linux-kernel - Re: [PATCH clocksource 1/3] clocksource: Reject bogus watchdog clocksource measurements

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:   Sat, 3 Dec 2022 10:51:48 +0800
From:   Feng Tang <feng.tang@...el.com>
To:     "Paul E. McKenney" <paulmck@...nel.org>
CC:     Thomas Gleixner <tglx@...utronix.de>,
        <linux-kernel@...r.kernel.org>, <sboyd@...nel.org>,
        <corbet@....net>, <Mark.Rutland@....com>, <maz@...nel.org>,
        <kernel-team@...a.com>, <ak@...ux.intel.com>,
        <zhengjun.xing@...el.com>, Chris Mason <clm@...a.com>,
        John Stultz <jstultz@...gle.com>,
        Waiman Long <longman@...hat.com>
Subject: Re: [PATCH clocksource 1/3] clocksource: Reject bogus watchdog
 clocksource measurements

On Fri, Dec 02, 2022 at 02:24:02PM -0800, Paul E. McKenney wrote:
[...]
> > > > > > > (My next step involves a pillow, but will follow up tomorrow morning
> > > > > > > Pacific Time.)
> > > > > >  
> > > > > > Really thanks for checking this through late night!
> > > > > 
> > > > > No problem, and I guess it is instead the day after tomorrow, but
> > > > > I thought you might be interested in chronyd's opinion:
> > > > > 
> > > > > [root@...test1029.snc8 ~]# cat /var/lib/chrony/drift
> > > > >         40001.074911             0.002098
> > > > > 
> > > > > In contrast, on my Fedora laptop:
> > > > > 
> > > > > $ sudo cat /var/lib/chrony/drift
> > > > >             2.074313             0.186606
> > > > > 
> > > > > I am (perhaps incorrectly) taking this to indicate that TSC is in fact
> > > > > drifting with respect to standard time.
> > > >  
> > > > This info is very useful! It further confirms the CPUID(0x15) gave
> > > > the wrong frequency info. 
> > > 
> > > So the TSC is just doing what it is told.  ;-)
> > > 
> > > This indicates a firmware problem?
> > > 
> > > > Also I don't think TSC itself is drifting, and the drift some from
> > > > the wrong match calculation(1896 MHz), if we give it the correct
> > > > number (likely 1975 MHz here), there shouldn't be big chrony drift
> > > > like your Fedora laptop.
> > > 
> > > Resetting so that the clocksource watchdog once again gets rid of TSC,
> > > but leaving nohpet:
> > > 
> > > [    0.000000] tsc: using CPUID[0x15] crystal_khz= 24000 kHz ebx=158 eax=2
> > > [    0.000000] tsc: Detected 1900.000 MHz processor
> > > [    0.000000] tsc: Detected 1896.000 MHz TSC
> > > [    5.287750] clocksource: refined-jiffies: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 1910969940391419 ns
> > > [   17.963947] clocksource: tsc-early: mask: 0xffffffffffffffff max_cycles: 0x36a8d32ce31, max_idle_ns: 881590731004 ns
> > > [   19.996287] clocksource: timekeeping watchdog on CPU3: Marking clocksource 'tsc-early' as unstable because the skew is too large:
> > > [   20.040287] clocksource:                       'refined-jiffies' wd_nsec: 503923392 wd_now: fffb73f8 wd_last: fffb7200 mask: ffffffff
> > > [   20.067286] clocksource:                       'tsc-early' cs_nsec: 588021368 cs_now: 581c1eb378 cs_last: 57d9aad9e8 mask: ffffffffffffffff
> > >  [   20.096286] clocksource:                       No current clocksource.
> > >  [   20.111286] tsc: Marking TSC unstable due to clocksource watchdog
> > >  [   24.582541] clocksource: jiffies: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 1911260446275000 ns
> > >  [   49.739301] clocksource: Switched to clocksource refined-jiffies
> > >  [   50.046356] clocksource: acpi_pm: mask: 0xffffff max_cycles: 0xffffff, max_idle_ns: 2085701024 ns
> > >  [   50.066475] clocksource: Switched to clocksource acpi_pm
> > > 
> > > # cat /var/lib/chrony/drift 
> > >             1.372570             0.020049
> > > 
> > > I interpret this to mean that acpi_pm (and thus from prior observations,
> > > HPET as well) are counting at the correct rate.
> > 
> > Correct. And this is a good news! that 1975 MHz seems to be the right
> > number.
> > 
> > Could you try below patch, it should override the CPUID calculation
> > and forced to use HW timer calibrated number:
> > 
> > ---
> > diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
> > index a78e73da4a74..68a2fea4961d 100644
> > --- a/arch/x86/kernel/tsc.c
> > +++ b/arch/x86/kernel/tsc.c
> > @@ -1417,7 +1417,8 @@ static int __init init_tsc_clocksource(void)
> >  	 * 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)) {
> > +	if (false) {
> >  		if (boot_cpu_has(X86_FEATURE_ART))
> >  			art_related_clocksource = &clocksource_tsc;
> >  		clocksource_register_khz(&clocksource_tsc, tsc_khz);
> > ---
> > 
> > And together with b50db7095fe0 "x86/tsc: Disable clocksource watchdog
> > for TSC on qualified platorms". I assume this will tell TSC to use
> > 1975 MHZ as its frequency.
> 
> This did not change things,

Maybe the kernel is with the tsc reclibrate patch? it will not use
the reclibrated frequency for a KNOWN_FREQ case, but just print
out the info. Anyway this doesn't matter, as it's just for debug,
and your way already proved the 1975 MHz is the right number :)

> but when I hardcoded tsc_khz to 1975, the
> clocksource watchdog no longer disables TSC and chrony shows drifts of
> less than 2.0.  (As opposed to about 40,000 without the hardcoding.)
> 
> So yes, forcing 1975 makes TSC work nicely.  Yay!  ;-)

Great to know it works and we have some workaround before a final
solution :)

Thanks,
Feng

> 							Thanx, Paul