On Sat, Apr 5, 2025 at 2:40 PM Thomas Gleixner tglx@linutronix.de wrote:
Lei Chen raised an issue with CLOCK_MONOTONIC_COARSE seeing time inconsistencies. Lei tracked down that this was being caused by the adjustment
tk->tkr_mono.xtime_nsec -= offset;which is made to compensate for the unaccumulated cycles in offset when the multiplicator is adjusted forward, so that the non-_COARSE clockids don't see inconsistencies.
However, the _COARSE clockid getter functions use the adjusted xtime_nsec value directly and do not compensate the negative offset via the clocksource delta multiplied with the new multiplicator. In that case the caller can observe time going backwards in consecutive calls.
Hey Thomas! Thanks so much for reviving this version of the fix and my apologies for apparently taking us down the wrong path with the earlier solution.
Looking over the patch, it seems ok to me, but in a test run with it, I've seen an error with CLOCK_REALTIME_COARSE during the clocksource-switch test (as well as some seemingly unrelated test errors, which I need to investigate) so I'm looking at the zeroing done in timekeeping_forward_now and will try to look more into this tomorrow.
thanks -john