Quoting Chunyan Zhang (2019-11-24 18:10:58) > On Mon, 25 Nov 2019 at 09:33, Stephen Boyd <sboyd@xxxxxxxxxx> wrote: > > > > Quoting Chunyan Zhang (2019-11-17 03:27:15) > > > > > > Not sure if I understand correctly - do you mean that switch to use a > > > reference to clk_parent_data.hw in the driver instead? > > > like: > > > https://elixir.bootlin.com/linux/v5.4-rc7/source/drivers/clk/qcom/gcc-sm8150.c#L136 > > > > > > > Yes something like that. > > > > > Since if I have to define many clk_parent_data.fw_name strings > > > instead, it seems not able to reduce the code size, right? > > > > Ideally there are some internal only clks that can be linked to their > > If the *internal* clks should be in the same base address, then we > have many external clks as parents, since most of our clks are not > located according to modules which clks serve, but according to clk > type. > > > parent with a single clk_hw pointer. That will hopefully keep the size > > Since all clks used for a chip are defined in the same driver file, I > actually can use clk_hw pointer directly, that will cut down the size > of this driver code, and also easier for users to look for parents for > one clk (only need to look at driver file). > > But not sure if you aggree this way? If all clks are in the same file then it sounds fine to just use clk_hw pointers everywhere. > > > down somewhat. And if there are any external clks, they can be described > > in DT and then only the .fw_name field can be used and the fallback > > field .name can be left assigned to NULL. > > Yes, I noticed that. But I still need to add many .fw_name, that will > not be a small count. > Sure. The plan is to get rid of the array of string names to specify parents. We can debate the size and cost associated with moving away from that, but it won't change the overall message here that I'd like to see new drivers use the new way of specifying parents instead of using strings for topology description.
