[PATCH] improve lmk to avoid deadlock issue

[PATCH] improve lmk to avoid deadlock issue

[Wang, Biao]

From: "wang, biao" <biao.wang@intel.com>
Date: Thu, 30 Jul 2015 14:14:44 +0800
Subject: [PATCH] improve lmk to avoid deadlock issue

Consider the following case:
Task A trigger lmk with a lock held, while process B try to
get this lock, but unfortunately B is the very culprit Task lmk select to
kill.
So B will never be killed, and A will forever select B to kill and
such dead lock trigger softlock up issue.
This patch try to pick the next task to break this loop.

Signed-off-by: wang, biao <biao.wang@intel.com>
Signed-off-by: Zhang Di <di.zhang@intel.com>
---
 drivers/staging/android/lowmemorykiller.c |   14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/drivers/staging/android/lowmemorykiller.c b/drivers/staging/android/lowmemorykiller.c
index feafa17..efabeb7 100644
--- a/drivers/staging/android/lowmemorykiller.c
+++ b/drivers/staging/android/lowmemorykiller.c
@@ -127,11 +127,15 @@ static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
 		if (!p)
 			continue;
 
-		if (test_tsk_thread_flag(p, TIF_MEMDIE) &&
-		    time_before_eq(jiffies, lowmem_deathpending_timeout)) {
-			task_unlock(p);
-			rcu_read_unlock();
-			return 0;
+		if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
+			if (time_before_eq(jiffies, lowmem_deathpending_timeout)) {
+				task_unlock(p);
+				rcu_read_unlock();
+				return 0;
+			} else {
+				task_unlock(p);
+				continue;
+			}
 		}
 		oom_score_adj = p->signal->oom_score_adj;
 		if (oom_score_adj < min_score_adj) {
-- 
1.7.9.5

Re: [PATCH] improve lmk to avoid deadlock issue

[Dan Carpenter]

I'm not an android dev but this patch seems reasonable enough.  There
are some process issues though.

1) The subject should have a subsystem prefix:

[PATCH] Staing: android: lowmemorykiller: improve lmk to avoid deadlock issue


On Thu, Jul 30, 2015 at 06:49:53AM +0000, Wang, Biao wrote:
> From: "wang, biao" <biao.wang@intel.com>
> Date: Thu, 30 Jul 2015 14:14:44 +0800
> Subject: [PATCH] improve lmk to avoid deadlock issue

2)  Don't put these lines in the patch.

> 
> Consider the following case:
> Task A trigger lmk with a lock held, while process B try to
> get this lock, but unfortunately B is the very culprit Task lmk select to
> kill.
> So B will never be killed, and A will forever select B to kill and
> such dead lock trigger softlock up issue.
> This patch try to pick the next task to break this loop.
> 
> Signed-off-by: wang, biao <biao.wang@intel.com>

3) Use capital letters for your name here.  It should match your email
address.

> Signed-off-by: Zhang Di <di.zhang@intel.com>

4) Did Zhang write this patch?  If so then add a From: tag at the top
of the email to give authorship credit.  Signed-off-by tags mean that
the patch went through your hands somehow.  Otherwise use the
Reviewed-by, Reported-by, or Acked-by tag whichever is appropriate.

> ---
>  drivers/staging/android/lowmemorykiller.c |   14 +++++++++-----
>  1 file changed, 9 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/staging/android/lowmemorykiller.c b/drivers/staging/android/lowmemorykiller.c
> index feafa17..efabeb7 100644
> --- a/drivers/staging/android/lowmemorykiller.c
> +++ b/drivers/staging/android/lowmemorykiller.c
> @@ -127,11 +127,15 @@ static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
>  		if (!p)
>  			continue;
>  
> -		if (test_tsk_thread_flag(p, TIF_MEMDIE) &&
> -		    time_before_eq(jiffies, lowmem_deathpending_timeout)) {
> -			task_unlock(p);
> -			rcu_read_unlock();
> -			return 0;
> +		if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
> +			if (time_before_eq(jiffies, lowmem_deathpending_timeout)) {

5) This goes over the 80 character limit.  Could you break it up like
this:

		if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
			if (time_before_eq(jiffies,
					   lowmem_deathpending_timeout)) {
				task_unlock(p);

Anyway, thank for this patch.  Please fix these small process issues
and resend.

regards,
dan carpenter

Re: [PATCH] improve lmk to avoid deadlock issue

[Joe Perches]

On Thu, 2015-07-30 at 13:09 +0300, Dan Carpenter wrote:
> > diff --git a/drivers/staging/android/lowmemorykiller.c b/drivers/staging/android/lowmemorykiller.c
[]
> > @@ -127,11 +127,15 @@ static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
> >  		if (!p)
> >  			continue;
> >  
> > -		if (test_tsk_thread_flag(p, TIF_MEMDIE) &&
> > -		    time_before_eq(jiffies, lowmem_deathpending_timeout)) {
> > -			task_unlock(p);
> > -			rcu_read_unlock();
> > -			return 0;
> > +		if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
> > +			if (time_before_eq(jiffies, lowmem_deathpending_timeout)) {
> 
> 5) This goes over the 80 character limit.  Could you break it up like
> this:
> 
> 		if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
> 			if (time_before_eq(jiffies,
> 					   lowmem_deathpending_timeout)) {
> 				task_unlock(p);
> 
> Anyway, thank for this patch.  Please fix these small process issues
> and resend.

Can the task_unlock in each branch be hoisted?

Another way to write this might be to use time_after:

		if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
			task_unlock(p);
			if (time_after(jiffies, lowmem_deathpending_timeout))
				continue;
			rcu_read_unlock();
			return 0;
		}