On 08/07/2021 04:54, Bjorn Andersson wrote:

Bryan had a previous patch where the glue layer was notified about role
switching (iirc) and as soon as we hit a probe deferal in the core
driver we'd dereference some pointer in the glue layer. I don't find the
patch right now, but I suspect it might have been caused by the same
platform_get_drvdata() as we see in qcom_dwc3_resume_irq().

Here

https://lore.kernel.org/linux-usb/20200311191501.8165-7-bryan.odonoghue@linaro.org/ (local)

and here

https://lore.kernel.org/linux-usb/20200311191501.8165-8-bryan.odonoghue@linaro.org/ (local)

one thing about that I don't think is right now in retrospect is having 
to find a DT connector in the core, meaning it incorrectly assumes you 
have a node named "connector" as a child of dwc3-core

https://lore.kernel.org/linux-usb/158463604559.152100.9219030962819234620@swboyd.mtv.corp.google.com/ (local)

Having done some work with TCPM on pm8150b silicon I see what Stephen 
was saying about that

That's one solid reason I like the USB role-switch API - because it gets 
you out of the business of trying to discern from dwc3-qcom if dwc3-core 
has role-switching turned on by iterating through its range of DT nodes 
and looking for a special one named "connector"

The initial and imperfect solution I had for that looked like this

if (dwc3_qcom_find_gpio_usb_connector(qcom->dwc3)) {}

Wesley had another iteration on that that was a little better

https://lore.kernel.org/linux-usb/20201009082843.28503-4-wcheng@codeaurora.org/ (local)

+static int dwc3_qcom_connector_check(struct fwnode_handle *fwnode)
+{
+	if (fwnode && (!fwnode_property_match_string(fwnode, "compatible",
+						     "gpio-usb-b-connector") ||
+	    !fwnode_property_match_string(fwnode, "compatible",
+					  "usb-c-connector")))
+		return 1;
+
+	return 0;
+}

All we are really doing there is replicating functionality that the 
role-switch API already provides


---
bod