Commit 30765b92 ("slab, lockdep: Annotate the locks before using
them") moves the init_lock_keys() call from after g_cpucache_up =
FULL, to before it. And overlooks the fact that init_node_lock_keys()
tests for it and ignores everything !FULL.
Introduce a LATE stage and change the lockdep test to be <LATE.
Acked-by: Christoph Lameter <[email protected]>
Cc: Pekka Enberg <[email protected]>
Cc: [email protected]
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
When you do:
$ perf record -e cycles,cycles,cycles noploop 10
You expect about 10,000 samples for each event, i.e., 10s at
1000samples/sec. However, this is not what's happening. You
get much fewer samples, maybe 3700 samples/event:
$ perf report -D | tail -15
Aggregated stats:
TOTAL events: 10998
MMAP events: 66
COMM events: 2
SAMPLE events: 10930
cycles stats:
TOTAL events: 3644
SAMPLE events: 3644
cycles stats:
TOTAL events: 3642
SAMPLE events: 3642
cycles stats:
TOTAL events: 3644
SAMPLE events: 3644
On a Intel Nehalem or even AMD64, there are 4 counters capable
of measuring cycles, so there is plenty of space to measure those
events without multiplexing (even with the NMI watchdog active).
And even with multiplexing, we'd expect roughly the same number
of samples per event.
The root of the problem was that when the event that caused the buffer
to become full was not the first event passed on the cmdline, the user
notification would get lost. The notification was sent to the file
descriptor of the overflowed event but the perf tool was not polling
on it. The perf tool aggregates all samples into a single buffer,
i.e., the buffer of the first event. Consequently, it assumes
notifications for any event will come via that descriptor.
The seemingly straight forward solution of moving the waitq into the
ringbuffer object doesn't work because of life-time issues. One could
perf_event_set_output() on a fd that you're also blocking on and cause
the old rb object to be freed while its waitq would still be
referenced by the blocked thread -> FAIL.
Therefore link all events to the ringbuffer and broadcast the wakeup
from the ringbuffer object to all possible events that could be waited
upon. This is rather ugly, and we're open to better solutions but it
works for now.
Reported-by: Stephane Eranian <[email protected]>
Finished-by: Stephane Eranian <[email protected]>
Reviewed-by: Stephane Eranian <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/20111126014731.GA7030@quad
Signed-off-by: Ingo Molnar <[email protected]>
Now that the core offcore support is fixed up (thanks Stephane) and we
have sane generic events utilizing them, re-enable the raw access to
the feature as well.
Note that it doesn't matter if you use event 0x1b7 or 0x1bb to specify
an offcore event, either one works and neither guarantees you'll end
up on a particular offcore MSR.
Based on original patch from: Vince Weaver <[email protected]>.
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Vince Weaver <[email protected]>.
Cc: Stephane Eranian <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Masami spotted that we always try to decode the instruction stream as
64bit instructions when running a 64bit kernel, this doesn't work for
ia32-compat proglets.
Use TIF_IA32 to detect if we need to use the 32bit instruction
decoder.
Reported-by: Masami Hiramatsu <[email protected]>
Cc: [email protected]
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Normally the RT bandwidth scheme will share bandwidth across the
entire root_domain. However sometimes its convenient to disable this
sharing for debug purposes. Provide a simple feature switch to this
end.
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
There's a lock inversion between the cputimer->lock and rq->lock;
notably the two callchains involved are:
update_rlimit_cpu()
sighand->siglock
set_process_cpu_timer()
cpu_timer_sample_group()
thread_group_cputimer()
cputimer->lock
thread_group_cputime()
task_sched_runtime()
->pi_lock
rq->lock
scheduler_tick()
rq->lock
task_tick_fair()
update_curr()
account_group_exec()
cputimer->lock
Where the first one is enabling a CLOCK_PROCESS_CPUTIME_ID timer, and
the second one is keeping up-to-date.
This problem was introduced by e8abccb719 ("posix-cpu-timers: Cure
SMP accounting oddities").
Cure the problem by removing the cputimer->lock and rq->lock nesting,
this leaves concurrent enablers doing duplicate work, but the time
wasted should be on the same order otherwise wasted spinning on the
lock and the greater-than assignment filter should ensure we preserve
monotonicity.
