nohz: Move nohz kick out of scheduler IPI, v9
Ingo,
Please pull the timers/nohz-irq-work-v7 branch that can be found at:
git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
timers/nohz-irq-work-v7
It has been acked by Peterz.
--- Summary ---
This set moves the nohz kick, used to notify a full dynticks CPU when events require tick rescheduling, out of the scheduler tick to a dedicated IPI.
This debloats a bit the scheduler IPI from off-topic work that was abusing that scheduler fast path for its convenient asynchronous properties. Now the nohz kick uses irq work for its own needs.
Of course this implied quite some background infrastructure rework, including:
* Clean up some irq work internals
* Implement remote irq work
* Implement nohz kick on top of remote irq work
* Move full dynticks timer enqueue notification to new kick
* Move multi-task notification to new kick
* Remove unecessary barriers on multi-task notification
Thanks,
Frederic
---
Frederic Weisbecker (6):
irq_work: Split raised and lazy lists
irq_work: Implement remote queueing
nohz: Support nohz full remote kick
nohz: Switch to nohz full remote kick on timer enqueue
nohz: Use nohz own full kick on 2nd task enqueue
nohz: Use IPI implicit full barrier against rq->nr_running r/w
include/linux/irq_work.h | 5 ++++ include/linux/tick.h | 9 +++++- kernel/irq_work.c | 76 ++++++++++++++++++++++++++++++------------------ kernel/sched/core.c | 22 ++++++++------ kernel/sched/sched.h | 12 ++++++-- kernel/smp.c | 9 ++++++ kernel/time/tick-sched.c | 10 ++++--- 7 files changed, 97 insertions(+), 46 deletions(-)
Quoted 42 lines · Frederic Weisbecker
Ingo,
Please pull the timers/nohz-irq-work-v7 branch that can be found at:
git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
timers/nohz-irq-work-v7It has been acked by Peterz.
--- Summary ---
This set moves the nohz kick, used to notify a full dynticks CPU when events require tick rescheduling, out of the scheduler tick to a dedicated IPI.
This debloats a bit the scheduler IPI from off-topic work that was abusing that scheduler fast path for its convenient asynchronous properties. Now the nohz kick uses irq work for its own needs.
Of course this implied quite some background infrastructure rework, including:
* Clean up some irq work internals
* Implement remote irq work
* Implement nohz kick on top of remote irq work
* Move full dynticks timer enqueue notification to new kick
* Move multi-task notification to new kick
* Remove unecessary barriers on multi-task notificationThanks,
Frederic
---Frederic Weisbecker (6):
irq_work: Split raised and lazy lists
irq_work: Implement remote queueing
nohz: Support nohz full remote kick
nohz: Switch to nohz full remote kick on timer enqueue
nohz: Use nohz own full kick on 2nd task enqueue
nohz: Use IPI implicit full barrier againstrq->nr_runningr/w
Quoted 8 lines
include/linux/irq_work.h | 5 ++++ include/linux/tick.h | 9 +++++- kernel/irq_work.c | 76 ++++++++++++++++++++++++++++++------------------ kernel/sched/core.c | 22 ++++++++------ kernel/sched/sched.h | 12 ++++++-- kernel/smp.c | 9 ++++++ kernel/time/tick-sched.c | 10 ++++--- 7 files changed, 97 insertions(+), 46 deletions(-)
Pulled, thanks a lot Frederic!
Ingo
Quoted 6 lines · Ingo Molnar
Quoted 2 lines · Frederic Weisbecker
Please pull the timers/nohz-irq-work-v7 branch that can be found at:
Pulled, thanks a lot Frederic!
Ingo, the previous version had one of your machines fall over, is that machine good now?
Quoted 10 lines · Peter Zijlstra
Quoted 6 lines · Ingo Molnar
Quoted 2 lines · Frederic Weisbecker
Please pull the timers/nohz-irq-work-v7 branch that can be found at:
Pulled, thanks a lot Frederic!
Ingo, the previous version had one of your machines fall over, is that machine good now?
It appears so. Will let you know if it still goes bad.
Thanks,
Ingo