On 26/02/2025 06:10, Ryan Chen wrote: >> Subject: Re: [PATCH v9 1/3] dt-binding: clock: ast2700: modify soc0/1 clock >> define >> >> On 25/02/2025 10:49, Ryan Chen wrote: >>>>>> Subject: Re: [PATCH v9 1/3] dt-binding: clock: ast2700: modify >>>>>> soc0/1 clock define >>>>>> >>>>>> On 24/02/2025 10:55, Ryan Chen wrote: >>>>>>> -remove redundant SOC0_CLK_UART_DIV13: >>>>>>> SOC0_CLK_UART_DIV13 is not use at clk-ast2700.c, the clock source >>>>>>> tree is uart clk src -> uart_div_table -> uart clk. >>>>>>> >>>>>>> -Change SOC0_CLK_HPLL_DIV_AHB to SOC0_CLK_AHBMUX: >>>>>>> modify clock tree implement. >>>>>>> older CLK_AHB use mpll_div_ahb/hpll_div_ahb to be ahb clock source. >>>>>>> mpll->mpll_div_ahb >>>>>>> -> clk_ahb >>>>>>> hpll->hpll_div_ahb >>>>>> >>>>>> >>>>>> I can barely understand it and from the pieces I got, it does not >>>>>> explain need for ABI break. >>>>>> >>>>> >>>>> #1. SCU0_CLK_UART_DIV13 is redundant, it does not impact ABI break >>>> >>>> You did not explain how it does not impact. Clock was exported, there >>>> was a user and now there is no clock. User stops working. ABI break. >>>> >>> >>> Sorry, SCU0_CLK_UART_DIV13 was defined, but was never referenced in any >> upstream device trees. >> >> >> That's incomplete definition of ABI >> >>> Since there is no in-tree usage of `SCU0_CLK_UART_DIV13`, its removal does >> not cause an ABI break. >> >> >> You ignored out of tree users. Please read carefully ABI docs. >> >> >>> >>>>> #2. Change SOC0_CLK_HPLL_DIV_AHB to SOC0_CLK_AHBMUX Older >>>> implement >>>>> where `mpll_div_ahb` and `hpll_div_ahb` were **hardcoded dividers** >>>>> for >>>> AHB. >>>>> In **the new approach (v8)**, I refactored the clock tree to clock tree. >>>> >>>> I still cannot parse sentences like "refactoring A to A". It's meaningless to >> me. >>>> >>>>> It should be ABI-safe change >>>> >>>> No, you do not understand the ABI. You removed a clock ID, that's the >>>> ABI change. >>>> >>>> Otherwise explain how this is not changing ABI. >>>> >>>> >>>>> >>>>> Or you want to keep original SOC0_CLK_HPLL_DIV_AHB define and then >>>>> add >>>> SOC0_CLK_AHBMUX. >>>>> To be 1st patch, then 2n patch remove redundant >>>> SOC0_CLK_HPLL_DIV_AHB? >>>> >>>> If you break the ABI you need to clearly explain why. We have long >>>> conversations and you still did not say why. >>>> >>> Sorry, my point will be following steps to avoid potential ABI issues, >>> I can modify the patch series as follows: >>> 1. **Patch 1:** Add `SOC0_CLK_AHBMUX` without removing >> `SOC0_CLK_HPLL_DIV_AHB`. >>> 2. **Patch 2:** Finally remove `SOC0_CLK_HPLL_DIV_AHB`. >> >> >> I do not understand what changed here. You remove exported clock which is >> ABI, so how is this answering my question. >> >> You keep dodging my questions. Here I asked "why". I do not see any answer >> why. > > Apology, I can't catch your point. But I still need your guideline. > > **Regarding `SCU0_CLK_UART_DIV13`:** > - This clock ID was originally defined but was never used in any in-tree device trees. (i2c-ast2700.c v1 ~ v9) > - So there should not have ABI-break am I correct? No, because there are other users of bindings. All forks, out of tree DTS, other projects etc. You defined ABI for them. > > **Regarding `SOC0_CLK_HPLL_DIV_AHB` → `SOC0_CLK_AHBMUX`:** > - The previous implementation used dividers (`mpll_div_ahb`, `hpll_div_ahb`) for AHB clock selection. (i2c-ast2700.c v1 ~ v8) > mpll->mpll_div_ahb > -> clk_ahb > hpll->hpll_div_ahb > - The new approach use `SOC0_CLK_AHBMUX`, which AHB clock sources via a mux. (i2c-ast2700.c v9) > mpll-> > ahb_mux -> div_table -> clk_ahb > hpll-> Your formatting is one of the problems I have troubles understanding it. Above is not wrapped or wrapped oddly. You keep using bold * but double **, which is not a standard markup. Please switch to standard mailing list formatting - proper wrapping, only text mode and use client which actually can parse and create that. What I don't understand is how clocks could change in hardware. You described implementation, but the clock IDS do not describe implementation but the the mapping to real hardware clocks. So how SOC0_CLK_HPLL_DIV_AHB clock disappeared from hardware? > - And since i2c-ast2700.c (v8) is not applied, so there should also no one use this ABI. Am I correct? If binding was not applied, then what are you changing here? Does it mean header described clock which was in the hardware but its handling was not yet implemented in the driver? > > If this is not acceptable, my next patch will keep SCU0_CLK_UART_DIV13/SCU0_CLK_HPLL_DIV_AHB define. Maybe my last message was not clear, so: you need to explain why you are breaking ABI and/or explain the ABI impact in the commit msg. > But add new SCU0_CLK_AHBMUX define for avoid your point ABI break. Is this acceptable? Best regards, Krzysztof