Reported-by: Dave Jones <[email protected]>
Reported-by: Simon Kirby <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: [email protected]
Cc: Linus Torvalds <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Link: http://lkml.kernel.org/r/1318928713.21167.4.camel@twins
Signed-off-by: Thomas Gleixner <[email protected]>
Initial benchmarks show they're a net loss:
$ for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i; done
$ echo 4096 32000 64 128 > /proc/sys/kernel/sem
$ ./sembench -t 2048 -w 1900 -o 0
Pre:
run time 30 seconds 778936 worker burns per second
run time 30 seconds 912190 worker burns per second
run time 30 seconds 817506 worker burns per second
run time 30 seconds 830870 worker burns per second
run time 30 seconds 845056 worker burns per second
Post:
run time 30 seconds 905920 worker burns per second
run time 30 seconds 849046 worker burns per second
run time 30 seconds 886286 worker burns per second
run time 30 seconds 822320 worker burns per second
run time 30 seconds 900283 worker burns per second
So about 4% faster. (!)
cpu_relax() stalls the pipeline, therefore, when used in a tight loop
it has the following benefits:
- allows SMT siblings to have a go;
- reduces pressure on the CPU interconnect.
However, cmpxchg loops are unfair and thus have unbounded completion
time, therefore we should avoid getting in such heavily contended
situations where the above benefits make any difference.
A typical cmpxchg loop should not go round more than a handfull of
times at worst, therefore adding extra delays just slows things down.
Since the llist primitives are new, there aren't any bad users yet,
and we should avoid growing them. Heavily contended sites should
generally be better off using the ticket locks for serialization since
they provide bounded completion times (fifo-fair over the cpus).
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Huang Ying <[email protected]>
Cc: Andrew Morton <[email protected]>
Link: http://lkml.kernel.org/r/1315836358.26517.43.camel@twins
Signed-off-by: Ingo Molnar <[email protected]>
Use the generic llist primitives.
We had a private lockless list implementation in the scheduler in the wake-list
code, now that we have a generic llist implementation that provides all required
operations, switch to it.
This patch is not expected to change any behavior.
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Huang Ying <[email protected]>
Cc: Andrew Morton <[email protected]>
Link: http://lkml.kernel.org/r/1315836353.26517.42.camel@twins
Signed-off-by: Ingo Molnar <[email protected]>
David reported:
Attached below is a watered-down version of rt/tst-cpuclock2.c from
GLIBC. Just build it with "gcc -o test test.c -lpthread -lrt" or
similar.
Run it several times, and you will see cases where the main thread
will measure a process clock difference before and after the nanosleep
which is smaller than the cpu-burner thread's individual thread clock
difference. This doesn't make any sense since the cpu-burner thread
is part of the top-level process's thread group.
I've reproduced this on both x86-64 and sparc64 (using both 32-bit and
64-bit binaries).
For example:
[davem@boricha build-x86_64-linux]$ ./test
process: before(0.001221967) after(0.498624371) diff(497402404)
thread: before(0.000081692) after(0.498316431) diff(498234739)
self: before(0.001223521) after(0.001240219) diff(16698)
[davem@boricha build-x86_64-linux]$
The diff of 'process' should always be >= the diff of 'thread'.
I make sure to wrap the 'thread' clock measurements the most tightly
around the nanosleep() call, and that the 'process' clock measurements
are the outer-most ones.
