host, bhyve vm and ntpd
Ian Lepore
ian at freebsd.org
Fri Oct 20 18:04:33 UTC 2017
On Fri, 2017-10-20 at 20:20 +0300, Boris Samorodov wrote:
> (CC to freebsd-virtualization@)
>
> 20.10.2017 19:32, Ian Lepore пишет:
> >
> > On Fri, 2017-10-20 at 18:36 +0300, Boris Samorodov wrote:
> > >
> > > 20.10.2017 18:31, Boris Samorodov пишет:
> > > >
> > > >
> > > > 20.10.2017 18:12, Ian Lepore пишет:
> > > > >
> > > > >
> > > > > On Fri, 2017-10-20 at 14:46 +0300, Boris Samorodov wrote:
> > > > > >
> > > > > >
> > > > > > Hi All,
> > > > > >
> > > > > > I have got a host:
> > > > > > ---
> > > > > > bhyve-host% uname -a
> > > > > > FreeBSD sm.bsnet 12.0-CURRENT FreeBSD 12.0-CURRENT #3 r322868: Fri Aug
> > > > > > 25 05:25:26 MSK 2017
> > > > > > bsam at builder.bsnet:/usr/obj/usr/src/sys/GENERIC-FAST amd64 amd64
> > > > > > ---
> > > > > >
> > > > > > And a bhyve vm:
> > > > > > ---
> > > > > > bhyve-vm: uname -a
> > > > > > FreeBSD builder.bsnet 12.0-CURRENT FreeBSD 12.0-CURRENT #58 r324782: Fri
> > > > > > Oct 20 05:12:17 MSK 2017
> > > > > > bsam at builder.bsnet:/usr/obj/usr/src/sys/PKG64X amd64 amd64
> > > > > > ---
> > > > > >
> > > > > > The only difference at kernel configs is a colored console. :-)
> > > > > >
> > > > > > And here I get some weird (is it?) result at the VM (I expect ntpd to be
> > > > > > more stable):
> > > > > > ---
> > > > > > bhyve-vm% for t in `jot 10`; do ntpq -pn; sleep 64; done
> > > > > > remote refid st t when poll reach delay offset
> > > > > > jitter
> > > > > > ==============================================================================
> > > > > > XX.XX.XX.1 XX.XX.XX.245 4 u 9 64 3 0.605 -1.202
> > > > > > 316.407
> > > > > > XX.XX.XX.1 XX.XX.XX.245 4 u 7 64 7 0.605 -1.202
> > > > > > 358.395
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 5 64 17 0.615 -328.42
> > > > > > 181.405
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 3 64 37 0.615 -328.42
> > > > > > 214.868
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 67 64 37 0.615 -328.42
> > > > > > 214.868
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 63 64 77 0.615 -328.42
> > > > > > 268.618
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 60 64 177 0.615 -328.42
> > > > > > 333.175
> > > > > > XX.XX.XX.1 .STEP. 16 u 1910 64 0 0.000 0.000
> > > > > > 0.000
> > > > > > XX.XX.XX.1 XX.XX.XX.245 4 u 27 64 1 0.703 -262.63
> > > > > > 0.004
> > > > > > XX.XX.XX.1 XX.XX.XX.245 4 u 31 64 1 0.649 -331.43
> > > > > > 68.800
> > > > > > ---
> > > > > >
> > > > > > At the same time host's results are very stable:
> > > > > > ---
> > > > > > bhyve-host% for t in `jot 10`; do ntpq -pn; sleep 64; done
> > > > > > remote refid st t when poll reach delay offset
> > > > > > jitter
> > > > > > ==============================================================================
> > > > > >
> > > > > >
> > > > > >
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 1 64 1 0.401 0.176
> > > > > > 0.106
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 6 64 3 0.401 0.176
> > > > > > 0.459
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 3 64 7 0.401 0.176
> > > > > > 0.940
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 67 64 7 0.401 0.176
> > > > > > 0.940
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 64 64 17 0.401 0.176
> > > > > > 1.566
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 60 64 37 0.448 1.275
> > > > > > 1.739
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 55 64 77 0.448 1.275
> > > > > > 2.365
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 53 64 177 0.448 1.275
> > > > > > 3.110
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 50 64 377 0.448 1.275
> > > > > > 3.929
> > > > > > *XX.XX.XX.1 XX.XX.XX.245 4 u 45 64 377 0.443 8.750
> > > > > > 4.722
> > > > > > ---
> > > > > >
> > > > > > The network is organized via bridge -- host igb and vm tap interfaces
> > > > > > are members of one bridge.
> > > > > >
> > > > > > Are those results expected? Does it smell like a bug? Should I dig
> > > > > > furter?
> > > > > >
> > > > > So it is repeatedly stepping the clock in the VM? (Set
> > > > > kern.timecounter.stepwarnings=1 to log steps).
> > > > No kernel/ntpd messages for 20 minutes after setting this sysctl.
> > > >
> > > > >
> > > > >
> > > > > That is usually a sign
> > > > > that the chosen timecounter is running at a different frequency than it
> > > > > claimed to be when it registered itself -- the host may not be
> > > > > emulating the timer hardware properly in the guest. What is the output
> > > > > of sysctl kern.timecounter in the vm?
> > > > ---
> > > > bhyve-vm% sysctl kern.timecounter
> > > >
> > > > kern.timecounter.tsc_shift: 1
> > > > kern.timecounter.smp_tsc_adjust: 0
> > > > kern.timecounter.smp_tsc: 0
> > > > kern.timecounter.invariant_tsc: 1
> > > > kern.timecounter.fast_gettime: 1
> > > > kern.timecounter.tick: 1
> > > > kern.timecounter.choice: ACPI-fast(900) HPET(950) i8254(0) TSC-low(-100)
> > > > dummy(-1000000)
> > > > kern.timecounter.hardware: HPET
> > > > kern.timecounter.alloweddeviation: 5
> > > > kern.timecounter.stepwarnings: 1
> > > > kern.timecounter.tc.ACPI-fast.quality: 900
> > > > kern.timecounter.tc.ACPI-fast.frequency: 3579545
> > > > kern.timecounter.tc.ACPI-fast.counter: 4161213491
> > > > kern.timecounter.tc.ACPI-fast.mask: 4294967295
> > > > kern.timecounter.tc.HPET.quality: 950
> > > > kern.timecounter.tc.HPET.frequency: 10000000
> > > > kern.timecounter.tc.HPET.counter: 3518036865
> > > > kern.timecounter.tc.HPET.mask: 4294967295
> > > > kern.timecounter.tc.i8254.quality: 0
> > > > kern.timecounter.tc.i8254.frequency: 1193182
> > > > kern.timecounter.tc.i8254.counter: 47597
> > > > kern.timecounter.tc.i8254.mask: 65535
> > > > kern.timecounter.tc.TSC-low.quality: -100
> > > > kern.timecounter.tc.TSC-low.frequency: 1199886114
> > > > kern.timecounter.tc.TSC-low.counter: 1274338278
> > > > kern.timecounter.tc.TSC-low.mask: 4294967295
> > > > ---
> > > BTW, here is the host kern.timecounter:
> > > ---
> > > kern.timecounter.tsc_shift: 1
> > > kern.timecounter.smp_tsc_adjust: 0
> > > kern.timecounter.smp_tsc: 1
> > > kern.timecounter.invariant_tsc: 1
> > > kern.timecounter.fast_gettime: 1
> > > kern.timecounter.tick: 1
> > > kern.timecounter.choice: ACPI-fast(900) HPET(950) i8254(0) TSC-low(1000)
> > > dummy(-1000000)
> > > kern.timecounter.hardware: TSC-low
> > > kern.timecounter.alloweddeviation: 5
> > > kern.timecounter.stepwarnings: 0
> > > kern.timecounter.tc.ACPI-fast.quality: 900
> > > kern.timecounter.tc.ACPI-fast.frequency: 3579545
> > > kern.timecounter.tc.ACPI-fast.counter: 9047194
> > > kern.timecounter.tc.ACPI-fast.mask: 16777215
> > > kern.timecounter.tc.HPET.quality: 950
> > > kern.timecounter.tc.HPET.frequency: 14318180
> > > kern.timecounter.tc.HPET.counter: 2232552795
> > > kern.timecounter.tc.HPET.mask: 4294967295
> > > kern.timecounter.tc.i8254.quality: 0
> > > kern.timecounter.tc.i8254.frequency: 1193182
> > > kern.timecounter.tc.i8254.counter: 43410
> > > kern.timecounter.tc.i8254.mask: 65535
> > > kern.timecounter.tc.TSC-low.quality: 1000
> > > kern.timecounter.tc.TSC-low.frequency: 1200067168
> > > kern.timecounter.tc.TSC-low.counter: 2463146362
> > > kern.timecounter.tc.TSC-low.mask: 4294967295
> > > ---
> > >
> > > >
> > > >
> > > > >
> > > > >
> > > > > Also, what is the output of ntptime(8) in the vm?
> > > > ---
> > > > bhyve-vm% ntptime
> > > >
> > > > ntp_gettime() returns code 0 (OK)
> > > > time dd94930f.20ea2900 Fri, Oct 20 2017 18:21:51.128, (.128573699),
> > > > maximum error 1309110 us, estimated error 3 us, TAI offset 37
> > > > ntp_adjtime() returns code 0 (OK)
> > > > modes 0x0 (),
> > > > offset 0.000 us, frequency 0.000 ppm, interval 1 s,
> > > > maximum error 1309110 us, estimated error 3 us,
> > > > status 0x2001 (PLL,NANO),
> > > > time constant 6, precision 0.001 us, tolerance 496 ppm,
> > > > ---
> > > >
> > > > Ian, thank you for your help!
> > > >
> > It seems odd to me that the frequency of the host HPET is 14.3mhz and
> > that of the guest is 10.0mhz, but maybe that's a normal condition for
> > bhyve. I did find some google hits[1] for bhyve guest timekeeping
> > trouble with the HPET timer which was solved by switching to a
> > different timecounter. Timecounter choices can't be controlled from
> > loader.conf, so I guess a sysctl.conf entry of
> > kern.timecounter.hardware="ACPI-fast" is the only way to fix that. You
> > can also just do that command interactively first and see if it stops
> > the time steps and ntp settles down.
> The process seems to become more monotonic. But steps nevertheless:
> ---
> *XX.1 XX.245 4 u 20 64 1 0.717 -12.771 4.193
> *XX.1 XX.245 4 u 28 64 3 0.751 -41.970 32.342
> *XX.1 XX.245 4 u 23 64 7 0.748 -59.505 46.624
> *XX.1 XX.245 4 u 18 64 17 0.699 -75.164 56.482
> *XX.1 XX.245 4 u 14 64 37 0.669 -90.112 63.767
> *XX.1 XX.245 4 u 11 64 77 0.605 -10.567 60.914
> *XX.1 XX.245 4 u 7 64 177 0.591 -169.54 116.762
> *XX.1 XX.245 4 u 3 64 377 0.591 -169.54 102.107
> *XX.1 XX.245 4 u 68 64 377 0.591 -169.54 102.107
> *XX.1 XX.245 4 u 63 64 377 0.591 -169.54 88.424
> *XX.1 XX.245 4 u 58 64 377 0.591 -169.54 92.949
> *XX.1 XX.245 4 u 55 64 377 0.591 -169.54 111.512
> *XX.1 XX.245 4 u 50 64 377 0.591 -169.54 140.827
> *XX.1 XX.245 4 u 45 64 377 0.591 -169.54 177.360
> *XX.1 XX.245 4 u 43 64 377 0.591 -169.54 219.057
> XX.1 .STEP. 16 u 588 64 0 0.000 0.000 0.000
> ---
>
> >
> > This would be a workaround, not a fix per se. If the time steps go
> > away, then something in bhyve's emulation of HPET (maybe only on some
> > hardware?) must be buggy.
> >
> >
> > [1] https://lists.freebsd.org/pipermail/freebsd-virtualization/2015-April/003492.html
> Also thanks for the link. Unfortunately the problem seems to persist.
>
Hmm, that almost looks normal... it noticed the clock was drifting
fast, so it went into a frequency-training cycle (at the point where
the offset stopped changing at -169.54) and then once it had figured
out the frequency it did a step to get realigned, and (presumably)
adjusted the frequency of the kernel clock. The output of ntptime
before and after the step would show that... before the step the
frequency would show as zero (which was the case in your ntptime output
earlier), and after the step it would be non-zero. No more steps would
occur after the first one, if that's what is happening here.
-- Ian
More information about the freebsd-virtualization
mailing list