Re: git: cb1f5d11366d - main - Reduce minimum idle hardclock rate from 2Hz to 1Hz.
Date: Mon, 10 Jan 2022 02:43:30 UTC
On 09.01.2022 21:33, Konstantin Belousov wrote: > On Mon, Jan 10, 2022 at 12:26:01AM +0000, Alexander Motin wrote: >> The branch main has been updated by mav: >> >> URL: https://cgit.FreeBSD.org/src/commit/?id=cb1f5d11366dc9b803f2755d83fe02599830882a >> >> commit cb1f5d11366dc9b803f2755d83fe02599830882a >> Author: Alexander Motin <mav@FreeBSD.org> >> AuthorDate: 2022-01-10 00:17:43 +0000 >> Commit: Alexander Motin <mav@FreeBSD.org> >> CommitDate: 2022-01-10 00:25:56 +0000 >> >> Reduce minimum idle hardclock rate from 2Hz to 1Hz. >> >> On idle 80-thread system it allows to improve package-level idle state >> residency and so power consumption by several percent. > Are you sure that this is safe? > > Our timecounters are tailored to wrap at 1sec, at least the most precise/ > important TSC-low is. If idle system interrupts with 1Hz frequency, > it probably comes dangerously close to situations where tc_windup() could > be missed. That particular timecounters wrap around should be handled there long ago with tc_min_ticktock_freq. When needed it should increase hardclock on BSP (could be any other CPU, so why not BSP) up to the minimally safe rate. -- Alexander Motin