From: Dave Chinner <david@fromorbit.com>
To: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: akpm@linux-foundation.org, tkhai@ya.ru, vbabka@suse.cz,
roman.gushchin@linux.dev, djwong@kernel.org, brauner@kernel.org,
paulmck@kernel.org, tytso@mit.edu, steven.price@arm.com,
cel@kernel.org, senozhatsky@chromium.org, yujie.liu@intel.com,
gregkh@linuxfoundation.org, muchun.song@linux.dev,
linux-kernel@vger.kernel.org, linux-mm@kvack.org, x86@kernel.org,
kvm@vger.kernel.org, xen-devel@lists.xenproject.org,
linux-erofs@lists.ozlabs.org,
linux-f2fs-devel@lists.sourceforge.net, cluster-devel@redhat.com,
linux-nfs@vger.kernel.org, linux-mtd@lists.infradead.org,
rcu@vger.kernel.org, netdev@vger.kernel.org,
dri-devel@lists.freedesktop.org, linux-arm-msm@vger.kernel.org,
dm-devel@redhat.com, linux-raid@vger.kernel.org,
linux-bcache@vger.kernel.org,
virtualization@lists.linux-foundation.org,
linux-fsdevel@vger.kernel.org, linux-ext4@vger.kernel.org,
linux-xfs@vger.kernel.org, linux-btrfs@vger.kernel.org
Subject: Re: [PATCH v2 44/47] mm: shrinker: make global slab shrink lockless
Date: Thu, 27 Jul 2023 09:09:42 +1000 [thread overview]
Message-ID: <ZMGnthZAh48JF+eV@dread.disaster.area> (raw)
In-Reply-To: <19ad6d06-8a14-6102-5eae-2134dc2c5061@bytedance.com>
On Wed, Jul 26, 2023 at 05:14:09PM +0800, Qi Zheng wrote:
> On 2023/7/26 16:08, Dave Chinner wrote:
> > On Mon, Jul 24, 2023 at 05:43:51PM +0800, Qi Zheng wrote:
> > > @@ -122,6 +126,13 @@ void shrinker_free_non_registered(struct shrinker *shrinker);
> > > void shrinker_register(struct shrinker *shrinker);
> > > void shrinker_unregister(struct shrinker *shrinker);
> > > +static inline bool shrinker_try_get(struct shrinker *shrinker)
> > > +{
> > > + return READ_ONCE(shrinker->registered) &&
> > > + refcount_inc_not_zero(&shrinker->refcount);
> > > +}
> >
> > Why do we care about shrinker->registered here? If we don't set
> > the refcount to 1 until we have fully initialised everything, then
> > the shrinker code can key entirely off the reference count and
> > none of the lookup code needs to care about whether the shrinker is
> > registered or not.
>
> The purpose of checking shrinker->registered here is to stop running
> shrinker after calling shrinker_free(), which can prevent the following
> situations from happening:
>
> CPU 0 CPU 1
>
> shrinker_try_get()
>
> shrinker_try_get()
>
> shrinker_put()
> shrinker_try_get()
> shrinker_put()
I don't see any race here? What is wrong with having multiple active
users at once?
> >
> > This should use a completion, then it is always safe under
> > rcu_read_lock(). This also gets rid of the shrinker_lock spin lock,
> > which only exists because we can't take a blocking lock under
> > rcu_read_lock(). i.e:
> >
> >
> > void shrinker_put(struct shrinker *shrinker)
> > {
> > if (refcount_dec_and_test(&shrinker->refcount))
> > complete(&shrinker->done);
> > }
> >
> > void shrinker_free()
> > {
> > .....
> > refcount_dec(&shrinker->refcount);
>
> I guess what you mean is shrinker_put(), because here may be the last
> refcount.
Yes, I did.
> > wait_for_completion(&shrinker->done);
> > /*
> > * lookups on the shrinker will now all fail as refcount has
> > * fallen to zero. We can now remove it from the lists and
> > * free it.
> > */
> > down_write(shrinker_rwsem);
> > list_del_rcu(&shrinker->list);
> > up_write(&shrinker_rwsem);
> > call_rcu(shrinker->rcu_head, shrinker_free_rcu_cb);
> > }
> >
> > ....
