2022-07-27 03:16:44

by Yu Zhe

Subject: [PATCH] x86/aperfmperf: use time_is_before_jiffies(a + b) to replace "jiffies - a > b"

time_is_before_jiffies deals with timer wrapping correctly.

Signed-off-by: Yu Zhe <[email protected]>
---
arch/x86/kernel/cpu/aperfmperf.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/kernel/cpu/aperfmperf.c b/arch/x86/kernel/cpu/aperfmperf.c
index 1f60a2b27936..22e0bac3fffe 100644
--- a/arch/x86/kernel/cpu/aperfmperf.c
+++ b/arch/x86/kernel/cpu/aperfmperf.c
@@ -423,7 +423,7 @@ unsigned int arch_freq_get_on_cpu(int cpu)
* Bail on invalid count and when the last update was too long ago,
* which covers idle and NOHZ full CPUs.
*/
- if (!mcnt || (jiffies - last) > MAX_SAMPLE_AGE)
+ if (!mcnt || time_is_before_jiffies(last + MAX_SAMPLE_AGE))
goto fallback;

return div64_u64((cpu_khz * acnt), mcnt);
--
2.11.0

2022-07-27 16:19:33

by Peter Zijlstra

Subject: Re: [PATCH] x86/aperfmperf: use time_is_before_jiffies(a + b) to replace "jiffies - a > b"

On Wed, Jul 27, 2022 at 11:14:05AM +0800, Yu Zhe wrote:
> time_is_before_jiffies deals with timer wrapping correctly.

Please explain how the current code does not.

>
> Signed-off-by: Yu Zhe <[email protected]>
> ---
> arch/x86/kernel/cpu/aperfmperf.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/kernel/cpu/aperfmperf.c b/arch/x86/kernel/cpu/aperfmperf.c
> index 1f60a2b27936..22e0bac3fffe 100644
> --- a/arch/x86/kernel/cpu/aperfmperf.c
> +++ b/arch/x86/kernel/cpu/aperfmperf.c
> @@ -423,7 +423,7 @@ unsigned int arch_freq_get_on_cpu(int cpu)
> * Bail on invalid count and when the last update was too long ago,
> * which covers idle and NOHZ full CPUs.
> */
> - if (!mcnt || (jiffies - last) > MAX_SAMPLE_AGE)
> + if (!mcnt || time_is_before_jiffies(last + MAX_SAMPLE_AGE))
> goto fallback;
>
> return div64_u64((cpu_khz * acnt), mcnt);
> --
> 2.11.0
>

2022-07-28 03:04:42

by Yu Zhe

Subject: Re: [PATCH] x86/aperfmperf: use time_is_before_jiffies(a + b) to replace "jiffies - a > b"

在 2022年07月28日 00:00, Peter Zijlstra 写道:

> On Wed, Jul 27, 2022 at 11:14:05AM +0800, Yu Zhe wrote:
>> time_is_before_jiffies deals with timer wrapping correctly.
> Please explain how the current code does not.

1. If the timer wrap changes in the future you won't have to alter your code.

2. unsigned long ut;

ut = ULONG_MAX + 4;

printf("time_after(ut, ULONG_MAX), ut:%d, %d --> %d\n", ut,
time_after(ut, ULONG_MAX), (ULONG_MAX - ut) < 0);



In this case, time_after returns true, it's correct.

>> Signed-off-by: Yu Zhe <[email protected]>
>> ---
>> arch/x86/kernel/cpu/aperfmperf.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kernel/cpu/aperfmperf.c b/arch/x86/kernel/cpu/aperfmperf.c
>> index 1f60a2b27936..22e0bac3fffe 100644
>> --- a/arch/x86/kernel/cpu/aperfmperf.c
>> +++ b/arch/x86/kernel/cpu/aperfmperf.c
>> @@ -423,7 +423,7 @@ unsigned int arch_freq_get_on_cpu(int cpu)
>> * Bail on invalid count and when the last update was too long ago,
>> * which covers idle and NOHZ full CPUs.
>> */
>> - if (!mcnt || (jiffies - last) > MAX_SAMPLE_AGE)
>> + if (!mcnt || time_is_before_jiffies(last + MAX_SAMPLE_AGE))
>> goto fallback;
>>
>> return div64_u64((cpu_khz * acnt), mcnt);
>> --
>> 2.11.0
>>
>

2022-07-28 10:39:28

by Peter Zijlstra

Subject: Re: [PATCH] x86/aperfmperf: use time_is_before_jiffies(a + b) to replace "jiffies - a > b"

On Thu, Jul 28, 2022 at 10:55:51AM +0800, Yu Zhe wrote:
> 在 2022年07月28日 00:00, Peter Zijlstra 写道:
>
> > On Wed, Jul 27, 2022 at 11:14:05AM +0800, Yu Zhe wrote:
> > > time_is_before_jiffies deals with timer wrapping correctly.
> > Please explain how the current code does not.
>
> 1. If the timer wrap changes in the future you won't have to alter your code.
>
> 2. unsigned long ut;
>
> ut = ULONG_MAX + 4;
>
> printf("time_after(ut, ULONG_MAX), ut:%d, %d --> %d\n", ut,
> time_after(ut, ULONG_MAX), (ULONG_MAX - ut) < 0);
>
>
> In this case, time_after returns true, it's correct.

Now try the same with the existing code...