---
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <fcntl.h>
#include <string.h>
#include <errno.h>
#include <pthread.h>
static pthread_barrier_t barrier;
static void *chew_cpu(void *arg)
{
pthread_barrier_wait(&barrier);
while (1)
__asm__ __volatile__("" : : : "memory");
return NULL;
}
int main(void)
{
clockid_t process_clock, my_thread_clock, th_clock;
struct timespec process_before, process_after;
struct timespec me_before, me_after;
struct timespec th_before, th_after;
struct timespec sleeptime;
unsigned long diff;
pthread_t th;
int err;
err = clock_getcpuclockid(0, &process_clock);
if (err)
return 1;
err = pthread_getcpuclockid(pthread_self(), &my_thread_clock);
if (err)
return 1;
pthread_barrier_init(&barrier, NULL, 2);
err = pthread_create(&th, NULL, chew_cpu, NULL);
if (err)
return 1;
err = pthread_getcpuclockid(th, &th_clock);
if (err)
return 1;
pthread_barrier_wait(&barrier);
err = clock_gettime(process_clock, &process_before);
if (err)
return 1;
err = clock_gettime(my_thread_clock, &me_before);
if (err)
return 1;
err = clock_gettime(th_clock, &th_before);
if (err)
return 1;
sleeptime.tv_sec = 0;
sleeptime.tv_nsec = 500000000;
nanosleep(&sleeptime, NULL);
err = clock_gettime(th_clock, &th_after);
if (err)
return 1;
err = clock_gettime(my_thread_clock, &me_after);
if (err)
return 1;
err = clock_gettime(process_clock, &process_after);
if (err)
return 1;
diff = process_after.tv_nsec - process_before.tv_nsec;
printf("process: before(%lu.%.9lu) after(%lu.%.9lu) diff(%lu)\n",
process_before.tv_sec, process_before.tv_nsec,
process_after.tv_sec, process_after.tv_nsec, diff);
diff = th_after.tv_nsec - th_before.tv_nsec;
printf("thread: before(%lu.%.9lu) after(%lu.%.9lu) diff(%lu)\n",
th_before.tv_sec, th_before.tv_nsec,
th_after.tv_sec, th_after.tv_nsec, diff);
diff = me_after.tv_nsec - me_before.tv_nsec;
printf("self: before(%lu.%.9lu) after(%lu.%.9lu) diff(%lu)\n",
me_before.tv_sec, me_before.tv_nsec,
me_after.tv_sec, me_after.tv_nsec, diff);
return 0;
}
This is due to us using p->se.sum_exec_runtime in
thread_group_cputime() where we iterate the thread group and sum all
data. This does not take time since the last schedule operation (tick
or otherwise) into account. We can cure this by using
task_sched_runtime() at the cost of having to take locks.
This also means we can (and must) do away with
thread_group_sched_runtime() since the modified thread_group_cputime()
is now more accurate and would deadlock when called from
thread_group_sched_runtime().
Aside of that it makes the function safe on 32 bit systems. The old
code added t->se.sum_exec_runtime unprotected. sum_exec_runtime is a
64bit value and could be changed on another cpu at the same time.
Reported-by: David Miller <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/1314874459.7945.22.camel@twins
Tested-by: David Miller <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
David reported:
Attached below is a watered-down version of rt/tst-cpuclock2.c from
GLIBC. Just build it with "gcc -o test test.c -lpthread -lrt" or
similar.
Run it several times, and you will see cases where the main thread
will measure a process clock difference before and after the nanosleep
which is smaller than the cpu-burner thread's individual thread clock
difference. This doesn't make any sense since the cpu-burner thread
is part of the top-level process's thread group.
I've reproduced this on both x86-64 and sparc64 (using both 32-bit and
64-bit binaries).
For example:
[davem@boricha build-x86_64-linux]$ ./test
process: before(0.001221967) after(0.498624371) diff(497402404)
thread: before(0.000081692) after(0.498316431) diff(498234739)
self: before(0.001223521) after(0.001240219) diff(16698)
[davem@boricha build-x86_64-linux]$
The diff of 'process' should always be >= the diff of 'thread'.
I make sure to wrap the 'thread' clock measurements the most tightly
around the nanosleep() call, and that the 'process' clock measurements
are the outer-most ones.