> >
> > > @@ -686,11 +711,14 @@ EXPORT_SYMBOL(shrinker_free_non_registered);
> > > void shrinker_register(struct shrinker *shrinker)
> > > {
> > > - down_write(&shrinker_rwsem);
> > > - list_add_tail(&shrinker->list, &shrinker_list);
> > > - shrinker->flags |= SHRINKER_REGISTERED;
> > > + refcount_set(&shrinker->refcount, 1);
> > > +
> > > + spin_lock(&shrinker_lock);
> > > + list_add_tail_rcu(&shrinker->list, &shrinker_list);
> > > + spin_unlock(&shrinker_lock);
> > > +
> > > shrinker_debugfs_add(shrinker);
> > > - up_write(&shrinker_rwsem);
> > > + WRITE_ONCE(shrinker->registered, true);
> > > }
> > > EXPORT_SYMBOL(shrinker_register);
> >
> > This just looks wrong - you are trying to use WRITE_ONCE() as a
> > release barrier to indicate that the shrinker is now set up fully.
> > That's not necessary - the refcount is an atomic and along with the
> > rcu locks they should provides all the barriers we need. i.e.
>
> The reason I used WRITE_ONCE() here is because the shrinker->registered
> will be read and written concurrently (read in shrinker_try_get() and
> written in shrinker_free()), which is why I added shrinker::registered
> field instead of using SHRINKER_REGISTERED flag (this can reduce the
> addition of WRITE_ONCE()/READ_ONCE()).
Using WRITE_ONCE/READ_ONCE doesn't provide memory barriers needed to
use the field like this. You need release/acquire memory ordering
here. i.e. smp_store_release()/smp_load_acquire().
As it is, the refcount_inc_not_zero() provides a control dependency,
as documented in include/linux/refcount.h, refcount_dec_and_test()
provides release memory ordering. The only thing I think we may need
is a write barrier before refcount_set(), such that if
refcount_inc_not_zero() sees a non-zero value, it is guaranteed to
see an initialised structure...
i.e. refcounts provide all the existence and initialisation
guarantees. Hence I don't see the need to use shrinker->registered
like this and it can remain a bit flag protected by the
shrinker_rwsem().
> > void shrinker_register(struct shrinker *shrinker)
> > {
> > down_write(&shrinker_rwsem);
> > list_add_tail_rcu(&shrinker->list, &shrinker_list);
> > shrinker->flags |= SHRINKER_REGISTERED;
> > shrinker_debugfs_add(shrinker);
> > up_write(&shrinker_rwsem);
> >
> > /*
> > * now the shrinker is fully set up, take the first
> > * reference to it to indicate that lookup operations are
> > * now allowed to use it via shrinker_try_get().
> > */
> > refcount_set(&shrinker->refcount, 1);
> > }
> >
> > > diff --git a/mm/shrinker_debug.c b/mm/shrinker_debug.c
> > > index f1becfd45853..c5573066adbf 100644
> > > --- a/mm/shrinker_debug.c
> > > +++ b/mm/shrinker_debug.c
> > > @@ -5,6 +5,7 @@
> > > #include <linux/seq_file.h>
> > > #include <linux/shrinker.h>
> > > #include <linux/memcontrol.h>
> > > +#include <linux/rculist.h>
> > > /* defined in vmscan.c */
> > > extern struct rw_semaphore shrinker_rwsem;
> > > @@ -161,17 +162,21 @@ int shrinker_debugfs_add(struct shrinker *shrinker)
> > > {
> > > struct dentry *entry;
> > > char buf[128];
> > > - int id;
> > > -
> > > - lockdep_assert_held(&shrinker_rwsem);
> > > + int id, ret = 0;
> > > /* debugfs isn't initialized yet, add debugfs entries later. */
> > > if (!shrinker_debugfs_root)
> > > return 0;
> > > + down_write(&shrinker_rwsem);
> > > + if (shrinker->debugfs_entry)
> > > + goto fail;
> > > +
> > > id = ida_alloc(&shrinker_debugfs_ida, GFP_KERNEL);
> > > - if (id < 0)
> > > - return id;
> > > + if (id < 0) {
> > > + ret = id;
> > > + goto fail;
> > > + }
> > > shrinker->debugfs_id = id;
> > > snprintf(buf, sizeof(buf), "%s-%d", shrinker->name, id);
> > > @@ -180,7 +185,8 @@ int shrinker_debugfs_add(struct shrinker *shrinker)
> > > entry = debugfs_create_dir(buf, shrinker_debugfs_root);
> > > if (IS_ERR(entry)) {
> > > ida_free(&shrinker_debugfs_ida, id);
> > > - return PTR_ERR(entry);
> > > + ret = PTR_ERR(entry);
> > > + goto fail;
> > > }
> > > shrinker->debugfs_entry = entry;
> > > @@ -188,7 +194,10 @@ int shrinker_debugfs_add(struct shrinker *shrinker)
> > > &shrinker_debugfs_count_fops);
> > > debugfs_create_file("scan", 0220, entry, shrinker,
> > > &shrinker_debugfs_scan_fops);
> > > - return 0;
> > > +
> > > +fail:
> > > + up_write(&shrinker_rwsem);
> > > + return ret;
> > > }
> > > int shrinker_debugfs_rename(struct shrinker *shrinker, const char *fmt, ...)
