Re: Regression: onboard-usb-hub breaks USB on RPi 3

[Stefan Wahren <stefan.wahren@i2se.com>]

Hi Matthias,

Am 20.12.22 um 01:30 schrieb Matthias Kaehlcke:
> On Mon, Dec 19, 2022 at 11:32:58PM +0100, Stefan Wahren wrote:
>> Hi Matthias,
>>
>> Am 19.12.22 um 18:44 schrieb Matthias Kaehlcke:
>>> Hi Stefan,
>>>
>>> Sorry for the regression.
>>>
>>> What seems to happen is this:
>>>
>>> arch/arm/boot/dts/bcm283x-rpi-lan7515.dtsi specifies device nodes for the
>>> two (nested) USB hubs (which is done rarely since USB devices (including
>>> hubs) are autodetected). The DT nodes were most likely only added to be
>>> able to configure the LED modes of the USB to Ethernet adapter. With
>>> 43993626de00 ("usb: misc: onboard-hub: add support for Microchip USB2514B
>>> USB 2.0 hub") the onboard_usb_hub driver gained support for the hubs on
>>> the RPi3. The onboard_usb_hub driver expects a regulator ("vdd") in the DT
>>> node of the USB hub, which isn't present for the RPi3 (this isn't an error
>>> per se). Without the regulator the onboard_hub platform driver fails to
>>> probe, when the USB driver of the hub is probed (onboard_hub_usbdev_probe())
>>> it doesn't find the corresponding platform driver instance
>>> (_find_onboard_hub()) and defers probing. When the deferred probe runs it
>>> encounters the same situation, rinse and repeat.
>> I forgot to mention that in error case /sys/kernel/debug/devices_deferred
>> was empty.
>>> One possible fix would be to specify the 'missing' "vdd" property, however
>>> that wouln't fix the issue for other boards with a similar configurations.
>>> Instead the driver could check if "vdd" exists in the DT node of the hub,
>>> and not defer probing if it doesn't.
>>>
>>> Could you please try if the below patch fixes the issue on the Rpi 3?
>> Yes, this prevents probing of onboard-usb-hub and the issue.
> Thanks for the confirmation, I'll send out a proper patch to get this fixed
> upstream.

sorry, i accidentally disabled this driver during testing of the patch. 
So i erroneously assumed the patch is working, but it's not. I seems 
that the change is never reached (add dev_info around your change).

The following worked on my Raspberry Pi 3 B+

diff --git a/drivers/usb/misc/onboard_usb_hub.c 
b/drivers/usb/misc/onboard_usb_hub.c
index de3627af3c84..570e9f3d2d89 100644
--- a/drivers/usb/misc/onboard_usb_hub.c
+++ b/drivers/usb/misc/onboard_usb_hub.c
@@ -227,6 +227,9 @@ static int onboard_hub_probe(struct platform_device 
*pdev)
      if (!hub)
          return -ENOMEM;

+    if (!of_get_property(dev->of_node, "vdd", NULL))
+        return -ENODEV;
+
      hub->vdd = devm_regulator_get(dev, "vdd");
      if (IS_ERR(hub->vdd))
          return PTR_ERR(hub->vdd);
@@ -340,6 +343,15 @@ static struct onboard_hub *_find_onboard_hub(struct 
device *dev)
      hub = dev_get_drvdata(&pdev->dev);
      put_device(&pdev->dev);

+    /*
+     * Some boards have device tree nodes for USB hubs supported by this
+     * driver, but the nodes don't have all properties needed for the 
driver
+     * to work properly. Use the absence of the "vdd" regulator as an
+     * indicator of such nodes.
+     */
+    if (!of_get_property(pdev->dev.of_node, "vdd", NULL))
+        return ERR_PTR(-ENODEV);
+
      /*
       * The presence of drvdata ('hub') indicates that the platform driver
       * finished probing. This handles the case where (conceivably) we 
couldThe following changes worked for me: