Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1684698 > unrolled thread

Re: [RFC PATCH] mm, oom: allow oom reaper to race with exit_mmap

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [RFC PATCH] mm, oom: allow oom reaper to race with exit_mmap David Rientjes <rientjes@google.com> - 2017-07-11 02:00 +0200
    Re: [RFC PATCH] mm, oom: allow oom reaper to race with exit_mmap Michal Hocko <mhocko@kernel.org> - 2017-07-11 09:00 +0200
      Re: [RFC PATCH] mm, oom: allow oom reaper to race with exit_mmap David Rientjes <rientjes@google.com> - 2017-07-11 22:50 +0200
        Re: [RFC PATCH] mm, oom: allow oom reaper to race with exit_mmap Michal Hocko <mhocko@kernel.org> - 2017-07-12 09:20 +0200

#1684698 — Re: [RFC PATCH] mm, oom: allow oom reaper to race with exit_mmap

On Mon, 26 Jun 2017, Michal Hocko wrote:

> diff --git a/mm/mmap.c b/mm/mmap.c
> index 3bd5ecd20d4d..253808e716dc 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -2962,6 +2962,11 @@ void exit_mmap(struct mm_struct *mm)
>  	/* Use -1 here to ensure all VMAs in the mm are unmapped */
>  	unmap_vmas(&tlb, vma, 0, -1);
>  
> +	/*
> +	 * oom reaper might race with exit_mmap so make sure we won't free
> +	 * page tables or unmap VMAs under its feet
> +	 */
> +	down_write(&mm->mmap_sem);
>  	free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
>  	tlb_finish_mmu(&tlb, 0, -1);
>  
> @@ -2974,7 +2979,9 @@ void exit_mmap(struct mm_struct *mm)
>  			nr_accounted += vma_pages(vma);
>  		vma = remove_vma(vma);
>  	}
> +	mm->mmap = NULL;
>  	vm_unacct_memory(nr_accounted);
> +	up_write(&mm->mmap_sem);
>  }
>  
>  /* Insert vm structure into process list sorted by address
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index 0e2c925e7826..5dc0ff22d567 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -472,36 +472,8 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
>  	struct vm_area_struct *vma;
>  	bool ret = true;
>  
> -	/*
> -	 * We have to make sure to not race with the victim exit path
> -	 * and cause premature new oom victim selection:
> -	 * __oom_reap_task_mm		exit_mm
> -	 *   mmget_not_zero
> -	 *				  mmput
> -	 *				    atomic_dec_and_test
> -	 *				  exit_oom_victim
> -	 *				[...]
> -	 *				out_of_memory
> -	 *				  select_bad_process
> -	 *				    # no TIF_MEMDIE task selects new victim
> -	 *  unmap_page_range # frees some memory
> -	 */
> -	mutex_lock(&oom_lock);
> -
> -	if (!down_read_trylock(&mm->mmap_sem)) {
> -		ret = false;
> -		goto unlock_oom;
> -	}
> -
> -	/*
> -	 * increase mm_users only after we know we will reap something so
> -	 * that the mmput_async is called only when we have reaped something
> -	 * and delayed __mmput doesn't matter that much
> -	 */
> -	if (!mmget_not_zero(mm)) {
> -		up_read(&mm->mmap_sem);
> -		goto unlock_oom;
> -	}
> +	if (!down_read_trylock(&mm->mmap_sem))
> +		return false;

I think this should return true if mm->mmap == NULL here.

