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.
Unified · Split
| @@ -1128,6 +1128,7 @@ static int tsc_cs_enable(struct clocksource *cs) · 1127 unmodified lines | ||
| 1128 | 1128 | static struct clocksource clocksource_tsc_early = { |
| 1129 | 1129 | .name = "tsc-early", |
| 1130 | 1130 | .rating = 299, |
| 1131 | + .uncertainty_margin = 32 * NSEC_PER_MSEC, | |
| 1131 | 1132 | .read = read_tsc, |
| 1132 | 1133 | .mask = CLOCKSOURCE_MASK(64), |
| 1133 | 1134 | .flags = CLOCK_SOURCE_IS_CONTINUOUS | |
| @@ -43,6 +43,8 @@ struct module; · 42 unmodified lines | ||
| 43 | 43 | * @shift: Cycle to nanosecond divisor (power of two) |
| 44 | 44 | * @max_idle_ns: Maximum idle time permitted by the clocksource (nsecs) |
| 45 | 45 | * @maxadj: Maximum adjustment value to mult (~11%) |
| 46 | + * @uncertainty_margin: Maximum uncertainty in nanoseconds per half second. | |
| 47 | + * Zero says to use default WATCHDOG_THRESHOLD. | |
| 46 | 48 | * @archdata: Optional arch-specific data |
| 47 | 49 | * @max_cycles: Maximum safe cycle value which won't overflow on |
| 48 | 50 | * multiplication |
| @@ -98,6 +100,7 @@ struct clocksource { · 49 unmodified lines | ||
| 98 | 100 | u32 shift; |
| 99 | 101 | u64 max_idle_ns; |
| 100 | 102 | u32 maxadj; |
| 103 | + u32 uncertainty_margin; | |
| 101 | 104 | #ifdef CONFIG_ARCH_CLOCKSOURCE_DATA |
| 102 | 105 | struct arch_clocksource_data archdata; |
| 103 | 106 | #endif |
| @@ -95,6 +95,20 @@ static char override_name[CS_NAME_LEN]; · 94 unmodified lines | ||
| 95 | 95 | static int finished_booting; |
| 96 | 96 | static u64 suspend_start; |
| 97 | 97 | |
| 98 | +/* | |
| 99 | + * Threshold: 0.0312s, when doubled: 0.0625s. | |
| 100 | + * Also a default for cs->uncertainty_margin when registering clocks. | |
| 101 | + */ | |
| 102 | +#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5) | |
| 103 | + | |
| 104 | +/* | |
| 105 | + * Maximum permissible delay between two readouts of the watchdog | |
| 106 | + * clocksource surrounding a read of the clocksource being validated. | |
| 107 | + * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as | |
| 108 | + * a lower bound for cs->uncertainty_margin values when registering clocks. | |
| 109 | + */ | |
| 110 | +#define WATCHDOG_MAX_SKEW (50 * NSEC_PER_USEC) | |
| 111 | + | |
| 98 | 112 | #ifdef CONFIG_CLOCKSOURCE_WATCHDOG |
| 99 | 113 | static void clocksource_watchdog_work(struct work_struct *work); |
| 100 | 114 | static void clocksource_select(void); |
| @@ -121,17 +135,9 @@ static int clocksource_watchdog_kthread(void *data); · 20 unmodified lines | ||
| 121 | 135 | static void __clocksource_change_rating(struct clocksource *cs, int rating); |
| 122 | 136 | |
| 123 | 137 | /* |
| 124 | − * Interval: 0.5sec Threshold: 0.0625s | |
| 138 | + * Interval: 0.5sec. | |
| 125 | 139 | */ |
| 126 | 140 | #define WATCHDOG_INTERVAL (HZ >> 1) |
| 127 | −#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4) | |
| 128 | − | |
| 129 | −/* | |
| 130 | − * Maximum permissible delay between two readouts of the watchdog | |
| 131 | − * clocksource surrounding a read of the clocksource being validated. | |
| 132 | − * This delay could be due to SMIs, NMIs, or to VCPU preemptions. | |
| 133 | − */ | |
| 134 | −#define WATCHDOG_MAX_SKEW (100 * NSEC_PER_USEC) | |
| 135 | 141 | |
| 136 | 142 | static void clocksource_watchdog_work(struct work_struct *work) |
| 137 | 143 | { |
