On Wed, Jul 31, 2024 at 10:57 PM Andrew Lunn [off-list ref] wrote:

quoted

quoted

quoted

+    /// `phydev` must be passed by the corresponding callback in `phy_driver`.
+    unsafe extern "C" fn probe_callback(phydev: *mut bindings::phy_device) -> core::ffi::c_int {
+        from_result(|| {
+            // SAFETY: This callback is called only in contexts
+            // where we can exclusively access to `phy_device`, so the accessors on
+            // `Device` are okay to call.

This one is slightly different to other callbacks. probe is called
without the mutex. Instead, probe is called before the device is
published. So the comment is correct, but given how important Rust
people take these SAFETY comments, maybe it should indicate it is
different to others?

Interesting. Given that we don't hold the mutex, does that mean that
some of the methods on Device are not safe to call in this context? Or
is there something else that makes it okay to call them despite not
holding the mutex?

probe is always the first method called on a device driver to match it
to a device. Traditionally, if probe fails, the device is destroyed,
since there is no driver to drive it. probe needs to complete
successfully before the phy_device structure is published so a MAC
driver can reference it. If it is not published, nothing can have
access to it, so you don't need to worry about parallel activities on
it.

And a PHY driver does not need a probe function. Historically, probe
was all about, can this driver drive this hardware. However, since we
have ID registers in the hardware, we already know the driver can
drive the hardware. So probe is now about setting up whatever needs
setting up. For PHY drivers, there is often nothing, no local state
needed, etc. So the probe is optional.

Makes sense. Thanks for the explanation!

Alice