Guenter reported boot stalls on a emulated ARM 32-bit platform, which has a 24-bit wide clocksource. It turns out that the calculated maximal idle time, which limits idle sleeps to prevent clocksource wrap arounds, is close to the point where the negative motion detection triggers. max_idle_ns: 597268854 ns negative motion tripping point: 671088640 ns If the idle wakeup is delayed beyond that point, the clocksource advances far enough to trigger the negative motion detection. This prevents the clock to advance and in the worst case the system stalls completely if the consecutive sleeps based on the stale clock are delayed as well. Cure this by calculating a more robust cut-off value for negative motion, which covers 87.5% of the actual clocksource counter width. Compare the delta against this value to catch negative motion. This is specifically for clock sources with a small counter width as their wrap around time is close to the half counter width. For clock sources with wide counters this is not a problem because the maximum idle time is far from the half counter width due to the math overflow protection constraints. For the case at hand this results in a tripping point of 1174405120ns. Note, that this cannot prevent issues when the delay exceeds the 87.5% margin, but that's not different from the previous unchecked version which allowed arbitrary time jumps. Systems with small counter width are prone to invalid results, but this problem is unlikely to be seen on real hardware. If such a system completely stalls for more than half a second, then there are other more urgent problems than the counter wrapping around.
Unified · Split
| @@ -49,6 +49,7 @@ struct module; · 48 unmodified lines | |||
| 49 | * @archdata: Optional arch-specific data | 49 | * @archdata: Optional arch-specific data |
| 50 | * @max_cycles: Maximum safe cycle value which won't overflow on | 50 | * @max_cycles: Maximum safe cycle value which won't overflow on |
| 51 | * multiplication | 51 | * multiplication |
| 52 | * @max_raw_delta: Maximum safe delta value for negative motion detection | ||
| 52 | * @name: Pointer to clocksource name | 53 | * @name: Pointer to clocksource name |
| 53 | * @list: List head for registration (internal) | 54 | * @list: List head for registration (internal) |
| 54 | * @freq_khz: Clocksource frequency in khz. | 55 | * @freq_khz: Clocksource frequency in khz. |
| @@ -109,6 +110,7 @@ struct clocksource { · 54 unmodified lines | |||
| 109 | struct arch_clocksource_data archdata; | 110 | struct arch_clocksource_data archdata; |
| 110 | #endif | 111 | #endif |
| 111 | u64 max_cycles; | 112 | u64 max_cycles; |
| 113 | u64 max_raw_delta; | ||
| 112 | const char *name; | 114 | const char *name; |
| 113 | struct list_head list; | 115 | struct list_head list; |
| 114 | u32 freq_khz; | 116 | u32 freq_khz; |
| @@ -24,7 +24,7 @@ static void clocksource_enqueue(struct clocksource *cs); · 23 unmodified lines | |||
| 24 | 24 | ||
| 25 | static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end) | 25 | static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end) |
| 26 | { | 26 | { |
| 27 | u64 delta = clocksource_delta(end, start, cs->mask); | 27 | u64 delta = clocksource_delta(end, start, cs->mask, cs->max_raw_delta); |
| 28 | 28 | ||
| 29 | if (likely(delta < cs->max_cycles)) | 29 | if (likely(delta < cs->max_cycles)) |
| 30 | return clocksource_cyc2ns(delta, cs->mult, cs->shift); | 30 | return clocksource_cyc2ns(delta, cs->mult, cs->shift); |
| @@ -993,6 +993,15 @@ static inline void clocksource_update_max_deferment(struct clocksource *cs) · 962 unmodified lines | |||
| 993 | cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift, | 993 | cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift, |
| 994 | cs->maxadj, cs->mask, | 994 | cs->maxadj, cs->mask, |
| 995 | &cs->max_cycles); | 995 | &cs->max_cycles); |
| 996 | |||
| 997 | /* | ||
| 998 | * Threshold for detecting negative motion in clocksource_delta(). | ||
| 999 | * | ||
| 1000 | * Allow for 0.875 of the counter width so that overly long idle | ||
| 1001 | * sleeps, which go slightly over mask/2, do not trigger the | ||
| 1002 | * negative motion detection. | ||
| 1003 | */ | ||
| 1004 | cs->max_raw_delta = (cs->mask >> 1) + (cs->mask >> 2) + (cs->mask >> 3); | ||
| 996 | } | 1005 | } |
| 997 | 1006 | ||
| 998 | static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur) | 1007 | static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur) |
| @@ -755,7 +755,8 @@ static void timekeeping_forward_now(struct timekeeper *tk) · 754 unmodified lines | |||
| 755 | u64 cycle_now, delta; | 755 | u64 cycle_now, delta; |
| 756 | 756 | ||
| 757 | cycle_now = tk_clock_read(&tk->tkr_mono); | 757 | cycle_now = tk_clock_read(&tk->tkr_mono); |
| 758 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | 758 | delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask, |
| 759 | tk->tkr_mono.clock->max_raw_delta); | ||
| 759 | tk->tkr_mono.cycle_last = cycle_now; | 760 | tk->tkr_mono.cycle_last = cycle_now; |
| 760 | tk->tkr_raw.cycle_last = cycle_now; | 761 | tk->tkr_raw.cycle_last = cycle_now; |
| 761 | 762 | ||
| @@ -2230,7 +2231,8 @@ static bool timekeeping_advance(enum timekeeping_adv_mode mode) · 1468 unmodified lines | |||
| 2230 | return false; | 2231 | return false; |
| 2231 | 2232 | ||
| 2232 | offset = clocksource_delta(tk_clock_read(&tk->tkr_mono), | 2233 | offset = clocksource_delta(tk_clock_read(&tk->tkr_mono), |
| 2233 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask); | 2234 | tk->tkr_mono.cycle_last, tk->tkr_mono.mask, |
| 2235 | tk->tkr_mono.clock->max_raw_delta); | ||
| 2234 | 2236 | ||
| 2235 | /* Check if there's really nothing to do */ | 2237 | /* Check if there's really nothing to do */ |
| 2236 | if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK) | 2238 | if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK) |
| @@ -30,15 +30,15 @@ static inline void timekeeping_inc_mg_floor_swaps(void) · 29 unmodified lines | |||
| 30 | 30 | ||
| 31 | #endif | 31 | #endif |
| 32 | 32 | ||
| 33 | static inline u64 clocksource_delta(u64 now, u64 last, u64 mask) | 33 | static inline u64 clocksource_delta(u64 now, u64 last, u64 mask, u64 max_delta) |
| 34 | { | 34 | { |
| 35 | u64 ret = (now - last) & mask; | 35 | u64 ret = (now - last) & mask; |
| 36 | 36 | ||
| 37 | /* | 37 | /* |
| 38 | * Prevent time going backwards by checking the MSB of mask in | 38 | * Prevent time going backwards by checking the result against |
| 39 | * the result. If set, return 0. | 39 | * @max_delta. If greater, return 0. |
| 40 | */ | 40 | */ |
| 41 | return ret & ~(mask >> 1) ? 0 : ret; | 41 | return ret > max_delta ? 0 : ret; |
| 42 | } | 42 | } |
| 43 | 43 | ||
| 44 | /* Semi public for serialization of non timekeeper VDSO updates. */ | 44 | /* Semi public for serialization of non timekeeper VDSO updates. */ |