Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 7 May 2020 05:49:19 -0700
From:   Jaegeuk Kim <jaegeuk@kernel.org>
To:     Chao Yu <yuchao0@huawei.com>
Cc:     linux-f2fs-devel@lists.sourceforge.net,
        linux-kernel@vger.kernel.org, chao@kernel.org
Subject: Re: [PATCH v2] f2fs: shrink spinlock coverage
Message-ID: <20200507124919.GA197114@google.com>
References: <20200506104542.123575-1-yuchao0@huawei.com>
 <20200506150521.GE107238@google.com>
 <f577a7a8-298f-cdda-7627-1273f83c261c@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <f577a7a8-298f-cdda-7627-1273f83c261c@huawei.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 05/07, Chao Yu wrote:
> On 2020/5/6 23:05, Jaegeuk Kim wrote:
> > On 05/06, Chao Yu wrote:
> >> In f2fs_try_to_free_nids(), .nid_list_lock spinlock critical region will
> >> increase as expected shrink number increase, to avoid spining other CPUs
> >> for long time, it's better to implement like extent cache and nats
> >> shrinker.
> >>
> >> Signed-off-by: Chao Yu <yuchao0@huawei.com>
> >> ---
> >> v2:
> >> - fix unlock wrong spinlock.
> >>  fs/f2fs/node.c | 15 +++++++++++----
> >>  1 file changed, 11 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> >> index 4da0d8713df5..ad0b14f4dab8 100644
> >> --- a/fs/f2fs/node.c
> >> +++ b/fs/f2fs/node.c
> >> @@ -2488,7 +2488,6 @@ void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
> >>  int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
> >>  {
> >>  	struct f2fs_nm_info *nm_i = NM_I(sbi);
> >> -	struct free_nid *i, *next;
> >>  	int nr = nr_shrink;
> >>  
> >>  	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
> >> @@ -2498,14 +2497,22 @@ int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
> >>  		return 0;
> >>  
> >>  	spin_lock(&nm_i->nid_list_lock);
> >> -	list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
> >> -		if (nr_shrink <= 0 ||
> >> -				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
> >> +	while (nr_shrink) {
> >> +		struct free_nid *i;
> >> +
> >> +		if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
> >>  			break;
> >>  
> >> +		i = list_first_entry(&nm_i->free_nid_list,
> >> +					struct free_nid, list);
> >> +		list_del(&i->list);
> >> +		spin_unlock(&nm_i->nid_list_lock);
> >> +
> >>  		__remove_free_nid(sbi, i, FREE_NID);
> > 
> > __remove_free_nid() will do list_del again. btw, how about just splitting out
> 
> Oh, my bad.
> 
> How about moving __remove_free_nid into .nid_list_lock coverage?
> 
> > given nr_shrink into multiple trials?
> 
> Like this?

Yes.

> 
> 	while (shrink) {
> 		batch = DEFAULT_BATCH_NUMBER; // 16
> 		spinlock();
> 		list_for_each_entry_safe() {
> 			if (!shrink || !batch)
> 				break;
> 			remove_item_from_list;
> 			shrink--;
> 			batch--;
> 		}
> 		spin_unlock();
> 	}
> 
> Thanks,
> 
> > 
> >>  		kmem_cache_free(free_nid_slab, i);
> >>  		nr_shrink--;
> >> +
> >> +		spin_lock(&nm_i->nid_list_lock);
> >>  	}
> >>  	spin_unlock(&nm_i->nid_list_lock);
> >>  	mutex_unlock(&nm_i->build_lock);
> >> -- 
> >> 2.18.0.rc1
> > .
> >