Hi Anson, On 27/06/2019 10:11, Anson Huang wrote: > Hi, Daniel > >> On 27/06/2019 02:43, Anson Huang wrote: >>> Hi, Daniel >>> >>>> On 26/06/2019 03:42, Anson Huang wrote: >>>>> Hi, Daniel >>>>> >>>>>> On 23/06/2019 14:38, Anson.Huang@xxxxxxx wrote: >>>>>>> From: Anson Huang <Anson.Huang@xxxxxxx> >>>>>>> >>>>>>> Systems which use platform driver model for clock driver require >>>>>>> the clock frequency to be supplied via device tree when system >>>>>>> counter driver is enabled. >>>>>>> >>>>>>> This is necessary as in the platform driver model the of_clk >>>>>>> operations do not work correctly because system counter driver is >>>>>>> initialized in early phase of system boot up, and clock driver >>>>>>> using platform driver model is NOT ready at that time, it will >>>>>>> cause system counter driver initialization failed. >>>>>>> >>>>>>> Add the optinal clock-frequency to the device tree bindings of the >>>>>>> NXP system counter, so the frequency can be handed in and the >>>>>>> of_clk operations can be skipped. >>>>>> >>>>>> Isn't it possible to create a fixed-clock and refer to it? So no >>>>>> need to create a specific action before calling timer_of_init() ? >>>>>> >>>>> >>>>> As the clock must be ready before the TIMER_OF_DECLARE, so adding a >>>>> CLK_OF_DECLARE_DRIVER in clock driver to ONLY register a fixed-clock? >>>>> The system counter's frequency are different on different platforms, >>>>> so adding fixed clock in system counter driver is NOT a good idea, >>>>> ONLY the DT node or the clock driver can create this fixed clock >>>>> according to >>>> platforms, can you advise where to create this fixed clock is better? >>>> >>>> Can you point me to a DT with the "nxp,sysctr-timer" ? >>> >>> The DT node of system counter is new added in 3/3 of this patch >>> series, also can be found from below link: >>> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatc >>> >> hwork.kernel.org%2Fpatch%2F11011703%2F&data=02%7C01%7Canson. >> huang% >>> >> 40nxp.com%7C8b9519ecceb346712be808d6fad675e4%7C686ea1d3bc2b4c6f >> a92cd99 >>> >> c5c301635%7C0%7C0%7C636972196338405582&sdata=sOQQzDFxoCqe >> VuHFuYPHh >>> F8Bdj2Zu9WS7Go%2FV9lrWa8%3D&reserved=0 >> >> Sorry, I was unclear. I meant a patch with the timer defined using a clock as >> defined currently in the binding (no clock-frequency). > > OK, for i.MX8MM, we use clocks, check below patch series: > > https://patchwork.kernel.org/patch/11008519/ > > code piece as below: > > + system_counter: timer@306a0000 { > + compatible = "nxp,sysctr-timer"; > + reg = <0x306a0000 0x30000>; > + interrupts = <GIC_SPI 47 IRQ_TYPE_LEVEL_HIGH>, > + <GIC_SPI 48 IRQ_TYPE_LEVEL_HIGH>; > + clocks = <&clk IMX8MM_CLK_SYS_CTR>; > + clock-names = "per"; > + }; Thanks, The fixed-clock can help to keep the code and the DT definition for the timer untouched as the 'clocks' above will refer to it. But that means we describe a fake clock. So it is up to you to decide if you want to stick the clock-frequency or use a fixed-clock. -- <http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
