Robert Hancock wrote: > On Tue, 2022-01-18 at 20:00 +0000, Thinh Nguyen wrote: >> Sean Anderson wrote: >>> Hi Thinh, >>> >>> On 1/18/22 2:46 PM, Thinh Nguyen wrote: >>>> Hi Sean, >>>> >>>> Baruch Siach wrote: >>>>> Hi Sean, Thinh, >>>>> >>>>> On Fri, Jan 14 2022, Sean Anderson wrote: >>>>>> This is a rework of patches 3-5 of [1]. It attempts to correctly >>>>>> program >>>>>> REFCLKPER and REFCLK_FLADJ based on the reference clock frequency. >>>>>> Since >>>>>> we no longer need a special property duplicating this configuration, >>>>>> snps,ref-clock-period-ns is deprecated. >>>>>> >>>>>> Please test this! Patches 3/4 in this series have the effect of >>>>>> programming REFCLKPER and REFCLK_FLADJ on boards which already >>>>>> configure >>>>>> the "ref" clock. I have build tested, but not much else. >>>>>> >>>>>> [1] >>>>>> https://urldefense.com/v3/__https://lore.kernel.org/linux-usb/20220114044230.2677283-1-robert.hancock@xxxxxxxxxx/__;!!A4F2R9G_pg!M3zKxDZC9a_etqzXo7GSEMTHRWfc1wR_84wwM4-fShiA35CsGcxcTEffHPbprbdC4d2R$ >>>>>> >>>>> >>>>> Thinh, you suggested the dedicated DT property for the reference clock: >>>>> >>>>> >>>>> https://urldefense.com/v3/__https://lore.kernel.org/all/d5acb192-80b9-36f7-43f5-81f21c4e6ba0@xxxxxxxxxxxx/__;!!A4F2R9G_pg!M3zKxDZC9a_etqzXo7GSEMTHRWfc1wR_84wwM4-fShiA35CsGcxcTEffHPbprbpOFmvX$ >>>>> >>>>> >>>>> Can you comment on this series? >>>>> >>>> >>>> Unless there's a good way to pass this information for PCI devices, my >>>> opinion hasn't changed. (Btw, I don't think creating a dummy clock >>>> provider and its dummy ops is a good solution as seems to complicate and >>>> bloat the PCI glue drivers). >>> >>> Can you explain your situation a bit more? I'm not sure how you can >>> access a device tree property but not add a fixed-rate clock. >>> >>> --Sean >> >> Currently for dwc3 pci devices, we have glue drivers that create a >> platform_device with specific properties to pass to the dwc3 core >> driver. Without a ref clock property, we would need another way to pass >> this information to the core driver or another way for the dwc3 core >> driver to check for specific pci device's properties and quirks. > > We've used the device tree to instantiate/configure devices inside of a PCI > device, though obviously that only works on DT-based platforms, and for > hardware that's part of the board itself, not an add-in card. > > We've also used the MFD infrastructure to instantiate devices and device > properties inside a PCI device on x86, which can be used if the driver you are > instantiating uses the generic device property accessors and not the DT- > specific ones. That gets a bit dirty however - I don't think there's an easy > way to create properties that are references to other nodes, or more than a > single level deep heirarchy of nodes. > > For a use case like you're describing, it sounds like it would be better to > abstract away some of the core DWC3 code from reading the settings from DT > directly, so that the PCI devices can instantiate it and set the configuration > however they want, without having to worry about creating fake properties for > the core to read. I think that pattern has been used with some other drivers > such as AHCI? > Yes. It would be great if we can rework and abstract this. I'd need to review how AHCI handles it. It doesn't seem like a small task as it touches on multiple drivers. But if anyone can start this off, I can help further contribute/review. Thanks, Thinh