> > > @@ -243,6 +252,11 @@ struct dentry *shrinker_debugfs_detach(struct shrinker *shrinker,
> > > shrinker->name = NULL;
> > > *debugfs_id = entry ? shrinker->debugfs_id : -1;
> > > + /*
> > > + * Ensure that shrinker->registered has been set to false before
> > > + * shrinker->debugfs_entry is set to NULL.
> > > + */
> > > + smp_wmb();
> > > shrinker->debugfs_entry = NULL;
> > > return entry;
> > > @@ -266,14 +280,26 @@ static int __init shrinker_debugfs_init(void)
> > > shrinker_debugfs_root = dentry;
> > > /* Create debugfs entries for shrinkers registered at boot */
> > > - down_write(&shrinker_rwsem);
> > > - list_for_each_entry(shrinker, &shrinker_list, list)
> > > + rcu_read_lock();
> > > + list_for_each_entry_rcu(shrinker, &shrinker_list, list) {
> > > + if (!shrinker_try_get(shrinker))
> > > + continue;
> > > + rcu_read_unlock();
> > > +
> > > if (!shrinker->debugfs_entry) {
> > > - ret = shrinker_debugfs_add(shrinker);
> > > - if (ret)
> > > - break;
> > > + /* Paired with smp_wmb() in shrinker_debugfs_detach() */
> > > + smp_rmb();
> > > + if (READ_ONCE(shrinker->registered))
> > > + ret = shrinker_debugfs_add(shrinker);
> > > }
> > > - up_write(&shrinker_rwsem);
> > > +
> > > + rcu_read_lock();
> > > + shrinker_put(shrinker);
> > > +
> > > + if (ret)
> > > + break;
> > > + }
> > > + rcu_read_unlock();
> > > return ret;
> > > }
> >
> > And all this churn and complexity can go away because the
> > shrinker_rwsem is still used to protect shrinker_register()
> > entirely....
>
> My consideration is that during this process, there may be a
> driver probe failure and then shrinker_free() is called (the
> shrinker_debugfs_init() is called in late_initcall stage). In
> this case, we need to use RCU+refcount to ensure that the shrinker
> is not freed.
Yeah, you're trying to work around the lack of a
wait_for_completion() call in shrinker_free().
With that, this doesn't need RCU at all, and the iteration can be
done fully under the shrinker_rwsem() safely and so none of this
code needs to change.
Cheers,
Dave.
