Pull scheduler updates from Ingo Molnar:
- Move the nohz kick code out of the scheduler tick to a dedicated IPI,
from Frederic Weisbecker.
This necessiated 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
- Remove proliferation of wait_on_bit() action functions and allow
wait_on_bit_action() functions to support a timeout. (Neil Brown)
- Another round of sched/numa improvements, cleanups and fixes. (Rik
van Riel)
- Implement fast idling of CPUs when the system is partially loaded,
for better scalability. (Tim Chen)
- Restructure and fix the CPU hotplug handling code that may leave
cfs_rq and rt_rq's throttled when tasks are migrated away from a dead
cpu. (Kirill Tkhai)
- Robustify the sched topology setup code. (Peterz Zijlstra)
- Improve sched_feat() handling wrt. static_keys (Jason Baron)
- Misc fixes.
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (37 commits)
sched/fair: Fix 'make xmldocs' warning caused by missing description
sched: Use macro for magic number of -1 for setparam
sched: Robustify topology setup
sched: Fix sched_setparam() policy == -1 logic
sched: Allow wait_on_bit_action() functions to support a timeout
sched: Remove proliferation of wait_on_bit() action functions
sched/numa: Revert "Use effective_load() to balance NUMA loads"
sched: Fix static_key race with sched_feat()
sched: Remove extra static_key*() function indirection
sched/rt: Fix replenish_dl_entity() comments to match the current upstream code
sched: Transform resched_task() into resched_curr()
sched/deadline: Kill task_struct->pi_top_task
sched: Rework check_for_tasks()
sched/rt: Enqueue just unthrottled rt_rq back on the stack in __disable_runtime()
sched/fair: Disable runtime_enabled on dying rq
sched/numa: Change scan period code to match intent
sched/numa: Rework best node setting in task_numa_migrate()
sched/numa: Examine a task move when examining a task swap
sched/numa: Simplify task_numa_compare()
sched/numa: Use effective_load() to balance NUMA loads
...
Binding the grace-period kthreads to the timekeeping CPU resulted in
significant performance decreases for some workloads. For more detail,
see:
https://lkml.org/lkml/2014/6/3/395 for benchmark numbers
https://lkml.org/lkml/2014/6/4/218 for CPU statistics
It turns out that it is necessary to bind the grace-period kthreads
to the timekeeping CPU only when all but CPU 0 is a nohz_full CPU
on the one hand or if CONFIG_NO_HZ_FULL_SYSIDLE=y on the other.
In other cases, it suffices to bind the grace-period kthreads to the
set of non-nohz_full CPUs.
This commit therefore creates a tick_nohz_not_full_mask that is the
complement of tick_nohz_full_mask, and then binds the grace-period
kthread to the set of CPUs indicated by this new mask, which covers
the CONFIG_NO_HZ_FULL_SYSIDLE=n case. The CONFIG_NO_HZ_FULL_SYSIDLE=y
case still binds the grace-period kthreads to the timekeeping CPU.
This commit also includes the tick_nohz_full_enabled() check suggested
by Frederic Weisbecker.
Reported-by: Jet Chen <jet.chen@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
[ paulmck: Created housekeeping_affine() and housekeeping_mask per
fweisbec feedback. ]
Remotely kicking a full nohz CPU in order to make it re-evaluate its
next tick is currently implemented using the scheduler IPI.
However this bloats a scheduler fast path with an off-topic feature.
The scheduler tick was abused here for its cool "callable
anywhere/anytime" properties.
But now that the irq work subsystem can queue remote callbacks, it's
a perfect fit to safely queue IPIs when interrupts are disabled
without worrying about concurrent callers.
So lets implement remote kick on top of irq work. This is going to
be used when a new event requires the next tick to be recalculated:
more than 1 task competing on the CPU, timer armed, ...
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Kevin Hilman <khilman@linaro.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
A few functions use remote per CPU access APIs when they
deal with local values.
Just do the right conversion to improve performance, code
readability and debug checks.
While at it, lets extend some of these function names with *_this_cpu()
suffix in order to display their purpose more clearly.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Pull nohz improvements from Frederic Weisbecker:
" It mostly contains fixes and full dynticks off-case optimizations. I believe that
distros want to enable this feature so it seems important to optimize the case
where the "nohz_full=" parameter is empty. ie: I'm trying to remove any performance
regression that comes with NO_HZ_FULL=y when the feature is not used.
This patchset improves the current situation a lot (off-case appears to be around 11% faster
with hackbench, although I guess it may vary depending on the configuration but it should be
significantly faster in any case) now there is still some work to do: I can still observe a
remaining loss of 1.6% throughput seen with hackbench compared to CONFIG_NO_HZ_FULL=n. "
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We are planning to convert the dynticks Kconfig options layout
into a choice menu. The user must be able to easily pick
any of the following implementations: constant periodic tick,
idle dynticks, full dynticks.