[toc] | [next] | [standalone]

#1684811

On Mon 10-07-17 16:55:22, David Rientjes wrote:
> On Mon, 26 Jun 2017, Michal Hocko wrote:
> 
> > diff --git a/mm/mmap.c b/mm/mmap.c
> > index 3bd5ecd20d4d..253808e716dc 100644
> > --- a/mm/mmap.c
> > +++ b/mm/mmap.c
> > @@ -2962,6 +2962,11 @@ void exit_mmap(struct mm_struct *mm)
> >  	/* Use -1 here to ensure all VMAs in the mm are unmapped */
> >  	unmap_vmas(&tlb, vma, 0, -1);
> >  
> > +	/*
> > +	 * oom reaper might race with exit_mmap so make sure we won't free
> > +	 * page tables or unmap VMAs under its feet
> > +	 */
> > +	down_write(&mm->mmap_sem);
> >  	free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
> >  	tlb_finish_mmu(&tlb, 0, -1);
> >  
> > @@ -2974,7 +2979,9 @@ void exit_mmap(struct mm_struct *mm)
> >  			nr_accounted += vma_pages(vma);
> >  		vma = remove_vma(vma);
> >  	}
> > +	mm->mmap = NULL;
> >  	vm_unacct_memory(nr_accounted);
> > +	up_write(&mm->mmap_sem);
> >  }
> >  
> >  /* Insert vm structure into process list sorted by address
> > diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> > index 0e2c925e7826..5dc0ff22d567 100644
> > --- a/mm/oom_kill.c
> > +++ b/mm/oom_kill.c
> > @@ -472,36 +472,8 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
> >  	struct vm_area_struct *vma;
> >  	bool ret = true;
> >  
> > -	/*
> > -	 * We have to make sure to not race with the victim exit path
> > -	 * and cause premature new oom victim selection:
> > -	 * __oom_reap_task_mm		exit_mm
> > -	 *   mmget_not_zero
> > -	 *				  mmput
> > -	 *				    atomic_dec_and_test
> > -	 *				  exit_oom_victim
> > -	 *				[...]
> > -	 *				out_of_memory
> > -	 *				  select_bad_process
> > -	 *				    # no TIF_MEMDIE task selects new victim
> > -	 *  unmap_page_range # frees some memory
> > -	 */
> > -	mutex_lock(&oom_lock);
> > -
> > -	if (!down_read_trylock(&mm->mmap_sem)) {
> > -		ret = false;
> > -		goto unlock_oom;
> > -	}
> > -
> > -	/*
> > -	 * increase mm_users only after we know we will reap something so
> > -	 * that the mmput_async is called only when we have reaped something
> > -	 * and delayed __mmput doesn't matter that much
> > -	 */
> > -	if (!mmget_not_zero(mm)) {
> > -		up_read(&mm->mmap_sem);
> > -		goto unlock_oom;
> > -	}
> > +	if (!down_read_trylock(&mm->mmap_sem))
> > +		return false;
> 
> I think this should return true if mm->mmap == NULL here.

This?
---
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index 5dc0ff22d567..e155d1d8064f 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -470,11 +470,14 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
 {
 	struct mmu_gather tlb;
 	struct vm_area_struct *vma;
-	bool ret = true;
 
 	if (!down_read_trylock(&mm->mmap_sem))
 		return false;
 
+	/* There is nothing to reap so bail out without signs in the log */
+	if (!mm->mmap)
+		goto unlock;
+
 	/*
 	 * Tell all users of get_user/copy_from_user etc... that the content
 	 * is no longer stable. No barriers really needed because unmapping
@@ -508,9 +511,10 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
 			K(get_mm_counter(mm, MM_ANONPAGES)),
 			K(get_mm_counter(mm, MM_FILEPAGES)),
 			K(get_mm_counter(mm, MM_SHMEMPAGES)));
+unlock:
 	up_read(&mm->mmap_sem);
 
-	return ret;
+	return true;
 }
 
 #define MAX_OOM_REAP_RETRIES 10
-- 
Michal Hocko
SUSE Labs

[toc] | [prev] | [next] | [standalone]

#1685336

On Tue, 11 Jul 2017, Michal Hocko wrote:

> This?
> ---
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index 5dc0ff22d567..e155d1d8064f 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -470,11 +470,14 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
>  {
>  	struct mmu_gather tlb;
>  	struct vm_area_struct *vma;
> -	bool ret = true;
>  
>  	if (!down_read_trylock(&mm->mmap_sem))
>  		return false;
>  
> +	/* There is nothing to reap so bail out without signs in the log */
> +	if (!mm->mmap)
> +		goto unlock;
> +
>  	/*
>  	 * Tell all users of get_user/copy_from_user etc... that the content
>  	 * is no longer stable. No barriers really needed because unmapping
> @@ -508,9 +511,10 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
>  			K(get_mm_counter(mm, MM_ANONPAGES)),
>  			K(get_mm_counter(mm, MM_FILEPAGES)),
>  			K(get_mm_counter(mm, MM_SHMEMPAGES)));
> +unlock:
>  	up_read(&mm->mmap_sem);
>  
> -	return ret;
> +	return true;
>  }
>  
>  #define MAX_OOM_REAP_RETRIES 10

Yes, this folded in with the original RFC patch appears to work better 
with light testing.

However, I think MAX_OOM_REAP_RETRIES and/or the timeout of HZ/10 needs to 
be increased as well to address the issue that Tetsuo pointed out.  The 
oom reaper shouldn't be required to do any work unless it is resolving a 
livelock, and that scenario should be relatively rare.  The oom killer 
being a natural ultra slow path, I think it would be justifiable to wait 
longer or retry more times than simply 1 second before declaring that 
reaping is not possible.  It reduces the likelihood of additional oom 
killing.

[toc] | [prev] | [next] | [standalone]

#1685560

On Tue 11-07-17 13:40:04, David Rientjes wrote:
> On Tue, 11 Jul 2017, Michal Hocko wrote:
> 
> > This?
> > ---
> > diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> > index 5dc0ff22d567..e155d1d8064f 100644
> > --- a/mm/oom_kill.c
> > +++ b/mm/oom_kill.c
> > @@ -470,11 +470,14 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
> >  {
> >  	struct mmu_gather tlb;
> >  	struct vm_area_struct *vma;
> > -	bool ret = true;
> >  
> >  	if (!down_read_trylock(&mm->mmap_sem))
> >  		return false;
> >  
> > +	/* There is nothing to reap so bail out without signs in the log */
> > +	if (!mm->mmap)
> > +		goto unlock;
> > +
> >  	/*
> >  	 * Tell all users of get_user/copy_from_user etc... that the content
> >  	 * is no longer stable. No barriers really needed because unmapping
> > @@ -508,9 +511,10 @@ static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
> >  			K(get_mm_counter(mm, MM_ANONPAGES)),
> >  			K(get_mm_counter(mm, MM_FILEPAGES)),
> >  			K(get_mm_counter(mm, MM_SHMEMPAGES)));
> > +unlock:
> >  	up_read(&mm->mmap_sem);
> >  
> > -	return ret;
> > +	return true;
> >  }
> >  
> >  #define MAX_OOM_REAP_RETRIES 10
> 
> Yes, this folded in with the original RFC patch appears to work better 
> with light testing.

Yes folding it into the original patch was the plan. I would really
appreciate some Tested-by here.

> However, I think MAX_OOM_REAP_RETRIES and/or the timeout of HZ/10 needs to 
> be increased as well to address the issue that Tetsuo pointed out.  The 
> oom reaper shouldn't be required to do any work unless it is resolving a 
> livelock, and that scenario should be relatively rare.  The oom killer 
> being a natural ultra slow path, I think it would be justifiable to wait 
> longer or retry more times than simply 1 second before declaring that 
> reaping is not possible.  It reduces the likelihood of additional oom 
> killing.

I believe that this is an independent issue and as such it should be
addressed separately along with some data backing up that decision. I am
not against improving the waiting logic. We would need some requeuing
when we cannot reap the victim because we cannot really wait too much
time on a single oom victim considering there might be many victims
queued (because of memcg ooms). This would obviously need some more code
and I am willing to implement that but I would like to see that this is
something that is a real problem first.

Thanks!
-- 
Michal Hocko
SUSE Labs

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web