Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux

firmware: arm_scmi: Mark VirtIO ready before registering scmi_virtio_driver

After commit 20bda12a0ea0 (“firmware: arm_scmi: Make VirtIO transport a
standalone driver”), the VirtIO transport probes independently. During
scmi_virtio_probe, scmi_probe() is called, which intune invokes
scmi_protocol_acquire() that sends a message over the virtqueue and
waits for a reply.

Previously, DRIVER_OK was only set after scmi_vio_probe, in the core
virtio via virtio_dev_probe(). According to the Virtio spec (3.1 Device
Initialization):
| The driver MUST NOT send any buffer available notifications to the
| device before setting DRIVER_OK.

Some type-1 hypervisors block available-buffer notifications until the
driver is marked OK. In such cases, scmi_vio_probe stalls in
scmi_wait_for_reply(), and the probe never completes.

Resolve this by setting DRIVER_OK immediately after the device-specific
setup, so scmi_probe() can safely send notifications.

Note after splitting the transports into modules, the probe sequence
changed a bit. We can no longer rely on virtio_device_ready() being
called by the core in virtio_dev_probe(), because scmi_vio_probe()
doesn’t complete until the core SCMI stack runs scmi_probe(), which
immediately issues the initial BASE protocol exchanges.

Fixes: 20bda12a0ea0 ("firmware: arm_scmi: Make VirtIO transport a standalone driver")
Signed-off-by: Junnan Wu <junnan01.wu@samsung.com>
Message-Id: <20250812075343.3201365-1-junnan01.wu@samsung.com>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>

authored by

Junnan Wu and committed by
Sudeep Holla
e8faa8a4 224dcf29

+3
+3
drivers/firmware/arm_scmi/transports/virtio.c
··· 871 871 /* Ensure initialized scmi_vdev is visible */ 872 872 smp_store_mb(scmi_vdev, vdev); 873 873 874 + /* Set device ready */ 875 + virtio_device_ready(vdev); 876 + 874 877 ret = platform_driver_register(&scmi_virtio_driver); 875 878 if (ret) { 876 879 vdev->priv = NULL;