virtio-mmio: don't break lifecycle of vm_dev

[ Upstream commit 55c91fe ]

vm_dev has a separate lifecycle because it has a 'struct device'
embedded. Thus, having a release callback for it is correct.

Allocating the vm_dev struct with devres totally breaks this protection,
though. Instead of waiting for the vm_dev release callback, the memory
is freed when the platform_device is removed. Resulting in a
use-after-free when finally the callback is to be called.

To easily see the problem, compile the kernel with
CONFIG_DEBUG_KOBJECT_RELEASE and unbind with sysfs.

The fix is easy, don't use devres in this case.

Found during my research about object lifetime problems.

Fixes: 7eb781b ("virtio_mmio: add cleanup for virtio_mmio_probe")
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Message-Id: <20230629120526.7184-1-wsa+renesas@sang-engineering.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

drivers/virtio/virtio_mmio.c

@@ -590,9 +590,8 @@ static void virtio_mmio_release_dev(struct device *_d)
 	struct virtio_device *vdev =
 			container_of(_d, struct virtio_device, dev);
 	struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev);
-	struct platform_device *pdev = vm_dev->pdev;
 
-	devm_kfree(&pdev->dev, vm_dev);
+	kfree(vm_dev);
 }
 
 /* Platform device */
@@ -603,7 +602,7 @@ static int virtio_mmio_probe(struct platform_device *pdev)
 	unsigned long magic;
 	int rc;
 
-	vm_dev = devm_kzalloc(&pdev->dev, sizeof(*vm_dev), GFP_KERNEL);
+	vm_dev = kzalloc(sizeof(*vm_dev), GFP_KERNEL);
 	if (!vm_dev)
 		return -ENOMEM;