| @@ -348,6 +354,7 @@ static void clocksource_watchdog(struct timer_list *unused) · 210 unmodified lines | ||
| 348 | 354 | int next_cpu, reset_pending; |
| 349 | 355 | int64_t wd_nsec, cs_nsec; |
| 350 | 356 | struct clocksource *cs; |
| 357 | + u32 md; | |
| 351 | 358 | |
| 352 | 359 | spin_lock(&watchdog_lock); |
| 353 | 360 | if (!watchdog_running) |
| @@ -394,7 +401,8 @@ static void clocksource_watchdog(struct timer_list *unused) · 40 unmodified lines | ||
| 394 | 401 | continue; |
| 395 | 402 | |
| 396 | 403 | /* Check the deviation from the watchdog clocksource. */ |
| 397 | − if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) { | |
| 404 | + md = cs->uncertainty_margin + watchdog->uncertainty_margin; | |
| 405 | + if (abs(cs_nsec - wd_nsec) > md) { | |
| 398 | 406 | pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n", |
| 399 | 407 | smp_processor_id(), cs->name); |
| 400 | 408 | pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n", |
| @@ -1047,6 +1055,26 @@ void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq · 646 unmodified lines | ||
| 1047 | 1055 | clocks_calc_mult_shift(&cs->mult, &cs->shift, freq, |
| 1048 | 1056 | NSEC_PER_SEC / scale, sec * scale); |
| 1049 | 1057 | } |
| 1058 | + | |
| 1059 | + /* | |
| 1060 | + * If the uncertainty margin is not specified, calculate it. | |
| 1061 | + * If both scale and freq are non-zero, calculate the clock | |
| 1062 | + * period, but bound below at 2*WATCHDOG_MAX_SKEW. However, | |
| 1063 | + * if either of scale or freq is zero, be very conservative and | |
| 1064 | + * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the | |
| 1065 | + * uncertainty margin. Allow stupidly small uncertainty margins | |
| 1066 | + * to be specified by the caller for testing purposes, but warn | |
| 1067 | + * to discourage production use of this capability. | |
| 1068 | + */ | |
| 1069 | + if (scale && freq && !cs->uncertainty_margin) { | |
| 1070 | + cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq); | |
| 1071 | + if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW) | |
| 1072 | + cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW; | |
| 1073 | + } else if (!cs->uncertainty_margin) { | |
| 1074 | + cs->uncertainty_margin = WATCHDOG_THRESHOLD; | |
| 1075 | + } | |
| 1076 | + WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW); | |
| 1077 | + | |
| 1050 | 1078 | /* |
| 1051 | 1079 | * Ensure clocksources that have large 'mult' values don't overflow |
| 1052 | 1080 | * when adjusted. |
| @@ -49,13 +49,14 @@ static u64 jiffies_read(struct clocksource *cs) · 48 unmodified lines | ||
| 49 | 49 | * for "tick-less" systems. |
| 50 | 50 | */ |
| 51 | 51 | static struct clocksource clocksource_jiffies = { |
| 52 | − .name = "jiffies", | |
| 53 | − .rating = 1, /* lowest valid rating*/ | |
| 54 | − .read = jiffies_read, | |
| 55 | − .mask = CLOCKSOURCE_MASK(32), | |
| 56 | − .mult = TICK_NSEC << JIFFIES_SHIFT, /* details above */ | |
| 57 | − .shift = JIFFIES_SHIFT, | |
| 58 | − .max_cycles = 10, | |
| 52 | + .name = "jiffies", | |
| 53 | + .rating = 1, /* lowest valid rating*/ | |
| 54 | + .uncertainty_margin = 32 * NSEC_PER_MSEC, | |
| 55 | + .read = jiffies_read, | |
| 56 | + .mask = CLOCKSOURCE_MASK(32), | |
| 57 | + .mult = TICK_NSEC << JIFFIES_SHIFT, /* details above */ | |
| 58 | + .shift = JIFFIES_SHIFT, | |
| 59 | + .max_cycles = 10, | |
| 59 | 60 | }; |
| 60 | 61 | |
| 61 | 62 | __cacheline_aligned_in_smp DEFINE_RAW_SPINLOCK(jiffies_lock); |