Re: [PATCH v3 7/7] clocksource/drivers/arm_arch_timer: validate arch_timer_rate

[Lukasz Luba <lukasz.luba@arm.com>]

On 2/12/20 11:10 AM, Marc Zyngier wrote:
> On 2020-02-12 10:55, Lukasz Luba wrote:
>> On 2/12/20 10:12 AM, Marc Zyngier wrote:
>>> On 2020-02-12 10:01, Lukasz Luba wrote:
>>>> Hi Ionela, Valentin
>>>>
>>>> On 2/11/20 6:45 PM, Ionela Voinescu wrote:
>>>>> From: Valentin Schneider <valentin.schneider@arm.com>
>>>>>
>>>>> Using an arch timer with a frequency of less than 1MHz can result 
>>>>> in an
>>>>> incorrect functionality of the system which assumes a reasonable rate.
>>>>>
>>>>> One example is the use of activity monitors for frequency invariance
>>>>> which uses the rate of the arch timer as the known rate of the 
>>>>> constant
>>>>> cycle counter in computing its ratio compared to the maximum frequency
>>>>> of a CPU. For arch timer frequencies less than 1MHz this ratio could
>>>>> end up being 0 which is an invalid value for its use.
>>>>>
>>>>> Therefore, warn if the arch timer rate is below 1MHz which contravenes
>>>>> the recommended architecture interval of 1 to 50MHz.
>>>>>
>>>>> Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
>>>>> Cc: Mark Rutland <mark.rutland@arm.com>
>>>>> Cc: Marc Zyngier <maz@kernel.org>
>>>>> ---
>>>>>   drivers/clocksource/arm_arch_timer.c | 18 +++++++++++++++---
>>>>>   1 file changed, 15 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/drivers/clocksource/arm_arch_timer.c 
>>>>> b/drivers/clocksource/arm_arch_timer.c
>>>>> index 9a5464c625b4..4faa930eabf8 100644
>>>>> --- a/drivers/clocksource/arm_arch_timer.c
>>>>> +++ b/drivers/clocksource/arm_arch_timer.c
>>>>> @@ -885,6 +885,17 @@ static int arch_timer_starting_cpu(unsigned 
>>>>> int cpu)
>>>>>       return 0;
>>>>>   }
>>>>>   +static int validate_timer_rate(void)
>>>>> +{
>>>>> +    if (!arch_timer_rate)
>>>>> +        return -EINVAL;
>>>>> +
>>>>> +    /* Arch timer frequency < 1MHz can cause trouble */
>>>>> +    WARN_ON(arch_timer_rate < 1000000);
>>>>
>>>> I don't see a big value of having a patch just to add one extra 
>>>> warning,
>>>> in a situation which we handle in our code with in 6/7 with:
>>>>
>>>> +    if (!ratio) {
>>>> +        pr_err("System timer frequency too low.\n");
>>>> +        return -EINVAL;
>>>> +    }
>>>>
>>>> Furthermore, the value '100000' here is because of our code and
>>>> calculation in there, so it does not belong to arch timer. Someone
>>>> might ask why it's not 200000 or a define in our header...
>>>> Or questions asking why do you warn when that arch timer and cpu is not
>>>> AMU capable...
>>>
>>> Because, as the commit message outlines it, such a frequency is terribly
>>> out of spec?
>>
>> I don't see in the RM that < 1MHz is terribly out of spec.
>> 'Frequency
>> Increments at a fixed frequency, typically in the range 1-50MHz.
>> Can support one or more alternative operating modes in which it
>> increments by larger amounts at a
>> lower frequency, typically for power-saving.'
> 
> Hint: constant apparent frequency.
> 
>> There is even an example how to operate at 20kHz and increment by 500.
>>
>> I don't know the code if it's supported, thought.
> 
> You're completely missing the point, I'm afraid. Nobody has to know that
> this is happening. For all intent and purposes, the counter has always
> the same frequency, even if the HW does fewer ticks of larger increments.

Fair enough. As I said I don't know details of that code.

> 
> 
> [...]
> 
>>>> Lastly, this is arch timer.
>>>> To increase chances of getting merge soon, I would recommend to drop
>>>> the patch from this series.
>>>
>>> And? It seems to address a potential issue where the time frequency
>>> is out of spec, and makes sure we don't end up with additional problems
>>> in the AMU code.
>>
>> This patch just prints warning, does not change anything in booting or
>> in any code related to AMU.
> 
> It seems to solve an issue with an assumption made in the AMU driver,
> and would help debugging the problem on broken systems. Are you saying
> that this is not the case and that the AMU code can perfectly cope with
> the frequency being less than 1MHz?

What I was saying is that patch 6/7 has the code which checks the rate
and reacts, so it does not need this patch. In case of helping with
debugging, the patch 6/7 also prints error
"System timer frequency too low" and bails out.
The commit message could have better emphasize it.

Regards,
Lukasz