Messages in this thread | | | Date | Mon, 14 Nov 2022 15:28:07 -0800 | From | "Paul E. McKenney" <> | Subject | [PATCH clocksource 0/3] Reject bogus watchdog clocksource measurements |
| |
Hello!
This series adds comments classifying bogus measurements and adds capped exponential backoff for messages that are likely caused by overly busy systems.
1. Reject bogus watchdog clocksource measurements.
2. Add comments to classify bogus measurements.
3. Exponential backoff for load-induced bogus watchdog reads.
Changes since v1:
o Merge the "Reject bogus watchdog clocksource measurements" [1] series with the "Clocksource-watchdog classification and backoff" [2] series.
o Updated messages to indicate timer bug for too-short watchdog intervals based on Feng Tang feedback.
o Added explanatory comments before error messages based on Feng Tang feedback.
o Check both the reference clocksource and the clocksource under test to handle clocksource-internal counter wrap, based on Feng Tang feedback.
o Reversed the order of the checks, so that the too-long check preceeds the too-short check. This avoids too-short false positives due to counter wrap.
o Simplified exponential-backoff time checks based on feedback from Waiman Long.
o Updated the exponentially backed off messages to more clearly indicate that the counts are instances since the last printed message.
o Applied Reviewed-by tags.
[1] https://lore.kernel.org/all/20221019230904.GA2502730@paulmck-ThinkPad-P17-Gen-1 [2] https://lore.kernel.org/all/20221102184001.GA1306489@paulmck-ThinkPad-P17-Gen-1/
Thanx, Paul
------------------------------------------------------------------------
b/include/linux/clocksource.h | 3 +++ b/kernel/time/clocksource.c | 13 ++++++++++++- kernel/time/clocksource.c | 33 +++++++++++++++++++++++++++------ 3 files changed, 42 insertions(+), 7 deletions(-)
| |