Re: [f2fs-dev] [PATCH 1/2] f2fs: handle failed bio allocation

[Jaegeuk Kim]

Hi Chao,

[snip]

> > > >
> > > > -	/* No failure on bio allocation */
> > > > -	bio = bio_alloc(GFP_NOIO, npages);
> > >
> > > How about using __GFP_NOFAIL flag to avoid failing in bio_alloc instead
> > > of adding opencode endless loop in code?
> > >
> > > We can see the reason in this commit 	647757197cd3
> > > ("mm: clarify __GFP_NOFAIL deprecation status ")
> > >
> > > "__GFP_NOFAIL is documented as a deprecated flag since commit
> > > 478352e789f5 ("mm: add comment about deprecation of __GFP_NOFAIL").
> > >
> > > This has discouraged people from using it but in some cases an opencoded
> > > endless loop around allocator has been used instead. So the allocator
> > > is not aware of the de facto __GFP_NOFAIL allocation because this
> > > information was not communicated properly.
> > >
> > > Let's make clear that if the allocation context really cannot afford
> > > failure because there is no good failure policy then using __GFP_NOFAIL
> > > is preferable to opencoding the loop outside of the allocator."
> > >
> > > BTW, I found that f2fs_kmem_cache_alloc also could be replaced, we could
> > > fix them together.
> > 
> > Agreed. I think that can be another patch like this.
> > 
> > From 1579e0d1ada96994c4ec6619fb5b5d9386e77ab3 Mon Sep 17 00:00:00 2001
> > From: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
> > Date: Thu, 20 Aug 2015 08:51:56 -0700
> > Subject: [PATCH] f2fs: use __GFP_NOFAIL to avoid infinite loop
> > 
> > __GFP_NOFAIL can avoid retrying the whole path of kmem_cache_alloc and
> > bio_alloc.
> > 
> > Suggested-by: Chao Yu <chao2.yu@xxxxxxxxxxx>
> > Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
> > ---
> >  fs/f2fs/f2fs.h | 16 +++++-----------
> >  1 file changed, 5 insertions(+), 11 deletions(-)
> > 
> > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > index 00591f7..c78b599 100644
> > --- a/fs/f2fs/f2fs.h
> > +++ b/fs/f2fs/f2fs.h
> > @@ -1244,13 +1244,10 @@ static inline void *f2fs_kmem_cache_alloc(struct kmem_cache *cachep,
> >  						gfp_t flags)
> >  {
> >  	void *entry;
> > -retry:
> > -	entry = kmem_cache_alloc(cachep, flags);
> > -	if (!entry) {
> > -		cond_resched();
> > -		goto retry;
> > -	}
> > 
> > +	entry = kmem_cache_alloc(cachep, flags);
> > +	if (!entry)
> > +		entry = kmem_cache_alloc(cachep, flags | __GFP_NOFAIL);
> 
> The fast + slow path model looks good to me, expect one thing:
> In several paths of checkpoint, caller will grab slab cache with GFP_ATOMIC,
> so in slow path, our flags will be GFP_ATOMIC | __GFP_NOFAIL, I'm not sure
> that the two flags can be used together.
> 
> Should we replace GFP_ATOMIC with GFP_NOFS in flags if caller passed
> GFP_ATOMIC?

Indeed, we need to avoid GFP_ATOMIC as much as possible to mitigate memory
pressure at this moment. Too much abused.

I wrote a patch like this.