Messages in this thread | | | From | John Stultz <> | Subject | [PATCH 0/9] Reduce timekeeping lock hold time v2 | Date | Thu, 1 Mar 2012 23:12:39 -0800 |
| |
Once again, here is the second half of the timekeeping cleanups (the first half are already in -tip) as well as some changes to reduce the timekeeping lock hold times.
This work grew out of some of Eric Dumazet's and Thomas Gleixner's suggestions, after they noticed the xtime_lock hold time could be on the long side.
The basic idea is that we keep a shadow copy of the timekeeper stucture, which can be updated while readers are accessing the time. Then we only have to block readers as we switch to the newly updated structure.
Changelog: v2: This revsion includes some formatting changes suggested by Ingo, as well as a patch converting shift values from ints to u32 to avoid any unintended extra effort caused by the compiler. There is also a few minor bug fixes that I caught with some additional testing.
Let me know if you have any further issues with the patchset.
If not, feel free to pull the tree from: git://git.linaro.org/people/jstultz/linux.git fortglx/3.4/time
John Stultz (9): time: Condense timekeeper.xtime into xtime_sec time: Rework timekeeping functions to take timekeeper ptr as argument time: Split timekeeper lock into separate reader/writer locks time: Update timekeeper structure using a local shadow time: Shadow cycle_last in timekeeper structure time: Reduce timekeeper read lock hold time time: Convert the timekeeper's wlock to a raw_spin_lock time: Whitespace cleanups per Ingo's requests time: Explicitly use u32 instead of int for shift values
kernel/time/timekeeping.c | 506 ++++++++++++++++++++++++++------------------- 1 files changed, 290 insertions(+), 216 deletions(-)
CC: Ingo Molnar <mingo@elte.hu> CC: Thomas Gleixner <tglx@linutronix.de> CC: Eric Dumazet <eric.dumazet@gmail.com> CC: Richard Cochran <richardcochran@gmail.com> -- 1.7.3.2.146.gca209
| |