As this implies a mutual exclusion, the two dynticks implementions
need to converge on the selection of a common Kconfig option in order
to ease the sharing of a common infrastructure.
It would thus seem pretty natural to reuse CONFIG_NO_HZ to
that end. It already implements all the idle dynticks code
and the full dynticks depends on all that code for now.
So ideally the choice menu would propose CONFIG_NO_HZ_IDLE and
CONFIG_NO_HZ_EXTENDED then both would select CONFIG_NO_HZ.
On the other hand we want to stay backward compatible: if
CONFIG_NO_HZ is set in an older config file, we want to
enable CONFIG_NO_HZ_IDLE by default.
But we can't afford both at the same time or we run into
a circular dependency:
1) CONFIG_NO_HZ_IDLE and CONFIG_NO_HZ_EXTENDED both select
CONFIG_NO_HZ
2) If CONFIG_NO_HZ is set, we default to CONFIG_NO_HZ_IDLE
We might be able to support that from Kconfig/Kbuild but it
may not be wise to introduce such a confusing behaviour.
So to solve this, create a new CONFIG_NO_HZ_COMMON option
which gathers the common code between idle and full dynticks
(that common code for now is simply the idle dynticks code)
and select it from their referring Kconfig.
Then we'll later create CONFIG_NO_HZ_IDLE and map CONFIG_NO_HZ
to it for backward compatibility.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Geoff Levand <geoff@infradead.org>
Cc: Gilad Ben Yossef <gilad@benyossef.com>
Cc: Hakan Akkan <hakanakkan@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Kevin Hilman <khilman@linaro.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
For extreme usecases such as Real Time or HPC, having
the ability to shutdown the tick when a single task runs
on a CPU is a desired feature:
* Reducing the amount of interrupts improves throughput
for CPU-bound tasks. The CPU is less distracted from its
real job, from an execution time and from the cache point
of views.
* This also improve latency response as we have less critical
sections.
Start with introducing a very simple interface to define
full dynticks CPU: use a boot time option defined cpumask
through the "nohz_extended=" kernel parameter. CPUs that
are part of this range will have their tick shutdown
whenever possible: provided they run a single task and
they don't do kernel activity that require the periodic
tick. These details will be later documented in
Documentation/*
An online CPU must be kept outside this range to handle the
timekeeping.
Suggested-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Geoff Levand <geoff@infradead.org>
Cc: Gilad Ben Yossef <gilad@benyossef.com>
Cc: Hakan Akkan <hakanakkan@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Kevin Hilman <khilman@linaro.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Conflicts:
kernel/irq_work.c
Add support for printk in full dynticks CPU.
* Don't stop tick with irq works pending. This
fix is generally useful and concerns archs that
can't raise self IPIs.
* Flush irq works before CPU offlining.
* Introduce "lazy" irq works that can wait for the
next tick to be executed, unless it's stopped.
* Implement klogd wake up using irq work. This
removes the ad-hoc printk_tick()/printk_needs_cpu()
hooks and make it working even in dynticks mode.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
The prediction for future is difficult and when the cpuidle governor prediction
fails and govenor possibly choose the shallower C-state than it should. How to
quickly notice and find the failure becomes important for power saving.
cpuidle menu governor has a method to predict the repeat pattern if there are 8
C-states residency which are continuous and the same or very close, so it will
predict the next C-states residency will keep same residency time.
There is a real case that turbostat utility (tools/power/x86/turbostat)
at kernel 3.3 or early. turbostat utility will read 10 registers one by one at
Sandybridge, so it will generate 10 IPIs to wake up idle CPUs. So cpuidle menu
governor will predict it is repeat mode and there is another IPI wake up idle
CPU soon, so it keeps idle CPU stay at C1 state even though CPU is totally
idle. However, in the turbostat, following 10 registers reading is sleep 5
seconds by default, so the idle CPU will keep at C1 for a long time though it is
idle until break event occurs.
In a idle Sandybridge system, run "./turbostat -v", we will notice that deep
C-state dangles between "70% ~ 99%". After patched the kernel, we will notice
deep C-state stays at >99.98%.
In the patch, a timer is added when menu governor detects a repeat mode and
choose a shallow C-state. The timer is set to a time out value that greater
than predicted time, and we conclude repeat mode prediction failure if timer is
triggered. When repeat mode happens as expected, the timer is not triggered
and CPU waken up from C-states and it will cancel the timer initiatively.
When repeat mode does not happen, the timer will be time out and menu governor
will quickly notice that the repeat mode prediction fails and then re-evaluates
deeper C-states possibility.
