Hi Yang, On Tue, 14 Oct 2025 at 23:03, Tom Rini <[email protected]> wrote: > > On Sat, Jun 21, 2025 at 06:54:17PM +0800, Yang Xiwen wrote: > > > Currently U-Boot clk framework mandates the sequence of registering clks > > to begin from root, and then its children. This is an addtional > > requirement that doesn't exist in Linux kernel, which makes clk driver > > porting more difficult. > > > > Let's handle this entirely in clk framework and give driver the freedom > > to register clocks in any order. > > > > This is done by assigning the parent "lazily". It will cache the parent > > name in the core clk struct and try to refresh it when clk consumers > > call clk_get_parent(). It's transparent to clk consumers as long as they > > only use APIs provided by clk framework. > > > > I've ran `ut dm clk*` and verified these commits does not break any > > existing testcases. Though no new testcase is added because I think > > existing testcases already test the logic very well. > > Sorry for the very late review. When I build this on say > rock-4c-plus-rk3399 this is the size growth I see: > rock-4c-plus-rk3399: all +228 spl/u-boot-spl:all +500 > spl/u-boot-spl:text +500 text +228 tpl/u-boot-tpl:all +552 > tpl/u-boot-tpl:text +552 > u-boot: add: 1/0, grow: 1/0 bytes: 228/0 (228) > function old new > delta > device_reparent - 120 > +120 > clk_get_parent 56 164 > +108 > spl-u-boot-spl: add: 4/0, grow: 1/0 bytes: 500/0 (500) > function old new > delta > uclass_find_device_by_namelen - 172 > +172 > device_reparent - 120 > +120 > clk_get_parent 56 144 > +88 > uclass_find_device_by_name - 68 > +68 > uclass_get_device_by_name - 52 > +52 > tpl-u-boot-tpl: add: 5/0, grow: 1/0 bytes: 552/0 (552) > function old new > delta > uclass_find_device_by_namelen - 172 > +172 > device_reparent - 120 > +120 > clk_get_parent 56 144 > +88 > uclass_find_device_by_name - 68 > +68 > uclass_get_device_by_name - 52 > +52 > strncmp - 52 > +52 > > (using buildman's 'bloat' features to report size change). I'm fine with > the growth in U-Boot itself, but that's just too much in SPL and TPL and > we need to figure out if we can avoid that somehow.
You could add a Kconfig for this option, like CLK_LAZY_PARENT and then use CONFIG_IS_ENABLED() so it is separately enabled in SPL, and default off. Also please add a test case to test/dm/clk.c (or clk_ccf.c) that covers your changes. Regards, Simon