---
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <fcntl.h>
#include <string.h>
#include <errno.h>
#include <pthread.h>
static pthread_barrier_t barrier;
static void *chew_cpu(void *arg)
{
pthread_barrier_wait(&barrier);
while (1)
__asm__ __volatile__("" : : : "memory");
return NULL;
}
int main(void)
{
clockid_t process_clock, my_thread_clock, th_clock;
struct timespec process_before, process_after;
struct timespec me_before, me_after;
struct timespec th_before, th_after;
struct timespec sleeptime;
unsigned long diff;
pthread_t th;
int err;
err = clock_getcpuclockid(0, &process_clock);
if (err)
return 1;
err = pthread_getcpuclockid(pthread_self(), &my_thread_clock);
if (err)
return 1;
pthread_barrier_init(&barrier, NULL, 2);
err = pthread_create(&th, NULL, chew_cpu, NULL);
if (err)
return 1;
err = pthread_getcpuclockid(th, &th_clock);
if (err)
return 1;
pthread_barrier_wait(&barrier);
err = clock_gettime(process_clock, &process_before);
if (err)
return 1;
err = clock_gettime(my_thread_clock, &me_before);
if (err)
return 1;
err = clock_gettime(th_clock, &th_before);
if (err)
return 1;
sleeptime.tv_sec = 0;
sleeptime.tv_nsec = 500000000;
nanosleep(&sleeptime, NULL);
err = clock_gettime(th_clock, &th_after);
if (err)
return 1;
err = clock_gettime(my_thread_clock, &me_after);
if (err)
return 1;
err = clock_gettime(process_clock, &process_after);
if (err)
return 1;
diff = process_after.tv_nsec - process_before.tv_nsec;
printf("process: before(%lu.%.9lu) after(%lu.%.9lu) diff(%lu)\n",
process_before.tv_sec, process_before.tv_nsec,
process_after.tv_sec, process_after.tv_nsec, diff);
diff = th_after.tv_nsec - th_before.tv_nsec;
printf("thread: before(%lu.%.9lu) after(%lu.%.9lu) diff(%lu)\n",
th_before.tv_sec, th_before.tv_nsec,
th_after.tv_sec, th_after.tv_nsec, diff);
diff = me_after.tv_nsec - me_before.tv_nsec;
printf("self: before(%lu.%.9lu) after(%lu.%.9lu) diff(%lu)\n",
me_before.tv_sec, me_before.tv_nsec,
me_after.tv_sec, me_after.tv_nsec, diff);
return 0;
}
This is due to us using p->se.sum_exec_runtime in
thread_group_cputime() where we iterate the thread group and sum all
data. This does not take time since the last schedule operation (tick
or otherwise) into account. We can cure this by using
task_sched_runtime() at the cost of having to take locks.
This also means we can (and must) do away with
thread_group_sched_runtime() since the modified thread_group_cputime()
is now more accurate and would deadlock when called from
thread_group_sched_runtime().
Reported-by: David Miller <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/1314874459.7945.22.camel@twins
Cc: [email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Francis reports that s2r gets him spurious NMIs, this is because the
suspend code leaves the boot cpu up and running.
Cure this by adding a suspend notifier. The problem is that hotplug
and suspend are completely un-serialized and the PM notifiers run
before the suspend cpu unplug of all but the boot cpu.
This leaves a window where the user can initialize another hotplug
operation (either remove or add a cpu) resulting in either one too
many or one too few hotplug ops. Thus we cannot use the hotplug code
for the suspend case.
There's another reason to not use the hotplug code, which is that the
hotplug code totally destroys the perf state, we can do better for
suspend and simply remove all counters from the PMU so that we can
re-instate them on resume.
Reported-by: Francis Moreau <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
On -rt kfree() can schedule, but CPU_STARTING is before the CPU is
fully up and running. These are contradictory, so avoid it. Instead
push the kfree() to CPU_ONLINE where we're free to schedule.
Reported-by: Thomas Gleixner <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
match_held_lock() was assuming it was being called on a lock class
that had already seen usage.
This condition was true for bug-free code using lockdep_assert_held(),
since you're in fact holding the lock when calling it. However the
assumption fails the moment you assume the assertion can fail, which
is the whole point of having the assertion in the first place.
Anyway, now that there's more lockdep_is_held() users, notably
__rcu_dereference_check(), its much easier to trigger this since we
test for a number of locks and we only need to hold any one of them to
be good.
Reported-by: Sergey Senozhatsky <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/1312547787.28695.2.camel@twins
Signed-off-by: Ingo Molnar <[email protected]>
On Sun, 2011-07-24 at 21:06 -0400, Arnaud Lacombe wrote:
> /src/linux/linux/kernel/lockdep.c: In function 'mark_held_locks':
> /src/linux/linux/kernel/lockdep.c:2471:31: warning: comparison of
> distinct pointer types lacks a cast
The warning is harmless in this case, but the below makes it go away.
Reported-by: Arnaud Lacombe <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/1311588599.2617.56.camel@laptop
Signed-off-by: Ingo Molnar <[email protected]>
Thomas noticed that a lock marked with lockdep_set_novalidate_class()
will still trigger warnings for IRQ inversions. Cure this by skipping
those when marking irq state.
