+	{ 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
 	{ 0 },
 };
 

@@ -303,6 +302,10 @@ static int __devinit virtio_pci_probe(st
 	struct virtio_pci_device *vp_dev;
 	int err;
 
+	/* We only own device ids >= 0x1000 */
+	if (pci_dev->device < 0x1000)
+		return -ENODEV;
+
 	/* allocate our structure and fill it out */
 	vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 	if (vp_dev == NULL)

Quoted 4 lines · Rusty Russell

Qumranet let us use their PCI vendor ID, with device ids >= 0x1000. We can specify that we accept all of them in the device ID table, and then return -ENODEV in the probe routine.

I thought the device id range was smaller. Avi?

Quoted 4 lines · Rusty Russell

Qumranet let us use their PCI vendor ID, with device ids >= 0x1000. We can specify that we accept all of them in the device ID table, and then return -ENODEV in the probe routine.

I thought the device id range was smaller. Avi?

Quoted 5 lines · Anthony Liguori

Quoted 3 lines · Rusty Russell

Qumranet let us use their PCI vendor ID, with device ids >= 0x1000. We can specify that we accept all of them in the device ID table, and then return -ENODEV in the probe routine.

Quoted 1 line

I thought the device id range was smaller. Avi?

Yes, 0x1000-0x10ff IIRC.

Also, if we break compatibility in the future, won't this claim devices we don't actually support?

Quoted 5 lines · Anthony Liguori

Quoted 3 lines · Rusty Russell

Qumranet let us use their PCI vendor ID, with device ids >= 0x1000. We can specify that we accept all of them in the device ID table, and then return -ENODEV in the probe routine.

Quoted 1 line

I thought the device id range was smaller. Avi?

Yes, 0x1000-0x10ff IIRC.

Also, if we break compatibility in the future, won't this claim devices we don't actually support?