On 2021/3/23 17:27, Arnd Bergmann wrote:

On Tue, Mar 23, 2021 at 9:33 AM Jie Deng [off-list ref] wrote:

quoted

On 2021/3/23 15:27, Viresh Kumar wrote:

quoted

On 23-03-21, 22:19, Jie Deng wrote:

quoted

+static int __maybe_unused virtio_i2c_freeze(struct virtio_device *vdev)
+{
+    virtio_i2c_del_vqs(vdev);
+    return 0;
+}
+
+static int __maybe_unused virtio_i2c_restore(struct virtio_device *vdev)
+{
+    return virtio_i2c_setup_vqs(vdev->priv);
+}

Sorry for not looking at this earlier, but shouldn't we enclose the above two
within #ifdef CONFIG_PM_SLEEP instead and drop the __maybe_unused ?

I remembered I was suggested to use "__maybe_unused" instead of "#ifdef".

You may check this https://lore.kernel.org/patchwork/patch/732981/

The reason may be something like that.

I usually recommend the use of __maybe_unused for the suspend/resume
callbacks for drivers that use SIMPLE_DEV_PM_OPS() or similar helpers
that hide the exact conditions under which the functions get called.

In this driver, there is an explicit #ifdef in the reference to the
functions, so
it would make sense to use the same #ifdef around the definition.

A better question to ask is whether you could use the helpers instead,
and drop the other #ifdef.

        Arnd

I didn't see the "struct virtio_driver" has a member "struct dev_pm_ops *pm"

It defines its own hooks (freeze and restore) though it includes "struct 
device_driver"

which has a "struct dev_pm_ops *pm".

I just follow other virtio drivers to directly use the hooks defined in 
"struct virtio_driver".

For this driver, Both __maybe_unused and #ifdef are OK to me.