--
Dave Chinner
david@fromorbit.com
______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/
next prev parent reply other threads:[~2023-07-26 23:10 UTC|newest]
Thread overview: 110+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-07-24 9:43 [PATCH v2 00/47] use refcount+RCU method to implement lockless slab shrink Qi Zheng
2023-07-24 9:43 ` [PATCH v2 01/47] mm: vmscan: move shrinker-related code into a separate file Qi Zheng
2023-07-25 2:35 ` Muchun Song
2023-07-25 3:09 ` Qi Zheng
2023-07-25 3:23 ` Muchun Song
2023-07-25 3:27 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 02/47] mm: shrinker: remove redundant shrinker_rwsem in debugfs operations Qi Zheng
2023-07-25 3:17 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 03/47] mm: shrinker: add infrastructure for dynamically allocating shrinker Qi Zheng
2023-07-24 12:25 ` Peter Zijlstra
2023-07-25 3:01 ` Qi Zheng
2023-07-25 9:02 ` Muchun Song
2023-07-25 9:56 ` Qi Zheng
2023-07-26 7:26 ` Dave Chinner
2023-07-26 9:20 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 04/47] kvm: mmu: dynamically allocate the x86-mmu shrinker Qi Zheng
2023-07-25 9:16 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 05/47] binder: dynamically allocate the android-binder shrinker Qi Zheng
2023-07-24 9:56 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 06/47] drm/ttm: dynamically allocate the drm-ttm_pool shrinker Qi Zheng
2023-07-25 9:19 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 07/47] xenbus/backend: dynamically allocate the xen-backend shrinker Qi Zheng
2023-07-25 9:22 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 08/47] erofs: dynamically allocate the erofs-shrinker Qi Zheng
2023-07-25 9:24 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 09/47] f2fs: dynamically allocate the f2fs-shrinker Qi Zheng
2023-07-25 9:25 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 10/47] gfs2: dynamically allocate the gfs2-glock shrinker Qi Zheng
2023-07-26 6:43 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 11/47] gfs2: dynamically allocate the gfs2-qd shrinker Qi Zheng
2023-07-26 6:49 ` Muchun Song
2023-07-26 9:22 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 12/47] NFSv4.2: dynamically allocate the nfs-xattr shrinkers Qi Zheng
2023-07-26 6:55 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 13/47] nfs: dynamically allocate the nfs-acl shrinker Qi Zheng
2023-07-26 6:57 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 14/47] nfsd: dynamically allocate the nfsd-filecache shrinker Qi Zheng
2023-07-26 6:59 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 15/47] quota: dynamically allocate the dquota-cache shrinker Qi Zheng
2023-07-26 6:59 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 16/47] ubifs: dynamically allocate the ubifs-slab shrinker Qi Zheng
2023-07-26 7:00 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 17/47] rcu: dynamically allocate the rcu-lazy shrinker Qi Zheng
2023-07-26 7:04 ` Muchun Song
2023-07-26 9:24 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 18/47] rcu: dynamically allocate the rcu-kfree shrinker Qi Zheng
2023-07-26 7:05 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 19/47] mm: thp: dynamically allocate the thp-related shrinkers Qi Zheng
2023-07-26 7:10 ` Muchun Song
2023-07-26 9:27 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 20/47] sunrpc: dynamically allocate the sunrpc_cred shrinker Qi Zheng
2023-07-26 7:11 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 21/47] mm: workingset: dynamically allocate the mm-shadow shrinker Qi Zheng
2023-07-26 7:13 ` Muchun Song
2023-07-26 9:28 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 22/47] drm/i915: dynamically allocate the i915_gem_mm shrinker Qi Zheng
2023-07-26 7:15 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 23/47] drm/msm: dynamically allocate the drm-msm_gem shrinker Qi Zheng
2023-07-26 7:24 ` Muchun Song
2023-07-26 9:31 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 24/47] drm/panfrost: dynamically allocate the drm-panfrost shrinker Qi Zheng
2023-07-24 11:17 ` Steven Price
2023-07-25 3:05 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 25/47] dm: dynamically allocate the dm-bufio shrinker Qi Zheng
