Rob Herring wrote: > On Thu, Jun 09, 2022 at 06:11:51PM +0000, Thinh Nguyen wrote: >> Hi Rob, >> >> Rob Herring wrote: >>> On Fri, Jun 03, 2022 at 07:48:08PM -0700, Thinh Nguyen wrote: >>>> Set this property to use xhci-snps extension to handle common Synopsys >>>> DWC_usb3x host quirks. >>> >>> I don't see why this needs to be in DT. >>> >>> The DWC3 stuff is a mess of quirks which doesn't work well. Quirk >>> properties in DT require either knowing the quirk up front (You don't >>> always) or updating the DT on a platform when you find one. Quirks >>> should be implied by the compatible string(s) instead. >>> >> >> Since different vendors share the same Synopsys DWC_usb3x IPs, the >> controller's behavior is predictable based on the IP versions. > > Not really, have you looked at the existing bindings. How does the same > IP have different numbers of clocks, resets, etc.? It's a huge mess for > these licensed IPs and partly because they don't have publicly available > specs where we can check what resources there really are. I'm not claiming every platform has the same setup. I'm talking about the controller's common behavior base on IP versions. > >> Just >> using the compatible strings will become unmanageable when we have the >> common behavior across different vendors. > > This IP is not special. We use compatible everywhere else and nowhere is > it unmanageable. And again, that's the only way you can handle quirks in > the OS without changing the DT. Try removing all the version checks in dwc3 driver and replace them with compatibility strings. Not only will it explode the driver, now we'd somehow need to notify all the vendors to add their compatibility string where it can/should be a simple version check. > >> Can we rename this property to "xhci-snps-DWC_usb3x-ip" or something >> similar? >> >> The main goal for this device property is to indicate that it's >> Synopsys's DWC_usb3x IP. As long as we know this, the xhci-snps glue >> extension can handle the fine tuning for the controller's behavior. > > Can't you just check the parent device compatible is 'snps,dwc3'? (I'm > assuming the struct device hierarchy is glue device -> dwc3 -> xhci.) Some dwc3 host devices also exist as pci devices, there's no compatible string for that. As noted below, if the xhci pci device goes through the dwc3 code path, then the dwc3 driver will do some initialization and wrap the device as a xhci platform device before passing it to the xhci platform glue driver. But since it's a pci device underneath, there's no compatibility string. > >> We could use compatible string for this goal also, but that would mean >> the host devices that go through the dwc3 driver path may not have the >> compatible string. (e.g. host on pci bus but get recreated as platform >> device). Then we would need a different way to determine that. We could >> match the parent device driver for "dwc3", but that implementation looks >> fragile. >> >> So, will the device property "xhci-snps-DWC_usb3x-ip" work for you? > > Looking at the driver, you are just creating a property within the > kernel. It's never in DT, so why are you documenting it? You can do > whatever DWC3 and XHCI maintainers will allow and as DT maintainer I > don't care. I don't really think it is a great approach, but ... > I figure that we should document any new property should we use it... I just want to note that this RFC series is to garner feedbacks. In no way that I want to force any one approach. So, your feedback is highly appreciated. Since we're introducing a new glue extension, if possible, it'd be great if we can start off with a proper approach and build on it. If it's not a good approach, let's help come up with one that can work for _all_ device types. Thanks, Thinh