Below is another case which will clearly show the patch much benefit:
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <signal.h>
#include <sys/time.h>
#include <time.h>
#include <pthread.h>
volatile int * shutdown;
volatile long * count;
int delay = 20;
int loop = 8;
void usage(void)
{
fprintf(stderr,
"Usage: idle_predict [options]\n"
" --help -h Print this help\n"
" --thread -n Thread number\n"
" --loop -l Loop times in shallow Cstate\n"
" --delay -t Sleep time (uS)in shallow Cstate\n");
}
void *simple_loop() {
int idle_num = 1;
while (!(*shutdown)) {
*count = *count + 1;
if (idle_num % loop)
usleep(delay);
else {
/* sleep 1 second */
usleep(1000000);
idle_num = 0;
}
idle_num++;
}
}
static void sighand(int sig)
{
*shutdown = 1;
}
int main(int argc, char *argv[])
{
sigset_t sigset;
int signum = SIGALRM;
int i, c, er = 0, thread_num = 8;
pthread_t pt[1024];
static char optstr[] = "n:l:t:h:";
while ((c = getopt(argc, argv, optstr)) != EOF)
switch (c) {
case 'n':
thread_num = atoi(optarg);
break;
case 'l':
loop = atoi(optarg);
break;
case 't':
delay = atoi(optarg);
break;
case 'h':
default:
usage();
exit(1);
}
printf("thread=%d,loop=%d,delay=%d\n",thread_num,loop,delay);
count = malloc(sizeof(long));
shutdown = malloc(sizeof(int));
*count = 0;
*shutdown = 0;
sigemptyset(&sigset);
sigaddset(&sigset, signum);
sigprocmask (SIG_BLOCK, &sigset, NULL);
signal(SIGINT, sighand);
signal(SIGTERM, sighand);
for(i = 0; i < thread_num ; i++)
pthread_create(&pt[i], NULL, simple_loop, NULL);
for (i = 0; i < thread_num; i++)
pthread_join(pt[i], NULL);
exit(0);
}
Get powertop V2 from git://github.com/fenrus75/powertop, build powertop.
After build the above test application, then run it.
Test plaform can be Intel Sandybridge or other recent platforms.
#./idle_predict -l 10 &
#./powertop
We will find that deep C-state will dangle between 40%~100% and much time spent
on C1 state. It is because menu governor wrongly predict that repeat mode
is kept, so it will choose the C1 shallow C-state even though it has chance to
sleep 1 second in deep C-state.
While after patched the kernel, we find that deep C-state will keep >99.6%.
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Youquan Song <youquan.song@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Those two APIs were provided to optimize the calls of
tick_nohz_idle_enter() and rcu_idle_enter() into a single
irq disabled section. This way no interrupt happening in-between would
needlessly process any RCU job.
Now we are talking about an optimization for which benefits
have yet to be measured. Let's start simple and completely decouple
idle rcu and dyntick idle logics to simplify.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
It is assumed that rcu won't be used once we switch to tickless
mode and until we restart the tick. However this is not always
true, as in x86-64 where we dereference the idle notifiers after
the tick is stopped.
To prepare for fixing this, add two new APIs:
tick_nohz_idle_enter_norcu() and tick_nohz_idle_exit_norcu().
If no use of RCU is made in the idle loop between
tick_nohz_enter_idle() and tick_nohz_exit_idle() calls, the arch
must instead call the new *_norcu() version such that the arch doesn't
need to call rcu_idle_enter() and rcu_idle_exit().
Otherwise the arch must call tick_nohz_enter_idle() and
tick_nohz_exit_idle() and also call explicitly:
- rcu_idle_enter() after its last use of RCU before the CPU is put
to sleep.
- rcu_idle_exit() before the first use of RCU after the CPU is woken
up.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: David Miller <davem@davemloft.net>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Hans-Christian Egtvedt <hans-christian.egtvedt@atmel.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The tick_nohz_stop_sched_tick() function, which tries to delay
the next timer tick as long as possible, can be called from two
places:
- From the idle loop to start the dytick idle mode
- From interrupt exit if we have interrupted the dyntick
idle mode, so that we reprogram the next tick event in
case the irq changed some internal state that requires this
action.
There are only few minor differences between both that
are handled by that function, driven by the ts->inidle
cpu variable and the inidle parameter. The whole guarantees
that we only update the dyntick mode on irq exit if we actually
interrupted the dyntick idle mode, and that we enter in RCU extended
quiescent state from idle loop entry only.
Split this function into:
- tick_nohz_idle_enter(), which sets ts->inidle to 1, enters
dynticks idle mode unconditionally if it can, and enters into RCU
extended quiescent state.
- tick_nohz_irq_exit() which only updates the dynticks idle mode
when ts->inidle is set (ie: if tick_nohz_idle_enter() has been called).
To maintain symmetry, tick_nohz_restart_sched_tick() has been renamed
into tick_nohz_idle_exit().
This simplifies the code and micro-optimize the irq exit path (no need
for local_irq_save there). This also prepares for the split between
dynticks and rcu extended quiescent state logics. We'll need this split to
further fix illegal uses of RCU in extended quiescent states in the idle
loop.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: David Miller <davem@davemloft.net>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Hans-Christian Egtvedt <hans-christian.egtvedt@atmel.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
