On 26/08/2020 16:55:14+0800, dinghao.liu@xxxxxxxxxx wrote: > > On Sun, Aug 23, 2020 at 3:59 PM Dinghao Liu <dinghao.liu@xxxxxxxxxx> wrote: > > > > > > When clk_hw_register_fixed_rate_with_accuracy() fails, > > > clk_data should be freed. It's the same for the subsequent > > > error paths. > > > > I suppose you should also unregister the already registered clocks > > in the latter two error paths? > > > > Sounds reasonable. But I find that the existing kernel code takes different > strategies for this case. of_sama5d4_sckc_setup() uses clk_hw_unregister() > after clk_hw_register_fixed_rate_with_accuracy(), while _of_fixed_clk_setup() > uses clk_hw_unregister_fixed_rate(). But at91sam926x_pmc_setup() just does > nothing in this case. I guess you should use clk_hw_unregister_fixed_rate after clk_hw_register_fixed_rate_with_accuracy. clk_hw_unregister will leak the struct clk_fixed_rate. It doesn't matter too much for of_sama5d4_sckc_setup and at91sam926x_pmc_setup because if th clock can't be registered, the platform will not boot. > > Also, tcon_ch1_setup() uses clk_unregister() after clk_register(), while > clk_register_vco_pll() just does nothing. > > So I'm not sure if we should register here and which unregister function to > use. Would you please give me more specific advice about this problem? > > Regards, > Dinghao -- Alexandre Belloni, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
