于 2020年10月25日 GMT+08:00 下午10:36:08, Andrew Lunn <andrew@xxxxxxx> 写到: >On Sun, Oct 25, 2020 at 10:27:05PM +0800, Icenowy Zheng wrote: >> >> >> 于 2020年10月25日 GMT+08:00 下午10:18:25, Andrew Lunn <andrew@xxxxxxx> 写到: >> >On Sun, Oct 25, 2020 at 04:55:56PM +0800, Icenowy Zheng wrote: >> >> Currently there are many boards that just set "rgmii" as phy-mode >in >> >the >> >> device tree, and leave the hardware [TR]XDLY pins to set PHY delay >> >mode. >> >> >> >> In order to keep old device tree working, omit setting delay for >just >> >> "RGMII" without any internal delay suffix, otherwise many devices >are >> >> broken. >> > >> >Hi Icenowy >> > >> >We have been here before with the Atheros PHY. It did not correctly >> >implement one of the delay modes, until somebody really did need >that >> >mode. So the driver was fixed. And we then found a number of device >> >trees were also buggy. It was painful for a while, but all the >device >> >trees got fixed. >> >> 1. As the PHY chip has hardware configuration for configuring delays, >> we should at least have a mode that respects what's set on the >hardware. > >Yes, that is PHY_INTERFACE_MODE_NA. In DT, set the phy-mode to "". Or >for most MAC drivers, don't list a phy-mode at all. However, we still need to tell the MAC it's RGMII mode that is in use, not MII/RMII/*MII. So the phy-mode still needs to be something that contains rgmii. > >> 2. As I know, at least Fedora ships a device tree with their >bootloader, and >> the DT will not be updated with kernel. > >I would check that. Debian does the exact opposite, the last time i >looked. It always uses the DT that come with the kernel because it >understands DT can have bugs, like all software. > > Andrew
