Re: rt14: strace -> migrate_disable_atomic imbalance

From: Peter Zijlstra
Date: Thu Sep 22 2011 - 07:46:32 EST

On Thu, 2011-09-22 at 13:31 +0200, Peter Zijlstra wrote:
> On Thu, 2011-09-22 at 12:00 +0200, Peter Zijlstra wrote:
> > OK, this one seems to be better.. But its quite vile, not sure I
> > actually like it anymore.
> >
> There's also a small matter of it actually being slower than what we had
> when tested with: strace hackbench > /dev/null 2>&1.
>
> curses.. even adding a busy wait on p->on_cpu in front.

Gah, its one of those benchmarks where if you rebuild+reboot the numbers
change. Latest patch here..

---
Subject: sched: Rewrite wait_task_inactive()
From: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
Date: Wed Sep 21 21:34:05 CEST 2011

Instead of using a combination of busy-waits and sleeps to poll a
task's state, ensure we get a notification and properly block.

This makes preempt notifiers unconditional, but since pretty much
everybody already has those enabled in their config it shouldn't be a
big deal (I hope).

The notification is somewhat horrid since the sched_out notifier is
called with rq->lock held and interrupts disabled, so we can't
actually do the wakeup from there. However the sched_in notifier is
called without holding rq->lock and with interrupts enable.

So what we do is we register a preempt notifier on the task we want,
that notifier will move itself to the next task passed into the
sched_out function, the sched_in notification of the next task, will do
the actual work... most horrid.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
---
arch/ia64/kvm/Kconfig | 1
arch/powerpc/kvm/Kconfig | 1
arch/s390/kvm/Kconfig | 1
arch/tile/kvm/Kconfig | 1
arch/x86/kvm/Kconfig | 1
include/linux/kvm_host.h | 2
include/linux/preempt.h | 4
include/linux/sched.h | 2
init/Kconfig | 3
kernel/sched.c | 191 +++++++++++++++++++++--------------------------
10 files changed, 86 insertions(+), 121 deletions(-)
Index: linux-2.6/arch/ia64/kvm/Kconfig
===================================================================
--- linux-2.6.orig/arch/ia64/kvm/Kconfig
+++ linux-2.6/arch/ia64/kvm/Kconfig
@@ -22,7 +22,6 @@ config KVM
depends on HAVE_KVM && MODULES && EXPERIMENTAL
# for device assignment:
depends on PCI
- select PREEMPT_NOTIFIERS
select ANON_INODES
select HAVE_KVM_IRQCHIP
select KVM_APIC_ARCHITECTURE
Index: linux-2.6/arch/powerpc/kvm/Kconfig
===================================================================
--- linux-2.6.orig/arch/powerpc/kvm/Kconfig
+++ linux-2.6/arch/powerpc/kvm/Kconfig
@@ -18,7 +18,6 @@ if VIRTUALIZATION

config KVM
bool
- select PREEMPT_NOTIFIERS
select ANON_INODES

config KVM_BOOK3S_HANDLER
Index: linux-2.6/arch/s390/kvm/Kconfig
===================================================================
--- linux-2.6.orig/arch/s390/kvm/Kconfig
+++ linux-2.6/arch/s390/kvm/Kconfig
@@ -19,7 +19,6 @@ config KVM
def_tristate y
prompt "Kernel-based Virtual Machine (KVM) support"
depends on HAVE_KVM && EXPERIMENTAL
- select PREEMPT_NOTIFIERS
select ANON_INODES
---help---
Support hosting paravirtualized guest machines using the SIE
Index: linux-2.6/arch/tile/kvm/Kconfig
===================================================================
--- linux-2.6.orig/arch/tile/kvm/Kconfig
+++ linux-2.6/arch/tile/kvm/Kconfig
@@ -19,7 +19,6 @@ if VIRTUALIZATION
config KVM
tristate "Kernel-based Virtual Machine (KVM) support"
depends on HAVE_KVM && MODULES && EXPERIMENTAL
- select PREEMPT_NOTIFIERS
select ANON_INODES
---help---
Support hosting paravirtualized guest machines.
Index: linux-2.6/arch/x86/kvm/Kconfig
===================================================================
--- linux-2.6.orig/arch/x86/kvm/Kconfig
+++ linux-2.6/arch/x86/kvm/Kconfig
@@ -24,7 +24,6 @@ config KVM
depends on PCI
# for TASKSTATS/TASK_DELAY_ACCT:
depends on NET
- select PREEMPT_NOTIFIERS
select MMU_NOTIFIER
select ANON_INODES
select HAVE_KVM_IRQCHIP
Index: linux-2.6/include/linux/kvm_host.h
===================================================================
--- linux-2.6.orig/include/linux/kvm_host.h
+++ linux-2.6/include/linux/kvm_host.h
@@ -111,9 +111,7 @@ enum {

struct kvm_vcpu {
struct kvm *kvm;
-#ifdef CONFIG_PREEMPT_NOTIFIERS
struct preempt_notifier preempt_notifier;
-#endif
int cpu;
int vcpu_id;
int srcu_idx;
Index: linux-2.6/include/linux/preempt.h
===================================================================
--- linux-2.6.orig/include/linux/preempt.h
+++ linux-2.6/include/linux/preempt.h
@@ -101,8 +101,6 @@ do { \

