Date: Tue, 8 Jun 2021 11:00:52 +0200 (CEST) From: freebsd@oldach.net (Helge Oldach) To: freebsd-stable@freebsd.org Subject: Re: Where might memory be reported? Message-ID: <202106081042.158AgIEZ099674@nuc.oldach.net> In-Reply-To: <bba08248-f898-cead-8df5-a52b1e14257a@heuristicsystems.com.au> from Dewayne Geraghty at "8 Jun 2021 15:58:37"
next in thread | previous in thread | raw e-mail | index | archive | help
Dewayne Geraghty wrote on Tue, 08 Jun 2021 07:58:37 +0200 (CEST): > More relevant is > https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=234559 > where Mark Johnson advised (in Feb '19) > "I've thought about this some more and decided to close this bug WONTFIX > for now. Basically, we started lazily dequeuing wired pages to avoid > the page queue synchronization cost incurred by frequent wirings. The > result is that wired pages may temporarily be counted as > active/inactive/laundry despite not being reclaimable. Maintaining the > previous behaviour wrt statistics would introduce synchronization > overhead that we've been trying to avoid in past and future work. I > don't plan to change the current behaviour and there doesn't seem to be > a compelling reason to do so, at least based on this report. If you or > anyone else would like to discuss this further or describe a use-case > that depends on the old behaviour, please feel free to re-open the bug." Not sure if it applies here. The situation in that ticket (which was, coincidentally, raised by me) was that the sum of all specific page counts was *higher* than the total page count. The explanation is, briefly, double counting some pages. The situation here is different: > Consider this output from a 12.2-STABLE box (r368820) on amd64: > > # sysctl vm.stats.vm | grep count > vm.stats.vm.v_cache_count: 0 > vm.stats.vm.v_user_wire_count: 0 > vm.stats.vm.v_laundry_count: 0 > vm.stats.vm.v_inactive_count: 121191 > vm.stats.vm.v_active_count: 20836 > vm.stats.vm.v_wire_count: 754310 > vm.stats.vm.v_free_count: 254711 > vm.stats.vm.v_page_count: 3993253 > > It should be pretty clear that there are missing pages. Where might they be and > how might I find out? The sum of all specific page counts here is much *lower* than the total v_page_count. We are *missing* pages. Did I misunderstand the case perhaps? Kind regards Helge
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202106081042.158AgIEZ099674>