Date: Sat, 5 Mar 2005 14:06:23 +0100 (CET) From: Antoine Brodin <antoine.brodin@laposte.net> To: FreeBSD-gnats-submit@FreeBSD.org Subject: kern/78444: sched_ule: doesn't keep track of the sleep time of a process Message-ID: <200503051306.j25D6NFG000989@barton.dreadbsd.org> Resent-Message-ID: <200503051310.j25DADlk032761@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
>Number: 78444 >Category: kern >Synopsis: sched_ule: doesn't keep track of the sleep time of a process >Confidential: no >Severity: non-critical >Priority: low >Responsible: freebsd-bugs >State: open >Quarter: >Keywords: >Date-Required: >Class: sw-bug >Submitter-Id: current-users >Arrival-Date: Sat Mar 05 13:10:13 GMT 2005 >Closed-Date: >Last-Modified: >Originator: Antoine Brodin >Release: FreeBSD 6.0-CURRENT i386 >Organization: none >Environment: System: FreeBSD barton.dreadbsd.org 6.0-CURRENT FreeBSD 6.0-CURRENT #1: Thu Mar 3 15:04:32 CET 2005 antoine@barton.dreadbsd.org:/usr/obj/usr/src/sys/BARTON i386 >Description: When using the sched_ule scheduler, the sleep time of a process isn't computed. kg_slptime is zeroed during the ksegrp creation and after that, it's never modified. In sched_ule.c, something named kg_slptime is manipulated but it's not the sleep time of a process: there's the macro #define kg_slptime kg_sched->skg_slptime It represents an history of the process's sleep and not a sleep time and it would be bogus to use it as the sleep time. Side effects: swapout doesn't work as intended: a process can never be swaped out since its sleep time is 0 and is below swap_idle_threshold1. >How-To-Repeat: Use ps: $ ps -o sl SL 0 0 0 0 0 0 0 0 0 0 0 0 >Fix: Having a lightweight schedcpu routine running every hz ticks like in sched_4bsd ? The problem is that such a routine would use a O(n) algorithm and would penalize quite unnecessarily sched_ule. Another idea: add a sched_slptime routine in sched_ule.c for swapout and fill_kinfo_thread to use ? add something like this to sched_ule.c: %% u_int sched_slptime(struct ksegrp *kg) { struct thread *td; u_int slptime = UINT_MAX; u_int hzticks; PROC_LOCK_ASSERT(kg->kg_proc, MA_OWNED); mtx_assert(&sched_lock, MA_OWNED); FOREACH_THREAD_IN_GROUP(kg, td) { if (td->td_slptime) { hzticks = ticks - td->td_slptime; slptime = min(slptime, hzticks); } else return (0); } return (slptime / hz); } %% add something like this to sched_4bsd.c: %% u_int sched_slptime(struct ksegrp *kg) { return (kg->kg_slptime); } %% add something like this to sched.h: %% u_int sched_slptime(struct ksegrp *kg); %% and use sched_slptime(kg) in vm_glue.c and kern_proc.c instead of kg->kg_slptime >Release-Note: >Audit-Trail: >Unformatted:
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200503051306.j25D6NFG000989>