From: linux@rempel-privat.de (Oleksij Rempel) Date: Sat, 20 Sep 2014 20:06:21 +0200 Subject: [PATCH 4/7] clk: add clk-asm9260 driver In-Reply-To: <20140918075650.GB29620@piout.net> References: <1410646408-28901-1-git-send-email-linux@rempel-privat.de> <1410646408-28901-5-git-send-email-linux@rempel-privat.de> <20140917135440.GG27054@piout.net> <541A7FE1.8060202@rempel-privat.de> <20140918075650.GB29620@piout.net> Message-ID: <541DC21D.7050303@rempel-privat.de> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Am 18.09.2014 um 09:56 schrieb Alexandre Belloni: > Hi, > > On 18/09/2014 at 08:46:57 +0200, Oleksij Rempel wrote : >> Am 17.09.2014 um 15:54 schrieb Alexandre Belloni: >>> Hi, >>> >>> Please have a look at https://lkml.org/lkml/2014/5/14/598 which describe >>> the preferred way of implementing clocks in the CCF where you only >>> declare the clock generator in the DT instead of each separate clocks. >> >> hm... if i see it correctly. i will need to move almost everything from >> DT to clk-driver. And i will need to create separate driver for other >> SoC provided by AlphaScale. On other side with current (generic) driver >> i will need only to change DT and it will work. >> >> May be instead of going qcom way, it will be better to have generic >> gate, div and mux bindings for DT? Sure it will make DT a bit fuzzy, but >> it will dramatically reduce kernel code and reduce the time of providing >> code to upstream. >> >> I don't think drivers/clk/qcom/gcc-msm8974.c is less complicated solution. > > IT is not less complicated but it is more flexible (for exemple when > setting flags on individual clocks, like CLK_SET_RATE_PARENT or > CLK_IGNORE_UNUSED). Describing all the individual clocks in DT is a > mistake we made in at91 and it will definitely bother you in the future, > for example when you realize that your clock controller is also taking > care of reset or power management. > > Also, this is the kind of driver you write only once per SoC so it is > about the same doing it in DT or in C with the added advantage that > doing it in C takes less memory and is probably faster. > > You can have a look at how it has been done for the berlin SoCs to see > how you can easily reuse code between drivers. What is the correct way to handle/define i2s MCLK input? I have I2S mux clock with choice of 3 sources: Xtal, PLL and MCLK. One of pins can be configured as MCLK src. Should i define fixedrate-clk? -- Regards, Oleksij -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 213 bytes Desc: OpenPGP digital signature URL: