On 20/11/2018 14:40, Pavel Tatashin wrote:
>>> +static __init void sched_clock_early_init(void)
>>> +{
>>> + u64 freq = arch_timer_get_cntfrq();
>>> + u64 (*read_time)(void) = arch_counter_get_cntvct;
>>
>> We already have arch_timer_read_counter which is exposed from
>> arm_arch_timer.h.
>
> OK
>
>>
>>> +
>>> + /* Early clock is available only on platforms with known freqs */
>>
>> This comment is misleading. It should read something like:
>>
>> /*
>> * The arm64 boot protocol mandates that CNTFRQ_EL0 reflects
>> * the timer frequency. To avoid breakage on misconfigured
>> * systems, do not register the early sched_clock if the
>> * programmed value if zero. Other random values will just
>> * result in random output.
>> */
>>
>
> OK
>
>>> + if (!freq)
>>> + return;
>>> +
>>> + sched_clock_register(read_time, BITS_PER_LONG, freq);
>>
>> This doesn't seem right. The counter has an architected minimum of 56
>> bits, and you can't assume that it is going to be more than that.
>
> Yeah, I saw 56 is used in arm_arch_timer.c, but I could not find where
> this minimum is defined in aarch64 specs. I will change it to 56.
See D10.1.2 (The system counter) in the ARM ARM[1], which says:
<quote>
The Generic Timer provides a system counter with the following
specification:
Width At least 56 bits wide.
The value returned by any 64-bit read of the counter is
zero-extended to 64 bits.
</quote>
This is not AArch64-specific though, and 32bit systems implementing the
generic timers have the exact same requirements.
> I will send v2 soon.
Please wait a bit and give others a chance to review this too.
Thanks,
M.
[1] https://static.docs.arm.com/ddi0487/da/DDI0487D_a_armv8_arm.pdf
--
Jazz is not dead. It just smells funny...
