Re: [PATCH v14 clocksource 4/6] clocksource: Reduce clocksource-skew threshold for TSC

From: Paul E. McKenney
Date: Tue May 11 2021 - 23:51:59 EST


On Wed, May 12, 2021 at 10:18:49AM +0800, Feng Tang wrote:
> Hi Paul,
>
> On Tue, May 11, 2021 at 04:34:53PM -0700, Paul E. McKenney wrote:
> > Currently, WATCHDOG_THRESHOLD is set to detect a 62.5-millisecond skew in
> > a 500-millisecond WATCHDOG_INTERVAL. This requires that clocks be skewed
> > by more than 12.5% in order to be marked unstable. Except that a clock
> > that is skewed by that much is probably destroying unsuspecting software
> > right and left. And given that there are now checks for false-positive
> > skews due to delays between reading the two clocks, it should be possible
> > to greatly decrease WATCHDOG_THRESHOLD, at least for fine-grained clocks
> > such as TSC.
> >
> > Therefore, add a new uncertainty_margin field to the clocksource
> > structure that contains the maximum uncertainty in nanoseconds for
> > the corresponding clock. This field may be initialized manually,
> > as it is for clocksource_tsc_early and clocksource_jiffies, which
> > is copied to refined_jiffies. If the field is not initialized
> > manually, it will be computed at clock-registry time as the period
> > of the clock in question based on the scale and freq parameters to
> > __clocksource_update_freq_scale() function. If either of those two
> > parameters are zero, the tens-of-milliseconds WATCHDOG_THRESHOLD is
> > used as a cowardly alternative to dividing by zero. No matter how the
> > uncertainty_margin field is calculated, it is bounded below by twice
> > WATCHDOG_MAX_SKEW, that is, by 100 microseconds.
> >
> > Note that manually initialized uncertainty_margin fields are not adjusted,
> > but there is a WARN_ON_ONCE() that triggers if any such field is less than
> > twice WATCHDOG_MAX_SKEW. This WARN_ON_ONCE() is intended to discourage
> > production use of the one-nanosecond uncertainty_margin values that are
> > used to test the clock-skew code itself.
> >
> > The actual clock-skew check uses the sum of the uncertainty_margin fields
> > of the two clocksource structures being compared. Integer overflow is
> > avoided because the largest computed value of the uncertainty_margin
> > fields is one billion (10^9), and double that value fits into an
> > unsigned int. However, if someone manually specifies (say) UINT_MAX,
> > they will get what they deserve.
> >
> > Note that the refined_jiffies uncertainty_margin field is initialized to
> > TICK_NSEC, which means that skew checks involving this clocksource will
> > be sufficently forgiving. In a similar vein, the clocksource_tsc_early
> > uncertainty_margin field is initialized to 32*NSEC_PER_MSEC, which
> > replicates the current behavior and allows custom setting if needed
> > in order to address the rare skews detected for this clocksource in
> > current mainline.
> >
> > Link: https://lore.kernel.org/lkml/202104291438.PuHsxRkl-lkp@xxxxxxxxx/
> > Link: https://lore.kernel.org/lkml/20210429140440.GT975577@paulmck-ThinkPad-P17-Gen-1
> > Link: https://lore.kernel.org/lkml/20210425224540.GA1312438@paulmck-ThinkPad-P17-Gen-1/
> > Link: https://lore.kernel.org/lkml/20210420064934.GE31773@xsang-OptiPlex-9020/
> > Link: https://lore.kernel.org/lkml/20210106004013.GA11179@paulmck-ThinkPad-P72/
> > Link: https://lore.kernel.org/lkml/20210414043435.GA2812539@paulmck-ThinkPad-P17-Gen-1/
> > Link: https://lore.kernel.org/lkml/20210419045155.GA596058@paulmck-ThinkPad-P17-Gen-1/
> > Suggested-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > Cc: John Stultz <john.stultz@xxxxxxxxxx>
> > Cc: Stephen Boyd <sboyd@xxxxxxxxxx>
> > Cc: Jonathan Corbet <corbet@xxxxxxx>