Reported-and-tested-by: Thomas Gleixner <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Use for_each_leaf_cfs_rq() instead of list_for_each_entry_rcu(), this
achieves that load_balance_fair() only iterates those task_groups that
actually have tasks on busiest, and that we iterate bottom-up, trying to
move light groups before the heavier ones.
No idea if it will actually work out to be beneficial in practice, does
anybody have a cgroup workload that might show a difference one way or
the other?
[ Also move update_h_load to sched_fair.c, loosing #ifdef-ery ]
Signed-off-by: Peter Zijlstra <[email protected]>
Reviewed-by: Paul Turner <[email protected]>
Link: http://lkml.kernel.org/r/1310557009.2586.28.camel@twins
Signed-off-by: Ingo Molnar <[email protected]>
The rcu_read_unlock_special() function relies on in_irq() to exclude
scheduler activity from interrupt level. This fails because exit_irq()
can invoke the scheduler after clearing the preempt_count() bits that
in_irq() uses to determine that it is at interrupt level. This situation
can result in failures as follows:
$task IRQ SoftIRQ
rcu_read_lock()
/* do stuff */
<preempt> |= UNLOCK_BLOCKED
rcu_read_unlock()
--t->rcu_read_lock_nesting
irq_enter();
/* do stuff, don't use RCU */
irq_exit();
sub_preempt_count(IRQ_EXIT_OFFSET);
invoke_softirq()
ttwu();
spin_lock_irq(&pi->lock)
rcu_read_lock();
/* do stuff */
rcu_read_unlock();
rcu_read_unlock_special()
rcu_report_exp_rnp()
ttwu()
spin_lock_irq(&pi->lock) /* deadlock */
rcu_read_unlock_special(t);
Ed can simply trigger this 'easy' because invoke_softirq() immediately
does a ttwu() of ksoftirqd/# instead of doing the in-place softirq stuff
first, but even without that the above happens.
Cure this by also excluding softirqs from the
rcu_read_unlock_special() handler and ensuring the force_irqthreads
ksoftirqd/# wakeup is done from full softirq context.
[ Alternatively, delaying the ->rcu_read_lock_nesting decrement
until after the special handling would make the thing more robust
in the face of interrupts as well. And there is a separate patch
for that. ]
Cc: Thomas Gleixner <[email protected]>
Reported-and-tested-by: Ed Tomlinson <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>
Ensure scheduler_ipi() calls irq_{enter,exit} when it does some actual
work. Traditionally we never did any actual work from the resched IPI
and all magic happened in the return from interrupt path.
Now that we do do some work, we need to ensure irq_{enter,exit} are
called so that we don't confuse things.
This affects things like timekeeping, NO_HZ and RCU, basically
everything with a hook in irq_enter/exit.
Explicit examples of things going wrong are:
sched_clock_cpu() -- has a callback when leaving NO_HZ state to take
a new reading from GTOD and TSC. Without this
callback, time is stuck in the past.
RCU -- needs in_irq() to work in order to avoid some nasty deadlocks
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>
Allow for sched_domain spans that overlap by giving such domains their
own sched_group list instead of sharing the sched_groups amongst
each-other.
This is needed for machines with more than 16 nodes, because
sched_domain_node_span() will generate a node mask from the
16 nearest nodes without regard if these masks have any overlap.
Currently sched_domains have a sched_group that maps to their child
sched_domain span, and since there is no overlap we share the
sched_group between the sched_domains of the various CPUs. If however
there is overlap, we would need to link the sched_group list in
different ways for each cpu, and hence sharing isn't possible.
In order to solve this, allocate private sched_groups for each CPU's
sched_domain but have the sched_groups share a sched_group_power
structure such that we can uniquely track the power.
Reported-and-tested-by: Anton Blanchard <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Commit 3fe1698b7f ("sched: Deal with non-atomic min_vruntime reads
on 32bit") forgot to initialize min_vruntime_copy which could lead to
an infinite while loop in task_waking_fair() under some circumstances
(early boot, lucky timing).
[ This bug was also reported by others that blamed it on the RCU
initialization problems ]
Reported-and-tested-by: Bruno Wolff III <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Reviewed-by: Paul E. McKenney <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Since only samples call perf_output_sample() its much saner (and more
correct) to put the sample logic in there than in the
perf_output_begin()/perf_output_end() pair.
Saves a useless argument, reduces conditionals and shrinks
struct perf_output_handle, win!
