[lkml]   [2013]   [Oct]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Subject[RFC PATCH 0/5] nohz: Fix racy sleeptime stats v2

So this is a new iteration with some more changes on top of the latest discussions
we had.

Unfortunately it's not yet a real viable solution. Check out patch 4/5 for details,
I think that's too much overhead. Also it changes enough the semantics to break the
idle sleeptime accounting along the way. This is because moving the accounting to
io_schedule() makes it account _all_ IO sleeptime, even if the CPU is still busy,
not just when it is idle like before. As a result, the idle sleeptime, that relies on
a diff against the io sleeptime, is broken.

Anyway, my hope with this new take is mostly to launch a discussion with a fresher
base and see if we can find better ideas.




Frederic Weisbecker (5):
nohz: Convert a few places to use local per cpu accesses
nohz: Only update sleeptime stats locally
timer: Change idle/iowait accounting semantics
sched: Refactor iowait accounting
nohz: Synchronize sleep time stats with seqlock

include/linux/ktime.h | 7 +++
include/linux/sched.h | 1 +
include/linux/tick.h | 11 ++--
kernel/sched/core.c | 68 ++++++++++++++++++---
kernel/sched/cputime.c | 2 +-
kernel/sched/sched.h | 5 +-
kernel/softirq.c | 4 +-
kernel/time/tick-broadcast.c | 6 +-
kernel/time/tick-internal.h | 4 +-
kernel/time/tick-sched.c | 138 ++++++++++++++-----------------------------
kernel/time/timer_list.c | 3 +-
11 files changed, 130 insertions(+), 119 deletions(-)

 \ /
  Last update: 2013-10-19 17:41    [W:0.095 / U:2.696 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site