[RT PATCH 1/2] kernel/cpu_chill: use schedule_hrtimeout()

[RT PATCH 1/2] kernel/cpu_chill: use schedule_hrtimeout()

[Sebastian Andrzej Siewior]

cpu_chill() uses __hrtimer_nanosleep() with a hack to get
TASK_UNINTERRUPTIBLE state. It seems to work but it does not.
The problem is that we may have a signal pending and receive a sporadic
wake up _before_ the timer expires. At that point the code will look at
->restart_block and use it based on its value from the previous syscall
resulting in a possible copy-to-user.
Instead all this trouble, we could use schedule_hrtimeout() which is not
an user-space interface and does not have this side effects.

Cc: stable-rt@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 kernel/time/hrtimer.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 6c77643eaf02..c0efc22ba635 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1873,14 +1873,13 @@ COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
  */
 void cpu_chill(void)
 {
-	struct timespec64 tu = {
-		.tv_nsec = NSEC_PER_MSEC,
-	};
+	ktime_t chill_time;
 	unsigned int freeze_flag = current->flags & PF_NOFREEZE;
 
+	chill_time = ktime_set(0, NSEC_PER_MSEC);
+	set_current_state(TASK_UNINTERRUPTIBLE);
 	current->flags |= PF_NOFREEZE;
-	__hrtimer_nanosleep(&tu, HRTIMER_MODE_REL_HARD, CLOCK_MONOTONIC,
-			    TASK_UNINTERRUPTIBLE);
+	schedule_hrtimeout(&chill_time, HRTIMER_MODE_REL_HARD);
 	if (!freeze_flag)
 		current->flags &= ~PF_NOFREEZE;
 }
-- 
2.16.2

[RT PATCH 2/2] Revert "cpu_chill: Add a UNINTERRUPTIBLE hrtimer_nanosleep"

[Sebastian Andrzej Siewior]

This reverts commit "cpu_chill: Add a UNINTERRUPTIBLE
hrtimer_nanosleep". Since cpu_chill() is now using schedule_hrtimeout()
we can remove that change since we have no users for it.

Cc: stable-rt@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 kernel/time/hrtimer.c | 21 ++++++---------------
 1 file changed, 6 insertions(+), 15 deletions(-)

diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index c0efc22ba635..ce2c2d04cbaa 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1742,13 +1742,12 @@ int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
 	return -ERESTART_RESTARTBLOCK;
 }
 
-static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode,
-				unsigned long state)
+static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
 {
 	struct restart_block *restart;
 
 	do {
-		set_current_state(state);
+		set_current_state(TASK_INTERRUPTIBLE);
 		hrtimer_start_expires(&t->timer, mode);
 
 		if (likely(t->task))
@@ -1786,15 +1785,13 @@ static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
 	hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
 				      HRTIMER_MODE_ABS, current);
 	hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
-	/* cpu_chill() does not care about restart state. */
-	ret = do_nanosleep(&t, HRTIMER_MODE_ABS, TASK_INTERRUPTIBLE);
+	ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
 	destroy_hrtimer_on_stack(&t.timer);
 	return ret;
 }
 
-static long __hrtimer_nanosleep(const struct timespec64 *rqtp,
-				const enum hrtimer_mode mode, const clockid_t clockid,
-				unsigned long state)
+long hrtimer_nanosleep(const struct timespec64 *rqtp,
+		       const enum hrtimer_mode mode, const clockid_t clockid)
 {
 	struct restart_block *restart;
 	struct hrtimer_sleeper t;
@@ -1807,7 +1804,7 @@ static long __hrtimer_nanosleep(const struct timespec64 *rqtp,
 
 	hrtimer_init_sleeper_on_stack(&t, clockid, mode, current);
 	hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
-	ret = do_nanosleep(&t, mode, state);
+	ret = do_nanosleep(&t, mode);
 	if (ret != -ERESTART_RESTARTBLOCK)
 		goto out;
 
@@ -1826,12 +1823,6 @@ static long __hrtimer_nanosleep(const struct timespec64 *rqtp,
 	return ret;
 }
 
-long hrtimer_nanosleep(const struct timespec64 *rqtp,
-		       const enum hrtimer_mode mode, const clockid_t clockid)
-{
-	return __hrtimer_nanosleep(rqtp, mode, clockid, TASK_INTERRUPTIBLE);
-}
-
 SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
 		struct timespec __user *, rmtp)
 {
-- 
2.16.2

Re: [RT PATCH 1/2] kernel/cpu_chill: use schedule_hrtimeout()

[Thomas Gleixner]

On Fri, 16 Mar 2018, Sebastian Andrzej Siewior wrote:

> cpu_chill() uses __hrtimer_nanosleep() with a hack to get
> TASK_UNINTERRUPTIBLE state. It seems to work but it does not.
> The problem is that we may have a signal pending and receive a sporadic
> wake up _before_ the timer expires. At that point the code will look at

_We_ have nothing pending. Please describe it in a factual way and not like
you are running the code yourself.