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
The nmi parameter indicated if we could do wakeups from the current
context, if not, we would set some state and self-IPI and let the
resulting interrupt do the wakeup.
For the various event classes:
- hardware: nmi=0; PMI is in fact an NMI or we run irq_work_run from
the PMI-tail (ARM etc.)
- tracepoint: nmi=0; since tracepoint could be from NMI context.
- software: nmi=[0,1]; some, like the schedule thing cannot
perform wakeups, and hence need 0.
As one can see, there is very little nmi=1 usage, and the down-side of
not using it is that on some platforms some software events can have a
jiffy delay in wakeup (when arch_irq_work_raise isn't implemented).
The up-side however is that we can remove the nmi parameter and save a
bunch of conditionals in fast paths.
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Michael Cree <[email protected]>
Cc: Will Deacon <[email protected]>
Cc: Deng-Cheng Zhu <[email protected]>
Cc: Anton Blanchard <[email protected]>
Cc: Eric B Munson <[email protected]>
Cc: Heiko Carstens <[email protected]>
Cc: Paul Mundt <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Jason Wessel <[email protected]>
Cc: Don Zickus <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Release console_sem after unlocking the logbuf_lock so that we don't
generate wakeups while holding logbuf_lock. This avoids some lock
inversion troubles once we remove the lockdep_off bits between
logbuf_lock and rq->lock (prints while holding rq->lock vs doing
wakeups while holding logbuf_lock).
There's of course still an actual deadlock where the printk()s under
rq->lock will issue a wakeup from the up() call, but lockdep won't
warn about that since semaphores are not tracked.
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
The main lock_is_held() user is lockdep_assert_held(), avoid false
assertions in lockdep_off() sections by unconditionally reporting the
lock is taken.
[ the reason this is important is a lockdep_assert_held() in ttwu()
which triggers a warning under lockdep_off() as in printk() which
can trigger another wakeup and lock up due to spinlock
recursion, as reported and heroically debugged by Arne Jansen ]
Reported-and-tested-by: Arne Jansen <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: <[email protected]>
Link: http://lkml.kernel.org/r/1307398759.2497.966.camel@laptop
Signed-off-by: Ingo Molnar <[email protected]>
While looking over the code I found that with the ttwu rework the
nr_wakeups_migrate test broke since we now switch cpus prior to
calling ttwu_stat(), hence the test is always true.
Cure this by passing the migration state in wake_flags. Also move the
whole test under CONFIG_SMP, its hard to migrate tasks on UP :-)
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Markus reported that commit 317f394160 ("sched: Move the second half
of ttwu() to the remote cpu") caused some accounting funnies on his AMD
Phenom II X4, such as weird 'top' results.
It turns out that this is due to non-synced TSC and the queued remote
wakeups stopped coupeling the two relevant cpu clocks, which leads to
wakeups seeing time jumps, which in turn lead to skewed runtime stats.
Add an explicit call to sched_clock_cpu() to couple the per-cpu clocks
to restore the normal flow of time.
Reported-and-tested-by: Markus Trippelsdorf <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/1306835745.2353.3.camel@twins
Signed-off-by: Ingo Molnar <[email protected]>
Inspired by an analysis from Hugh on why again all this doesn't explode
in our face.
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
In order to always call list_del_event() on the correct cpu if the
event is part of an active context and avoid having to do two IPIs,
change the close() semantics slightly.
The current perf_event_disable() call would disable a whole group if
the event that's being closed is the group leader, whereas the new
code keeps the group siblings enabled.
People should not rely on this behaviour and I don't think they do,
but in case we find they do, the fix is easy and we have to take the
double IPI cost.
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Vince Weaver <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Instead of tracking if a context is active or not, track which events
of the context are active. By making it a bitmask of
EVENT_PINNED|EVENT_FLEXIBLE we can simplify some of the scheduling
routines since it can avoid adding events that are already active.
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Currently __perf_install_in_context() will try and schedule in the
event irrespective of our event scheduling rules, that is, we try to
schedule CPU-pinned, TASK-pinned, CPU-flexible, TASK-flexible, but
when creating a new event we simply try and schedule it on top of
whatever is already on the PMU, this can lead to errors for pinned
events.
Therefore, simplify things and simply schedule everything out, add the
event to the corresponding context and schedule everything back in.
