Re: [PATCH 08/32] riscv: Fix kernel time_init()
From: Stephen Boyd <sboyd@kernel.org>
Date: 2020-11-13 08:11:23
Also in:
linux-clk, linux-riscv
Quoting Damien Le Moal (2020-11-12 23:57:19)
On 2020/11/13 16:53, Stephen Boyd wrote:quoted
Yes it's fine. Just the commit text reads as "If of_clk_init() is not called in time_init() then nothing works" which is true but made me wonder if it was because it needed to be super early or not. The commit text could be a little clearer here.OK. I will clarify the commit message in V2. Working on it now.
Thanks!
quoted
We don't have any good solution for a fallback to call of_clk_init() somewhere later. I do wonder if we should generalize this though and call of_clk_init() from start_kernel() directly via some Kconfig that architectures select if they need it for their timer and then move it to an initcall if architectures don't select the config. Or throw it into the of_platform_default_populate_init() hook if the architecture doesn't need to call it early.This last idea seems reasonable and probably the easiest. And I think it could be done unconditionally even if the arch calls of_clk_init() early as the already populated clock provider nodes would not be initialized again.
Also of_clk_init() is for the CLK_OF_DECLARE() users and if they can wait to platform bus population time then they could be converted to regular old platform device drivers. Maybe the problem is the clk driver you're using is only using CLK_OF_DECLARE() and not registering a platform driver?