lkml.org 
[lkml]   [2021]   [Jul]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] use 64bit timer for hpet
From
Date
Gleixner,

> Seriously? The wrap-around time for 32bit HPET @24MHz is ~3 minutes.

In some cases, our system will be very busy, and the timeout of 3 minutes
is not an exaggeration. Then, the system considers that the tsc clock is
inaccurate and switches the tsc clock to the hpet clock, which brings
greater performance overhead.

> Aside of that the reason why the kernel does not support 64bit HPET is
> that there are HPETs which advertise 64bit support, but the
> implementation is buggy.

Can you tell me what is the buggy with the 64-bit hpet clock? In my opinion,
it is unreasonable to use a lower-bit width clock to calibrate a higher-bit width
clock, and the hardware already supports the higher-bit width.


> 2021年7月7日 下午6:04,Thomas Gleixner <tglx@linutronix.de> 写道:
>
> Liao,
>
> On Fri, Jul 02 2021 at 16:13, zhaoyan liao wrote:
>> The kernel judges whether the tsc clock is accurate in the
>> clocksource_watchdog background thread function. The hpet clock source
>> is 32-bit, but tsc is 64-bit. Therefore, when the system is busy and the
>> clocksource_watchdog cannot be scheduled in time, the hpet clock may
>> overflow and cause the system to misjudge tsc as unreliable.
>
> Seriously? The wrap-around time for 32bit HPET @24MHz is ~3 minutes.
>
>> In this case, we recommend that the kernel adopts the 64-bit hpet clock
>> by default to keep the width of the two clock sources the same to reduce
>> misjudgment. Some CPU models may not support 64-bit hpet, but according
>> to the description of the CPU's register manual, it does not affect our
>> reading action.
>
> So much for the theory.
>
>> -#define HPET_MASK CLOCKSOURCE_MASK(32)
>> +#define HPET_MASK CLOCKSOURCE_MASK(64)
>
> How is that valid for a 32bit HPET? This breaks the clocksource.
>
>> +inline unsigned long hpet_readq(unsigned int a)
>> +{
>> + return readq(hpet_virt_address + a);
>
> Breaks 32bit build immediately.
>
> Aside of that the reason why the kernel does not support 64bit HPET is
> that there are HPETs which advertise 64bit support, but the
> implementation is buggy.
>
> IOW, while this works for your hardware this breaks quite some parts of
> the universe. Not really a good approach.
>
> Thanks,
>
> tglx

\
 
 \ /
  Last update: 2021-07-08 05:12    [W:0.120 / U:0.260 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site