On 6/12/25 2:57 AM, Jason Wang wrote:

On Fri, Jun 6, 2025 at 7:46 PM Paolo Abeni [off-list ref] wrote:

quoted

-u64 vp_modern_get_features(struct virtio_pci_modern_device *mdev);
-u64 vp_modern_get_driver_features(struct virtio_pci_modern_device *mdev);
-void vp_modern_set_features(struct virtio_pci_modern_device *mdev,
-                    u64 features);
+void
+vp_modern_get_driver_extended_features(struct virtio_pci_modern_device *mdev,
+                                      u64 *features);
+void vp_modern_get_extended_features(struct virtio_pci_modern_device *mdev,
+                                    u64 *features);
+void vp_modern_set_extended_features(struct virtio_pci_modern_device *mdev,
+                                    const u64 *features);
+
+static inline u64
+vp_modern_get_features(struct virtio_pci_modern_device *mdev)
+{
+       u64 features_array[VIRTIO_FEATURES_DWORDS];
+       int i;
+
+       vp_modern_get_extended_features(mdev, features_array);
+       for (i = 1; i < VIRTIO_FEATURES_DWORDS; ++i)
+               WARN_ON_ONCE(features_array[i]);

It looks to me it's sufficient and safe to just return
featuers_array[0] here. Or maybe we need some comment to explain why
we need WARN here.

vp_modern_get_extended_features() can return a 'features' array
including extended features. Callers of vp_modern_get_features() can
deal with/expect features to be present only in the lower 64 bit range.

This check is intended to catch early device bug/inconsistencies, but
I'm unsure if e.g. syzkaller could hit with some fancy setup.

I can drop the check in the next revision.

/P