#endif /* CONFIG_PREEMPT_COUNT */

-#ifdef CONFIG_PREEMPT_NOTIFIERS
-
struct preempt_notifier;

/**
@@ -147,6 +145,4 @@ static inline void preempt_notifier_init
notifier->ops = ops;
}

-#endif
-
#endif /* __LINUX_PREEMPT_H */
Index: linux-2.6/include/linux/sched.h
===================================================================
--- linux-2.6.orig/include/linux/sched.h
+++ linux-2.6/include/linux/sched.h
@@ -1236,10 +1236,8 @@ struct task_struct {
struct sched_entity se;
struct sched_rt_entity rt;

-#ifdef CONFIG_PREEMPT_NOTIFIERS
/* list of struct preempt_notifier: */
struct hlist_head preempt_notifiers;
-#endif

/*
* fpu_counter contains the number of consecutive context switches
Index: linux-2.6/init/Kconfig
===================================================================
--- linux-2.6.orig/init/Kconfig
+++ linux-2.6/init/Kconfig
@@ -1403,9 +1403,6 @@ config STOP_MACHINE

source "block/Kconfig"

-config PREEMPT_NOTIFIERS
- bool
-
config PADATA
depends on SMP
bool
Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -2387,6 +2387,54 @@ struct migration_arg {

static int migration_cpu_stop(void *data);

+struct wait_task_inactive_blocked {
+ struct preempt_notifier notifier;
+ struct task_struct *waiter;
+};
+
+static void wait_task_inactive_sched_in(struct preempt_notifier *n, int cpu)
+{
+ struct task_struct *p;
+ struct wait_task_inactive_blocked *blocked =
+ container_of(n, struct wait_task_inactive_blocked, notifier);
+
+ hlist_del(&n->link);
+
+ p = ACCESS_ONCE(blocked->waiter);
+ blocked->waiter = NULL;
+ wake_up_process(p);
+}
+
+static void
+preempt_ops_sched_out_nop(struct preempt_notifier *n, struct task_struct *next)
+{
+}
+
+static struct preempt_ops wait_task_inactive_ops_post = {
+ .sched_in = wait_task_inactive_sched_in,
+ .sched_out = preempt_ops_sched_out_nop,
+};
+
+static void preempt_ops_sched_in_nop(struct preempt_notifier *n, int cpu)
+{
+}
+
+static void
+wait_task_inactive_sched_out(struct preempt_notifier *n, struct task_struct *next)
+{
+ if (current->on_rq) /* we're not inactive yet */
+ return;
+
+ hlist_del(&n->link);
+ n->ops = &wait_task_inactive_ops_post;
+ hlist_add_head(&n->link, &next->preempt_notifiers);
+}
+
+static struct preempt_ops wait_task_inactive_ops_pre = {
+ .sched_in = preempt_ops_sched_in_nop,
+ .sched_out = wait_task_inactive_sched_out,
+};
+
/*
* wait_task_inactive - wait for a thread to unschedule.
*
@@ -2405,93 +2453,49 @@ static int migration_cpu_stop(void *data
*/
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
{
+ unsigned long ncsw = 0;
unsigned long flags;
- int running, on_rq;
- unsigned long ncsw;
struct rq *rq;

- for (;;) {
- /*
- * We do the initial early heuristics without holding
- * any task-queue locks at all. We'll only try to get
- * the runqueue lock when things look like they will
- * work out!
- */
- rq = task_rq(p);
-
- /*
- * If the task is actively running on another CPU
- * still, just relax and busy-wait without holding
- * any locks.
- *
- * NOTE! Since we don't hold any locks, it's not
- * even sure that "rq" stays as the right runqueue!
- * But we don't care, since "task_running()" will
- * return false if the runqueue has changed and p
- * is actually now running somewhere else!
- */
- while (task_running(rq, p)) {
- if (match_state && unlikely(p->state != match_state))
- return 0;
- cpu_relax();
- }
-
- /*
- * Ok, time to look more closely! We need the rq
- * lock now, to be *sure*. If we're wrong, we'll
- * just go back and repeat.
- */
- rq = task_rq_lock(p, &flags);
- trace_sched_wait_task(p);
- running = task_running(rq, p);
- on_rq = p->on_rq;
- ncsw = 0;
- if (!match_state || p->state == match_state)
- ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
- task_rq_unlock(rq, p, &flags);
-
- /*
- * If it changed from the expected state, bail out now.
- */
- if (unlikely(!ncsw))
- break;
+ struct wait_task_inactive_blocked blocked = {
+ .notifier = {
+ .ops = &wait_task_inactive_ops_pre,
+ },
+ .waiter = current,
+ };

