Hi! On Thu, Nov 12, 2020 at 12:30 AM Sergio Paracuellos <sergio.paracuellos@xxxxxxxxx> wrote: > > This patchset ports CPU clock detection for MT7621 from OpenWrt > and adds a complete clock plan for the mt7621 SOC. > > The documentation for this SOC only talks about two registers > regarding to the clocks: > * SYSC_REG_CPLL_CLKCFG0 - provides some information about boostrapped > refclock. PLL and dividers used for CPU and some sort of BUS (AHB?). > * SYSC_REG_CPLL_CLKCFG1 - a banch of gates to enable/disable clocks for > all or some ip cores. > > No documentation about a probably existant set of dividers for each ip > core is included in the datasheets. So we cannot make anything better, > AFAICT. > > Looking into driver code, there is another frequency which is used in > some drivers (uart, sd...) which for any reason is always hardcoded to > 50 MHz. Taking this into account this patchset provides three main fixed > clocks to the SOC in 'mt7621-pll' which are: > - "cpu": with detected frequency (900 MHz in my board). > - "ahb": cpu / 4 = 225 Mhz. > - "apb": 50 Mhz. > > PLL controller cannot be manipulatedbecause there is no info about > how to do it. Because of this, there is nothing related with registers > in the included binding. > > It also provides a clock gate driver 'mt7621-clk' as a platform driver > to allow to enable and disable some clocks in the different ip cores. > The parent clocks for this clock gates have also set taking into account > existant device tree and driver code resulting in the followings: > - "hsdma": "ahb" > - "fe": "ahb" > - "sp_divtx": "ahb" > - "timer": "cpu" > - "int": "cpu" > - "mc": "ahb" > - "pcm": "ahb" > - "pio": "ahb" > - "gdma": "ahb" > - "nand": "ahb" > - "i2c": "ahb" > - "i2s": "ahb" > - "spi": "ahb" > - "uart1": "apb" > - "uart2": "apb" > - "uart3": "apb" > - "eth": "ahb" > - "pcie0": "ahb" > - "pcie1": "ahb" > - "pcie2": "ahb" > - "crypto": "ahb" > - "shxc": "ahb" > > There was a previous attempt of doing this here[0] but the author > did not wanted to make assumptions of a clock plan for the platform. I've already said in previous threads that clock assignment in current linux kernel is not trustworthy. I've got the clock plan for mt7621 now. (Can't share it, sorry.) Most of your clock assumptions above are incorrect. I've made a clock driver with gate support a few months ago.[0] but I don't have much time to really finish it. Maybe you could rework your clock gate driver based on it. [0] https://github.com/981213/linux/commit/2eca1f045e4c3db18c941135464c0d7422ad8133 -- Regards, Chuanhong Guo
