Re: [RFC PATCH net-next 0/3] Make the PHY library stop being so greedy when binding the generic PHY driver

[Vladimir Oltean <olteanv@gmail.com>]

On Thu, Sep 02, 2021 at 10:07:49PM +0200, Andrew Lunn wrote:
> > The interrupt controller _has_ been set up. The trouble is that the
> > interrupt controller has the same OF node as the switch itself, and the
> > same OF node. Therefore, fw_devlink waits for the _entire_ switch to
> > finish probing, it doesn't have insight into the fact that the
> > dependency is just on the interrupt controller.
>
> That seems to be the problem. fw_devlink appears to think probe is an
> atomic operation. A device is not probed, or full probed. Where as the
> drivers are making use of it being non atomic.
>
> Maybe fw_devlink needs the third state, probing. And when deciding if
> a device can be probed and depends on a device which is currently
> probing, it looks deeper, follows the phandle and see if the resource
> is actually available?

This is interesting because there already exists a device link state for
when the consumer is "probing", but for the supplier, it's binary:

/**
 * enum device_link_state - Device link states.
 * @DL_STATE_NONE: The presence of the drivers is not being tracked.
 * @DL_STATE_DORMANT: None of the supplier/consumer drivers is present.
 * @DL_STATE_AVAILABLE: The supplier driver is present, but the consumer is not.
 * @DL_STATE_CONSUMER_PROBE: The consumer is probing (supplier driver present).
 * @DL_STATE_ACTIVE: Both the supplier and consumer drivers are present.
 * @DL_STATE_SUPPLIER_UNBIND: The supplier driver is unbinding.
 */

The check that's killing us is in device_links_check_suppliers, and is
for DL_STATE_AVAILABLE:

	list_for_each_entry(link, &dev->links.suppliers, c_node) {
		if (!(link->flags & DL_FLAG_MANAGED))
			continue;

		if (link->status != DL_STATE_AVAILABLE &&
		    !(link->flags & DL_FLAG_SYNC_STATE_ONLY)) {
			device_links_missing_supplier(dev);
			dev_err(dev, "probe deferral - supplier %s not ready\n",
				dev_name(link->supplier));
			ret = -EPROBE_DEFER;
			break;
		}
		WRITE_ONCE(link->status, DL_STATE_CONSUMER_PROBE);
	}

Anyway, I was expecting quite a different reaction from this patch
series, and especially one from Saravana. We are essentially battling to
handle an -EPROBE_DEFER we don't need (the battle might be worth it
though, in the general case, which is one of the reasons I posted them).

But these patches also solve DSA's issue with the circular dependency
between the switch and its internal PHYs, and nobody seems to have asked
the most important question: why?

The PHY should return -EPROBE_DEFER ad infinitum, since its supplier has
never finished probing by the time it calls phy_attach_direct.