- /*
- * Was it really running after all now that we
- * checked with the proper locks actually held?
- *
- * Oops. Go back and try again..
- */
- if (unlikely(running)) {
- cpu_relax();
- continue;
- }
+ /*
+ * Busy wait for the task to stop running, the caller promised it
+ * wouldn't be long.
+ */
+ while (p->on_cpu) {
+ /* if we don't match the expected state, bail */
+ if (match_state && unlikely(p->state != match_state))
+ return 0;
+ cpu_relax();
+ }

- /*
- * It's not enough that it's not actively running,
- * it must be off the runqueue _entirely_, and not
- * preempted!
- *
- * So if it was still runnable (but just not actively
- * running right now), it's preempted, and we should
- * yield - it could be a while.
- */
- if (unlikely(on_rq)) {
- ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);
+ rq = task_rq_lock(p, &flags);
+ trace_sched_wait_task(p);
+ if (!p->on_rq) /* we're already blocked */
+ goto done;

- set_current_state(TASK_UNINTERRUPTIBLE);
- schedule_hrtimeout(&to, HRTIMER_MODE_REL);
- continue;
- }
+ hlist_add_head(&blocked.notifier.link, &p->preempt_notifiers);
+ task_rq_unlock(rq, p, &flags);

- /*
- * Ahh, all good. It wasn't running, and it wasn't
- * runnable, which means that it will never become
- * running in the future either. We're all done!
- */
- break;
+ for (;;) {
+ set_current_state(TASK_UNINTERRUPTIBLE);
+ if (!blocked.waiter)
+ break;
+ schedule();
}
+ __set_current_state(TASK_RUNNING);

+ rq = task_rq_lock(p, &flags);
+done:
+ if (!match_state || p->state == match_state)
+ ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
+ task_rq_unlock(rq, p, &flags);
return ncsw;
}

@@ -2519,9 +2523,7 @@ void kick_process(struct task_struct *p)
preempt_enable();
}
EXPORT_SYMBOL_GPL(kick_process);
-#endif /* CONFIG_SMP */

-#ifdef CONFIG_SMP
/*
* ->cpus_allowed is protected by both rq->lock and p->pi_lock
*/
@@ -2967,10 +2969,7 @@ static void __sched_fork(struct task_str
#endif

INIT_LIST_HEAD(&p->rt.run_list);
-
-#ifdef CONFIG_PREEMPT_NOTIFIERS
INIT_HLIST_HEAD(&p->preempt_notifiers);
-#endif
}

/*
@@ -3084,8 +3083,6 @@ void wake_up_new_task(struct task_struct
task_rq_unlock(rq, p, &flags);
}

-#ifdef CONFIG_PREEMPT_NOTIFIERS
-
/**
* preempt_notifier_register - tell me when current is being preempted & rescheduled
* @notifier: notifier struct to register
@@ -3111,9 +3108,9 @@ EXPORT_SYMBOL_GPL(preempt_notifier_unreg
static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
struct preempt_notifier *notifier;
- struct hlist_node *node;
+ struct hlist_node *node, *n;

- hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
+ hlist_for_each_entry_safe(notifier, node, n, &curr->preempt_notifiers, link)
notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

@@ -3122,26 +3119,12 @@ fire_sched_out_preempt_notifiers(struct
struct task_struct *next)
{
struct preempt_notifier *notifier;
- struct hlist_node *node;
+ struct hlist_node *node, *n;

- hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link)
+ hlist_for_each_entry_safe(notifier, node, n, &curr->preempt_notifiers, link)
notifier->ops->sched_out(notifier, next);
}

-#else /* !CONFIG_PREEMPT_NOTIFIERS */
-
-static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
-{
-}
-
-static void
-fire_sched_out_preempt_notifiers(struct task_struct *curr,
- struct task_struct *next)
-{
-}
-
-#endif /* CONFIG_PREEMPT_NOTIFIERS */
-
/**
* prepare_task_switch - prepare to switch tasks
* @rq: the runqueue preparing to switch
@@ -8312,9 +8295,7 @@ void __init sched_init(void)

set_load_weight(&init_task);

-#ifdef CONFIG_PREEMPT_NOTIFIERS
INIT_HLIST_HEAD(&init_task.preempt_notifiers);
-#endif

#ifdef CONFIG_SMP
open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/