> ->restart_block and use it based on its value from the previous syscall
> resulting in a possible copy-to-user.

What you want to describe is:

If a task calls cpu_chill() and gets woken up by a regular or spurious
wakeup and has a signal pending, then it exits the sleep loop in
do_nanosleep() and sets up the restart block. If restart->nanosleep.type is
not TI_NONE then this results in accessing a stale user pointer from a
previously interrupted syscall and a copy to user based on the stale
pointer or a BUG() when 'type' is not supported in nanosleep_copyout().

> instead all this trouble, we could use schedule_hrtimeout() which is not

Sentences start with an uppercase letter.

> instead all this trouble, we could use schedule_hrtimeout() which is not

s/we could// ....

> an user-space interface and does not have this side effects.

hmm?

> Cc: stable-rt@vger.kernel.org
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> ---
>  kernel/time/hrtimer.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
> index 6c77643eaf02..c0efc22ba635 100644
> --- a/kernel/time/hrtimer.c
> +++ b/kernel/time/hrtimer.c
> @@ -1873,14 +1873,13 @@ COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
>   */
>  void cpu_chill(void)
>  {
> -	struct timespec64 tu = {
> -		.tv_nsec = NSEC_PER_MSEC,
> -	};
> +	ktime_t chill_time;
>  	unsigned int freeze_flag = current->flags & PF_NOFREEZE;
>  
> +	chill_time = ktime_set(0, NSEC_PER_MSEC);
> +	set_current_state(TASK_UNINTERRUPTIBLE);
>  	current->flags |= PF_NOFREEZE;
> -	__hrtimer_nanosleep(&tu, HRTIMER_MODE_REL_HARD, CLOCK_MONOTONIC,
> -			    TASK_UNINTERRUPTIBLE);
> +	schedule_hrtimeout(&chill_time, HRTIMER_MODE_REL_HARD);
>  	if (!freeze_flag)
>  		current->flags &= ~PF_NOFREEZE;
>  }
> -- 
> 2.16.2
> 
> 

[RT PATCH 1/2 v2] kernel/cpu_chill: use schedule_hrtimeout()

[Sebastian Andrzej Siewior]

If a task calls cpu_chill() and gets woken up by a regular or spurious
wakeup and has a signal pending, then it exits the sleep loop in
do_nanosleep() and sets up the restart block. If restart->nanosleep.type is
not TI_NONE then this results in accessing a stale user pointer from a
previously interrupted syscall and a copy to user based on the stale
pointer or a BUG() when 'type' is not supported in nanosleep_copyout().

Instead all this trouble, use schedule_hrtimeout().

Cc: stable-rt@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 kernel/time/hrtimer.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 6c77643eaf02..c0efc22ba635 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1873,14 +1873,13 @@ COMPAT_SYSCALL_DEFINE2(nanosleep, struct compat_timespec __user *, rqtp,
  */
 void cpu_chill(void)
 {
-	struct timespec64 tu = {
-		.tv_nsec = NSEC_PER_MSEC,
-	};
+	ktime_t chill_time;
 	unsigned int freeze_flag = current->flags & PF_NOFREEZE;
 
+	chill_time = ktime_set(0, NSEC_PER_MSEC);
+	set_current_state(TASK_UNINTERRUPTIBLE);
 	current->flags |= PF_NOFREEZE;
-	__hrtimer_nanosleep(&tu, HRTIMER_MODE_REL_HARD, CLOCK_MONOTONIC,
-			    TASK_UNINTERRUPTIBLE);
+	schedule_hrtimeout(&chill_time, HRTIMER_MODE_REL_HARD);
 	if (!freeze_flag)
 		current->flags &= ~PF_NOFREEZE;
 }
-- 
2.16.2