> > Cc: Mark Rutland <Mark.Rutland@xxxxxxx>
> > Cc: Marc Zyngier <maz@xxxxxxxxxx>
> > Cc: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> > Cc: Xing Zhengjun <zhengjun.xing@xxxxxxxxxxxxxxx>
> > Cc: Feng Tang <feng.tang@xxxxxxxxx>
> > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > ---
> > arch/x86/kernel/tsc.c | 1 +
> > include/linux/clocksource.h | 3 +++
> > kernel/time/clocksource.c | 48 +++++++++++++++++++++++++++++--------
> > kernel/time/jiffies.c | 15 ++++++------
> > 4 files changed, 50 insertions(+), 17 deletions(-)
> >
> > diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
> > index 6eb1b097e97e..2e076a459a0c 100644
> > --- a/arch/x86/kernel/tsc.c
> > +++ b/arch/x86/kernel/tsc.c
> > @@ -1128,6 +1128,7 @@ static int tsc_cs_enable(struct clocksource *cs)
> > static struct clocksource clocksource_tsc_early = {
> > .name = "tsc-early",
> > .rating = 299,
> > + .uncertainty_margin = 32 * NSEC_PER_MSEC,
> > .read = read_tsc,
> > .mask = CLOCKSOURCE_MASK(64),
> > .flags = CLOCK_SOURCE_IS_CONTINUOUS |
> > diff --git a/include/linux/clocksource.h b/include/linux/clocksource.h
> > index 7f83d51c0fd7..895203727cb5 100644
> > --- a/include/linux/clocksource.h
> > +++ b/include/linux/clocksource.h
> > @@ -43,6 +43,8 @@ struct module;
> > * @shift: Cycle to nanosecond divisor (power of two)
> > * @max_idle_ns: Maximum idle time permitted by the clocksource (nsecs)
> > * @maxadj: Maximum adjustment value to mult (~11%)
> > + * @uncertainty_margin: Maximum uncertainty in nanoseconds per half second.
> > + * Zero says to use default WATCHDOG_THRESHOLD.
> > * @archdata: Optional arch-specific data
> > * @max_cycles: Maximum safe cycle value which won't overflow on
> > * multiplication
> > @@ -98,6 +100,7 @@ struct clocksource {
> > u32 shift;
> > u64 max_idle_ns;
> > u32 maxadj;
> > + u32 uncertainty_margin;
> > #ifdef CONFIG_ARCH_CLOCKSOURCE_DATA
> > struct arch_clocksource_data archdata;
> > #endif
> > diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c
> > index 66243da2dadb..9ebf9931f3d6 100644
> > --- a/kernel/time/clocksource.c
> > +++ b/kernel/time/clocksource.c
> > @@ -95,6 +95,20 @@ static char override_name[CS_NAME_LEN];
> > static int finished_booting;
> > static u64 suspend_start;
> >
> > +/*
> > + * Threshold: 0.0312s, when doubled: 0.0625s.
> > + * Also a default for cs->uncertainty_margin when registering clocks.
> > + */
> > +#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
> > +
> > +/*
> > + * Maximum permissible delay between two readouts of the watchdog
> > + * clocksource surrounding a read of the clocksource being validated.
> > + * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as
> > + * a lower bound for cs->uncertainty_margin values when registering clocks.
> > + */
> > +#define WATCHDOG_MAX_SKEW (50 * NSEC_PER_USEC)
> > +
> > #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
> > static void clocksource_watchdog_work(struct work_struct *work);
> > static void clocksource_select(void);
> > @@ -121,17 +135,9 @@ static int clocksource_watchdog_kthread(void *data);
> > static void __clocksource_change_rating(struct clocksource *cs, int rating);
> >
> > /*
> > - * Interval: 0.5sec Threshold: 0.0625s
> > + * Interval: 0.5sec.
> > */
> > #define WATCHDOG_INTERVAL (HZ >> 1)
> > -#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
> > -
> > -/*
> > - * Maximum permissible delay between two readouts of the watchdog
> > - * clocksource surrounding a read of the clocksource being validated.
> > - * This delay could be due to SMIs, NMIs, or to VCPU preemptions.
> > - */
> > -#define WATCHDOG_MAX_SKEW (100 * NSEC_PER_USEC)
> >
> > static void clocksource_watchdog_work(struct work_struct *work)
> > {
> > @@ -347,6 +353,7 @@ static void clocksource_watchdog(struct timer_list *unused)
> > int next_cpu, reset_pending;
> > int64_t wd_nsec, cs_nsec;
> > struct clocksource *cs;
> > + u32 md;
> >
> > spin_lock(&watchdog_lock);
> > if (!watchdog_running)
> > @@ -393,7 +400,8 @@ static void clocksource_watchdog(struct timer_list *unused)
> > continue;
> >
> > /* Check the deviation from the watchdog clocksource. */
> > - if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
> > + md = cs->uncertainty_margin + watchdog->uncertainty_margin;
> > + if (abs(cs_nsec - wd_nsec) > md) {
> > pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
> > smp_processor_id(), cs->name);
> > pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
> > @@ -1046,6 +1054,26 @@ void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq
> > clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
> > NSEC_PER_SEC / scale, sec * scale);
> > }
> > +
> > + /*
> > + * If the uncertainty margin is not specified, calculate it.
> > + * If both scale and freq are non-zero, calculate the clock
> > + * period, but bound below at 2*WATCHDOG_MAX_SKEW. However,
> > + * if either of scale or freq is zero, be very conservative and
> > + * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
> > + * uncertainty margin. Allow stupidly small uncertainty margins
> > + * to be specified by the caller for testing purposes, but warn
> > + * to discourage production use of this capability.
> > + */
> > + if (scale && freq && !cs->uncertainty_margin) {
> > + cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
> > + if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
> > + cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
> > + } else if (!cs->uncertainty_margin) {
> > + cs->uncertainty_margin = WATCHDOG_THRESHOLD;
> > + }
> > + WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
> > +
> > /*
> > * Ensure clocksources that have large 'mult' values don't overflow
> > * when adjusted.
> > diff --git a/kernel/time/jiffies.c b/kernel/time/jiffies.c
> > index a492e4da69ba..b3f608c2b936 100644
> > --- a/kernel/time/jiffies.c
> > +++ b/kernel/time/jiffies.c
> > @@ -49,13 +49,14 @@ static u64 jiffies_read(struct clocksource *cs)
> > * for "tick-less" systems.
> > */
> > static struct clocksource clocksource_jiffies = {
> > - .name = "jiffies",
> > - .rating = 1, /* lowest valid rating*/
> > - .read = jiffies_read,
> > - .mask = CLOCKSOURCE_MASK(32),
> > - .mult = TICK_NSEC << JIFFIES_SHIFT, /* details above */
> > - .shift = JIFFIES_SHIFT,
> > - .max_cycles = 10,
> > + .name = "jiffies",
> > + .rating = 1, /* lowest valid rating*/
> > + .uncertainty_margin = TICK_NSEC,
>
> 'jiffies' is known to be very bad as a watchdog ("worse bandaid" in
> Thomas' words :)), and TICK_NSEC just turns to 1ms for HZ=1000 case.
> Maybe we should give it a bigger margin, like the 32ms margin for
> 'tsc-early'?
>
> Other than this, it looks good to me, thanks!

As in the fixup diff below?

Would you be willing to provide an ack or tested-by for the rest
of the series? (I have your ack on 1/6.)

Thanx, Paul

------------------------------------------------------------------------

diff --git a/kernel/time/jiffies.c b/kernel/time/jiffies.c
index b3f608c2b936..01935aafdb46 100644
--- a/kernel/time/jiffies.c
+++ b/kernel/time/jiffies.c
@@ -51,7 +51,7 @@ static u64 jiffies_read(struct clocksource *cs)
static struct clocksource clocksource_jiffies = {
.name = "jiffies",
.rating = 1, /* lowest valid rating*/
- .uncertainty_margin = TICK_NSEC,
+ .uncertainty_margin = 32 * NSEC_PER_MSEC,
.read = jiffies_read,
.mask = CLOCKSOURCE_MASK(32),
.mult = TICK_NSEC << JIFFIES_SHIFT, /* details above */