On 5/1/2018 6:49 PM, Peter Zijlstra wrote:
On 5/1/2018 5:01 PM, Peter Zijlstra wrote:
Let me ponder what the best solution is, it's a bit of a mess.
So there's:
- TASK_PARKED, which we can make a special state; this solves the
problem because then wait_task_inactive() is guaranteed to see
TASK_PARKED and DTRT.
Yes this should work, as it takes pi-lock and blocking kthread_parkme.
- complete(&kthread->parked), which we can do inside schedule(); this
solves the problem because then kthread_park() will not return early
and the task really is blocked.
I think complete will not help, as problem is like below :
Control Thread CPUHP thread
cpuhp_thread_fun
Wake control thread
complete(&st->done);
takedown_cpu
kthread_park
set_bit(KTHREAD_SHOULD_PARK
Here cpuhp is looping,
//success case
Generally when issue is not
coming
it schedule out by below :
ht->thread_should_run(td->cpu
scheduler
//failure case
before schedule
loop check
(kthread_should_park()
enter here as PARKED set
wake_up_process(k)
__kthread_parkme
complete(&self->parked);
SETS RUNNING
schedule
wait_for_completion(&kthread->parked);
So even we protect complete, it may see TASK_RUNNING as final state.
That's why we took pi-lock , So wake_up_process either see TASK_RUNNING
so return or it will exit early before parkme call.
Please correct me , if i misunderstood it. With pi-lock approach we are
not seeing issue. SO you first solution should fix this(put parked in
special state).
Regards
Gaurav
and I'm fairly sure I thought of a 3rd way to cure things, but now that
I'm writing things down I cannot seem to remember :/ -- could be we muck
with wait_task_inactive().
In any case, I hate all of them, but I think the completion one is the
least horrible because it gives the strongest guarantees and cleans up
most. But promoting TASK_PARKED to special certainly is the earier
patch.
The below boots, but that's about all I did with it. Opinions?
---
--- a/include/linux/kthread.h
+++ b/include/linux/kthread.h
@@ -62,6 +62,7 @@ void *kthread_probe_data(struct task_str
int kthread_park(struct task_struct *k);
void kthread_unpark(struct task_struct *k);
void kthread_parkme(void);
+void kthread_park_complete(struct task_struct *k);
int kthreadd(void *unused);
extern struct task_struct *kthreadd_task;
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -55,7 +55,6 @@ enum KTHREAD_BITS {
KTHREAD_IS_PER_CPU = 0,
KTHREAD_SHOULD_STOP,
KTHREAD_SHOULD_PARK,
- KTHREAD_IS_PARKED,
};
static inline void set_kthread_struct(void *kthread)
@@ -177,14 +176,12 @@ void *kthread_probe_data(struct task_str
static void __kthread_parkme(struct kthread *self)
{
- __set_current_state(TASK_PARKED);
- while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) {
- if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags))
- complete(&self->parked);
+ for (;;) {
+ set_current_state(TASK_PARKED);
+ if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
+ break;
schedule();
- __set_current_state(TASK_PARKED);
}
- clear_bit(KTHREAD_IS_PARKED, &self->flags);
__set_current_state(TASK_RUNNING);
}
@@ -194,6 +191,11 @@ void kthread_parkme(void)
}
EXPORT_SYMBOL_GPL(kthread_parkme);
+void kthread_park_complete(struct task_struct *k)
+{
+ complete(&to_kthread(k)->parked);
+}
+
static int kthread(void *_create)
{
/* Copy data: it's on kthread's stack */
@@ -450,22 +452,15 @@ void kthread_unpark(struct task_struct *
{
struct kthread *kthread = to_kthread(k);
- clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
/*
- * We clear the IS_PARKED bit here as we don't wait
- * until the task has left the park code. So if we'd
- * park before that happens we'd see the IS_PARKED bit
- * which might be about to be cleared.
+ * Newly created kthread was parked when the CPU was offline.
+ * The binding was lost and we need to set it again.
*/
- if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
- /*
- * Newly created kthread was parked when the CPU was offline.
- * The binding was lost and we need to set it again.
- */
- if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
- __kthread_bind(k, kthread->cpu, TASK_PARKED);
- wake_up_state(k, TASK_PARKED);
- }
+ if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
+ __kthread_bind(k, kthread->cpu, TASK_PARKED);
+
+ clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
+ wake_up_state(k, TASK_PARKED);
}
EXPORT_SYMBOL_GPL(kthread_unpark);
@@ -488,12 +483,13 @@ int kthread_park(struct task_struct *k)
if (WARN_ON(k->flags & PF_EXITING))
return -ENOSYS;
- if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
- set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
- if (k != current) {
- wake_up_process(k);
- wait_for_completion(&kthread->parked);
- }
+ if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
+ return -EBUSY;
+
+ set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
+ if (k != current) {
+ wake_up_process(k);
+ wait_for_completion(&kthread->parked);
}
return 0;
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7,6 +7,8 @@
*/
#include "sched.h"
+#include <linux/kthread.h>
+
#include <asm/switch_to.h>
#include <asm/tlb.h>
@@ -2718,20 +2720,28 @@ static struct rq *finish_task_switch(str
membarrier_mm_sync_core_before_usermode(mm);
mmdrop(mm);
}
- if (unlikely(prev_state == TASK_DEAD)) {
- if (prev->sched_class->task_dead)
- prev->sched_class->task_dead(prev);
+ if (unlikely(prev_state & (TASK_DEAD|TASK_PARKED))) {
+ switch (prev_state) {
+ case TASK_DEAD:
+ if (prev->sched_class->task_dead)
+ prev->sched_class->task_dead(prev);
+
+ /*
+ * Remove function-return probe instances associated with this
+ * task and put them back on the free list.
+ */
+ kprobe_flush_task(prev);
- /*
- * Remove function-return probe instances associated with this
- * task and put them back on the free list.
- */
- kprobe_flush_task(prev);
+ /* Task is done with its stack. */
+ put_task_stack(prev);
- /* Task is done with its stack. */
- put_task_stack(prev);
+ put_task_struct(prev);
+ break;
- put_task_struct(prev);
+ case TASK_PARKED:
+ kthread_park_complete(prev);
+ break;
+ }
}
tick_nohz_task_switch();
--
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center,
Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project.
--
To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