2023-07-26 7:24 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 26/47] dm zoned: dynamically allocate the dm-zoned-meta shrinker Qi Zheng
2023-07-26 7:25 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 27/47] md/raid5: dynamically allocate the md-raid5 shrinker Qi Zheng
2023-07-26 7:27 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 28/47] bcache: dynamically allocate the md-bcache shrinker Qi Zheng
2023-07-26 7:32 ` Muchun Song
2023-07-26 9:33 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 29/47] vmw_balloon: dynamically allocate the vmw-balloon shrinker Qi Zheng
2023-07-26 7:35 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 30/47] virtio_balloon: dynamically allocate the virtio-balloon shrinker Qi Zheng
2023-07-26 7:36 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 31/47] mbcache: dynamically allocate the mbcache shrinker Qi Zheng
2023-07-26 7:39 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 32/47] ext4: dynamically allocate the ext4-es shrinker Qi Zheng
2023-07-26 7:40 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 33/47] jbd2,ext4: dynamically allocate the jbd2-journal shrinker Qi Zheng
2023-07-26 7:41 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 34/47] nfsd: dynamically allocate the nfsd-client shrinker Qi Zheng
2023-07-24 12:17 ` Jeff Layton
2023-07-24 9:43 ` [PATCH v2 35/47] nfsd: dynamically allocate the nfsd-reply shrinker Qi Zheng
2023-07-24 12:17 ` Jeff Layton
2023-07-24 9:43 ` [PATCH v2 36/47] xfs: dynamically allocate the xfs-buf shrinker Qi Zheng
2023-07-26 7:42 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 37/47] xfs: dynamically allocate the xfs-inodegc shrinker Qi Zheng
2023-07-26 7:42 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 38/47] xfs: dynamically allocate the xfs-qm shrinker Qi Zheng
2023-07-26 7:43 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 39/47] zsmalloc: dynamically allocate the mm-zspool shrinker Qi Zheng
2023-07-26 7:43 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 40/47] fs: super: dynamically allocate the s_shrink Qi Zheng
2023-07-26 7:45 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 41/47] mm: shrinker: remove old APIs Qi Zheng
2023-07-26 7:46 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 42/47] drm/ttm: introduce pool_shrink_rwsem Qi Zheng
2023-07-26 7:56 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 43/47] mm: shrinker: add a secondary array for shrinker_info::{map, nr_deferred} Qi Zheng
2023-07-26 9:30 ` Muchun Song
2023-07-24 9:43 ` [PATCH v2 44/47] mm: shrinker: make global slab shrink lockless Qi Zheng
2023-07-26 8:08 ` Dave Chinner
2023-07-26 9:14 ` Qi Zheng
2023-07-26 23:09 ` Dave Chinner [this message]
2023-07-27 3:34 ` Qi Zheng
2023-07-24 9:43 ` [PATCH v2 45/47] mm: shrinker: make memcg " Qi Zheng
2023-07-24 9:43 ` [PATCH v2 46/47] mm: shrinker: hold write lock to reparent shrinker nr_deferred Qi Zheng
2023-07-24 9:43 ` [PATCH v2 47/47] mm: shrinker: convert shrinker_rwsem to mutex Qi Zheng
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ZMGnthZAh48JF+eV@dread.disaster.area \
--to=david@fromorbit.com \
--cc=akpm@linux-foundation.org \
--cc=brauner@kernel.org \
--cc=cel@kernel.org \
--cc=cluster-devel@redhat.com \
--cc=djwong@kernel.org \
--cc=dm-devel@redhat.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=gregkh@linuxfoundation.org \
--cc=kvm@vger.kernel.org \
--cc=linux-arm-msm@vger.kernel.org \
--cc=linux-bcache@vger.kernel.org \
--cc=linux-btrfs@vger.kernel.org \
--cc=linux-erofs@lists.ozlabs.org \
--cc=linux-ext4@vger.kernel.org \
--cc=linux-f2fs-devel@lists.sourceforge.net \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=linux-mtd@lists.infradead.org \
--cc=linux-nfs@vger.kernel.org \
--cc=linux-raid@vger.kernel.org \
--cc=linux-xfs@vger.kernel.org \
--cc=muchun.song@linux.dev \
--cc=netdev@vger.kernel.org \
--cc=paulmck@kernel.org \
--cc=rcu@vger.kernel.org \
--cc=roman.gushchin@linux.dev \
--cc=senozhatsky@chromium.org \
--cc=steven.price@arm.com \
--cc=tkhai@ya.ru \
--cc=tytso@mit.edu \
--cc=vbabka@suse.cz \
--cc=virtualization@lists.linux-foundation.org \
--cc=x86@kernel.org \
--cc=xen-devel@lists.xenproject.org \
--cc=yujie.liu@intel.com \
--cc=zhengqi.arch@bytedance.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox