[PATCH RFC] clocksource: add detection for screw of watchdog

[Feng Tang]

Signed-off-by: Feng Tang <feng.tang@xxxxxxxxx>
---
 kernel/time/clocksource.c | 39 +++++++++++++++++++++++++++++++++------
 1 file changed, 33 insertions(+), 6 deletions(-)

diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c
index b8a14d2fb5ba6..6fbef74b0ed74 100644
--- a/kernel/time/clocksource.c
+++ b/kernel/time/clocksource.c
@@ -205,34 +205,54 @@ EXPORT_SYMBOL_GPL(max_cswd_read_retries);
 static int verify_n_cpus = 8;
 module_param(verify_n_cpus, int, 0644);
 
-static bool cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
+enum {
+	WATCHDOG_READ_SCREWED,
+	CLOCKSROUCE_READ_SCREWED,
+	BOTH_READ_FINE,
+};
+
+static int cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
 {
 	unsigned int nretries;
-	u64 wd_end, wd_delta;
-	int64_t wd_delay;
+	u64 wd_end, wd_delta, wd_third;
+	int64_t wd_delay, wd_seq_delay;
+
 
 	for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) {
 		local_irq_disable();
 		*wdnow = watchdog->read(watchdog);
 		*csnow = cs->read(cs);
 		wd_end = watchdog->read(watchdog);
+		wd_third = watchdog->read(watchdog);
 		local_irq_enable();
 
 		wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask);
 		wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
 					      watchdog->shift);
+		/*
+		 * If a big delay is seen between 2 consecutive read of watchdog,
+		 * then there must be something wrong with watchdog itself, or
+		 * the system is under some extreme pressures, no need to judge
+		 * clocksource for this run.
+		 */
+		wd_delta = clocksource_delta(wd_third, wd_end, watchdog->mask);
+		wd_seq_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
+						watchdog->shift);
+		if (wd_seq_delay > WATCHDOG_MAX_SKEW / 2)
+			return WATCHDOG_READ_SCREWED;
+
 		if (wd_delay <= WATCHDOG_MAX_SKEW) {
 			if (nretries > 1 || nretries >= max_cswd_read_retries) {
 				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
 					smp_processor_id(), watchdog->name, nretries);
 			}
-			return true;
+			return BOTH_READ_FINE;
 		}
 	}
 
 	pr_warn("timekeeping watchdog on CPU%d: %s read-back delay of %lldns, attempt %d, marking unstable\n",
 		smp_processor_id(), watchdog->name, wd_delay, nretries);
-	return false;
+	return CLOCKSROUCE_READ_SCREWED;
 }
 
 static u64 csnow_mid;
@@ -373,10 +393,17 @@ static void clocksource_watchdog(struct timer_list *unused)
 			continue;
 		}
 
-		if (!cs_watchdog_read(cs, &csnow, &wdnow)) {
+		switch (cs_watchdog_read(cs, &csnow, &wdnow)) {
+		case WATCHDOG_READ_SCREWED:
+			/* skip checking this 'cs' */
+			continue;
+		case CLOCKSROUCE_READ_SCREWED:
 			/* Clock readout unreliable, so give it up. */
 			__clocksource_unstable(cs);
 			continue;
+		case BOTH_READ_FINE:
+		default:
+			;
 		}
 
 		/* Clocksource initialized ? */
-- 
2.27.0

Thanks,
Feng

> The current watchdog clocksource skew threshold of 50us is found to be
> insufficient. So it is changed back to 100us before commit 2e27e793e280
> ("clocksource: Reduce clocksource-skew threshold") in patch 1. Patch 2
> adds a Kconfig option to allow kernel builder to control the actual
> threshold to be used.
> 
> Waiman Long (2):
>   clocksource: Avoid accidental unstable marking of clocksources
>   clocksource: Add a Kconfig option for WATCHDOG_MAX_SKEW
> 
>  kernel/time/Kconfig       |  9 ++++++
>  kernel/time/clocksource.c | 58 ++++++++++++++++++++++++++++++---------
>  2 files changed, 54 insertions(+), 13 deletions(-)
> 
> -- 
> 2.27.0