This also nicely handles the case where with
__ARCH_WANT_INTERRUPTS_ON_CTXSW the IPI can come right in the middle
of schedule, before we managed to call perf_event_task_sched_in().
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Make task_ctx_sched_*() imply EVENT_ALL, since anything less will not
actually have scheduled the task in/out at all.
Since there's no site that schedules all of a task in (due to the
interleave with flexible cpuctx) we can remove this function.
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Currently we only hold one ctx->lock at a time, which results in us
flipping back and forth between cpuctx->ctx.lock and task_ctx->lock.
Avoid this and gain large atomic regions by holding both locks. We
nest the task lock inside the cpu lock, since with task scheduling we
might have to change task ctx while holding the cpu ctx lock.
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
It is not necessary to use waitqueues for the RCU kthreads because
we always know exactly which thread is to be awakened. In addition,
wake_up() only issues an actual wakeup when there is a thread waiting on
the queue, which was why there was an extra explicit wake_up_process()
to get the RCU kthreads started.
Eliminating the waitqueues (and wake_up()) in favor of wake_up_process()
eliminates the need for the initial wake_up_process() and also shrinks
the data structure size a bit. The wakeup logic is placed in a new
rcu_wait() macro.
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Marc reported that e4a52bcb9 (sched: Remove rq->lock from the first
half of ttwu()) broke his ARM-SMP machine. Now ARM is one of the few
__ARCH_WANT_INTERRUPTS_ON_CTXSW users, so that exception in the ttwu()
code was suspect.
Yong found that the interrupt could hit after context_switch() changes
current but before it clears p->on_cpu, if that interrupt were to
attempt a wake-up of p we would indeed find ourselves spinning in IRQ
context.
Fix this by reverting to the old behaviour for this situation and
perform a full remote wake-up.
Cc: Frank Rowand <[email protected]>
Cc: Yong Zhang <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Reported-by: Marc Zyngier <[email protected]>
Tested-by: Marc Zyngier <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Hugh says:
"The only significant loser, I think, would be page reclaim (when
concurrent with truncation): could spin for a long time waiting for
the i_mmap_mutex it expects would soon be dropped? "
Counter points:
- cpu contention makes the spin stop (need_resched())
- zap pages should be freeing pages at a higher rate than reclaim
ever can
I think the simplification of the truncate code is definitely worth it.
Effectively reverts: 2aa15890f3 ("mm: prevent concurrent
unmap_mapping_range() on the same inode") and takes out the code that
caused its problem.
Signed-off-by: Peter Zijlstra <[email protected]>
Reviewed-by: KAMEZAWA Hiroyuki <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: David Miller <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Cc: Russell King <[email protected]>
Cc: Paul Mundt <[email protected]>
Cc: Jeff Dike <[email protected]>
Cc: Richard Weinberger <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: KOSAKI Motohiro <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Namhyung Kim <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Rework the existing mmu_gather infrastructure.
The direct purpose of these patches was to allow preemptible mmu_gather,
but even without that I think these patches provide an improvement to the
status quo.
The first 9 patches rework the mmu_gather infrastructure. For review
purpose I've split them into generic and per-arch patches with the last of
those a generic cleanup.
The next patch provides generic RCU page-table freeing, and the followup
is a patch converting s390 to use this. I've also got 4 patches from
DaveM lined up (not included in this series) that uses this to implement
gup_fast() for sparc64.
Then there is one patch that extends the generic mmu_gather batching.
After that follow the mm preemptibility patches, these make part of the mm
a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to
mutexes which together with the mmu_gather rework makes mmu_gather
preemptible as well.
Making i_mmap_lock a mutex also enables a clean-up of the truncate code.
This also allows for preemptible mmu_notifiers, something that XPMEM I
think wants.
Furthermore, it removes the new and universially detested unmap_mutex.
This patch:
Remove the first obstacle towards a fully preemptible mmu_gather.
The current scheme assumes mmu_gather is always done with preemption
disabled and uses per-cpu storage for the page batches. Change this to
try and allocate a page for batching and in case of failure, use a small
on-stack array to make some progress.
Preemptible mmu_gather is desired in general and usable once i_mmap_lock
becomes a mutex. Doing it before the mutex conversion saves us from
having to rework the code by moving the mmu_gather bits inside the
pte_lock.
