Re: [PATCH v2] x86/tsc: Have tsc=recalibrate override things

[Feng Tang]

Add Paul as he also met similar issue before.

On Wed, Nov 01, 2023 at 12:16:21PM +0100, Peter Zijlstra wrote:
> Subject: x86/tsc: Have tsc=recalibrate override things
> From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Date: Mon, 30 Oct 2023 17:00:50 +0100
> 
> My brand-spanking new SPR supermicro workstation was reporting NTP
> failures:
> 
> Oct 30 13:00:26 spr ntpd[3517]: CLOCK: kernel reports TIME_ERROR: 0x41: Clock Unsynchronized
> Oct 30 13:00:58 spr ntpd[3517]: CLOCK: time stepped by 32.316775
> Oct 30 13:00:58 spr ntpd[3517]: CLOCK: frequency error 41699 PPM exceeds tolerance 500 PPM
> 
> CPUID provides:
> 
>     Time Stamp Counter/Core Crystal Clock Information (0x15):
>        TSC/clock ratio = 200/2
>        nominal core crystal clock = 25000000 Hz
>     Processor Frequency Information (0x16):
>        Core Base Frequency (MHz) = 0x9c4 (2500)
>        Core Maximum Frequency (MHz) = 0x12c0 (4800)
>        Bus (Reference) Frequency (MHz) = 0x64 (100)
> 
> and the kernel believes this. Since commit a7ec817d5542 ("x86/tsc: Add
> option to force frequency recalibration with HW timer") there is the
> tsc=recalibrate option, which forces the recalibrate.
> 
> This duely reports:
> 
> Oct 30 12:42:39 spr kernel: tsc: Warning: TSC freq calibrated by CPUID/MSR differs from what is calibrated by HW timer, please check with vendor!!
> Oct 30 12:42:39 spr kernel: tsc: Previous calibrated TSC freq:         2500.000 MHz
> Oct 30 12:42:39 spr kernel: tsc: TSC freq recalibrated by [HPET]:         2399.967 MHz
> 
> but then continues running at 2500, offering no solace and keeping NTP
> upset -- drifting ~30 seconds for every 15 minutes.
> 
> Have tsc=recalibrate override the CPUID provided numbers. This makes the
> machine usable and keeps NTP 'happy':
> 
> Oct 30 16:48:44 spr ntpd[2200]: CLOCK: time stepped by -0.768679
 
Hi Peter,

The patch looks fine to me.

One thought is the original semantics of  'tsc=recalibrate' is just
to recalibrate and print out the frequency, and with this change,
it adds 'override' semantics, so the description for 'recalibrate'
in kernel-parameters.txt may also need some update. 

Thanks,
Feng

> Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> ---
>  arch/x86/kernel/tsc.c |   15 ++++++---------
>  1 file changed, 6 insertions(+), 9 deletions(-)
> 
> --- a/arch/x86/kernel/tsc.c
> +++ b/arch/x86/kernel/tsc.c
> @@ -1430,14 +1430,13 @@ static void tsc_refine_calibration_work(
>  			hpet ? "HPET" : "PM_TIMER",
>  			(unsigned long)freq / 1000,
>  			(unsigned long)freq % 1000);
> +	} else {
>  
> -		return;
> +		/* Make sure we're within 1% */
> +		if (abs(tsc_khz - freq) > tsc_khz/100)
> +			goto out;
>  	}
>  
> -	/* Make sure we're within 1% */
> -	if (abs(tsc_khz - freq) > tsc_khz/100)
> -		goto out;
> -
>  	tsc_khz = freq;
>  	pr_info("Refined TSC clocksource calibration: %lu.%03lu MHz\n",
>  		(unsigned long)tsc_khz / 1000,
> @@ -1479,14 +1478,12 @@ static int __init init_tsc_clocksource(v
>  	 * When TSC frequency is known (retrieved via MSR or CPUID), we skip
>  	 * the refined calibration and directly register it as a clocksource.
>  	 */
> -	if (boot_cpu_has(X86_FEATURE_TSC_KNOWN_FREQ)) {
> +	if (boot_cpu_has(X86_FEATURE_TSC_KNOWN_FREQ) && !tsc_force_recalibrate) {
>  		if (boot_cpu_has(X86_FEATURE_ART))
>  			art_related_clocksource = &clocksource_tsc;
>  		clocksource_register_khz(&clocksource_tsc, tsc_khz);
>  		clocksource_unregister(&clocksource_tsc_early);
> -
> -		if (!tsc_force_recalibrate)
> -			return 0;
> +		return 0;
>  	}
>  
>  	schedule_delayed_work(&tsc_irqwork, 0);