From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Wu, Songjun" Subject: Re: [PATCH v2 14/18] serial: intel: Add CCF support Date: Mon, 6 Aug 2018 15:05:27 +0800 Message-ID: References: <20180803030237.3366-1-songjun.wu@linux.intel.com> <20180803030237.3366-15-songjun.wu@linux.intel.com> <20180803055640.GA32226@kroah.com> <763bba56-3701-7fe9-9b31-4710594b40d5@linux.intel.com> <20180803103023.GA6557@kroah.com> <3360edd2-f3d8-b860-13fa-ce680edbfd0a@hauke-m.de> <20180804124309.GB4920@kroah.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 8bit Return-path: In-Reply-To: Content-Language: en-US Sender: linux-kernel-owner@vger.kernel.org To: Arnd Bergmann , Greg Kroah-Hartman Cc: Hauke Mehrtens , hua.ma@linux.intel.com, yixin.zhu@linux.intel.com, chuanhua.lei@linux.intel.com, qi-ming.wu@intel.com, "open list:RALINK MIPS ARCHITECTURE" , linux-clk , linux-serial@vger.kernel.org, DTML , Linux Kernel Mailing List , Jiri Slaby List-Id: devicetree@vger.kernel.org On 8/5/2018 5:03 AM, Arnd Bergmann wrote: > On Sat, Aug 4, 2018 at 2:43 PM, Greg Kroah-Hartman > wrote: >> On Sat, Aug 04, 2018 at 12:54:22PM +0200, Hauke Mehrtens wrote: >>> On 08/03/2018 12:30 PM, Greg Kroah-Hartman wrote: >>>> On Fri, Aug 03, 2018 at 03:33:38PM +0800, Wu, Songjun wrote: >>> This patch makes it possible to use it with the legacy lantiq code and >>> also with the common clock framework. I see multiple options to fix this >>> problem. >>> >>> 1. The current approach to have it as a compile variant for a) legacy >>> lantiq arch code without common clock framework and b) support for SoCs >>> using the common clock framework. >>> 2. Convert the lantiq arch code to the common clock framework. This >>> would be a good approach, but it need some efforts. >>> 3. Remove the arch/mips/lantiq code. There are still users of this code. >>> 4. Use the old APIs also for the new xRX500 SoC, I do not like this >>> approach. >>> 5. Move lantiq_soc.h to somewhere in include/linux/ so it is globally >>> available and provide some better wrapper code. >> I don't really care what you do at this point in time, but you all >> should know better than the crazy #ifdef is not allowed to try to >> prevent/allow the inclusion of a .h file. Checkpatch might have even >> warned you about it, right? >> >> So do it correctly, odds are #5 is correct, as that makes it work like >> any other device in the kernel. You are not unique here. > The best approach here would clearly be 2. We don't want platform > specific header files for doing things that should be completely generic. > > Converting lantiq to the common-clk framework obviously requires > some work, but then again the whole arch/mips/lantiq/clk.c file > is fairly short and maybe not that hard to convert. > > >From looking at arch/mips/lantiq/xway/sysctrl.c, it appears that you > already use the clkdev lookup mechanism for some devices without > using COMMON_CLK, so I would assume that you can also use those > for the remaining clks, which would be much simpler. It registers > one anonymous clk there as > > clkdev_add_pmu("1e100c00.serial", NULL, 0, 0, PMU_ASC1); > > so why not add replace that with two named clocks and just use > the same names in the DT for the newer chip? > > Arnd We discussed internally and have another solution for this issue. Add one lantiq.h in the serial folder, and use "#ifdef preprocessor" in lantiq.h, also providing no-op stub functions in the #else case, then call those functions unconditionally from lantiq.c to avoid #ifdef in C file. To support CCF in legacy product is another topic, is not included in this patch. The implementation is as following： #ifdef CONFIG_LANTIQ #include #else #define LTQ_EARLY_ASC 0 #define CPHYSADDR(_val) 0 static inline struct clk *clk_get_fpi(void) {     return NULL; } #endif