On 01-02-21, 10:44, Dominik Brodowski wrote:

IIRC, it was required on various ARM systems,[*] as CPUs were registered as
subsys_initcall(), while cpufreq used to be initialized only later, as an

s/later/earlier ? arch happens before subsys not at least and that is
the only way we can break cpufreq here, i.e. when the driver comes up
before the CPUs are registered.

arch_initcall(). If the ordering is opposite now on all architectures (it
wasn't on ARM back then), we should be fine.

[*] https://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git/commit/arch/arm/mach-sa1100/cpu-sa1100.c?id=f59d3bbe35f6268d729f51be82af8325d62f20f5

Thanks for your reply, it made me look at that aspect in some more
detail to confirm I don't end up breaking anything. Unless I am making
a mistake in reading the code, this is the code flow that we have
right now:

start_kernel()
-> kernel_init()
   -> kernel_init_freeable()
      -> do_basic_setup()
         -> driver_init()
            -> cpu_dev_init()
               -> subsys_system_register(for-CPUs)
 
         -> do_initcalls()
            -> register-cpufreq-driver from any level

And so CPUs should always be there for a cpufreq driver.

Makes sense ?

-- 
viresh