Also avoid flushing the tlb batches from under the pte lock, this is
useful even without the i_mmap_lock conversion as it significantly reduces
pte lock hold times.
[[email protected]: fix comment tpyo]
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: David Miller <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Cc: Russell King <[email protected]>
Cc: Paul Mundt <[email protected]>
Cc: Jeff Dike <[email protected]>
Cc: Richard Weinberger <[email protected]>
Cc: Tony Luck <[email protected]>
Reviewed-by: KAMEZAWA Hiroyuki <[email protected]>
Acked-by: Hugh Dickins <[email protected]>
Acked-by: Mel Gorman <[email protected]>
Cc: KOSAKI Motohiro <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Namhyung Kim <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently the x86 backend incorrectly assumes that any BRANCH_INSN
with sample_period==1 is a BTS request. This is not true when we do
frequency driven profiling such as 'perf record -e branches'.
Solves this error:
$ perf record -e branches ./array
Error: sys_perf_event_open() syscall returned with 95 (Operation not supported).
Signed-off-by: Peter Zijlstra <[email protected]>
Reported-by: Ingo Molnar <[email protected]>
Cc: "Metzger, Markus T" <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Link: http://lkml.kernel.org/n/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Vladis Kletnieks reported a new RCU debug warning in the scheduler.
Since commit dce840a087 ("sched: Dynamically allocate sched_domain/
sched_group data-structures") the sched_domain trees are protected by
RCU instead of RCU-sched.
This means that we need to include rcu_read_lock() protection when we
iterate them since disabling preemption doesn't suffice anymore.
Reported-by: [email protected]
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/1302882741.2388.241.camel@twins
Signed-off-by: Ingo Molnar <[email protected]>
Now that we've removed the rq->lock requirement from the first part of
ttwu() and can compute placement without holding any rq->lock, ensure
we execute the second half of ttwu() on the actual cpu we want the
task to run on.
This avoids having to take rq->lock and doing the task enqueue
remotely, saving lots on cacheline transfers.
As measured using: http://oss.oracle.com/~mason/sembench.c
$ for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i; done
$ echo 4096 32000 64 128 > /proc/sys/kernel/sem
$ ./sembench -t 2048 -w 1900 -o 0
unpatched: run time 30 seconds 647278 worker burns per second
patched: run time 30 seconds 816715 worker burns per second
Reviewed-by: Frank Rowand <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Currently ttwu() does two rq->lock acquisitions, once on the task's
old rq, holding it over the p->state fiddling and load-balance pass.
Then it drops the old rq->lock to acquire the new rq->lock.
By having serialized ttwu(), p->sched_class, p->cpus_allowed with
p->pi_lock, we can now drop the whole first rq->lock acquisition.
The p->pi_lock serializing concurrent ttwu() calls protects p->state,
which we will set to TASK_WAKING to bridge possible p->pi_lock to
rq->lock gaps and serialize set_task_cpu() calls against
task_rq_lock().
The p->pi_lock serialization of p->sched_class allows us to call
scheduling class methods without holding the rq->lock, and the
serialization of p->cpus_allowed allows us to do the load-balancing
bits without races.
Reviewed-by: Frank Rowand <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
In order to be able to call set_task_cpu() while either holding
p->pi_lock or task_rq(p)->lock we need to hold both locks in order to
stabilize task_rq().
This makes task_rq_lock() acquire both locks, and have
__task_rq_lock() validate that p->pi_lock is held. This increases the
locking overhead for most scheduler syscalls but allows reduction of
rq->lock contention for some scheduler hot paths (ttwu).
Reviewed-by: Frank Rowand <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Currently p->pi_lock already serializes p->sched_class, also put
p->cpus_allowed and try_to_wake_up() under it, this prepares the way
to do the first part of ttwu() without holding rq->lock.
By having p->sched_class and p->cpus_allowed serialized by p->pi_lock,
we prepare the way to call select_task_rq() without holding rq->lock.
Reviewed-by: Frank Rowand <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Provide a generic p->on_rq because the p->se.on_rq semantics are
unfavourable for lockless wakeups but needed for sched_fair.
In particular, p->on_rq is only cleared when we actually dequeue the
task in schedule() and not on any random dequeue as done by things
like __migrate_task() and __sched_setscheduler().
This also allows us to remove p->se usage from !sched_fair code.
Reviewed-by: Frank Rowand <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Peter Zijlstra