On 2/8/21 5:43 AM, Marek Vasut wrote:
On 2/8/21 6:57 AM, Samuel Holland wrote:
Resetting an XHCI controller inside xhci_register undoes any register setup performed by the platform driver. And at least on the Allwinner H6, resetting the XHCI controller also resets the PHY, which prevents the controller from working. That means the controller must be taken out of reset before initializing the PHY, which must be done before calling xhci_register.
The logic in the XHCI core was added to support the Raspberry Pi 4 (although this was not mentioned in the commit log!), which uses the xhci-pci platform driver. Move the reset logic to the platform driver, where it belongs, and where it cannot interfere with other platform drivers.
Are there any other XHCI drivers using the XHCI core code which might stop resetting correctly due to this patch ?
Since commit 0b80371b350e was merged, the only new call to xhci_register is in USB_MTU3. Neither the binding for "mediatek,mtu3" nor any device tree node containing that compatible string has a "resets" property. And there have been no calls to reset_(de)assert since then, either. So I do not see any other drivers that would be likely to break.
Samuel